blob: a368db2431a596020a98a3e64f67685ccef9c6cc [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
NeilBrown76073052011-05-11 14:34:56 +1000526 || test_bit(Faulty, &rdev->flags))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000527 continue;
NeilBrown76073052011-05-11 14:34:56 +1000528 if (!test_bit(In_sync, &rdev->flags) &&
529 rdev->recovery_offset < this_sector + sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 continue;
NeilBrown76073052011-05-11 14:34:56 +1000531 if (test_bit(WriteMostly, &rdev->flags)) {
532 /* Don't balance among write-mostly, just
533 * use the first as a last resort */
NeilBrown307729c2012-01-09 01:41:51 +1100534 if (best_disk < 0) {
535 if (is_badblock(rdev, this_sector, sectors,
536 &first_bad, &bad_sectors)) {
537 if (first_bad < this_sector)
538 /* Cannot use this */
539 continue;
540 best_good_sectors = first_bad - this_sector;
541 } else
542 best_good_sectors = sectors;
NeilBrown76073052011-05-11 14:34:56 +1000543 best_disk = disk;
NeilBrown307729c2012-01-09 01:41:51 +1100544 }
NeilBrown76073052011-05-11 14:34:56 +1000545 continue;
546 }
547 /* This is a reasonable device to use. It might
548 * even be best.
549 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000550 if (is_badblock(rdev, this_sector, sectors,
551 &first_bad, &bad_sectors)) {
552 if (best_dist < MaxSector)
553 /* already have a better device */
554 continue;
555 if (first_bad <= this_sector) {
556 /* cannot read here. If this is the 'primary'
557 * device, then we must not read beyond
558 * bad_sectors from another device..
559 */
560 bad_sectors -= (this_sector - first_bad);
561 if (choose_first && sectors > bad_sectors)
562 sectors = bad_sectors;
563 if (best_good_sectors > sectors)
564 best_good_sectors = sectors;
565
566 } else {
567 sector_t good_sectors = first_bad - this_sector;
568 if (good_sectors > best_good_sectors) {
569 best_good_sectors = good_sectors;
570 best_disk = disk;
571 }
572 if (choose_first)
573 break;
574 }
575 continue;
576 } else
577 best_good_sectors = sectors;
578
NeilBrown76073052011-05-11 14:34:56 +1000579 dist = abs(this_sector - conf->mirrors[disk].head_position);
580 if (choose_first
581 /* Don't change to another disk for sequential reads */
582 || conf->next_seq_sect == this_sector
583 || dist == 0
584 /* If device is idle, use it */
585 || atomic_read(&rdev->nr_pending) == 0) {
586 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 break;
588 }
NeilBrown76073052011-05-11 14:34:56 +1000589 if (dist < best_dist) {
590 best_dist = dist;
591 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000593 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
NeilBrown76073052011-05-11 14:34:56 +1000595 if (best_disk >= 0) {
596 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700597 if (!rdev)
598 goto retry;
599 atomic_inc(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000600 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 /* cannot risk returning a device that failed
602 * before we inc'ed nr_pending
603 */
NeilBrown03c902e2006-01-06 00:20:46 -0800604 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 goto retry;
606 }
NeilBrownd2eb35a2011-07-28 11:31:48 +1000607 sectors = best_good_sectors;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700608 conf->next_seq_sect = this_sector + sectors;
NeilBrown76073052011-05-11 14:34:56 +1000609 conf->last_used = best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 }
611 rcu_read_unlock();
NeilBrownd2eb35a2011-07-28 11:31:48 +1000612 *max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
NeilBrown76073052011-05-11 14:34:56 +1000614 return best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615}
616
NeilBrownfd01b882011-10-11 16:47:53 +1100617int md_raid1_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700618{
NeilBrowne8096362011-10-11 16:49:05 +1100619 struct r1conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700620 int i, ret = 0;
621
NeilBrown34db0cd2011-10-11 16:50:01 +1100622 if ((bits & (1 << BDI_async_congested)) &&
623 conf->pending_count >= max_queued_requests)
624 return 1;
625
NeilBrown0d129222006-10-03 01:15:54 -0700626 rcu_read_lock();
NeilBrown30194632011-12-23 10:17:56 +1100627 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100628 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700629 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200630 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700631
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500632 BUG_ON(!q);
633
NeilBrown0d129222006-10-03 01:15:54 -0700634 /* Note the '|| 1' - when read_balance prefers
635 * non-congested targets, it can be removed
636 */
Alexander Beregalov91a9e992009-04-07 01:35:56 +0400637 if ((bits & (1<<BDI_async_congested)) || 1)
NeilBrown0d129222006-10-03 01:15:54 -0700638 ret |= bdi_congested(&q->backing_dev_info, bits);
639 else
640 ret &= bdi_congested(&q->backing_dev_info, bits);
641 }
642 }
643 rcu_read_unlock();
644 return ret;
645}
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500646EXPORT_SYMBOL_GPL(md_raid1_congested);
NeilBrown0d129222006-10-03 01:15:54 -0700647
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500648static int raid1_congested(void *data, int bits)
649{
NeilBrownfd01b882011-10-11 16:47:53 +1100650 struct mddev *mddev = data;
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500651
652 return mddev_congested(mddev, bits) ||
653 md_raid1_congested(mddev, bits);
654}
NeilBrown0d129222006-10-03 01:15:54 -0700655
NeilBrowne8096362011-10-11 16:49:05 +1100656static void flush_pending_writes(struct r1conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800657{
658 /* Any writes that have been queued but are awaiting
659 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800660 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800661 spin_lock_irq(&conf->device_lock);
662
663 if (conf->pending_bio_list.head) {
664 struct bio *bio;
665 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100666 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800667 spin_unlock_irq(&conf->device_lock);
668 /* flush any pending bitmap writes to
669 * disk before proceeding w/ I/O */
670 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100671 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800672
673 while (bio) { /* submit pending writes */
674 struct bio *next = bio->bi_next;
675 bio->bi_next = NULL;
676 generic_make_request(bio);
677 bio = next;
678 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800679 } else
680 spin_unlock_irq(&conf->device_lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100681}
682
NeilBrown17999be2006-01-06 00:20:12 -0800683/* Barriers....
684 * Sometimes we need to suspend IO while we do something else,
685 * either some resync/recovery, or reconfigure the array.
686 * To do this we raise a 'barrier'.
687 * The 'barrier' is a counter that can be raised multiple times
688 * to count how many activities are happening which preclude
689 * normal IO.
690 * We can only raise the barrier if there is no pending IO.
691 * i.e. if nr_pending == 0.
692 * We choose only to raise the barrier if no-one is waiting for the
693 * barrier to go down. This means that as soon as an IO request
694 * is ready, no other operations which require a barrier will start
695 * until the IO request has had a chance.
696 *
697 * So: regular IO calls 'wait_barrier'. When that returns there
698 * is no backgroup IO happening, It must arrange to call
699 * allow_barrier when it has finished its IO.
700 * backgroup IO calls must call raise_barrier. Once that returns
701 * there is no normal IO happeing. It must arrange to call
702 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 */
704#define RESYNC_DEPTH 32
705
NeilBrowne8096362011-10-11 16:49:05 +1100706static void raise_barrier(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
708 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800709
710 /* Wait until no block IO is waiting */
711 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
NeilBrownc3b328a2011-04-18 18:25:43 +1000712 conf->resync_lock, );
NeilBrown17999be2006-01-06 00:20:12 -0800713
714 /* block any new IO from starting */
715 conf->barrier++;
716
NeilBrown046abee2010-10-26 15:46:20 +1100717 /* Now wait for all pending IO to complete */
NeilBrown17999be2006-01-06 00:20:12 -0800718 wait_event_lock_irq(conf->wait_barrier,
719 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
NeilBrownc3b328a2011-04-18 18:25:43 +1000720 conf->resync_lock, );
NeilBrown17999be2006-01-06 00:20:12 -0800721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 spin_unlock_irq(&conf->resync_lock);
723}
724
NeilBrowne8096362011-10-11 16:49:05 +1100725static void lower_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800726{
727 unsigned long flags;
NeilBrown709ae482009-12-14 12:49:51 +1100728 BUG_ON(conf->barrier <= 0);
NeilBrown17999be2006-01-06 00:20:12 -0800729 spin_lock_irqsave(&conf->resync_lock, flags);
730 conf->barrier--;
731 spin_unlock_irqrestore(&conf->resync_lock, flags);
732 wake_up(&conf->wait_barrier);
733}
734
NeilBrowne8096362011-10-11 16:49:05 +1100735static void wait_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800736{
737 spin_lock_irq(&conf->resync_lock);
738 if (conf->barrier) {
739 conf->nr_waiting++;
740 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
741 conf->resync_lock,
NeilBrownc3b328a2011-04-18 18:25:43 +1000742 );
NeilBrown17999be2006-01-06 00:20:12 -0800743 conf->nr_waiting--;
744 }
745 conf->nr_pending++;
746 spin_unlock_irq(&conf->resync_lock);
747}
748
NeilBrowne8096362011-10-11 16:49:05 +1100749static void allow_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800750{
751 unsigned long flags;
752 spin_lock_irqsave(&conf->resync_lock, flags);
753 conf->nr_pending--;
754 spin_unlock_irqrestore(&conf->resync_lock, flags);
755 wake_up(&conf->wait_barrier);
756}
757
NeilBrowne8096362011-10-11 16:49:05 +1100758static void freeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800759{
760 /* stop syncio and normal IO and wait for everything to
761 * go quite.
762 * We increment barrier and nr_waiting, and then
NeilBrown1c830532008-03-04 14:29:35 -0800763 * wait until nr_pending match nr_queued+1
764 * This is called in the context of one normal IO request
765 * that has failed. Thus any sync request that might be pending
766 * will be blocked by nr_pending, and we need to wait for
767 * pending IO requests to complete or be queued for re-try.
768 * Thus the number queued (nr_queued) plus this request (1)
769 * must match the number of pending IOs (nr_pending) before
770 * we continue.
NeilBrownddaf22a2006-01-06 00:20:19 -0800771 */
772 spin_lock_irq(&conf->resync_lock);
773 conf->barrier++;
774 conf->nr_waiting++;
775 wait_event_lock_irq(conf->wait_barrier,
NeilBrown1c830532008-03-04 14:29:35 -0800776 conf->nr_pending == conf->nr_queued+1,
NeilBrownddaf22a2006-01-06 00:20:19 -0800777 conf->resync_lock,
NeilBrownc3b328a2011-04-18 18:25:43 +1000778 flush_pending_writes(conf));
NeilBrownddaf22a2006-01-06 00:20:19 -0800779 spin_unlock_irq(&conf->resync_lock);
780}
NeilBrowne8096362011-10-11 16:49:05 +1100781static void unfreeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800782{
783 /* reverse the effect of the freeze */
784 spin_lock_irq(&conf->resync_lock);
785 conf->barrier--;
786 conf->nr_waiting--;
787 wake_up(&conf->wait_barrier);
788 spin_unlock_irq(&conf->resync_lock);
789}
790
NeilBrown17999be2006-01-06 00:20:12 -0800791
NeilBrown4e780642010-10-19 12:54:01 +1100792/* duplicate the data pages for behind I/O
NeilBrown4e780642010-10-19 12:54:01 +1100793 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100794static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
NeilBrown4b6d2872005-09-09 16:23:47 -0700795{
796 int i;
797 struct bio_vec *bvec;
NeilBrown2ca68f52011-07-28 11:32:10 +1000798 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
NeilBrown4b6d2872005-09-09 16:23:47 -0700799 GFP_NOIO);
NeilBrown2ca68f52011-07-28 11:32:10 +1000800 if (unlikely(!bvecs))
NeilBrownaf6d7b72011-05-11 14:51:19 +1000801 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700802
NeilBrown4b6d2872005-09-09 16:23:47 -0700803 bio_for_each_segment(bvec, bio, i) {
NeilBrown2ca68f52011-07-28 11:32:10 +1000804 bvecs[i] = *bvec;
805 bvecs[i].bv_page = alloc_page(GFP_NOIO);
806 if (unlikely(!bvecs[i].bv_page))
NeilBrown4b6d2872005-09-09 16:23:47 -0700807 goto do_sync_io;
NeilBrown2ca68f52011-07-28 11:32:10 +1000808 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
809 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
810 kunmap(bvecs[i].bv_page);
NeilBrown4b6d2872005-09-09 16:23:47 -0700811 kunmap(bvec->bv_page);
812 }
NeilBrown2ca68f52011-07-28 11:32:10 +1000813 r1_bio->behind_bvecs = bvecs;
NeilBrownaf6d7b72011-05-11 14:51:19 +1000814 r1_bio->behind_page_count = bio->bi_vcnt;
815 set_bit(R1BIO_BehindIO, &r1_bio->state);
816 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700817
818do_sync_io:
NeilBrownaf6d7b72011-05-11 14:51:19 +1000819 for (i = 0; i < bio->bi_vcnt; i++)
NeilBrown2ca68f52011-07-28 11:32:10 +1000820 if (bvecs[i].bv_page)
821 put_page(bvecs[i].bv_page);
822 kfree(bvecs);
NeilBrown36a4e1f2011-10-07 14:23:17 +1100823 pr_debug("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
NeilBrown4b6d2872005-09-09 16:23:47 -0700824}
825
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -0700826static void make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
NeilBrowne8096362011-10-11 16:49:05 +1100828 struct r1conf *conf = mddev->private;
NeilBrown0f6d02d2011-10-11 16:48:46 +1100829 struct mirror_info *mirror;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100830 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 struct bio *read_bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +1000832 int i, disks;
NeilBrown84255d12008-05-23 13:04:32 -0700833 struct bitmap *bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -0700834 unsigned long flags;
Jens Axboea3623572005-11-01 09:26:16 +0100835 const int rw = bio_data_dir(bio);
NeilBrown2c7d46e2010-08-18 16:16:05 +1000836 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
Tejun Heoe9c74692010-09-03 11:56:18 +0200837 const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA));
NeilBrown3cb03002011-10-11 16:45:26 +1100838 struct md_rdev *blocked_rdev;
NeilBrownc3b328a2011-04-18 18:25:43 +1000839 int plugged;
NeilBrown1f68f0c2011-07-28 11:31:48 +1000840 int first_clone;
841 int sectors_handled;
842 int max_sectors;
NeilBrown191ea9b2005-06-21 17:17:23 -0700843
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 /*
845 * Register the new request and wait if the reconstruction
846 * thread has put up a bar for new requests.
847 * Continue immediately if no resync is active currently.
848 */
NeilBrown62de6082006-05-01 12:15:47 -0700849
NeilBrown3d310eb2005-06-21 17:17:26 -0700850 md_write_start(mddev, bio); /* wait on superblock update early */
851
NeilBrown6eef4b22009-12-14 12:49:51 +1100852 if (bio_data_dir(bio) == WRITE &&
853 bio->bi_sector + bio->bi_size/512 > mddev->suspend_lo &&
854 bio->bi_sector < mddev->suspend_hi) {
855 /* As the suspend_* range is controlled by
856 * userspace, we want an interruptible
857 * wait.
858 */
859 DEFINE_WAIT(w);
860 for (;;) {
861 flush_signals(current);
862 prepare_to_wait(&conf->wait_barrier,
863 &w, TASK_INTERRUPTIBLE);
864 if (bio->bi_sector + bio->bi_size/512 <= mddev->suspend_lo ||
865 bio->bi_sector >= mddev->suspend_hi)
866 break;
867 schedule();
868 }
869 finish_wait(&conf->wait_barrier, &w);
870 }
NeilBrown62de6082006-05-01 12:15:47 -0700871
NeilBrown17999be2006-01-06 00:20:12 -0800872 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
NeilBrown84255d12008-05-23 13:04:32 -0700874 bitmap = mddev->bitmap;
875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 /*
877 * make_request() can abort the operation when READA is being
878 * used and no empty request is available.
879 *
880 */
881 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
882
883 r1_bio->master_bio = bio;
884 r1_bio->sectors = bio->bi_size >> 9;
NeilBrown191ea9b2005-06-21 17:17:23 -0700885 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 r1_bio->mddev = mddev;
887 r1_bio->sector = bio->bi_sector;
888
NeilBrownd2eb35a2011-07-28 11:31:48 +1000889 /* We might need to issue multiple reads to different
890 * devices if there are bad blocks around, so we keep
891 * track of the number of reads in bio->bi_phys_segments.
892 * If this is 0, there is only one r1_bio and no locking
893 * will be needed when requests complete. If it is
894 * non-zero, then it is the number of not-completed requests.
895 */
896 bio->bi_phys_segments = 0;
897 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
898
Jens Axboea3623572005-11-01 09:26:16 +0100899 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 /*
901 * read balancing logic:
902 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000903 int rdisk;
904
905read_again:
906 rdisk = read_balance(conf, r1_bio, &max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 if (rdisk < 0) {
909 /* couldn't find anywhere to read from */
910 raid_end_bio_io(r1_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200911 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 }
913 mirror = conf->mirrors + rdisk;
914
NeilBrowne5551902010-03-31 11:21:44 +1100915 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
916 bitmap) {
917 /* Reading from a write-mostly device must
918 * take care not to over-take any writes
919 * that are 'behind'
920 */
921 wait_event(bitmap->behind_wait,
922 atomic_read(&bitmap->behind_writes) == 0);
923 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 r1_bio->read_disk = rdisk;
925
NeilBrowna167f662010-10-26 18:31:13 +1100926 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000927 md_trim_bio(read_bio, r1_bio->sector - bio->bi_sector,
928 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930 r1_bio->bios[rdisk] = read_bio;
931
932 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
933 read_bio->bi_bdev = mirror->rdev->bdev;
934 read_bio->bi_end_io = raid1_end_read_request;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200935 read_bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 read_bio->bi_private = r1_bio;
937
NeilBrownd2eb35a2011-07-28 11:31:48 +1000938 if (max_sectors < r1_bio->sectors) {
939 /* could not read all from this device, so we will
940 * need another r1_bio.
941 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000942
943 sectors_handled = (r1_bio->sector + max_sectors
944 - bio->bi_sector);
945 r1_bio->sectors = max_sectors;
946 spin_lock_irq(&conf->device_lock);
947 if (bio->bi_phys_segments == 0)
948 bio->bi_phys_segments = 2;
949 else
950 bio->bi_phys_segments++;
951 spin_unlock_irq(&conf->device_lock);
952 /* Cannot call generic_make_request directly
953 * as that will be queued in __make_request
954 * and subsequent mempool_alloc might block waiting
955 * for it. So hand bio over to raid1d.
956 */
957 reschedule_retry(r1_bio);
958
959 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
960
961 r1_bio->master_bio = bio;
962 r1_bio->sectors = (bio->bi_size >> 9) - sectors_handled;
963 r1_bio->state = 0;
964 r1_bio->mddev = mddev;
965 r1_bio->sector = bio->bi_sector + sectors_handled;
966 goto read_again;
967 } else
968 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200969 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
971
972 /*
973 * WRITE:
974 */
NeilBrown34db0cd2011-10-11 16:50:01 +1100975 if (conf->pending_count >= max_queued_requests) {
976 md_wakeup_thread(mddev->thread);
977 wait_event(conf->wait_barrier,
978 conf->pending_count < max_queued_requests);
979 }
NeilBrown1f68f0c2011-07-28 11:31:48 +1000980 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 * inc refcount on their rdev. Record them by setting
982 * bios[x] to bio
NeilBrown1f68f0c2011-07-28 11:31:48 +1000983 * If there are known/acknowledged bad blocks on any device on
984 * which we have seen a write error, we want to avoid writing those
985 * blocks.
986 * This potentially requires several writes to write around
987 * the bad blocks. Each set of writes gets it's own r1bio
988 * with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 */
NeilBrownc3b328a2011-04-18 18:25:43 +1000990 plugged = mddev_check_plugged(mddev);
991
NeilBrown8f19ccb2011-12-23 10:17:56 +1100992 disks = conf->raid_disks * 2;
Dan Williams6bfe0b42008-04-30 00:52:32 -0700993 retry_write:
994 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 rcu_read_lock();
NeilBrown1f68f0c2011-07-28 11:31:48 +1000996 max_sectors = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100998 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -0700999 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1000 atomic_inc(&rdev->nr_pending);
1001 blocked_rdev = rdev;
1002 break;
1003 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001004 r1_bio->bios[i] = NULL;
1005 if (!rdev || test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11001006 if (i < conf->raid_disks)
1007 set_bit(R1BIO_Degraded, &r1_bio->state);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001008 continue;
1009 }
1010
1011 atomic_inc(&rdev->nr_pending);
1012 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1013 sector_t first_bad;
1014 int bad_sectors;
1015 int is_bad;
1016
1017 is_bad = is_badblock(rdev, r1_bio->sector,
1018 max_sectors,
1019 &first_bad, &bad_sectors);
1020 if (is_bad < 0) {
1021 /* mustn't write here until the bad block is
1022 * acknowledged*/
1023 set_bit(BlockedBadBlocks, &rdev->flags);
1024 blocked_rdev = rdev;
1025 break;
NeilBrown964147d2010-05-18 15:27:13 +10001026 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001027 if (is_bad && first_bad <= r1_bio->sector) {
1028 /* Cannot write here at all */
1029 bad_sectors -= (r1_bio->sector - first_bad);
1030 if (bad_sectors < max_sectors)
1031 /* mustn't write more than bad_sectors
1032 * to other devices yet
1033 */
1034 max_sectors = bad_sectors;
1035 rdev_dec_pending(rdev, mddev);
1036 /* We don't set R1BIO_Degraded as that
1037 * only applies if the disk is
1038 * missing, so it might be re-added,
1039 * and we want to know to recover this
1040 * chunk.
1041 * In this case the device is here,
1042 * and the fact that this chunk is not
1043 * in-sync is recorded in the bad
1044 * block log
1045 */
1046 continue;
1047 }
1048 if (is_bad) {
1049 int good_sectors = first_bad - r1_bio->sector;
1050 if (good_sectors < max_sectors)
1051 max_sectors = good_sectors;
1052 }
1053 }
1054 r1_bio->bios[i] = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 }
1056 rcu_read_unlock();
1057
Dan Williams6bfe0b42008-04-30 00:52:32 -07001058 if (unlikely(blocked_rdev)) {
1059 /* Wait for this device to become unblocked */
1060 int j;
1061
1062 for (j = 0; j < i; j++)
1063 if (r1_bio->bios[j])
1064 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001065 r1_bio->state = 0;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001066 allow_barrier(conf);
1067 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1068 wait_barrier(conf);
1069 goto retry_write;
1070 }
1071
NeilBrown1f68f0c2011-07-28 11:31:48 +10001072 if (max_sectors < r1_bio->sectors) {
1073 /* We are splitting this write into multiple parts, so
1074 * we need to prepare for allocating another r1_bio.
1075 */
1076 r1_bio->sectors = max_sectors;
1077 spin_lock_irq(&conf->device_lock);
1078 if (bio->bi_phys_segments == 0)
1079 bio->bi_phys_segments = 2;
1080 else
1081 bio->bi_phys_segments++;
1082 spin_unlock_irq(&conf->device_lock);
NeilBrown191ea9b2005-06-21 17:17:23 -07001083 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001084 sectors_handled = r1_bio->sector + max_sectors - bio->bi_sector;
NeilBrown4b6d2872005-09-09 16:23:47 -07001085
NeilBrown4e780642010-10-19 12:54:01 +11001086 atomic_set(&r1_bio->remaining, 1);
NeilBrown4b6d2872005-09-09 16:23:47 -07001087 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -07001088
NeilBrown1f68f0c2011-07-28 11:31:48 +10001089 first_clone = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 for (i = 0; i < disks; i++) {
1091 struct bio *mbio;
1092 if (!r1_bio->bios[i])
1093 continue;
1094
NeilBrowna167f662010-10-26 18:31:13 +11001095 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001096 md_trim_bio(mbio, r1_bio->sector - bio->bi_sector, max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
NeilBrown1f68f0c2011-07-28 11:31:48 +10001098 if (first_clone) {
1099 /* do behind I/O ?
1100 * Not if there are too many, or cannot
1101 * allocate memory, or a reader on WriteMostly
1102 * is waiting for behind writes to flush */
1103 if (bitmap &&
1104 (atomic_read(&bitmap->behind_writes)
1105 < mddev->bitmap_info.max_write_behind) &&
1106 !waitqueue_active(&bitmap->behind_wait))
1107 alloc_behind_pages(mbio, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
NeilBrown1f68f0c2011-07-28 11:31:48 +10001109 bitmap_startwrite(bitmap, r1_bio->sector,
1110 r1_bio->sectors,
1111 test_bit(R1BIO_BehindIO,
1112 &r1_bio->state));
1113 first_clone = 0;
1114 }
NeilBrown2ca68f52011-07-28 11:32:10 +10001115 if (r1_bio->behind_bvecs) {
NeilBrown4b6d2872005-09-09 16:23:47 -07001116 struct bio_vec *bvec;
1117 int j;
1118
1119 /* Yes, I really want the '__' version so that
1120 * we clear any unused pointer in the io_vec, rather
1121 * than leave them unchanged. This is important
1122 * because when we come to free the pages, we won't
NeilBrown046abee2010-10-26 15:46:20 +11001123 * know the original bi_idx, so we just free
NeilBrown4b6d2872005-09-09 16:23:47 -07001124 * them all
1125 */
1126 __bio_for_each_segment(bvec, mbio, j, 0)
NeilBrown2ca68f52011-07-28 11:32:10 +10001127 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
NeilBrown4b6d2872005-09-09 16:23:47 -07001128 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1129 atomic_inc(&r1_bio->behind_remaining);
1130 }
1131
NeilBrown1f68f0c2011-07-28 11:31:48 +10001132 r1_bio->bios[i] = mbio;
1133
1134 mbio->bi_sector = (r1_bio->sector +
1135 conf->mirrors[i].rdev->data_offset);
1136 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1137 mbio->bi_end_io = raid1_end_write_request;
1138 mbio->bi_rw = WRITE | do_flush_fua | do_sync;
1139 mbio->bi_private = r1_bio;
1140
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 atomic_inc(&r1_bio->remaining);
NeilBrown4e780642010-10-19 12:54:01 +11001142 spin_lock_irqsave(&conf->device_lock, flags);
1143 bio_list_add(&conf->pending_bio_list, mbio);
NeilBrown34db0cd2011-10-11 16:50:01 +11001144 conf->pending_count++;
NeilBrown4e780642010-10-19 12:54:01 +11001145 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 }
NeilBrown079fa162011-09-10 17:21:23 +10001147 /* Mustn't call r1_bio_write_done before this next test,
1148 * as it could result in the bio being freed.
1149 */
NeilBrown1f68f0c2011-07-28 11:31:48 +10001150 if (sectors_handled < (bio->bi_size >> 9)) {
NeilBrown079fa162011-09-10 17:21:23 +10001151 r1_bio_write_done(r1_bio);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001152 /* We need another r1_bio. It has already been counted
1153 * in bio->bi_phys_segments
1154 */
1155 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1156 r1_bio->master_bio = bio;
1157 r1_bio->sectors = (bio->bi_size >> 9) - sectors_handled;
1158 r1_bio->state = 0;
1159 r1_bio->mddev = mddev;
1160 r1_bio->sector = bio->bi_sector + sectors_handled;
1161 goto retry_write;
1162 }
1163
NeilBrown079fa162011-09-10 17:21:23 +10001164 r1_bio_write_done(r1_bio);
1165
1166 /* In case raid1d snuck in to freeze_array */
1167 wake_up(&conf->wait_barrier);
1168
NeilBrownc3b328a2011-04-18 18:25:43 +10001169 if (do_sync || !bitmap || !plugged)
Lars Ellenberge3881a62007-01-10 23:15:37 -08001170 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171}
1172
NeilBrownfd01b882011-10-11 16:47:53 +11001173static void status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174{
NeilBrowne8096362011-10-11 16:49:05 +11001175 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 int i;
1177
1178 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001179 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001180 rcu_read_lock();
1181 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001182 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001184 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1185 }
1186 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 seq_printf(seq, "]");
1188}
1189
1190
NeilBrownfd01b882011-10-11 16:47:53 +11001191static void error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192{
1193 char b[BDEVNAME_SIZE];
NeilBrowne8096362011-10-11 16:49:05 +11001194 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
1196 /*
1197 * If it is not operational, then we have already marked it as dead
1198 * else if it is the last working disks, ignore the error, let the
1199 * next level up know.
1200 * else mark the drive as failed
1201 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001202 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001203 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 /*
1205 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001206 * normal single drive.
1207 * However don't try a recovery from this drive as
1208 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 */
NeilBrown53890422011-07-27 11:00:36 +10001210 conf->recovery_disabled = mddev->recovery_disabled;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001212 }
NeilBrownde393cd2011-07-28 11:31:48 +10001213 set_bit(Blocked, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001214 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1215 unsigned long flags;
1216 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -07001218 set_bit(Faulty, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001219 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 /*
1221 * if recovery is running, make sure it aborts.
1222 */
NeilBrowndfc70642008-05-23 13:04:39 -07001223 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrowndd00a992007-05-10 03:15:50 -07001224 } else
1225 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001226 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Joe Perches067032b2011-01-14 09:14:33 +11001227 printk(KERN_ALERT
1228 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1229 "md/raid1:%s: Operation continuing on %d devices.\n",
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001230 mdname(mddev), bdevname(rdev->bdev, b),
1231 mdname(mddev), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232}
1233
NeilBrowne8096362011-10-11 16:49:05 +11001234static void print_conf(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235{
1236 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001238 printk(KERN_DEBUG "RAID1 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 if (!conf) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001240 printk(KERN_DEBUG "(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 return;
1242 }
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001243 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 conf->raid_disks);
1245
NeilBrownddac7c72006-08-31 21:27:36 -07001246 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 for (i = 0; i < conf->raid_disks; i++) {
1248 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001249 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownddac7c72006-08-31 21:27:36 -07001250 if (rdev)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001251 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -07001252 i, !test_bit(In_sync, &rdev->flags),
1253 !test_bit(Faulty, &rdev->flags),
1254 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
NeilBrownddac7c72006-08-31 21:27:36 -07001256 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
NeilBrowne8096362011-10-11 16:49:05 +11001259static void close_sync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260{
NeilBrown17999be2006-01-06 00:20:12 -08001261 wait_barrier(conf);
1262 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264 mempool_destroy(conf->r1buf_pool);
1265 conf->r1buf_pool = NULL;
1266}
1267
NeilBrownfd01b882011-10-11 16:47:53 +11001268static int raid1_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269{
1270 int i;
NeilBrowne8096362011-10-11 16:49:05 +11001271 struct r1conf *conf = mddev->private;
NeilBrown6b965622010-08-18 11:56:59 +10001272 int count = 0;
1273 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
1275 /*
1276 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001277 * and mark them readable.
1278 * Called under mddev lock, so rcu protection not needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 */
1280 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001281 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001282 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1283 if (repl
1284 && repl->recovery_offset == MaxSector
1285 && !test_bit(Faulty, &repl->flags)
1286 && !test_and_set_bit(In_sync, &repl->flags)) {
1287 /* replacement has just become active */
1288 if (!rdev ||
1289 !test_and_clear_bit(In_sync, &rdev->flags))
1290 count++;
1291 if (rdev) {
1292 /* Replaced device not technically
1293 * faulty, but we need to be sure
1294 * it gets removed and never re-added
1295 */
1296 set_bit(Faulty, &rdev->flags);
1297 sysfs_notify_dirent_safe(
1298 rdev->sysfs_state);
1299 }
1300 }
NeilBrownddac7c72006-08-31 21:27:36 -07001301 if (rdev
1302 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001303 && !test_and_set_bit(In_sync, &rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001304 count++;
Jonathan Brassow654e8b52011-07-27 11:00:36 +10001305 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 }
1307 }
NeilBrown6b965622010-08-18 11:56:59 +10001308 spin_lock_irqsave(&conf->device_lock, flags);
1309 mddev->degraded -= count;
1310 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001313 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
1315
1316
NeilBrownfd01b882011-10-11 16:47:53 +11001317static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318{
NeilBrowne8096362011-10-11 16:49:05 +11001319 struct r1conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001320 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001321 int mirror = 0;
NeilBrown0f6d02d2011-10-11 16:48:46 +11001322 struct mirror_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001323 int first = 0;
NeilBrown30194632011-12-23 10:17:56 +11001324 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
NeilBrown53890422011-07-27 11:00:36 +10001326 if (mddev->recovery_disabled == conf->recovery_disabled)
1327 return -EBUSY;
1328
Neil Brown6c2fce22008-06-28 08:31:31 +10001329 if (rdev->raid_disk >= 0)
1330 first = last = rdev->raid_disk;
1331
NeilBrown7ef449d2011-12-23 10:17:57 +11001332 for (mirror = first; mirror <= last; mirror++) {
1333 p = conf->mirrors+mirror;
1334 if (!p->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10001336 disk_stack_limits(mddev->gendisk, rdev->bdev,
1337 rdev->data_offset << 9);
NeilBrown627a2d32010-03-08 16:44:38 +11001338 /* as we don't honour merge_bvec_fn, we must
1339 * never risk violating it, so limit
1340 * ->max_segments to one lying with a single
1341 * page, as a one page request is never in
1342 * violation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 */
NeilBrown627a2d32010-03-08 16:44:38 +11001344 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
1345 blk_queue_max_segments(mddev->queue, 1);
1346 blk_queue_segment_boundary(mddev->queue,
1347 PAGE_CACHE_SIZE - 1);
1348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 p->head_position = 0;
1351 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001352 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001353 /* As all devices are equivalent, we don't need a full recovery
1354 * if this was recently any drive of the array
1355 */
1356 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001357 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001358 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 break;
1360 }
NeilBrown7ef449d2011-12-23 10:17:57 +11001361 if (test_bit(WantReplacement, &p->rdev->flags) &&
1362 p[conf->raid_disks].rdev == NULL) {
1363 /* Add this device as a replacement */
1364 clear_bit(In_sync, &rdev->flags);
1365 set_bit(Replacement, &rdev->flags);
1366 rdev->raid_disk = mirror;
1367 err = 0;
1368 conf->fullsync = 1;
1369 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1370 break;
1371 }
1372 }
Andre Nollac5e7112009-08-03 10:59:47 +10001373 md_integrity_add_rdev(rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001375 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376}
1377
NeilBrownb8321b62011-12-23 10:17:51 +11001378static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
NeilBrowne8096362011-10-11 16:49:05 +11001380 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001382 int number = rdev->raid_disk;
NeilBrown0f6d02d2011-10-11 16:48:46 +11001383 struct mirror_info *p = conf->mirrors+ number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
NeilBrownb014f142011-12-23 10:17:56 +11001385 if (rdev != p->rdev)
1386 p = conf->mirrors + conf->raid_disks + number;
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 print_conf(conf);
NeilBrownb8321b62011-12-23 10:17:51 +11001389 if (rdev == p->rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001390 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 atomic_read(&rdev->nr_pending)) {
1392 err = -EBUSY;
1393 goto abort;
1394 }
NeilBrown046abee2010-10-26 15:46:20 +11001395 /* Only remove non-faulty devices if recovery
NeilBrowndfc70642008-05-23 13:04:39 -07001396 * is not possible.
1397 */
1398 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown53890422011-07-27 11:00:36 +10001399 mddev->recovery_disabled != conf->recovery_disabled &&
NeilBrowndfc70642008-05-23 13:04:39 -07001400 mddev->degraded < conf->raid_disks) {
1401 err = -EBUSY;
1402 goto abort;
1403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001405 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 if (atomic_read(&rdev->nr_pending)) {
1407 /* lost the race, try later */
1408 err = -EBUSY;
1409 p->rdev = rdev;
Andre Nollac5e7112009-08-03 10:59:47 +10001410 goto abort;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001411 } else if (conf->mirrors[conf->raid_disks + number].rdev) {
1412 /* We just removed a device that is being replaced.
1413 * Move down the replacement. We drain all IO before
1414 * doing this to avoid confusion.
1415 */
1416 struct md_rdev *repl =
1417 conf->mirrors[conf->raid_disks + number].rdev;
1418 raise_barrier(conf);
1419 clear_bit(Replacement, &repl->flags);
1420 p->rdev = repl;
1421 conf->mirrors[conf->raid_disks + number].rdev = NULL;
1422 lower_barrier(conf);
NeilBrownb014f142011-12-23 10:17:56 +11001423 clear_bit(WantReplacement, &rdev->flags);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001424 } else
1425 clear_bit(WantReplacement, &rdev->flags);
Martin K. Petersena91a2782011-03-17 11:11:05 +01001426 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 }
1428abort:
1429
1430 print_conf(conf);
1431 return err;
1432}
1433
1434
NeilBrown6712ecf2007-09-27 12:47:43 +02001435static void end_sync_read(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001437 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
NeilBrown0fc280f2011-10-07 14:22:55 +11001439 update_head_pos(r1_bio->read_disk, r1_bio);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 /*
1442 * we have read a block, now it needs to be re-written,
1443 * or re-read if the read failed.
1444 * We don't do much here, just schedule handling by raid1d
1445 */
NeilBrown69382e82006-01-06 00:20:22 -08001446 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001448
1449 if (atomic_dec_and_test(&r1_bio->remaining))
1450 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451}
1452
NeilBrown6712ecf2007-09-27 12:47:43 +02001453static void end_sync_write(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454{
1455 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +11001456 struct r1bio *r1_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11001457 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001458 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 int mirror=0;
NeilBrown4367af52011-07-28 11:31:49 +10001460 sector_t first_bad;
1461 int bad_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001463 mirror = find_bio_disk(r1_bio, bio);
1464
NeilBrown6b1117d2006-03-31 02:31:57 -08001465 if (!uptodate) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001466 sector_t sync_blocks = 0;
NeilBrown6b1117d2006-03-31 02:31:57 -08001467 sector_t s = r1_bio->sector;
1468 long sectors_to_go = r1_bio->sectors;
1469 /* make sure these bits doesn't get cleared. */
1470 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001471 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001472 &sync_blocks, 1);
1473 s += sync_blocks;
1474 sectors_to_go -= sync_blocks;
1475 } while (sectors_to_go > 0);
NeilBrownd8f05d22011-07-28 11:33:00 +10001476 set_bit(WriteErrorSeen,
1477 &conf->mirrors[mirror].rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001478 if (!test_and_set_bit(WantReplacement,
1479 &conf->mirrors[mirror].rdev->flags))
1480 set_bit(MD_RECOVERY_NEEDED, &
1481 mddev->recovery);
NeilBrownd8f05d22011-07-28 11:33:00 +10001482 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +10001483 } else if (is_badblock(conf->mirrors[mirror].rdev,
1484 r1_bio->sector,
1485 r1_bio->sectors,
NeilBrown3a9f28a2011-07-28 11:33:42 +10001486 &first_bad, &bad_sectors) &&
1487 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1488 r1_bio->sector,
1489 r1_bio->sectors,
1490 &first_bad, &bad_sectors)
1491 )
NeilBrown4367af52011-07-28 11:31:49 +10001492 set_bit(R1BIO_MadeGood, &r1_bio->state);
NeilBrowne3b97032005-08-04 12:53:34 -07001493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown4367af52011-07-28 11:31:49 +10001495 int s = r1_bio->sectors;
NeilBrownd8f05d22011-07-28 11:33:00 +10001496 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1497 test_bit(R1BIO_WriteError, &r1_bio->state))
NeilBrown4367af52011-07-28 11:31:49 +10001498 reschedule_retry(r1_bio);
1499 else {
1500 put_buf(r1_bio);
1501 md_done_sync(mddev, s, uptodate);
1502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504}
1505
NeilBrown3cb03002011-10-11 16:45:26 +11001506static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrownd8f05d22011-07-28 11:33:00 +10001507 int sectors, struct page *page, int rw)
1508{
1509 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
1510 /* success */
1511 return 1;
NeilBrown19d67162011-12-23 10:17:57 +11001512 if (rw == WRITE) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001513 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001514 if (!test_and_set_bit(WantReplacement,
1515 &rdev->flags))
1516 set_bit(MD_RECOVERY_NEEDED, &
1517 rdev->mddev->recovery);
1518 }
NeilBrownd8f05d22011-07-28 11:33:00 +10001519 /* need to record an error - either for the block or the device */
1520 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1521 md_error(rdev->mddev, rdev);
1522 return 0;
1523}
1524
NeilBrown9f2c9d12011-10-11 16:48:43 +11001525static int fix_sync_read_error(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001526{
1527 /* Try some synchronous reads of other devices to get
1528 * good data, much like with normal read errors. Only
1529 * read into the pages we already have so we don't
1530 * need to re-issue the read request.
1531 * We don't need to freeze the array, because being in an
1532 * active sync request, there is no normal IO, and
1533 * no overlapping syncs.
NeilBrown06f60382011-07-28 11:31:48 +10001534 * We don't need to check is_badblock() again as we
1535 * made sure that anything with a bad block in range
1536 * will have bi_end_io clear.
NeilBrowna68e5872011-05-11 14:40:44 +10001537 */
NeilBrownfd01b882011-10-11 16:47:53 +11001538 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001539 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001540 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1541 sector_t sect = r1_bio->sector;
1542 int sectors = r1_bio->sectors;
1543 int idx = 0;
1544
1545 while(sectors) {
1546 int s = sectors;
1547 int d = r1_bio->read_disk;
1548 int success = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11001549 struct md_rdev *rdev;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001550 int start;
NeilBrowna68e5872011-05-11 14:40:44 +10001551
1552 if (s > (PAGE_SIZE>>9))
1553 s = PAGE_SIZE >> 9;
1554 do {
1555 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1556 /* No rcu protection needed here devices
1557 * can only be removed when no resync is
1558 * active, and resync is currently active
1559 */
1560 rdev = conf->mirrors[d].rdev;
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001561 if (sync_page_io(rdev, sect, s<<9,
NeilBrowna68e5872011-05-11 14:40:44 +10001562 bio->bi_io_vec[idx].bv_page,
1563 READ, false)) {
1564 success = 1;
1565 break;
1566 }
1567 }
1568 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001569 if (d == conf->raid_disks * 2)
NeilBrowna68e5872011-05-11 14:40:44 +10001570 d = 0;
1571 } while (!success && d != r1_bio->read_disk);
1572
NeilBrown78d7f5f2011-05-11 14:48:56 +10001573 if (!success) {
NeilBrowna68e5872011-05-11 14:40:44 +10001574 char b[BDEVNAME_SIZE];
NeilBrown3a9f28a2011-07-28 11:33:42 +10001575 int abort = 0;
1576 /* Cannot read from anywhere, this block is lost.
1577 * Record a bad block on each device. If that doesn't
1578 * work just disable and interrupt the recovery.
1579 * Don't fail devices as that won't really help.
1580 */
NeilBrowna68e5872011-05-11 14:40:44 +10001581 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1582 " for block %llu\n",
1583 mdname(mddev),
1584 bdevname(bio->bi_bdev, b),
1585 (unsigned long long)r1_bio->sector);
NeilBrown8f19ccb2011-12-23 10:17:56 +11001586 for (d = 0; d < conf->raid_disks * 2; d++) {
NeilBrown3a9f28a2011-07-28 11:33:42 +10001587 rdev = conf->mirrors[d].rdev;
1588 if (!rdev || test_bit(Faulty, &rdev->flags))
1589 continue;
1590 if (!rdev_set_badblocks(rdev, sect, s, 0))
1591 abort = 1;
1592 }
1593 if (abort) {
NeilBrownd890fa22011-10-26 11:54:39 +11001594 conf->recovery_disabled =
1595 mddev->recovery_disabled;
NeilBrown3a9f28a2011-07-28 11:33:42 +10001596 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1597 md_done_sync(mddev, r1_bio->sectors, 0);
1598 put_buf(r1_bio);
1599 return 0;
1600 }
1601 /* Try next page */
1602 sectors -= s;
1603 sect += s;
1604 idx++;
1605 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001606 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001607
1608 start = d;
1609 /* write it back and re-read */
1610 while (d != r1_bio->read_disk) {
1611 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001612 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001613 d--;
1614 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1615 continue;
1616 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001617 if (r1_sync_page_io(rdev, sect, s,
1618 bio->bi_io_vec[idx].bv_page,
1619 WRITE) == 0) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001620 r1_bio->bios[d]->bi_end_io = NULL;
1621 rdev_dec_pending(rdev, mddev);
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001622 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001623 }
1624 d = start;
1625 while (d != r1_bio->read_disk) {
1626 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001627 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001628 d--;
1629 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1630 continue;
1631 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001632 if (r1_sync_page_io(rdev, sect, s,
1633 bio->bi_io_vec[idx].bv_page,
1634 READ) != 0)
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001635 atomic_add(s, &rdev->corrected_errors);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001636 }
NeilBrowna68e5872011-05-11 14:40:44 +10001637 sectors -= s;
1638 sect += s;
1639 idx ++;
1640 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001641 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrown7ca78d52011-05-11 14:50:37 +10001642 set_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrowna68e5872011-05-11 14:40:44 +10001643 return 1;
1644}
1645
NeilBrown9f2c9d12011-10-11 16:48:43 +11001646static int process_checks(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001647{
1648 /* We have read all readable devices. If we haven't
1649 * got the block, then there is no hope left.
1650 * If we have, then we want to do a comparison
1651 * and skip the write if everything is the same.
1652 * If any blocks failed to read, then we need to
1653 * attempt an over-write
1654 */
NeilBrownfd01b882011-10-11 16:47:53 +11001655 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001656 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001657 int primary;
1658 int i;
1659
NeilBrown8f19ccb2011-12-23 10:17:56 +11001660 for (primary = 0; primary < conf->raid_disks * 2; primary++)
NeilBrowna68e5872011-05-11 14:40:44 +10001661 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1662 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1663 r1_bio->bios[primary]->bi_end_io = NULL;
1664 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1665 break;
1666 }
1667 r1_bio->read_disk = primary;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001668 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001669 int j;
1670 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1671 struct bio *pbio = r1_bio->bios[primary];
1672 struct bio *sbio = r1_bio->bios[i];
1673 int size;
NeilBrowna68e5872011-05-11 14:40:44 +10001674
NeilBrown78d7f5f2011-05-11 14:48:56 +10001675 if (r1_bio->bios[i]->bi_end_io != end_sync_read)
1676 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001677
NeilBrown78d7f5f2011-05-11 14:48:56 +10001678 if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) {
1679 for (j = vcnt; j-- ; ) {
1680 struct page *p, *s;
1681 p = pbio->bi_io_vec[j].bv_page;
1682 s = sbio->bi_io_vec[j].bv_page;
1683 if (memcmp(page_address(p),
1684 page_address(s),
1685 PAGE_SIZE))
1686 break;
NeilBrowna68e5872011-05-11 14:40:44 +10001687 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001688 } else
1689 j = 0;
1690 if (j >= 0)
1691 mddev->resync_mismatches += r1_bio->sectors;
1692 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1693 && test_bit(BIO_UPTODATE, &sbio->bi_flags))) {
1694 /* No need to write to this device. */
1695 sbio->bi_end_io = NULL;
1696 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1697 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001698 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001699 /* fixup the bio for reuse */
1700 sbio->bi_vcnt = vcnt;
1701 sbio->bi_size = r1_bio->sectors << 9;
1702 sbio->bi_idx = 0;
1703 sbio->bi_phys_segments = 0;
1704 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1705 sbio->bi_flags |= 1 << BIO_UPTODATE;
1706 sbio->bi_next = NULL;
1707 sbio->bi_sector = r1_bio->sector +
1708 conf->mirrors[i].rdev->data_offset;
1709 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1710 size = sbio->bi_size;
1711 for (j = 0; j < vcnt ; j++) {
1712 struct bio_vec *bi;
1713 bi = &sbio->bi_io_vec[j];
1714 bi->bv_offset = 0;
1715 if (size > PAGE_SIZE)
1716 bi->bv_len = PAGE_SIZE;
1717 else
1718 bi->bv_len = size;
1719 size -= PAGE_SIZE;
1720 memcpy(page_address(bi->bv_page),
1721 page_address(pbio->bi_io_vec[j].bv_page),
1722 PAGE_SIZE);
1723 }
1724 }
NeilBrowna68e5872011-05-11 14:40:44 +10001725 return 0;
1726}
1727
NeilBrown9f2c9d12011-10-11 16:48:43 +11001728static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729{
NeilBrowne8096362011-10-11 16:49:05 +11001730 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 int i;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001732 int disks = conf->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 struct bio *bio, *wbio;
1734
1735 bio = r1_bio->bios[r1_bio->read_disk];
1736
NeilBrowna68e5872011-05-11 14:40:44 +10001737 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
1738 /* ouch - failed to read all of that. */
1739 if (!fix_sync_read_error(r1_bio))
1740 return;
NeilBrown7ca78d52011-05-11 14:50:37 +10001741
1742 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1743 if (process_checks(r1_bio) < 0)
1744 return;
NeilBrownd11c1712006-01-06 00:20:26 -08001745 /*
1746 * schedule writes
1747 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 atomic_set(&r1_bio->remaining, 1);
1749 for (i = 0; i < disks ; i++) {
1750 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08001751 if (wbio->bi_end_io == NULL ||
1752 (wbio->bi_end_io == end_sync_read &&
1753 (i == r1_bio->read_disk ||
1754 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 continue;
1756
NeilBrown3e198f72006-01-06 00:20:21 -08001757 wbio->bi_rw = WRITE;
1758 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 atomic_inc(&r1_bio->remaining);
1760 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
NeilBrown191ea9b2005-06-21 17:17:23 -07001761
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 generic_make_request(wbio);
1763 }
1764
1765 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001766 /* if we're here, all write(s) have completed, so clean up */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 md_done_sync(mddev, r1_bio->sectors, 1);
1768 put_buf(r1_bio);
1769 }
1770}
1771
1772/*
1773 * This is a kernel thread which:
1774 *
1775 * 1. Retries failed read operations on working mirrors.
1776 * 2. Updates the raid superblock when problems encounter.
NeilBrownd2eb35a2011-07-28 11:31:48 +10001777 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 */
1779
NeilBrowne8096362011-10-11 16:49:05 +11001780static void fix_read_error(struct r1conf *conf, int read_disk,
NeilBrown867868f2006-10-03 01:15:51 -07001781 sector_t sect, int sectors)
1782{
NeilBrownfd01b882011-10-11 16:47:53 +11001783 struct mddev *mddev = conf->mddev;
NeilBrown867868f2006-10-03 01:15:51 -07001784 while(sectors) {
1785 int s = sectors;
1786 int d = read_disk;
1787 int success = 0;
1788 int start;
NeilBrown3cb03002011-10-11 16:45:26 +11001789 struct md_rdev *rdev;
NeilBrown867868f2006-10-03 01:15:51 -07001790
1791 if (s > (PAGE_SIZE>>9))
1792 s = PAGE_SIZE >> 9;
1793
1794 do {
1795 /* Note: no rcu protection needed here
1796 * as this is synchronous in the raid1d thread
1797 * which is the thread that might remove
1798 * a device. If raid1d ever becomes multi-threaded....
1799 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001800 sector_t first_bad;
1801 int bad_sectors;
1802
NeilBrown867868f2006-10-03 01:15:51 -07001803 rdev = conf->mirrors[d].rdev;
1804 if (rdev &&
1805 test_bit(In_sync, &rdev->flags) &&
NeilBrownd2eb35a2011-07-28 11:31:48 +10001806 is_badblock(rdev, sect, s,
1807 &first_bad, &bad_sectors) == 0 &&
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11001808 sync_page_io(rdev, sect, s<<9,
1809 conf->tmppage, READ, false))
NeilBrown867868f2006-10-03 01:15:51 -07001810 success = 1;
1811 else {
1812 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001813 if (d == conf->raid_disks * 2)
NeilBrown867868f2006-10-03 01:15:51 -07001814 d = 0;
1815 }
1816 } while (!success && d != read_disk);
1817
1818 if (!success) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001819 /* Cannot read from anywhere - mark it bad */
NeilBrown3cb03002011-10-11 16:45:26 +11001820 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001821 if (!rdev_set_badblocks(rdev, sect, s, 0))
1822 md_error(mddev, rdev);
NeilBrown867868f2006-10-03 01:15:51 -07001823 break;
1824 }
1825 /* write it back and re-read */
1826 start = d;
1827 while (d != read_disk) {
1828 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001829 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07001830 d--;
1831 rdev = conf->mirrors[d].rdev;
1832 if (rdev &&
NeilBrownd8f05d22011-07-28 11:33:00 +10001833 test_bit(In_sync, &rdev->flags))
1834 r1_sync_page_io(rdev, sect, s,
1835 conf->tmppage, WRITE);
NeilBrown867868f2006-10-03 01:15:51 -07001836 }
1837 d = start;
1838 while (d != read_disk) {
1839 char b[BDEVNAME_SIZE];
1840 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001841 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07001842 d--;
1843 rdev = conf->mirrors[d].rdev;
1844 if (rdev &&
1845 test_bit(In_sync, &rdev->flags)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001846 if (r1_sync_page_io(rdev, sect, s,
1847 conf->tmppage, READ)) {
NeilBrown867868f2006-10-03 01:15:51 -07001848 atomic_add(s, &rdev->corrected_errors);
1849 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001850 "md/raid1:%s: read error corrected "
NeilBrown867868f2006-10-03 01:15:51 -07001851 "(%d sectors at %llu on %s)\n",
1852 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07001853 (unsigned long long)(sect +
1854 rdev->data_offset),
NeilBrown867868f2006-10-03 01:15:51 -07001855 bdevname(rdev->bdev, b));
1856 }
1857 }
1858 }
1859 sectors -= s;
1860 sect += s;
1861 }
1862}
1863
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001864static void bi_complete(struct bio *bio, int error)
1865{
1866 complete((struct completion *)bio->bi_private);
1867}
1868
1869static int submit_bio_wait(int rw, struct bio *bio)
1870{
1871 struct completion event;
1872 rw |= REQ_SYNC;
1873
1874 init_completion(&event);
1875 bio->bi_private = &event;
1876 bio->bi_end_io = bi_complete;
1877 submit_bio(rw, bio);
1878 wait_for_completion(&event);
1879
1880 return test_bit(BIO_UPTODATE, &bio->bi_flags);
1881}
1882
NeilBrown9f2c9d12011-10-11 16:48:43 +11001883static int narrow_write_error(struct r1bio *r1_bio, int i)
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001884{
NeilBrownfd01b882011-10-11 16:47:53 +11001885 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001886 struct r1conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11001887 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001888 int vcnt, idx;
1889 struct bio_vec *vec;
1890
1891 /* bio has the data to be written to device 'i' where
1892 * we just recently had a write error.
1893 * We repeatedly clone the bio and trim down to one block,
1894 * then try the write. Where the write fails we record
1895 * a bad block.
1896 * It is conceivable that the bio doesn't exactly align with
1897 * blocks. We must handle this somehow.
1898 *
1899 * We currently own a reference on the rdev.
1900 */
1901
1902 int block_sectors;
1903 sector_t sector;
1904 int sectors;
1905 int sect_to_write = r1_bio->sectors;
1906 int ok = 1;
1907
1908 if (rdev->badblocks.shift < 0)
1909 return 0;
1910
1911 block_sectors = 1 << rdev->badblocks.shift;
1912 sector = r1_bio->sector;
1913 sectors = ((sector + block_sectors)
1914 & ~(sector_t)(block_sectors - 1))
1915 - sector;
1916
1917 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
1918 vcnt = r1_bio->behind_page_count;
1919 vec = r1_bio->behind_bvecs;
1920 idx = 0;
1921 while (vec[idx].bv_page == NULL)
1922 idx++;
1923 } else {
1924 vcnt = r1_bio->master_bio->bi_vcnt;
1925 vec = r1_bio->master_bio->bi_io_vec;
1926 idx = r1_bio->master_bio->bi_idx;
1927 }
1928 while (sect_to_write) {
1929 struct bio *wbio;
1930 if (sectors > sect_to_write)
1931 sectors = sect_to_write;
1932 /* Write at 'sector' for 'sectors'*/
1933
1934 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
1935 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
1936 wbio->bi_sector = r1_bio->sector;
1937 wbio->bi_rw = WRITE;
1938 wbio->bi_vcnt = vcnt;
1939 wbio->bi_size = r1_bio->sectors << 9;
1940 wbio->bi_idx = idx;
1941
1942 md_trim_bio(wbio, sector - r1_bio->sector, sectors);
1943 wbio->bi_sector += rdev->data_offset;
1944 wbio->bi_bdev = rdev->bdev;
1945 if (submit_bio_wait(WRITE, wbio) == 0)
1946 /* failure! */
1947 ok = rdev_set_badblocks(rdev, sector,
1948 sectors, 0)
1949 && ok;
1950
1951 bio_put(wbio);
1952 sect_to_write -= sectors;
1953 sector += sectors;
1954 sectors = block_sectors;
1955 }
1956 return ok;
1957}
1958
NeilBrowne8096362011-10-11 16:49:05 +11001959static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10001960{
1961 int m;
1962 int s = r1_bio->sectors;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001963 for (m = 0; m < conf->raid_disks * 2 ; m++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001964 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10001965 struct bio *bio = r1_bio->bios[m];
1966 if (bio->bi_end_io == NULL)
1967 continue;
1968 if (test_bit(BIO_UPTODATE, &bio->bi_flags) &&
1969 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
1970 rdev_clear_badblocks(rdev, r1_bio->sector, s);
1971 }
1972 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
1973 test_bit(R1BIO_WriteError, &r1_bio->state)) {
1974 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
1975 md_error(conf->mddev, rdev);
1976 }
1977 }
1978 put_buf(r1_bio);
1979 md_done_sync(conf->mddev, s, 1);
1980}
1981
NeilBrowne8096362011-10-11 16:49:05 +11001982static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10001983{
1984 int m;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001985 for (m = 0; m < conf->raid_disks * 2 ; m++)
NeilBrown62096bc2011-07-28 11:38:13 +10001986 if (r1_bio->bios[m] == IO_MADE_GOOD) {
NeilBrown3cb03002011-10-11 16:45:26 +11001987 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10001988 rdev_clear_badblocks(rdev,
1989 r1_bio->sector,
1990 r1_bio->sectors);
1991 rdev_dec_pending(rdev, conf->mddev);
1992 } else if (r1_bio->bios[m] != NULL) {
1993 /* This drive got a write error. We need to
1994 * narrow down and record precise write
1995 * errors.
1996 */
1997 if (!narrow_write_error(r1_bio, m)) {
1998 md_error(conf->mddev,
1999 conf->mirrors[m].rdev);
2000 /* an I/O failed, we can't clear the bitmap */
2001 set_bit(R1BIO_Degraded, &r1_bio->state);
2002 }
2003 rdev_dec_pending(conf->mirrors[m].rdev,
2004 conf->mddev);
2005 }
2006 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2007 close_write(r1_bio);
2008 raid_end_bio_io(r1_bio);
2009}
2010
NeilBrowne8096362011-10-11 16:49:05 +11002011static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002012{
2013 int disk;
2014 int max_sectors;
NeilBrownfd01b882011-10-11 16:47:53 +11002015 struct mddev *mddev = conf->mddev;
NeilBrown62096bc2011-07-28 11:38:13 +10002016 struct bio *bio;
2017 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11002018 struct md_rdev *rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002019
2020 clear_bit(R1BIO_ReadError, &r1_bio->state);
2021 /* we got a read error. Maybe the drive is bad. Maybe just
2022 * the block and we can fix it.
2023 * We freeze all other IO, and try reading the block from
2024 * other devices. When we find one, we re-write
2025 * and check it that fixes the read error.
2026 * This is all done synchronously while the array is
2027 * frozen
2028 */
2029 if (mddev->ro == 0) {
2030 freeze_array(conf);
2031 fix_read_error(conf, r1_bio->read_disk,
2032 r1_bio->sector, r1_bio->sectors);
2033 unfreeze_array(conf);
2034 } else
2035 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
2036
2037 bio = r1_bio->bios[r1_bio->read_disk];
2038 bdevname(bio->bi_bdev, b);
2039read_more:
2040 disk = read_balance(conf, r1_bio, &max_sectors);
2041 if (disk == -1) {
2042 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
2043 " read error for block %llu\n",
2044 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2045 raid_end_bio_io(r1_bio);
2046 } else {
2047 const unsigned long do_sync
2048 = r1_bio->master_bio->bi_rw & REQ_SYNC;
2049 if (bio) {
2050 r1_bio->bios[r1_bio->read_disk] =
2051 mddev->ro ? IO_BLOCKED : NULL;
2052 bio_put(bio);
2053 }
2054 r1_bio->read_disk = disk;
2055 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2056 md_trim_bio(bio, r1_bio->sector - bio->bi_sector, max_sectors);
2057 r1_bio->bios[r1_bio->read_disk] = bio;
2058 rdev = conf->mirrors[disk].rdev;
2059 printk_ratelimited(KERN_ERR
2060 "md/raid1:%s: redirecting sector %llu"
2061 " to other mirror: %s\n",
2062 mdname(mddev),
2063 (unsigned long long)r1_bio->sector,
2064 bdevname(rdev->bdev, b));
2065 bio->bi_sector = r1_bio->sector + rdev->data_offset;
2066 bio->bi_bdev = rdev->bdev;
2067 bio->bi_end_io = raid1_end_read_request;
2068 bio->bi_rw = READ | do_sync;
2069 bio->bi_private = r1_bio;
2070 if (max_sectors < r1_bio->sectors) {
2071 /* Drat - have to split this up more */
2072 struct bio *mbio = r1_bio->master_bio;
2073 int sectors_handled = (r1_bio->sector + max_sectors
2074 - mbio->bi_sector);
2075 r1_bio->sectors = max_sectors;
2076 spin_lock_irq(&conf->device_lock);
2077 if (mbio->bi_phys_segments == 0)
2078 mbio->bi_phys_segments = 2;
2079 else
2080 mbio->bi_phys_segments++;
2081 spin_unlock_irq(&conf->device_lock);
2082 generic_make_request(bio);
2083 bio = NULL;
2084
2085 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2086
2087 r1_bio->master_bio = mbio;
2088 r1_bio->sectors = (mbio->bi_size >> 9)
2089 - sectors_handled;
2090 r1_bio->state = 0;
2091 set_bit(R1BIO_ReadError, &r1_bio->state);
2092 r1_bio->mddev = mddev;
2093 r1_bio->sector = mbio->bi_sector + sectors_handled;
2094
2095 goto read_more;
2096 } else
2097 generic_make_request(bio);
2098 }
2099}
2100
NeilBrownfd01b882011-10-11 16:47:53 +11002101static void raid1d(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102{
NeilBrown9f2c9d12011-10-11 16:48:43 +11002103 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 unsigned long flags;
NeilBrowne8096362011-10-11 16:49:05 +11002105 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002107 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
2109 md_check_recovery(mddev);
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002110
2111 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002113
NeilBrownc3b328a2011-04-18 18:25:43 +10002114 if (atomic_read(&mddev->plug_cnt) == 0)
2115 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002116
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002118 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002119 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002121 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002122 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08002124 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 spin_unlock_irqrestore(&conf->device_lock, flags);
2126
2127 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002128 conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10002129 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002130 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002131 test_bit(R1BIO_WriteError, &r1_bio->state))
2132 handle_sync_write_finished(conf, r1_bio);
2133 else
NeilBrown4367af52011-07-28 11:31:49 +10002134 sync_request_write(mddev, r1_bio);
NeilBrowncd5ff9a2011-07-28 11:32:41 +10002135 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002136 test_bit(R1BIO_WriteError, &r1_bio->state))
2137 handle_write_finished(conf, r1_bio);
2138 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2139 handle_read_error(conf, r1_bio);
2140 else
NeilBrownd2eb35a2011-07-28 11:31:48 +10002141 /* just a partial read to be scheduled from separate
2142 * context
2143 */
2144 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
NeilBrown62096bc2011-07-28 11:38:13 +10002145
NeilBrown1d9d5242009-10-16 15:55:32 +11002146 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002147 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2148 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002150 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151}
2152
2153
NeilBrowne8096362011-10-11 16:49:05 +11002154static int init_resync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155{
2156 int buffs;
2157
2158 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002159 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2161 conf->poolinfo);
2162 if (!conf->r1buf_pool)
2163 return -ENOMEM;
2164 conf->next_resync = 0;
2165 return 0;
2166}
2167
2168/*
2169 * perform a "sync" on one "block"
2170 *
2171 * We need to make sure that no normal I/O request - particularly write
2172 * requests - conflict with active sync requests.
2173 *
2174 * This is achieved by tracking pending requests and a 'barrier' concept
2175 * that can be installed to exclude normal IO requests.
2176 */
2177
NeilBrownfd01b882011-10-11 16:47:53 +11002178static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179{
NeilBrowne8096362011-10-11 16:49:05 +11002180 struct r1conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002181 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 struct bio *bio;
2183 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08002184 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08002186 int wonly = -1;
2187 int write_targets = 0, read_targets = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11002188 sector_t sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07002189 int still_degraded = 0;
NeilBrown06f60382011-07-28 11:31:48 +10002190 int good_sectors = RESYNC_SECTORS;
2191 int min_bad = 0; /* number of sectors that are bad in all devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
2193 if (!conf->r1buf_pool)
2194 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002195 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
Andre Noll58c0fed2009-03-31 14:33:13 +11002197 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002199 /* If we aborted, we need to abort the
2200 * sync on the 'current' bitmap chunk (there will
2201 * only be one in raid1 resync.
2202 * We can find the current addess in mddev->curr_resync
2203 */
NeilBrown6a806c52005-07-15 03:56:35 -07002204 if (mddev->curr_resync < max_sector) /* aborted */
2205 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07002206 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07002207 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07002208 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07002209
2210 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 close_sync(conf);
2212 return 0;
2213 }
2214
NeilBrown07d84d102006-06-26 00:27:56 -07002215 if (mddev->bitmap == NULL &&
2216 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07002217 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07002218 conf->fullsync == 0) {
2219 *skipped = 1;
2220 return max_sector - sector_nr;
2221 }
NeilBrown6394cca2006-08-27 01:23:50 -07002222 /* before building a request, check if we can skip these blocks..
2223 * This call the bitmap_start_sync doesn't actually record anything
2224 */
NeilBrowne3b97032005-08-04 12:53:34 -07002225 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de4852005-11-08 21:39:38 -08002226 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002227 /* We can skip this block, and probably several more */
2228 *skipped = 1;
2229 return sync_blocks;
2230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 /*
NeilBrown17999be2006-01-06 00:20:12 -08002232 * If there is non-resync activity waiting for a turn,
2233 * and resync is going fast enough,
2234 * then let it though before starting on this new sync request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 */
NeilBrown17999be2006-01-06 00:20:12 -08002236 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 msleep_interruptible(1000);
NeilBrown17999be2006-01-06 00:20:12 -08002238
NeilBrownb47490c2008-02-06 01:39:50 -08002239 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
NeilBrown1c4588e2010-10-26 17:41:22 +11002240 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown17999be2006-01-06 00:20:12 -08002241 raise_barrier(conf);
2242
2243 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
NeilBrown3e198f72006-01-06 00:20:21 -08002245 rcu_read_lock();
2246 /*
2247 * If we get a correctably read error during resync or recovery,
2248 * we might want to read from a different device. So we
2249 * flag all drives that could conceivably be read from for READ,
2250 * and any others (which will be non-In_sync devices) for WRITE.
2251 * If a read fails, we try reading from something else for which READ
2252 * is OK.
2253 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 r1_bio->mddev = mddev;
2256 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07002257 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259
NeilBrown8f19ccb2011-12-23 10:17:56 +11002260 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002261 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 bio = r1_bio->bios[i];
2263
2264 /* take from bio_init */
2265 bio->bi_next = NULL;
NeilBrowndb8d9d32010-10-07 12:00:50 +11002266 bio->bi_flags &= ~(BIO_POOL_MASK-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 bio->bi_flags |= 1 << BIO_UPTODATE;
NeilBrown802ba062006-12-13 00:34:13 -08002268 bio->bi_rw = READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 bio->bi_vcnt = 0;
2270 bio->bi_idx = 0;
2271 bio->bi_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 bio->bi_size = 0;
2273 bio->bi_end_io = NULL;
2274 bio->bi_private = NULL;
2275
NeilBrown3e198f72006-01-06 00:20:21 -08002276 rdev = rcu_dereference(conf->mirrors[i].rdev);
2277 if (rdev == NULL ||
NeilBrown06f60382011-07-28 11:31:48 +10002278 test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11002279 if (i < conf->raid_disks)
2280 still_degraded = 1;
NeilBrown3e198f72006-01-06 00:20:21 -08002281 } else if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 bio->bi_rw = WRITE;
2283 bio->bi_end_io = end_sync_write;
2284 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08002285 } else {
2286 /* may need to read from here */
NeilBrown06f60382011-07-28 11:31:48 +10002287 sector_t first_bad = MaxSector;
2288 int bad_sectors;
2289
2290 if (is_badblock(rdev, sector_nr, good_sectors,
2291 &first_bad, &bad_sectors)) {
2292 if (first_bad > sector_nr)
2293 good_sectors = first_bad - sector_nr;
2294 else {
2295 bad_sectors -= (sector_nr - first_bad);
2296 if (min_bad == 0 ||
2297 min_bad > bad_sectors)
2298 min_bad = bad_sectors;
2299 }
NeilBrown3e198f72006-01-06 00:20:21 -08002300 }
NeilBrown06f60382011-07-28 11:31:48 +10002301 if (sector_nr < first_bad) {
2302 if (test_bit(WriteMostly, &rdev->flags)) {
2303 if (wonly < 0)
2304 wonly = i;
2305 } else {
2306 if (disk < 0)
2307 disk = i;
2308 }
2309 bio->bi_rw = READ;
2310 bio->bi_end_io = end_sync_read;
2311 read_targets++;
2312 }
NeilBrown3e198f72006-01-06 00:20:21 -08002313 }
NeilBrown06f60382011-07-28 11:31:48 +10002314 if (bio->bi_end_io) {
2315 atomic_inc(&rdev->nr_pending);
2316 bio->bi_sector = sector_nr + rdev->data_offset;
2317 bio->bi_bdev = rdev->bdev;
2318 bio->bi_private = r1_bio;
2319 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 }
NeilBrown3e198f72006-01-06 00:20:21 -08002321 rcu_read_unlock();
2322 if (disk < 0)
2323 disk = wonly;
2324 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07002325
NeilBrown06f60382011-07-28 11:31:48 +10002326 if (read_targets == 0 && min_bad > 0) {
2327 /* These sectors are bad on all InSync devices, so we
2328 * need to mark them bad on all write targets
2329 */
2330 int ok = 1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002331 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
NeilBrown06f60382011-07-28 11:31:48 +10002332 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
NeilBrown3cb03002011-10-11 16:45:26 +11002333 struct md_rdev *rdev =
NeilBrown06f60382011-07-28 11:31:48 +10002334 rcu_dereference(conf->mirrors[i].rdev);
2335 ok = rdev_set_badblocks(rdev, sector_nr,
2336 min_bad, 0
2337 ) && ok;
2338 }
2339 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2340 *skipped = 1;
2341 put_buf(r1_bio);
2342
2343 if (!ok) {
2344 /* Cannot record the badblocks, so need to
2345 * abort the resync.
2346 * If there are multiple read targets, could just
2347 * fail the really bad ones ???
2348 */
2349 conf->recovery_disabled = mddev->recovery_disabled;
2350 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2351 return 0;
2352 } else
2353 return min_bad;
2354
2355 }
2356 if (min_bad > 0 && min_bad < good_sectors) {
2357 /* only resync enough to reach the next bad->good
2358 * transition */
2359 good_sectors = min_bad;
2360 }
2361
NeilBrown3e198f72006-01-06 00:20:21 -08002362 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2363 /* extra read targets are also write targets */
2364 write_targets += read_targets-1;
2365
2366 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 /* There is nowhere to write, so all non-sync
2368 * drives must be failed - so we are finished
2369 */
NeilBrown57afd892005-06-21 17:17:13 -07002370 sector_t rv = max_sector - sector_nr;
2371 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 return rv;
2374 }
2375
NeilBrownc6207272008-02-06 01:39:52 -08002376 if (max_sector > mddev->resync_max)
2377 max_sector = mddev->resync_max; /* Don't do IO beyond here */
NeilBrown06f60382011-07-28 11:31:48 +10002378 if (max_sector > sector_nr + good_sectors)
2379 max_sector = sector_nr + good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07002381 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 do {
2383 struct page *page;
2384 int len = PAGE_SIZE;
2385 if (sector_nr + (len>>9) > max_sector)
2386 len = (max_sector - sector_nr) << 9;
2387 if (len == 0)
2388 break;
NeilBrown6a806c52005-07-15 03:56:35 -07002389 if (sync_blocks == 0) {
2390 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de4852005-11-08 21:39:38 -08002391 &sync_blocks, still_degraded) &&
2392 !conf->fullsync &&
2393 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07002394 break;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002395 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
NeilBrown7571ae82010-10-07 11:54:46 +11002396 if ((len >> 9) > sync_blocks)
NeilBrown6a806c52005-07-15 03:56:35 -07002397 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07002398 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002399
NeilBrown8f19ccb2011-12-23 10:17:56 +11002400 for (i = 0 ; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 bio = r1_bio->bios[i];
2402 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08002403 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 if (bio_add_page(bio, page, len, 0) == 0) {
2405 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08002406 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 while (i > 0) {
2408 i--;
2409 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07002410 if (bio->bi_end_io==NULL)
2411 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 /* remove last page from this bio */
2413 bio->bi_vcnt--;
2414 bio->bi_size -= len;
2415 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
2416 }
2417 goto bio_full;
2418 }
2419 }
2420 }
2421 nr_sectors += len>>9;
2422 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07002423 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2425 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 r1_bio->sectors = nr_sectors;
2427
NeilBrownd11c1712006-01-06 00:20:26 -08002428 /* For a user-requested sync, we read all readable devices and do a
2429 * compare
2430 */
2431 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2432 atomic_set(&r1_bio->remaining, read_targets);
NeilBrown8f19ccb2011-12-23 10:17:56 +11002433 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrownd11c1712006-01-06 00:20:26 -08002434 bio = r1_bio->bios[i];
2435 if (bio->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07002436 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002437 generic_make_request(bio);
2438 }
2439 }
2440 } else {
2441 atomic_set(&r1_bio->remaining, 1);
2442 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07002443 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002444 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
NeilBrownd11c1712006-01-06 00:20:26 -08002446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 return nr_sectors;
2448}
2449
NeilBrownfd01b882011-10-11 16:47:53 +11002450static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07002451{
2452 if (sectors)
2453 return sectors;
2454
2455 return mddev->dev_sectors;
2456}
2457
NeilBrowne8096362011-10-11 16:49:05 +11002458static struct r1conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459{
NeilBrowne8096362011-10-11 16:49:05 +11002460 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002461 int i;
NeilBrown0f6d02d2011-10-11 16:48:46 +11002462 struct mirror_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11002463 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002464 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
NeilBrowne8096362011-10-11 16:49:05 +11002466 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11002468 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469
NeilBrown8f19ccb2011-12-23 10:17:56 +11002470 conf->mirrors = kzalloc(sizeof(struct mirror_info)
2471 * mddev->raid_disks * 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 GFP_KERNEL);
2473 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11002474 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475
NeilBrownddaf22a2006-01-06 00:20:19 -08002476 conf->tmppage = alloc_page(GFP_KERNEL);
2477 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11002478 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08002479
NeilBrown709ae482009-12-14 12:49:51 +11002480 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11002482 goto abort;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002483 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2485 r1bio_pool_free,
2486 conf->poolinfo);
2487 if (!conf->r1bio_pool)
NeilBrown709ae482009-12-14 12:49:51 +11002488 goto abort;
2489
NeilBrowned9bfdf2009-10-16 15:55:44 +11002490 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
NeilBrownc19d5792011-12-23 10:17:57 +11002492 err = -EINVAL;
Neil Browne7e72bf2008-05-14 16:05:54 -07002493 spin_lock_init(&conf->device_lock);
Cheng Renquan159ec1f2009-01-09 08:31:08 +11002494 list_for_each_entry(rdev, &mddev->disks, same_set) {
NeilBrown709ae482009-12-14 12:49:51 +11002495 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 if (disk_idx >= mddev->raid_disks
2497 || disk_idx < 0)
2498 continue;
NeilBrownc19d5792011-12-23 10:17:57 +11002499 if (test_bit(Replacement, &rdev->flags))
2500 disk = conf->mirrors + conf->raid_disks + disk_idx;
2501 else
2502 disk = conf->mirrors + disk_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503
NeilBrownc19d5792011-12-23 10:17:57 +11002504 if (disk->rdev)
2505 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 disk->rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
2508 disk->head_position = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 }
2510 conf->raid_disks = mddev->raid_disks;
2511 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 INIT_LIST_HEAD(&conf->retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513
2514 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08002515 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
NeilBrown191ea9b2005-06-21 17:17:23 -07002517 bio_list_init(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +11002518 conf->pending_count = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11002519 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown191ea9b2005-06-21 17:17:23 -07002520
NeilBrownc19d5792011-12-23 10:17:57 +11002521 err = -EIO;
NeilBrown709ae482009-12-14 12:49:51 +11002522 conf->last_used = -1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002523 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
2525 disk = conf->mirrors + i;
2526
NeilBrownc19d5792011-12-23 10:17:57 +11002527 if (i < conf->raid_disks &&
2528 disk[conf->raid_disks].rdev) {
2529 /* This slot has a replacement. */
2530 if (!disk->rdev) {
2531 /* No original, just make the replacement
2532 * a recovering spare
2533 */
2534 disk->rdev =
2535 disk[conf->raid_disks].rdev;
2536 disk[conf->raid_disks].rdev = NULL;
2537 } else if (!test_bit(In_sync, &disk->rdev->flags))
2538 /* Original is not in_sync - bad */
2539 goto abort;
2540 }
2541
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002542 if (!disk->rdev ||
2543 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 disk->head_position = 0;
NeilBrown918f0232007-08-22 14:01:52 -07002545 if (disk->rdev)
2546 conf->fullsync = 1;
NeilBrown709ae482009-12-14 12:49:51 +11002547 } else if (conf->last_used < 0)
2548 /*
2549 * The first working device is used as a
2550 * starting point to read balancing.
2551 */
2552 conf->last_used = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 }
NeilBrown709ae482009-12-14 12:49:51 +11002554
NeilBrown709ae482009-12-14 12:49:51 +11002555 if (conf->last_used < 0) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002556 printk(KERN_ERR "md/raid1:%s: no operational mirrors\n",
NeilBrown709ae482009-12-14 12:49:51 +11002557 mdname(mddev));
2558 goto abort;
NeilBrown11ce99e2006-10-03 01:15:52 -07002559 }
NeilBrown709ae482009-12-14 12:49:51 +11002560 err = -ENOMEM;
2561 conf->thread = md_register_thread(raid1d, mddev, NULL);
2562 if (!conf->thread) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002563 printk(KERN_ERR
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002564 "md/raid1:%s: couldn't allocate thread\n",
NeilBrown191ea9b2005-06-21 17:17:23 -07002565 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11002566 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002568
NeilBrown709ae482009-12-14 12:49:51 +11002569 return conf;
2570
2571 abort:
2572 if (conf) {
2573 if (conf->r1bio_pool)
2574 mempool_destroy(conf->r1bio_pool);
2575 kfree(conf->mirrors);
2576 safe_put_page(conf->tmppage);
2577 kfree(conf->poolinfo);
2578 kfree(conf);
2579 }
2580 return ERR_PTR(err);
2581}
2582
NeilBrownfd01b882011-10-11 16:47:53 +11002583static int run(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002584{
NeilBrowne8096362011-10-11 16:49:05 +11002585 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002586 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11002587 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002588
2589 if (mddev->level != 1) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002590 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
NeilBrown709ae482009-12-14 12:49:51 +11002591 mdname(mddev), mddev->level);
2592 return -EIO;
2593 }
2594 if (mddev->reshape_position != MaxSector) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002595 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
NeilBrown709ae482009-12-14 12:49:51 +11002596 mdname(mddev));
2597 return -EIO;
2598 }
2599 /*
2600 * copy the already verified devices into our private RAID1
2601 * bookkeeping area. [whatever we allocate in run(),
2602 * should be freed in stop()]
2603 */
2604 if (mddev->private == NULL)
2605 conf = setup_conf(mddev);
2606 else
2607 conf = mddev->private;
2608
2609 if (IS_ERR(conf))
2610 return PTR_ERR(conf);
2611
NeilBrown709ae482009-12-14 12:49:51 +11002612 list_for_each_entry(rdev, &mddev->disks, same_set) {
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002613 if (!mddev->gendisk)
2614 continue;
NeilBrown709ae482009-12-14 12:49:51 +11002615 disk_stack_limits(mddev->gendisk, rdev->bdev,
2616 rdev->data_offset << 9);
2617 /* as we don't honour merge_bvec_fn, we must never risk
NeilBrown627a2d32010-03-08 16:44:38 +11002618 * violating it, so limit ->max_segments to 1 lying within
2619 * a single page, as a one page request is never in violation.
NeilBrown709ae482009-12-14 12:49:51 +11002620 */
NeilBrown627a2d32010-03-08 16:44:38 +11002621 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
2622 blk_queue_max_segments(mddev->queue, 1);
2623 blk_queue_segment_boundary(mddev->queue,
2624 PAGE_CACHE_SIZE - 1);
2625 }
NeilBrown709ae482009-12-14 12:49:51 +11002626 }
2627
2628 mddev->degraded = 0;
2629 for (i=0; i < conf->raid_disks; i++)
2630 if (conf->mirrors[i].rdev == NULL ||
2631 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2632 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2633 mddev->degraded++;
2634
2635 if (conf->raid_disks - mddev->degraded == 1)
2636 mddev->recovery_cp = MaxSector;
2637
Andre Noll8c6ac862009-06-18 08:48:06 +10002638 if (mddev->recovery_cp != MaxSector)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002639 printk(KERN_NOTICE "md/raid1:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10002640 " -- starting background reconstruction\n",
2641 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002643 "md/raid1:%s: active with %d out of %d mirrors\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 mdname(mddev), mddev->raid_disks - mddev->degraded,
2645 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11002646
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 /*
2648 * Ok, everything is just fine now
2649 */
NeilBrown709ae482009-12-14 12:49:51 +11002650 mddev->thread = conf->thread;
2651 conf->thread = NULL;
2652 mddev->private = conf;
2653
Dan Williams1f403622009-03-31 14:59:03 +11002654 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002656 if (mddev->queue) {
2657 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2658 mddev->queue->backing_dev_info.congested_data = mddev;
2659 }
Martin K. Petersena91a2782011-03-17 11:11:05 +01002660 return md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661}
2662
NeilBrownfd01b882011-10-11 16:47:53 +11002663static int stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664{
NeilBrowne8096362011-10-11 16:49:05 +11002665 struct r1conf *conf = mddev->private;
NeilBrown4b6d2872005-09-09 16:23:47 -07002666 struct bitmap *bitmap = mddev->bitmap;
NeilBrown4b6d2872005-09-09 16:23:47 -07002667
2668 /* wait for behind writes to complete */
NeilBrowne5551902010-03-31 11:21:44 +11002669 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002670 printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n",
2671 mdname(mddev));
NeilBrown4b6d2872005-09-09 16:23:47 -07002672 /* need to kick something here to make sure I/O goes? */
NeilBrowne5551902010-03-31 11:21:44 +11002673 wait_event(bitmap->behind_wait,
2674 atomic_read(&bitmap->behind_writes) == 0);
NeilBrown4b6d2872005-09-09 16:23:47 -07002675 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
NeilBrown409c57f2009-03-31 14:39:39 +11002677 raise_barrier(conf);
2678 lower_barrier(conf);
2679
NeilBrown01f96c02011-09-21 15:30:20 +10002680 md_unregister_thread(&mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 if (conf->r1bio_pool)
2682 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002683 kfree(conf->mirrors);
2684 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 kfree(conf);
2686 mddev->private = NULL;
2687 return 0;
2688}
2689
NeilBrownfd01b882011-10-11 16:47:53 +11002690static int raid1_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691{
2692 /* no resync is happening, and there is enough space
2693 * on all devices, so we can resize.
2694 * We need to make sure resync covers any new space.
2695 * If the array is shrinking we should possibly wait until
2696 * any io in the removed space completes, but it hardly seems
2697 * worth it.
2698 */
Dan Williams1f403622009-03-31 14:59:03 +11002699 md_set_array_sectors(mddev, raid1_size(mddev, sectors, 0));
Dan Williamsb522adc2009-03-31 15:00:31 +11002700 if (mddev->array_sectors > raid1_size(mddev, sectors, 0))
2701 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10002702 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10002703 revalidate_disk(mddev->gendisk);
Dan Williamsb522adc2009-03-31 15:00:31 +11002704 if (sectors > mddev->dev_sectors &&
NeilBrownb0986362011-05-11 15:52:21 +10002705 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11002706 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2708 }
Dan Williamsb522adc2009-03-31 15:00:31 +11002709 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07002710 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 return 0;
2712}
2713
NeilBrownfd01b882011-10-11 16:47:53 +11002714static int raid1_reshape(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715{
2716 /* We need to:
2717 * 1/ resize the r1bio_pool
2718 * 2/ resize conf->mirrors
2719 *
2720 * We allocate a new r1bio_pool if we can.
2721 * Then raise a device barrier and wait until all IO stops.
2722 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07002723 *
2724 * At the same time, we "pack" the devices so that all the missing
2725 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 */
2727 mempool_t *newpool, *oldpool;
2728 struct pool_info *newpoolinfo;
NeilBrown0f6d02d2011-10-11 16:48:46 +11002729 struct mirror_info *newmirrors;
NeilBrowne8096362011-10-11 16:49:05 +11002730 struct r1conf *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08002731 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07002732 unsigned long flags;
Dan Williamsb5470dc2008-06-27 21:44:04 -07002733 int d, d2, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734
NeilBrown63c70c42006-03-27 01:18:13 -08002735 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10002736 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08002737 mddev->layout != mddev->new_layout ||
2738 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10002739 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08002740 mddev->new_layout = mddev->layout;
2741 mddev->new_level = mddev->level;
2742 return -EINVAL;
2743 }
2744
Dan Williamsb5470dc2008-06-27 21:44:04 -07002745 err = md_allow_write(mddev);
2746 if (err)
2747 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08002748
NeilBrown63c70c42006-03-27 01:18:13 -08002749 raid_disks = mddev->raid_disks + mddev->delta_disks;
2750
NeilBrown6ea9c072005-06-21 17:17:09 -07002751 if (raid_disks < conf->raid_disks) {
2752 cnt=0;
2753 for (d= 0; d < conf->raid_disks; d++)
2754 if (conf->mirrors[d].rdev)
2755 cnt++;
2756 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07002758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759
2760 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2761 if (!newpoolinfo)
2762 return -ENOMEM;
2763 newpoolinfo->mddev = mddev;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002764 newpoolinfo->raid_disks = raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765
2766 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2767 r1bio_pool_free, newpoolinfo);
2768 if (!newpool) {
2769 kfree(newpoolinfo);
2770 return -ENOMEM;
2771 }
NeilBrown8f19ccb2011-12-23 10:17:56 +11002772 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks * 2,
2773 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 if (!newmirrors) {
2775 kfree(newpoolinfo);
2776 mempool_destroy(newpool);
2777 return -ENOMEM;
2778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
NeilBrown17999be2006-01-06 00:20:12 -08002780 raise_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781
2782 /* ok, everything is stopped */
2783 oldpool = conf->r1bio_pool;
2784 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07002785
NeilBrowna88aa782007-08-22 14:01:53 -07002786 for (d = d2 = 0; d < conf->raid_disks; d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002787 struct md_rdev *rdev = conf->mirrors[d].rdev;
NeilBrowna88aa782007-08-22 14:01:53 -07002788 if (rdev && rdev->raid_disk != d2) {
Namhyung Kim36fad852011-07-27 11:00:36 +10002789 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna88aa782007-08-22 14:01:53 -07002790 rdev->raid_disk = d2;
Namhyung Kim36fad852011-07-27 11:00:36 +10002791 sysfs_unlink_rdev(mddev, rdev);
2792 if (sysfs_link_rdev(mddev, rdev))
NeilBrowna88aa782007-08-22 14:01:53 -07002793 printk(KERN_WARNING
Namhyung Kim36fad852011-07-27 11:00:36 +10002794 "md/raid1:%s: cannot register rd%d\n",
2795 mdname(mddev), rdev->raid_disk);
NeilBrown6ea9c072005-06-21 17:17:09 -07002796 }
NeilBrowna88aa782007-08-22 14:01:53 -07002797 if (rdev)
2798 newmirrors[d2++].rdev = rdev;
2799 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 kfree(conf->mirrors);
2801 conf->mirrors = newmirrors;
2802 kfree(conf->poolinfo);
2803 conf->poolinfo = newpoolinfo;
2804
NeilBrownc04be0a2006-10-03 01:15:53 -07002805 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07002807 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08002809 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
NeilBrown6ea9c072005-06-21 17:17:09 -07002811 conf->last_used = 0; /* just make sure it is in-range */
NeilBrown17999be2006-01-06 00:20:12 -08002812 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813
2814 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2815 md_wakeup_thread(mddev->thread);
2816
2817 mempool_destroy(oldpool);
2818 return 0;
2819}
2820
NeilBrownfd01b882011-10-11 16:47:53 +11002821static void raid1_quiesce(struct mddev *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07002822{
NeilBrowne8096362011-10-11 16:49:05 +11002823 struct r1conf *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07002824
2825 switch(state) {
NeilBrown6eef4b22009-12-14 12:49:51 +11002826 case 2: /* wake for suspend */
2827 wake_up(&conf->wait_barrier);
2828 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002829 case 1:
NeilBrown17999be2006-01-06 00:20:12 -08002830 raise_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002831 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002832 case 0:
NeilBrown17999be2006-01-06 00:20:12 -08002833 lower_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002834 break;
2835 }
NeilBrown36fa3062005-09-09 16:23:45 -07002836}
2837
NeilBrownfd01b882011-10-11 16:47:53 +11002838static void *raid1_takeover(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002839{
2840 /* raid1 can take over:
2841 * raid5 with 2 devices, any layout or chunk size
2842 */
2843 if (mddev->level == 5 && mddev->raid_disks == 2) {
NeilBrowne8096362011-10-11 16:49:05 +11002844 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002845 mddev->new_level = 1;
2846 mddev->new_layout = 0;
2847 mddev->new_chunk_sectors = 0;
2848 conf = setup_conf(mddev);
2849 if (!IS_ERR(conf))
2850 conf->barrier = 1;
2851 return conf;
2852 }
2853 return ERR_PTR(-EINVAL);
2854}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855
NeilBrown84fc4b52011-10-11 16:49:58 +11002856static struct md_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857{
2858 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08002859 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 .owner = THIS_MODULE,
2861 .make_request = make_request,
2862 .run = run,
2863 .stop = stop,
2864 .status = status,
2865 .error_handler = error,
2866 .hot_add_disk = raid1_add_disk,
2867 .hot_remove_disk= raid1_remove_disk,
2868 .spare_active = raid1_spare_active,
2869 .sync_request = sync_request,
2870 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07002871 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08002872 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07002873 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11002874 .takeover = raid1_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875};
2876
2877static int __init raid_init(void)
2878{
NeilBrown2604b702006-01-06 00:20:36 -08002879 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880}
2881
2882static void raid_exit(void)
2883{
NeilBrown2604b702006-01-06 00:20:36 -08002884 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885}
2886
2887module_init(raid_init);
2888module_exit(raid_exit);
2889MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11002890MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08002892MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08002893MODULE_ALIAS("md-level-1");
NeilBrown34db0cd2011-10-11 16:50:01 +11002894
2895module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);