blob: 15dd59b84e94442c50ae164cb2ec1c130a74bc52 [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:
NeilBrown8f19ccb2011-12-23 10:17:56 +1100138 while (++j < pi->raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 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
NeilBrown8f19ccb2011-12-23 10:17:56 +1100167 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 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
NeilBrown8f19ccb2011-12-23 10:17:56 +1100188 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3e198f72006-01-06 00:20:21 -0800189 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;
NeilBrown30194632011-12-23 10:17:56 +1100280 struct r1conf *conf = r1_bio->mddev->private;
281 int raid_disks = conf->raid_disks;
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100282
NeilBrown8f19ccb2011-12-23 10:17:56 +1100283 for (mirror = 0; mirror < raid_disks * 2; mirror++)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100284 if (r1_bio->bios[mirror] == bio)
285 break;
286
NeilBrown8f19ccb2011-12-23 10:17:56 +1100287 BUG_ON(mirror == raid_disks * 2);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100288 update_head_pos(mirror, r1_bio);
289
290 return mirror;
291}
292
NeilBrown6712ecf2007-09-27 12:47:43 +0200293static void raid1_end_read_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
295 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +1100296 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 int mirror;
NeilBrowne8096362011-10-11 16:49:05 +1100298 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 mirror = r1_bio->read_disk;
301 /*
302 * this branch is our 'one mirror IO has finished' event handler:
303 */
NeilBrownddaf22a2006-01-06 00:20:19 -0800304 update_head_pos(mirror, r1_bio);
305
NeilBrowndd00a992007-05-10 03:15:50 -0700306 if (uptodate)
307 set_bit(R1BIO_Uptodate, &r1_bio->state);
308 else {
309 /* If all other devices have failed, we want to return
310 * the error upwards rather than fail the last device.
311 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 */
NeilBrowndd00a992007-05-10 03:15:50 -0700313 unsigned long flags;
314 spin_lock_irqsave(&conf->device_lock, flags);
315 if (r1_bio->mddev->degraded == conf->raid_disks ||
316 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
317 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)))
318 uptodate = 1;
319 spin_unlock_irqrestore(&conf->device_lock, flags);
320 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
NeilBrowndd00a992007-05-10 03:15:50 -0700322 if (uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 raid_end_bio_io(r1_bio);
NeilBrowndd00a992007-05-10 03:15:50 -0700324 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 /*
326 * oops, read error:
327 */
328 char b[BDEVNAME_SIZE];
Christian Dietrich8bda4702011-07-27 11:00:36 +1000329 printk_ratelimited(
330 KERN_ERR "md/raid1:%s: %s: "
331 "rescheduling sector %llu\n",
332 mdname(conf->mddev),
333 bdevname(conf->mirrors[mirror].rdev->bdev,
334 b),
335 (unsigned long long)r1_bio->sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000336 set_bit(R1BIO_ReadError, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 reschedule_retry(r1_bio);
338 }
339
340 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
342
NeilBrown9f2c9d12011-10-11 16:48:43 +1100343static void close_write(struct r1bio *r1_bio)
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000344{
345 /* it really is the end of this request */
346 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
347 /* free extra copy of the data pages */
348 int i = r1_bio->behind_page_count;
349 while (i--)
350 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
351 kfree(r1_bio->behind_bvecs);
352 r1_bio->behind_bvecs = NULL;
353 }
354 /* clear the bitmap if all writes complete successfully */
355 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
356 r1_bio->sectors,
357 !test_bit(R1BIO_Degraded, &r1_bio->state),
358 test_bit(R1BIO_BehindIO, &r1_bio->state));
359 md_write_end(r1_bio->mddev);
360}
361
NeilBrown9f2c9d12011-10-11 16:48:43 +1100362static void r1_bio_write_done(struct r1bio *r1_bio)
NeilBrown4e780642010-10-19 12:54:01 +1100363{
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000364 if (!atomic_dec_and_test(&r1_bio->remaining))
365 return;
366
367 if (test_bit(R1BIO_WriteError, &r1_bio->state))
368 reschedule_retry(r1_bio);
369 else {
370 close_write(r1_bio);
NeilBrown4367af52011-07-28 11:31:49 +1000371 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
372 reschedule_retry(r1_bio);
373 else
374 raid_end_bio_io(r1_bio);
NeilBrown4e780642010-10-19 12:54:01 +1100375 }
376}
377
NeilBrown6712ecf2007-09-27 12:47:43 +0200378static void raid1_end_write_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
380 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +1100381 struct r1bio *r1_bio = bio->bi_private;
NeilBrowna9701a32005-11-08 21:39:34 -0800382 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
NeilBrowne8096362011-10-11 16:49:05 +1100383 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown04b857f2006-03-09 17:33:46 -0800384 struct bio *to_put = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100386 mirror = find_bio_disk(r1_bio, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
Tejun Heoe9c74692010-09-03 11:56:18 +0200388 /*
389 * 'one mirror IO has finished' event handler:
390 */
Tejun Heoe9c74692010-09-03 11:56:18 +0200391 if (!uptodate) {
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000392 set_bit(WriteErrorSeen,
393 &conf->mirrors[mirror].rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +1100394 if (!test_and_set_bit(WantReplacement,
395 &conf->mirrors[mirror].rdev->flags))
396 set_bit(MD_RECOVERY_NEEDED, &
397 conf->mddev->recovery);
398
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000399 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +1000400 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 /*
Tejun Heoe9c74692010-09-03 11:56:18 +0200402 * Set R1BIO_Uptodate in our master bio, so that we
403 * will return a good error code for to the higher
404 * levels even if IO on some other mirrored buffer
405 * fails.
406 *
407 * The 'master' represents the composite IO operation
408 * to user-side. So if something waits for IO, then it
409 * will wait for the 'master' bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 */
NeilBrown4367af52011-07-28 11:31:49 +1000411 sector_t first_bad;
412 int bad_sectors;
413
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000414 r1_bio->bios[mirror] = NULL;
415 to_put = bio;
Tejun Heoe9c74692010-09-03 11:56:18 +0200416 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
NeilBrown4367af52011-07-28 11:31:49 +1000418 /* Maybe we can clear some bad blocks. */
419 if (is_badblock(conf->mirrors[mirror].rdev,
420 r1_bio->sector, r1_bio->sectors,
421 &first_bad, &bad_sectors)) {
422 r1_bio->bios[mirror] = IO_MADE_GOOD;
423 set_bit(R1BIO_MadeGood, &r1_bio->state);
424 }
425 }
426
Tejun Heoe9c74692010-09-03 11:56:18 +0200427 if (behind) {
428 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
429 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700430
Tejun Heoe9c74692010-09-03 11:56:18 +0200431 /*
432 * In behind mode, we ACK the master bio once the I/O
433 * has safely reached all non-writemostly
434 * disks. Setting the Returned bit ensures that this
435 * gets done only once -- we don't ever want to return
436 * -EIO here, instead we'll wait
437 */
438 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
439 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
440 /* Maybe we can return now */
441 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
442 struct bio *mbio = r1_bio->master_bio;
NeilBrown36a4e1f2011-10-07 14:23:17 +1100443 pr_debug("raid1: behind end write sectors"
444 " %llu-%llu\n",
445 (unsigned long long) mbio->bi_sector,
446 (unsigned long long) mbio->bi_sector +
447 (mbio->bi_size >> 9) - 1);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000448 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700449 }
450 }
451 }
NeilBrown4367af52011-07-28 11:31:49 +1000452 if (r1_bio->bios[mirror] == NULL)
453 rdev_dec_pending(conf->mirrors[mirror].rdev,
454 conf->mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200455
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 * Let's see if all mirrored write operations have finished
458 * already.
459 */
NeilBrownaf6d7b72011-05-11 14:51:19 +1000460 r1_bio_write_done(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700461
NeilBrown04b857f2006-03-09 17:33:46 -0800462 if (to_put)
463 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
466
467/*
468 * This routine returns the disk from which the requested read should
469 * be done. There is a per-array 'next expected sequential IO' sector
470 * number - if this matches on the next IO then we use the last disk.
471 * There is also a per-disk 'last know head position' sector that is
472 * maintained from IRQ contexts, both the normal and the resync IO
473 * completion handlers update this position correctly. If there is no
474 * perfect sequential match then we pick the disk whose head is closest.
475 *
476 * If there are 2 mirrors in the same 2 devices, performance degrades
477 * because position is mirror, not device based.
478 *
479 * The rdev for the device selected will have nr_pending incremented.
480 */
NeilBrowne8096362011-10-11 16:49:05 +1100481static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000483 const sector_t this_sector = r1_bio->sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000484 int sectors;
485 int best_good_sectors;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000486 int start_disk;
NeilBrown76073052011-05-11 14:34:56 +1000487 int best_disk;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000488 int i;
NeilBrown76073052011-05-11 14:34:56 +1000489 sector_t best_dist;
NeilBrown3cb03002011-10-11 16:45:26 +1100490 struct md_rdev *rdev;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000491 int choose_first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493 rcu_read_lock();
494 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700495 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 * device if no resync is going on, or below the resync window.
497 * We take the first readable disk when above the resync window.
498 */
499 retry:
NeilBrownd2eb35a2011-07-28 11:31:48 +1000500 sectors = r1_bio->sectors;
NeilBrown76073052011-05-11 14:34:56 +1000501 best_disk = -1;
502 best_dist = MaxSector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000503 best_good_sectors = 0;
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 if (conf->mddev->recovery_cp < MaxSector &&
506 (this_sector + sectors >= conf->next_resync)) {
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000507 choose_first = 1;
508 start_disk = 0;
509 } else {
510 choose_first = 0;
511 start_disk = conf->last_used;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 }
513
NeilBrown8f19ccb2011-12-23 10:17:56 +1100514 for (i = 0 ; i < conf->raid_disks * 2 ; i++) {
NeilBrown76073052011-05-11 14:34:56 +1000515 sector_t dist;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000516 sector_t first_bad;
517 int bad_sectors;
518
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000519 int disk = start_disk + i;
520 if (disk >= conf->raid_disks)
521 disk -= conf->raid_disks;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700522
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000523 rdev = rcu_dereference(conf->mirrors[disk].rdev);
524 if (r1_bio->bios[disk] == IO_BLOCKED
525 || rdev == NULL
NeilBrown6b740b82012-03-19 12:46:39 +1100526 || test_bit(Unmerged, &rdev->flags)
NeilBrown76073052011-05-11 14:34:56 +1000527 || test_bit(Faulty, &rdev->flags))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000528 continue;
NeilBrown76073052011-05-11 14:34:56 +1000529 if (!test_bit(In_sync, &rdev->flags) &&
530 rdev->recovery_offset < this_sector + sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 continue;
NeilBrown76073052011-05-11 14:34:56 +1000532 if (test_bit(WriteMostly, &rdev->flags)) {
533 /* Don't balance among write-mostly, just
534 * use the first as a last resort */
NeilBrown307729c2012-01-09 01:41:51 +1100535 if (best_disk < 0) {
536 if (is_badblock(rdev, this_sector, sectors,
537 &first_bad, &bad_sectors)) {
538 if (first_bad < this_sector)
539 /* Cannot use this */
540 continue;
541 best_good_sectors = first_bad - this_sector;
542 } else
543 best_good_sectors = sectors;
NeilBrown76073052011-05-11 14:34:56 +1000544 best_disk = disk;
NeilBrown307729c2012-01-09 01:41:51 +1100545 }
NeilBrown76073052011-05-11 14:34:56 +1000546 continue;
547 }
548 /* This is a reasonable device to use. It might
549 * even be best.
550 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000551 if (is_badblock(rdev, this_sector, sectors,
552 &first_bad, &bad_sectors)) {
553 if (best_dist < MaxSector)
554 /* already have a better device */
555 continue;
556 if (first_bad <= this_sector) {
557 /* cannot read here. If this is the 'primary'
558 * device, then we must not read beyond
559 * bad_sectors from another device..
560 */
561 bad_sectors -= (this_sector - first_bad);
562 if (choose_first && sectors > bad_sectors)
563 sectors = bad_sectors;
564 if (best_good_sectors > sectors)
565 best_good_sectors = sectors;
566
567 } else {
568 sector_t good_sectors = first_bad - this_sector;
569 if (good_sectors > best_good_sectors) {
570 best_good_sectors = good_sectors;
571 best_disk = disk;
572 }
573 if (choose_first)
574 break;
575 }
576 continue;
577 } else
578 best_good_sectors = sectors;
579
NeilBrown76073052011-05-11 14:34:56 +1000580 dist = abs(this_sector - conf->mirrors[disk].head_position);
581 if (choose_first
582 /* Don't change to another disk for sequential reads */
583 || conf->next_seq_sect == this_sector
584 || dist == 0
585 /* If device is idle, use it */
586 || atomic_read(&rdev->nr_pending) == 0) {
587 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 break;
589 }
NeilBrown76073052011-05-11 14:34:56 +1000590 if (dist < best_dist) {
591 best_dist = dist;
592 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 }
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
NeilBrown76073052011-05-11 14:34:56 +1000596 if (best_disk >= 0) {
597 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700598 if (!rdev)
599 goto retry;
600 atomic_inc(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000601 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 /* cannot risk returning a device that failed
603 * before we inc'ed nr_pending
604 */
NeilBrown03c902e2006-01-06 00:20:46 -0800605 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 goto retry;
607 }
NeilBrownd2eb35a2011-07-28 11:31:48 +1000608 sectors = best_good_sectors;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700609 conf->next_seq_sect = this_sector + sectors;
NeilBrown76073052011-05-11 14:34:56 +1000610 conf->last_used = best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 }
612 rcu_read_unlock();
NeilBrownd2eb35a2011-07-28 11:31:48 +1000613 *max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
NeilBrown76073052011-05-11 14:34:56 +1000615 return best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616}
617
NeilBrown6b740b82012-03-19 12:46:39 +1100618static int raid1_mergeable_bvec(struct request_queue *q,
619 struct bvec_merge_data *bvm,
620 struct bio_vec *biovec)
621{
622 struct mddev *mddev = q->queuedata;
623 struct r1conf *conf = mddev->private;
624 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
625 int max = biovec->bv_len;
626
627 if (mddev->merge_check_needed) {
628 int disk;
629 rcu_read_lock();
630 for (disk = 0; disk < conf->raid_disks * 2; disk++) {
631 struct md_rdev *rdev = rcu_dereference(
632 conf->mirrors[disk].rdev);
633 if (rdev && !test_bit(Faulty, &rdev->flags)) {
634 struct request_queue *q =
635 bdev_get_queue(rdev->bdev);
636 if (q->merge_bvec_fn) {
637 bvm->bi_sector = sector +
638 rdev->data_offset;
639 bvm->bi_bdev = rdev->bdev;
640 max = min(max, q->merge_bvec_fn(
641 q, bvm, biovec));
642 }
643 }
644 }
645 rcu_read_unlock();
646 }
647 return max;
648
649}
650
NeilBrownfd01b882011-10-11 16:47:53 +1100651int md_raid1_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700652{
NeilBrowne8096362011-10-11 16:49:05 +1100653 struct r1conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700654 int i, ret = 0;
655
NeilBrown34db0cd2011-10-11 16:50:01 +1100656 if ((bits & (1 << BDI_async_congested)) &&
657 conf->pending_count >= max_queued_requests)
658 return 1;
659
NeilBrown0d129222006-10-03 01:15:54 -0700660 rcu_read_lock();
NeilBrownf53e29f2012-02-13 14:24:05 +1100661 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100662 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700663 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200664 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700665
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500666 BUG_ON(!q);
667
NeilBrown0d129222006-10-03 01:15:54 -0700668 /* Note the '|| 1' - when read_balance prefers
669 * non-congested targets, it can be removed
670 */
Alexander Beregalov91a9e992009-04-07 01:35:56 +0400671 if ((bits & (1<<BDI_async_congested)) || 1)
NeilBrown0d129222006-10-03 01:15:54 -0700672 ret |= bdi_congested(&q->backing_dev_info, bits);
673 else
674 ret &= bdi_congested(&q->backing_dev_info, bits);
675 }
676 }
677 rcu_read_unlock();
678 return ret;
679}
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500680EXPORT_SYMBOL_GPL(md_raid1_congested);
NeilBrown0d129222006-10-03 01:15:54 -0700681
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500682static int raid1_congested(void *data, int bits)
683{
NeilBrownfd01b882011-10-11 16:47:53 +1100684 struct mddev *mddev = data;
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500685
686 return mddev_congested(mddev, bits) ||
687 md_raid1_congested(mddev, bits);
688}
NeilBrown0d129222006-10-03 01:15:54 -0700689
NeilBrowne8096362011-10-11 16:49:05 +1100690static void flush_pending_writes(struct r1conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800691{
692 /* Any writes that have been queued but are awaiting
693 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800694 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800695 spin_lock_irq(&conf->device_lock);
696
697 if (conf->pending_bio_list.head) {
698 struct bio *bio;
699 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100700 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800701 spin_unlock_irq(&conf->device_lock);
702 /* flush any pending bitmap writes to
703 * disk before proceeding w/ I/O */
704 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100705 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800706
707 while (bio) { /* submit pending writes */
708 struct bio *next = bio->bi_next;
709 bio->bi_next = NULL;
710 generic_make_request(bio);
711 bio = next;
712 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800713 } else
714 spin_unlock_irq(&conf->device_lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100715}
716
NeilBrown17999be2006-01-06 00:20:12 -0800717/* Barriers....
718 * Sometimes we need to suspend IO while we do something else,
719 * either some resync/recovery, or reconfigure the array.
720 * To do this we raise a 'barrier'.
721 * The 'barrier' is a counter that can be raised multiple times
722 * to count how many activities are happening which preclude
723 * normal IO.
724 * We can only raise the barrier if there is no pending IO.
725 * i.e. if nr_pending == 0.
726 * We choose only to raise the barrier if no-one is waiting for the
727 * barrier to go down. This means that as soon as an IO request
728 * is ready, no other operations which require a barrier will start
729 * until the IO request has had a chance.
730 *
731 * So: regular IO calls 'wait_barrier'. When that returns there
732 * is no backgroup IO happening, It must arrange to call
733 * allow_barrier when it has finished its IO.
734 * backgroup IO calls must call raise_barrier. Once that returns
735 * there is no normal IO happeing. It must arrange to call
736 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 */
738#define RESYNC_DEPTH 32
739
NeilBrowne8096362011-10-11 16:49:05 +1100740static void raise_barrier(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
742 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800743
744 /* Wait until no block IO is waiting */
745 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
NeilBrownc3b328a2011-04-18 18:25:43 +1000746 conf->resync_lock, );
NeilBrown17999be2006-01-06 00:20:12 -0800747
748 /* block any new IO from starting */
749 conf->barrier++;
750
NeilBrown046abee2010-10-26 15:46:20 +1100751 /* Now wait for all pending IO to complete */
NeilBrown17999be2006-01-06 00:20:12 -0800752 wait_event_lock_irq(conf->wait_barrier,
753 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
NeilBrownc3b328a2011-04-18 18:25:43 +1000754 conf->resync_lock, );
NeilBrown17999be2006-01-06 00:20:12 -0800755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 spin_unlock_irq(&conf->resync_lock);
757}
758
NeilBrowne8096362011-10-11 16:49:05 +1100759static void lower_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800760{
761 unsigned long flags;
NeilBrown709ae482009-12-14 12:49:51 +1100762 BUG_ON(conf->barrier <= 0);
NeilBrown17999be2006-01-06 00:20:12 -0800763 spin_lock_irqsave(&conf->resync_lock, flags);
764 conf->barrier--;
765 spin_unlock_irqrestore(&conf->resync_lock, flags);
766 wake_up(&conf->wait_barrier);
767}
768
NeilBrowne8096362011-10-11 16:49:05 +1100769static void wait_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800770{
771 spin_lock_irq(&conf->resync_lock);
772 if (conf->barrier) {
773 conf->nr_waiting++;
NeilBrownd6b42dc2012-03-19 12:46:38 +1100774 /* Wait for the barrier to drop.
775 * However if there are already pending
776 * requests (preventing the barrier from
777 * rising completely), and the
778 * pre-process bio queue isn't empty,
779 * then don't wait, as we need to empty
780 * that queue to get the nr_pending
781 * count down.
782 */
783 wait_event_lock_irq(conf->wait_barrier,
784 !conf->barrier ||
785 (conf->nr_pending &&
786 current->bio_list &&
787 !bio_list_empty(current->bio_list)),
NeilBrown17999be2006-01-06 00:20:12 -0800788 conf->resync_lock,
NeilBrownd6b42dc2012-03-19 12:46:38 +1100789 );
NeilBrown17999be2006-01-06 00:20:12 -0800790 conf->nr_waiting--;
791 }
792 conf->nr_pending++;
793 spin_unlock_irq(&conf->resync_lock);
794}
795
NeilBrowne8096362011-10-11 16:49:05 +1100796static void allow_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800797{
798 unsigned long flags;
799 spin_lock_irqsave(&conf->resync_lock, flags);
800 conf->nr_pending--;
801 spin_unlock_irqrestore(&conf->resync_lock, flags);
802 wake_up(&conf->wait_barrier);
803}
804
NeilBrowne8096362011-10-11 16:49:05 +1100805static void freeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800806{
807 /* stop syncio and normal IO and wait for everything to
808 * go quite.
809 * We increment barrier and nr_waiting, and then
NeilBrown1c830532008-03-04 14:29:35 -0800810 * wait until nr_pending match nr_queued+1
811 * This is called in the context of one normal IO request
812 * that has failed. Thus any sync request that might be pending
813 * will be blocked by nr_pending, and we need to wait for
814 * pending IO requests to complete or be queued for re-try.
815 * Thus the number queued (nr_queued) plus this request (1)
816 * must match the number of pending IOs (nr_pending) before
817 * we continue.
NeilBrownddaf22a2006-01-06 00:20:19 -0800818 */
819 spin_lock_irq(&conf->resync_lock);
820 conf->barrier++;
821 conf->nr_waiting++;
822 wait_event_lock_irq(conf->wait_barrier,
NeilBrown1c830532008-03-04 14:29:35 -0800823 conf->nr_pending == conf->nr_queued+1,
NeilBrownddaf22a2006-01-06 00:20:19 -0800824 conf->resync_lock,
NeilBrownc3b328a2011-04-18 18:25:43 +1000825 flush_pending_writes(conf));
NeilBrownddaf22a2006-01-06 00:20:19 -0800826 spin_unlock_irq(&conf->resync_lock);
827}
NeilBrowne8096362011-10-11 16:49:05 +1100828static void unfreeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800829{
830 /* reverse the effect of the freeze */
831 spin_lock_irq(&conf->resync_lock);
832 conf->barrier--;
833 conf->nr_waiting--;
834 wake_up(&conf->wait_barrier);
835 spin_unlock_irq(&conf->resync_lock);
836}
837
NeilBrown17999be2006-01-06 00:20:12 -0800838
NeilBrown4e780642010-10-19 12:54:01 +1100839/* duplicate the data pages for behind I/O
NeilBrown4e780642010-10-19 12:54:01 +1100840 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100841static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
NeilBrown4b6d2872005-09-09 16:23:47 -0700842{
843 int i;
844 struct bio_vec *bvec;
NeilBrown2ca68f52011-07-28 11:32:10 +1000845 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
NeilBrown4b6d2872005-09-09 16:23:47 -0700846 GFP_NOIO);
NeilBrown2ca68f52011-07-28 11:32:10 +1000847 if (unlikely(!bvecs))
NeilBrownaf6d7b72011-05-11 14:51:19 +1000848 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700849
NeilBrown4b6d2872005-09-09 16:23:47 -0700850 bio_for_each_segment(bvec, bio, i) {
NeilBrown2ca68f52011-07-28 11:32:10 +1000851 bvecs[i] = *bvec;
852 bvecs[i].bv_page = alloc_page(GFP_NOIO);
853 if (unlikely(!bvecs[i].bv_page))
NeilBrown4b6d2872005-09-09 16:23:47 -0700854 goto do_sync_io;
NeilBrown2ca68f52011-07-28 11:32:10 +1000855 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
856 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
857 kunmap(bvecs[i].bv_page);
NeilBrown4b6d2872005-09-09 16:23:47 -0700858 kunmap(bvec->bv_page);
859 }
NeilBrown2ca68f52011-07-28 11:32:10 +1000860 r1_bio->behind_bvecs = bvecs;
NeilBrownaf6d7b72011-05-11 14:51:19 +1000861 r1_bio->behind_page_count = bio->bi_vcnt;
862 set_bit(R1BIO_BehindIO, &r1_bio->state);
863 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700864
865do_sync_io:
NeilBrownaf6d7b72011-05-11 14:51:19 +1000866 for (i = 0; i < bio->bi_vcnt; i++)
NeilBrown2ca68f52011-07-28 11:32:10 +1000867 if (bvecs[i].bv_page)
868 put_page(bvecs[i].bv_page);
869 kfree(bvecs);
NeilBrown36a4e1f2011-10-07 14:23:17 +1100870 pr_debug("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
NeilBrown4b6d2872005-09-09 16:23:47 -0700871}
872
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -0700873static void make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874{
NeilBrowne8096362011-10-11 16:49:05 +1100875 struct r1conf *conf = mddev->private;
NeilBrown0f6d02d2011-10-11 16:48:46 +1100876 struct mirror_info *mirror;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100877 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 struct bio *read_bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +1000879 int i, disks;
NeilBrown84255d12008-05-23 13:04:32 -0700880 struct bitmap *bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -0700881 unsigned long flags;
Jens Axboea3623572005-11-01 09:26:16 +0100882 const int rw = bio_data_dir(bio);
NeilBrown2c7d46e2010-08-18 16:16:05 +1000883 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
Tejun Heoe9c74692010-09-03 11:56:18 +0200884 const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA));
NeilBrown3cb03002011-10-11 16:45:26 +1100885 struct md_rdev *blocked_rdev;
NeilBrownc3b328a2011-04-18 18:25:43 +1000886 int plugged;
NeilBrown1f68f0c2011-07-28 11:31:48 +1000887 int first_clone;
888 int sectors_handled;
889 int max_sectors;
NeilBrown191ea9b2005-06-21 17:17:23 -0700890
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 /*
892 * Register the new request and wait if the reconstruction
893 * thread has put up a bar for new requests.
894 * Continue immediately if no resync is active currently.
895 */
NeilBrown62de6082006-05-01 12:15:47 -0700896
NeilBrown3d310eb2005-06-21 17:17:26 -0700897 md_write_start(mddev, bio); /* wait on superblock update early */
898
NeilBrown6eef4b22009-12-14 12:49:51 +1100899 if (bio_data_dir(bio) == WRITE &&
900 bio->bi_sector + bio->bi_size/512 > mddev->suspend_lo &&
901 bio->bi_sector < mddev->suspend_hi) {
902 /* As the suspend_* range is controlled by
903 * userspace, we want an interruptible
904 * wait.
905 */
906 DEFINE_WAIT(w);
907 for (;;) {
908 flush_signals(current);
909 prepare_to_wait(&conf->wait_barrier,
910 &w, TASK_INTERRUPTIBLE);
911 if (bio->bi_sector + bio->bi_size/512 <= mddev->suspend_lo ||
912 bio->bi_sector >= mddev->suspend_hi)
913 break;
914 schedule();
915 }
916 finish_wait(&conf->wait_barrier, &w);
917 }
NeilBrown62de6082006-05-01 12:15:47 -0700918
NeilBrown17999be2006-01-06 00:20:12 -0800919 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
NeilBrown84255d12008-05-23 13:04:32 -0700921 bitmap = mddev->bitmap;
922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 /*
924 * make_request() can abort the operation when READA is being
925 * used and no empty request is available.
926 *
927 */
928 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
929
930 r1_bio->master_bio = bio;
931 r1_bio->sectors = bio->bi_size >> 9;
NeilBrown191ea9b2005-06-21 17:17:23 -0700932 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 r1_bio->mddev = mddev;
934 r1_bio->sector = bio->bi_sector;
935
NeilBrownd2eb35a2011-07-28 11:31:48 +1000936 /* We might need to issue multiple reads to different
937 * devices if there are bad blocks around, so we keep
938 * track of the number of reads in bio->bi_phys_segments.
939 * If this is 0, there is only one r1_bio and no locking
940 * will be needed when requests complete. If it is
941 * non-zero, then it is the number of not-completed requests.
942 */
943 bio->bi_phys_segments = 0;
944 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
945
Jens Axboea3623572005-11-01 09:26:16 +0100946 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 /*
948 * read balancing logic:
949 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000950 int rdisk;
951
952read_again:
953 rdisk = read_balance(conf, r1_bio, &max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
955 if (rdisk < 0) {
956 /* couldn't find anywhere to read from */
957 raid_end_bio_io(r1_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200958 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
960 mirror = conf->mirrors + rdisk;
961
NeilBrowne5551902010-03-31 11:21:44 +1100962 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
963 bitmap) {
964 /* Reading from a write-mostly device must
965 * take care not to over-take any writes
966 * that are 'behind'
967 */
968 wait_event(bitmap->behind_wait,
969 atomic_read(&bitmap->behind_writes) == 0);
970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 r1_bio->read_disk = rdisk;
972
NeilBrowna167f662010-10-26 18:31:13 +1100973 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000974 md_trim_bio(read_bio, r1_bio->sector - bio->bi_sector,
975 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
977 r1_bio->bios[rdisk] = read_bio;
978
979 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
980 read_bio->bi_bdev = mirror->rdev->bdev;
981 read_bio->bi_end_io = raid1_end_read_request;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200982 read_bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 read_bio->bi_private = r1_bio;
984
NeilBrownd2eb35a2011-07-28 11:31:48 +1000985 if (max_sectors < r1_bio->sectors) {
986 /* could not read all from this device, so we will
987 * need another r1_bio.
988 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000989
990 sectors_handled = (r1_bio->sector + max_sectors
991 - bio->bi_sector);
992 r1_bio->sectors = max_sectors;
993 spin_lock_irq(&conf->device_lock);
994 if (bio->bi_phys_segments == 0)
995 bio->bi_phys_segments = 2;
996 else
997 bio->bi_phys_segments++;
998 spin_unlock_irq(&conf->device_lock);
999 /* Cannot call generic_make_request directly
1000 * as that will be queued in __make_request
1001 * and subsequent mempool_alloc might block waiting
1002 * for it. So hand bio over to raid1d.
1003 */
1004 reschedule_retry(r1_bio);
1005
1006 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1007
1008 r1_bio->master_bio = bio;
1009 r1_bio->sectors = (bio->bi_size >> 9) - sectors_handled;
1010 r1_bio->state = 0;
1011 r1_bio->mddev = mddev;
1012 r1_bio->sector = bio->bi_sector + sectors_handled;
1013 goto read_again;
1014 } else
1015 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001016 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 }
1018
1019 /*
1020 * WRITE:
1021 */
NeilBrown34db0cd2011-10-11 16:50:01 +11001022 if (conf->pending_count >= max_queued_requests) {
1023 md_wakeup_thread(mddev->thread);
1024 wait_event(conf->wait_barrier,
1025 conf->pending_count < max_queued_requests);
1026 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001027 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 * inc refcount on their rdev. Record them by setting
1029 * bios[x] to bio
NeilBrown1f68f0c2011-07-28 11:31:48 +10001030 * If there are known/acknowledged bad blocks on any device on
1031 * which we have seen a write error, we want to avoid writing those
1032 * blocks.
1033 * This potentially requires several writes to write around
1034 * the bad blocks. Each set of writes gets it's own r1bio
1035 * with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001037 plugged = mddev_check_plugged(mddev);
1038
NeilBrown8f19ccb2011-12-23 10:17:56 +11001039 disks = conf->raid_disks * 2;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001040 retry_write:
1041 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 rcu_read_lock();
NeilBrown1f68f0c2011-07-28 11:31:48 +10001043 max_sectors = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001045 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001046 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1047 atomic_inc(&rdev->nr_pending);
1048 blocked_rdev = rdev;
1049 break;
1050 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001051 r1_bio->bios[i] = NULL;
NeilBrown6b740b82012-03-19 12:46:39 +11001052 if (!rdev || test_bit(Faulty, &rdev->flags)
1053 || test_bit(Unmerged, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11001054 if (i < conf->raid_disks)
1055 set_bit(R1BIO_Degraded, &r1_bio->state);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001056 continue;
1057 }
1058
1059 atomic_inc(&rdev->nr_pending);
1060 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1061 sector_t first_bad;
1062 int bad_sectors;
1063 int is_bad;
1064
1065 is_bad = is_badblock(rdev, r1_bio->sector,
1066 max_sectors,
1067 &first_bad, &bad_sectors);
1068 if (is_bad < 0) {
1069 /* mustn't write here until the bad block is
1070 * acknowledged*/
1071 set_bit(BlockedBadBlocks, &rdev->flags);
1072 blocked_rdev = rdev;
1073 break;
NeilBrown964147d2010-05-18 15:27:13 +10001074 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001075 if (is_bad && first_bad <= r1_bio->sector) {
1076 /* Cannot write here at all */
1077 bad_sectors -= (r1_bio->sector - first_bad);
1078 if (bad_sectors < max_sectors)
1079 /* mustn't write more than bad_sectors
1080 * to other devices yet
1081 */
1082 max_sectors = bad_sectors;
1083 rdev_dec_pending(rdev, mddev);
1084 /* We don't set R1BIO_Degraded as that
1085 * only applies if the disk is
1086 * missing, so it might be re-added,
1087 * and we want to know to recover this
1088 * chunk.
1089 * In this case the device is here,
1090 * and the fact that this chunk is not
1091 * in-sync is recorded in the bad
1092 * block log
1093 */
1094 continue;
1095 }
1096 if (is_bad) {
1097 int good_sectors = first_bad - r1_bio->sector;
1098 if (good_sectors < max_sectors)
1099 max_sectors = good_sectors;
1100 }
1101 }
1102 r1_bio->bios[i] = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 }
1104 rcu_read_unlock();
1105
Dan Williams6bfe0b42008-04-30 00:52:32 -07001106 if (unlikely(blocked_rdev)) {
1107 /* Wait for this device to become unblocked */
1108 int j;
1109
1110 for (j = 0; j < i; j++)
1111 if (r1_bio->bios[j])
1112 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001113 r1_bio->state = 0;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001114 allow_barrier(conf);
1115 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1116 wait_barrier(conf);
1117 goto retry_write;
1118 }
1119
NeilBrown1f68f0c2011-07-28 11:31:48 +10001120 if (max_sectors < r1_bio->sectors) {
1121 /* We are splitting this write into multiple parts, so
1122 * we need to prepare for allocating another r1_bio.
1123 */
1124 r1_bio->sectors = max_sectors;
1125 spin_lock_irq(&conf->device_lock);
1126 if (bio->bi_phys_segments == 0)
1127 bio->bi_phys_segments = 2;
1128 else
1129 bio->bi_phys_segments++;
1130 spin_unlock_irq(&conf->device_lock);
NeilBrown191ea9b2005-06-21 17:17:23 -07001131 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001132 sectors_handled = r1_bio->sector + max_sectors - bio->bi_sector;
NeilBrown4b6d2872005-09-09 16:23:47 -07001133
NeilBrown4e780642010-10-19 12:54:01 +11001134 atomic_set(&r1_bio->remaining, 1);
NeilBrown4b6d2872005-09-09 16:23:47 -07001135 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -07001136
NeilBrown1f68f0c2011-07-28 11:31:48 +10001137 first_clone = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 for (i = 0; i < disks; i++) {
1139 struct bio *mbio;
1140 if (!r1_bio->bios[i])
1141 continue;
1142
NeilBrowna167f662010-10-26 18:31:13 +11001143 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001144 md_trim_bio(mbio, r1_bio->sector - bio->bi_sector, max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
NeilBrown1f68f0c2011-07-28 11:31:48 +10001146 if (first_clone) {
1147 /* do behind I/O ?
1148 * Not if there are too many, or cannot
1149 * allocate memory, or a reader on WriteMostly
1150 * is waiting for behind writes to flush */
1151 if (bitmap &&
1152 (atomic_read(&bitmap->behind_writes)
1153 < mddev->bitmap_info.max_write_behind) &&
1154 !waitqueue_active(&bitmap->behind_wait))
1155 alloc_behind_pages(mbio, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
NeilBrown1f68f0c2011-07-28 11:31:48 +10001157 bitmap_startwrite(bitmap, r1_bio->sector,
1158 r1_bio->sectors,
1159 test_bit(R1BIO_BehindIO,
1160 &r1_bio->state));
1161 first_clone = 0;
1162 }
NeilBrown2ca68f52011-07-28 11:32:10 +10001163 if (r1_bio->behind_bvecs) {
NeilBrown4b6d2872005-09-09 16:23:47 -07001164 struct bio_vec *bvec;
1165 int j;
1166
1167 /* Yes, I really want the '__' version so that
1168 * we clear any unused pointer in the io_vec, rather
1169 * than leave them unchanged. This is important
1170 * because when we come to free the pages, we won't
NeilBrown046abee2010-10-26 15:46:20 +11001171 * know the original bi_idx, so we just free
NeilBrown4b6d2872005-09-09 16:23:47 -07001172 * them all
1173 */
1174 __bio_for_each_segment(bvec, mbio, j, 0)
NeilBrown2ca68f52011-07-28 11:32:10 +10001175 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
NeilBrown4b6d2872005-09-09 16:23:47 -07001176 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1177 atomic_inc(&r1_bio->behind_remaining);
1178 }
1179
NeilBrown1f68f0c2011-07-28 11:31:48 +10001180 r1_bio->bios[i] = mbio;
1181
1182 mbio->bi_sector = (r1_bio->sector +
1183 conf->mirrors[i].rdev->data_offset);
1184 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1185 mbio->bi_end_io = raid1_end_write_request;
1186 mbio->bi_rw = WRITE | do_flush_fua | do_sync;
1187 mbio->bi_private = r1_bio;
1188
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 atomic_inc(&r1_bio->remaining);
NeilBrown4e780642010-10-19 12:54:01 +11001190 spin_lock_irqsave(&conf->device_lock, flags);
1191 bio_list_add(&conf->pending_bio_list, mbio);
NeilBrown34db0cd2011-10-11 16:50:01 +11001192 conf->pending_count++;
NeilBrown4e780642010-10-19 12:54:01 +11001193 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 }
NeilBrown079fa162011-09-10 17:21:23 +10001195 /* Mustn't call r1_bio_write_done before this next test,
1196 * as it could result in the bio being freed.
1197 */
NeilBrown1f68f0c2011-07-28 11:31:48 +10001198 if (sectors_handled < (bio->bi_size >> 9)) {
NeilBrown079fa162011-09-10 17:21:23 +10001199 r1_bio_write_done(r1_bio);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001200 /* We need another r1_bio. It has already been counted
1201 * in bio->bi_phys_segments
1202 */
1203 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1204 r1_bio->master_bio = bio;
1205 r1_bio->sectors = (bio->bi_size >> 9) - sectors_handled;
1206 r1_bio->state = 0;
1207 r1_bio->mddev = mddev;
1208 r1_bio->sector = bio->bi_sector + sectors_handled;
1209 goto retry_write;
1210 }
1211
NeilBrown079fa162011-09-10 17:21:23 +10001212 r1_bio_write_done(r1_bio);
1213
1214 /* In case raid1d snuck in to freeze_array */
1215 wake_up(&conf->wait_barrier);
1216
NeilBrownc3b328a2011-04-18 18:25:43 +10001217 if (do_sync || !bitmap || !plugged)
Lars Ellenberge3881a62007-01-10 23:15:37 -08001218 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
NeilBrownfd01b882011-10-11 16:47:53 +11001221static void status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222{
NeilBrowne8096362011-10-11 16:49:05 +11001223 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 int i;
1225
1226 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001227 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001228 rcu_read_lock();
1229 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001230 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001232 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1233 }
1234 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 seq_printf(seq, "]");
1236}
1237
1238
NeilBrownfd01b882011-10-11 16:47:53 +11001239static void error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240{
1241 char b[BDEVNAME_SIZE];
NeilBrowne8096362011-10-11 16:49:05 +11001242 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
1244 /*
1245 * If it is not operational, then we have already marked it as dead
1246 * else if it is the last working disks, ignore the error, let the
1247 * next level up know.
1248 * else mark the drive as failed
1249 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001250 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001251 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 /*
1253 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001254 * normal single drive.
1255 * However don't try a recovery from this drive as
1256 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 */
NeilBrown53890422011-07-27 11:00:36 +10001258 conf->recovery_disabled = mddev->recovery_disabled;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001260 }
NeilBrownde393cd2011-07-28 11:31:48 +10001261 set_bit(Blocked, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001262 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1263 unsigned long flags;
1264 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -07001266 set_bit(Faulty, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001267 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 /*
1269 * if recovery is running, make sure it aborts.
1270 */
NeilBrowndfc70642008-05-23 13:04:39 -07001271 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrowndd00a992007-05-10 03:15:50 -07001272 } else
1273 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001274 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Joe Perches067032b2011-01-14 09:14:33 +11001275 printk(KERN_ALERT
1276 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1277 "md/raid1:%s: Operation continuing on %d devices.\n",
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001278 mdname(mddev), bdevname(rdev->bdev, b),
1279 mdname(mddev), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280}
1281
NeilBrowne8096362011-10-11 16:49:05 +11001282static void print_conf(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283{
1284 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001286 printk(KERN_DEBUG "RAID1 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 if (!conf) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001288 printk(KERN_DEBUG "(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 return;
1290 }
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001291 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 conf->raid_disks);
1293
NeilBrownddac7c72006-08-31 21:27:36 -07001294 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 for (i = 0; i < conf->raid_disks; i++) {
1296 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001297 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownddac7c72006-08-31 21:27:36 -07001298 if (rdev)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001299 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -07001300 i, !test_bit(In_sync, &rdev->flags),
1301 !test_bit(Faulty, &rdev->flags),
1302 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 }
NeilBrownddac7c72006-08-31 21:27:36 -07001304 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305}
1306
NeilBrowne8096362011-10-11 16:49:05 +11001307static void close_sync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308{
NeilBrown17999be2006-01-06 00:20:12 -08001309 wait_barrier(conf);
1310 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 mempool_destroy(conf->r1buf_pool);
1313 conf->r1buf_pool = NULL;
1314}
1315
NeilBrownfd01b882011-10-11 16:47:53 +11001316static int raid1_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317{
1318 int i;
NeilBrowne8096362011-10-11 16:49:05 +11001319 struct r1conf *conf = mddev->private;
NeilBrown6b965622010-08-18 11:56:59 +10001320 int count = 0;
1321 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
1323 /*
1324 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001325 * and mark them readable.
1326 * Called under mddev lock, so rcu protection not needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 */
1328 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001329 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001330 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1331 if (repl
1332 && repl->recovery_offset == MaxSector
1333 && !test_bit(Faulty, &repl->flags)
1334 && !test_and_set_bit(In_sync, &repl->flags)) {
1335 /* replacement has just become active */
1336 if (!rdev ||
1337 !test_and_clear_bit(In_sync, &rdev->flags))
1338 count++;
1339 if (rdev) {
1340 /* Replaced device not technically
1341 * faulty, but we need to be sure
1342 * it gets removed and never re-added
1343 */
1344 set_bit(Faulty, &rdev->flags);
1345 sysfs_notify_dirent_safe(
1346 rdev->sysfs_state);
1347 }
1348 }
NeilBrownddac7c72006-08-31 21:27:36 -07001349 if (rdev
1350 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001351 && !test_and_set_bit(In_sync, &rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001352 count++;
Jonathan Brassow654e8b52011-07-27 11:00:36 +10001353 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 }
1355 }
NeilBrown6b965622010-08-18 11:56:59 +10001356 spin_lock_irqsave(&conf->device_lock, flags);
1357 mddev->degraded -= count;
1358 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001361 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362}
1363
1364
NeilBrownfd01b882011-10-11 16:47:53 +11001365static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
NeilBrowne8096362011-10-11 16:49:05 +11001367 struct r1conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001368 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001369 int mirror = 0;
NeilBrown0f6d02d2011-10-11 16:48:46 +11001370 struct mirror_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001371 int first = 0;
NeilBrown30194632011-12-23 10:17:56 +11001372 int last = conf->raid_disks - 1;
NeilBrown6b740b82012-03-19 12:46:39 +11001373 struct request_queue *q = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
NeilBrown53890422011-07-27 11:00:36 +10001375 if (mddev->recovery_disabled == conf->recovery_disabled)
1376 return -EBUSY;
1377
Neil Brown6c2fce22008-06-28 08:31:31 +10001378 if (rdev->raid_disk >= 0)
1379 first = last = rdev->raid_disk;
1380
NeilBrown6b740b82012-03-19 12:46:39 +11001381 if (q->merge_bvec_fn) {
1382 set_bit(Unmerged, &rdev->flags);
1383 mddev->merge_check_needed = 1;
1384 }
1385
NeilBrown7ef449d2011-12-23 10:17:57 +11001386 for (mirror = first; mirror <= last; mirror++) {
1387 p = conf->mirrors+mirror;
1388 if (!p->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10001390 disk_stack_limits(mddev->gendisk, rdev->bdev,
1391 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
1393 p->head_position = 0;
1394 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001395 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001396 /* As all devices are equivalent, we don't need a full recovery
1397 * if this was recently any drive of the array
1398 */
1399 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001400 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001401 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 break;
1403 }
NeilBrown7ef449d2011-12-23 10:17:57 +11001404 if (test_bit(WantReplacement, &p->rdev->flags) &&
1405 p[conf->raid_disks].rdev == NULL) {
1406 /* Add this device as a replacement */
1407 clear_bit(In_sync, &rdev->flags);
1408 set_bit(Replacement, &rdev->flags);
1409 rdev->raid_disk = mirror;
1410 err = 0;
1411 conf->fullsync = 1;
1412 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1413 break;
1414 }
1415 }
NeilBrown6b740b82012-03-19 12:46:39 +11001416 if (err == 0 && test_bit(Unmerged, &rdev->flags)) {
1417 /* Some requests might not have seen this new
1418 * merge_bvec_fn. We must wait for them to complete
1419 * before merging the device fully.
1420 * First we make sure any code which has tested
1421 * our function has submitted the request, then
1422 * we wait for all outstanding requests to complete.
1423 */
1424 synchronize_sched();
1425 raise_barrier(conf);
1426 lower_barrier(conf);
1427 clear_bit(Unmerged, &rdev->flags);
1428 }
Andre Nollac5e7112009-08-03 10:59:47 +10001429 md_integrity_add_rdev(rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001431 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432}
1433
NeilBrownb8321b62011-12-23 10:17:51 +11001434static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
NeilBrowne8096362011-10-11 16:49:05 +11001436 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001438 int number = rdev->raid_disk;
NeilBrown0f6d02d2011-10-11 16:48:46 +11001439 struct mirror_info *p = conf->mirrors+ number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
NeilBrownb014f142011-12-23 10:17:56 +11001441 if (rdev != p->rdev)
1442 p = conf->mirrors + conf->raid_disks + number;
1443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 print_conf(conf);
NeilBrownb8321b62011-12-23 10:17:51 +11001445 if (rdev == p->rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001446 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 atomic_read(&rdev->nr_pending)) {
1448 err = -EBUSY;
1449 goto abort;
1450 }
NeilBrown046abee2010-10-26 15:46:20 +11001451 /* Only remove non-faulty devices if recovery
NeilBrowndfc70642008-05-23 13:04:39 -07001452 * is not possible.
1453 */
1454 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown53890422011-07-27 11:00:36 +10001455 mddev->recovery_disabled != conf->recovery_disabled &&
NeilBrowndfc70642008-05-23 13:04:39 -07001456 mddev->degraded < conf->raid_disks) {
1457 err = -EBUSY;
1458 goto abort;
1459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001461 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 if (atomic_read(&rdev->nr_pending)) {
1463 /* lost the race, try later */
1464 err = -EBUSY;
1465 p->rdev = rdev;
Andre Nollac5e7112009-08-03 10:59:47 +10001466 goto abort;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001467 } else if (conf->mirrors[conf->raid_disks + number].rdev) {
1468 /* We just removed a device that is being replaced.
1469 * Move down the replacement. We drain all IO before
1470 * doing this to avoid confusion.
1471 */
1472 struct md_rdev *repl =
1473 conf->mirrors[conf->raid_disks + number].rdev;
1474 raise_barrier(conf);
1475 clear_bit(Replacement, &repl->flags);
1476 p->rdev = repl;
1477 conf->mirrors[conf->raid_disks + number].rdev = NULL;
1478 lower_barrier(conf);
NeilBrownb014f142011-12-23 10:17:56 +11001479 clear_bit(WantReplacement, &rdev->flags);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001480 } else
1481 clear_bit(WantReplacement, &rdev->flags);
Martin K. Petersena91a2782011-03-17 11:11:05 +01001482 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 }
1484abort:
1485
1486 print_conf(conf);
1487 return err;
1488}
1489
1490
NeilBrown6712ecf2007-09-27 12:47:43 +02001491static void end_sync_read(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001493 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
NeilBrown0fc280f2011-10-07 14:22:55 +11001495 update_head_pos(r1_bio->read_disk, r1_bio);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 /*
1498 * we have read a block, now it needs to be re-written,
1499 * or re-read if the read failed.
1500 * We don't do much here, just schedule handling by raid1d
1501 */
NeilBrown69382e82006-01-06 00:20:22 -08001502 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001504
1505 if (atomic_dec_and_test(&r1_bio->remaining))
1506 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
1508
NeilBrown6712ecf2007-09-27 12:47:43 +02001509static void end_sync_write(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510{
1511 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +11001512 struct r1bio *r1_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11001513 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001514 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 int mirror=0;
NeilBrown4367af52011-07-28 11:31:49 +10001516 sector_t first_bad;
1517 int bad_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001519 mirror = find_bio_disk(r1_bio, bio);
1520
NeilBrown6b1117d2006-03-31 02:31:57 -08001521 if (!uptodate) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001522 sector_t sync_blocks = 0;
NeilBrown6b1117d2006-03-31 02:31:57 -08001523 sector_t s = r1_bio->sector;
1524 long sectors_to_go = r1_bio->sectors;
1525 /* make sure these bits doesn't get cleared. */
1526 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001527 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001528 &sync_blocks, 1);
1529 s += sync_blocks;
1530 sectors_to_go -= sync_blocks;
1531 } while (sectors_to_go > 0);
NeilBrownd8f05d22011-07-28 11:33:00 +10001532 set_bit(WriteErrorSeen,
1533 &conf->mirrors[mirror].rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001534 if (!test_and_set_bit(WantReplacement,
1535 &conf->mirrors[mirror].rdev->flags))
1536 set_bit(MD_RECOVERY_NEEDED, &
1537 mddev->recovery);
NeilBrownd8f05d22011-07-28 11:33:00 +10001538 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +10001539 } else if (is_badblock(conf->mirrors[mirror].rdev,
1540 r1_bio->sector,
1541 r1_bio->sectors,
NeilBrown3a9f28a2011-07-28 11:33:42 +10001542 &first_bad, &bad_sectors) &&
1543 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1544 r1_bio->sector,
1545 r1_bio->sectors,
1546 &first_bad, &bad_sectors)
1547 )
NeilBrown4367af52011-07-28 11:31:49 +10001548 set_bit(R1BIO_MadeGood, &r1_bio->state);
NeilBrowne3b97032005-08-04 12:53:34 -07001549
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown4367af52011-07-28 11:31:49 +10001551 int s = r1_bio->sectors;
NeilBrownd8f05d22011-07-28 11:33:00 +10001552 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1553 test_bit(R1BIO_WriteError, &r1_bio->state))
NeilBrown4367af52011-07-28 11:31:49 +10001554 reschedule_retry(r1_bio);
1555 else {
1556 put_buf(r1_bio);
1557 md_done_sync(mddev, s, uptodate);
1558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560}
1561
NeilBrown3cb03002011-10-11 16:45:26 +11001562static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrownd8f05d22011-07-28 11:33:00 +10001563 int sectors, struct page *page, int rw)
1564{
1565 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
1566 /* success */
1567 return 1;
NeilBrown19d67162011-12-23 10:17:57 +11001568 if (rw == WRITE) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001569 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001570 if (!test_and_set_bit(WantReplacement,
1571 &rdev->flags))
1572 set_bit(MD_RECOVERY_NEEDED, &
1573 rdev->mddev->recovery);
1574 }
NeilBrownd8f05d22011-07-28 11:33:00 +10001575 /* need to record an error - either for the block or the device */
1576 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1577 md_error(rdev->mddev, rdev);
1578 return 0;
1579}
1580
NeilBrown9f2c9d12011-10-11 16:48:43 +11001581static int fix_sync_read_error(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001582{
1583 /* Try some synchronous reads of other devices to get
1584 * good data, much like with normal read errors. Only
1585 * read into the pages we already have so we don't
1586 * need to re-issue the read request.
1587 * We don't need to freeze the array, because being in an
1588 * active sync request, there is no normal IO, and
1589 * no overlapping syncs.
NeilBrown06f60382011-07-28 11:31:48 +10001590 * We don't need to check is_badblock() again as we
1591 * made sure that anything with a bad block in range
1592 * will have bi_end_io clear.
NeilBrowna68e5872011-05-11 14:40:44 +10001593 */
NeilBrownfd01b882011-10-11 16:47:53 +11001594 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001595 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001596 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1597 sector_t sect = r1_bio->sector;
1598 int sectors = r1_bio->sectors;
1599 int idx = 0;
1600
1601 while(sectors) {
1602 int s = sectors;
1603 int d = r1_bio->read_disk;
1604 int success = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11001605 struct md_rdev *rdev;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001606 int start;
NeilBrowna68e5872011-05-11 14:40:44 +10001607
1608 if (s > (PAGE_SIZE>>9))
1609 s = PAGE_SIZE >> 9;
1610 do {
1611 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1612 /* No rcu protection needed here devices
1613 * can only be removed when no resync is
1614 * active, and resync is currently active
1615 */
1616 rdev = conf->mirrors[d].rdev;
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001617 if (sync_page_io(rdev, sect, s<<9,
NeilBrowna68e5872011-05-11 14:40:44 +10001618 bio->bi_io_vec[idx].bv_page,
1619 READ, false)) {
1620 success = 1;
1621 break;
1622 }
1623 }
1624 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001625 if (d == conf->raid_disks * 2)
NeilBrowna68e5872011-05-11 14:40:44 +10001626 d = 0;
1627 } while (!success && d != r1_bio->read_disk);
1628
NeilBrown78d7f5f2011-05-11 14:48:56 +10001629 if (!success) {
NeilBrowna68e5872011-05-11 14:40:44 +10001630 char b[BDEVNAME_SIZE];
NeilBrown3a9f28a2011-07-28 11:33:42 +10001631 int abort = 0;
1632 /* Cannot read from anywhere, this block is lost.
1633 * Record a bad block on each device. If that doesn't
1634 * work just disable and interrupt the recovery.
1635 * Don't fail devices as that won't really help.
1636 */
NeilBrowna68e5872011-05-11 14:40:44 +10001637 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1638 " for block %llu\n",
1639 mdname(mddev),
1640 bdevname(bio->bi_bdev, b),
1641 (unsigned long long)r1_bio->sector);
NeilBrown8f19ccb2011-12-23 10:17:56 +11001642 for (d = 0; d < conf->raid_disks * 2; d++) {
NeilBrown3a9f28a2011-07-28 11:33:42 +10001643 rdev = conf->mirrors[d].rdev;
1644 if (!rdev || test_bit(Faulty, &rdev->flags))
1645 continue;
1646 if (!rdev_set_badblocks(rdev, sect, s, 0))
1647 abort = 1;
1648 }
1649 if (abort) {
NeilBrownd890fa22011-10-26 11:54:39 +11001650 conf->recovery_disabled =
1651 mddev->recovery_disabled;
NeilBrown3a9f28a2011-07-28 11:33:42 +10001652 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1653 md_done_sync(mddev, r1_bio->sectors, 0);
1654 put_buf(r1_bio);
1655 return 0;
1656 }
1657 /* Try next page */
1658 sectors -= s;
1659 sect += s;
1660 idx++;
1661 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001662 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001663
1664 start = d;
1665 /* write it back and re-read */
1666 while (d != r1_bio->read_disk) {
1667 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001668 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001669 d--;
1670 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1671 continue;
1672 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001673 if (r1_sync_page_io(rdev, sect, s,
1674 bio->bi_io_vec[idx].bv_page,
1675 WRITE) == 0) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001676 r1_bio->bios[d]->bi_end_io = NULL;
1677 rdev_dec_pending(rdev, mddev);
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001678 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001679 }
1680 d = start;
1681 while (d != r1_bio->read_disk) {
1682 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001683 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001684 d--;
1685 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1686 continue;
1687 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001688 if (r1_sync_page_io(rdev, sect, s,
1689 bio->bi_io_vec[idx].bv_page,
1690 READ) != 0)
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001691 atomic_add(s, &rdev->corrected_errors);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001692 }
NeilBrowna68e5872011-05-11 14:40:44 +10001693 sectors -= s;
1694 sect += s;
1695 idx ++;
1696 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001697 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrown7ca78d52011-05-11 14:50:37 +10001698 set_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrowna68e5872011-05-11 14:40:44 +10001699 return 1;
1700}
1701
NeilBrown9f2c9d12011-10-11 16:48:43 +11001702static int process_checks(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001703{
1704 /* We have read all readable devices. If we haven't
1705 * got the block, then there is no hope left.
1706 * If we have, then we want to do a comparison
1707 * and skip the write if everything is the same.
1708 * If any blocks failed to read, then we need to
1709 * attempt an over-write
1710 */
NeilBrownfd01b882011-10-11 16:47:53 +11001711 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001712 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001713 int primary;
1714 int i;
majianpengf4380a92012-04-12 16:04:47 +10001715 int vcnt;
NeilBrowna68e5872011-05-11 14:40:44 +10001716
NeilBrown8f19ccb2011-12-23 10:17:56 +11001717 for (primary = 0; primary < conf->raid_disks * 2; primary++)
NeilBrowna68e5872011-05-11 14:40:44 +10001718 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1719 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1720 r1_bio->bios[primary]->bi_end_io = NULL;
1721 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1722 break;
1723 }
1724 r1_bio->read_disk = primary;
majianpengf4380a92012-04-12 16:04:47 +10001725 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
NeilBrown8f19ccb2011-12-23 10:17:56 +11001726 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001727 int j;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001728 struct bio *pbio = r1_bio->bios[primary];
1729 struct bio *sbio = r1_bio->bios[i];
1730 int size;
NeilBrowna68e5872011-05-11 14:40:44 +10001731
NeilBrown78d7f5f2011-05-11 14:48:56 +10001732 if (r1_bio->bios[i]->bi_end_io != end_sync_read)
1733 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001734
NeilBrown78d7f5f2011-05-11 14:48:56 +10001735 if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) {
1736 for (j = vcnt; j-- ; ) {
1737 struct page *p, *s;
1738 p = pbio->bi_io_vec[j].bv_page;
1739 s = sbio->bi_io_vec[j].bv_page;
1740 if (memcmp(page_address(p),
1741 page_address(s),
NeilBrown5020ad72012-04-02 01:39:05 +10001742 sbio->bi_io_vec[j].bv_len))
NeilBrown78d7f5f2011-05-11 14:48:56 +10001743 break;
NeilBrowna68e5872011-05-11 14:40:44 +10001744 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001745 } else
1746 j = 0;
1747 if (j >= 0)
1748 mddev->resync_mismatches += r1_bio->sectors;
1749 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1750 && test_bit(BIO_UPTODATE, &sbio->bi_flags))) {
1751 /* No need to write to this device. */
1752 sbio->bi_end_io = NULL;
1753 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1754 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001755 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001756 /* fixup the bio for reuse */
1757 sbio->bi_vcnt = vcnt;
1758 sbio->bi_size = r1_bio->sectors << 9;
1759 sbio->bi_idx = 0;
1760 sbio->bi_phys_segments = 0;
1761 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1762 sbio->bi_flags |= 1 << BIO_UPTODATE;
1763 sbio->bi_next = NULL;
1764 sbio->bi_sector = r1_bio->sector +
1765 conf->mirrors[i].rdev->data_offset;
1766 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1767 size = sbio->bi_size;
1768 for (j = 0; j < vcnt ; j++) {
1769 struct bio_vec *bi;
1770 bi = &sbio->bi_io_vec[j];
1771 bi->bv_offset = 0;
1772 if (size > PAGE_SIZE)
1773 bi->bv_len = PAGE_SIZE;
1774 else
1775 bi->bv_len = size;
1776 size -= PAGE_SIZE;
1777 memcpy(page_address(bi->bv_page),
1778 page_address(pbio->bi_io_vec[j].bv_page),
1779 PAGE_SIZE);
1780 }
1781 }
NeilBrowna68e5872011-05-11 14:40:44 +10001782 return 0;
1783}
1784
NeilBrown9f2c9d12011-10-11 16:48:43 +11001785static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786{
NeilBrowne8096362011-10-11 16:49:05 +11001787 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 int i;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001789 int disks = conf->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 struct bio *bio, *wbio;
1791
1792 bio = r1_bio->bios[r1_bio->read_disk];
1793
NeilBrowna68e5872011-05-11 14:40:44 +10001794 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
1795 /* ouch - failed to read all of that. */
1796 if (!fix_sync_read_error(r1_bio))
1797 return;
NeilBrown7ca78d52011-05-11 14:50:37 +10001798
1799 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1800 if (process_checks(r1_bio) < 0)
1801 return;
NeilBrownd11c1712006-01-06 00:20:26 -08001802 /*
1803 * schedule writes
1804 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 atomic_set(&r1_bio->remaining, 1);
1806 for (i = 0; i < disks ; i++) {
1807 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08001808 if (wbio->bi_end_io == NULL ||
1809 (wbio->bi_end_io == end_sync_read &&
1810 (i == r1_bio->read_disk ||
1811 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 continue;
1813
NeilBrown3e198f72006-01-06 00:20:21 -08001814 wbio->bi_rw = WRITE;
1815 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 atomic_inc(&r1_bio->remaining);
1817 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
NeilBrown191ea9b2005-06-21 17:17:23 -07001818
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 generic_make_request(wbio);
1820 }
1821
1822 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001823 /* if we're here, all write(s) have completed, so clean up */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 md_done_sync(mddev, r1_bio->sectors, 1);
1825 put_buf(r1_bio);
1826 }
1827}
1828
1829/*
1830 * This is a kernel thread which:
1831 *
1832 * 1. Retries failed read operations on working mirrors.
1833 * 2. Updates the raid superblock when problems encounter.
NeilBrownd2eb35a2011-07-28 11:31:48 +10001834 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 */
1836
NeilBrowne8096362011-10-11 16:49:05 +11001837static void fix_read_error(struct r1conf *conf, int read_disk,
NeilBrown867868f2006-10-03 01:15:51 -07001838 sector_t sect, int sectors)
1839{
NeilBrownfd01b882011-10-11 16:47:53 +11001840 struct mddev *mddev = conf->mddev;
NeilBrown867868f2006-10-03 01:15:51 -07001841 while(sectors) {
1842 int s = sectors;
1843 int d = read_disk;
1844 int success = 0;
1845 int start;
NeilBrown3cb03002011-10-11 16:45:26 +11001846 struct md_rdev *rdev;
NeilBrown867868f2006-10-03 01:15:51 -07001847
1848 if (s > (PAGE_SIZE>>9))
1849 s = PAGE_SIZE >> 9;
1850
1851 do {
1852 /* Note: no rcu protection needed here
1853 * as this is synchronous in the raid1d thread
1854 * which is the thread that might remove
1855 * a device. If raid1d ever becomes multi-threaded....
1856 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001857 sector_t first_bad;
1858 int bad_sectors;
1859
NeilBrown867868f2006-10-03 01:15:51 -07001860 rdev = conf->mirrors[d].rdev;
1861 if (rdev &&
1862 test_bit(In_sync, &rdev->flags) &&
NeilBrownd2eb35a2011-07-28 11:31:48 +10001863 is_badblock(rdev, sect, s,
1864 &first_bad, &bad_sectors) == 0 &&
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11001865 sync_page_io(rdev, sect, s<<9,
1866 conf->tmppage, READ, false))
NeilBrown867868f2006-10-03 01:15:51 -07001867 success = 1;
1868 else {
1869 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001870 if (d == conf->raid_disks * 2)
NeilBrown867868f2006-10-03 01:15:51 -07001871 d = 0;
1872 }
1873 } while (!success && d != read_disk);
1874
1875 if (!success) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001876 /* Cannot read from anywhere - mark it bad */
NeilBrown3cb03002011-10-11 16:45:26 +11001877 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001878 if (!rdev_set_badblocks(rdev, sect, s, 0))
1879 md_error(mddev, rdev);
NeilBrown867868f2006-10-03 01:15:51 -07001880 break;
1881 }
1882 /* write it back and re-read */
1883 start = d;
1884 while (d != read_disk) {
1885 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001886 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07001887 d--;
1888 rdev = conf->mirrors[d].rdev;
1889 if (rdev &&
NeilBrownd8f05d22011-07-28 11:33:00 +10001890 test_bit(In_sync, &rdev->flags))
1891 r1_sync_page_io(rdev, sect, s,
1892 conf->tmppage, WRITE);
NeilBrown867868f2006-10-03 01:15:51 -07001893 }
1894 d = start;
1895 while (d != read_disk) {
1896 char b[BDEVNAME_SIZE];
1897 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001898 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07001899 d--;
1900 rdev = conf->mirrors[d].rdev;
1901 if (rdev &&
1902 test_bit(In_sync, &rdev->flags)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001903 if (r1_sync_page_io(rdev, sect, s,
1904 conf->tmppage, READ)) {
NeilBrown867868f2006-10-03 01:15:51 -07001905 atomic_add(s, &rdev->corrected_errors);
1906 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001907 "md/raid1:%s: read error corrected "
NeilBrown867868f2006-10-03 01:15:51 -07001908 "(%d sectors at %llu on %s)\n",
1909 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07001910 (unsigned long long)(sect +
1911 rdev->data_offset),
NeilBrown867868f2006-10-03 01:15:51 -07001912 bdevname(rdev->bdev, b));
1913 }
1914 }
1915 }
1916 sectors -= s;
1917 sect += s;
1918 }
1919}
1920
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001921static void bi_complete(struct bio *bio, int error)
1922{
1923 complete((struct completion *)bio->bi_private);
1924}
1925
1926static int submit_bio_wait(int rw, struct bio *bio)
1927{
1928 struct completion event;
1929 rw |= REQ_SYNC;
1930
1931 init_completion(&event);
1932 bio->bi_private = &event;
1933 bio->bi_end_io = bi_complete;
1934 submit_bio(rw, bio);
1935 wait_for_completion(&event);
1936
1937 return test_bit(BIO_UPTODATE, &bio->bi_flags);
1938}
1939
NeilBrown9f2c9d12011-10-11 16:48:43 +11001940static int narrow_write_error(struct r1bio *r1_bio, int i)
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001941{
NeilBrownfd01b882011-10-11 16:47:53 +11001942 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001943 struct r1conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11001944 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001945 int vcnt, idx;
1946 struct bio_vec *vec;
1947
1948 /* bio has the data to be written to device 'i' where
1949 * we just recently had a write error.
1950 * We repeatedly clone the bio and trim down to one block,
1951 * then try the write. Where the write fails we record
1952 * a bad block.
1953 * It is conceivable that the bio doesn't exactly align with
1954 * blocks. We must handle this somehow.
1955 *
1956 * We currently own a reference on the rdev.
1957 */
1958
1959 int block_sectors;
1960 sector_t sector;
1961 int sectors;
1962 int sect_to_write = r1_bio->sectors;
1963 int ok = 1;
1964
1965 if (rdev->badblocks.shift < 0)
1966 return 0;
1967
1968 block_sectors = 1 << rdev->badblocks.shift;
1969 sector = r1_bio->sector;
1970 sectors = ((sector + block_sectors)
1971 & ~(sector_t)(block_sectors - 1))
1972 - sector;
1973
1974 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
1975 vcnt = r1_bio->behind_page_count;
1976 vec = r1_bio->behind_bvecs;
1977 idx = 0;
1978 while (vec[idx].bv_page == NULL)
1979 idx++;
1980 } else {
1981 vcnt = r1_bio->master_bio->bi_vcnt;
1982 vec = r1_bio->master_bio->bi_io_vec;
1983 idx = r1_bio->master_bio->bi_idx;
1984 }
1985 while (sect_to_write) {
1986 struct bio *wbio;
1987 if (sectors > sect_to_write)
1988 sectors = sect_to_write;
1989 /* Write at 'sector' for 'sectors'*/
1990
1991 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
1992 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
1993 wbio->bi_sector = r1_bio->sector;
1994 wbio->bi_rw = WRITE;
1995 wbio->bi_vcnt = vcnt;
1996 wbio->bi_size = r1_bio->sectors << 9;
1997 wbio->bi_idx = idx;
1998
1999 md_trim_bio(wbio, sector - r1_bio->sector, sectors);
2000 wbio->bi_sector += rdev->data_offset;
2001 wbio->bi_bdev = rdev->bdev;
2002 if (submit_bio_wait(WRITE, wbio) == 0)
2003 /* failure! */
2004 ok = rdev_set_badblocks(rdev, sector,
2005 sectors, 0)
2006 && ok;
2007
2008 bio_put(wbio);
2009 sect_to_write -= sectors;
2010 sector += sectors;
2011 sectors = block_sectors;
2012 }
2013 return ok;
2014}
2015
NeilBrowne8096362011-10-11 16:49:05 +11002016static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002017{
2018 int m;
2019 int s = r1_bio->sectors;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002020 for (m = 0; m < conf->raid_disks * 2 ; m++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002021 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002022 struct bio *bio = r1_bio->bios[m];
2023 if (bio->bi_end_io == NULL)
2024 continue;
2025 if (test_bit(BIO_UPTODATE, &bio->bi_flags) &&
2026 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
2027 rdev_clear_badblocks(rdev, r1_bio->sector, s);
2028 }
2029 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
2030 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2031 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2032 md_error(conf->mddev, rdev);
2033 }
2034 }
2035 put_buf(r1_bio);
2036 md_done_sync(conf->mddev, s, 1);
2037}
2038
NeilBrowne8096362011-10-11 16:49:05 +11002039static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002040{
2041 int m;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002042 for (m = 0; m < conf->raid_disks * 2 ; m++)
NeilBrown62096bc2011-07-28 11:38:13 +10002043 if (r1_bio->bios[m] == IO_MADE_GOOD) {
NeilBrown3cb03002011-10-11 16:45:26 +11002044 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002045 rdev_clear_badblocks(rdev,
2046 r1_bio->sector,
2047 r1_bio->sectors);
2048 rdev_dec_pending(rdev, conf->mddev);
2049 } else if (r1_bio->bios[m] != NULL) {
2050 /* This drive got a write error. We need to
2051 * narrow down and record precise write
2052 * errors.
2053 */
2054 if (!narrow_write_error(r1_bio, m)) {
2055 md_error(conf->mddev,
2056 conf->mirrors[m].rdev);
2057 /* an I/O failed, we can't clear the bitmap */
2058 set_bit(R1BIO_Degraded, &r1_bio->state);
2059 }
2060 rdev_dec_pending(conf->mirrors[m].rdev,
2061 conf->mddev);
2062 }
2063 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2064 close_write(r1_bio);
2065 raid_end_bio_io(r1_bio);
2066}
2067
NeilBrowne8096362011-10-11 16:49:05 +11002068static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002069{
2070 int disk;
2071 int max_sectors;
NeilBrownfd01b882011-10-11 16:47:53 +11002072 struct mddev *mddev = conf->mddev;
NeilBrown62096bc2011-07-28 11:38:13 +10002073 struct bio *bio;
2074 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11002075 struct md_rdev *rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002076
2077 clear_bit(R1BIO_ReadError, &r1_bio->state);
2078 /* we got a read error. Maybe the drive is bad. Maybe just
2079 * the block and we can fix it.
2080 * We freeze all other IO, and try reading the block from
2081 * other devices. When we find one, we re-write
2082 * and check it that fixes the read error.
2083 * This is all done synchronously while the array is
2084 * frozen
2085 */
2086 if (mddev->ro == 0) {
2087 freeze_array(conf);
2088 fix_read_error(conf, r1_bio->read_disk,
2089 r1_bio->sector, r1_bio->sectors);
2090 unfreeze_array(conf);
2091 } else
2092 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
2093
2094 bio = r1_bio->bios[r1_bio->read_disk];
2095 bdevname(bio->bi_bdev, b);
2096read_more:
2097 disk = read_balance(conf, r1_bio, &max_sectors);
2098 if (disk == -1) {
2099 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
2100 " read error for block %llu\n",
2101 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2102 raid_end_bio_io(r1_bio);
2103 } else {
2104 const unsigned long do_sync
2105 = r1_bio->master_bio->bi_rw & REQ_SYNC;
2106 if (bio) {
2107 r1_bio->bios[r1_bio->read_disk] =
2108 mddev->ro ? IO_BLOCKED : NULL;
2109 bio_put(bio);
2110 }
2111 r1_bio->read_disk = disk;
2112 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2113 md_trim_bio(bio, r1_bio->sector - bio->bi_sector, max_sectors);
2114 r1_bio->bios[r1_bio->read_disk] = bio;
2115 rdev = conf->mirrors[disk].rdev;
2116 printk_ratelimited(KERN_ERR
2117 "md/raid1:%s: redirecting sector %llu"
2118 " to other mirror: %s\n",
2119 mdname(mddev),
2120 (unsigned long long)r1_bio->sector,
2121 bdevname(rdev->bdev, b));
2122 bio->bi_sector = r1_bio->sector + rdev->data_offset;
2123 bio->bi_bdev = rdev->bdev;
2124 bio->bi_end_io = raid1_end_read_request;
2125 bio->bi_rw = READ | do_sync;
2126 bio->bi_private = r1_bio;
2127 if (max_sectors < r1_bio->sectors) {
2128 /* Drat - have to split this up more */
2129 struct bio *mbio = r1_bio->master_bio;
2130 int sectors_handled = (r1_bio->sector + max_sectors
2131 - mbio->bi_sector);
2132 r1_bio->sectors = max_sectors;
2133 spin_lock_irq(&conf->device_lock);
2134 if (mbio->bi_phys_segments == 0)
2135 mbio->bi_phys_segments = 2;
2136 else
2137 mbio->bi_phys_segments++;
2138 spin_unlock_irq(&conf->device_lock);
2139 generic_make_request(bio);
2140 bio = NULL;
2141
2142 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2143
2144 r1_bio->master_bio = mbio;
2145 r1_bio->sectors = (mbio->bi_size >> 9)
2146 - sectors_handled;
2147 r1_bio->state = 0;
2148 set_bit(R1BIO_ReadError, &r1_bio->state);
2149 r1_bio->mddev = mddev;
2150 r1_bio->sector = mbio->bi_sector + sectors_handled;
2151
2152 goto read_more;
2153 } else
2154 generic_make_request(bio);
2155 }
2156}
2157
NeilBrownfd01b882011-10-11 16:47:53 +11002158static void raid1d(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159{
NeilBrown9f2c9d12011-10-11 16:48:43 +11002160 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 unsigned long flags;
NeilBrowne8096362011-10-11 16:49:05 +11002162 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002164 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165
2166 md_check_recovery(mddev);
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002167
2168 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002170
NeilBrownc3b328a2011-04-18 18:25:43 +10002171 if (atomic_read(&mddev->plug_cnt) == 0)
2172 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002173
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002175 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002176 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002178 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002179 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08002181 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 spin_unlock_irqrestore(&conf->device_lock, flags);
2183
2184 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002185 conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10002186 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002187 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002188 test_bit(R1BIO_WriteError, &r1_bio->state))
2189 handle_sync_write_finished(conf, r1_bio);
2190 else
NeilBrown4367af52011-07-28 11:31:49 +10002191 sync_request_write(mddev, r1_bio);
NeilBrowncd5ff9a2011-07-28 11:32:41 +10002192 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002193 test_bit(R1BIO_WriteError, &r1_bio->state))
2194 handle_write_finished(conf, r1_bio);
2195 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2196 handle_read_error(conf, r1_bio);
2197 else
NeilBrownd2eb35a2011-07-28 11:31:48 +10002198 /* just a partial read to be scheduled from separate
2199 * context
2200 */
2201 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
NeilBrown62096bc2011-07-28 11:38:13 +10002202
NeilBrown1d9d5242009-10-16 15:55:32 +11002203 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002204 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2205 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002207 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208}
2209
2210
NeilBrowne8096362011-10-11 16:49:05 +11002211static int init_resync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
2213 int buffs;
2214
2215 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002216 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2218 conf->poolinfo);
2219 if (!conf->r1buf_pool)
2220 return -ENOMEM;
2221 conf->next_resync = 0;
2222 return 0;
2223}
2224
2225/*
2226 * perform a "sync" on one "block"
2227 *
2228 * We need to make sure that no normal I/O request - particularly write
2229 * requests - conflict with active sync requests.
2230 *
2231 * This is achieved by tracking pending requests and a 'barrier' concept
2232 * that can be installed to exclude normal IO requests.
2233 */
2234
NeilBrownfd01b882011-10-11 16:47:53 +11002235static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236{
NeilBrowne8096362011-10-11 16:49:05 +11002237 struct r1conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002238 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 struct bio *bio;
2240 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08002241 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08002243 int wonly = -1;
2244 int write_targets = 0, read_targets = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11002245 sector_t sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07002246 int still_degraded = 0;
NeilBrown06f60382011-07-28 11:31:48 +10002247 int good_sectors = RESYNC_SECTORS;
2248 int min_bad = 0; /* number of sectors that are bad in all devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249
2250 if (!conf->r1buf_pool)
2251 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002252 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
Andre Noll58c0fed2009-03-31 14:33:13 +11002254 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002256 /* If we aborted, we need to abort the
2257 * sync on the 'current' bitmap chunk (there will
2258 * only be one in raid1 resync.
2259 * We can find the current addess in mddev->curr_resync
2260 */
NeilBrown6a806c52005-07-15 03:56:35 -07002261 if (mddev->curr_resync < max_sector) /* aborted */
2262 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07002263 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07002264 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07002265 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07002266
2267 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 close_sync(conf);
2269 return 0;
2270 }
2271
NeilBrown07d84d102006-06-26 00:27:56 -07002272 if (mddev->bitmap == NULL &&
2273 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07002274 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07002275 conf->fullsync == 0) {
2276 *skipped = 1;
2277 return max_sector - sector_nr;
2278 }
NeilBrown6394cca2006-08-27 01:23:50 -07002279 /* before building a request, check if we can skip these blocks..
2280 * This call the bitmap_start_sync doesn't actually record anything
2281 */
NeilBrowne3b97032005-08-04 12:53:34 -07002282 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de4852005-11-08 21:39:38 -08002283 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002284 /* We can skip this block, and probably several more */
2285 *skipped = 1;
2286 return sync_blocks;
2287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 /*
NeilBrown17999be2006-01-06 00:20:12 -08002289 * If there is non-resync activity waiting for a turn,
2290 * and resync is going fast enough,
2291 * then let it though before starting on this new sync request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 */
NeilBrown17999be2006-01-06 00:20:12 -08002293 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 msleep_interruptible(1000);
NeilBrown17999be2006-01-06 00:20:12 -08002295
NeilBrownb47490c2008-02-06 01:39:50 -08002296 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
NeilBrown1c4588e2010-10-26 17:41:22 +11002297 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown17999be2006-01-06 00:20:12 -08002298 raise_barrier(conf);
2299
2300 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
NeilBrown3e198f72006-01-06 00:20:21 -08002302 rcu_read_lock();
2303 /*
2304 * If we get a correctably read error during resync or recovery,
2305 * we might want to read from a different device. So we
2306 * flag all drives that could conceivably be read from for READ,
2307 * and any others (which will be non-In_sync devices) for WRITE.
2308 * If a read fails, we try reading from something else for which READ
2309 * is OK.
2310 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 r1_bio->mddev = mddev;
2313 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07002314 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316
NeilBrown8f19ccb2011-12-23 10:17:56 +11002317 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002318 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 bio = r1_bio->bios[i];
2320
2321 /* take from bio_init */
2322 bio->bi_next = NULL;
NeilBrowndb8d9d32010-10-07 12:00:50 +11002323 bio->bi_flags &= ~(BIO_POOL_MASK-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 bio->bi_flags |= 1 << BIO_UPTODATE;
NeilBrown802ba062006-12-13 00:34:13 -08002325 bio->bi_rw = READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 bio->bi_vcnt = 0;
2327 bio->bi_idx = 0;
2328 bio->bi_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 bio->bi_size = 0;
2330 bio->bi_end_io = NULL;
2331 bio->bi_private = NULL;
2332
NeilBrown3e198f72006-01-06 00:20:21 -08002333 rdev = rcu_dereference(conf->mirrors[i].rdev);
2334 if (rdev == NULL ||
NeilBrown06f60382011-07-28 11:31:48 +10002335 test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11002336 if (i < conf->raid_disks)
2337 still_degraded = 1;
NeilBrown3e198f72006-01-06 00:20:21 -08002338 } else if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 bio->bi_rw = WRITE;
2340 bio->bi_end_io = end_sync_write;
2341 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08002342 } else {
2343 /* may need to read from here */
NeilBrown06f60382011-07-28 11:31:48 +10002344 sector_t first_bad = MaxSector;
2345 int bad_sectors;
2346
2347 if (is_badblock(rdev, sector_nr, good_sectors,
2348 &first_bad, &bad_sectors)) {
2349 if (first_bad > sector_nr)
2350 good_sectors = first_bad - sector_nr;
2351 else {
2352 bad_sectors -= (sector_nr - first_bad);
2353 if (min_bad == 0 ||
2354 min_bad > bad_sectors)
2355 min_bad = bad_sectors;
2356 }
NeilBrown3e198f72006-01-06 00:20:21 -08002357 }
NeilBrown06f60382011-07-28 11:31:48 +10002358 if (sector_nr < first_bad) {
2359 if (test_bit(WriteMostly, &rdev->flags)) {
2360 if (wonly < 0)
2361 wonly = i;
2362 } else {
2363 if (disk < 0)
2364 disk = i;
2365 }
2366 bio->bi_rw = READ;
2367 bio->bi_end_io = end_sync_read;
2368 read_targets++;
2369 }
NeilBrown3e198f72006-01-06 00:20:21 -08002370 }
NeilBrown06f60382011-07-28 11:31:48 +10002371 if (bio->bi_end_io) {
2372 atomic_inc(&rdev->nr_pending);
2373 bio->bi_sector = sector_nr + rdev->data_offset;
2374 bio->bi_bdev = rdev->bdev;
2375 bio->bi_private = r1_bio;
2376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 }
NeilBrown3e198f72006-01-06 00:20:21 -08002378 rcu_read_unlock();
2379 if (disk < 0)
2380 disk = wonly;
2381 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07002382
NeilBrown06f60382011-07-28 11:31:48 +10002383 if (read_targets == 0 && min_bad > 0) {
2384 /* These sectors are bad on all InSync devices, so we
2385 * need to mark them bad on all write targets
2386 */
2387 int ok = 1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002388 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
NeilBrown06f60382011-07-28 11:31:48 +10002389 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
majianpenga42f9d82012-04-02 01:04:19 +10002390 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown06f60382011-07-28 11:31:48 +10002391 ok = rdev_set_badblocks(rdev, sector_nr,
2392 min_bad, 0
2393 ) && ok;
2394 }
2395 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2396 *skipped = 1;
2397 put_buf(r1_bio);
2398
2399 if (!ok) {
2400 /* Cannot record the badblocks, so need to
2401 * abort the resync.
2402 * If there are multiple read targets, could just
2403 * fail the really bad ones ???
2404 */
2405 conf->recovery_disabled = mddev->recovery_disabled;
2406 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2407 return 0;
2408 } else
2409 return min_bad;
2410
2411 }
2412 if (min_bad > 0 && min_bad < good_sectors) {
2413 /* only resync enough to reach the next bad->good
2414 * transition */
2415 good_sectors = min_bad;
2416 }
2417
NeilBrown3e198f72006-01-06 00:20:21 -08002418 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2419 /* extra read targets are also write targets */
2420 write_targets += read_targets-1;
2421
2422 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 /* There is nowhere to write, so all non-sync
2424 * drives must be failed - so we are finished
2425 */
NeilBrown57afd892005-06-21 17:17:13 -07002426 sector_t rv = max_sector - sector_nr;
2427 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 return rv;
2430 }
2431
NeilBrownc6207272008-02-06 01:39:52 -08002432 if (max_sector > mddev->resync_max)
2433 max_sector = mddev->resync_max; /* Don't do IO beyond here */
NeilBrown06f60382011-07-28 11:31:48 +10002434 if (max_sector > sector_nr + good_sectors)
2435 max_sector = sector_nr + good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07002437 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 do {
2439 struct page *page;
2440 int len = PAGE_SIZE;
2441 if (sector_nr + (len>>9) > max_sector)
2442 len = (max_sector - sector_nr) << 9;
2443 if (len == 0)
2444 break;
NeilBrown6a806c52005-07-15 03:56:35 -07002445 if (sync_blocks == 0) {
2446 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de4852005-11-08 21:39:38 -08002447 &sync_blocks, still_degraded) &&
2448 !conf->fullsync &&
2449 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07002450 break;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002451 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
NeilBrown7571ae82010-10-07 11:54:46 +11002452 if ((len >> 9) > sync_blocks)
NeilBrown6a806c52005-07-15 03:56:35 -07002453 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07002454 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002455
NeilBrown8f19ccb2011-12-23 10:17:56 +11002456 for (i = 0 ; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 bio = r1_bio->bios[i];
2458 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08002459 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 if (bio_add_page(bio, page, len, 0) == 0) {
2461 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08002462 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 while (i > 0) {
2464 i--;
2465 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07002466 if (bio->bi_end_io==NULL)
2467 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 /* remove last page from this bio */
2469 bio->bi_vcnt--;
2470 bio->bi_size -= len;
2471 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
2472 }
2473 goto bio_full;
2474 }
2475 }
2476 }
2477 nr_sectors += len>>9;
2478 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07002479 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2481 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 r1_bio->sectors = nr_sectors;
2483
NeilBrownd11c1712006-01-06 00:20:26 -08002484 /* For a user-requested sync, we read all readable devices and do a
2485 * compare
2486 */
2487 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2488 atomic_set(&r1_bio->remaining, read_targets);
NeilBrown8f19ccb2011-12-23 10:17:56 +11002489 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrownd11c1712006-01-06 00:20:26 -08002490 bio = r1_bio->bios[i];
2491 if (bio->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07002492 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002493 generic_make_request(bio);
2494 }
2495 }
2496 } else {
2497 atomic_set(&r1_bio->remaining, 1);
2498 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07002499 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002500 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501
NeilBrownd11c1712006-01-06 00:20:26 -08002502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 return nr_sectors;
2504}
2505
NeilBrownfd01b882011-10-11 16:47:53 +11002506static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07002507{
2508 if (sectors)
2509 return sectors;
2510
2511 return mddev->dev_sectors;
2512}
2513
NeilBrowne8096362011-10-11 16:49:05 +11002514static struct r1conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515{
NeilBrowne8096362011-10-11 16:49:05 +11002516 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002517 int i;
NeilBrown0f6d02d2011-10-11 16:48:46 +11002518 struct mirror_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11002519 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002520 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
NeilBrowne8096362011-10-11 16:49:05 +11002522 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11002524 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
NeilBrown8f19ccb2011-12-23 10:17:56 +11002526 conf->mirrors = kzalloc(sizeof(struct mirror_info)
2527 * mddev->raid_disks * 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 GFP_KERNEL);
2529 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11002530 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
NeilBrownddaf22a2006-01-06 00:20:19 -08002532 conf->tmppage = alloc_page(GFP_KERNEL);
2533 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11002534 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08002535
NeilBrown709ae482009-12-14 12:49:51 +11002536 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11002538 goto abort;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002539 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2541 r1bio_pool_free,
2542 conf->poolinfo);
2543 if (!conf->r1bio_pool)
NeilBrown709ae482009-12-14 12:49:51 +11002544 goto abort;
2545
NeilBrowned9bfdf2009-10-16 15:55:44 +11002546 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
NeilBrownc19d5792011-12-23 10:17:57 +11002548 err = -EINVAL;
Neil Browne7e72bf2008-05-14 16:05:54 -07002549 spin_lock_init(&conf->device_lock);
NeilBrowndafb20f2012-03-19 12:46:39 +11002550 rdev_for_each(rdev, mddev) {
NeilBrown709ae482009-12-14 12:49:51 +11002551 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 if (disk_idx >= mddev->raid_disks
2553 || disk_idx < 0)
2554 continue;
NeilBrownc19d5792011-12-23 10:17:57 +11002555 if (test_bit(Replacement, &rdev->flags))
2556 disk = conf->mirrors + conf->raid_disks + disk_idx;
2557 else
2558 disk = conf->mirrors + disk_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
NeilBrownc19d5792011-12-23 10:17:57 +11002560 if (disk->rdev)
2561 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 disk->rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
2564 disk->head_position = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 }
2566 conf->raid_disks = mddev->raid_disks;
2567 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 INIT_LIST_HEAD(&conf->retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569
2570 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08002571 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572
NeilBrown191ea9b2005-06-21 17:17:23 -07002573 bio_list_init(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +11002574 conf->pending_count = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11002575 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown191ea9b2005-06-21 17:17:23 -07002576
NeilBrownc19d5792011-12-23 10:17:57 +11002577 err = -EIO;
NeilBrown709ae482009-12-14 12:49:51 +11002578 conf->last_used = -1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002579 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
2581 disk = conf->mirrors + i;
2582
NeilBrownc19d5792011-12-23 10:17:57 +11002583 if (i < conf->raid_disks &&
2584 disk[conf->raid_disks].rdev) {
2585 /* This slot has a replacement. */
2586 if (!disk->rdev) {
2587 /* No original, just make the replacement
2588 * a recovering spare
2589 */
2590 disk->rdev =
2591 disk[conf->raid_disks].rdev;
2592 disk[conf->raid_disks].rdev = NULL;
2593 } else if (!test_bit(In_sync, &disk->rdev->flags))
2594 /* Original is not in_sync - bad */
2595 goto abort;
2596 }
2597
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002598 if (!disk->rdev ||
2599 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 disk->head_position = 0;
NeilBrown918f0232007-08-22 14:01:52 -07002601 if (disk->rdev)
2602 conf->fullsync = 1;
NeilBrown709ae482009-12-14 12:49:51 +11002603 } else if (conf->last_used < 0)
2604 /*
2605 * The first working device is used as a
2606 * starting point to read balancing.
2607 */
2608 conf->last_used = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 }
NeilBrown709ae482009-12-14 12:49:51 +11002610
NeilBrown709ae482009-12-14 12:49:51 +11002611 if (conf->last_used < 0) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002612 printk(KERN_ERR "md/raid1:%s: no operational mirrors\n",
NeilBrown709ae482009-12-14 12:49:51 +11002613 mdname(mddev));
2614 goto abort;
NeilBrown11ce99e2006-10-03 01:15:52 -07002615 }
NeilBrown709ae482009-12-14 12:49:51 +11002616 err = -ENOMEM;
2617 conf->thread = md_register_thread(raid1d, mddev, NULL);
2618 if (!conf->thread) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002619 printk(KERN_ERR
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002620 "md/raid1:%s: couldn't allocate thread\n",
NeilBrown191ea9b2005-06-21 17:17:23 -07002621 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11002622 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002624
NeilBrown709ae482009-12-14 12:49:51 +11002625 return conf;
2626
2627 abort:
2628 if (conf) {
2629 if (conf->r1bio_pool)
2630 mempool_destroy(conf->r1bio_pool);
2631 kfree(conf->mirrors);
2632 safe_put_page(conf->tmppage);
2633 kfree(conf->poolinfo);
2634 kfree(conf);
2635 }
2636 return ERR_PTR(err);
2637}
2638
majianpeng5220ea12012-04-02 09:48:38 +10002639static int stop(struct mddev *mddev);
NeilBrownfd01b882011-10-11 16:47:53 +11002640static int run(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002641{
NeilBrowne8096362011-10-11 16:49:05 +11002642 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002643 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11002644 struct md_rdev *rdev;
majianpeng5220ea12012-04-02 09:48:38 +10002645 int ret;
NeilBrown709ae482009-12-14 12:49:51 +11002646
2647 if (mddev->level != 1) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002648 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
NeilBrown709ae482009-12-14 12:49:51 +11002649 mdname(mddev), mddev->level);
2650 return -EIO;
2651 }
2652 if (mddev->reshape_position != MaxSector) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002653 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
NeilBrown709ae482009-12-14 12:49:51 +11002654 mdname(mddev));
2655 return -EIO;
2656 }
2657 /*
2658 * copy the already verified devices into our private RAID1
2659 * bookkeeping area. [whatever we allocate in run(),
2660 * should be freed in stop()]
2661 */
2662 if (mddev->private == NULL)
2663 conf = setup_conf(mddev);
2664 else
2665 conf = mddev->private;
2666
2667 if (IS_ERR(conf))
2668 return PTR_ERR(conf);
2669
NeilBrowndafb20f2012-03-19 12:46:39 +11002670 rdev_for_each(rdev, mddev) {
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002671 if (!mddev->gendisk)
2672 continue;
NeilBrown709ae482009-12-14 12:49:51 +11002673 disk_stack_limits(mddev->gendisk, rdev->bdev,
2674 rdev->data_offset << 9);
NeilBrown709ae482009-12-14 12:49:51 +11002675 }
2676
2677 mddev->degraded = 0;
2678 for (i=0; i < conf->raid_disks; i++)
2679 if (conf->mirrors[i].rdev == NULL ||
2680 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2681 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2682 mddev->degraded++;
2683
2684 if (conf->raid_disks - mddev->degraded == 1)
2685 mddev->recovery_cp = MaxSector;
2686
Andre Noll8c6ac862009-06-18 08:48:06 +10002687 if (mddev->recovery_cp != MaxSector)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002688 printk(KERN_NOTICE "md/raid1:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10002689 " -- starting background reconstruction\n",
2690 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002692 "md/raid1:%s: active with %d out of %d mirrors\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 mdname(mddev), mddev->raid_disks - mddev->degraded,
2694 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11002695
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 /*
2697 * Ok, everything is just fine now
2698 */
NeilBrown709ae482009-12-14 12:49:51 +11002699 mddev->thread = conf->thread;
2700 conf->thread = NULL;
2701 mddev->private = conf;
2702
Dan Williams1f403622009-03-31 14:59:03 +11002703 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002705 if (mddev->queue) {
2706 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2707 mddev->queue->backing_dev_info.congested_data = mddev;
NeilBrown6b740b82012-03-19 12:46:39 +11002708 blk_queue_merge_bvec(mddev->queue, raid1_mergeable_bvec);
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002709 }
majianpeng5220ea12012-04-02 09:48:38 +10002710
2711 ret = md_integrity_register(mddev);
2712 if (ret)
2713 stop(mddev);
2714 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715}
2716
NeilBrownfd01b882011-10-11 16:47:53 +11002717static int stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718{
NeilBrowne8096362011-10-11 16:49:05 +11002719 struct r1conf *conf = mddev->private;
NeilBrown4b6d2872005-09-09 16:23:47 -07002720 struct bitmap *bitmap = mddev->bitmap;
NeilBrown4b6d2872005-09-09 16:23:47 -07002721
2722 /* wait for behind writes to complete */
NeilBrowne5551902010-03-31 11:21:44 +11002723 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002724 printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n",
2725 mdname(mddev));
NeilBrown4b6d2872005-09-09 16:23:47 -07002726 /* need to kick something here to make sure I/O goes? */
NeilBrowne5551902010-03-31 11:21:44 +11002727 wait_event(bitmap->behind_wait,
2728 atomic_read(&bitmap->behind_writes) == 0);
NeilBrown4b6d2872005-09-09 16:23:47 -07002729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730
NeilBrown409c57f2009-03-31 14:39:39 +11002731 raise_barrier(conf);
2732 lower_barrier(conf);
2733
NeilBrown01f96c02011-09-21 15:30:20 +10002734 md_unregister_thread(&mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 if (conf->r1bio_pool)
2736 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002737 kfree(conf->mirrors);
2738 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 kfree(conf);
2740 mddev->private = NULL;
2741 return 0;
2742}
2743
NeilBrownfd01b882011-10-11 16:47:53 +11002744static int raid1_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745{
2746 /* no resync is happening, and there is enough space
2747 * on all devices, so we can resize.
2748 * We need to make sure resync covers any new space.
2749 * If the array is shrinking we should possibly wait until
2750 * any io in the removed space completes, but it hardly seems
2751 * worth it.
2752 */
Dan Williams1f403622009-03-31 14:59:03 +11002753 md_set_array_sectors(mddev, raid1_size(mddev, sectors, 0));
Dan Williamsb522adc2009-03-31 15:00:31 +11002754 if (mddev->array_sectors > raid1_size(mddev, sectors, 0))
2755 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10002756 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10002757 revalidate_disk(mddev->gendisk);
Dan Williamsb522adc2009-03-31 15:00:31 +11002758 if (sectors > mddev->dev_sectors &&
NeilBrownb0986362011-05-11 15:52:21 +10002759 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11002760 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2762 }
Dan Williamsb522adc2009-03-31 15:00:31 +11002763 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07002764 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 return 0;
2766}
2767
NeilBrownfd01b882011-10-11 16:47:53 +11002768static int raid1_reshape(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769{
2770 /* We need to:
2771 * 1/ resize the r1bio_pool
2772 * 2/ resize conf->mirrors
2773 *
2774 * We allocate a new r1bio_pool if we can.
2775 * Then raise a device barrier and wait until all IO stops.
2776 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07002777 *
2778 * At the same time, we "pack" the devices so that all the missing
2779 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 */
2781 mempool_t *newpool, *oldpool;
2782 struct pool_info *newpoolinfo;
NeilBrown0f6d02d2011-10-11 16:48:46 +11002783 struct mirror_info *newmirrors;
NeilBrowne8096362011-10-11 16:49:05 +11002784 struct r1conf *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08002785 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07002786 unsigned long flags;
Dan Williamsb5470dc2008-06-27 21:44:04 -07002787 int d, d2, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
NeilBrown63c70c42006-03-27 01:18:13 -08002789 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10002790 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08002791 mddev->layout != mddev->new_layout ||
2792 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10002793 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08002794 mddev->new_layout = mddev->layout;
2795 mddev->new_level = mddev->level;
2796 return -EINVAL;
2797 }
2798
Dan Williamsb5470dc2008-06-27 21:44:04 -07002799 err = md_allow_write(mddev);
2800 if (err)
2801 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08002802
NeilBrown63c70c42006-03-27 01:18:13 -08002803 raid_disks = mddev->raid_disks + mddev->delta_disks;
2804
NeilBrown6ea9c072005-06-21 17:17:09 -07002805 if (raid_disks < conf->raid_disks) {
2806 cnt=0;
2807 for (d= 0; d < conf->raid_disks; d++)
2808 if (conf->mirrors[d].rdev)
2809 cnt++;
2810 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07002812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813
2814 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2815 if (!newpoolinfo)
2816 return -ENOMEM;
2817 newpoolinfo->mddev = mddev;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002818 newpoolinfo->raid_disks = raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819
2820 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2821 r1bio_pool_free, newpoolinfo);
2822 if (!newpool) {
2823 kfree(newpoolinfo);
2824 return -ENOMEM;
2825 }
NeilBrown8f19ccb2011-12-23 10:17:56 +11002826 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks * 2,
2827 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 if (!newmirrors) {
2829 kfree(newpoolinfo);
2830 mempool_destroy(newpool);
2831 return -ENOMEM;
2832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833
NeilBrown17999be2006-01-06 00:20:12 -08002834 raise_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835
2836 /* ok, everything is stopped */
2837 oldpool = conf->r1bio_pool;
2838 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07002839
NeilBrowna88aa782007-08-22 14:01:53 -07002840 for (d = d2 = 0; d < conf->raid_disks; d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002841 struct md_rdev *rdev = conf->mirrors[d].rdev;
NeilBrowna88aa782007-08-22 14:01:53 -07002842 if (rdev && rdev->raid_disk != d2) {
Namhyung Kim36fad852011-07-27 11:00:36 +10002843 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna88aa782007-08-22 14:01:53 -07002844 rdev->raid_disk = d2;
Namhyung Kim36fad852011-07-27 11:00:36 +10002845 sysfs_unlink_rdev(mddev, rdev);
2846 if (sysfs_link_rdev(mddev, rdev))
NeilBrowna88aa782007-08-22 14:01:53 -07002847 printk(KERN_WARNING
Namhyung Kim36fad852011-07-27 11:00:36 +10002848 "md/raid1:%s: cannot register rd%d\n",
2849 mdname(mddev), rdev->raid_disk);
NeilBrown6ea9c072005-06-21 17:17:09 -07002850 }
NeilBrowna88aa782007-08-22 14:01:53 -07002851 if (rdev)
2852 newmirrors[d2++].rdev = rdev;
2853 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 kfree(conf->mirrors);
2855 conf->mirrors = newmirrors;
2856 kfree(conf->poolinfo);
2857 conf->poolinfo = newpoolinfo;
2858
NeilBrownc04be0a2006-10-03 01:15:53 -07002859 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07002861 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08002863 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
NeilBrown6ea9c072005-06-21 17:17:09 -07002865 conf->last_used = 0; /* just make sure it is in-range */
NeilBrown17999be2006-01-06 00:20:12 -08002866 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
2868 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2869 md_wakeup_thread(mddev->thread);
2870
2871 mempool_destroy(oldpool);
2872 return 0;
2873}
2874
NeilBrownfd01b882011-10-11 16:47:53 +11002875static void raid1_quiesce(struct mddev *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07002876{
NeilBrowne8096362011-10-11 16:49:05 +11002877 struct r1conf *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07002878
2879 switch(state) {
NeilBrown6eef4b22009-12-14 12:49:51 +11002880 case 2: /* wake for suspend */
2881 wake_up(&conf->wait_barrier);
2882 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002883 case 1:
NeilBrown17999be2006-01-06 00:20:12 -08002884 raise_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002885 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002886 case 0:
NeilBrown17999be2006-01-06 00:20:12 -08002887 lower_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002888 break;
2889 }
NeilBrown36fa3062005-09-09 16:23:45 -07002890}
2891
NeilBrownfd01b882011-10-11 16:47:53 +11002892static void *raid1_takeover(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002893{
2894 /* raid1 can take over:
2895 * raid5 with 2 devices, any layout or chunk size
2896 */
2897 if (mddev->level == 5 && mddev->raid_disks == 2) {
NeilBrowne8096362011-10-11 16:49:05 +11002898 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002899 mddev->new_level = 1;
2900 mddev->new_layout = 0;
2901 mddev->new_chunk_sectors = 0;
2902 conf = setup_conf(mddev);
2903 if (!IS_ERR(conf))
2904 conf->barrier = 1;
2905 return conf;
2906 }
2907 return ERR_PTR(-EINVAL);
2908}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909
NeilBrown84fc4b52011-10-11 16:49:58 +11002910static struct md_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911{
2912 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08002913 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 .owner = THIS_MODULE,
2915 .make_request = make_request,
2916 .run = run,
2917 .stop = stop,
2918 .status = status,
2919 .error_handler = error,
2920 .hot_add_disk = raid1_add_disk,
2921 .hot_remove_disk= raid1_remove_disk,
2922 .spare_active = raid1_spare_active,
2923 .sync_request = sync_request,
2924 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07002925 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08002926 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07002927 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11002928 .takeover = raid1_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929};
2930
2931static int __init raid_init(void)
2932{
NeilBrown2604b702006-01-06 00:20:36 -08002933 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934}
2935
2936static void raid_exit(void)
2937{
NeilBrown2604b702006-01-06 00:20:36 -08002938 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939}
2940
2941module_init(raid_init);
2942module_exit(raid_exit);
2943MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11002944MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08002946MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08002947MODULE_ALIAS("md-level-1");
NeilBrown34db0cd2011-10-11 16:50:01 +11002948
2949module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);