blob: 4e3843f7d24592cc596df1fc6d6c4577a9a6f57f [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
Jonathan Brassow473e87c2012-07-31 10:03:52 +100049/* when we get a read error on a read-only array, we redirect to another
50 * device without failing the first device, or trying to over-write to
51 * correct the read error. To keep track of bad blocks on a per-bio
52 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
53 */
54#define IO_BLOCKED ((struct bio *)1)
55/* When we successfully write to a known bad-block, we need to remove the
56 * bad-block marking which must be done from process context. So we record
57 * the success by setting devs[n].bio to IO_MADE_GOOD
58 */
59#define IO_MADE_GOOD ((struct bio *)2)
60
61#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
62
NeilBrown34db0cd2011-10-11 16:50:01 +110063/* When there are this many requests queue to be written by
64 * the raid1 thread, we become 'congested' to provide back-pressure
65 * for writeback.
66 */
67static int max_queued_requests = 1024;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
majianpeng79ef3a82013-11-15 14:55:02 +080069static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
70 sector_t bi_sector);
NeilBrowne8096362011-10-11 16:49:05 +110071static void lower_barrier(struct r1conf *conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Al Virodd0fc662005-10-07 07:46:04 +010073static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 struct pool_info *pi = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +110076 int size = offsetof(struct r1bio, bios[pi->raid_disks]);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78 /* allocate a r1bio with room for raid_disks entries in the bios array */
Jens Axboe7eaceac2011-03-10 08:52:07 +010079 return kzalloc(size, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080}
81
82static void r1bio_pool_free(void *r1_bio, void *data)
83{
84 kfree(r1_bio);
85}
86
87#define RESYNC_BLOCK_SIZE (64*1024)
majianpeng8e005f72013-11-14 15:16:18 +110088#define RESYNC_DEPTH 32
Linus Torvalds1da177e2005-04-16 15:20:36 -070089#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
90#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
majianpeng8e005f72013-11-14 15:16:18 +110091#define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
92#define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +100093#define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)
94#define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
majianpeng8e005f72013-11-14 15:16:18 +110095#define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Al Virodd0fc662005-10-07 07:46:04 +010097static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
99 struct pool_info *pi = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100100 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 struct bio *bio;
NeilBrownda1aab32014-04-09 12:25:43 +1000102 int need_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 int i, j;
104
105 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100106 if (!r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
109 /*
110 * Allocate bios : 1 for reading, n-1 for writing
111 */
112 for (j = pi->raid_disks ; j-- ; ) {
NeilBrown67465572010-10-26 17:33:54 +1100113 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 if (!bio)
115 goto out_free_bio;
116 r1_bio->bios[j] = bio;
117 }
118 /*
119 * Allocate RESYNC_PAGES data pages and attach them to
NeilBrownd11c1712006-01-06 00:20:26 -0800120 * the first bio.
121 * If this is a user-requested check/repair, allocate
122 * RESYNC_PAGES for each bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 */
NeilBrownd11c1712006-01-06 00:20:26 -0800124 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
NeilBrownda1aab32014-04-09 12:25:43 +1000125 need_pages = pi->raid_disks;
NeilBrownd11c1712006-01-06 00:20:26 -0800126 else
NeilBrownda1aab32014-04-09 12:25:43 +1000127 need_pages = 1;
128 for (j = 0; j < need_pages; j++) {
NeilBrownd11c1712006-01-06 00:20:26 -0800129 bio = r1_bio->bios[j];
Kent Overstreeta0787602012-09-10 14:03:28 -0700130 bio->bi_vcnt = RESYNC_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Kent Overstreeta0787602012-09-10 14:03:28 -0700132 if (bio_alloc_pages(bio, gfp_flags))
NeilBrownda1aab32014-04-09 12:25:43 +1000133 goto out_free_pages;
NeilBrownd11c1712006-01-06 00:20:26 -0800134 }
135 /* If not user-requests, copy the page pointers to all bios */
136 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
137 for (i=0; i<RESYNC_PAGES ; i++)
138 for (j=1; j<pi->raid_disks; j++)
139 r1_bio->bios[j]->bi_io_vec[i].bv_page =
140 r1_bio->bios[0]->bi_io_vec[i].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 }
142
143 r1_bio->master_bio = NULL;
144
145 return r1_bio;
146
NeilBrownda1aab32014-04-09 12:25:43 +1000147out_free_pages:
148 while (--j >= 0) {
149 struct bio_vec *bv;
150
151 bio_for_each_segment_all(bv, r1_bio->bios[j], i)
152 __free_page(bv->bv_page);
153 }
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155out_free_bio:
NeilBrown8f19ccb2011-12-23 10:17:56 +1100156 while (++j < pi->raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 bio_put(r1_bio->bios[j]);
158 r1bio_pool_free(r1_bio, data);
159 return NULL;
160}
161
162static void r1buf_pool_free(void *__r1_bio, void *data)
163{
164 struct pool_info *pi = data;
NeilBrownd11c1712006-01-06 00:20:26 -0800165 int i,j;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100166 struct r1bio *r1bio = __r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
NeilBrownd11c1712006-01-06 00:20:26 -0800168 for (i = 0; i < RESYNC_PAGES; i++)
169 for (j = pi->raid_disks; j-- ;) {
170 if (j == 0 ||
171 r1bio->bios[j]->bi_io_vec[i].bv_page !=
172 r1bio->bios[0]->bi_io_vec[i].bv_page)
NeilBrown1345b1d2006-01-06 00:20:40 -0800173 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800174 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 for (i=0 ; i < pi->raid_disks; i++)
176 bio_put(r1bio->bios[i]);
177
178 r1bio_pool_free(r1bio, data);
179}
180
NeilBrowne8096362011-10-11 16:49:05 +1100181static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
183 int i;
184
NeilBrown8f19ccb2011-12-23 10:17:56 +1100185 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 struct bio **bio = r1_bio->bios + i;
NeilBrown4367af52011-07-28 11:31:49 +1000187 if (!BIO_SPECIAL(*bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 bio_put(*bio);
189 *bio = NULL;
190 }
191}
192
NeilBrown9f2c9d12011-10-11 16:48:43 +1100193static void free_r1bio(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
NeilBrowne8096362011-10-11 16:49:05 +1100195 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 put_all_bios(conf, r1_bio);
198 mempool_free(r1_bio, conf->r1bio_pool);
199}
200
NeilBrown9f2c9d12011-10-11 16:48:43 +1100201static void put_buf(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
NeilBrowne8096362011-10-11 16:49:05 +1100203 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown3e198f72006-01-06 00:20:21 -0800204 int i;
205
NeilBrown8f19ccb2011-12-23 10:17:56 +1100206 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3e198f72006-01-06 00:20:21 -0800207 struct bio *bio = r1_bio->bios[i];
208 if (bio->bi_end_io)
209 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212 mempool_free(r1_bio, conf->r1buf_pool);
213
NeilBrown17999be2006-01-06 00:20:12 -0800214 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
NeilBrown9f2c9d12011-10-11 16:48:43 +1100217static void reschedule_retry(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
219 unsigned long flags;
NeilBrownfd01b882011-10-11 16:47:53 +1100220 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +1100221 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
223 spin_lock_irqsave(&conf->device_lock, flags);
224 list_add(&r1_bio->retry_list, &conf->retry_list);
NeilBrownddaf22a2006-01-06 00:20:19 -0800225 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 spin_unlock_irqrestore(&conf->device_lock, flags);
227
NeilBrown17999be2006-01-06 00:20:12 -0800228 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 md_wakeup_thread(mddev->thread);
230}
231
232/*
233 * raid_end_bio_io() is called when we have finished servicing a mirrored
234 * operation and are ready to return a success/failure code to the buffer
235 * cache layer.
236 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100237static void call_bio_endio(struct r1bio *r1_bio)
NeilBrownd2eb35a2011-07-28 11:31:48 +1000238{
239 struct bio *bio = r1_bio->master_bio;
240 int done;
NeilBrowne8096362011-10-11 16:49:05 +1100241 struct r1conf *conf = r1_bio->mddev->private;
majianpeng79ef3a82013-11-15 14:55:02 +0800242 sector_t start_next_window = r1_bio->start_next_window;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700243 sector_t bi_sector = bio->bi_iter.bi_sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000244
245 if (bio->bi_phys_segments) {
246 unsigned long flags;
247 spin_lock_irqsave(&conf->device_lock, flags);
248 bio->bi_phys_segments--;
249 done = (bio->bi_phys_segments == 0);
250 spin_unlock_irqrestore(&conf->device_lock, flags);
majianpeng79ef3a82013-11-15 14:55:02 +0800251 /*
252 * make_request() might be waiting for
253 * bi_phys_segments to decrease
254 */
255 wake_up(&conf->wait_barrier);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000256 } else
257 done = 1;
258
259 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200260 bio->bi_error = -EIO;
261
NeilBrownd2eb35a2011-07-28 11:31:48 +1000262 if (done) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200263 bio_endio(bio);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000264 /*
265 * Wake up any possible resync thread that waits for the device
266 * to go idle.
267 */
majianpeng79ef3a82013-11-15 14:55:02 +0800268 allow_barrier(conf, start_next_window, bi_sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000269 }
270}
271
NeilBrown9f2c9d12011-10-11 16:48:43 +1100272static void raid_end_bio_io(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 struct bio *bio = r1_bio->master_bio;
275
NeilBrown4b6d2872005-09-09 16:23:47 -0700276 /* if nobody has done the final endio yet, do it now */
277 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
NeilBrown36a4e1f2011-10-07 14:23:17 +1100278 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
279 (bio_data_dir(bio) == WRITE) ? "write" : "read",
Kent Overstreet4f024f32013-10-11 15:44:27 -0700280 (unsigned long long) bio->bi_iter.bi_sector,
281 (unsigned long long) bio_end_sector(bio) - 1);
NeilBrown4b6d2872005-09-09 16:23:47 -0700282
NeilBrownd2eb35a2011-07-28 11:31:48 +1000283 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 free_r1bio(r1_bio);
286}
287
288/*
289 * Update disk head position estimator based on IRQ completion info.
290 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100291static inline void update_head_pos(int disk, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
NeilBrowne8096362011-10-11 16:49:05 +1100293 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
295 conf->mirrors[disk].head_position =
296 r1_bio->sector + (r1_bio->sectors);
297}
298
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100299/*
300 * Find the disk number which triggered given bio
301 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100302static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100303{
304 int mirror;
NeilBrown30194632011-12-23 10:17:56 +1100305 struct r1conf *conf = r1_bio->mddev->private;
306 int raid_disks = conf->raid_disks;
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100307
NeilBrown8f19ccb2011-12-23 10:17:56 +1100308 for (mirror = 0; mirror < raid_disks * 2; mirror++)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100309 if (r1_bio->bios[mirror] == bio)
310 break;
311
NeilBrown8f19ccb2011-12-23 10:17:56 +1100312 BUG_ON(mirror == raid_disks * 2);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100313 update_head_pos(mirror, r1_bio);
314
315 return mirror;
316}
317
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200318static void raid1_end_read_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200320 int uptodate = !bio->bi_error;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100321 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 int mirror;
NeilBrowne8096362011-10-11 16:49:05 +1100323 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 mirror = r1_bio->read_disk;
326 /*
327 * this branch is our 'one mirror IO has finished' event handler:
328 */
NeilBrownddaf22a2006-01-06 00:20:19 -0800329 update_head_pos(mirror, r1_bio);
330
NeilBrowndd00a992007-05-10 03:15:50 -0700331 if (uptodate)
332 set_bit(R1BIO_Uptodate, &r1_bio->state);
333 else {
334 /* If all other devices have failed, we want to return
335 * the error upwards rather than fail the last device.
336 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 */
NeilBrowndd00a992007-05-10 03:15:50 -0700338 unsigned long flags;
339 spin_lock_irqsave(&conf->device_lock, flags);
340 if (r1_bio->mddev->degraded == conf->raid_disks ||
341 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
NeilBrown34cab6f2015-07-24 09:22:16 +1000342 test_bit(In_sync, &conf->mirrors[mirror].rdev->flags)))
NeilBrowndd00a992007-05-10 03:15:50 -0700343 uptodate = 1;
344 spin_unlock_irqrestore(&conf->device_lock, flags);
345 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100347 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 raid_end_bio_io(r1_bio);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100349 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
350 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 /*
352 * oops, read error:
353 */
354 char b[BDEVNAME_SIZE];
Christian Dietrich8bda4702011-07-27 11:00:36 +1000355 printk_ratelimited(
356 KERN_ERR "md/raid1:%s: %s: "
357 "rescheduling sector %llu\n",
358 mdname(conf->mddev),
359 bdevname(conf->mirrors[mirror].rdev->bdev,
360 b),
361 (unsigned long long)r1_bio->sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000362 set_bit(R1BIO_ReadError, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 reschedule_retry(r1_bio);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100364 /* don't drop the reference on read_disk yet */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366}
367
NeilBrown9f2c9d12011-10-11 16:48:43 +1100368static void close_write(struct r1bio *r1_bio)
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000369{
370 /* it really is the end of this request */
371 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
372 /* free extra copy of the data pages */
373 int i = r1_bio->behind_page_count;
374 while (i--)
375 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
376 kfree(r1_bio->behind_bvecs);
377 r1_bio->behind_bvecs = NULL;
378 }
379 /* clear the bitmap if all writes complete successfully */
380 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
381 r1_bio->sectors,
382 !test_bit(R1BIO_Degraded, &r1_bio->state),
383 test_bit(R1BIO_BehindIO, &r1_bio->state));
384 md_write_end(r1_bio->mddev);
385}
386
NeilBrown9f2c9d12011-10-11 16:48:43 +1100387static void r1_bio_write_done(struct r1bio *r1_bio)
NeilBrown4e780642010-10-19 12:54:01 +1100388{
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000389 if (!atomic_dec_and_test(&r1_bio->remaining))
390 return;
391
392 if (test_bit(R1BIO_WriteError, &r1_bio->state))
393 reschedule_retry(r1_bio);
394 else {
395 close_write(r1_bio);
NeilBrown4367af52011-07-28 11:31:49 +1000396 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
397 reschedule_retry(r1_bio);
398 else
399 raid_end_bio_io(r1_bio);
NeilBrown4e780642010-10-19 12:54:01 +1100400 }
401}
402
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200403static void raid1_end_write_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
NeilBrown9f2c9d12011-10-11 16:48:43 +1100405 struct r1bio *r1_bio = bio->bi_private;
NeilBrowna9701a32005-11-08 21:39:34 -0800406 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
NeilBrowne8096362011-10-11 16:49:05 +1100407 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown04b857f2006-03-09 17:33:46 -0800408 struct bio *to_put = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100410 mirror = find_bio_disk(r1_bio, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Tejun Heoe9c74692010-09-03 11:56:18 +0200412 /*
413 * 'one mirror IO has finished' event handler:
414 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200415 if (bio->bi_error) {
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000416 set_bit(WriteErrorSeen,
417 &conf->mirrors[mirror].rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +1100418 if (!test_and_set_bit(WantReplacement,
419 &conf->mirrors[mirror].rdev->flags))
420 set_bit(MD_RECOVERY_NEEDED, &
421 conf->mddev->recovery);
422
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000423 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +1000424 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 /*
Tejun Heoe9c74692010-09-03 11:56:18 +0200426 * Set R1BIO_Uptodate in our master bio, so that we
427 * will return a good error code for to the higher
428 * levels even if IO on some other mirrored buffer
429 * fails.
430 *
431 * The 'master' represents the composite IO operation
432 * to user-side. So if something waits for IO, then it
433 * will wait for the 'master' bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 */
NeilBrown4367af52011-07-28 11:31:49 +1000435 sector_t first_bad;
436 int bad_sectors;
437
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000438 r1_bio->bios[mirror] = NULL;
439 to_put = bio;
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300440 /*
441 * Do not set R1BIO_Uptodate if the current device is
442 * rebuilding or Faulty. This is because we cannot use
443 * such device for properly reading the data back (we could
444 * potentially use it, if the current write would have felt
445 * before rdev->recovery_offset, but for simplicity we don't
446 * check this here.
447 */
448 if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) &&
449 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))
450 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
NeilBrown4367af52011-07-28 11:31:49 +1000452 /* Maybe we can clear some bad blocks. */
453 if (is_badblock(conf->mirrors[mirror].rdev,
454 r1_bio->sector, r1_bio->sectors,
455 &first_bad, &bad_sectors)) {
456 r1_bio->bios[mirror] = IO_MADE_GOOD;
457 set_bit(R1BIO_MadeGood, &r1_bio->state);
458 }
459 }
460
Tejun Heoe9c74692010-09-03 11:56:18 +0200461 if (behind) {
462 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
463 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700464
Tejun Heoe9c74692010-09-03 11:56:18 +0200465 /*
466 * In behind mode, we ACK the master bio once the I/O
467 * has safely reached all non-writemostly
468 * disks. Setting the Returned bit ensures that this
469 * gets done only once -- we don't ever want to return
470 * -EIO here, instead we'll wait
471 */
472 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
473 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
474 /* Maybe we can return now */
475 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
476 struct bio *mbio = r1_bio->master_bio;
NeilBrown36a4e1f2011-10-07 14:23:17 +1100477 pr_debug("raid1: behind end write sectors"
478 " %llu-%llu\n",
Kent Overstreet4f024f32013-10-11 15:44:27 -0700479 (unsigned long long) mbio->bi_iter.bi_sector,
480 (unsigned long long) bio_end_sector(mbio) - 1);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000481 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700482 }
483 }
484 }
NeilBrown4367af52011-07-28 11:31:49 +1000485 if (r1_bio->bios[mirror] == NULL)
486 rdev_dec_pending(conf->mirrors[mirror].rdev,
487 conf->mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200488
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 * Let's see if all mirrored write operations have finished
491 * already.
492 */
NeilBrownaf6d7b72011-05-11 14:51:19 +1000493 r1_bio_write_done(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700494
NeilBrown04b857f2006-03-09 17:33:46 -0800495 if (to_put)
496 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499/*
500 * This routine returns the disk from which the requested read should
501 * be done. There is a per-array 'next expected sequential IO' sector
502 * number - if this matches on the next IO then we use the last disk.
503 * There is also a per-disk 'last know head position' sector that is
504 * maintained from IRQ contexts, both the normal and the resync IO
505 * completion handlers update this position correctly. If there is no
506 * perfect sequential match then we pick the disk whose head is closest.
507 *
508 * If there are 2 mirrors in the same 2 devices, performance degrades
509 * because position is mirror, not device based.
510 *
511 * The rdev for the device selected will have nr_pending incremented.
512 */
NeilBrowne8096362011-10-11 16:49:05 +1100513static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000515 const sector_t this_sector = r1_bio->sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000516 int sectors;
517 int best_good_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000518 int best_disk, best_dist_disk, best_pending_disk;
519 int has_nonrot_disk;
Shaohua Libe4d3282012-07-31 10:03:53 +1000520 int disk;
NeilBrown76073052011-05-11 14:34:56 +1000521 sector_t best_dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000522 unsigned int min_pending;
NeilBrown3cb03002011-10-11 16:45:26 +1100523 struct md_rdev *rdev;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000524 int choose_first;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000525 int choose_next_idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
527 rcu_read_lock();
528 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700529 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 * device if no resync is going on, or below the resync window.
531 * We take the first readable disk when above the resync window.
532 */
533 retry:
NeilBrownd2eb35a2011-07-28 11:31:48 +1000534 sectors = r1_bio->sectors;
NeilBrown76073052011-05-11 14:34:56 +1000535 best_disk = -1;
Shaohua Li9dedf602012-07-31 10:03:53 +1000536 best_dist_disk = -1;
NeilBrown76073052011-05-11 14:34:56 +1000537 best_dist = MaxSector;
Shaohua Li9dedf602012-07-31 10:03:53 +1000538 best_pending_disk = -1;
539 min_pending = UINT_MAX;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000540 best_good_sectors = 0;
Shaohua Li9dedf602012-07-31 10:03:53 +1000541 has_nonrot_disk = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000542 choose_next_idle = 0;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000543
Goldwyn Rodrigues7d49ffc2014-08-12 10:13:19 -0500544 if ((conf->mddev->recovery_cp < this_sector + sectors) ||
545 (mddev_is_clustered(conf->mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -0500546 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
Goldwyn Rodrigues7d49ffc2014-08-12 10:13:19 -0500547 this_sector + sectors)))
548 choose_first = 1;
549 else
550 choose_first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Shaohua Libe4d3282012-07-31 10:03:53 +1000552 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
NeilBrown76073052011-05-11 14:34:56 +1000553 sector_t dist;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000554 sector_t first_bad;
555 int bad_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000556 unsigned int pending;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000557 bool nonrot;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000558
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000559 rdev = rcu_dereference(conf->mirrors[disk].rdev);
560 if (r1_bio->bios[disk] == IO_BLOCKED
561 || rdev == NULL
NeilBrown76073052011-05-11 14:34:56 +1000562 || test_bit(Faulty, &rdev->flags))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000563 continue;
NeilBrown76073052011-05-11 14:34:56 +1000564 if (!test_bit(In_sync, &rdev->flags) &&
565 rdev->recovery_offset < this_sector + sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 continue;
NeilBrown76073052011-05-11 14:34:56 +1000567 if (test_bit(WriteMostly, &rdev->flags)) {
568 /* Don't balance among write-mostly, just
569 * use the first as a last resort */
Tomáš Hodekd1901ef2015-02-23 11:00:38 +1100570 if (best_dist_disk < 0) {
NeilBrown307729c2012-01-09 01:41:51 +1100571 if (is_badblock(rdev, this_sector, sectors,
572 &first_bad, &bad_sectors)) {
573 if (first_bad < this_sector)
574 /* Cannot use this */
575 continue;
576 best_good_sectors = first_bad - this_sector;
577 } else
578 best_good_sectors = sectors;
Tomáš Hodekd1901ef2015-02-23 11:00:38 +1100579 best_dist_disk = disk;
580 best_pending_disk = disk;
NeilBrown307729c2012-01-09 01:41:51 +1100581 }
NeilBrown76073052011-05-11 14:34:56 +1000582 continue;
583 }
584 /* This is a reasonable device to use. It might
585 * even be best.
586 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000587 if (is_badblock(rdev, this_sector, sectors,
588 &first_bad, &bad_sectors)) {
589 if (best_dist < MaxSector)
590 /* already have a better device */
591 continue;
592 if (first_bad <= this_sector) {
593 /* cannot read here. If this is the 'primary'
594 * device, then we must not read beyond
595 * bad_sectors from another device..
596 */
597 bad_sectors -= (this_sector - first_bad);
598 if (choose_first && sectors > bad_sectors)
599 sectors = bad_sectors;
600 if (best_good_sectors > sectors)
601 best_good_sectors = sectors;
602
603 } else {
604 sector_t good_sectors = first_bad - this_sector;
605 if (good_sectors > best_good_sectors) {
606 best_good_sectors = good_sectors;
607 best_disk = disk;
608 }
609 if (choose_first)
610 break;
611 }
612 continue;
613 } else
614 best_good_sectors = sectors;
615
Shaohua Li12cee5a2012-07-31 10:03:53 +1000616 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
617 has_nonrot_disk |= nonrot;
Shaohua Li9dedf602012-07-31 10:03:53 +1000618 pending = atomic_read(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000619 dist = abs(this_sector - conf->mirrors[disk].head_position);
Shaohua Li12cee5a2012-07-31 10:03:53 +1000620 if (choose_first) {
NeilBrown76073052011-05-11 14:34:56 +1000621 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 break;
623 }
Shaohua Li12cee5a2012-07-31 10:03:53 +1000624 /* Don't change to another disk for sequential reads */
625 if (conf->mirrors[disk].next_seq_sect == this_sector
626 || dist == 0) {
627 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
628 struct raid1_info *mirror = &conf->mirrors[disk];
629
630 best_disk = disk;
631 /*
632 * If buffered sequential IO size exceeds optimal
633 * iosize, check if there is idle disk. If yes, choose
634 * the idle disk. read_balance could already choose an
635 * idle disk before noticing it's a sequential IO in
636 * this disk. This doesn't matter because this disk
637 * will idle, next time it will be utilized after the
638 * first disk has IO size exceeds optimal iosize. In
639 * this way, iosize of the first disk will be optimal
640 * iosize at least. iosize of the second disk might be
641 * small, but not a big deal since when the second disk
642 * starts IO, the first disk is likely still busy.
643 */
644 if (nonrot && opt_iosize > 0 &&
645 mirror->seq_start != MaxSector &&
646 mirror->next_seq_sect > opt_iosize &&
647 mirror->next_seq_sect - opt_iosize >=
648 mirror->seq_start) {
649 choose_next_idle = 1;
650 continue;
651 }
652 break;
653 }
654 /* If device is idle, use it */
655 if (pending == 0) {
656 best_disk = disk;
657 break;
658 }
659
660 if (choose_next_idle)
661 continue;
Shaohua Li9dedf602012-07-31 10:03:53 +1000662
663 if (min_pending > pending) {
664 min_pending = pending;
665 best_pending_disk = disk;
666 }
667
NeilBrown76073052011-05-11 14:34:56 +1000668 if (dist < best_dist) {
669 best_dist = dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000670 best_dist_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 }
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Shaohua Li9dedf602012-07-31 10:03:53 +1000674 /*
675 * If all disks are rotational, choose the closest disk. If any disk is
676 * non-rotational, choose the disk with less pending request even the
677 * disk is rotational, which might/might not be optimal for raids with
678 * mixed ratation/non-rotational disks depending on workload.
679 */
680 if (best_disk == -1) {
681 if (has_nonrot_disk)
682 best_disk = best_pending_disk;
683 else
684 best_disk = best_dist_disk;
685 }
686
NeilBrown76073052011-05-11 14:34:56 +1000687 if (best_disk >= 0) {
688 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700689 if (!rdev)
690 goto retry;
691 atomic_inc(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000692 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 /* cannot risk returning a device that failed
694 * before we inc'ed nr_pending
695 */
NeilBrown03c902e2006-01-06 00:20:46 -0800696 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 goto retry;
698 }
NeilBrownd2eb35a2011-07-28 11:31:48 +1000699 sectors = best_good_sectors;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000700
701 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
702 conf->mirrors[best_disk].seq_start = this_sector;
703
Shaohua Libe4d3282012-07-31 10:03:53 +1000704 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 }
706 rcu_read_unlock();
NeilBrownd2eb35a2011-07-28 11:31:48 +1000707 *max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
NeilBrown76073052011-05-11 14:34:56 +1000709 return best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
NeilBrown5c675f82014-12-15 12:56:56 +1100712static int raid1_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700713{
NeilBrowne8096362011-10-11 16:49:05 +1100714 struct r1conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700715 int i, ret = 0;
716
Tejun Heo44522262015-05-22 17:13:26 -0400717 if ((bits & (1 << WB_async_congested)) &&
NeilBrown34db0cd2011-10-11 16:50:01 +1100718 conf->pending_count >= max_queued_requests)
719 return 1;
720
NeilBrown0d129222006-10-03 01:15:54 -0700721 rcu_read_lock();
NeilBrownf53e29f2012-02-13 14:24:05 +1100722 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100723 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700724 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200725 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700726
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500727 BUG_ON(!q);
728
NeilBrown0d129222006-10-03 01:15:54 -0700729 /* Note the '|| 1' - when read_balance prefers
730 * non-congested targets, it can be removed
731 */
Tejun Heo44522262015-05-22 17:13:26 -0400732 if ((bits & (1 << WB_async_congested)) || 1)
NeilBrown0d129222006-10-03 01:15:54 -0700733 ret |= bdi_congested(&q->backing_dev_info, bits);
734 else
735 ret &= bdi_congested(&q->backing_dev_info, bits);
736 }
737 }
738 rcu_read_unlock();
739 return ret;
740}
NeilBrown0d129222006-10-03 01:15:54 -0700741
NeilBrowne8096362011-10-11 16:49:05 +1100742static void flush_pending_writes(struct r1conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800743{
744 /* Any writes that have been queued but are awaiting
745 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800746 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800747 spin_lock_irq(&conf->device_lock);
748
749 if (conf->pending_bio_list.head) {
750 struct bio *bio;
751 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100752 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800753 spin_unlock_irq(&conf->device_lock);
754 /* flush any pending bitmap writes to
755 * disk before proceeding w/ I/O */
756 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100757 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800758
759 while (bio) { /* submit pending writes */
760 struct bio *next = bio->bi_next;
761 bio->bi_next = NULL;
Shaohua Li2ff8cc22012-10-11 13:28:54 +1100762 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
763 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
764 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200765 bio_endio(bio);
Shaohua Li2ff8cc22012-10-11 13:28:54 +1100766 else
767 generic_make_request(bio);
NeilBrowna35e63e2008-03-04 14:29:29 -0800768 bio = next;
769 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800770 } else
771 spin_unlock_irq(&conf->device_lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100772}
773
NeilBrown17999be2006-01-06 00:20:12 -0800774/* Barriers....
775 * Sometimes we need to suspend IO while we do something else,
776 * either some resync/recovery, or reconfigure the array.
777 * To do this we raise a 'barrier'.
778 * The 'barrier' is a counter that can be raised multiple times
779 * to count how many activities are happening which preclude
780 * normal IO.
781 * We can only raise the barrier if there is no pending IO.
782 * i.e. if nr_pending == 0.
783 * We choose only to raise the barrier if no-one is waiting for the
784 * barrier to go down. This means that as soon as an IO request
785 * is ready, no other operations which require a barrier will start
786 * until the IO request has had a chance.
787 *
788 * So: regular IO calls 'wait_barrier'. When that returns there
789 * is no backgroup IO happening, It must arrange to call
790 * allow_barrier when it has finished its IO.
791 * backgroup IO calls must call raise_barrier. Once that returns
792 * there is no normal IO happeing. It must arrange to call
793 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 */
NeilBrownc2fd4c92014-09-10 16:01:24 +1000795static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796{
797 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800798
799 /* Wait until no block IO is waiting */
800 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
Lukas Czernereed8c022012-11-30 11:42:40 +0100801 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800802
803 /* block any new IO from starting */
804 conf->barrier++;
NeilBrownc2fd4c92014-09-10 16:01:24 +1000805 conf->next_resync = sector_nr;
NeilBrown17999be2006-01-06 00:20:12 -0800806
majianpeng79ef3a82013-11-15 14:55:02 +0800807 /* For these conditions we must wait:
808 * A: while the array is in frozen state
809 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
810 * the max count which allowed.
811 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
812 * next resync will reach to the window which normal bios are
813 * handling.
NeilBrown2f73d3c2014-09-10 15:01:49 +1000814 * D: while there are any active requests in the current window.
majianpeng79ef3a82013-11-15 14:55:02 +0800815 */
NeilBrown17999be2006-01-06 00:20:12 -0800816 wait_event_lock_irq(conf->wait_barrier,
majianpengb364e3d2013-11-14 15:16:18 +1100817 !conf->array_frozen &&
majianpeng79ef3a82013-11-15 14:55:02 +0800818 conf->barrier < RESYNC_DEPTH &&
NeilBrown2f73d3c2014-09-10 15:01:49 +1000819 conf->current_window_requests == 0 &&
majianpeng79ef3a82013-11-15 14:55:02 +0800820 (conf->start_next_window >=
821 conf->next_resync + RESYNC_SECTORS),
Lukas Czernereed8c022012-11-30 11:42:40 +0100822 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800823
NeilBrown34e97f12014-09-16 12:14:14 +1000824 conf->nr_pending++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 spin_unlock_irq(&conf->resync_lock);
826}
827
NeilBrowne8096362011-10-11 16:49:05 +1100828static void lower_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800829{
830 unsigned long flags;
NeilBrown709ae482009-12-14 12:49:51 +1100831 BUG_ON(conf->barrier <= 0);
NeilBrown17999be2006-01-06 00:20:12 -0800832 spin_lock_irqsave(&conf->resync_lock, flags);
833 conf->barrier--;
NeilBrown34e97f12014-09-16 12:14:14 +1000834 conf->nr_pending--;
NeilBrown17999be2006-01-06 00:20:12 -0800835 spin_unlock_irqrestore(&conf->resync_lock, flags);
836 wake_up(&conf->wait_barrier);
837}
838
majianpeng79ef3a82013-11-15 14:55:02 +0800839static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
NeilBrown17999be2006-01-06 00:20:12 -0800840{
majianpeng79ef3a82013-11-15 14:55:02 +0800841 bool wait = false;
842
843 if (conf->array_frozen || !bio)
844 wait = true;
845 else if (conf->barrier && bio_data_dir(bio) == WRITE) {
NeilBrown23554962014-09-10 15:56:57 +1000846 if ((conf->mddev->curr_resync_completed
847 >= bio_end_sector(bio)) ||
848 (conf->next_resync + NEXT_NORMALIO_DISTANCE
849 <= bio->bi_iter.bi_sector))
majianpeng79ef3a82013-11-15 14:55:02 +0800850 wait = false;
851 else
852 wait = true;
853 }
854
855 return wait;
856}
857
858static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
859{
860 sector_t sector = 0;
861
NeilBrown17999be2006-01-06 00:20:12 -0800862 spin_lock_irq(&conf->resync_lock);
majianpeng79ef3a82013-11-15 14:55:02 +0800863 if (need_to_wait_for_sync(conf, bio)) {
NeilBrown17999be2006-01-06 00:20:12 -0800864 conf->nr_waiting++;
NeilBrownd6b42dc2012-03-19 12:46:38 +1100865 /* Wait for the barrier to drop.
866 * However if there are already pending
867 * requests (preventing the barrier from
868 * rising completely), and the
NeilBrown5965b642014-09-04 15:51:44 +1000869 * per-process bio queue isn't empty,
NeilBrownd6b42dc2012-03-19 12:46:38 +1100870 * then don't wait, as we need to empty
NeilBrown5965b642014-09-04 15:51:44 +1000871 * that queue to allow conf->start_next_window
872 * to increase.
NeilBrownd6b42dc2012-03-19 12:46:38 +1100873 */
874 wait_event_lock_irq(conf->wait_barrier,
majianpengb364e3d2013-11-14 15:16:18 +1100875 !conf->array_frozen &&
876 (!conf->barrier ||
NeilBrown5965b642014-09-04 15:51:44 +1000877 ((conf->start_next_window <
878 conf->next_resync + RESYNC_SECTORS) &&
879 current->bio_list &&
880 !bio_list_empty(current->bio_list))),
Lukas Czernereed8c022012-11-30 11:42:40 +0100881 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800882 conf->nr_waiting--;
883 }
majianpeng79ef3a82013-11-15 14:55:02 +0800884
885 if (bio && bio_data_dir(bio) == WRITE) {
Jes Sorensene8ff8bf2015-09-16 10:20:05 -0400886 if (bio->bi_iter.bi_sector >= conf->next_resync) {
majianpeng79ef3a82013-11-15 14:55:02 +0800887 if (conf->start_next_window == MaxSector)
888 conf->start_next_window =
889 conf->next_resync +
890 NEXT_NORMALIO_DISTANCE;
891
892 if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
Kent Overstreet4f024f32013-10-11 15:44:27 -0700893 <= bio->bi_iter.bi_sector)
majianpeng79ef3a82013-11-15 14:55:02 +0800894 conf->next_window_requests++;
895 else
896 conf->current_window_requests++;
majianpeng79ef3a82013-11-15 14:55:02 +0800897 sector = conf->start_next_window;
NeilBrown41a336e2014-01-14 11:56:14 +1100898 }
majianpeng79ef3a82013-11-15 14:55:02 +0800899 }
900
NeilBrown17999be2006-01-06 00:20:12 -0800901 conf->nr_pending++;
902 spin_unlock_irq(&conf->resync_lock);
majianpeng79ef3a82013-11-15 14:55:02 +0800903 return sector;
NeilBrown17999be2006-01-06 00:20:12 -0800904}
905
majianpeng79ef3a82013-11-15 14:55:02 +0800906static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
907 sector_t bi_sector)
NeilBrown17999be2006-01-06 00:20:12 -0800908{
909 unsigned long flags;
majianpeng79ef3a82013-11-15 14:55:02 +0800910
NeilBrown17999be2006-01-06 00:20:12 -0800911 spin_lock_irqsave(&conf->resync_lock, flags);
912 conf->nr_pending--;
majianpeng79ef3a82013-11-15 14:55:02 +0800913 if (start_next_window) {
914 if (start_next_window == conf->start_next_window) {
915 if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
916 <= bi_sector)
917 conf->next_window_requests--;
918 else
919 conf->current_window_requests--;
920 } else
921 conf->current_window_requests--;
922
923 if (!conf->current_window_requests) {
924 if (conf->next_window_requests) {
925 conf->current_window_requests =
926 conf->next_window_requests;
927 conf->next_window_requests = 0;
928 conf->start_next_window +=
929 NEXT_NORMALIO_DISTANCE;
930 } else
931 conf->start_next_window = MaxSector;
932 }
933 }
NeilBrown17999be2006-01-06 00:20:12 -0800934 spin_unlock_irqrestore(&conf->resync_lock, flags);
935 wake_up(&conf->wait_barrier);
936}
937
NeilBrowne2d59922013-06-12 11:01:22 +1000938static void freeze_array(struct r1conf *conf, int extra)
NeilBrownddaf22a2006-01-06 00:20:19 -0800939{
940 /* stop syncio and normal IO and wait for everything to
941 * go quite.
majianpengb364e3d2013-11-14 15:16:18 +1100942 * We wait until nr_pending match nr_queued+extra
NeilBrown1c830532008-03-04 14:29:35 -0800943 * This is called in the context of one normal IO request
944 * that has failed. Thus any sync request that might be pending
945 * will be blocked by nr_pending, and we need to wait for
946 * pending IO requests to complete or be queued for re-try.
NeilBrowne2d59922013-06-12 11:01:22 +1000947 * Thus the number queued (nr_queued) plus this request (extra)
NeilBrown1c830532008-03-04 14:29:35 -0800948 * must match the number of pending IOs (nr_pending) before
949 * we continue.
NeilBrownddaf22a2006-01-06 00:20:19 -0800950 */
951 spin_lock_irq(&conf->resync_lock);
majianpengb364e3d2013-11-14 15:16:18 +1100952 conf->array_frozen = 1;
Lukas Czernereed8c022012-11-30 11:42:40 +0100953 wait_event_lock_irq_cmd(conf->wait_barrier,
NeilBrowne2d59922013-06-12 11:01:22 +1000954 conf->nr_pending == conf->nr_queued+extra,
Lukas Czernereed8c022012-11-30 11:42:40 +0100955 conf->resync_lock,
956 flush_pending_writes(conf));
NeilBrownddaf22a2006-01-06 00:20:19 -0800957 spin_unlock_irq(&conf->resync_lock);
958}
NeilBrowne8096362011-10-11 16:49:05 +1100959static void unfreeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800960{
961 /* reverse the effect of the freeze */
962 spin_lock_irq(&conf->resync_lock);
majianpengb364e3d2013-11-14 15:16:18 +1100963 conf->array_frozen = 0;
NeilBrownddaf22a2006-01-06 00:20:19 -0800964 wake_up(&conf->wait_barrier);
965 spin_unlock_irq(&conf->resync_lock);
966}
967
NeilBrownf72ffdd2014-09-30 14:23:59 +1000968/* duplicate the data pages for behind I/O
NeilBrown4e780642010-10-19 12:54:01 +1100969 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100970static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
NeilBrown4b6d2872005-09-09 16:23:47 -0700971{
972 int i;
973 struct bio_vec *bvec;
NeilBrown2ca68f52011-07-28 11:32:10 +1000974 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
NeilBrown4b6d2872005-09-09 16:23:47 -0700975 GFP_NOIO);
NeilBrown2ca68f52011-07-28 11:32:10 +1000976 if (unlikely(!bvecs))
NeilBrownaf6d7b72011-05-11 14:51:19 +1000977 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700978
Kent Overstreetcb34e052012-09-05 15:22:02 -0700979 bio_for_each_segment_all(bvec, bio, i) {
NeilBrown2ca68f52011-07-28 11:32:10 +1000980 bvecs[i] = *bvec;
981 bvecs[i].bv_page = alloc_page(GFP_NOIO);
982 if (unlikely(!bvecs[i].bv_page))
NeilBrown4b6d2872005-09-09 16:23:47 -0700983 goto do_sync_io;
NeilBrown2ca68f52011-07-28 11:32:10 +1000984 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
985 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
986 kunmap(bvecs[i].bv_page);
NeilBrown4b6d2872005-09-09 16:23:47 -0700987 kunmap(bvec->bv_page);
988 }
NeilBrown2ca68f52011-07-28 11:32:10 +1000989 r1_bio->behind_bvecs = bvecs;
NeilBrownaf6d7b72011-05-11 14:51:19 +1000990 r1_bio->behind_page_count = bio->bi_vcnt;
991 set_bit(R1BIO_BehindIO, &r1_bio->state);
992 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700993
994do_sync_io:
NeilBrownaf6d7b72011-05-11 14:51:19 +1000995 for (i = 0; i < bio->bi_vcnt; i++)
NeilBrown2ca68f52011-07-28 11:32:10 +1000996 if (bvecs[i].bv_page)
997 put_page(bvecs[i].bv_page);
998 kfree(bvecs);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700999 pr_debug("%dB behind alloc failed, doing sync I/O\n",
1000 bio->bi_iter.bi_size);
NeilBrown4b6d2872005-09-09 16:23:47 -07001001}
1002
NeilBrownf54a9d02012-08-02 08:33:20 +10001003struct raid1_plug_cb {
1004 struct blk_plug_cb cb;
1005 struct bio_list pending;
1006 int pending_cnt;
1007};
1008
1009static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
1010{
1011 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1012 cb);
1013 struct mddev *mddev = plug->cb.data;
1014 struct r1conf *conf = mddev->private;
1015 struct bio *bio;
1016
NeilBrown874807a2012-11-27 12:14:40 +11001017 if (from_schedule || current->bio_list) {
NeilBrownf54a9d02012-08-02 08:33:20 +10001018 spin_lock_irq(&conf->device_lock);
1019 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1020 conf->pending_count += plug->pending_cnt;
1021 spin_unlock_irq(&conf->device_lock);
NeilBrownee0b0242013-02-25 12:38:29 +11001022 wake_up(&conf->wait_barrier);
NeilBrownf54a9d02012-08-02 08:33:20 +10001023 md_wakeup_thread(mddev->thread);
1024 kfree(plug);
1025 return;
1026 }
1027
1028 /* we aren't scheduling, so we can do the write-out directly. */
1029 bio = bio_list_get(&plug->pending);
1030 bitmap_unplug(mddev->bitmap);
1031 wake_up(&conf->wait_barrier);
1032
1033 while (bio) { /* submit pending writes */
1034 struct bio *next = bio->bi_next;
1035 bio->bi_next = NULL;
Shaohua Li32f9f572013-04-28 18:26:38 +08001036 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
1037 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1038 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001039 bio_endio(bio);
Shaohua Li32f9f572013-04-28 18:26:38 +08001040 else
1041 generic_make_request(bio);
NeilBrownf54a9d02012-08-02 08:33:20 +10001042 bio = next;
1043 }
1044 kfree(plug);
1045}
1046
Shaohua Li849674e2016-01-20 13:52:20 -08001047static void raid1_make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
NeilBrowne8096362011-10-11 16:49:05 +11001049 struct r1conf *conf = mddev->private;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001050 struct raid1_info *mirror;
NeilBrown9f2c9d12011-10-11 16:48:43 +11001051 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 struct bio *read_bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001053 int i, disks;
NeilBrown84255d12008-05-23 13:04:32 -07001054 struct bitmap *bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -07001055 unsigned long flags;
Jens Axboea3623572005-11-01 09:26:16 +01001056 const int rw = bio_data_dir(bio);
NeilBrown2c7d46e2010-08-18 16:16:05 +10001057 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
Tejun Heoe9c74692010-09-03 11:56:18 +02001058 const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA));
Shaohua Li2ff8cc22012-10-11 13:28:54 +11001059 const unsigned long do_discard = (bio->bi_rw
1060 & (REQ_DISCARD | REQ_SECURE));
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001061 const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME);
NeilBrown3cb03002011-10-11 16:45:26 +11001062 struct md_rdev *blocked_rdev;
NeilBrownf54a9d02012-08-02 08:33:20 +10001063 struct blk_plug_cb *cb;
1064 struct raid1_plug_cb *plug = NULL;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001065 int first_clone;
1066 int sectors_handled;
1067 int max_sectors;
majianpeng79ef3a82013-11-15 14:55:02 +08001068 sector_t start_next_window;
NeilBrown191ea9b2005-06-21 17:17:23 -07001069
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 /*
1071 * Register the new request and wait if the reconstruction
1072 * thread has put up a bar for new requests.
1073 * Continue immediately if no resync is active currently.
1074 */
NeilBrown62de6082006-05-01 12:15:47 -07001075
NeilBrown3d310eb2005-06-21 17:17:26 -07001076 md_write_start(mddev, bio); /* wait on superblock update early */
1077
NeilBrown6eef4b22009-12-14 12:49:51 +11001078 if (bio_data_dir(bio) == WRITE &&
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001079 ((bio_end_sector(bio) > mddev->suspend_lo &&
1080 bio->bi_iter.bi_sector < mddev->suspend_hi) ||
1081 (mddev_is_clustered(mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -05001082 md_cluster_ops->area_resyncing(mddev, WRITE,
1083 bio->bi_iter.bi_sector, bio_end_sector(bio))))) {
NeilBrown6eef4b22009-12-14 12:49:51 +11001084 /* As the suspend_* range is controlled by
1085 * userspace, we want an interruptible
1086 * wait.
1087 */
1088 DEFINE_WAIT(w);
1089 for (;;) {
1090 flush_signals(current);
1091 prepare_to_wait(&conf->wait_barrier,
1092 &w, TASK_INTERRUPTIBLE);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001093 if (bio_end_sector(bio) <= mddev->suspend_lo ||
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001094 bio->bi_iter.bi_sector >= mddev->suspend_hi ||
1095 (mddev_is_clustered(mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -05001096 !md_cluster_ops->area_resyncing(mddev, WRITE,
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001097 bio->bi_iter.bi_sector, bio_end_sector(bio))))
NeilBrown6eef4b22009-12-14 12:49:51 +11001098 break;
1099 schedule();
1100 }
1101 finish_wait(&conf->wait_barrier, &w);
1102 }
NeilBrown62de6082006-05-01 12:15:47 -07001103
majianpeng79ef3a82013-11-15 14:55:02 +08001104 start_next_window = wait_barrier(conf, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
NeilBrown84255d12008-05-23 13:04:32 -07001106 bitmap = mddev->bitmap;
1107
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 /*
1109 * make_request() can abort the operation when READA is being
1110 * used and no empty request is available.
1111 *
1112 */
1113 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1114
1115 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001116 r1_bio->sectors = bio_sectors(bio);
NeilBrown191ea9b2005-06-21 17:17:23 -07001117 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001119 r1_bio->sector = bio->bi_iter.bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
NeilBrownd2eb35a2011-07-28 11:31:48 +10001121 /* We might need to issue multiple reads to different
1122 * devices if there are bad blocks around, so we keep
1123 * track of the number of reads in bio->bi_phys_segments.
1124 * If this is 0, there is only one r1_bio and no locking
1125 * will be needed when requests complete. If it is
1126 * non-zero, then it is the number of not-completed requests.
1127 */
1128 bio->bi_phys_segments = 0;
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001129 bio_clear_flag(bio, BIO_SEG_VALID);
NeilBrownd2eb35a2011-07-28 11:31:48 +10001130
Jens Axboea3623572005-11-01 09:26:16 +01001131 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 /*
1133 * read balancing logic:
1134 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001135 int rdisk;
1136
1137read_again:
1138 rdisk = read_balance(conf, r1_bio, &max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140 if (rdisk < 0) {
1141 /* couldn't find anywhere to read from */
1142 raid_end_bio_io(r1_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001143 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 }
1145 mirror = conf->mirrors + rdisk;
1146
NeilBrowne5551902010-03-31 11:21:44 +11001147 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1148 bitmap) {
1149 /* Reading from a write-mostly device must
1150 * take care not to over-take any writes
1151 * that are 'behind'
1152 */
1153 wait_event(bitmap->behind_wait,
1154 atomic_read(&bitmap->behind_writes) == 0);
1155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 r1_bio->read_disk = rdisk;
NeilBrownf0cc9a02014-09-22 10:06:23 +10001157 r1_bio->start_next_window = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
NeilBrowna167f662010-10-26 18:31:13 +11001159 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001160 bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
Kent Overstreet6678d832013-08-07 11:14:32 -07001161 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
1163 r1_bio->bios[rdisk] = read_bio;
1164
Kent Overstreet4f024f32013-10-11 15:44:27 -07001165 read_bio->bi_iter.bi_sector = r1_bio->sector +
1166 mirror->rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 read_bio->bi_bdev = mirror->rdev->bdev;
1168 read_bio->bi_end_io = raid1_end_read_request;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001169 read_bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 read_bio->bi_private = r1_bio;
1171
NeilBrownd2eb35a2011-07-28 11:31:48 +10001172 if (max_sectors < r1_bio->sectors) {
1173 /* could not read all from this device, so we will
1174 * need another r1_bio.
1175 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001176
1177 sectors_handled = (r1_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07001178 - bio->bi_iter.bi_sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +10001179 r1_bio->sectors = max_sectors;
1180 spin_lock_irq(&conf->device_lock);
1181 if (bio->bi_phys_segments == 0)
1182 bio->bi_phys_segments = 2;
1183 else
1184 bio->bi_phys_segments++;
1185 spin_unlock_irq(&conf->device_lock);
1186 /* Cannot call generic_make_request directly
1187 * as that will be queued in __make_request
1188 * and subsequent mempool_alloc might block waiting
1189 * for it. So hand bio over to raid1d.
1190 */
1191 reschedule_retry(r1_bio);
1192
1193 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1194
1195 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001196 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrownd2eb35a2011-07-28 11:31:48 +10001197 r1_bio->state = 0;
1198 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001199 r1_bio->sector = bio->bi_iter.bi_sector +
1200 sectors_handled;
NeilBrownd2eb35a2011-07-28 11:31:48 +10001201 goto read_again;
1202 } else
1203 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001204 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 }
1206
1207 /*
1208 * WRITE:
1209 */
NeilBrown34db0cd2011-10-11 16:50:01 +11001210 if (conf->pending_count >= max_queued_requests) {
1211 md_wakeup_thread(mddev->thread);
1212 wait_event(conf->wait_barrier,
1213 conf->pending_count < max_queued_requests);
1214 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001215 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 * inc refcount on their rdev. Record them by setting
1217 * bios[x] to bio
NeilBrown1f68f0c2011-07-28 11:31:48 +10001218 * If there are known/acknowledged bad blocks on any device on
1219 * which we have seen a write error, we want to avoid writing those
1220 * blocks.
1221 * This potentially requires several writes to write around
1222 * the bad blocks. Each set of writes gets it's own r1bio
1223 * with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001225
NeilBrown8f19ccb2011-12-23 10:17:56 +11001226 disks = conf->raid_disks * 2;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001227 retry_write:
majianpeng79ef3a82013-11-15 14:55:02 +08001228 r1_bio->start_next_window = start_next_window;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001229 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 rcu_read_lock();
NeilBrown1f68f0c2011-07-28 11:31:48 +10001231 max_sectors = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001233 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001234 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1235 atomic_inc(&rdev->nr_pending);
1236 blocked_rdev = rdev;
1237 break;
1238 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001239 r1_bio->bios[i] = NULL;
Kent Overstreet8ae12662015-04-27 23:48:34 -07001240 if (!rdev || test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11001241 if (i < conf->raid_disks)
1242 set_bit(R1BIO_Degraded, &r1_bio->state);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001243 continue;
1244 }
1245
1246 atomic_inc(&rdev->nr_pending);
1247 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1248 sector_t first_bad;
1249 int bad_sectors;
1250 int is_bad;
1251
1252 is_bad = is_badblock(rdev, r1_bio->sector,
1253 max_sectors,
1254 &first_bad, &bad_sectors);
1255 if (is_bad < 0) {
1256 /* mustn't write here until the bad block is
1257 * acknowledged*/
1258 set_bit(BlockedBadBlocks, &rdev->flags);
1259 blocked_rdev = rdev;
1260 break;
NeilBrown964147d2010-05-18 15:27:13 +10001261 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001262 if (is_bad && first_bad <= r1_bio->sector) {
1263 /* Cannot write here at all */
1264 bad_sectors -= (r1_bio->sector - first_bad);
1265 if (bad_sectors < max_sectors)
1266 /* mustn't write more than bad_sectors
1267 * to other devices yet
1268 */
1269 max_sectors = bad_sectors;
1270 rdev_dec_pending(rdev, mddev);
1271 /* We don't set R1BIO_Degraded as that
1272 * only applies if the disk is
1273 * missing, so it might be re-added,
1274 * and we want to know to recover this
1275 * chunk.
1276 * In this case the device is here,
1277 * and the fact that this chunk is not
1278 * in-sync is recorded in the bad
1279 * block log
1280 */
1281 continue;
1282 }
1283 if (is_bad) {
1284 int good_sectors = first_bad - r1_bio->sector;
1285 if (good_sectors < max_sectors)
1286 max_sectors = good_sectors;
1287 }
1288 }
1289 r1_bio->bios[i] = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 }
1291 rcu_read_unlock();
1292
Dan Williams6bfe0b42008-04-30 00:52:32 -07001293 if (unlikely(blocked_rdev)) {
1294 /* Wait for this device to become unblocked */
1295 int j;
majianpeng79ef3a82013-11-15 14:55:02 +08001296 sector_t old = start_next_window;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001297
1298 for (j = 0; j < i; j++)
1299 if (r1_bio->bios[j])
1300 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001301 r1_bio->state = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001302 allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001303 md_wait_for_blocked_rdev(blocked_rdev, mddev);
majianpeng79ef3a82013-11-15 14:55:02 +08001304 start_next_window = wait_barrier(conf, bio);
1305 /*
1306 * We must make sure the multi r1bios of bio have
1307 * the same value of bi_phys_segments
1308 */
1309 if (bio->bi_phys_segments && old &&
1310 old != start_next_window)
1311 /* Wait for the former r1bio(s) to complete */
1312 wait_event(conf->wait_barrier,
1313 bio->bi_phys_segments == 1);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001314 goto retry_write;
1315 }
1316
NeilBrown1f68f0c2011-07-28 11:31:48 +10001317 if (max_sectors < r1_bio->sectors) {
1318 /* We are splitting this write into multiple parts, so
1319 * we need to prepare for allocating another r1_bio.
1320 */
1321 r1_bio->sectors = max_sectors;
1322 spin_lock_irq(&conf->device_lock);
1323 if (bio->bi_phys_segments == 0)
1324 bio->bi_phys_segments = 2;
1325 else
1326 bio->bi_phys_segments++;
1327 spin_unlock_irq(&conf->device_lock);
NeilBrown191ea9b2005-06-21 17:17:23 -07001328 }
Kent Overstreet4f024f32013-10-11 15:44:27 -07001329 sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
NeilBrown4b6d2872005-09-09 16:23:47 -07001330
NeilBrown4e780642010-10-19 12:54:01 +11001331 atomic_set(&r1_bio->remaining, 1);
NeilBrown4b6d2872005-09-09 16:23:47 -07001332 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -07001333
NeilBrown1f68f0c2011-07-28 11:31:48 +10001334 first_clone = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 for (i = 0; i < disks; i++) {
1336 struct bio *mbio;
1337 if (!r1_bio->bios[i])
1338 continue;
1339
NeilBrowna167f662010-10-26 18:31:13 +11001340 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001341 bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
NeilBrown1f68f0c2011-07-28 11:31:48 +10001343 if (first_clone) {
1344 /* do behind I/O ?
1345 * Not if there are too many, or cannot
1346 * allocate memory, or a reader on WriteMostly
1347 * is waiting for behind writes to flush */
1348 if (bitmap &&
1349 (atomic_read(&bitmap->behind_writes)
1350 < mddev->bitmap_info.max_write_behind) &&
1351 !waitqueue_active(&bitmap->behind_wait))
1352 alloc_behind_pages(mbio, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
NeilBrown1f68f0c2011-07-28 11:31:48 +10001354 bitmap_startwrite(bitmap, r1_bio->sector,
1355 r1_bio->sectors,
1356 test_bit(R1BIO_BehindIO,
1357 &r1_bio->state));
1358 first_clone = 0;
1359 }
NeilBrown2ca68f52011-07-28 11:32:10 +10001360 if (r1_bio->behind_bvecs) {
NeilBrown4b6d2872005-09-09 16:23:47 -07001361 struct bio_vec *bvec;
1362 int j;
1363
Kent Overstreetcb34e052012-09-05 15:22:02 -07001364 /*
1365 * We trimmed the bio, so _all is legit
NeilBrown4b6d2872005-09-09 16:23:47 -07001366 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001367 bio_for_each_segment_all(bvec, mbio, j)
NeilBrown2ca68f52011-07-28 11:32:10 +10001368 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
NeilBrown4b6d2872005-09-09 16:23:47 -07001369 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1370 atomic_inc(&r1_bio->behind_remaining);
1371 }
1372
NeilBrown1f68f0c2011-07-28 11:31:48 +10001373 r1_bio->bios[i] = mbio;
1374
Kent Overstreet4f024f32013-10-11 15:44:27 -07001375 mbio->bi_iter.bi_sector = (r1_bio->sector +
NeilBrown1f68f0c2011-07-28 11:31:48 +10001376 conf->mirrors[i].rdev->data_offset);
1377 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1378 mbio->bi_end_io = raid1_end_write_request;
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001379 mbio->bi_rw =
1380 WRITE | do_flush_fua | do_sync | do_discard | do_same;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001381 mbio->bi_private = r1_bio;
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 atomic_inc(&r1_bio->remaining);
NeilBrownf54a9d02012-08-02 08:33:20 +10001384
1385 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1386 if (cb)
1387 plug = container_of(cb, struct raid1_plug_cb, cb);
1388 else
1389 plug = NULL;
NeilBrown4e780642010-10-19 12:54:01 +11001390 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001391 if (plug) {
1392 bio_list_add(&plug->pending, mbio);
1393 plug->pending_cnt++;
1394 } else {
1395 bio_list_add(&conf->pending_bio_list, mbio);
1396 conf->pending_count++;
1397 }
NeilBrown4e780642010-10-19 12:54:01 +11001398 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001399 if (!plug)
NeilBrownb357f042012-07-03 17:45:31 +10001400 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 }
NeilBrown079fa162011-09-10 17:21:23 +10001402 /* Mustn't call r1_bio_write_done before this next test,
1403 * as it could result in the bio being freed.
1404 */
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001405 if (sectors_handled < bio_sectors(bio)) {
NeilBrown079fa162011-09-10 17:21:23 +10001406 r1_bio_write_done(r1_bio);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001407 /* We need another r1_bio. It has already been counted
1408 * in bio->bi_phys_segments
1409 */
1410 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1411 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001412 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001413 r1_bio->state = 0;
1414 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001415 r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001416 goto retry_write;
1417 }
1418
NeilBrown079fa162011-09-10 17:21:23 +10001419 r1_bio_write_done(r1_bio);
1420
1421 /* In case raid1d snuck in to freeze_array */
1422 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423}
1424
Shaohua Li849674e2016-01-20 13:52:20 -08001425static void raid1_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426{
NeilBrowne8096362011-10-11 16:49:05 +11001427 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 int i;
1429
1430 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001431 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001432 rcu_read_lock();
1433 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001434 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001436 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1437 }
1438 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 seq_printf(seq, "]");
1440}
1441
Shaohua Li849674e2016-01-20 13:52:20 -08001442static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443{
1444 char b[BDEVNAME_SIZE];
NeilBrowne8096362011-10-11 16:49:05 +11001445 struct r1conf *conf = mddev->private;
NeilBrown423f04d2015-07-27 11:48:52 +10001446 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
1448 /*
1449 * If it is not operational, then we have already marked it as dead
1450 * else if it is the last working disks, ignore the error, let the
1451 * next level up know.
1452 * else mark the drive as failed
1453 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001454 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001455 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 /*
1457 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001458 * normal single drive.
1459 * However don't try a recovery from this drive as
1460 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 */
NeilBrown53890422011-07-27 11:00:36 +10001462 conf->recovery_disabled = mddev->recovery_disabled;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001464 }
NeilBrownde393cd2011-07-28 11:31:48 +10001465 set_bit(Blocked, &rdev->flags);
NeilBrown423f04d2015-07-27 11:48:52 +10001466 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001467 if (test_and_clear_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -07001469 set_bit(Faulty, &rdev->flags);
NeilBrowndd00a992007-05-10 03:15:50 -07001470 } else
1471 set_bit(Faulty, &rdev->flags);
NeilBrown423f04d2015-07-27 11:48:52 +10001472 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrown2446dba2014-07-31 10:16:29 +10001473 /*
1474 * if recovery is running, make sure it aborts.
1475 */
1476 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrown850b2b42006-10-03 01:15:46 -07001477 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown55ce74d2015-08-14 11:11:10 +10001478 set_bit(MD_CHANGE_PENDING, &mddev->flags);
Joe Perches067032b2011-01-14 09:14:33 +11001479 printk(KERN_ALERT
1480 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1481 "md/raid1:%s: Operation continuing on %d devices.\n",
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001482 mdname(mddev), bdevname(rdev->bdev, b),
1483 mdname(mddev), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484}
1485
NeilBrowne8096362011-10-11 16:49:05 +11001486static void print_conf(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487{
1488 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001490 printk(KERN_DEBUG "RAID1 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 if (!conf) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001492 printk(KERN_DEBUG "(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return;
1494 }
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001495 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 conf->raid_disks);
1497
NeilBrownddac7c72006-08-31 21:27:36 -07001498 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 for (i = 0; i < conf->raid_disks; i++) {
1500 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001501 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownddac7c72006-08-31 21:27:36 -07001502 if (rdev)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001503 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -07001504 i, !test_bit(In_sync, &rdev->flags),
1505 !test_bit(Faulty, &rdev->flags),
1506 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 }
NeilBrownddac7c72006-08-31 21:27:36 -07001508 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509}
1510
NeilBrowne8096362011-10-11 16:49:05 +11001511static void close_sync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
majianpeng79ef3a82013-11-15 14:55:02 +08001513 wait_barrier(conf, NULL);
1514 allow_barrier(conf, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 mempool_destroy(conf->r1buf_pool);
1517 conf->r1buf_pool = NULL;
majianpeng79ef3a82013-11-15 14:55:02 +08001518
NeilBrown669cc7b2014-09-04 16:30:38 +10001519 spin_lock_irq(&conf->resync_lock);
Jes Sorensene8ff8bf2015-09-16 10:20:05 -04001520 conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE;
majianpeng79ef3a82013-11-15 14:55:02 +08001521 conf->start_next_window = MaxSector;
NeilBrown669cc7b2014-09-04 16:30:38 +10001522 conf->current_window_requests +=
1523 conf->next_window_requests;
1524 conf->next_window_requests = 0;
1525 spin_unlock_irq(&conf->resync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526}
1527
NeilBrownfd01b882011-10-11 16:47:53 +11001528static int raid1_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
1530 int i;
NeilBrowne8096362011-10-11 16:49:05 +11001531 struct r1conf *conf = mddev->private;
NeilBrown6b965622010-08-18 11:56:59 +10001532 int count = 0;
1533 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
1535 /*
NeilBrownf72ffdd2014-09-30 14:23:59 +10001536 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001537 * and mark them readable.
1538 * Called under mddev lock, so rcu protection not needed.
NeilBrown423f04d2015-07-27 11:48:52 +10001539 * device_lock used to avoid races with raid1_end_read_request
1540 * which expects 'In_sync' flags and ->degraded to be consistent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 */
NeilBrown423f04d2015-07-27 11:48:52 +10001542 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001544 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001545 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1546 if (repl
Goldwyn Rodrigues1aee41f2014-10-29 18:51:31 -05001547 && !test_bit(Candidate, &repl->flags)
NeilBrown8c7a2c22011-12-23 10:17:57 +11001548 && repl->recovery_offset == MaxSector
1549 && !test_bit(Faulty, &repl->flags)
1550 && !test_and_set_bit(In_sync, &repl->flags)) {
1551 /* replacement has just become active */
1552 if (!rdev ||
1553 !test_and_clear_bit(In_sync, &rdev->flags))
1554 count++;
1555 if (rdev) {
1556 /* Replaced device not technically
1557 * faulty, but we need to be sure
1558 * it gets removed and never re-added
1559 */
1560 set_bit(Faulty, &rdev->flags);
1561 sysfs_notify_dirent_safe(
1562 rdev->sysfs_state);
1563 }
1564 }
NeilBrownddac7c72006-08-31 21:27:36 -07001565 if (rdev
Lukasz Dorau61e49472013-10-24 12:55:17 +11001566 && rdev->recovery_offset == MaxSector
NeilBrownddac7c72006-08-31 21:27:36 -07001567 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001568 && !test_and_set_bit(In_sync, &rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001569 count++;
Jonathan Brassow654e8b52011-07-27 11:00:36 +10001570 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
1572 }
NeilBrown6b965622010-08-18 11:56:59 +10001573 mddev->degraded -= count;
1574 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001577 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578}
1579
NeilBrownfd01b882011-10-11 16:47:53 +11001580static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581{
NeilBrowne8096362011-10-11 16:49:05 +11001582 struct r1conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001583 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001584 int mirror = 0;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001585 struct raid1_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001586 int first = 0;
NeilBrown30194632011-12-23 10:17:56 +11001587 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
NeilBrown53890422011-07-27 11:00:36 +10001589 if (mddev->recovery_disabled == conf->recovery_disabled)
1590 return -EBUSY;
1591
Dan Williams1501efa2016-01-13 16:00:07 -08001592 if (md_integrity_add_rdev(rdev, mddev))
1593 return -ENXIO;
1594
Neil Brown6c2fce22008-06-28 08:31:31 +10001595 if (rdev->raid_disk >= 0)
1596 first = last = rdev->raid_disk;
1597
Goldwyn Rodrigues70bcecd2015-08-21 10:33:39 -05001598 /*
1599 * find the disk ... but prefer rdev->saved_raid_disk
1600 * if possible.
1601 */
1602 if (rdev->saved_raid_disk >= 0 &&
1603 rdev->saved_raid_disk >= first &&
1604 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1605 first = last = rdev->saved_raid_disk;
1606
NeilBrown7ef449d2011-12-23 10:17:57 +11001607 for (mirror = first; mirror <= last; mirror++) {
1608 p = conf->mirrors+mirror;
1609 if (!p->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
Jonathan Brassow9092c022013-05-02 14:19:24 -05001611 if (mddev->gendisk)
1612 disk_stack_limits(mddev->gendisk, rdev->bdev,
1613 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
1615 p->head_position = 0;
1616 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001617 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001618 /* As all devices are equivalent, we don't need a full recovery
1619 * if this was recently any drive of the array
1620 */
1621 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001622 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001623 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 break;
1625 }
NeilBrown7ef449d2011-12-23 10:17:57 +11001626 if (test_bit(WantReplacement, &p->rdev->flags) &&
1627 p[conf->raid_disks].rdev == NULL) {
1628 /* Add this device as a replacement */
1629 clear_bit(In_sync, &rdev->flags);
1630 set_bit(Replacement, &rdev->flags);
1631 rdev->raid_disk = mirror;
1632 err = 0;
1633 conf->fullsync = 1;
1634 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1635 break;
1636 }
1637 }
Jonathan Brassow9092c022013-05-02 14:19:24 -05001638 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
Shaohua Li2ff8cc22012-10-11 13:28:54 +11001639 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001641 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642}
1643
NeilBrownb8321b62011-12-23 10:17:51 +11001644static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645{
NeilBrowne8096362011-10-11 16:49:05 +11001646 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001648 int number = rdev->raid_disk;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001649 struct raid1_info *p = conf->mirrors + number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
NeilBrownb014f142011-12-23 10:17:56 +11001651 if (rdev != p->rdev)
1652 p = conf->mirrors + conf->raid_disks + number;
1653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 print_conf(conf);
NeilBrownb8321b62011-12-23 10:17:51 +11001655 if (rdev == p->rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001656 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 atomic_read(&rdev->nr_pending)) {
1658 err = -EBUSY;
1659 goto abort;
1660 }
NeilBrown046abee2010-10-26 15:46:20 +11001661 /* Only remove non-faulty devices if recovery
NeilBrowndfc70642008-05-23 13:04:39 -07001662 * is not possible.
1663 */
1664 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown53890422011-07-27 11:00:36 +10001665 mddev->recovery_disabled != conf->recovery_disabled &&
NeilBrowndfc70642008-05-23 13:04:39 -07001666 mddev->degraded < conf->raid_disks) {
1667 err = -EBUSY;
1668 goto abort;
1669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001671 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 if (atomic_read(&rdev->nr_pending)) {
1673 /* lost the race, try later */
1674 err = -EBUSY;
1675 p->rdev = rdev;
Andre Nollac5e7112009-08-03 10:59:47 +10001676 goto abort;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001677 } else if (conf->mirrors[conf->raid_disks + number].rdev) {
1678 /* We just removed a device that is being replaced.
1679 * Move down the replacement. We drain all IO before
1680 * doing this to avoid confusion.
1681 */
1682 struct md_rdev *repl =
1683 conf->mirrors[conf->raid_disks + number].rdev;
NeilBrowne2d59922013-06-12 11:01:22 +10001684 freeze_array(conf, 0);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001685 clear_bit(Replacement, &repl->flags);
1686 p->rdev = repl;
1687 conf->mirrors[conf->raid_disks + number].rdev = NULL;
NeilBrowne2d59922013-06-12 11:01:22 +10001688 unfreeze_array(conf);
NeilBrownb014f142011-12-23 10:17:56 +11001689 clear_bit(WantReplacement, &rdev->flags);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001690 } else
1691 clear_bit(WantReplacement, &rdev->flags);
Martin K. Petersena91a2782011-03-17 11:11:05 +01001692 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 }
1694abort:
1695
1696 print_conf(conf);
1697 return err;
1698}
1699
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001700static void end_sync_read(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001702 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
NeilBrown0fc280f2011-10-07 14:22:55 +11001704 update_head_pos(r1_bio->read_disk, r1_bio);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001705
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 /*
1707 * we have read a block, now it needs to be re-written,
1708 * or re-read if the read failed.
1709 * We don't do much here, just schedule handling by raid1d
1710 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001711 if (!bio->bi_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001713
1714 if (atomic_dec_and_test(&r1_bio->remaining))
1715 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716}
1717
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001718static void end_sync_write(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001720 int uptodate = !bio->bi_error;
NeilBrown9f2c9d12011-10-11 16:48:43 +11001721 struct r1bio *r1_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11001722 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001723 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 int mirror=0;
NeilBrown4367af52011-07-28 11:31:49 +10001725 sector_t first_bad;
1726 int bad_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001728 mirror = find_bio_disk(r1_bio, bio);
1729
NeilBrown6b1117d2006-03-31 02:31:57 -08001730 if (!uptodate) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001731 sector_t sync_blocks = 0;
NeilBrown6b1117d2006-03-31 02:31:57 -08001732 sector_t s = r1_bio->sector;
1733 long sectors_to_go = r1_bio->sectors;
1734 /* make sure these bits doesn't get cleared. */
1735 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001736 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001737 &sync_blocks, 1);
1738 s += sync_blocks;
1739 sectors_to_go -= sync_blocks;
1740 } while (sectors_to_go > 0);
NeilBrownd8f05d22011-07-28 11:33:00 +10001741 set_bit(WriteErrorSeen,
1742 &conf->mirrors[mirror].rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001743 if (!test_and_set_bit(WantReplacement,
1744 &conf->mirrors[mirror].rdev->flags))
1745 set_bit(MD_RECOVERY_NEEDED, &
1746 mddev->recovery);
NeilBrownd8f05d22011-07-28 11:33:00 +10001747 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +10001748 } else if (is_badblock(conf->mirrors[mirror].rdev,
1749 r1_bio->sector,
1750 r1_bio->sectors,
NeilBrown3a9f28a2011-07-28 11:33:42 +10001751 &first_bad, &bad_sectors) &&
1752 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1753 r1_bio->sector,
1754 r1_bio->sectors,
1755 &first_bad, &bad_sectors)
1756 )
NeilBrown4367af52011-07-28 11:31:49 +10001757 set_bit(R1BIO_MadeGood, &r1_bio->state);
NeilBrowne3b97032005-08-04 12:53:34 -07001758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown4367af52011-07-28 11:31:49 +10001760 int s = r1_bio->sectors;
NeilBrownd8f05d22011-07-28 11:33:00 +10001761 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1762 test_bit(R1BIO_WriteError, &r1_bio->state))
NeilBrown4367af52011-07-28 11:31:49 +10001763 reschedule_retry(r1_bio);
1764 else {
1765 put_buf(r1_bio);
1766 md_done_sync(mddev, s, uptodate);
1767 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769}
1770
NeilBrown3cb03002011-10-11 16:45:26 +11001771static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrownd8f05d22011-07-28 11:33:00 +10001772 int sectors, struct page *page, int rw)
1773{
1774 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
1775 /* success */
1776 return 1;
NeilBrown19d67162011-12-23 10:17:57 +11001777 if (rw == WRITE) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001778 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001779 if (!test_and_set_bit(WantReplacement,
1780 &rdev->flags))
1781 set_bit(MD_RECOVERY_NEEDED, &
1782 rdev->mddev->recovery);
1783 }
NeilBrownd8f05d22011-07-28 11:33:00 +10001784 /* need to record an error - either for the block or the device */
1785 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1786 md_error(rdev->mddev, rdev);
1787 return 0;
1788}
1789
NeilBrown9f2c9d12011-10-11 16:48:43 +11001790static int fix_sync_read_error(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001791{
1792 /* Try some synchronous reads of other devices to get
1793 * good data, much like with normal read errors. Only
1794 * read into the pages we already have so we don't
1795 * need to re-issue the read request.
1796 * We don't need to freeze the array, because being in an
1797 * active sync request, there is no normal IO, and
1798 * no overlapping syncs.
NeilBrown06f60382011-07-28 11:31:48 +10001799 * We don't need to check is_badblock() again as we
1800 * made sure that anything with a bad block in range
1801 * will have bi_end_io clear.
NeilBrowna68e5872011-05-11 14:40:44 +10001802 */
NeilBrownfd01b882011-10-11 16:47:53 +11001803 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001804 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001805 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1806 sector_t sect = r1_bio->sector;
1807 int sectors = r1_bio->sectors;
1808 int idx = 0;
1809
1810 while(sectors) {
1811 int s = sectors;
1812 int d = r1_bio->read_disk;
1813 int success = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11001814 struct md_rdev *rdev;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001815 int start;
NeilBrowna68e5872011-05-11 14:40:44 +10001816
1817 if (s > (PAGE_SIZE>>9))
1818 s = PAGE_SIZE >> 9;
1819 do {
1820 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1821 /* No rcu protection needed here devices
1822 * can only be removed when no resync is
1823 * active, and resync is currently active
1824 */
1825 rdev = conf->mirrors[d].rdev;
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001826 if (sync_page_io(rdev, sect, s<<9,
NeilBrowna68e5872011-05-11 14:40:44 +10001827 bio->bi_io_vec[idx].bv_page,
1828 READ, false)) {
1829 success = 1;
1830 break;
1831 }
1832 }
1833 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001834 if (d == conf->raid_disks * 2)
NeilBrowna68e5872011-05-11 14:40:44 +10001835 d = 0;
1836 } while (!success && d != r1_bio->read_disk);
1837
NeilBrown78d7f5f2011-05-11 14:48:56 +10001838 if (!success) {
NeilBrowna68e5872011-05-11 14:40:44 +10001839 char b[BDEVNAME_SIZE];
NeilBrown3a9f28a2011-07-28 11:33:42 +10001840 int abort = 0;
1841 /* Cannot read from anywhere, this block is lost.
1842 * Record a bad block on each device. If that doesn't
1843 * work just disable and interrupt the recovery.
1844 * Don't fail devices as that won't really help.
1845 */
NeilBrowna68e5872011-05-11 14:40:44 +10001846 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1847 " for block %llu\n",
1848 mdname(mddev),
1849 bdevname(bio->bi_bdev, b),
1850 (unsigned long long)r1_bio->sector);
NeilBrown8f19ccb2011-12-23 10:17:56 +11001851 for (d = 0; d < conf->raid_disks * 2; d++) {
NeilBrown3a9f28a2011-07-28 11:33:42 +10001852 rdev = conf->mirrors[d].rdev;
1853 if (!rdev || test_bit(Faulty, &rdev->flags))
1854 continue;
1855 if (!rdev_set_badblocks(rdev, sect, s, 0))
1856 abort = 1;
1857 }
1858 if (abort) {
NeilBrownd890fa22011-10-26 11:54:39 +11001859 conf->recovery_disabled =
1860 mddev->recovery_disabled;
NeilBrown3a9f28a2011-07-28 11:33:42 +10001861 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1862 md_done_sync(mddev, r1_bio->sectors, 0);
1863 put_buf(r1_bio);
1864 return 0;
1865 }
1866 /* Try next page */
1867 sectors -= s;
1868 sect += s;
1869 idx++;
1870 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001871 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001872
1873 start = d;
1874 /* write it back and re-read */
1875 while (d != r1_bio->read_disk) {
1876 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001877 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001878 d--;
1879 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1880 continue;
1881 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001882 if (r1_sync_page_io(rdev, sect, s,
1883 bio->bi_io_vec[idx].bv_page,
1884 WRITE) == 0) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001885 r1_bio->bios[d]->bi_end_io = NULL;
1886 rdev_dec_pending(rdev, mddev);
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001887 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001888 }
1889 d = start;
1890 while (d != r1_bio->read_disk) {
1891 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001892 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001893 d--;
1894 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1895 continue;
1896 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001897 if (r1_sync_page_io(rdev, sect, s,
1898 bio->bi_io_vec[idx].bv_page,
1899 READ) != 0)
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001900 atomic_add(s, &rdev->corrected_errors);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001901 }
NeilBrowna68e5872011-05-11 14:40:44 +10001902 sectors -= s;
1903 sect += s;
1904 idx ++;
1905 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001906 set_bit(R1BIO_Uptodate, &r1_bio->state);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001907 bio->bi_error = 0;
NeilBrowna68e5872011-05-11 14:40:44 +10001908 return 1;
1909}
1910
NeilBrownc95e6382014-09-09 13:54:11 +10001911static void process_checks(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001912{
1913 /* We have read all readable devices. If we haven't
1914 * got the block, then there is no hope left.
1915 * If we have, then we want to do a comparison
1916 * and skip the write if everything is the same.
1917 * If any blocks failed to read, then we need to
1918 * attempt an over-write
1919 */
NeilBrownfd01b882011-10-11 16:47:53 +11001920 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001921 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001922 int primary;
1923 int i;
majianpengf4380a92012-04-12 16:04:47 +10001924 int vcnt;
NeilBrowna68e5872011-05-11 14:40:44 +10001925
NeilBrown30bc9b52013-07-17 15:19:29 +10001926 /* Fix variable parts of all bios */
1927 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1928 for (i = 0; i < conf->raid_disks * 2; i++) {
1929 int j;
1930 int size;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001931 int error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001932 struct bio *b = r1_bio->bios[i];
1933 if (b->bi_end_io != end_sync_read)
1934 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001935 /* fixup the bio for reuse, but preserve errno */
1936 error = b->bi_error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001937 bio_reset(b);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001938 b->bi_error = error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001939 b->bi_vcnt = vcnt;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001940 b->bi_iter.bi_size = r1_bio->sectors << 9;
1941 b->bi_iter.bi_sector = r1_bio->sector +
NeilBrown30bc9b52013-07-17 15:19:29 +10001942 conf->mirrors[i].rdev->data_offset;
1943 b->bi_bdev = conf->mirrors[i].rdev->bdev;
1944 b->bi_end_io = end_sync_read;
1945 b->bi_private = r1_bio;
1946
Kent Overstreet4f024f32013-10-11 15:44:27 -07001947 size = b->bi_iter.bi_size;
NeilBrown30bc9b52013-07-17 15:19:29 +10001948 for (j = 0; j < vcnt ; j++) {
1949 struct bio_vec *bi;
1950 bi = &b->bi_io_vec[j];
1951 bi->bv_offset = 0;
1952 if (size > PAGE_SIZE)
1953 bi->bv_len = PAGE_SIZE;
1954 else
1955 bi->bv_len = size;
1956 size -= PAGE_SIZE;
1957 }
1958 }
NeilBrown8f19ccb2011-12-23 10:17:56 +11001959 for (primary = 0; primary < conf->raid_disks * 2; primary++)
NeilBrowna68e5872011-05-11 14:40:44 +10001960 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001961 !r1_bio->bios[primary]->bi_error) {
NeilBrowna68e5872011-05-11 14:40:44 +10001962 r1_bio->bios[primary]->bi_end_io = NULL;
1963 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1964 break;
1965 }
1966 r1_bio->read_disk = primary;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001967 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001968 int j;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001969 struct bio *pbio = r1_bio->bios[primary];
1970 struct bio *sbio = r1_bio->bios[i];
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001971 int error = sbio->bi_error;
NeilBrowna68e5872011-05-11 14:40:44 +10001972
Kent Overstreet2aabaa62012-09-11 11:26:12 -07001973 if (sbio->bi_end_io != end_sync_read)
NeilBrown78d7f5f2011-05-11 14:48:56 +10001974 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001975 /* Now we can 'fixup' the error value */
1976 sbio->bi_error = 0;
NeilBrowna68e5872011-05-11 14:40:44 +10001977
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001978 if (!error) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001979 for (j = vcnt; j-- ; ) {
1980 struct page *p, *s;
1981 p = pbio->bi_io_vec[j].bv_page;
1982 s = sbio->bi_io_vec[j].bv_page;
1983 if (memcmp(page_address(p),
1984 page_address(s),
NeilBrown5020ad72012-04-02 01:39:05 +10001985 sbio->bi_io_vec[j].bv_len))
NeilBrown78d7f5f2011-05-11 14:48:56 +10001986 break;
NeilBrowna68e5872011-05-11 14:40:44 +10001987 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001988 } else
1989 j = 0;
1990 if (j >= 0)
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11001991 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001992 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001993 && !error)) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001994 /* No need to write to this device. */
1995 sbio->bi_end_io = NULL;
1996 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1997 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001998 }
Kent Overstreetd3b45c22012-09-10 13:49:33 -07001999
2000 bio_copy_data(sbio, pbio);
NeilBrown78d7f5f2011-05-11 14:48:56 +10002001 }
NeilBrowna68e5872011-05-11 14:40:44 +10002002}
2003
NeilBrown9f2c9d12011-10-11 16:48:43 +11002004static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005{
NeilBrowne8096362011-10-11 16:49:05 +11002006 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 int i;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002008 int disks = conf->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 struct bio *bio, *wbio;
2010
2011 bio = r1_bio->bios[r1_bio->read_disk];
2012
NeilBrowna68e5872011-05-11 14:40:44 +10002013 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
2014 /* ouch - failed to read all of that. */
2015 if (!fix_sync_read_error(r1_bio))
2016 return;
NeilBrown7ca78d52011-05-11 14:50:37 +10002017
2018 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrownc95e6382014-09-09 13:54:11 +10002019 process_checks(r1_bio);
2020
NeilBrownd11c1712006-01-06 00:20:26 -08002021 /*
2022 * schedule writes
2023 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 atomic_set(&r1_bio->remaining, 1);
2025 for (i = 0; i < disks ; i++) {
2026 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08002027 if (wbio->bi_end_io == NULL ||
2028 (wbio->bi_end_io == end_sync_read &&
2029 (i == r1_bio->read_disk ||
2030 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 continue;
2032
NeilBrown3e198f72006-01-06 00:20:21 -08002033 wbio->bi_rw = WRITE;
2034 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 atomic_inc(&r1_bio->remaining);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002036 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
NeilBrown191ea9b2005-06-21 17:17:23 -07002037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 generic_make_request(wbio);
2039 }
2040
2041 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002042 /* if we're here, all write(s) have completed, so clean up */
NeilBrown58e94ae2012-07-19 15:59:18 +10002043 int s = r1_bio->sectors;
2044 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2045 test_bit(R1BIO_WriteError, &r1_bio->state))
2046 reschedule_retry(r1_bio);
2047 else {
2048 put_buf(r1_bio);
2049 md_done_sync(mddev, s, 1);
2050 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 }
2052}
2053
2054/*
2055 * This is a kernel thread which:
2056 *
2057 * 1. Retries failed read operations on working mirrors.
2058 * 2. Updates the raid superblock when problems encounter.
NeilBrownd2eb35a2011-07-28 11:31:48 +10002059 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 */
2061
NeilBrowne8096362011-10-11 16:49:05 +11002062static void fix_read_error(struct r1conf *conf, int read_disk,
NeilBrown867868f2006-10-03 01:15:51 -07002063 sector_t sect, int sectors)
2064{
NeilBrownfd01b882011-10-11 16:47:53 +11002065 struct mddev *mddev = conf->mddev;
NeilBrown867868f2006-10-03 01:15:51 -07002066 while(sectors) {
2067 int s = sectors;
2068 int d = read_disk;
2069 int success = 0;
2070 int start;
NeilBrown3cb03002011-10-11 16:45:26 +11002071 struct md_rdev *rdev;
NeilBrown867868f2006-10-03 01:15:51 -07002072
2073 if (s > (PAGE_SIZE>>9))
2074 s = PAGE_SIZE >> 9;
2075
2076 do {
2077 /* Note: no rcu protection needed here
2078 * as this is synchronous in the raid1d thread
2079 * which is the thread that might remove
2080 * a device. If raid1d ever becomes multi-threaded....
2081 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10002082 sector_t first_bad;
2083 int bad_sectors;
2084
NeilBrown867868f2006-10-03 01:15:51 -07002085 rdev = conf->mirrors[d].rdev;
2086 if (rdev &&
majianpengda8840a2012-05-22 13:55:03 +10002087 (test_bit(In_sync, &rdev->flags) ||
2088 (!test_bit(Faulty, &rdev->flags) &&
2089 rdev->recovery_offset >= sect + s)) &&
NeilBrownd2eb35a2011-07-28 11:31:48 +10002090 is_badblock(rdev, sect, s,
2091 &first_bad, &bad_sectors) == 0 &&
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11002092 sync_page_io(rdev, sect, s<<9,
2093 conf->tmppage, READ, false))
NeilBrown867868f2006-10-03 01:15:51 -07002094 success = 1;
2095 else {
2096 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002097 if (d == conf->raid_disks * 2)
NeilBrown867868f2006-10-03 01:15:51 -07002098 d = 0;
2099 }
2100 } while (!success && d != read_disk);
2101
2102 if (!success) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002103 /* Cannot read from anywhere - mark it bad */
NeilBrown3cb03002011-10-11 16:45:26 +11002104 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10002105 if (!rdev_set_badblocks(rdev, sect, s, 0))
2106 md_error(mddev, rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002107 break;
2108 }
2109 /* write it back and re-read */
2110 start = d;
2111 while (d != read_disk) {
2112 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002113 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002114 d--;
2115 rdev = conf->mirrors[d].rdev;
2116 if (rdev &&
NeilBrownb8cb6b42014-09-18 11:09:04 +10002117 !test_bit(Faulty, &rdev->flags))
NeilBrownd8f05d22011-07-28 11:33:00 +10002118 r1_sync_page_io(rdev, sect, s,
2119 conf->tmppage, WRITE);
NeilBrown867868f2006-10-03 01:15:51 -07002120 }
2121 d = start;
2122 while (d != read_disk) {
2123 char b[BDEVNAME_SIZE];
2124 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002125 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002126 d--;
2127 rdev = conf->mirrors[d].rdev;
2128 if (rdev &&
NeilBrownb8cb6b42014-09-18 11:09:04 +10002129 !test_bit(Faulty, &rdev->flags)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002130 if (r1_sync_page_io(rdev, sect, s,
2131 conf->tmppage, READ)) {
NeilBrown867868f2006-10-03 01:15:51 -07002132 atomic_add(s, &rdev->corrected_errors);
2133 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002134 "md/raid1:%s: read error corrected "
NeilBrown867868f2006-10-03 01:15:51 -07002135 "(%d sectors at %llu on %s)\n",
2136 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07002137 (unsigned long long)(sect +
2138 rdev->data_offset),
NeilBrown867868f2006-10-03 01:15:51 -07002139 bdevname(rdev->bdev, b));
2140 }
2141 }
2142 }
2143 sectors -= s;
2144 sect += s;
2145 }
2146}
2147
NeilBrown9f2c9d12011-10-11 16:48:43 +11002148static int narrow_write_error(struct r1bio *r1_bio, int i)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002149{
NeilBrownfd01b882011-10-11 16:47:53 +11002150 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11002151 struct r1conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11002152 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002153
2154 /* bio has the data to be written to device 'i' where
2155 * we just recently had a write error.
2156 * We repeatedly clone the bio and trim down to one block,
2157 * then try the write. Where the write fails we record
2158 * a bad block.
2159 * It is conceivable that the bio doesn't exactly align with
2160 * blocks. We must handle this somehow.
2161 *
2162 * We currently own a reference on the rdev.
2163 */
2164
2165 int block_sectors;
2166 sector_t sector;
2167 int sectors;
2168 int sect_to_write = r1_bio->sectors;
2169 int ok = 1;
2170
2171 if (rdev->badblocks.shift < 0)
2172 return 0;
2173
Nate Daileyab713cd2015-02-12 12:02:09 -05002174 block_sectors = roundup(1 << rdev->badblocks.shift,
2175 bdev_logical_block_size(rdev->bdev) >> 9);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002176 sector = r1_bio->sector;
2177 sectors = ((sector + block_sectors)
2178 & ~(sector_t)(block_sectors - 1))
2179 - sector;
2180
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002181 while (sect_to_write) {
2182 struct bio *wbio;
2183 if (sectors > sect_to_write)
2184 sectors = sect_to_write;
2185 /* Write at 'sector' for 'sectors'*/
2186
Kent Overstreetb7838632012-09-10 15:17:11 -07002187 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2188 unsigned vcnt = r1_bio->behind_page_count;
2189 struct bio_vec *vec = r1_bio->behind_bvecs;
2190
2191 while (!vec->bv_page) {
2192 vec++;
2193 vcnt--;
2194 }
2195
2196 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2197 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2198
2199 wbio->bi_vcnt = vcnt;
2200 } else {
2201 wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2202 }
2203
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002204 wbio->bi_rw = WRITE;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002205 wbio->bi_iter.bi_sector = r1_bio->sector;
2206 wbio->bi_iter.bi_size = r1_bio->sectors << 9;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002207
Kent Overstreet6678d832013-08-07 11:14:32 -07002208 bio_trim(wbio, sector - r1_bio->sector, sectors);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002209 wbio->bi_iter.bi_sector += rdev->data_offset;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002210 wbio->bi_bdev = rdev->bdev;
Jes Sorensen203d27b2015-10-20 12:09:12 -04002211 if (submit_bio_wait(WRITE, wbio) < 0)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002212 /* failure! */
2213 ok = rdev_set_badblocks(rdev, sector,
2214 sectors, 0)
2215 && ok;
2216
2217 bio_put(wbio);
2218 sect_to_write -= sectors;
2219 sector += sectors;
2220 sectors = block_sectors;
2221 }
2222 return ok;
2223}
2224
NeilBrowne8096362011-10-11 16:49:05 +11002225static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002226{
2227 int m;
2228 int s = r1_bio->sectors;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002229 for (m = 0; m < conf->raid_disks * 2 ; m++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002230 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002231 struct bio *bio = r1_bio->bios[m];
2232 if (bio->bi_end_io == NULL)
2233 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002234 if (!bio->bi_error &&
NeilBrown62096bc2011-07-28 11:38:13 +10002235 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
NeilBrownc6563a82012-05-21 09:27:00 +10002236 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002237 }
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002238 if (bio->bi_error &&
NeilBrown62096bc2011-07-28 11:38:13 +10002239 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2240 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2241 md_error(conf->mddev, rdev);
2242 }
2243 }
2244 put_buf(r1_bio);
2245 md_done_sync(conf->mddev, s, 1);
2246}
2247
NeilBrowne8096362011-10-11 16:49:05 +11002248static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002249{
2250 int m;
NeilBrown55ce74d2015-08-14 11:11:10 +10002251 bool fail = false;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002252 for (m = 0; m < conf->raid_disks * 2 ; m++)
NeilBrown62096bc2011-07-28 11:38:13 +10002253 if (r1_bio->bios[m] == IO_MADE_GOOD) {
NeilBrown3cb03002011-10-11 16:45:26 +11002254 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002255 rdev_clear_badblocks(rdev,
2256 r1_bio->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10002257 r1_bio->sectors, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002258 rdev_dec_pending(rdev, conf->mddev);
2259 } else if (r1_bio->bios[m] != NULL) {
2260 /* This drive got a write error. We need to
2261 * narrow down and record precise write
2262 * errors.
2263 */
NeilBrown55ce74d2015-08-14 11:11:10 +10002264 fail = true;
NeilBrown62096bc2011-07-28 11:38:13 +10002265 if (!narrow_write_error(r1_bio, m)) {
2266 md_error(conf->mddev,
2267 conf->mirrors[m].rdev);
2268 /* an I/O failed, we can't clear the bitmap */
2269 set_bit(R1BIO_Degraded, &r1_bio->state);
2270 }
2271 rdev_dec_pending(conf->mirrors[m].rdev,
2272 conf->mddev);
2273 }
NeilBrown55ce74d2015-08-14 11:11:10 +10002274 if (fail) {
2275 spin_lock_irq(&conf->device_lock);
2276 list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
2277 spin_unlock_irq(&conf->device_lock);
2278 md_wakeup_thread(conf->mddev->thread);
NeilBrownbd8688a2015-10-24 16:02:16 +11002279 } else {
2280 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2281 close_write(r1_bio);
NeilBrown55ce74d2015-08-14 11:11:10 +10002282 raid_end_bio_io(r1_bio);
NeilBrownbd8688a2015-10-24 16:02:16 +11002283 }
NeilBrown62096bc2011-07-28 11:38:13 +10002284}
2285
NeilBrowne8096362011-10-11 16:49:05 +11002286static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002287{
2288 int disk;
2289 int max_sectors;
NeilBrownfd01b882011-10-11 16:47:53 +11002290 struct mddev *mddev = conf->mddev;
NeilBrown62096bc2011-07-28 11:38:13 +10002291 struct bio *bio;
2292 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11002293 struct md_rdev *rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002294
2295 clear_bit(R1BIO_ReadError, &r1_bio->state);
2296 /* we got a read error. Maybe the drive is bad. Maybe just
2297 * the block and we can fix it.
2298 * We freeze all other IO, and try reading the block from
2299 * other devices. When we find one, we re-write
2300 * and check it that fixes the read error.
2301 * This is all done synchronously while the array is
2302 * frozen
2303 */
2304 if (mddev->ro == 0) {
NeilBrowne2d59922013-06-12 11:01:22 +10002305 freeze_array(conf, 1);
NeilBrown62096bc2011-07-28 11:38:13 +10002306 fix_read_error(conf, r1_bio->read_disk,
2307 r1_bio->sector, r1_bio->sectors);
2308 unfreeze_array(conf);
2309 } else
2310 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
NeilBrown7ad4d4a2012-10-11 13:44:30 +11002311 rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
NeilBrown62096bc2011-07-28 11:38:13 +10002312
2313 bio = r1_bio->bios[r1_bio->read_disk];
2314 bdevname(bio->bi_bdev, b);
2315read_more:
2316 disk = read_balance(conf, r1_bio, &max_sectors);
2317 if (disk == -1) {
2318 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
2319 " read error for block %llu\n",
2320 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2321 raid_end_bio_io(r1_bio);
2322 } else {
2323 const unsigned long do_sync
2324 = r1_bio->master_bio->bi_rw & REQ_SYNC;
2325 if (bio) {
2326 r1_bio->bios[r1_bio->read_disk] =
2327 mddev->ro ? IO_BLOCKED : NULL;
2328 bio_put(bio);
2329 }
2330 r1_bio->read_disk = disk;
2331 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002332 bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
2333 max_sectors);
NeilBrown62096bc2011-07-28 11:38:13 +10002334 r1_bio->bios[r1_bio->read_disk] = bio;
2335 rdev = conf->mirrors[disk].rdev;
2336 printk_ratelimited(KERN_ERR
2337 "md/raid1:%s: redirecting sector %llu"
2338 " to other mirror: %s\n",
2339 mdname(mddev),
2340 (unsigned long long)r1_bio->sector,
2341 bdevname(rdev->bdev, b));
Kent Overstreet4f024f32013-10-11 15:44:27 -07002342 bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
NeilBrown62096bc2011-07-28 11:38:13 +10002343 bio->bi_bdev = rdev->bdev;
2344 bio->bi_end_io = raid1_end_read_request;
2345 bio->bi_rw = READ | do_sync;
2346 bio->bi_private = r1_bio;
2347 if (max_sectors < r1_bio->sectors) {
2348 /* Drat - have to split this up more */
2349 struct bio *mbio = r1_bio->master_bio;
2350 int sectors_handled = (r1_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07002351 - mbio->bi_iter.bi_sector);
NeilBrown62096bc2011-07-28 11:38:13 +10002352 r1_bio->sectors = max_sectors;
2353 spin_lock_irq(&conf->device_lock);
2354 if (mbio->bi_phys_segments == 0)
2355 mbio->bi_phys_segments = 2;
2356 else
2357 mbio->bi_phys_segments++;
2358 spin_unlock_irq(&conf->device_lock);
2359 generic_make_request(bio);
2360 bio = NULL;
2361
2362 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2363
2364 r1_bio->master_bio = mbio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002365 r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
NeilBrown62096bc2011-07-28 11:38:13 +10002366 r1_bio->state = 0;
2367 set_bit(R1BIO_ReadError, &r1_bio->state);
2368 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002369 r1_bio->sector = mbio->bi_iter.bi_sector +
2370 sectors_handled;
NeilBrown62096bc2011-07-28 11:38:13 +10002371
2372 goto read_more;
2373 } else
2374 generic_make_request(bio);
2375 }
2376}
2377
Shaohua Li4ed87312012-10-11 13:34:00 +11002378static void raid1d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379{
Shaohua Li4ed87312012-10-11 13:34:00 +11002380 struct mddev *mddev = thread->mddev;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002381 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 unsigned long flags;
NeilBrowne8096362011-10-11 16:49:05 +11002383 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002385 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386
2387 md_check_recovery(mddev);
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002388
NeilBrown55ce74d2015-08-14 11:11:10 +10002389 if (!list_empty_careful(&conf->bio_end_io_list) &&
2390 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2391 LIST_HEAD(tmp);
2392 spin_lock_irqsave(&conf->device_lock, flags);
2393 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2394 list_add(&tmp, &conf->bio_end_io_list);
2395 list_del_init(&conf->bio_end_io_list);
2396 }
2397 spin_unlock_irqrestore(&conf->device_lock, flags);
2398 while (!list_empty(&tmp)) {
Mikulas Patockaa4527442015-10-01 15:17:43 -04002399 r1_bio = list_first_entry(&tmp, struct r1bio,
2400 retry_list);
NeilBrown55ce74d2015-08-14 11:11:10 +10002401 list_del(&r1_bio->retry_list);
NeilBrownbd8688a2015-10-24 16:02:16 +11002402 if (mddev->degraded)
2403 set_bit(R1BIO_Degraded, &r1_bio->state);
2404 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2405 close_write(r1_bio);
NeilBrown55ce74d2015-08-14 11:11:10 +10002406 raid_end_bio_io(r1_bio);
2407 }
2408 }
2409
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002410 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002412
NeilBrown0021b7b2012-07-31 09:08:14 +02002413 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002414
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002416 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002417 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002419 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002420 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08002422 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 spin_unlock_irqrestore(&conf->device_lock, flags);
2424
2425 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002426 conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10002427 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002428 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002429 test_bit(R1BIO_WriteError, &r1_bio->state))
2430 handle_sync_write_finished(conf, r1_bio);
2431 else
NeilBrown4367af52011-07-28 11:31:49 +10002432 sync_request_write(mddev, r1_bio);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002433 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002434 test_bit(R1BIO_WriteError, &r1_bio->state))
2435 handle_write_finished(conf, r1_bio);
2436 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2437 handle_read_error(conf, r1_bio);
2438 else
NeilBrownd2eb35a2011-07-28 11:31:48 +10002439 /* just a partial read to be scheduled from separate
2440 * context
2441 */
2442 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
NeilBrown62096bc2011-07-28 11:38:13 +10002443
NeilBrown1d9d5242009-10-16 15:55:32 +11002444 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002445 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2446 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002448 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449}
2450
NeilBrowne8096362011-10-11 16:49:05 +11002451static int init_resync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452{
2453 int buffs;
2454
2455 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002456 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2458 conf->poolinfo);
2459 if (!conf->r1buf_pool)
2460 return -ENOMEM;
2461 conf->next_resync = 0;
2462 return 0;
2463}
2464
2465/*
2466 * perform a "sync" on one "block"
2467 *
2468 * We need to make sure that no normal I/O request - particularly write
2469 * requests - conflict with active sync requests.
2470 *
2471 * This is achieved by tracking pending requests and a 'barrier' concept
2472 * that can be installed to exclude normal IO requests.
2473 */
2474
Shaohua Li849674e2016-01-20 13:52:20 -08002475static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2476 int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477{
NeilBrowne8096362011-10-11 16:49:05 +11002478 struct r1conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002479 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 struct bio *bio;
2481 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08002482 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08002484 int wonly = -1;
2485 int write_targets = 0, read_targets = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11002486 sector_t sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07002487 int still_degraded = 0;
NeilBrown06f60382011-07-28 11:31:48 +10002488 int good_sectors = RESYNC_SECTORS;
2489 int min_bad = 0; /* number of sectors that are bad in all devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490
2491 if (!conf->r1buf_pool)
2492 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002493 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494
Andre Noll58c0fed2009-03-31 14:33:13 +11002495 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002497 /* If we aborted, we need to abort the
2498 * sync on the 'current' bitmap chunk (there will
2499 * only be one in raid1 resync.
2500 * We can find the current addess in mddev->curr_resync
2501 */
NeilBrown6a806c52005-07-15 03:56:35 -07002502 if (mddev->curr_resync < max_sector) /* aborted */
2503 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07002504 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07002505 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07002506 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07002507
2508 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 close_sync(conf);
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002510
2511 if (mddev_is_clustered(mddev)) {
2512 conf->cluster_sync_low = 0;
2513 conf->cluster_sync_high = 0;
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 return 0;
2516 }
2517
NeilBrown07d84d102006-06-26 00:27:56 -07002518 if (mddev->bitmap == NULL &&
2519 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07002520 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07002521 conf->fullsync == 0) {
2522 *skipped = 1;
2523 return max_sector - sector_nr;
2524 }
NeilBrown6394cca2006-08-27 01:23:50 -07002525 /* before building a request, check if we can skip these blocks..
2526 * This call the bitmap_start_sync doesn't actually record anything
2527 */
NeilBrowne3b97032005-08-04 12:53:34 -07002528 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de485f2005-11-08 21:39:38 -08002529 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002530 /* We can skip this block, and probably several more */
2531 *skipped = 1;
2532 return sync_blocks;
2533 }
NeilBrown17999be2006-01-06 00:20:12 -08002534
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002535 /* we are incrementing sector_nr below. To be safe, we check against
2536 * sector_nr + two times RESYNC_SECTORS
2537 */
2538
2539 bitmap_cond_end_sync(mddev->bitmap, sector_nr,
2540 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
NeilBrown1c4588e2010-10-26 17:41:22 +11002541 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown17999be2006-01-06 00:20:12 -08002542
NeilBrownc2fd4c92014-09-10 16:01:24 +10002543 raise_barrier(conf, sector_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
NeilBrown3e198f72006-01-06 00:20:21 -08002545 rcu_read_lock();
2546 /*
2547 * If we get a correctably read error during resync or recovery,
2548 * we might want to read from a different device. So we
2549 * flag all drives that could conceivably be read from for READ,
2550 * and any others (which will be non-In_sync devices) for WRITE.
2551 * If a read fails, we try reading from something else for which READ
2552 * is OK.
2553 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 r1_bio->mddev = mddev;
2556 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07002557 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
NeilBrown8f19ccb2011-12-23 10:17:56 +11002560 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002561 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 bio = r1_bio->bios[i];
Kent Overstreet2aabaa62012-09-11 11:26:12 -07002563 bio_reset(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
NeilBrown3e198f72006-01-06 00:20:21 -08002565 rdev = rcu_dereference(conf->mirrors[i].rdev);
2566 if (rdev == NULL ||
NeilBrown06f60382011-07-28 11:31:48 +10002567 test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11002568 if (i < conf->raid_disks)
2569 still_degraded = 1;
NeilBrown3e198f72006-01-06 00:20:21 -08002570 } else if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 bio->bi_rw = WRITE;
2572 bio->bi_end_io = end_sync_write;
2573 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08002574 } else {
2575 /* may need to read from here */
NeilBrown06f60382011-07-28 11:31:48 +10002576 sector_t first_bad = MaxSector;
2577 int bad_sectors;
2578
2579 if (is_badblock(rdev, sector_nr, good_sectors,
2580 &first_bad, &bad_sectors)) {
2581 if (first_bad > sector_nr)
2582 good_sectors = first_bad - sector_nr;
2583 else {
2584 bad_sectors -= (sector_nr - first_bad);
2585 if (min_bad == 0 ||
2586 min_bad > bad_sectors)
2587 min_bad = bad_sectors;
2588 }
NeilBrown3e198f72006-01-06 00:20:21 -08002589 }
NeilBrown06f60382011-07-28 11:31:48 +10002590 if (sector_nr < first_bad) {
2591 if (test_bit(WriteMostly, &rdev->flags)) {
2592 if (wonly < 0)
2593 wonly = i;
2594 } else {
2595 if (disk < 0)
2596 disk = i;
2597 }
2598 bio->bi_rw = READ;
2599 bio->bi_end_io = end_sync_read;
2600 read_targets++;
Alexander Lyakasd57368a2012-07-17 13:17:55 +03002601 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2602 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2603 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2604 /*
2605 * The device is suitable for reading (InSync),
2606 * but has bad block(s) here. Let's try to correct them,
2607 * if we are doing resync or repair. Otherwise, leave
2608 * this device alone for this sync request.
2609 */
2610 bio->bi_rw = WRITE;
2611 bio->bi_end_io = end_sync_write;
2612 write_targets++;
NeilBrown06f60382011-07-28 11:31:48 +10002613 }
NeilBrown3e198f72006-01-06 00:20:21 -08002614 }
NeilBrown06f60382011-07-28 11:31:48 +10002615 if (bio->bi_end_io) {
2616 atomic_inc(&rdev->nr_pending);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002617 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
NeilBrown06f60382011-07-28 11:31:48 +10002618 bio->bi_bdev = rdev->bdev;
2619 bio->bi_private = r1_bio;
2620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 }
NeilBrown3e198f72006-01-06 00:20:21 -08002622 rcu_read_unlock();
2623 if (disk < 0)
2624 disk = wonly;
2625 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07002626
NeilBrown06f60382011-07-28 11:31:48 +10002627 if (read_targets == 0 && min_bad > 0) {
2628 /* These sectors are bad on all InSync devices, so we
2629 * need to mark them bad on all write targets
2630 */
2631 int ok = 1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002632 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
NeilBrown06f60382011-07-28 11:31:48 +10002633 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
majianpenga42f9d82012-04-02 01:04:19 +10002634 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown06f60382011-07-28 11:31:48 +10002635 ok = rdev_set_badblocks(rdev, sector_nr,
2636 min_bad, 0
2637 ) && ok;
2638 }
2639 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2640 *skipped = 1;
2641 put_buf(r1_bio);
2642
2643 if (!ok) {
2644 /* Cannot record the badblocks, so need to
2645 * abort the resync.
2646 * If there are multiple read targets, could just
2647 * fail the really bad ones ???
2648 */
2649 conf->recovery_disabled = mddev->recovery_disabled;
2650 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2651 return 0;
2652 } else
2653 return min_bad;
2654
2655 }
2656 if (min_bad > 0 && min_bad < good_sectors) {
2657 /* only resync enough to reach the next bad->good
2658 * transition */
2659 good_sectors = min_bad;
2660 }
2661
NeilBrown3e198f72006-01-06 00:20:21 -08002662 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2663 /* extra read targets are also write targets */
2664 write_targets += read_targets-1;
2665
2666 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 /* There is nowhere to write, so all non-sync
2668 * drives must be failed - so we are finished
2669 */
NeilBrownb7219cc2012-07-31 10:05:34 +10002670 sector_t rv;
2671 if (min_bad > 0)
2672 max_sector = sector_nr + min_bad;
2673 rv = max_sector - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07002674 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 return rv;
2677 }
2678
NeilBrownc6207272008-02-06 01:39:52 -08002679 if (max_sector > mddev->resync_max)
2680 max_sector = mddev->resync_max; /* Don't do IO beyond here */
NeilBrown06f60382011-07-28 11:31:48 +10002681 if (max_sector > sector_nr + good_sectors)
2682 max_sector = sector_nr + good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07002684 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 do {
2686 struct page *page;
2687 int len = PAGE_SIZE;
2688 if (sector_nr + (len>>9) > max_sector)
2689 len = (max_sector - sector_nr) << 9;
2690 if (len == 0)
2691 break;
NeilBrown6a806c52005-07-15 03:56:35 -07002692 if (sync_blocks == 0) {
2693 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de485f2005-11-08 21:39:38 -08002694 &sync_blocks, still_degraded) &&
2695 !conf->fullsync &&
2696 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07002697 break;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002698 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
NeilBrown7571ae82010-10-07 11:54:46 +11002699 if ((len >> 9) > sync_blocks)
NeilBrown6a806c52005-07-15 03:56:35 -07002700 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07002701 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002702
NeilBrown8f19ccb2011-12-23 10:17:56 +11002703 for (i = 0 ; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 bio = r1_bio->bios[i];
2705 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08002706 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 if (bio_add_page(bio, page, len, 0) == 0) {
2708 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08002709 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 while (i > 0) {
2711 i--;
2712 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07002713 if (bio->bi_end_io==NULL)
2714 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 /* remove last page from this bio */
2716 bio->bi_vcnt--;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002717 bio->bi_iter.bi_size -= len;
Jens Axboeb7c44ed2015-07-24 12:37:59 -06002718 bio_clear_flag(bio, BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 }
2720 goto bio_full;
2721 }
2722 }
2723 }
2724 nr_sectors += len>>9;
2725 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07002726 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2728 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 r1_bio->sectors = nr_sectors;
2730
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002731 if (mddev_is_clustered(mddev) &&
2732 conf->cluster_sync_high < sector_nr + nr_sectors) {
2733 conf->cluster_sync_low = mddev->curr_resync_completed;
2734 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2735 /* Send resync message */
2736 md_cluster_ops->resync_info_update(mddev,
2737 conf->cluster_sync_low,
2738 conf->cluster_sync_high);
2739 }
2740
NeilBrownd11c1712006-01-06 00:20:26 -08002741 /* For a user-requested sync, we read all readable devices and do a
2742 * compare
2743 */
2744 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2745 atomic_set(&r1_bio->remaining, read_targets);
NeilBrown2d4f4f32012-07-09 11:34:13 +10002746 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
NeilBrownd11c1712006-01-06 00:20:26 -08002747 bio = r1_bio->bios[i];
2748 if (bio->bi_end_io == end_sync_read) {
NeilBrown2d4f4f32012-07-09 11:34:13 +10002749 read_targets--;
NeilBrownddac7c72006-08-31 21:27:36 -07002750 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002751 generic_make_request(bio);
2752 }
2753 }
2754 } else {
2755 atomic_set(&r1_bio->remaining, 1);
2756 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07002757 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002758 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759
NeilBrownd11c1712006-01-06 00:20:26 -08002760 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 return nr_sectors;
2762}
2763
NeilBrownfd01b882011-10-11 16:47:53 +11002764static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07002765{
2766 if (sectors)
2767 return sectors;
2768
2769 return mddev->dev_sectors;
2770}
2771
NeilBrowne8096362011-10-11 16:49:05 +11002772static struct r1conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773{
NeilBrowne8096362011-10-11 16:49:05 +11002774 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002775 int i;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002776 struct raid1_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11002777 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002778 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
NeilBrowne8096362011-10-11 16:49:05 +11002780 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11002782 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002784 conf->mirrors = kzalloc(sizeof(struct raid1_info)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002785 * mddev->raid_disks * 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 GFP_KERNEL);
2787 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11002788 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789
NeilBrownddaf22a2006-01-06 00:20:19 -08002790 conf->tmppage = alloc_page(GFP_KERNEL);
2791 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11002792 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08002793
NeilBrown709ae482009-12-14 12:49:51 +11002794 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11002796 goto abort;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002797 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2799 r1bio_pool_free,
2800 conf->poolinfo);
2801 if (!conf->r1bio_pool)
NeilBrown709ae482009-12-14 12:49:51 +11002802 goto abort;
2803
NeilBrowned9bfdf2009-10-16 15:55:44 +11002804 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805
NeilBrownc19d5792011-12-23 10:17:57 +11002806 err = -EINVAL;
Neil Browne7e72bf2008-05-14 16:05:54 -07002807 spin_lock_init(&conf->device_lock);
NeilBrowndafb20f2012-03-19 12:46:39 +11002808 rdev_for_each(rdev, mddev) {
NeilBrownaba336b2012-05-31 15:39:11 +10002809 struct request_queue *q;
NeilBrown709ae482009-12-14 12:49:51 +11002810 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 if (disk_idx >= mddev->raid_disks
2812 || disk_idx < 0)
2813 continue;
NeilBrownc19d5792011-12-23 10:17:57 +11002814 if (test_bit(Replacement, &rdev->flags))
NeilBrown02b898f2012-10-31 11:42:03 +11002815 disk = conf->mirrors + mddev->raid_disks + disk_idx;
NeilBrownc19d5792011-12-23 10:17:57 +11002816 else
2817 disk = conf->mirrors + disk_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818
NeilBrownc19d5792011-12-23 10:17:57 +11002819 if (disk->rdev)
2820 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 disk->rdev = rdev;
NeilBrownaba336b2012-05-31 15:39:11 +10002822 q = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823
2824 disk->head_position = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +10002825 disk->seq_start = MaxSector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 }
2827 conf->raid_disks = mddev->raid_disks;
2828 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 INIT_LIST_HEAD(&conf->retry_list);
NeilBrown55ce74d2015-08-14 11:11:10 +10002830 INIT_LIST_HEAD(&conf->bio_end_io_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831
2832 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08002833 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834
NeilBrown191ea9b2005-06-21 17:17:23 -07002835 bio_list_init(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +11002836 conf->pending_count = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11002837 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown191ea9b2005-06-21 17:17:23 -07002838
majianpeng79ef3a82013-11-15 14:55:02 +08002839 conf->start_next_window = MaxSector;
2840 conf->current_window_requests = conf->next_window_requests = 0;
2841
NeilBrownc19d5792011-12-23 10:17:57 +11002842 err = -EIO;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002843 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844
2845 disk = conf->mirrors + i;
2846
NeilBrownc19d5792011-12-23 10:17:57 +11002847 if (i < conf->raid_disks &&
2848 disk[conf->raid_disks].rdev) {
2849 /* This slot has a replacement. */
2850 if (!disk->rdev) {
2851 /* No original, just make the replacement
2852 * a recovering spare
2853 */
2854 disk->rdev =
2855 disk[conf->raid_disks].rdev;
2856 disk[conf->raid_disks].rdev = NULL;
2857 } else if (!test_bit(In_sync, &disk->rdev->flags))
2858 /* Original is not in_sync - bad */
2859 goto abort;
2860 }
2861
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002862 if (!disk->rdev ||
2863 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 disk->head_position = 0;
Jonathan Brassow4f0a5e02012-05-22 13:55:31 +10002865 if (disk->rdev &&
2866 (disk->rdev->saved_raid_disk < 0))
NeilBrown918f0232007-08-22 14:01:52 -07002867 conf->fullsync = 1;
Shaohua Libe4d3282012-07-31 10:03:53 +10002868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 }
NeilBrown709ae482009-12-14 12:49:51 +11002870
NeilBrown709ae482009-12-14 12:49:51 +11002871 err = -ENOMEM;
NeilBrown02326052012-07-03 15:56:52 +10002872 conf->thread = md_register_thread(raid1d, mddev, "raid1");
NeilBrown709ae482009-12-14 12:49:51 +11002873 if (!conf->thread) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002874 printk(KERN_ERR
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002875 "md/raid1:%s: couldn't allocate thread\n",
NeilBrown191ea9b2005-06-21 17:17:23 -07002876 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11002877 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002879
NeilBrown709ae482009-12-14 12:49:51 +11002880 return conf;
2881
2882 abort:
2883 if (conf) {
Julia Lawall644df1a2015-09-13 14:15:10 +02002884 mempool_destroy(conf->r1bio_pool);
NeilBrown709ae482009-12-14 12:49:51 +11002885 kfree(conf->mirrors);
2886 safe_put_page(conf->tmppage);
2887 kfree(conf->poolinfo);
2888 kfree(conf);
2889 }
2890 return ERR_PTR(err);
2891}
2892
NeilBrownafa0f552014-12-15 12:56:58 +11002893static void raid1_free(struct mddev *mddev, void *priv);
Shaohua Li849674e2016-01-20 13:52:20 -08002894static int raid1_run(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002895{
NeilBrowne8096362011-10-11 16:49:05 +11002896 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002897 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11002898 struct md_rdev *rdev;
majianpeng5220ea12012-04-02 09:48:38 +10002899 int ret;
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002900 bool discard_supported = false;
NeilBrown709ae482009-12-14 12:49:51 +11002901
2902 if (mddev->level != 1) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002903 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
NeilBrown709ae482009-12-14 12:49:51 +11002904 mdname(mddev), mddev->level);
2905 return -EIO;
2906 }
2907 if (mddev->reshape_position != MaxSector) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002908 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
NeilBrown709ae482009-12-14 12:49:51 +11002909 mdname(mddev));
2910 return -EIO;
2911 }
2912 /*
2913 * copy the already verified devices into our private RAID1
2914 * bookkeeping area. [whatever we allocate in run(),
NeilBrownafa0f552014-12-15 12:56:58 +11002915 * should be freed in raid1_free()]
NeilBrown709ae482009-12-14 12:49:51 +11002916 */
2917 if (mddev->private == NULL)
2918 conf = setup_conf(mddev);
2919 else
2920 conf = mddev->private;
2921
2922 if (IS_ERR(conf))
2923 return PTR_ERR(conf);
2924
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11002925 if (mddev->queue)
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07002926 blk_queue_max_write_same_sectors(mddev->queue, 0);
2927
NeilBrowndafb20f2012-03-19 12:46:39 +11002928 rdev_for_each(rdev, mddev) {
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002929 if (!mddev->gendisk)
2930 continue;
NeilBrown709ae482009-12-14 12:49:51 +11002931 disk_stack_limits(mddev->gendisk, rdev->bdev,
2932 rdev->data_offset << 9);
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002933 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
2934 discard_supported = true;
NeilBrown709ae482009-12-14 12:49:51 +11002935 }
2936
2937 mddev->degraded = 0;
2938 for (i=0; i < conf->raid_disks; i++)
2939 if (conf->mirrors[i].rdev == NULL ||
2940 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2941 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2942 mddev->degraded++;
2943
2944 if (conf->raid_disks - mddev->degraded == 1)
2945 mddev->recovery_cp = MaxSector;
2946
Andre Noll8c6ac862009-06-18 08:48:06 +10002947 if (mddev->recovery_cp != MaxSector)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002948 printk(KERN_NOTICE "md/raid1:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10002949 " -- starting background reconstruction\n",
2950 mdname(mddev));
NeilBrownf72ffdd2014-09-30 14:23:59 +10002951 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002952 "md/raid1:%s: active with %d out of %d mirrors\n",
NeilBrownf72ffdd2014-09-30 14:23:59 +10002953 mdname(mddev), mddev->raid_disks - mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11002955
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 /*
2957 * Ok, everything is just fine now
2958 */
NeilBrown709ae482009-12-14 12:49:51 +11002959 mddev->thread = conf->thread;
2960 conf->thread = NULL;
2961 mddev->private = conf;
2962
Dan Williams1f403622009-03-31 14:59:03 +11002963 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002965 if (mddev->queue) {
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002966 if (discard_supported)
2967 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
2968 mddev->queue);
2969 else
2970 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
2971 mddev->queue);
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002972 }
majianpeng5220ea12012-04-02 09:48:38 +10002973
2974 ret = md_integrity_register(mddev);
NeilBrown5aa61f42014-12-15 12:56:57 +11002975 if (ret) {
2976 md_unregister_thread(&mddev->thread);
NeilBrownafa0f552014-12-15 12:56:58 +11002977 raid1_free(mddev, conf);
NeilBrown5aa61f42014-12-15 12:56:57 +11002978 }
majianpeng5220ea12012-04-02 09:48:38 +10002979 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980}
2981
NeilBrownafa0f552014-12-15 12:56:58 +11002982static void raid1_free(struct mddev *mddev, void *priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983{
NeilBrownafa0f552014-12-15 12:56:58 +11002984 struct r1conf *conf = priv;
NeilBrown4b6d2872005-09-09 16:23:47 -07002985
Julia Lawall644df1a2015-09-13 14:15:10 +02002986 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002987 kfree(conf->mirrors);
Hirokazu Takahashi0fea7ed2013-04-24 11:42:44 +10002988 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002989 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991}
2992
NeilBrownfd01b882011-10-11 16:47:53 +11002993static int raid1_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994{
2995 /* no resync is happening, and there is enough space
2996 * on all devices, so we can resize.
2997 * We need to make sure resync covers any new space.
2998 * If the array is shrinking we should possibly wait until
2999 * any io in the removed space completes, but it hardly seems
3000 * worth it.
3001 */
NeilBrowna4a61252012-05-22 13:55:27 +10003002 sector_t newsize = raid1_size(mddev, sectors, 0);
3003 if (mddev->external_size &&
3004 mddev->array_sectors > newsize)
Dan Williamsb522adc2009-03-31 15:00:31 +11003005 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10003006 if (mddev->bitmap) {
3007 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
3008 if (ret)
3009 return ret;
3010 }
3011 md_set_array_sectors(mddev, newsize);
Andre Nollf233ea52008-07-21 17:05:22 +10003012 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10003013 revalidate_disk(mddev->gendisk);
Dan Williamsb522adc2009-03-31 15:00:31 +11003014 if (sectors > mddev->dev_sectors &&
NeilBrownb0986362011-05-11 15:52:21 +10003015 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11003016 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3018 }
Dan Williamsb522adc2009-03-31 15:00:31 +11003019 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07003020 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 return 0;
3022}
3023
NeilBrownfd01b882011-10-11 16:47:53 +11003024static int raid1_reshape(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025{
3026 /* We need to:
3027 * 1/ resize the r1bio_pool
3028 * 2/ resize conf->mirrors
3029 *
3030 * We allocate a new r1bio_pool if we can.
3031 * Then raise a device barrier and wait until all IO stops.
3032 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07003033 *
3034 * At the same time, we "pack" the devices so that all the missing
3035 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 */
3037 mempool_t *newpool, *oldpool;
3038 struct pool_info *newpoolinfo;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10003039 struct raid1_info *newmirrors;
NeilBrowne8096362011-10-11 16:49:05 +11003040 struct r1conf *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08003041 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07003042 unsigned long flags;
Dan Williamsb5470dc2008-06-27 21:44:04 -07003043 int d, d2, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044
NeilBrown63c70c42006-03-27 01:18:13 -08003045 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10003046 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08003047 mddev->layout != mddev->new_layout ||
3048 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10003049 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08003050 mddev->new_layout = mddev->layout;
3051 mddev->new_level = mddev->level;
3052 return -EINVAL;
3053 }
3054
Goldwyn Rodrigues28c1b9f2015-10-22 16:01:25 +11003055 if (!mddev_is_clustered(mddev)) {
3056 err = md_allow_write(mddev);
3057 if (err)
3058 return err;
3059 }
NeilBrown2a2275d2007-01-26 00:57:11 -08003060
NeilBrown63c70c42006-03-27 01:18:13 -08003061 raid_disks = mddev->raid_disks + mddev->delta_disks;
3062
NeilBrown6ea9c072005-06-21 17:17:09 -07003063 if (raid_disks < conf->raid_disks) {
3064 cnt=0;
3065 for (d= 0; d < conf->raid_disks; d++)
3066 if (conf->mirrors[d].rdev)
3067 cnt++;
3068 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07003070 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071
3072 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3073 if (!newpoolinfo)
3074 return -ENOMEM;
3075 newpoolinfo->mddev = mddev;
NeilBrown8f19ccb2011-12-23 10:17:56 +11003076 newpoolinfo->raid_disks = raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077
3078 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3079 r1bio_pool_free, newpoolinfo);
3080 if (!newpool) {
3081 kfree(newpoolinfo);
3082 return -ENOMEM;
3083 }
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10003084 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
NeilBrown8f19ccb2011-12-23 10:17:56 +11003085 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 if (!newmirrors) {
3087 kfree(newpoolinfo);
3088 mempool_destroy(newpool);
3089 return -ENOMEM;
3090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091
NeilBrowne2d59922013-06-12 11:01:22 +10003092 freeze_array(conf, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093
3094 /* ok, everything is stopped */
3095 oldpool = conf->r1bio_pool;
3096 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07003097
NeilBrowna88aa782007-08-22 14:01:53 -07003098 for (d = d2 = 0; d < conf->raid_disks; d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11003099 struct md_rdev *rdev = conf->mirrors[d].rdev;
NeilBrowna88aa782007-08-22 14:01:53 -07003100 if (rdev && rdev->raid_disk != d2) {
Namhyung Kim36fad852011-07-27 11:00:36 +10003101 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna88aa782007-08-22 14:01:53 -07003102 rdev->raid_disk = d2;
Namhyung Kim36fad852011-07-27 11:00:36 +10003103 sysfs_unlink_rdev(mddev, rdev);
3104 if (sysfs_link_rdev(mddev, rdev))
NeilBrowna88aa782007-08-22 14:01:53 -07003105 printk(KERN_WARNING
Namhyung Kim36fad852011-07-27 11:00:36 +10003106 "md/raid1:%s: cannot register rd%d\n",
3107 mdname(mddev), rdev->raid_disk);
NeilBrown6ea9c072005-06-21 17:17:09 -07003108 }
NeilBrowna88aa782007-08-22 14:01:53 -07003109 if (rdev)
3110 newmirrors[d2++].rdev = rdev;
3111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112 kfree(conf->mirrors);
3113 conf->mirrors = newmirrors;
3114 kfree(conf->poolinfo);
3115 conf->poolinfo = newpoolinfo;
3116
NeilBrownc04be0a2006-10-03 01:15:53 -07003117 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07003119 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08003121 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122
NeilBrowne2d59922013-06-12 11:01:22 +10003123 unfreeze_array(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124
NeilBrown985ca972015-07-06 12:26:57 +10003125 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3127 md_wakeup_thread(mddev->thread);
3128
3129 mempool_destroy(oldpool);
3130 return 0;
3131}
3132
NeilBrownfd01b882011-10-11 16:47:53 +11003133static void raid1_quiesce(struct mddev *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07003134{
NeilBrowne8096362011-10-11 16:49:05 +11003135 struct r1conf *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07003136
3137 switch(state) {
NeilBrown6eef4b22009-12-14 12:49:51 +11003138 case 2: /* wake for suspend */
3139 wake_up(&conf->wait_barrier);
3140 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07003141 case 1:
majianpeng07169fd2013-11-14 15:16:18 +11003142 freeze_array(conf, 0);
NeilBrown36fa3062005-09-09 16:23:45 -07003143 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07003144 case 0:
majianpeng07169fd2013-11-14 15:16:18 +11003145 unfreeze_array(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07003146 break;
3147 }
NeilBrown36fa3062005-09-09 16:23:45 -07003148}
3149
NeilBrownfd01b882011-10-11 16:47:53 +11003150static void *raid1_takeover(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11003151{
3152 /* raid1 can take over:
3153 * raid5 with 2 devices, any layout or chunk size
3154 */
3155 if (mddev->level == 5 && mddev->raid_disks == 2) {
NeilBrowne8096362011-10-11 16:49:05 +11003156 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11003157 mddev->new_level = 1;
3158 mddev->new_layout = 0;
3159 mddev->new_chunk_sectors = 0;
3160 conf = setup_conf(mddev);
3161 if (!IS_ERR(conf))
majianpeng07169fd2013-11-14 15:16:18 +11003162 /* Array must appear to be quiesced */
3163 conf->array_frozen = 1;
NeilBrown709ae482009-12-14 12:49:51 +11003164 return conf;
3165 }
3166 return ERR_PTR(-EINVAL);
3167}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168
NeilBrown84fc4b52011-10-11 16:49:58 +11003169static struct md_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170{
3171 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08003172 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 .owner = THIS_MODULE,
Shaohua Li849674e2016-01-20 13:52:20 -08003174 .make_request = raid1_make_request,
3175 .run = raid1_run,
NeilBrownafa0f552014-12-15 12:56:58 +11003176 .free = raid1_free,
Shaohua Li849674e2016-01-20 13:52:20 -08003177 .status = raid1_status,
3178 .error_handler = raid1_error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 .hot_add_disk = raid1_add_disk,
3180 .hot_remove_disk= raid1_remove_disk,
3181 .spare_active = raid1_spare_active,
Shaohua Li849674e2016-01-20 13:52:20 -08003182 .sync_request = raid1_sync_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07003184 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08003185 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07003186 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11003187 .takeover = raid1_takeover,
NeilBrown5c675f82014-12-15 12:56:56 +11003188 .congested = raid1_congested,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189};
3190
3191static int __init raid_init(void)
3192{
NeilBrown2604b702006-01-06 00:20:36 -08003193 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194}
3195
3196static void raid_exit(void)
3197{
NeilBrown2604b702006-01-06 00:20:36 -08003198 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199}
3200
3201module_init(raid_init);
3202module_exit(raid_exit);
3203MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11003204MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08003206MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08003207MODULE_ALIAS("md-level-1");
NeilBrown34db0cd2011-10-11 16:50:01 +11003208
3209module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);