blob: 21dc00eb1989fd981e5c843097ca5315ca7d258d [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;
NeilBrowne8096362011-10-11 16:49:05 +1100322 struct r1conf *conf = r1_bio->mddev->private;
NeilBrowne5872d52016-06-02 16:19:52 +1000323 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 /*
326 * this branch is our 'one mirror IO has finished' event handler:
327 */
NeilBrowne5872d52016-06-02 16:19:52 +1000328 update_head_pos(r1_bio->read_disk, r1_bio);
NeilBrownddaf22a2006-01-06 00:20:19 -0800329
NeilBrowndd00a992007-05-10 03:15:50 -0700330 if (uptodate)
331 set_bit(R1BIO_Uptodate, &r1_bio->state);
332 else {
333 /* If all other devices have failed, we want to return
334 * the error upwards rather than fail the last device.
335 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 */
NeilBrowndd00a992007-05-10 03:15:50 -0700337 unsigned long flags;
338 spin_lock_irqsave(&conf->device_lock, flags);
339 if (r1_bio->mddev->degraded == conf->raid_disks ||
340 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
NeilBrowne5872d52016-06-02 16:19:52 +1000341 test_bit(In_sync, &rdev->flags)))
NeilBrowndd00a992007-05-10 03:15:50 -0700342 uptodate = 1;
343 spin_unlock_irqrestore(&conf->device_lock, flags);
344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100346 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 raid_end_bio_io(r1_bio);
NeilBrowne5872d52016-06-02 16:19:52 +1000348 rdev_dec_pending(rdev, conf->mddev);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100349 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 /*
351 * oops, read error:
352 */
353 char b[BDEVNAME_SIZE];
Christian Dietrich8bda4702011-07-27 11:00:36 +1000354 printk_ratelimited(
355 KERN_ERR "md/raid1:%s: %s: "
356 "rescheduling sector %llu\n",
357 mdname(conf->mddev),
NeilBrowne5872d52016-06-02 16:19:52 +1000358 bdevname(rdev->bdev,
Christian Dietrich8bda4702011-07-27 11:00:36 +1000359 b),
360 (unsigned long long)r1_bio->sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000361 set_bit(R1BIO_ReadError, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 reschedule_retry(r1_bio);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100363 /* don't drop the reference on read_disk yet */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365}
366
NeilBrown9f2c9d12011-10-11 16:48:43 +1100367static void close_write(struct r1bio *r1_bio)
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000368{
369 /* it really is the end of this request */
370 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
371 /* free extra copy of the data pages */
372 int i = r1_bio->behind_page_count;
373 while (i--)
374 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
375 kfree(r1_bio->behind_bvecs);
376 r1_bio->behind_bvecs = NULL;
377 }
378 /* clear the bitmap if all writes complete successfully */
379 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
380 r1_bio->sectors,
381 !test_bit(R1BIO_Degraded, &r1_bio->state),
382 test_bit(R1BIO_BehindIO, &r1_bio->state));
383 md_write_end(r1_bio->mddev);
384}
385
NeilBrown9f2c9d12011-10-11 16:48:43 +1100386static void r1_bio_write_done(struct r1bio *r1_bio)
NeilBrown4e780642010-10-19 12:54:01 +1100387{
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000388 if (!atomic_dec_and_test(&r1_bio->remaining))
389 return;
390
391 if (test_bit(R1BIO_WriteError, &r1_bio->state))
392 reschedule_retry(r1_bio);
393 else {
394 close_write(r1_bio);
NeilBrown4367af52011-07-28 11:31:49 +1000395 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
396 reschedule_retry(r1_bio);
397 else
398 raid_end_bio_io(r1_bio);
NeilBrown4e780642010-10-19 12:54:01 +1100399 }
400}
401
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200402static void raid1_end_write_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
NeilBrown9f2c9d12011-10-11 16:48:43 +1100404 struct r1bio *r1_bio = bio->bi_private;
NeilBrowne5872d52016-06-02 16:19:52 +1000405 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
NeilBrowne8096362011-10-11 16:49:05 +1100406 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown04b857f2006-03-09 17:33:46 -0800407 struct bio *to_put = NULL;
NeilBrowne5872d52016-06-02 16:19:52 +1000408 int mirror = find_bio_disk(r1_bio, bio);
409 struct md_rdev *rdev = conf->mirrors[mirror].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Tejun Heoe9c74692010-09-03 11:56:18 +0200411 /*
412 * 'one mirror IO has finished' event handler:
413 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200414 if (bio->bi_error) {
NeilBrowne5872d52016-06-02 16:19:52 +1000415 set_bit(WriteErrorSeen, &rdev->flags);
416 if (!test_and_set_bit(WantReplacement, &rdev->flags))
NeilBrown19d67162011-12-23 10:17:57 +1100417 set_bit(MD_RECOVERY_NEEDED, &
418 conf->mddev->recovery);
419
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000420 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +1000421 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 /*
Tejun Heoe9c74692010-09-03 11:56:18 +0200423 * Set R1BIO_Uptodate in our master bio, so that we
424 * will return a good error code for to the higher
425 * levels even if IO on some other mirrored buffer
426 * fails.
427 *
428 * The 'master' represents the composite IO operation
429 * to user-side. So if something waits for IO, then it
430 * will wait for the 'master' bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 */
NeilBrown4367af52011-07-28 11:31:49 +1000432 sector_t first_bad;
433 int bad_sectors;
434
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000435 r1_bio->bios[mirror] = NULL;
436 to_put = bio;
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300437 /*
438 * Do not set R1BIO_Uptodate if the current device is
439 * rebuilding or Faulty. This is because we cannot use
440 * such device for properly reading the data back (we could
441 * potentially use it, if the current write would have felt
442 * before rdev->recovery_offset, but for simplicity we don't
443 * check this here.
444 */
NeilBrowne5872d52016-06-02 16:19:52 +1000445 if (test_bit(In_sync, &rdev->flags) &&
446 !test_bit(Faulty, &rdev->flags))
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300447 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
NeilBrown4367af52011-07-28 11:31:49 +1000449 /* Maybe we can clear some bad blocks. */
NeilBrowne5872d52016-06-02 16:19:52 +1000450 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
NeilBrown4367af52011-07-28 11:31:49 +1000451 &first_bad, &bad_sectors)) {
452 r1_bio->bios[mirror] = IO_MADE_GOOD;
453 set_bit(R1BIO_MadeGood, &r1_bio->state);
454 }
455 }
456
Tejun Heoe9c74692010-09-03 11:56:18 +0200457 if (behind) {
NeilBrowne5872d52016-06-02 16:19:52 +1000458 if (test_bit(WriteMostly, &rdev->flags))
Tejun Heoe9c74692010-09-03 11:56:18 +0200459 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700460
Tejun Heoe9c74692010-09-03 11:56:18 +0200461 /*
462 * In behind mode, we ACK the master bio once the I/O
463 * has safely reached all non-writemostly
464 * disks. Setting the Returned bit ensures that this
465 * gets done only once -- we don't ever want to return
466 * -EIO here, instead we'll wait
467 */
468 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
469 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
470 /* Maybe we can return now */
471 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
472 struct bio *mbio = r1_bio->master_bio;
NeilBrown36a4e1f2011-10-07 14:23:17 +1100473 pr_debug("raid1: behind end write sectors"
474 " %llu-%llu\n",
Kent Overstreet4f024f32013-10-11 15:44:27 -0700475 (unsigned long long) mbio->bi_iter.bi_sector,
476 (unsigned long long) bio_end_sector(mbio) - 1);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000477 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700478 }
479 }
480 }
NeilBrown4367af52011-07-28 11:31:49 +1000481 if (r1_bio->bios[mirror] == NULL)
NeilBrowne5872d52016-06-02 16:19:52 +1000482 rdev_dec_pending(rdev, conf->mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 * Let's see if all mirrored write operations have finished
486 * already.
487 */
NeilBrownaf6d7b72011-05-11 14:51:19 +1000488 r1_bio_write_done(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700489
NeilBrown04b857f2006-03-09 17:33:46 -0800490 if (to_put)
491 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492}
493
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494/*
495 * This routine returns the disk from which the requested read should
496 * be done. There is a per-array 'next expected sequential IO' sector
497 * number - if this matches on the next IO then we use the last disk.
498 * There is also a per-disk 'last know head position' sector that is
499 * maintained from IRQ contexts, both the normal and the resync IO
500 * completion handlers update this position correctly. If there is no
501 * perfect sequential match then we pick the disk whose head is closest.
502 *
503 * If there are 2 mirrors in the same 2 devices, performance degrades
504 * because position is mirror, not device based.
505 *
506 * The rdev for the device selected will have nr_pending incremented.
507 */
NeilBrowne8096362011-10-11 16:49:05 +1100508static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000510 const sector_t this_sector = r1_bio->sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000511 int sectors;
512 int best_good_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000513 int best_disk, best_dist_disk, best_pending_disk;
514 int has_nonrot_disk;
Shaohua Libe4d3282012-07-31 10:03:53 +1000515 int disk;
NeilBrown76073052011-05-11 14:34:56 +1000516 sector_t best_dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000517 unsigned int min_pending;
NeilBrown3cb03002011-10-11 16:45:26 +1100518 struct md_rdev *rdev;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000519 int choose_first;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000520 int choose_next_idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
522 rcu_read_lock();
523 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700524 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 * device if no resync is going on, or below the resync window.
526 * We take the first readable disk when above the resync window.
527 */
528 retry:
NeilBrownd2eb35a2011-07-28 11:31:48 +1000529 sectors = r1_bio->sectors;
NeilBrown76073052011-05-11 14:34:56 +1000530 best_disk = -1;
Shaohua Li9dedf602012-07-31 10:03:53 +1000531 best_dist_disk = -1;
NeilBrown76073052011-05-11 14:34:56 +1000532 best_dist = MaxSector;
Shaohua Li9dedf602012-07-31 10:03:53 +1000533 best_pending_disk = -1;
534 min_pending = UINT_MAX;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000535 best_good_sectors = 0;
Shaohua Li9dedf602012-07-31 10:03:53 +1000536 has_nonrot_disk = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000537 choose_next_idle = 0;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000538
Goldwyn Rodrigues7d49ffc2014-08-12 10:13:19 -0500539 if ((conf->mddev->recovery_cp < this_sector + sectors) ||
540 (mddev_is_clustered(conf->mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -0500541 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
Goldwyn Rodrigues7d49ffc2014-08-12 10:13:19 -0500542 this_sector + sectors)))
543 choose_first = 1;
544 else
545 choose_first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Shaohua Libe4d3282012-07-31 10:03:53 +1000547 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
NeilBrown76073052011-05-11 14:34:56 +1000548 sector_t dist;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000549 sector_t first_bad;
550 int bad_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000551 unsigned int pending;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000552 bool nonrot;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000553
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000554 rdev = rcu_dereference(conf->mirrors[disk].rdev);
555 if (r1_bio->bios[disk] == IO_BLOCKED
556 || rdev == NULL
NeilBrown76073052011-05-11 14:34:56 +1000557 || test_bit(Faulty, &rdev->flags))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000558 continue;
NeilBrown76073052011-05-11 14:34:56 +1000559 if (!test_bit(In_sync, &rdev->flags) &&
560 rdev->recovery_offset < this_sector + sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 continue;
NeilBrown76073052011-05-11 14:34:56 +1000562 if (test_bit(WriteMostly, &rdev->flags)) {
563 /* Don't balance among write-mostly, just
564 * use the first as a last resort */
Tomáš Hodekd1901ef2015-02-23 11:00:38 +1100565 if (best_dist_disk < 0) {
NeilBrown307729c2012-01-09 01:41:51 +1100566 if (is_badblock(rdev, this_sector, sectors,
567 &first_bad, &bad_sectors)) {
Wei Fang816b0ac2016-03-21 19:18:32 +0800568 if (first_bad <= this_sector)
NeilBrown307729c2012-01-09 01:41:51 +1100569 /* Cannot use this */
570 continue;
571 best_good_sectors = first_bad - this_sector;
572 } else
573 best_good_sectors = sectors;
Tomáš Hodekd1901ef2015-02-23 11:00:38 +1100574 best_dist_disk = disk;
575 best_pending_disk = disk;
NeilBrown307729c2012-01-09 01:41:51 +1100576 }
NeilBrown76073052011-05-11 14:34:56 +1000577 continue;
578 }
579 /* This is a reasonable device to use. It might
580 * even be best.
581 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000582 if (is_badblock(rdev, this_sector, sectors,
583 &first_bad, &bad_sectors)) {
584 if (best_dist < MaxSector)
585 /* already have a better device */
586 continue;
587 if (first_bad <= this_sector) {
588 /* cannot read here. If this is the 'primary'
589 * device, then we must not read beyond
590 * bad_sectors from another device..
591 */
592 bad_sectors -= (this_sector - first_bad);
593 if (choose_first && sectors > bad_sectors)
594 sectors = bad_sectors;
595 if (best_good_sectors > sectors)
596 best_good_sectors = sectors;
597
598 } else {
599 sector_t good_sectors = first_bad - this_sector;
600 if (good_sectors > best_good_sectors) {
601 best_good_sectors = good_sectors;
602 best_disk = disk;
603 }
604 if (choose_first)
605 break;
606 }
607 continue;
608 } else
609 best_good_sectors = sectors;
610
Shaohua Li12cee5a2012-07-31 10:03:53 +1000611 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
612 has_nonrot_disk |= nonrot;
Shaohua Li9dedf602012-07-31 10:03:53 +1000613 pending = atomic_read(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000614 dist = abs(this_sector - conf->mirrors[disk].head_position);
Shaohua Li12cee5a2012-07-31 10:03:53 +1000615 if (choose_first) {
NeilBrown76073052011-05-11 14:34:56 +1000616 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 break;
618 }
Shaohua Li12cee5a2012-07-31 10:03:53 +1000619 /* Don't change to another disk for sequential reads */
620 if (conf->mirrors[disk].next_seq_sect == this_sector
621 || dist == 0) {
622 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
623 struct raid1_info *mirror = &conf->mirrors[disk];
624
625 best_disk = disk;
626 /*
627 * If buffered sequential IO size exceeds optimal
628 * iosize, check if there is idle disk. If yes, choose
629 * the idle disk. read_balance could already choose an
630 * idle disk before noticing it's a sequential IO in
631 * this disk. This doesn't matter because this disk
632 * will idle, next time it will be utilized after the
633 * first disk has IO size exceeds optimal iosize. In
634 * this way, iosize of the first disk will be optimal
635 * iosize at least. iosize of the second disk might be
636 * small, but not a big deal since when the second disk
637 * starts IO, the first disk is likely still busy.
638 */
639 if (nonrot && opt_iosize > 0 &&
640 mirror->seq_start != MaxSector &&
641 mirror->next_seq_sect > opt_iosize &&
642 mirror->next_seq_sect - opt_iosize >=
643 mirror->seq_start) {
644 choose_next_idle = 1;
645 continue;
646 }
647 break;
648 }
649 /* If device is idle, use it */
650 if (pending == 0) {
651 best_disk = disk;
652 break;
653 }
654
655 if (choose_next_idle)
656 continue;
Shaohua Li9dedf602012-07-31 10:03:53 +1000657
658 if (min_pending > pending) {
659 min_pending = pending;
660 best_pending_disk = disk;
661 }
662
NeilBrown76073052011-05-11 14:34:56 +1000663 if (dist < best_dist) {
664 best_dist = dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000665 best_dist_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 }
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Shaohua Li9dedf602012-07-31 10:03:53 +1000669 /*
670 * If all disks are rotational, choose the closest disk. If any disk is
671 * non-rotational, choose the disk with less pending request even the
672 * disk is rotational, which might/might not be optimal for raids with
673 * mixed ratation/non-rotational disks depending on workload.
674 */
675 if (best_disk == -1) {
676 if (has_nonrot_disk)
677 best_disk = best_pending_disk;
678 else
679 best_disk = best_dist_disk;
680 }
681
NeilBrown76073052011-05-11 14:34:56 +1000682 if (best_disk >= 0) {
683 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700684 if (!rdev)
685 goto retry;
686 atomic_inc(&rdev->nr_pending);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000687 sectors = best_good_sectors;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000688
689 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
690 conf->mirrors[best_disk].seq_start = this_sector;
691
Shaohua Libe4d3282012-07-31 10:03:53 +1000692 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 }
694 rcu_read_unlock();
NeilBrownd2eb35a2011-07-28 11:31:48 +1000695 *max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
NeilBrown76073052011-05-11 14:34:56 +1000697 return best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698}
699
NeilBrown5c675f82014-12-15 12:56:56 +1100700static int raid1_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700701{
NeilBrowne8096362011-10-11 16:49:05 +1100702 struct r1conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700703 int i, ret = 0;
704
Tejun Heo44522262015-05-22 17:13:26 -0400705 if ((bits & (1 << WB_async_congested)) &&
NeilBrown34db0cd2011-10-11 16:50:01 +1100706 conf->pending_count >= max_queued_requests)
707 return 1;
708
NeilBrown0d129222006-10-03 01:15:54 -0700709 rcu_read_lock();
NeilBrownf53e29f2012-02-13 14:24:05 +1100710 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100711 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700712 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200713 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700714
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500715 BUG_ON(!q);
716
NeilBrown0d129222006-10-03 01:15:54 -0700717 /* Note the '|| 1' - when read_balance prefers
718 * non-congested targets, it can be removed
719 */
Tejun Heo44522262015-05-22 17:13:26 -0400720 if ((bits & (1 << WB_async_congested)) || 1)
NeilBrown0d129222006-10-03 01:15:54 -0700721 ret |= bdi_congested(&q->backing_dev_info, bits);
722 else
723 ret &= bdi_congested(&q->backing_dev_info, bits);
724 }
725 }
726 rcu_read_unlock();
727 return ret;
728}
NeilBrown0d129222006-10-03 01:15:54 -0700729
NeilBrowne8096362011-10-11 16:49:05 +1100730static void flush_pending_writes(struct r1conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800731{
732 /* Any writes that have been queued but are awaiting
733 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800734 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800735 spin_lock_irq(&conf->device_lock);
736
737 if (conf->pending_bio_list.head) {
738 struct bio *bio;
739 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100740 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800741 spin_unlock_irq(&conf->device_lock);
742 /* flush any pending bitmap writes to
743 * disk before proceeding w/ I/O */
744 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100745 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800746
747 while (bio) { /* submit pending writes */
748 struct bio *next = bio->bi_next;
749 bio->bi_next = NULL;
Mike Christie796a5cf2016-06-05 14:32:07 -0500750 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
Shaohua Li2ff8cc22012-10-11 13:28:54 +1100751 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
752 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200753 bio_endio(bio);
Shaohua Li2ff8cc22012-10-11 13:28:54 +1100754 else
755 generic_make_request(bio);
NeilBrowna35e63e2008-03-04 14:29:29 -0800756 bio = next;
757 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800758 } else
759 spin_unlock_irq(&conf->device_lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100760}
761
NeilBrown17999be2006-01-06 00:20:12 -0800762/* Barriers....
763 * Sometimes we need to suspend IO while we do something else,
764 * either some resync/recovery, or reconfigure the array.
765 * To do this we raise a 'barrier'.
766 * The 'barrier' is a counter that can be raised multiple times
767 * to count how many activities are happening which preclude
768 * normal IO.
769 * We can only raise the barrier if there is no pending IO.
770 * i.e. if nr_pending == 0.
771 * We choose only to raise the barrier if no-one is waiting for the
772 * barrier to go down. This means that as soon as an IO request
773 * is ready, no other operations which require a barrier will start
774 * until the IO request has had a chance.
775 *
776 * So: regular IO calls 'wait_barrier'. When that returns there
777 * is no backgroup IO happening, It must arrange to call
778 * allow_barrier when it has finished its IO.
779 * backgroup IO calls must call raise_barrier. Once that returns
780 * there is no normal IO happeing. It must arrange to call
781 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 */
NeilBrownc2fd4c92014-09-10 16:01:24 +1000783static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
785 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800786
787 /* Wait until no block IO is waiting */
788 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
Lukas Czernereed8c022012-11-30 11:42:40 +0100789 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800790
791 /* block any new IO from starting */
792 conf->barrier++;
NeilBrownc2fd4c92014-09-10 16:01:24 +1000793 conf->next_resync = sector_nr;
NeilBrown17999be2006-01-06 00:20:12 -0800794
majianpeng79ef3a82013-11-15 14:55:02 +0800795 /* For these conditions we must wait:
796 * A: while the array is in frozen state
797 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
798 * the max count which allowed.
799 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
800 * next resync will reach to the window which normal bios are
801 * handling.
NeilBrown2f73d3c2014-09-10 15:01:49 +1000802 * D: while there are any active requests in the current window.
majianpeng79ef3a82013-11-15 14:55:02 +0800803 */
NeilBrown17999be2006-01-06 00:20:12 -0800804 wait_event_lock_irq(conf->wait_barrier,
majianpengb364e3d2013-11-14 15:16:18 +1100805 !conf->array_frozen &&
majianpeng79ef3a82013-11-15 14:55:02 +0800806 conf->barrier < RESYNC_DEPTH &&
NeilBrown2f73d3c2014-09-10 15:01:49 +1000807 conf->current_window_requests == 0 &&
majianpeng79ef3a82013-11-15 14:55:02 +0800808 (conf->start_next_window >=
809 conf->next_resync + RESYNC_SECTORS),
Lukas Czernereed8c022012-11-30 11:42:40 +0100810 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800811
NeilBrown34e97f12014-09-16 12:14:14 +1000812 conf->nr_pending++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 spin_unlock_irq(&conf->resync_lock);
814}
815
NeilBrowne8096362011-10-11 16:49:05 +1100816static void lower_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800817{
818 unsigned long flags;
NeilBrown709ae482009-12-14 12:49:51 +1100819 BUG_ON(conf->barrier <= 0);
NeilBrown17999be2006-01-06 00:20:12 -0800820 spin_lock_irqsave(&conf->resync_lock, flags);
821 conf->barrier--;
NeilBrown34e97f12014-09-16 12:14:14 +1000822 conf->nr_pending--;
NeilBrown17999be2006-01-06 00:20:12 -0800823 spin_unlock_irqrestore(&conf->resync_lock, flags);
824 wake_up(&conf->wait_barrier);
825}
826
majianpeng79ef3a82013-11-15 14:55:02 +0800827static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
NeilBrown17999be2006-01-06 00:20:12 -0800828{
majianpeng79ef3a82013-11-15 14:55:02 +0800829 bool wait = false;
830
831 if (conf->array_frozen || !bio)
832 wait = true;
833 else if (conf->barrier && bio_data_dir(bio) == WRITE) {
NeilBrown23554962014-09-10 15:56:57 +1000834 if ((conf->mddev->curr_resync_completed
835 >= bio_end_sector(bio)) ||
836 (conf->next_resync + NEXT_NORMALIO_DISTANCE
837 <= bio->bi_iter.bi_sector))
majianpeng79ef3a82013-11-15 14:55:02 +0800838 wait = false;
839 else
840 wait = true;
841 }
842
843 return wait;
844}
845
846static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
847{
848 sector_t sector = 0;
849
NeilBrown17999be2006-01-06 00:20:12 -0800850 spin_lock_irq(&conf->resync_lock);
majianpeng79ef3a82013-11-15 14:55:02 +0800851 if (need_to_wait_for_sync(conf, bio)) {
NeilBrown17999be2006-01-06 00:20:12 -0800852 conf->nr_waiting++;
NeilBrownd6b42dc2012-03-19 12:46:38 +1100853 /* Wait for the barrier to drop.
854 * However if there are already pending
855 * requests (preventing the barrier from
856 * rising completely), and the
NeilBrown5965b642014-09-04 15:51:44 +1000857 * per-process bio queue isn't empty,
NeilBrownd6b42dc2012-03-19 12:46:38 +1100858 * then don't wait, as we need to empty
NeilBrown5965b642014-09-04 15:51:44 +1000859 * that queue to allow conf->start_next_window
860 * to increase.
NeilBrownd6b42dc2012-03-19 12:46:38 +1100861 */
862 wait_event_lock_irq(conf->wait_barrier,
majianpengb364e3d2013-11-14 15:16:18 +1100863 !conf->array_frozen &&
864 (!conf->barrier ||
NeilBrown5965b642014-09-04 15:51:44 +1000865 ((conf->start_next_window <
866 conf->next_resync + RESYNC_SECTORS) &&
867 current->bio_list &&
868 !bio_list_empty(current->bio_list))),
Lukas Czernereed8c022012-11-30 11:42:40 +0100869 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800870 conf->nr_waiting--;
871 }
majianpeng79ef3a82013-11-15 14:55:02 +0800872
873 if (bio && bio_data_dir(bio) == WRITE) {
Jes Sorensene8ff8bf2015-09-16 10:20:05 -0400874 if (bio->bi_iter.bi_sector >= conf->next_resync) {
majianpeng79ef3a82013-11-15 14:55:02 +0800875 if (conf->start_next_window == MaxSector)
876 conf->start_next_window =
877 conf->next_resync +
878 NEXT_NORMALIO_DISTANCE;
879
880 if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
Kent Overstreet4f024f32013-10-11 15:44:27 -0700881 <= bio->bi_iter.bi_sector)
majianpeng79ef3a82013-11-15 14:55:02 +0800882 conf->next_window_requests++;
883 else
884 conf->current_window_requests++;
majianpeng79ef3a82013-11-15 14:55:02 +0800885 sector = conf->start_next_window;
NeilBrown41a336e2014-01-14 11:56:14 +1100886 }
majianpeng79ef3a82013-11-15 14:55:02 +0800887 }
888
NeilBrown17999be2006-01-06 00:20:12 -0800889 conf->nr_pending++;
890 spin_unlock_irq(&conf->resync_lock);
majianpeng79ef3a82013-11-15 14:55:02 +0800891 return sector;
NeilBrown17999be2006-01-06 00:20:12 -0800892}
893
majianpeng79ef3a82013-11-15 14:55:02 +0800894static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
895 sector_t bi_sector)
NeilBrown17999be2006-01-06 00:20:12 -0800896{
897 unsigned long flags;
majianpeng79ef3a82013-11-15 14:55:02 +0800898
NeilBrown17999be2006-01-06 00:20:12 -0800899 spin_lock_irqsave(&conf->resync_lock, flags);
900 conf->nr_pending--;
majianpeng79ef3a82013-11-15 14:55:02 +0800901 if (start_next_window) {
902 if (start_next_window == conf->start_next_window) {
903 if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
904 <= bi_sector)
905 conf->next_window_requests--;
906 else
907 conf->current_window_requests--;
908 } else
909 conf->current_window_requests--;
910
911 if (!conf->current_window_requests) {
912 if (conf->next_window_requests) {
913 conf->current_window_requests =
914 conf->next_window_requests;
915 conf->next_window_requests = 0;
916 conf->start_next_window +=
917 NEXT_NORMALIO_DISTANCE;
918 } else
919 conf->start_next_window = MaxSector;
920 }
921 }
NeilBrown17999be2006-01-06 00:20:12 -0800922 spin_unlock_irqrestore(&conf->resync_lock, flags);
923 wake_up(&conf->wait_barrier);
924}
925
NeilBrowne2d59922013-06-12 11:01:22 +1000926static void freeze_array(struct r1conf *conf, int extra)
NeilBrownddaf22a2006-01-06 00:20:19 -0800927{
928 /* stop syncio and normal IO and wait for everything to
929 * go quite.
majianpengb364e3d2013-11-14 15:16:18 +1100930 * We wait until nr_pending match nr_queued+extra
NeilBrown1c830532008-03-04 14:29:35 -0800931 * This is called in the context of one normal IO request
932 * that has failed. Thus any sync request that might be pending
933 * will be blocked by nr_pending, and we need to wait for
934 * pending IO requests to complete or be queued for re-try.
NeilBrowne2d59922013-06-12 11:01:22 +1000935 * Thus the number queued (nr_queued) plus this request (extra)
NeilBrown1c830532008-03-04 14:29:35 -0800936 * must match the number of pending IOs (nr_pending) before
937 * we continue.
NeilBrownddaf22a2006-01-06 00:20:19 -0800938 */
939 spin_lock_irq(&conf->resync_lock);
majianpengb364e3d2013-11-14 15:16:18 +1100940 conf->array_frozen = 1;
Lukas Czernereed8c022012-11-30 11:42:40 +0100941 wait_event_lock_irq_cmd(conf->wait_barrier,
NeilBrowne2d59922013-06-12 11:01:22 +1000942 conf->nr_pending == conf->nr_queued+extra,
Lukas Czernereed8c022012-11-30 11:42:40 +0100943 conf->resync_lock,
944 flush_pending_writes(conf));
NeilBrownddaf22a2006-01-06 00:20:19 -0800945 spin_unlock_irq(&conf->resync_lock);
946}
NeilBrowne8096362011-10-11 16:49:05 +1100947static void unfreeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800948{
949 /* reverse the effect of the freeze */
950 spin_lock_irq(&conf->resync_lock);
majianpengb364e3d2013-11-14 15:16:18 +1100951 conf->array_frozen = 0;
NeilBrownddaf22a2006-01-06 00:20:19 -0800952 wake_up(&conf->wait_barrier);
953 spin_unlock_irq(&conf->resync_lock);
954}
955
NeilBrownf72ffdd2014-09-30 14:23:59 +1000956/* duplicate the data pages for behind I/O
NeilBrown4e780642010-10-19 12:54:01 +1100957 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100958static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
NeilBrown4b6d2872005-09-09 16:23:47 -0700959{
960 int i;
961 struct bio_vec *bvec;
NeilBrown2ca68f52011-07-28 11:32:10 +1000962 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
NeilBrown4b6d2872005-09-09 16:23:47 -0700963 GFP_NOIO);
NeilBrown2ca68f52011-07-28 11:32:10 +1000964 if (unlikely(!bvecs))
NeilBrownaf6d7b72011-05-11 14:51:19 +1000965 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700966
Kent Overstreetcb34e052012-09-05 15:22:02 -0700967 bio_for_each_segment_all(bvec, bio, i) {
NeilBrown2ca68f52011-07-28 11:32:10 +1000968 bvecs[i] = *bvec;
969 bvecs[i].bv_page = alloc_page(GFP_NOIO);
970 if (unlikely(!bvecs[i].bv_page))
NeilBrown4b6d2872005-09-09 16:23:47 -0700971 goto do_sync_io;
NeilBrown2ca68f52011-07-28 11:32:10 +1000972 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
973 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
974 kunmap(bvecs[i].bv_page);
NeilBrown4b6d2872005-09-09 16:23:47 -0700975 kunmap(bvec->bv_page);
976 }
NeilBrown2ca68f52011-07-28 11:32:10 +1000977 r1_bio->behind_bvecs = bvecs;
NeilBrownaf6d7b72011-05-11 14:51:19 +1000978 r1_bio->behind_page_count = bio->bi_vcnt;
979 set_bit(R1BIO_BehindIO, &r1_bio->state);
980 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700981
982do_sync_io:
NeilBrownaf6d7b72011-05-11 14:51:19 +1000983 for (i = 0; i < bio->bi_vcnt; i++)
NeilBrown2ca68f52011-07-28 11:32:10 +1000984 if (bvecs[i].bv_page)
985 put_page(bvecs[i].bv_page);
986 kfree(bvecs);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700987 pr_debug("%dB behind alloc failed, doing sync I/O\n",
988 bio->bi_iter.bi_size);
NeilBrown4b6d2872005-09-09 16:23:47 -0700989}
990
NeilBrownf54a9d02012-08-02 08:33:20 +1000991struct raid1_plug_cb {
992 struct blk_plug_cb cb;
993 struct bio_list pending;
994 int pending_cnt;
995};
996
997static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
998{
999 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1000 cb);
1001 struct mddev *mddev = plug->cb.data;
1002 struct r1conf *conf = mddev->private;
1003 struct bio *bio;
1004
NeilBrown874807a2012-11-27 12:14:40 +11001005 if (from_schedule || current->bio_list) {
NeilBrownf54a9d02012-08-02 08:33:20 +10001006 spin_lock_irq(&conf->device_lock);
1007 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1008 conf->pending_count += plug->pending_cnt;
1009 spin_unlock_irq(&conf->device_lock);
NeilBrownee0b0242013-02-25 12:38:29 +11001010 wake_up(&conf->wait_barrier);
NeilBrownf54a9d02012-08-02 08:33:20 +10001011 md_wakeup_thread(mddev->thread);
1012 kfree(plug);
1013 return;
1014 }
1015
1016 /* we aren't scheduling, so we can do the write-out directly. */
1017 bio = bio_list_get(&plug->pending);
1018 bitmap_unplug(mddev->bitmap);
1019 wake_up(&conf->wait_barrier);
1020
1021 while (bio) { /* submit pending writes */
1022 struct bio *next = bio->bi_next;
1023 bio->bi_next = NULL;
Mike Christie796a5cf2016-06-05 14:32:07 -05001024 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
Shaohua Li32f9f572013-04-28 18:26:38 +08001025 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1026 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001027 bio_endio(bio);
Shaohua Li32f9f572013-04-28 18:26:38 +08001028 else
1029 generic_make_request(bio);
NeilBrownf54a9d02012-08-02 08:33:20 +10001030 bio = next;
1031 }
1032 kfree(plug);
1033}
1034
Shaohua Li849674e2016-01-20 13:52:20 -08001035static void raid1_make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036{
NeilBrowne8096362011-10-11 16:49:05 +11001037 struct r1conf *conf = mddev->private;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001038 struct raid1_info *mirror;
NeilBrown9f2c9d12011-10-11 16:48:43 +11001039 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 struct bio *read_bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001041 int i, disks;
NeilBrown84255d12008-05-23 13:04:32 -07001042 struct bitmap *bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -07001043 unsigned long flags;
Mike Christie796a5cf2016-06-05 14:32:07 -05001044 const int op = bio_op(bio);
Jens Axboea3623572005-11-01 09:26:16 +01001045 const int rw = bio_data_dir(bio);
Jens Axboe1eff9d32016-08-05 15:35:16 -06001046 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1047 const unsigned long do_flush_fua = (bio->bi_opf &
Mike Christie28a8f0d2016-06-05 14:32:25 -05001048 (REQ_PREFLUSH | REQ_FUA));
NeilBrown3cb03002011-10-11 16:45:26 +11001049 struct md_rdev *blocked_rdev;
NeilBrownf54a9d02012-08-02 08:33:20 +10001050 struct blk_plug_cb *cb;
1051 struct raid1_plug_cb *plug = NULL;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001052 int first_clone;
1053 int sectors_handled;
1054 int max_sectors;
majianpeng79ef3a82013-11-15 14:55:02 +08001055 sector_t start_next_window;
NeilBrown191ea9b2005-06-21 17:17:23 -07001056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 /*
1058 * Register the new request and wait if the reconstruction
1059 * thread has put up a bar for new requests.
1060 * Continue immediately if no resync is active currently.
1061 */
NeilBrown62de6082006-05-01 12:15:47 -07001062
NeilBrown3d310eb2005-06-21 17:17:26 -07001063 md_write_start(mddev, bio); /* wait on superblock update early */
1064
NeilBrown6eef4b22009-12-14 12:49:51 +11001065 if (bio_data_dir(bio) == WRITE &&
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001066 ((bio_end_sector(bio) > mddev->suspend_lo &&
1067 bio->bi_iter.bi_sector < mddev->suspend_hi) ||
1068 (mddev_is_clustered(mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -05001069 md_cluster_ops->area_resyncing(mddev, WRITE,
1070 bio->bi_iter.bi_sector, bio_end_sector(bio))))) {
NeilBrown6eef4b22009-12-14 12:49:51 +11001071 /* As the suspend_* range is controlled by
1072 * userspace, we want an interruptible
1073 * wait.
1074 */
1075 DEFINE_WAIT(w);
1076 for (;;) {
1077 flush_signals(current);
1078 prepare_to_wait(&conf->wait_barrier,
1079 &w, TASK_INTERRUPTIBLE);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001080 if (bio_end_sector(bio) <= mddev->suspend_lo ||
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001081 bio->bi_iter.bi_sector >= mddev->suspend_hi ||
1082 (mddev_is_clustered(mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -05001083 !md_cluster_ops->area_resyncing(mddev, WRITE,
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001084 bio->bi_iter.bi_sector, bio_end_sector(bio))))
NeilBrown6eef4b22009-12-14 12:49:51 +11001085 break;
1086 schedule();
1087 }
1088 finish_wait(&conf->wait_barrier, &w);
1089 }
NeilBrown62de6082006-05-01 12:15:47 -07001090
majianpeng79ef3a82013-11-15 14:55:02 +08001091 start_next_window = wait_barrier(conf, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
NeilBrown84255d12008-05-23 13:04:32 -07001093 bitmap = mddev->bitmap;
1094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 /*
Christoph Hellwig70246282016-07-19 11:28:41 +02001096 * make_request() can abort the operation when read-ahead is being
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 * used and no empty request is available.
1098 *
1099 */
1100 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1101
1102 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001103 r1_bio->sectors = bio_sectors(bio);
NeilBrown191ea9b2005-06-21 17:17:23 -07001104 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001106 r1_bio->sector = bio->bi_iter.bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
NeilBrownd2eb35a2011-07-28 11:31:48 +10001108 /* We might need to issue multiple reads to different
1109 * devices if there are bad blocks around, so we keep
1110 * track of the number of reads in bio->bi_phys_segments.
1111 * If this is 0, there is only one r1_bio and no locking
1112 * will be needed when requests complete. If it is
1113 * non-zero, then it is the number of not-completed requests.
1114 */
1115 bio->bi_phys_segments = 0;
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001116 bio_clear_flag(bio, BIO_SEG_VALID);
NeilBrownd2eb35a2011-07-28 11:31:48 +10001117
Jens Axboea3623572005-11-01 09:26:16 +01001118 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 /*
1120 * read balancing logic:
1121 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001122 int rdisk;
1123
1124read_again:
1125 rdisk = read_balance(conf, r1_bio, &max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 if (rdisk < 0) {
1128 /* couldn't find anywhere to read from */
1129 raid_end_bio_io(r1_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001130 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 }
1132 mirror = conf->mirrors + rdisk;
1133
NeilBrowne5551902010-03-31 11:21:44 +11001134 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1135 bitmap) {
1136 /* Reading from a write-mostly device must
1137 * take care not to over-take any writes
1138 * that are 'behind'
1139 */
1140 wait_event(bitmap->behind_wait,
1141 atomic_read(&bitmap->behind_writes) == 0);
1142 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 r1_bio->read_disk = rdisk;
NeilBrownf0cc9a02014-09-22 10:06:23 +10001144 r1_bio->start_next_window = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
NeilBrowna167f662010-10-26 18:31:13 +11001146 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001147 bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
Kent Overstreet6678d832013-08-07 11:14:32 -07001148 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
1150 r1_bio->bios[rdisk] = read_bio;
1151
Kent Overstreet4f024f32013-10-11 15:44:27 -07001152 read_bio->bi_iter.bi_sector = r1_bio->sector +
1153 mirror->rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 read_bio->bi_bdev = mirror->rdev->bdev;
1155 read_bio->bi_end_io = raid1_end_read_request;
Mike Christie796a5cf2016-06-05 14:32:07 -05001156 bio_set_op_attrs(read_bio, op, do_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 read_bio->bi_private = r1_bio;
1158
NeilBrownd2eb35a2011-07-28 11:31:48 +10001159 if (max_sectors < r1_bio->sectors) {
1160 /* could not read all from this device, so we will
1161 * need another r1_bio.
1162 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001163
1164 sectors_handled = (r1_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07001165 - bio->bi_iter.bi_sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +10001166 r1_bio->sectors = max_sectors;
1167 spin_lock_irq(&conf->device_lock);
1168 if (bio->bi_phys_segments == 0)
1169 bio->bi_phys_segments = 2;
1170 else
1171 bio->bi_phys_segments++;
1172 spin_unlock_irq(&conf->device_lock);
1173 /* Cannot call generic_make_request directly
1174 * as that will be queued in __make_request
1175 * and subsequent mempool_alloc might block waiting
1176 * for it. So hand bio over to raid1d.
1177 */
1178 reschedule_retry(r1_bio);
1179
1180 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1181
1182 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001183 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrownd2eb35a2011-07-28 11:31:48 +10001184 r1_bio->state = 0;
1185 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001186 r1_bio->sector = bio->bi_iter.bi_sector +
1187 sectors_handled;
NeilBrownd2eb35a2011-07-28 11:31:48 +10001188 goto read_again;
1189 } else
1190 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001191 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 }
1193
1194 /*
1195 * WRITE:
1196 */
NeilBrown34db0cd2011-10-11 16:50:01 +11001197 if (conf->pending_count >= max_queued_requests) {
1198 md_wakeup_thread(mddev->thread);
1199 wait_event(conf->wait_barrier,
1200 conf->pending_count < max_queued_requests);
1201 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001202 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 * inc refcount on their rdev. Record them by setting
1204 * bios[x] to bio
NeilBrown1f68f0c2011-07-28 11:31:48 +10001205 * If there are known/acknowledged bad blocks on any device on
1206 * which we have seen a write error, we want to avoid writing those
1207 * blocks.
1208 * This potentially requires several writes to write around
1209 * the bad blocks. Each set of writes gets it's own r1bio
1210 * with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001212
NeilBrown8f19ccb2011-12-23 10:17:56 +11001213 disks = conf->raid_disks * 2;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001214 retry_write:
majianpeng79ef3a82013-11-15 14:55:02 +08001215 r1_bio->start_next_window = start_next_window;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001216 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 rcu_read_lock();
NeilBrown1f68f0c2011-07-28 11:31:48 +10001218 max_sectors = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001220 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001221 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1222 atomic_inc(&rdev->nr_pending);
1223 blocked_rdev = rdev;
1224 break;
1225 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001226 r1_bio->bios[i] = NULL;
Kent Overstreet8ae12662015-04-27 23:48:34 -07001227 if (!rdev || test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11001228 if (i < conf->raid_disks)
1229 set_bit(R1BIO_Degraded, &r1_bio->state);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001230 continue;
1231 }
1232
1233 atomic_inc(&rdev->nr_pending);
1234 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1235 sector_t first_bad;
1236 int bad_sectors;
1237 int is_bad;
1238
1239 is_bad = is_badblock(rdev, r1_bio->sector,
1240 max_sectors,
1241 &first_bad, &bad_sectors);
1242 if (is_bad < 0) {
1243 /* mustn't write here until the bad block is
1244 * acknowledged*/
1245 set_bit(BlockedBadBlocks, &rdev->flags);
1246 blocked_rdev = rdev;
1247 break;
NeilBrown964147d2010-05-18 15:27:13 +10001248 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001249 if (is_bad && first_bad <= r1_bio->sector) {
1250 /* Cannot write here at all */
1251 bad_sectors -= (r1_bio->sector - first_bad);
1252 if (bad_sectors < max_sectors)
1253 /* mustn't write more than bad_sectors
1254 * to other devices yet
1255 */
1256 max_sectors = bad_sectors;
1257 rdev_dec_pending(rdev, mddev);
1258 /* We don't set R1BIO_Degraded as that
1259 * only applies if the disk is
1260 * missing, so it might be re-added,
1261 * and we want to know to recover this
1262 * chunk.
1263 * In this case the device is here,
1264 * and the fact that this chunk is not
1265 * in-sync is recorded in the bad
1266 * block log
1267 */
1268 continue;
1269 }
1270 if (is_bad) {
1271 int good_sectors = first_bad - r1_bio->sector;
1272 if (good_sectors < max_sectors)
1273 max_sectors = good_sectors;
1274 }
1275 }
1276 r1_bio->bios[i] = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 }
1278 rcu_read_unlock();
1279
Dan Williams6bfe0b42008-04-30 00:52:32 -07001280 if (unlikely(blocked_rdev)) {
1281 /* Wait for this device to become unblocked */
1282 int j;
majianpeng79ef3a82013-11-15 14:55:02 +08001283 sector_t old = start_next_window;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001284
1285 for (j = 0; j < i; j++)
1286 if (r1_bio->bios[j])
1287 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001288 r1_bio->state = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001289 allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001290 md_wait_for_blocked_rdev(blocked_rdev, mddev);
majianpeng79ef3a82013-11-15 14:55:02 +08001291 start_next_window = wait_barrier(conf, bio);
1292 /*
1293 * We must make sure the multi r1bios of bio have
1294 * the same value of bi_phys_segments
1295 */
1296 if (bio->bi_phys_segments && old &&
1297 old != start_next_window)
1298 /* Wait for the former r1bio(s) to complete */
1299 wait_event(conf->wait_barrier,
1300 bio->bi_phys_segments == 1);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001301 goto retry_write;
1302 }
1303
NeilBrown1f68f0c2011-07-28 11:31:48 +10001304 if (max_sectors < r1_bio->sectors) {
1305 /* We are splitting this write into multiple parts, so
1306 * we need to prepare for allocating another r1_bio.
1307 */
1308 r1_bio->sectors = max_sectors;
1309 spin_lock_irq(&conf->device_lock);
1310 if (bio->bi_phys_segments == 0)
1311 bio->bi_phys_segments = 2;
1312 else
1313 bio->bi_phys_segments++;
1314 spin_unlock_irq(&conf->device_lock);
NeilBrown191ea9b2005-06-21 17:17:23 -07001315 }
Kent Overstreet4f024f32013-10-11 15:44:27 -07001316 sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
NeilBrown4b6d2872005-09-09 16:23:47 -07001317
NeilBrown4e780642010-10-19 12:54:01 +11001318 atomic_set(&r1_bio->remaining, 1);
NeilBrown4b6d2872005-09-09 16:23:47 -07001319 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -07001320
NeilBrown1f68f0c2011-07-28 11:31:48 +10001321 first_clone = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 for (i = 0; i < disks; i++) {
1323 struct bio *mbio;
1324 if (!r1_bio->bios[i])
1325 continue;
1326
NeilBrowna167f662010-10-26 18:31:13 +11001327 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001328 bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
NeilBrown1f68f0c2011-07-28 11:31:48 +10001330 if (first_clone) {
1331 /* do behind I/O ?
1332 * Not if there are too many, or cannot
1333 * allocate memory, or a reader on WriteMostly
1334 * is waiting for behind writes to flush */
1335 if (bitmap &&
1336 (atomic_read(&bitmap->behind_writes)
1337 < mddev->bitmap_info.max_write_behind) &&
1338 !waitqueue_active(&bitmap->behind_wait))
1339 alloc_behind_pages(mbio, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
NeilBrown1f68f0c2011-07-28 11:31:48 +10001341 bitmap_startwrite(bitmap, r1_bio->sector,
1342 r1_bio->sectors,
1343 test_bit(R1BIO_BehindIO,
1344 &r1_bio->state));
1345 first_clone = 0;
1346 }
NeilBrown2ca68f52011-07-28 11:32:10 +10001347 if (r1_bio->behind_bvecs) {
NeilBrown4b6d2872005-09-09 16:23:47 -07001348 struct bio_vec *bvec;
1349 int j;
1350
Kent Overstreetcb34e052012-09-05 15:22:02 -07001351 /*
1352 * We trimmed the bio, so _all is legit
NeilBrown4b6d2872005-09-09 16:23:47 -07001353 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001354 bio_for_each_segment_all(bvec, mbio, j)
NeilBrown2ca68f52011-07-28 11:32:10 +10001355 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
NeilBrown4b6d2872005-09-09 16:23:47 -07001356 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1357 atomic_inc(&r1_bio->behind_remaining);
1358 }
1359
NeilBrown1f68f0c2011-07-28 11:31:48 +10001360 r1_bio->bios[i] = mbio;
1361
Kent Overstreet4f024f32013-10-11 15:44:27 -07001362 mbio->bi_iter.bi_sector = (r1_bio->sector +
NeilBrown1f68f0c2011-07-28 11:31:48 +10001363 conf->mirrors[i].rdev->data_offset);
1364 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1365 mbio->bi_end_io = raid1_end_write_request;
Christoph Hellwig288dab82016-06-09 16:00:36 +02001366 bio_set_op_attrs(mbio, op, do_flush_fua | do_sync);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001367 mbio->bi_private = r1_bio;
1368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 atomic_inc(&r1_bio->remaining);
NeilBrownf54a9d02012-08-02 08:33:20 +10001370
1371 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1372 if (cb)
1373 plug = container_of(cb, struct raid1_plug_cb, cb);
1374 else
1375 plug = NULL;
NeilBrown4e780642010-10-19 12:54:01 +11001376 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001377 if (plug) {
1378 bio_list_add(&plug->pending, mbio);
1379 plug->pending_cnt++;
1380 } else {
1381 bio_list_add(&conf->pending_bio_list, mbio);
1382 conf->pending_count++;
1383 }
NeilBrown4e780642010-10-19 12:54:01 +11001384 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001385 if (!plug)
NeilBrownb357f042012-07-03 17:45:31 +10001386 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 }
NeilBrown079fa162011-09-10 17:21:23 +10001388 /* Mustn't call r1_bio_write_done before this next test,
1389 * as it could result in the bio being freed.
1390 */
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001391 if (sectors_handled < bio_sectors(bio)) {
NeilBrown079fa162011-09-10 17:21:23 +10001392 r1_bio_write_done(r1_bio);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001393 /* We need another r1_bio. It has already been counted
1394 * in bio->bi_phys_segments
1395 */
1396 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1397 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001398 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001399 r1_bio->state = 0;
1400 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001401 r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001402 goto retry_write;
1403 }
1404
NeilBrown079fa162011-09-10 17:21:23 +10001405 r1_bio_write_done(r1_bio);
1406
1407 /* In case raid1d snuck in to freeze_array */
1408 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409}
1410
Shaohua Li849674e2016-01-20 13:52:20 -08001411static void raid1_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412{
NeilBrowne8096362011-10-11 16:49:05 +11001413 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 int i;
1415
1416 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001417 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001418 rcu_read_lock();
1419 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001420 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001422 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1423 }
1424 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 seq_printf(seq, "]");
1426}
1427
Shaohua Li849674e2016-01-20 13:52:20 -08001428static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429{
1430 char b[BDEVNAME_SIZE];
NeilBrowne8096362011-10-11 16:49:05 +11001431 struct r1conf *conf = mddev->private;
NeilBrown423f04d2015-07-27 11:48:52 +10001432 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
1434 /*
1435 * If it is not operational, then we have already marked it as dead
1436 * else if it is the last working disks, ignore the error, let the
1437 * next level up know.
1438 * else mark the drive as failed
1439 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001440 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001441 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 /*
1443 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001444 * normal single drive.
1445 * However don't try a recovery from this drive as
1446 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 */
NeilBrown53890422011-07-27 11:00:36 +10001448 conf->recovery_disabled = mddev->recovery_disabled;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001450 }
NeilBrownde393cd2011-07-28 11:31:48 +10001451 set_bit(Blocked, &rdev->flags);
NeilBrown423f04d2015-07-27 11:48:52 +10001452 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001453 if (test_and_clear_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -07001455 set_bit(Faulty, &rdev->flags);
NeilBrowndd00a992007-05-10 03:15:50 -07001456 } else
1457 set_bit(Faulty, &rdev->flags);
NeilBrown423f04d2015-07-27 11:48:52 +10001458 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrown2446dba2014-07-31 10:16:29 +10001459 /*
1460 * if recovery is running, make sure it aborts.
1461 */
1462 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Guoqing Jiang85ad1d12016-05-03 22:22:13 -04001463 set_mask_bits(&mddev->flags, 0,
1464 BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING));
Joe Perches067032b2011-01-14 09:14:33 +11001465 printk(KERN_ALERT
1466 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1467 "md/raid1:%s: Operation continuing on %d devices.\n",
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001468 mdname(mddev), bdevname(rdev->bdev, b),
1469 mdname(mddev), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
NeilBrowne8096362011-10-11 16:49:05 +11001472static void print_conf(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
1474 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001476 printk(KERN_DEBUG "RAID1 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (!conf) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001478 printk(KERN_DEBUG "(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 return;
1480 }
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001481 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 conf->raid_disks);
1483
NeilBrownddac7c72006-08-31 21:27:36 -07001484 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 for (i = 0; i < conf->raid_disks; i++) {
1486 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001487 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownddac7c72006-08-31 21:27:36 -07001488 if (rdev)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001489 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -07001490 i, !test_bit(In_sync, &rdev->flags),
1491 !test_bit(Faulty, &rdev->flags),
1492 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 }
NeilBrownddac7c72006-08-31 21:27:36 -07001494 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495}
1496
NeilBrowne8096362011-10-11 16:49:05 +11001497static void close_sync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498{
majianpeng79ef3a82013-11-15 14:55:02 +08001499 wait_barrier(conf, NULL);
1500 allow_barrier(conf, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
1502 mempool_destroy(conf->r1buf_pool);
1503 conf->r1buf_pool = NULL;
majianpeng79ef3a82013-11-15 14:55:02 +08001504
NeilBrown669cc7b2014-09-04 16:30:38 +10001505 spin_lock_irq(&conf->resync_lock);
Jes Sorensene8ff8bf2015-09-16 10:20:05 -04001506 conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE;
majianpeng79ef3a82013-11-15 14:55:02 +08001507 conf->start_next_window = MaxSector;
NeilBrown669cc7b2014-09-04 16:30:38 +10001508 conf->current_window_requests +=
1509 conf->next_window_requests;
1510 conf->next_window_requests = 0;
1511 spin_unlock_irq(&conf->resync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512}
1513
NeilBrownfd01b882011-10-11 16:47:53 +11001514static int raid1_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515{
1516 int i;
NeilBrowne8096362011-10-11 16:49:05 +11001517 struct r1conf *conf = mddev->private;
NeilBrown6b965622010-08-18 11:56:59 +10001518 int count = 0;
1519 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
1521 /*
NeilBrownf72ffdd2014-09-30 14:23:59 +10001522 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001523 * and mark them readable.
1524 * Called under mddev lock, so rcu protection not needed.
NeilBrown423f04d2015-07-27 11:48:52 +10001525 * device_lock used to avoid races with raid1_end_read_request
1526 * which expects 'In_sync' flags and ->degraded to be consistent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 */
NeilBrown423f04d2015-07-27 11:48:52 +10001528 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001530 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001531 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1532 if (repl
Goldwyn Rodrigues1aee41f2014-10-29 18:51:31 -05001533 && !test_bit(Candidate, &repl->flags)
NeilBrown8c7a2c22011-12-23 10:17:57 +11001534 && repl->recovery_offset == MaxSector
1535 && !test_bit(Faulty, &repl->flags)
1536 && !test_and_set_bit(In_sync, &repl->flags)) {
1537 /* replacement has just become active */
1538 if (!rdev ||
1539 !test_and_clear_bit(In_sync, &rdev->flags))
1540 count++;
1541 if (rdev) {
1542 /* Replaced device not technically
1543 * faulty, but we need to be sure
1544 * it gets removed and never re-added
1545 */
1546 set_bit(Faulty, &rdev->flags);
1547 sysfs_notify_dirent_safe(
1548 rdev->sysfs_state);
1549 }
1550 }
NeilBrownddac7c72006-08-31 21:27:36 -07001551 if (rdev
Lukasz Dorau61e49472013-10-24 12:55:17 +11001552 && rdev->recovery_offset == MaxSector
NeilBrownddac7c72006-08-31 21:27:36 -07001553 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001554 && !test_and_set_bit(In_sync, &rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001555 count++;
Jonathan Brassow654e8b52011-07-27 11:00:36 +10001556 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 }
1558 }
NeilBrown6b965622010-08-18 11:56:59 +10001559 mddev->degraded -= count;
1560 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001563 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564}
1565
NeilBrownfd01b882011-10-11 16:47:53 +11001566static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567{
NeilBrowne8096362011-10-11 16:49:05 +11001568 struct r1conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001569 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001570 int mirror = 0;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001571 struct raid1_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001572 int first = 0;
NeilBrown30194632011-12-23 10:17:56 +11001573 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
NeilBrown53890422011-07-27 11:00:36 +10001575 if (mddev->recovery_disabled == conf->recovery_disabled)
1576 return -EBUSY;
1577
Dan Williams1501efa2016-01-13 16:00:07 -08001578 if (md_integrity_add_rdev(rdev, mddev))
1579 return -ENXIO;
1580
Neil Brown6c2fce22008-06-28 08:31:31 +10001581 if (rdev->raid_disk >= 0)
1582 first = last = rdev->raid_disk;
1583
Goldwyn Rodrigues70bcecd2015-08-21 10:33:39 -05001584 /*
1585 * find the disk ... but prefer rdev->saved_raid_disk
1586 * if possible.
1587 */
1588 if (rdev->saved_raid_disk >= 0 &&
1589 rdev->saved_raid_disk >= first &&
1590 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1591 first = last = rdev->saved_raid_disk;
1592
NeilBrown7ef449d2011-12-23 10:17:57 +11001593 for (mirror = first; mirror <= last; mirror++) {
1594 p = conf->mirrors+mirror;
1595 if (!p->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
Jonathan Brassow9092c022013-05-02 14:19:24 -05001597 if (mddev->gendisk)
1598 disk_stack_limits(mddev->gendisk, rdev->bdev,
1599 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
1601 p->head_position = 0;
1602 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001603 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001604 /* As all devices are equivalent, we don't need a full recovery
1605 * if this was recently any drive of the array
1606 */
1607 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001608 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001609 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 break;
1611 }
NeilBrown7ef449d2011-12-23 10:17:57 +11001612 if (test_bit(WantReplacement, &p->rdev->flags) &&
1613 p[conf->raid_disks].rdev == NULL) {
1614 /* Add this device as a replacement */
1615 clear_bit(In_sync, &rdev->flags);
1616 set_bit(Replacement, &rdev->flags);
1617 rdev->raid_disk = mirror;
1618 err = 0;
1619 conf->fullsync = 1;
1620 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1621 break;
1622 }
1623 }
Jonathan Brassow9092c022013-05-02 14:19:24 -05001624 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
Shaohua Li2ff8cc22012-10-11 13:28:54 +11001625 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001627 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628}
1629
NeilBrownb8321b62011-12-23 10:17:51 +11001630static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631{
NeilBrowne8096362011-10-11 16:49:05 +11001632 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001634 int number = rdev->raid_disk;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001635 struct raid1_info *p = conf->mirrors + number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
NeilBrownb014f142011-12-23 10:17:56 +11001637 if (rdev != p->rdev)
1638 p = conf->mirrors + conf->raid_disks + number;
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 print_conf(conf);
NeilBrownb8321b62011-12-23 10:17:51 +11001641 if (rdev == p->rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001642 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 atomic_read(&rdev->nr_pending)) {
1644 err = -EBUSY;
1645 goto abort;
1646 }
NeilBrown046abee2010-10-26 15:46:20 +11001647 /* Only remove non-faulty devices if recovery
NeilBrowndfc70642008-05-23 13:04:39 -07001648 * is not possible.
1649 */
1650 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown53890422011-07-27 11:00:36 +10001651 mddev->recovery_disabled != conf->recovery_disabled &&
NeilBrowndfc70642008-05-23 13:04:39 -07001652 mddev->degraded < conf->raid_disks) {
1653 err = -EBUSY;
1654 goto abort;
1655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 p->rdev = NULL;
NeilBrownd787be42016-06-02 16:19:53 +10001657 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
1658 synchronize_rcu();
1659 if (atomic_read(&rdev->nr_pending)) {
1660 /* lost the race, try later */
1661 err = -EBUSY;
1662 p->rdev = rdev;
1663 goto abort;
1664 }
1665 }
1666 if (conf->mirrors[conf->raid_disks + number].rdev) {
NeilBrown8c7a2c22011-12-23 10:17:57 +11001667 /* We just removed a device that is being replaced.
1668 * Move down the replacement. We drain all IO before
1669 * doing this to avoid confusion.
1670 */
1671 struct md_rdev *repl =
1672 conf->mirrors[conf->raid_disks + number].rdev;
NeilBrowne2d59922013-06-12 11:01:22 +10001673 freeze_array(conf, 0);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001674 clear_bit(Replacement, &repl->flags);
1675 p->rdev = repl;
1676 conf->mirrors[conf->raid_disks + number].rdev = NULL;
NeilBrowne2d59922013-06-12 11:01:22 +10001677 unfreeze_array(conf);
NeilBrownb014f142011-12-23 10:17:56 +11001678 clear_bit(WantReplacement, &rdev->flags);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001679 } else
1680 clear_bit(WantReplacement, &rdev->flags);
Martin K. Petersena91a2782011-03-17 11:11:05 +01001681 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683abort:
1684
1685 print_conf(conf);
1686 return err;
1687}
1688
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001689static void end_sync_read(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001691 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
NeilBrown0fc280f2011-10-07 14:22:55 +11001693 update_head_pos(r1_bio->read_disk, r1_bio);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001694
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 /*
1696 * we have read a block, now it needs to be re-written,
1697 * or re-read if the read failed.
1698 * We don't do much here, just schedule handling by raid1d
1699 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001700 if (!bio->bi_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001702
1703 if (atomic_dec_and_test(&r1_bio->remaining))
1704 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705}
1706
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001707static void end_sync_write(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001709 int uptodate = !bio->bi_error;
NeilBrown9f2c9d12011-10-11 16:48:43 +11001710 struct r1bio *r1_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11001711 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001712 struct r1conf *conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10001713 sector_t first_bad;
1714 int bad_sectors;
NeilBrown854abd72016-06-02 16:19:52 +10001715 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001716
NeilBrown6b1117d2006-03-31 02:31:57 -08001717 if (!uptodate) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001718 sector_t sync_blocks = 0;
NeilBrown6b1117d2006-03-31 02:31:57 -08001719 sector_t s = r1_bio->sector;
1720 long sectors_to_go = r1_bio->sectors;
1721 /* make sure these bits doesn't get cleared. */
1722 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001723 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001724 &sync_blocks, 1);
1725 s += sync_blocks;
1726 sectors_to_go -= sync_blocks;
1727 } while (sectors_to_go > 0);
NeilBrown854abd72016-06-02 16:19:52 +10001728 set_bit(WriteErrorSeen, &rdev->flags);
1729 if (!test_and_set_bit(WantReplacement, &rdev->flags))
NeilBrown19d67162011-12-23 10:17:57 +11001730 set_bit(MD_RECOVERY_NEEDED, &
1731 mddev->recovery);
NeilBrownd8f05d22011-07-28 11:33:00 +10001732 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown854abd72016-06-02 16:19:52 +10001733 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
NeilBrown3a9f28a2011-07-28 11:33:42 +10001734 &first_bad, &bad_sectors) &&
1735 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1736 r1_bio->sector,
1737 r1_bio->sectors,
1738 &first_bad, &bad_sectors)
1739 )
NeilBrown4367af52011-07-28 11:31:49 +10001740 set_bit(R1BIO_MadeGood, &r1_bio->state);
NeilBrowne3b97032005-08-04 12:53:34 -07001741
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown4367af52011-07-28 11:31:49 +10001743 int s = r1_bio->sectors;
NeilBrownd8f05d22011-07-28 11:33:00 +10001744 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1745 test_bit(R1BIO_WriteError, &r1_bio->state))
NeilBrown4367af52011-07-28 11:31:49 +10001746 reschedule_retry(r1_bio);
1747 else {
1748 put_buf(r1_bio);
1749 md_done_sync(mddev, s, uptodate);
1750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752}
1753
NeilBrown3cb03002011-10-11 16:45:26 +11001754static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrownd8f05d22011-07-28 11:33:00 +10001755 int sectors, struct page *page, int rw)
1756{
Mike Christie796a5cf2016-06-05 14:32:07 -05001757 if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
NeilBrownd8f05d22011-07-28 11:33:00 +10001758 /* success */
1759 return 1;
NeilBrown19d67162011-12-23 10:17:57 +11001760 if (rw == WRITE) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001761 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001762 if (!test_and_set_bit(WantReplacement,
1763 &rdev->flags))
1764 set_bit(MD_RECOVERY_NEEDED, &
1765 rdev->mddev->recovery);
1766 }
NeilBrownd8f05d22011-07-28 11:33:00 +10001767 /* need to record an error - either for the block or the device */
1768 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1769 md_error(rdev->mddev, rdev);
1770 return 0;
1771}
1772
NeilBrown9f2c9d12011-10-11 16:48:43 +11001773static int fix_sync_read_error(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001774{
1775 /* Try some synchronous reads of other devices to get
1776 * good data, much like with normal read errors. Only
1777 * read into the pages we already have so we don't
1778 * need to re-issue the read request.
1779 * We don't need to freeze the array, because being in an
1780 * active sync request, there is no normal IO, and
1781 * no overlapping syncs.
NeilBrown06f60382011-07-28 11:31:48 +10001782 * We don't need to check is_badblock() again as we
1783 * made sure that anything with a bad block in range
1784 * will have bi_end_io clear.
NeilBrowna68e5872011-05-11 14:40:44 +10001785 */
NeilBrownfd01b882011-10-11 16:47:53 +11001786 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001787 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001788 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1789 sector_t sect = r1_bio->sector;
1790 int sectors = r1_bio->sectors;
1791 int idx = 0;
1792
1793 while(sectors) {
1794 int s = sectors;
1795 int d = r1_bio->read_disk;
1796 int success = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11001797 struct md_rdev *rdev;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001798 int start;
NeilBrowna68e5872011-05-11 14:40:44 +10001799
1800 if (s > (PAGE_SIZE>>9))
1801 s = PAGE_SIZE >> 9;
1802 do {
1803 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1804 /* No rcu protection needed here devices
1805 * can only be removed when no resync is
1806 * active, and resync is currently active
1807 */
1808 rdev = conf->mirrors[d].rdev;
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001809 if (sync_page_io(rdev, sect, s<<9,
NeilBrowna68e5872011-05-11 14:40:44 +10001810 bio->bi_io_vec[idx].bv_page,
Mike Christie796a5cf2016-06-05 14:32:07 -05001811 REQ_OP_READ, 0, false)) {
NeilBrowna68e5872011-05-11 14:40:44 +10001812 success = 1;
1813 break;
1814 }
1815 }
1816 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001817 if (d == conf->raid_disks * 2)
NeilBrowna68e5872011-05-11 14:40:44 +10001818 d = 0;
1819 } while (!success && d != r1_bio->read_disk);
1820
NeilBrown78d7f5f2011-05-11 14:48:56 +10001821 if (!success) {
NeilBrowna68e5872011-05-11 14:40:44 +10001822 char b[BDEVNAME_SIZE];
NeilBrown3a9f28a2011-07-28 11:33:42 +10001823 int abort = 0;
1824 /* Cannot read from anywhere, this block is lost.
1825 * Record a bad block on each device. If that doesn't
1826 * work just disable and interrupt the recovery.
1827 * Don't fail devices as that won't really help.
1828 */
NeilBrowna68e5872011-05-11 14:40:44 +10001829 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1830 " for block %llu\n",
1831 mdname(mddev),
1832 bdevname(bio->bi_bdev, b),
1833 (unsigned long long)r1_bio->sector);
NeilBrown8f19ccb2011-12-23 10:17:56 +11001834 for (d = 0; d < conf->raid_disks * 2; d++) {
NeilBrown3a9f28a2011-07-28 11:33:42 +10001835 rdev = conf->mirrors[d].rdev;
1836 if (!rdev || test_bit(Faulty, &rdev->flags))
1837 continue;
1838 if (!rdev_set_badblocks(rdev, sect, s, 0))
1839 abort = 1;
1840 }
1841 if (abort) {
NeilBrownd890fa22011-10-26 11:54:39 +11001842 conf->recovery_disabled =
1843 mddev->recovery_disabled;
NeilBrown3a9f28a2011-07-28 11:33:42 +10001844 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1845 md_done_sync(mddev, r1_bio->sectors, 0);
1846 put_buf(r1_bio);
1847 return 0;
1848 }
1849 /* Try next page */
1850 sectors -= s;
1851 sect += s;
1852 idx++;
1853 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001854 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001855
1856 start = d;
1857 /* write it back and re-read */
1858 while (d != r1_bio->read_disk) {
1859 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001860 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001861 d--;
1862 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1863 continue;
1864 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001865 if (r1_sync_page_io(rdev, sect, s,
1866 bio->bi_io_vec[idx].bv_page,
1867 WRITE) == 0) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001868 r1_bio->bios[d]->bi_end_io = NULL;
1869 rdev_dec_pending(rdev, mddev);
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001870 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001871 }
1872 d = start;
1873 while (d != r1_bio->read_disk) {
1874 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001875 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001876 d--;
1877 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1878 continue;
1879 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001880 if (r1_sync_page_io(rdev, sect, s,
1881 bio->bi_io_vec[idx].bv_page,
1882 READ) != 0)
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001883 atomic_add(s, &rdev->corrected_errors);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001884 }
NeilBrowna68e5872011-05-11 14:40:44 +10001885 sectors -= s;
1886 sect += s;
1887 idx ++;
1888 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001889 set_bit(R1BIO_Uptodate, &r1_bio->state);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001890 bio->bi_error = 0;
NeilBrowna68e5872011-05-11 14:40:44 +10001891 return 1;
1892}
1893
NeilBrownc95e6382014-09-09 13:54:11 +10001894static void process_checks(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001895{
1896 /* We have read all readable devices. If we haven't
1897 * got the block, then there is no hope left.
1898 * If we have, then we want to do a comparison
1899 * and skip the write if everything is the same.
1900 * If any blocks failed to read, then we need to
1901 * attempt an over-write
1902 */
NeilBrownfd01b882011-10-11 16:47:53 +11001903 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001904 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001905 int primary;
1906 int i;
majianpengf4380a92012-04-12 16:04:47 +10001907 int vcnt;
NeilBrowna68e5872011-05-11 14:40:44 +10001908
NeilBrown30bc9b52013-07-17 15:19:29 +10001909 /* Fix variable parts of all bios */
1910 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1911 for (i = 0; i < conf->raid_disks * 2; i++) {
1912 int j;
1913 int size;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001914 int error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001915 struct bio *b = r1_bio->bios[i];
1916 if (b->bi_end_io != end_sync_read)
1917 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001918 /* fixup the bio for reuse, but preserve errno */
1919 error = b->bi_error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001920 bio_reset(b);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001921 b->bi_error = error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001922 b->bi_vcnt = vcnt;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001923 b->bi_iter.bi_size = r1_bio->sectors << 9;
1924 b->bi_iter.bi_sector = r1_bio->sector +
NeilBrown30bc9b52013-07-17 15:19:29 +10001925 conf->mirrors[i].rdev->data_offset;
1926 b->bi_bdev = conf->mirrors[i].rdev->bdev;
1927 b->bi_end_io = end_sync_read;
1928 b->bi_private = r1_bio;
1929
Kent Overstreet4f024f32013-10-11 15:44:27 -07001930 size = b->bi_iter.bi_size;
NeilBrown30bc9b52013-07-17 15:19:29 +10001931 for (j = 0; j < vcnt ; j++) {
1932 struct bio_vec *bi;
1933 bi = &b->bi_io_vec[j];
1934 bi->bv_offset = 0;
1935 if (size > PAGE_SIZE)
1936 bi->bv_len = PAGE_SIZE;
1937 else
1938 bi->bv_len = size;
1939 size -= PAGE_SIZE;
1940 }
1941 }
NeilBrown8f19ccb2011-12-23 10:17:56 +11001942 for (primary = 0; primary < conf->raid_disks * 2; primary++)
NeilBrowna68e5872011-05-11 14:40:44 +10001943 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001944 !r1_bio->bios[primary]->bi_error) {
NeilBrowna68e5872011-05-11 14:40:44 +10001945 r1_bio->bios[primary]->bi_end_io = NULL;
1946 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1947 break;
1948 }
1949 r1_bio->read_disk = primary;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001950 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001951 int j;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001952 struct bio *pbio = r1_bio->bios[primary];
1953 struct bio *sbio = r1_bio->bios[i];
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001954 int error = sbio->bi_error;
NeilBrowna68e5872011-05-11 14:40:44 +10001955
Kent Overstreet2aabaa62012-09-11 11:26:12 -07001956 if (sbio->bi_end_io != end_sync_read)
NeilBrown78d7f5f2011-05-11 14:48:56 +10001957 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001958 /* Now we can 'fixup' the error value */
1959 sbio->bi_error = 0;
NeilBrowna68e5872011-05-11 14:40:44 +10001960
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001961 if (!error) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001962 for (j = vcnt; j-- ; ) {
1963 struct page *p, *s;
1964 p = pbio->bi_io_vec[j].bv_page;
1965 s = sbio->bi_io_vec[j].bv_page;
1966 if (memcmp(page_address(p),
1967 page_address(s),
NeilBrown5020ad72012-04-02 01:39:05 +10001968 sbio->bi_io_vec[j].bv_len))
NeilBrown78d7f5f2011-05-11 14:48:56 +10001969 break;
NeilBrowna68e5872011-05-11 14:40:44 +10001970 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001971 } else
1972 j = 0;
1973 if (j >= 0)
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11001974 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001975 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001976 && !error)) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001977 /* No need to write to this device. */
1978 sbio->bi_end_io = NULL;
1979 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1980 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001981 }
Kent Overstreetd3b45c22012-09-10 13:49:33 -07001982
1983 bio_copy_data(sbio, pbio);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001984 }
NeilBrowna68e5872011-05-11 14:40:44 +10001985}
1986
NeilBrown9f2c9d12011-10-11 16:48:43 +11001987static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988{
NeilBrowne8096362011-10-11 16:49:05 +11001989 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 int i;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001991 int disks = conf->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 struct bio *bio, *wbio;
1993
1994 bio = r1_bio->bios[r1_bio->read_disk];
1995
NeilBrowna68e5872011-05-11 14:40:44 +10001996 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
1997 /* ouch - failed to read all of that. */
1998 if (!fix_sync_read_error(r1_bio))
1999 return;
NeilBrown7ca78d52011-05-11 14:50:37 +10002000
2001 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrownc95e6382014-09-09 13:54:11 +10002002 process_checks(r1_bio);
2003
NeilBrownd11c1712006-01-06 00:20:26 -08002004 /*
2005 * schedule writes
2006 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 atomic_set(&r1_bio->remaining, 1);
2008 for (i = 0; i < disks ; i++) {
2009 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08002010 if (wbio->bi_end_io == NULL ||
2011 (wbio->bi_end_io == end_sync_read &&
2012 (i == r1_bio->read_disk ||
2013 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 continue;
2015
Mike Christie796a5cf2016-06-05 14:32:07 -05002016 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
NeilBrown3e198f72006-01-06 00:20:21 -08002017 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 atomic_inc(&r1_bio->remaining);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002019 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
NeilBrown191ea9b2005-06-21 17:17:23 -07002020
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 generic_make_request(wbio);
2022 }
2023
2024 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002025 /* if we're here, all write(s) have completed, so clean up */
NeilBrown58e94ae2012-07-19 15:59:18 +10002026 int s = r1_bio->sectors;
2027 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2028 test_bit(R1BIO_WriteError, &r1_bio->state))
2029 reschedule_retry(r1_bio);
2030 else {
2031 put_buf(r1_bio);
2032 md_done_sync(mddev, s, 1);
2033 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 }
2035}
2036
2037/*
2038 * This is a kernel thread which:
2039 *
2040 * 1. Retries failed read operations on working mirrors.
2041 * 2. Updates the raid superblock when problems encounter.
NeilBrownd2eb35a2011-07-28 11:31:48 +10002042 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 */
2044
NeilBrowne8096362011-10-11 16:49:05 +11002045static void fix_read_error(struct r1conf *conf, int read_disk,
NeilBrown867868f2006-10-03 01:15:51 -07002046 sector_t sect, int sectors)
2047{
NeilBrownfd01b882011-10-11 16:47:53 +11002048 struct mddev *mddev = conf->mddev;
NeilBrown867868f2006-10-03 01:15:51 -07002049 while(sectors) {
2050 int s = sectors;
2051 int d = read_disk;
2052 int success = 0;
2053 int start;
NeilBrown3cb03002011-10-11 16:45:26 +11002054 struct md_rdev *rdev;
NeilBrown867868f2006-10-03 01:15:51 -07002055
2056 if (s > (PAGE_SIZE>>9))
2057 s = PAGE_SIZE >> 9;
2058
2059 do {
NeilBrownd2eb35a2011-07-28 11:31:48 +10002060 sector_t first_bad;
2061 int bad_sectors;
2062
NeilBrown707a6a42016-06-02 16:19:52 +10002063 rcu_read_lock();
2064 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002065 if (rdev &&
majianpengda8840a2012-05-22 13:55:03 +10002066 (test_bit(In_sync, &rdev->flags) ||
2067 (!test_bit(Faulty, &rdev->flags) &&
2068 rdev->recovery_offset >= sect + s)) &&
NeilBrownd2eb35a2011-07-28 11:31:48 +10002069 is_badblock(rdev, sect, s,
NeilBrown707a6a42016-06-02 16:19:52 +10002070 &first_bad, &bad_sectors) == 0) {
2071 atomic_inc(&rdev->nr_pending);
2072 rcu_read_unlock();
2073 if (sync_page_io(rdev, sect, s<<9,
Mike Christie796a5cf2016-06-05 14:32:07 -05002074 conf->tmppage, REQ_OP_READ, 0, false))
NeilBrown707a6a42016-06-02 16:19:52 +10002075 success = 1;
2076 rdev_dec_pending(rdev, mddev);
2077 if (success)
2078 break;
2079 } else
2080 rcu_read_unlock();
2081 d++;
2082 if (d == conf->raid_disks * 2)
2083 d = 0;
NeilBrown867868f2006-10-03 01:15:51 -07002084 } while (!success && d != read_disk);
2085
2086 if (!success) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002087 /* Cannot read from anywhere - mark it bad */
NeilBrown3cb03002011-10-11 16:45:26 +11002088 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10002089 if (!rdev_set_badblocks(rdev, sect, s, 0))
2090 md_error(mddev, rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002091 break;
2092 }
2093 /* write it back and re-read */
2094 start = d;
2095 while (d != read_disk) {
2096 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002097 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002098 d--;
NeilBrown707a6a42016-06-02 16:19:52 +10002099 rcu_read_lock();
2100 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002101 if (rdev &&
NeilBrown707a6a42016-06-02 16:19:52 +10002102 !test_bit(Faulty, &rdev->flags)) {
2103 atomic_inc(&rdev->nr_pending);
2104 rcu_read_unlock();
NeilBrownd8f05d22011-07-28 11:33:00 +10002105 r1_sync_page_io(rdev, sect, s,
2106 conf->tmppage, WRITE);
NeilBrown707a6a42016-06-02 16:19:52 +10002107 rdev_dec_pending(rdev, mddev);
2108 } else
2109 rcu_read_unlock();
NeilBrown867868f2006-10-03 01:15:51 -07002110 }
2111 d = start;
2112 while (d != read_disk) {
2113 char b[BDEVNAME_SIZE];
2114 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002115 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002116 d--;
NeilBrown707a6a42016-06-02 16:19:52 +10002117 rcu_read_lock();
2118 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002119 if (rdev &&
NeilBrownb8cb6b42014-09-18 11:09:04 +10002120 !test_bit(Faulty, &rdev->flags)) {
NeilBrown707a6a42016-06-02 16:19:52 +10002121 atomic_inc(&rdev->nr_pending);
2122 rcu_read_unlock();
NeilBrownd8f05d22011-07-28 11:33:00 +10002123 if (r1_sync_page_io(rdev, sect, s,
2124 conf->tmppage, READ)) {
NeilBrown867868f2006-10-03 01:15:51 -07002125 atomic_add(s, &rdev->corrected_errors);
2126 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002127 "md/raid1:%s: read error corrected "
NeilBrown867868f2006-10-03 01:15:51 -07002128 "(%d sectors at %llu on %s)\n",
2129 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07002130 (unsigned long long)(sect +
NeilBrown707a6a42016-06-02 16:19:52 +10002131 rdev->data_offset),
NeilBrown867868f2006-10-03 01:15:51 -07002132 bdevname(rdev->bdev, b));
2133 }
NeilBrown707a6a42016-06-02 16:19:52 +10002134 rdev_dec_pending(rdev, mddev);
2135 } else
2136 rcu_read_unlock();
NeilBrown867868f2006-10-03 01:15:51 -07002137 }
2138 sectors -= s;
2139 sect += s;
2140 }
2141}
2142
NeilBrown9f2c9d12011-10-11 16:48:43 +11002143static int narrow_write_error(struct r1bio *r1_bio, int i)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002144{
NeilBrownfd01b882011-10-11 16:47:53 +11002145 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11002146 struct r1conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11002147 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002148
2149 /* bio has the data to be written to device 'i' where
2150 * we just recently had a write error.
2151 * We repeatedly clone the bio and trim down to one block,
2152 * then try the write. Where the write fails we record
2153 * a bad block.
2154 * It is conceivable that the bio doesn't exactly align with
2155 * blocks. We must handle this somehow.
2156 *
2157 * We currently own a reference on the rdev.
2158 */
2159
2160 int block_sectors;
2161 sector_t sector;
2162 int sectors;
2163 int sect_to_write = r1_bio->sectors;
2164 int ok = 1;
2165
2166 if (rdev->badblocks.shift < 0)
2167 return 0;
2168
Nate Daileyab713cd2015-02-12 12:02:09 -05002169 block_sectors = roundup(1 << rdev->badblocks.shift,
2170 bdev_logical_block_size(rdev->bdev) >> 9);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002171 sector = r1_bio->sector;
2172 sectors = ((sector + block_sectors)
2173 & ~(sector_t)(block_sectors - 1))
2174 - sector;
2175
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002176 while (sect_to_write) {
2177 struct bio *wbio;
2178 if (sectors > sect_to_write)
2179 sectors = sect_to_write;
2180 /* Write at 'sector' for 'sectors'*/
2181
Kent Overstreetb7838632012-09-10 15:17:11 -07002182 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2183 unsigned vcnt = r1_bio->behind_page_count;
2184 struct bio_vec *vec = r1_bio->behind_bvecs;
2185
2186 while (!vec->bv_page) {
2187 vec++;
2188 vcnt--;
2189 }
2190
2191 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2192 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2193
2194 wbio->bi_vcnt = vcnt;
2195 } else {
2196 wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2197 }
2198
Mike Christie796a5cf2016-06-05 14:32:07 -05002199 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002200 wbio->bi_iter.bi_sector = r1_bio->sector;
2201 wbio->bi_iter.bi_size = r1_bio->sectors << 9;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002202
Kent Overstreet6678d832013-08-07 11:14:32 -07002203 bio_trim(wbio, sector - r1_bio->sector, sectors);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002204 wbio->bi_iter.bi_sector += rdev->data_offset;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002205 wbio->bi_bdev = rdev->bdev;
Mike Christie4e49ea42016-06-05 14:31:41 -05002206
2207 if (submit_bio_wait(wbio) < 0)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002208 /* failure! */
2209 ok = rdev_set_badblocks(rdev, sector,
2210 sectors, 0)
2211 && ok;
2212
2213 bio_put(wbio);
2214 sect_to_write -= sectors;
2215 sector += sectors;
2216 sectors = block_sectors;
2217 }
2218 return ok;
2219}
2220
NeilBrowne8096362011-10-11 16:49:05 +11002221static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002222{
2223 int m;
2224 int s = r1_bio->sectors;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002225 for (m = 0; m < conf->raid_disks * 2 ; m++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002226 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002227 struct bio *bio = r1_bio->bios[m];
2228 if (bio->bi_end_io == NULL)
2229 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002230 if (!bio->bi_error &&
NeilBrown62096bc2011-07-28 11:38:13 +10002231 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
NeilBrownc6563a82012-05-21 09:27:00 +10002232 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002233 }
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002234 if (bio->bi_error &&
NeilBrown62096bc2011-07-28 11:38:13 +10002235 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2236 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2237 md_error(conf->mddev, rdev);
2238 }
2239 }
2240 put_buf(r1_bio);
2241 md_done_sync(conf->mddev, s, 1);
2242}
2243
NeilBrowne8096362011-10-11 16:49:05 +11002244static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002245{
2246 int m;
NeilBrown55ce74d2015-08-14 11:11:10 +10002247 bool fail = false;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002248 for (m = 0; m < conf->raid_disks * 2 ; m++)
NeilBrown62096bc2011-07-28 11:38:13 +10002249 if (r1_bio->bios[m] == IO_MADE_GOOD) {
NeilBrown3cb03002011-10-11 16:45:26 +11002250 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002251 rdev_clear_badblocks(rdev,
2252 r1_bio->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10002253 r1_bio->sectors, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002254 rdev_dec_pending(rdev, conf->mddev);
2255 } else if (r1_bio->bios[m] != NULL) {
2256 /* This drive got a write error. We need to
2257 * narrow down and record precise write
2258 * errors.
2259 */
NeilBrown55ce74d2015-08-14 11:11:10 +10002260 fail = true;
NeilBrown62096bc2011-07-28 11:38:13 +10002261 if (!narrow_write_error(r1_bio, m)) {
2262 md_error(conf->mddev,
2263 conf->mirrors[m].rdev);
2264 /* an I/O failed, we can't clear the bitmap */
2265 set_bit(R1BIO_Degraded, &r1_bio->state);
2266 }
2267 rdev_dec_pending(conf->mirrors[m].rdev,
2268 conf->mddev);
2269 }
NeilBrown55ce74d2015-08-14 11:11:10 +10002270 if (fail) {
2271 spin_lock_irq(&conf->device_lock);
2272 list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
Nate Daileyccfc7bf2016-02-29 10:43:58 -05002273 conf->nr_queued++;
NeilBrown55ce74d2015-08-14 11:11:10 +10002274 spin_unlock_irq(&conf->device_lock);
2275 md_wakeup_thread(conf->mddev->thread);
NeilBrownbd8688a2015-10-24 16:02:16 +11002276 } else {
2277 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2278 close_write(r1_bio);
NeilBrown55ce74d2015-08-14 11:11:10 +10002279 raid_end_bio_io(r1_bio);
NeilBrownbd8688a2015-10-24 16:02:16 +11002280 }
NeilBrown62096bc2011-07-28 11:38:13 +10002281}
2282
NeilBrowne8096362011-10-11 16:49:05 +11002283static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002284{
2285 int disk;
2286 int max_sectors;
NeilBrownfd01b882011-10-11 16:47:53 +11002287 struct mddev *mddev = conf->mddev;
NeilBrown62096bc2011-07-28 11:38:13 +10002288 struct bio *bio;
2289 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11002290 struct md_rdev *rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002291
2292 clear_bit(R1BIO_ReadError, &r1_bio->state);
2293 /* we got a read error. Maybe the drive is bad. Maybe just
2294 * the block and we can fix it.
2295 * We freeze all other IO, and try reading the block from
2296 * other devices. When we find one, we re-write
2297 * and check it that fixes the read error.
2298 * This is all done synchronously while the array is
2299 * frozen
2300 */
2301 if (mddev->ro == 0) {
NeilBrowne2d59922013-06-12 11:01:22 +10002302 freeze_array(conf, 1);
NeilBrown62096bc2011-07-28 11:38:13 +10002303 fix_read_error(conf, r1_bio->read_disk,
2304 r1_bio->sector, r1_bio->sectors);
2305 unfreeze_array(conf);
2306 } else
2307 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
NeilBrown7ad4d4a2012-10-11 13:44:30 +11002308 rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
NeilBrown62096bc2011-07-28 11:38:13 +10002309
2310 bio = r1_bio->bios[r1_bio->read_disk];
2311 bdevname(bio->bi_bdev, b);
2312read_more:
2313 disk = read_balance(conf, r1_bio, &max_sectors);
2314 if (disk == -1) {
2315 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
2316 " read error for block %llu\n",
2317 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2318 raid_end_bio_io(r1_bio);
2319 } else {
2320 const unsigned long do_sync
Jens Axboe1eff9d32016-08-05 15:35:16 -06002321 = r1_bio->master_bio->bi_opf & REQ_SYNC;
NeilBrown62096bc2011-07-28 11:38:13 +10002322 if (bio) {
2323 r1_bio->bios[r1_bio->read_disk] =
2324 mddev->ro ? IO_BLOCKED : NULL;
2325 bio_put(bio);
2326 }
2327 r1_bio->read_disk = disk;
2328 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002329 bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
2330 max_sectors);
NeilBrown62096bc2011-07-28 11:38:13 +10002331 r1_bio->bios[r1_bio->read_disk] = bio;
2332 rdev = conf->mirrors[disk].rdev;
2333 printk_ratelimited(KERN_ERR
2334 "md/raid1:%s: redirecting sector %llu"
2335 " to other mirror: %s\n",
2336 mdname(mddev),
2337 (unsigned long long)r1_bio->sector,
2338 bdevname(rdev->bdev, b));
Kent Overstreet4f024f32013-10-11 15:44:27 -07002339 bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
NeilBrown62096bc2011-07-28 11:38:13 +10002340 bio->bi_bdev = rdev->bdev;
2341 bio->bi_end_io = raid1_end_read_request;
Mike Christie796a5cf2016-06-05 14:32:07 -05002342 bio_set_op_attrs(bio, REQ_OP_READ, do_sync);
NeilBrown62096bc2011-07-28 11:38:13 +10002343 bio->bi_private = r1_bio;
2344 if (max_sectors < r1_bio->sectors) {
2345 /* Drat - have to split this up more */
2346 struct bio *mbio = r1_bio->master_bio;
2347 int sectors_handled = (r1_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07002348 - mbio->bi_iter.bi_sector);
NeilBrown62096bc2011-07-28 11:38:13 +10002349 r1_bio->sectors = max_sectors;
2350 spin_lock_irq(&conf->device_lock);
2351 if (mbio->bi_phys_segments == 0)
2352 mbio->bi_phys_segments = 2;
2353 else
2354 mbio->bi_phys_segments++;
2355 spin_unlock_irq(&conf->device_lock);
2356 generic_make_request(bio);
2357 bio = NULL;
2358
2359 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2360
2361 r1_bio->master_bio = mbio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002362 r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
NeilBrown62096bc2011-07-28 11:38:13 +10002363 r1_bio->state = 0;
2364 set_bit(R1BIO_ReadError, &r1_bio->state);
2365 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002366 r1_bio->sector = mbio->bi_iter.bi_sector +
2367 sectors_handled;
NeilBrown62096bc2011-07-28 11:38:13 +10002368
2369 goto read_more;
2370 } else
2371 generic_make_request(bio);
2372 }
2373}
2374
Shaohua Li4ed87312012-10-11 13:34:00 +11002375static void raid1d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376{
Shaohua Li4ed87312012-10-11 13:34:00 +11002377 struct mddev *mddev = thread->mddev;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002378 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 unsigned long flags;
NeilBrowne8096362011-10-11 16:49:05 +11002380 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002382 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383
2384 md_check_recovery(mddev);
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002385
NeilBrown55ce74d2015-08-14 11:11:10 +10002386 if (!list_empty_careful(&conf->bio_end_io_list) &&
2387 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2388 LIST_HEAD(tmp);
2389 spin_lock_irqsave(&conf->device_lock, flags);
2390 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
Nate Daileyccfc7bf2016-02-29 10:43:58 -05002391 while (!list_empty(&conf->bio_end_io_list)) {
2392 list_move(conf->bio_end_io_list.prev, &tmp);
2393 conf->nr_queued--;
2394 }
NeilBrown55ce74d2015-08-14 11:11:10 +10002395 }
2396 spin_unlock_irqrestore(&conf->device_lock, flags);
2397 while (!list_empty(&tmp)) {
Mikulas Patockaa4527442015-10-01 15:17:43 -04002398 r1_bio = list_first_entry(&tmp, struct r1bio,
2399 retry_list);
NeilBrown55ce74d2015-08-14 11:11:10 +10002400 list_del(&r1_bio->retry_list);
NeilBrownbd8688a2015-10-24 16:02:16 +11002401 if (mddev->degraded)
2402 set_bit(R1BIO_Degraded, &r1_bio->state);
2403 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2404 close_write(r1_bio);
NeilBrown55ce74d2015-08-14 11:11:10 +10002405 raid_end_bio_io(r1_bio);
2406 }
2407 }
2408
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002409 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002411
NeilBrown0021b7b2012-07-31 09:08:14 +02002412 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002413
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002415 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002416 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002418 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002419 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08002421 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 spin_unlock_irqrestore(&conf->device_lock, flags);
2423
2424 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002425 conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10002426 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002427 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002428 test_bit(R1BIO_WriteError, &r1_bio->state))
2429 handle_sync_write_finished(conf, r1_bio);
2430 else
NeilBrown4367af52011-07-28 11:31:49 +10002431 sync_request_write(mddev, r1_bio);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002432 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002433 test_bit(R1BIO_WriteError, &r1_bio->state))
2434 handle_write_finished(conf, r1_bio);
2435 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2436 handle_read_error(conf, r1_bio);
2437 else
NeilBrownd2eb35a2011-07-28 11:31:48 +10002438 /* just a partial read to be scheduled from separate
2439 * context
2440 */
2441 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
NeilBrown62096bc2011-07-28 11:38:13 +10002442
NeilBrown1d9d5242009-10-16 15:55:32 +11002443 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002444 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2445 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002447 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448}
2449
NeilBrowne8096362011-10-11 16:49:05 +11002450static int init_resync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451{
2452 int buffs;
2453
2454 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002455 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2457 conf->poolinfo);
2458 if (!conf->r1buf_pool)
2459 return -ENOMEM;
2460 conf->next_resync = 0;
2461 return 0;
2462}
2463
2464/*
2465 * perform a "sync" on one "block"
2466 *
2467 * We need to make sure that no normal I/O request - particularly write
2468 * requests - conflict with active sync requests.
2469 *
2470 * This is achieved by tracking pending requests and a 'barrier' concept
2471 * that can be installed to exclude normal IO requests.
2472 */
2473
Shaohua Li849674e2016-01-20 13:52:20 -08002474static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2475 int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476{
NeilBrowne8096362011-10-11 16:49:05 +11002477 struct r1conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002478 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 struct bio *bio;
2480 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08002481 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08002483 int wonly = -1;
2484 int write_targets = 0, read_targets = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11002485 sector_t sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07002486 int still_degraded = 0;
NeilBrown06f60382011-07-28 11:31:48 +10002487 int good_sectors = RESYNC_SECTORS;
2488 int min_bad = 0; /* number of sectors that are bad in all devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
2490 if (!conf->r1buf_pool)
2491 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002492 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493
Andre Noll58c0fed2009-03-31 14:33:13 +11002494 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002496 /* If we aborted, we need to abort the
2497 * sync on the 'current' bitmap chunk (there will
2498 * only be one in raid1 resync.
2499 * We can find the current addess in mddev->curr_resync
2500 */
NeilBrown6a806c52005-07-15 03:56:35 -07002501 if (mddev->curr_resync < max_sector) /* aborted */
2502 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07002503 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07002504 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07002505 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07002506
2507 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 close_sync(conf);
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002509
2510 if (mddev_is_clustered(mddev)) {
2511 conf->cluster_sync_low = 0;
2512 conf->cluster_sync_high = 0;
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 return 0;
2515 }
2516
NeilBrown07d84d102006-06-26 00:27:56 -07002517 if (mddev->bitmap == NULL &&
2518 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07002519 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07002520 conf->fullsync == 0) {
2521 *skipped = 1;
2522 return max_sector - sector_nr;
2523 }
NeilBrown6394cca2006-08-27 01:23:50 -07002524 /* before building a request, check if we can skip these blocks..
2525 * This call the bitmap_start_sync doesn't actually record anything
2526 */
NeilBrowne3b97032005-08-04 12:53:34 -07002527 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de485f2005-11-08 21:39:38 -08002528 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002529 /* We can skip this block, and probably several more */
2530 *skipped = 1;
2531 return sync_blocks;
2532 }
NeilBrown17999be2006-01-06 00:20:12 -08002533
Tomasz Majchrzak7ac50442016-06-13 15:51:19 +02002534 /*
2535 * If there is non-resync activity waiting for a turn, then let it
2536 * though before starting on this new sync request.
2537 */
2538 if (conf->nr_waiting)
2539 schedule_timeout_uninterruptible(1);
2540
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002541 /* we are incrementing sector_nr below. To be safe, we check against
2542 * sector_nr + two times RESYNC_SECTORS
2543 */
2544
2545 bitmap_cond_end_sync(mddev->bitmap, sector_nr,
2546 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
NeilBrown1c4588e2010-10-26 17:41:22 +11002547 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown17999be2006-01-06 00:20:12 -08002548
NeilBrownc2fd4c92014-09-10 16:01:24 +10002549 raise_barrier(conf, sector_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550
NeilBrown3e198f72006-01-06 00:20:21 -08002551 rcu_read_lock();
2552 /*
2553 * If we get a correctably read error during resync or recovery,
2554 * we might want to read from a different device. So we
2555 * flag all drives that could conceivably be read from for READ,
2556 * and any others (which will be non-In_sync devices) for WRITE.
2557 * If a read fails, we try reading from something else for which READ
2558 * is OK.
2559 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 r1_bio->mddev = mddev;
2562 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07002563 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565
NeilBrown8f19ccb2011-12-23 10:17:56 +11002566 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002567 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 bio = r1_bio->bios[i];
Kent Overstreet2aabaa62012-09-11 11:26:12 -07002569 bio_reset(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
NeilBrown3e198f72006-01-06 00:20:21 -08002571 rdev = rcu_dereference(conf->mirrors[i].rdev);
2572 if (rdev == NULL ||
NeilBrown06f60382011-07-28 11:31:48 +10002573 test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11002574 if (i < conf->raid_disks)
2575 still_degraded = 1;
NeilBrown3e198f72006-01-06 00:20:21 -08002576 } else if (!test_bit(In_sync, &rdev->flags)) {
Mike Christie796a5cf2016-06-05 14:32:07 -05002577 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 bio->bi_end_io = end_sync_write;
2579 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08002580 } else {
2581 /* may need to read from here */
NeilBrown06f60382011-07-28 11:31:48 +10002582 sector_t first_bad = MaxSector;
2583 int bad_sectors;
2584
2585 if (is_badblock(rdev, sector_nr, good_sectors,
2586 &first_bad, &bad_sectors)) {
2587 if (first_bad > sector_nr)
2588 good_sectors = first_bad - sector_nr;
2589 else {
2590 bad_sectors -= (sector_nr - first_bad);
2591 if (min_bad == 0 ||
2592 min_bad > bad_sectors)
2593 min_bad = bad_sectors;
2594 }
NeilBrown3e198f72006-01-06 00:20:21 -08002595 }
NeilBrown06f60382011-07-28 11:31:48 +10002596 if (sector_nr < first_bad) {
2597 if (test_bit(WriteMostly, &rdev->flags)) {
2598 if (wonly < 0)
2599 wonly = i;
2600 } else {
2601 if (disk < 0)
2602 disk = i;
2603 }
Mike Christie796a5cf2016-06-05 14:32:07 -05002604 bio_set_op_attrs(bio, REQ_OP_READ, 0);
NeilBrown06f60382011-07-28 11:31:48 +10002605 bio->bi_end_io = end_sync_read;
2606 read_targets++;
Alexander Lyakasd57368a2012-07-17 13:17:55 +03002607 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2608 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2609 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2610 /*
2611 * The device is suitable for reading (InSync),
2612 * but has bad block(s) here. Let's try to correct them,
2613 * if we are doing resync or repair. Otherwise, leave
2614 * this device alone for this sync request.
2615 */
Mike Christie796a5cf2016-06-05 14:32:07 -05002616 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Alexander Lyakasd57368a2012-07-17 13:17:55 +03002617 bio->bi_end_io = end_sync_write;
2618 write_targets++;
NeilBrown06f60382011-07-28 11:31:48 +10002619 }
NeilBrown3e198f72006-01-06 00:20:21 -08002620 }
NeilBrown06f60382011-07-28 11:31:48 +10002621 if (bio->bi_end_io) {
2622 atomic_inc(&rdev->nr_pending);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002623 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
NeilBrown06f60382011-07-28 11:31:48 +10002624 bio->bi_bdev = rdev->bdev;
2625 bio->bi_private = r1_bio;
2626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 }
NeilBrown3e198f72006-01-06 00:20:21 -08002628 rcu_read_unlock();
2629 if (disk < 0)
2630 disk = wonly;
2631 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07002632
NeilBrown06f60382011-07-28 11:31:48 +10002633 if (read_targets == 0 && min_bad > 0) {
2634 /* These sectors are bad on all InSync devices, so we
2635 * need to mark them bad on all write targets
2636 */
2637 int ok = 1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002638 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
NeilBrown06f60382011-07-28 11:31:48 +10002639 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
majianpenga42f9d82012-04-02 01:04:19 +10002640 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown06f60382011-07-28 11:31:48 +10002641 ok = rdev_set_badblocks(rdev, sector_nr,
2642 min_bad, 0
2643 ) && ok;
2644 }
2645 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2646 *skipped = 1;
2647 put_buf(r1_bio);
2648
2649 if (!ok) {
2650 /* Cannot record the badblocks, so need to
2651 * abort the resync.
2652 * If there are multiple read targets, could just
2653 * fail the really bad ones ???
2654 */
2655 conf->recovery_disabled = mddev->recovery_disabled;
2656 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2657 return 0;
2658 } else
2659 return min_bad;
2660
2661 }
2662 if (min_bad > 0 && min_bad < good_sectors) {
2663 /* only resync enough to reach the next bad->good
2664 * transition */
2665 good_sectors = min_bad;
2666 }
2667
NeilBrown3e198f72006-01-06 00:20:21 -08002668 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2669 /* extra read targets are also write targets */
2670 write_targets += read_targets-1;
2671
2672 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 /* There is nowhere to write, so all non-sync
2674 * drives must be failed - so we are finished
2675 */
NeilBrownb7219cc2012-07-31 10:05:34 +10002676 sector_t rv;
2677 if (min_bad > 0)
2678 max_sector = sector_nr + min_bad;
2679 rv = max_sector - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07002680 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 return rv;
2683 }
2684
NeilBrownc6207272008-02-06 01:39:52 -08002685 if (max_sector > mddev->resync_max)
2686 max_sector = mddev->resync_max; /* Don't do IO beyond here */
NeilBrown06f60382011-07-28 11:31:48 +10002687 if (max_sector > sector_nr + good_sectors)
2688 max_sector = sector_nr + good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07002690 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 do {
2692 struct page *page;
2693 int len = PAGE_SIZE;
2694 if (sector_nr + (len>>9) > max_sector)
2695 len = (max_sector - sector_nr) << 9;
2696 if (len == 0)
2697 break;
NeilBrown6a806c52005-07-15 03:56:35 -07002698 if (sync_blocks == 0) {
2699 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de485f2005-11-08 21:39:38 -08002700 &sync_blocks, still_degraded) &&
2701 !conf->fullsync &&
2702 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07002703 break;
NeilBrown7571ae82010-10-07 11:54:46 +11002704 if ((len >> 9) > sync_blocks)
NeilBrown6a806c52005-07-15 03:56:35 -07002705 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07002706 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002707
NeilBrown8f19ccb2011-12-23 10:17:56 +11002708 for (i = 0 ; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 bio = r1_bio->bios[i];
2710 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08002711 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 if (bio_add_page(bio, page, len, 0) == 0) {
2713 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08002714 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 while (i > 0) {
2716 i--;
2717 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07002718 if (bio->bi_end_io==NULL)
2719 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 /* remove last page from this bio */
2721 bio->bi_vcnt--;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002722 bio->bi_iter.bi_size -= len;
Jens Axboeb7c44ed2015-07-24 12:37:59 -06002723 bio_clear_flag(bio, BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 }
2725 goto bio_full;
2726 }
2727 }
2728 }
2729 nr_sectors += len>>9;
2730 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07002731 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2733 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 r1_bio->sectors = nr_sectors;
2735
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002736 if (mddev_is_clustered(mddev) &&
2737 conf->cluster_sync_high < sector_nr + nr_sectors) {
2738 conf->cluster_sync_low = mddev->curr_resync_completed;
2739 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2740 /* Send resync message */
2741 md_cluster_ops->resync_info_update(mddev,
2742 conf->cluster_sync_low,
2743 conf->cluster_sync_high);
2744 }
2745
NeilBrownd11c1712006-01-06 00:20:26 -08002746 /* For a user-requested sync, we read all readable devices and do a
2747 * compare
2748 */
2749 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2750 atomic_set(&r1_bio->remaining, read_targets);
NeilBrown2d4f4f32012-07-09 11:34:13 +10002751 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
NeilBrownd11c1712006-01-06 00:20:26 -08002752 bio = r1_bio->bios[i];
2753 if (bio->bi_end_io == end_sync_read) {
NeilBrown2d4f4f32012-07-09 11:34:13 +10002754 read_targets--;
NeilBrownddac7c72006-08-31 21:27:36 -07002755 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002756 generic_make_request(bio);
2757 }
2758 }
2759 } else {
2760 atomic_set(&r1_bio->remaining, 1);
2761 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07002762 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002763 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
NeilBrownd11c1712006-01-06 00:20:26 -08002765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 return nr_sectors;
2767}
2768
NeilBrownfd01b882011-10-11 16:47:53 +11002769static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07002770{
2771 if (sectors)
2772 return sectors;
2773
2774 return mddev->dev_sectors;
2775}
2776
NeilBrowne8096362011-10-11 16:49:05 +11002777static struct r1conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778{
NeilBrowne8096362011-10-11 16:49:05 +11002779 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002780 int i;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002781 struct raid1_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11002782 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002783 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784
NeilBrowne8096362011-10-11 16:49:05 +11002785 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11002787 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002789 conf->mirrors = kzalloc(sizeof(struct raid1_info)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002790 * mddev->raid_disks * 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 GFP_KERNEL);
2792 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11002793 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
NeilBrownddaf22a2006-01-06 00:20:19 -08002795 conf->tmppage = alloc_page(GFP_KERNEL);
2796 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11002797 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08002798
NeilBrown709ae482009-12-14 12:49:51 +11002799 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11002801 goto abort;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002802 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2804 r1bio_pool_free,
2805 conf->poolinfo);
2806 if (!conf->r1bio_pool)
NeilBrown709ae482009-12-14 12:49:51 +11002807 goto abort;
2808
NeilBrowned9bfdf2009-10-16 15:55:44 +11002809 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
NeilBrownc19d5792011-12-23 10:17:57 +11002811 err = -EINVAL;
Neil Browne7e72bf2008-05-14 16:05:54 -07002812 spin_lock_init(&conf->device_lock);
NeilBrowndafb20f2012-03-19 12:46:39 +11002813 rdev_for_each(rdev, mddev) {
NeilBrownaba336b2012-05-31 15:39:11 +10002814 struct request_queue *q;
NeilBrown709ae482009-12-14 12:49:51 +11002815 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 if (disk_idx >= mddev->raid_disks
2817 || disk_idx < 0)
2818 continue;
NeilBrownc19d5792011-12-23 10:17:57 +11002819 if (test_bit(Replacement, &rdev->flags))
NeilBrown02b898f2012-10-31 11:42:03 +11002820 disk = conf->mirrors + mddev->raid_disks + disk_idx;
NeilBrownc19d5792011-12-23 10:17:57 +11002821 else
2822 disk = conf->mirrors + disk_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823
NeilBrownc19d5792011-12-23 10:17:57 +11002824 if (disk->rdev)
2825 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 disk->rdev = rdev;
NeilBrownaba336b2012-05-31 15:39:11 +10002827 q = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828
2829 disk->head_position = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +10002830 disk->seq_start = MaxSector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 }
2832 conf->raid_disks = mddev->raid_disks;
2833 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 INIT_LIST_HEAD(&conf->retry_list);
NeilBrown55ce74d2015-08-14 11:11:10 +10002835 INIT_LIST_HEAD(&conf->bio_end_io_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836
2837 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08002838 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839
NeilBrown191ea9b2005-06-21 17:17:23 -07002840 bio_list_init(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +11002841 conf->pending_count = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11002842 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown191ea9b2005-06-21 17:17:23 -07002843
majianpeng79ef3a82013-11-15 14:55:02 +08002844 conf->start_next_window = MaxSector;
2845 conf->current_window_requests = conf->next_window_requests = 0;
2846
NeilBrownc19d5792011-12-23 10:17:57 +11002847 err = -EIO;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002848 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849
2850 disk = conf->mirrors + i;
2851
NeilBrownc19d5792011-12-23 10:17:57 +11002852 if (i < conf->raid_disks &&
2853 disk[conf->raid_disks].rdev) {
2854 /* This slot has a replacement. */
2855 if (!disk->rdev) {
2856 /* No original, just make the replacement
2857 * a recovering spare
2858 */
2859 disk->rdev =
2860 disk[conf->raid_disks].rdev;
2861 disk[conf->raid_disks].rdev = NULL;
2862 } else if (!test_bit(In_sync, &disk->rdev->flags))
2863 /* Original is not in_sync - bad */
2864 goto abort;
2865 }
2866
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002867 if (!disk->rdev ||
2868 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 disk->head_position = 0;
Jonathan Brassow4f0a5e02012-05-22 13:55:31 +10002870 if (disk->rdev &&
2871 (disk->rdev->saved_raid_disk < 0))
NeilBrown918f0232007-08-22 14:01:52 -07002872 conf->fullsync = 1;
Shaohua Libe4d3282012-07-31 10:03:53 +10002873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 }
NeilBrown709ae482009-12-14 12:49:51 +11002875
NeilBrown709ae482009-12-14 12:49:51 +11002876 err = -ENOMEM;
NeilBrown02326052012-07-03 15:56:52 +10002877 conf->thread = md_register_thread(raid1d, mddev, "raid1");
NeilBrown709ae482009-12-14 12:49:51 +11002878 if (!conf->thread) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002879 printk(KERN_ERR
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002880 "md/raid1:%s: couldn't allocate thread\n",
NeilBrown191ea9b2005-06-21 17:17:23 -07002881 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11002882 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002884
NeilBrown709ae482009-12-14 12:49:51 +11002885 return conf;
2886
2887 abort:
2888 if (conf) {
Julia Lawall644df1a2015-09-13 14:15:10 +02002889 mempool_destroy(conf->r1bio_pool);
NeilBrown709ae482009-12-14 12:49:51 +11002890 kfree(conf->mirrors);
2891 safe_put_page(conf->tmppage);
2892 kfree(conf->poolinfo);
2893 kfree(conf);
2894 }
2895 return ERR_PTR(err);
2896}
2897
NeilBrownafa0f552014-12-15 12:56:58 +11002898static void raid1_free(struct mddev *mddev, void *priv);
Shaohua Li849674e2016-01-20 13:52:20 -08002899static int raid1_run(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002900{
NeilBrowne8096362011-10-11 16:49:05 +11002901 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002902 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11002903 struct md_rdev *rdev;
majianpeng5220ea12012-04-02 09:48:38 +10002904 int ret;
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002905 bool discard_supported = false;
NeilBrown709ae482009-12-14 12:49:51 +11002906
2907 if (mddev->level != 1) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002908 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
NeilBrown709ae482009-12-14 12:49:51 +11002909 mdname(mddev), mddev->level);
2910 return -EIO;
2911 }
2912 if (mddev->reshape_position != MaxSector) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002913 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
NeilBrown709ae482009-12-14 12:49:51 +11002914 mdname(mddev));
2915 return -EIO;
2916 }
2917 /*
2918 * copy the already verified devices into our private RAID1
2919 * bookkeeping area. [whatever we allocate in run(),
NeilBrownafa0f552014-12-15 12:56:58 +11002920 * should be freed in raid1_free()]
NeilBrown709ae482009-12-14 12:49:51 +11002921 */
2922 if (mddev->private == NULL)
2923 conf = setup_conf(mddev);
2924 else
2925 conf = mddev->private;
2926
2927 if (IS_ERR(conf))
2928 return PTR_ERR(conf);
2929
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11002930 if (mddev->queue)
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07002931 blk_queue_max_write_same_sectors(mddev->queue, 0);
2932
NeilBrowndafb20f2012-03-19 12:46:39 +11002933 rdev_for_each(rdev, mddev) {
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002934 if (!mddev->gendisk)
2935 continue;
NeilBrown709ae482009-12-14 12:49:51 +11002936 disk_stack_limits(mddev->gendisk, rdev->bdev,
2937 rdev->data_offset << 9);
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002938 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
2939 discard_supported = true;
NeilBrown709ae482009-12-14 12:49:51 +11002940 }
2941
2942 mddev->degraded = 0;
2943 for (i=0; i < conf->raid_disks; i++)
2944 if (conf->mirrors[i].rdev == NULL ||
2945 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2946 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2947 mddev->degraded++;
2948
2949 if (conf->raid_disks - mddev->degraded == 1)
2950 mddev->recovery_cp = MaxSector;
2951
Andre Noll8c6ac862009-06-18 08:48:06 +10002952 if (mddev->recovery_cp != MaxSector)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002953 printk(KERN_NOTICE "md/raid1:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10002954 " -- starting background reconstruction\n",
2955 mdname(mddev));
NeilBrownf72ffdd2014-09-30 14:23:59 +10002956 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002957 "md/raid1:%s: active with %d out of %d mirrors\n",
NeilBrownf72ffdd2014-09-30 14:23:59 +10002958 mdname(mddev), mddev->raid_disks - mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11002960
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 /*
2962 * Ok, everything is just fine now
2963 */
NeilBrown709ae482009-12-14 12:49:51 +11002964 mddev->thread = conf->thread;
2965 conf->thread = NULL;
2966 mddev->private = conf;
2967
Dan Williams1f403622009-03-31 14:59:03 +11002968 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002970 if (mddev->queue) {
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002971 if (discard_supported)
2972 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
2973 mddev->queue);
2974 else
2975 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
2976 mddev->queue);
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002977 }
majianpeng5220ea12012-04-02 09:48:38 +10002978
2979 ret = md_integrity_register(mddev);
NeilBrown5aa61f42014-12-15 12:56:57 +11002980 if (ret) {
2981 md_unregister_thread(&mddev->thread);
NeilBrownafa0f552014-12-15 12:56:58 +11002982 raid1_free(mddev, conf);
NeilBrown5aa61f42014-12-15 12:56:57 +11002983 }
majianpeng5220ea12012-04-02 09:48:38 +10002984 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985}
2986
NeilBrownafa0f552014-12-15 12:56:58 +11002987static void raid1_free(struct mddev *mddev, void *priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988{
NeilBrownafa0f552014-12-15 12:56:58 +11002989 struct r1conf *conf = priv;
NeilBrown4b6d2872005-09-09 16:23:47 -07002990
Julia Lawall644df1a2015-09-13 14:15:10 +02002991 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002992 kfree(conf->mirrors);
Hirokazu Takahashi0fea7ed2013-04-24 11:42:44 +10002993 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002994 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996}
2997
NeilBrownfd01b882011-10-11 16:47:53 +11002998static int raid1_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999{
3000 /* no resync is happening, and there is enough space
3001 * on all devices, so we can resize.
3002 * We need to make sure resync covers any new space.
3003 * If the array is shrinking we should possibly wait until
3004 * any io in the removed space completes, but it hardly seems
3005 * worth it.
3006 */
NeilBrowna4a61252012-05-22 13:55:27 +10003007 sector_t newsize = raid1_size(mddev, sectors, 0);
3008 if (mddev->external_size &&
3009 mddev->array_sectors > newsize)
Dan Williamsb522adc2009-03-31 15:00:31 +11003010 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10003011 if (mddev->bitmap) {
3012 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
3013 if (ret)
3014 return ret;
3015 }
3016 md_set_array_sectors(mddev, newsize);
Andre Nollf233ea52008-07-21 17:05:22 +10003017 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10003018 revalidate_disk(mddev->gendisk);
Dan Williamsb522adc2009-03-31 15:00:31 +11003019 if (sectors > mddev->dev_sectors &&
NeilBrownb0986362011-05-11 15:52:21 +10003020 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11003021 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3023 }
Dan Williamsb522adc2009-03-31 15:00:31 +11003024 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07003025 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 return 0;
3027}
3028
NeilBrownfd01b882011-10-11 16:47:53 +11003029static int raid1_reshape(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030{
3031 /* We need to:
3032 * 1/ resize the r1bio_pool
3033 * 2/ resize conf->mirrors
3034 *
3035 * We allocate a new r1bio_pool if we can.
3036 * Then raise a device barrier and wait until all IO stops.
3037 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07003038 *
3039 * At the same time, we "pack" the devices so that all the missing
3040 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 */
3042 mempool_t *newpool, *oldpool;
3043 struct pool_info *newpoolinfo;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10003044 struct raid1_info *newmirrors;
NeilBrowne8096362011-10-11 16:49:05 +11003045 struct r1conf *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08003046 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07003047 unsigned long flags;
Dan Williamsb5470dc2008-06-27 21:44:04 -07003048 int d, d2, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049
NeilBrown63c70c42006-03-27 01:18:13 -08003050 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10003051 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08003052 mddev->layout != mddev->new_layout ||
3053 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10003054 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08003055 mddev->new_layout = mddev->layout;
3056 mddev->new_level = mddev->level;
3057 return -EINVAL;
3058 }
3059
Goldwyn Rodrigues28c1b9f2015-10-22 16:01:25 +11003060 if (!mddev_is_clustered(mddev)) {
3061 err = md_allow_write(mddev);
3062 if (err)
3063 return err;
3064 }
NeilBrown2a2275d2007-01-26 00:57:11 -08003065
NeilBrown63c70c42006-03-27 01:18:13 -08003066 raid_disks = mddev->raid_disks + mddev->delta_disks;
3067
NeilBrown6ea9c072005-06-21 17:17:09 -07003068 if (raid_disks < conf->raid_disks) {
3069 cnt=0;
3070 for (d= 0; d < conf->raid_disks; d++)
3071 if (conf->mirrors[d].rdev)
3072 cnt++;
3073 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07003075 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076
3077 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3078 if (!newpoolinfo)
3079 return -ENOMEM;
3080 newpoolinfo->mddev = mddev;
NeilBrown8f19ccb2011-12-23 10:17:56 +11003081 newpoolinfo->raid_disks = raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082
3083 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3084 r1bio_pool_free, newpoolinfo);
3085 if (!newpool) {
3086 kfree(newpoolinfo);
3087 return -ENOMEM;
3088 }
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10003089 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
NeilBrown8f19ccb2011-12-23 10:17:56 +11003090 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 if (!newmirrors) {
3092 kfree(newpoolinfo);
3093 mempool_destroy(newpool);
3094 return -ENOMEM;
3095 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
NeilBrowne2d59922013-06-12 11:01:22 +10003097 freeze_array(conf, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098
3099 /* ok, everything is stopped */
3100 oldpool = conf->r1bio_pool;
3101 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07003102
NeilBrowna88aa782007-08-22 14:01:53 -07003103 for (d = d2 = 0; d < conf->raid_disks; d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11003104 struct md_rdev *rdev = conf->mirrors[d].rdev;
NeilBrowna88aa782007-08-22 14:01:53 -07003105 if (rdev && rdev->raid_disk != d2) {
Namhyung Kim36fad852011-07-27 11:00:36 +10003106 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna88aa782007-08-22 14:01:53 -07003107 rdev->raid_disk = d2;
Namhyung Kim36fad852011-07-27 11:00:36 +10003108 sysfs_unlink_rdev(mddev, rdev);
3109 if (sysfs_link_rdev(mddev, rdev))
NeilBrowna88aa782007-08-22 14:01:53 -07003110 printk(KERN_WARNING
Namhyung Kim36fad852011-07-27 11:00:36 +10003111 "md/raid1:%s: cannot register rd%d\n",
3112 mdname(mddev), rdev->raid_disk);
NeilBrown6ea9c072005-06-21 17:17:09 -07003113 }
NeilBrowna88aa782007-08-22 14:01:53 -07003114 if (rdev)
3115 newmirrors[d2++].rdev = rdev;
3116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 kfree(conf->mirrors);
3118 conf->mirrors = newmirrors;
3119 kfree(conf->poolinfo);
3120 conf->poolinfo = newpoolinfo;
3121
NeilBrownc04be0a2006-10-03 01:15:53 -07003122 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07003124 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08003126 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127
NeilBrowne2d59922013-06-12 11:01:22 +10003128 unfreeze_array(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129
NeilBrown985ca972015-07-06 12:26:57 +10003130 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3132 md_wakeup_thread(mddev->thread);
3133
3134 mempool_destroy(oldpool);
3135 return 0;
3136}
3137
NeilBrownfd01b882011-10-11 16:47:53 +11003138static void raid1_quiesce(struct mddev *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07003139{
NeilBrowne8096362011-10-11 16:49:05 +11003140 struct r1conf *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07003141
3142 switch(state) {
NeilBrown6eef4b22009-12-14 12:49:51 +11003143 case 2: /* wake for suspend */
3144 wake_up(&conf->wait_barrier);
3145 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07003146 case 1:
majianpeng07169fd2013-11-14 15:16:18 +11003147 freeze_array(conf, 0);
NeilBrown36fa3062005-09-09 16:23:45 -07003148 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07003149 case 0:
majianpeng07169fd2013-11-14 15:16:18 +11003150 unfreeze_array(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07003151 break;
3152 }
NeilBrown36fa3062005-09-09 16:23:45 -07003153}
3154
NeilBrownfd01b882011-10-11 16:47:53 +11003155static void *raid1_takeover(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11003156{
3157 /* raid1 can take over:
3158 * raid5 with 2 devices, any layout or chunk size
3159 */
3160 if (mddev->level == 5 && mddev->raid_disks == 2) {
NeilBrowne8096362011-10-11 16:49:05 +11003161 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11003162 mddev->new_level = 1;
3163 mddev->new_layout = 0;
3164 mddev->new_chunk_sectors = 0;
3165 conf = setup_conf(mddev);
3166 if (!IS_ERR(conf))
majianpeng07169fd2013-11-14 15:16:18 +11003167 /* Array must appear to be quiesced */
3168 conf->array_frozen = 1;
NeilBrown709ae482009-12-14 12:49:51 +11003169 return conf;
3170 }
3171 return ERR_PTR(-EINVAL);
3172}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173
NeilBrown84fc4b52011-10-11 16:49:58 +11003174static struct md_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175{
3176 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08003177 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 .owner = THIS_MODULE,
Shaohua Li849674e2016-01-20 13:52:20 -08003179 .make_request = raid1_make_request,
3180 .run = raid1_run,
NeilBrownafa0f552014-12-15 12:56:58 +11003181 .free = raid1_free,
Shaohua Li849674e2016-01-20 13:52:20 -08003182 .status = raid1_status,
3183 .error_handler = raid1_error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 .hot_add_disk = raid1_add_disk,
3185 .hot_remove_disk= raid1_remove_disk,
3186 .spare_active = raid1_spare_active,
Shaohua Li849674e2016-01-20 13:52:20 -08003187 .sync_request = raid1_sync_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07003189 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08003190 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07003191 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11003192 .takeover = raid1_takeover,
NeilBrown5c675f82014-12-15 12:56:56 +11003193 .congested = raid1_congested,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194};
3195
3196static int __init raid_init(void)
3197{
NeilBrown2604b702006-01-06 00:20:36 -08003198 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199}
3200
3201static void raid_exit(void)
3202{
NeilBrown2604b702006-01-06 00:20:36 -08003203 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204}
3205
3206module_init(raid_init);
3207module_exit(raid_exit);
3208MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11003209MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08003211MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08003212MODULE_ALIAS("md-level-1");
NeilBrown34db0cd2011-10-11 16:50:01 +11003213
3214module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);