blob: 15f0b552bf4883ccdb1cbe44345f4a11bb41574d [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:
Guoqing Jiang491221f2016-09-22 03:10:01 -0400148 while (--j >= 0)
149 bio_free_pages(r1_bio->bios[j]);
NeilBrownda1aab32014-04-09 12:25:43 +1000150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151out_free_bio:
NeilBrown8f19ccb2011-12-23 10:17:56 +1100152 while (++j < pi->raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 bio_put(r1_bio->bios[j]);
154 r1bio_pool_free(r1_bio, data);
155 return NULL;
156}
157
158static void r1buf_pool_free(void *__r1_bio, void *data)
159{
160 struct pool_info *pi = data;
NeilBrownd11c1712006-01-06 00:20:26 -0800161 int i,j;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100162 struct r1bio *r1bio = __r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
NeilBrownd11c1712006-01-06 00:20:26 -0800164 for (i = 0; i < RESYNC_PAGES; i++)
165 for (j = pi->raid_disks; j-- ;) {
166 if (j == 0 ||
167 r1bio->bios[j]->bi_io_vec[i].bv_page !=
168 r1bio->bios[0]->bi_io_vec[i].bv_page)
NeilBrown1345b1d2006-01-06 00:20:40 -0800169 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 for (i=0 ; i < pi->raid_disks; i++)
172 bio_put(r1bio->bios[i]);
173
174 r1bio_pool_free(r1bio, data);
175}
176
NeilBrowne8096362011-10-11 16:49:05 +1100177static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
179 int i;
180
NeilBrown8f19ccb2011-12-23 10:17:56 +1100181 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 struct bio **bio = r1_bio->bios + i;
NeilBrown4367af52011-07-28 11:31:49 +1000183 if (!BIO_SPECIAL(*bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 bio_put(*bio);
185 *bio = NULL;
186 }
187}
188
NeilBrown9f2c9d12011-10-11 16:48:43 +1100189static void free_r1bio(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
NeilBrowne8096362011-10-11 16:49:05 +1100191 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 put_all_bios(conf, r1_bio);
194 mempool_free(r1_bio, conf->r1bio_pool);
195}
196
NeilBrown9f2c9d12011-10-11 16:48:43 +1100197static void put_buf(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
NeilBrowne8096362011-10-11 16:49:05 +1100199 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown3e198f72006-01-06 00:20:21 -0800200 int i;
201
NeilBrown8f19ccb2011-12-23 10:17:56 +1100202 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3e198f72006-01-06 00:20:21 -0800203 struct bio *bio = r1_bio->bios[i];
204 if (bio->bi_end_io)
205 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
208 mempool_free(r1_bio, conf->r1buf_pool);
209
NeilBrown17999be2006-01-06 00:20:12 -0800210 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
NeilBrown9f2c9d12011-10-11 16:48:43 +1100213static void reschedule_retry(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 unsigned long flags;
NeilBrownfd01b882011-10-11 16:47:53 +1100216 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +1100217 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219 spin_lock_irqsave(&conf->device_lock, flags);
220 list_add(&r1_bio->retry_list, &conf->retry_list);
NeilBrownddaf22a2006-01-06 00:20:19 -0800221 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 spin_unlock_irqrestore(&conf->device_lock, flags);
223
NeilBrown17999be2006-01-06 00:20:12 -0800224 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 md_wakeup_thread(mddev->thread);
226}
227
228/*
229 * raid_end_bio_io() is called when we have finished servicing a mirrored
230 * operation and are ready to return a success/failure code to the buffer
231 * cache layer.
232 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100233static void call_bio_endio(struct r1bio *r1_bio)
NeilBrownd2eb35a2011-07-28 11:31:48 +1000234{
235 struct bio *bio = r1_bio->master_bio;
236 int done;
NeilBrowne8096362011-10-11 16:49:05 +1100237 struct r1conf *conf = r1_bio->mddev->private;
majianpeng79ef3a82013-11-15 14:55:02 +0800238 sector_t start_next_window = r1_bio->start_next_window;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700239 sector_t bi_sector = bio->bi_iter.bi_sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000240
241 if (bio->bi_phys_segments) {
242 unsigned long flags;
243 spin_lock_irqsave(&conf->device_lock, flags);
244 bio->bi_phys_segments--;
245 done = (bio->bi_phys_segments == 0);
246 spin_unlock_irqrestore(&conf->device_lock, flags);
majianpeng79ef3a82013-11-15 14:55:02 +0800247 /*
248 * make_request() might be waiting for
249 * bi_phys_segments to decrease
250 */
251 wake_up(&conf->wait_barrier);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000252 } else
253 done = 1;
254
255 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200256 bio->bi_error = -EIO;
257
NeilBrownd2eb35a2011-07-28 11:31:48 +1000258 if (done) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200259 bio_endio(bio);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000260 /*
261 * Wake up any possible resync thread that waits for the device
262 * to go idle.
263 */
majianpeng79ef3a82013-11-15 14:55:02 +0800264 allow_barrier(conf, start_next_window, bi_sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000265 }
266}
267
NeilBrown9f2c9d12011-10-11 16:48:43 +1100268static void raid_end_bio_io(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
270 struct bio *bio = r1_bio->master_bio;
271
NeilBrown4b6d2872005-09-09 16:23:47 -0700272 /* if nobody has done the final endio yet, do it now */
273 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
NeilBrown36a4e1f2011-10-07 14:23:17 +1100274 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
275 (bio_data_dir(bio) == WRITE) ? "write" : "read",
Kent Overstreet4f024f32013-10-11 15:44:27 -0700276 (unsigned long long) bio->bi_iter.bi_sector,
277 (unsigned long long) bio_end_sector(bio) - 1);
NeilBrown4b6d2872005-09-09 16:23:47 -0700278
NeilBrownd2eb35a2011-07-28 11:31:48 +1000279 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 free_r1bio(r1_bio);
282}
283
284/*
285 * Update disk head position estimator based on IRQ completion info.
286 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100287static inline void update_head_pos(int disk, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
NeilBrowne8096362011-10-11 16:49:05 +1100289 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
291 conf->mirrors[disk].head_position =
292 r1_bio->sector + (r1_bio->sectors);
293}
294
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100295/*
296 * Find the disk number which triggered given bio
297 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100298static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100299{
300 int mirror;
NeilBrown30194632011-12-23 10:17:56 +1100301 struct r1conf *conf = r1_bio->mddev->private;
302 int raid_disks = conf->raid_disks;
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100303
NeilBrown8f19ccb2011-12-23 10:17:56 +1100304 for (mirror = 0; mirror < raid_disks * 2; mirror++)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100305 if (r1_bio->bios[mirror] == bio)
306 break;
307
NeilBrown8f19ccb2011-12-23 10:17:56 +1100308 BUG_ON(mirror == raid_disks * 2);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100309 update_head_pos(mirror, r1_bio);
310
311 return mirror;
312}
313
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200314static void raid1_end_read_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200316 int uptodate = !bio->bi_error;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100317 struct r1bio *r1_bio = bio->bi_private;
NeilBrowne8096362011-10-11 16:49:05 +1100318 struct r1conf *conf = r1_bio->mddev->private;
NeilBrowne5872d52016-06-02 16:19:52 +1000319 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 /*
322 * this branch is our 'one mirror IO has finished' event handler:
323 */
NeilBrowne5872d52016-06-02 16:19:52 +1000324 update_head_pos(r1_bio->read_disk, r1_bio);
NeilBrownddaf22a2006-01-06 00:20:19 -0800325
NeilBrowndd00a992007-05-10 03:15:50 -0700326 if (uptodate)
327 set_bit(R1BIO_Uptodate, &r1_bio->state);
328 else {
329 /* If all other devices have failed, we want to return
330 * the error upwards rather than fail the last device.
331 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 */
NeilBrowndd00a992007-05-10 03:15:50 -0700333 unsigned long flags;
334 spin_lock_irqsave(&conf->device_lock, flags);
335 if (r1_bio->mddev->degraded == conf->raid_disks ||
336 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
NeilBrowne5872d52016-06-02 16:19:52 +1000337 test_bit(In_sync, &rdev->flags)))
NeilBrowndd00a992007-05-10 03:15:50 -0700338 uptodate = 1;
339 spin_unlock_irqrestore(&conf->device_lock, flags);
340 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100342 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 raid_end_bio_io(r1_bio);
NeilBrowne5872d52016-06-02 16:19:52 +1000344 rdev_dec_pending(rdev, conf->mddev);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100345 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 /*
347 * oops, read error:
348 */
349 char b[BDEVNAME_SIZE];
NeilBrown1d41c212016-11-02 14:16:50 +1100350 pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n",
351 mdname(conf->mddev),
352 bdevname(rdev->bdev, b),
353 (unsigned long long)r1_bio->sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000354 set_bit(R1BIO_ReadError, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 reschedule_retry(r1_bio);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100356 /* don't drop the reference on read_disk yet */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358}
359
NeilBrown9f2c9d12011-10-11 16:48:43 +1100360static void close_write(struct r1bio *r1_bio)
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000361{
362 /* it really is the end of this request */
363 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
364 /* free extra copy of the data pages */
365 int i = r1_bio->behind_page_count;
366 while (i--)
367 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
368 kfree(r1_bio->behind_bvecs);
369 r1_bio->behind_bvecs = NULL;
370 }
371 /* clear the bitmap if all writes complete successfully */
372 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
373 r1_bio->sectors,
374 !test_bit(R1BIO_Degraded, &r1_bio->state),
375 test_bit(R1BIO_BehindIO, &r1_bio->state));
376 md_write_end(r1_bio->mddev);
377}
378
NeilBrown9f2c9d12011-10-11 16:48:43 +1100379static void r1_bio_write_done(struct r1bio *r1_bio)
NeilBrown4e780642010-10-19 12:54:01 +1100380{
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000381 if (!atomic_dec_and_test(&r1_bio->remaining))
382 return;
383
384 if (test_bit(R1BIO_WriteError, &r1_bio->state))
385 reschedule_retry(r1_bio);
386 else {
387 close_write(r1_bio);
NeilBrown4367af52011-07-28 11:31:49 +1000388 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
389 reschedule_retry(r1_bio);
390 else
391 raid_end_bio_io(r1_bio);
NeilBrown4e780642010-10-19 12:54:01 +1100392 }
393}
394
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200395static void raid1_end_write_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
NeilBrown9f2c9d12011-10-11 16:48:43 +1100397 struct r1bio *r1_bio = bio->bi_private;
NeilBrowne5872d52016-06-02 16:19:52 +1000398 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
NeilBrowne8096362011-10-11 16:49:05 +1100399 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown04b857f2006-03-09 17:33:46 -0800400 struct bio *to_put = NULL;
NeilBrowne5872d52016-06-02 16:19:52 +1000401 int mirror = find_bio_disk(r1_bio, bio);
402 struct md_rdev *rdev = conf->mirrors[mirror].rdev;
Shaohua Lie3f948c2016-10-06 14:09:16 -0700403 bool discard_error;
404
405 discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Tejun Heoe9c74692010-09-03 11:56:18 +0200407 /*
408 * 'one mirror IO has finished' event handler:
409 */
Shaohua Lie3f948c2016-10-06 14:09:16 -0700410 if (bio->bi_error && !discard_error) {
NeilBrowne5872d52016-06-02 16:19:52 +1000411 set_bit(WriteErrorSeen, &rdev->flags);
412 if (!test_and_set_bit(WantReplacement, &rdev->flags))
NeilBrown19d67162011-12-23 10:17:57 +1100413 set_bit(MD_RECOVERY_NEEDED, &
414 conf->mddev->recovery);
415
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000416 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +1000417 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 /*
Tejun Heoe9c74692010-09-03 11:56:18 +0200419 * Set R1BIO_Uptodate in our master bio, so that we
420 * will return a good error code for to the higher
421 * levels even if IO on some other mirrored buffer
422 * fails.
423 *
424 * The 'master' represents the composite IO operation
425 * to user-side. So if something waits for IO, then it
426 * will wait for the 'master' bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 */
NeilBrown4367af52011-07-28 11:31:49 +1000428 sector_t first_bad;
429 int bad_sectors;
430
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000431 r1_bio->bios[mirror] = NULL;
432 to_put = bio;
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300433 /*
434 * Do not set R1BIO_Uptodate if the current device is
435 * rebuilding or Faulty. This is because we cannot use
436 * such device for properly reading the data back (we could
437 * potentially use it, if the current write would have felt
438 * before rdev->recovery_offset, but for simplicity we don't
439 * check this here.
440 */
NeilBrowne5872d52016-06-02 16:19:52 +1000441 if (test_bit(In_sync, &rdev->flags) &&
442 !test_bit(Faulty, &rdev->flags))
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300443 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
NeilBrown4367af52011-07-28 11:31:49 +1000445 /* Maybe we can clear some bad blocks. */
NeilBrowne5872d52016-06-02 16:19:52 +1000446 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
Shaohua Lie3f948c2016-10-06 14:09:16 -0700447 &first_bad, &bad_sectors) && !discard_error) {
NeilBrown4367af52011-07-28 11:31:49 +1000448 r1_bio->bios[mirror] = IO_MADE_GOOD;
449 set_bit(R1BIO_MadeGood, &r1_bio->state);
450 }
451 }
452
Tejun Heoe9c74692010-09-03 11:56:18 +0200453 if (behind) {
NeilBrowne5872d52016-06-02 16:19:52 +1000454 if (test_bit(WriteMostly, &rdev->flags))
Tejun Heoe9c74692010-09-03 11:56:18 +0200455 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700456
Tejun Heoe9c74692010-09-03 11:56:18 +0200457 /*
458 * In behind mode, we ACK the master bio once the I/O
459 * has safely reached all non-writemostly
460 * disks. Setting the Returned bit ensures that this
461 * gets done only once -- we don't ever want to return
462 * -EIO here, instead we'll wait
463 */
464 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
465 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
466 /* Maybe we can return now */
467 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
468 struct bio *mbio = r1_bio->master_bio;
NeilBrown36a4e1f2011-10-07 14:23:17 +1100469 pr_debug("raid1: behind end write sectors"
470 " %llu-%llu\n",
Kent Overstreet4f024f32013-10-11 15:44:27 -0700471 (unsigned long long) mbio->bi_iter.bi_sector,
472 (unsigned long long) bio_end_sector(mbio) - 1);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000473 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700474 }
475 }
476 }
NeilBrown4367af52011-07-28 11:31:49 +1000477 if (r1_bio->bios[mirror] == NULL)
NeilBrowne5872d52016-06-02 16:19:52 +1000478 rdev_dec_pending(rdev, conf->mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200479
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 * Let's see if all mirrored write operations have finished
482 * already.
483 */
NeilBrownaf6d7b72011-05-11 14:51:19 +1000484 r1_bio_write_done(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700485
NeilBrown04b857f2006-03-09 17:33:46 -0800486 if (to_put)
487 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488}
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490/*
491 * This routine returns the disk from which the requested read should
492 * be done. There is a per-array 'next expected sequential IO' sector
493 * number - if this matches on the next IO then we use the last disk.
494 * There is also a per-disk 'last know head position' sector that is
495 * maintained from IRQ contexts, both the normal and the resync IO
496 * completion handlers update this position correctly. If there is no
497 * perfect sequential match then we pick the disk whose head is closest.
498 *
499 * If there are 2 mirrors in the same 2 devices, performance degrades
500 * because position is mirror, not device based.
501 *
502 * The rdev for the device selected will have nr_pending incremented.
503 */
NeilBrowne8096362011-10-11 16:49:05 +1100504static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000506 const sector_t this_sector = r1_bio->sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000507 int sectors;
508 int best_good_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000509 int best_disk, best_dist_disk, best_pending_disk;
510 int has_nonrot_disk;
Shaohua Libe4d3282012-07-31 10:03:53 +1000511 int disk;
NeilBrown76073052011-05-11 14:34:56 +1000512 sector_t best_dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000513 unsigned int min_pending;
NeilBrown3cb03002011-10-11 16:45:26 +1100514 struct md_rdev *rdev;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000515 int choose_first;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000516 int choose_next_idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
518 rcu_read_lock();
519 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700520 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 * device if no resync is going on, or below the resync window.
522 * We take the first readable disk when above the resync window.
523 */
524 retry:
NeilBrownd2eb35a2011-07-28 11:31:48 +1000525 sectors = r1_bio->sectors;
NeilBrown76073052011-05-11 14:34:56 +1000526 best_disk = -1;
Shaohua Li9dedf602012-07-31 10:03:53 +1000527 best_dist_disk = -1;
NeilBrown76073052011-05-11 14:34:56 +1000528 best_dist = MaxSector;
Shaohua Li9dedf602012-07-31 10:03:53 +1000529 best_pending_disk = -1;
530 min_pending = UINT_MAX;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000531 best_good_sectors = 0;
Shaohua Li9dedf602012-07-31 10:03:53 +1000532 has_nonrot_disk = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000533 choose_next_idle = 0;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000534
Goldwyn Rodrigues7d49ffc2014-08-12 10:13:19 -0500535 if ((conf->mddev->recovery_cp < this_sector + sectors) ||
536 (mddev_is_clustered(conf->mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -0500537 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
Goldwyn Rodrigues7d49ffc2014-08-12 10:13:19 -0500538 this_sector + sectors)))
539 choose_first = 1;
540 else
541 choose_first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Shaohua Libe4d3282012-07-31 10:03:53 +1000543 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
NeilBrown76073052011-05-11 14:34:56 +1000544 sector_t dist;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000545 sector_t first_bad;
546 int bad_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000547 unsigned int pending;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000548 bool nonrot;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000549
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000550 rdev = rcu_dereference(conf->mirrors[disk].rdev);
551 if (r1_bio->bios[disk] == IO_BLOCKED
552 || rdev == NULL
NeilBrown76073052011-05-11 14:34:56 +1000553 || test_bit(Faulty, &rdev->flags))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000554 continue;
NeilBrown76073052011-05-11 14:34:56 +1000555 if (!test_bit(In_sync, &rdev->flags) &&
556 rdev->recovery_offset < this_sector + sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 continue;
NeilBrown76073052011-05-11 14:34:56 +1000558 if (test_bit(WriteMostly, &rdev->flags)) {
559 /* Don't balance among write-mostly, just
560 * use the first as a last resort */
Tomáš Hodekd1901ef2015-02-23 11:00:38 +1100561 if (best_dist_disk < 0) {
NeilBrown307729c2012-01-09 01:41:51 +1100562 if (is_badblock(rdev, this_sector, sectors,
563 &first_bad, &bad_sectors)) {
Wei Fang816b0ac2016-03-21 19:18:32 +0800564 if (first_bad <= this_sector)
NeilBrown307729c2012-01-09 01:41:51 +1100565 /* Cannot use this */
566 continue;
567 best_good_sectors = first_bad - this_sector;
568 } else
569 best_good_sectors = sectors;
Tomáš Hodekd1901ef2015-02-23 11:00:38 +1100570 best_dist_disk = disk;
571 best_pending_disk = disk;
NeilBrown307729c2012-01-09 01:41:51 +1100572 }
NeilBrown76073052011-05-11 14:34:56 +1000573 continue;
574 }
575 /* This is a reasonable device to use. It might
576 * even be best.
577 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000578 if (is_badblock(rdev, this_sector, sectors,
579 &first_bad, &bad_sectors)) {
580 if (best_dist < MaxSector)
581 /* already have a better device */
582 continue;
583 if (first_bad <= this_sector) {
584 /* cannot read here. If this is the 'primary'
585 * device, then we must not read beyond
586 * bad_sectors from another device..
587 */
588 bad_sectors -= (this_sector - first_bad);
589 if (choose_first && sectors > bad_sectors)
590 sectors = bad_sectors;
591 if (best_good_sectors > sectors)
592 best_good_sectors = sectors;
593
594 } else {
595 sector_t good_sectors = first_bad - this_sector;
596 if (good_sectors > best_good_sectors) {
597 best_good_sectors = good_sectors;
598 best_disk = disk;
599 }
600 if (choose_first)
601 break;
602 }
603 continue;
604 } else
605 best_good_sectors = sectors;
606
Shaohua Li12cee5a2012-07-31 10:03:53 +1000607 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
608 has_nonrot_disk |= nonrot;
Shaohua Li9dedf602012-07-31 10:03:53 +1000609 pending = atomic_read(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000610 dist = abs(this_sector - conf->mirrors[disk].head_position);
Shaohua Li12cee5a2012-07-31 10:03:53 +1000611 if (choose_first) {
NeilBrown76073052011-05-11 14:34:56 +1000612 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 break;
614 }
Shaohua Li12cee5a2012-07-31 10:03:53 +1000615 /* Don't change to another disk for sequential reads */
616 if (conf->mirrors[disk].next_seq_sect == this_sector
617 || dist == 0) {
618 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
619 struct raid1_info *mirror = &conf->mirrors[disk];
620
621 best_disk = disk;
622 /*
623 * If buffered sequential IO size exceeds optimal
624 * iosize, check if there is idle disk. If yes, choose
625 * the idle disk. read_balance could already choose an
626 * idle disk before noticing it's a sequential IO in
627 * this disk. This doesn't matter because this disk
628 * will idle, next time it will be utilized after the
629 * first disk has IO size exceeds optimal iosize. In
630 * this way, iosize of the first disk will be optimal
631 * iosize at least. iosize of the second disk might be
632 * small, but not a big deal since when the second disk
633 * starts IO, the first disk is likely still busy.
634 */
635 if (nonrot && opt_iosize > 0 &&
636 mirror->seq_start != MaxSector &&
637 mirror->next_seq_sect > opt_iosize &&
638 mirror->next_seq_sect - opt_iosize >=
639 mirror->seq_start) {
640 choose_next_idle = 1;
641 continue;
642 }
643 break;
644 }
645 /* If device is idle, use it */
646 if (pending == 0) {
647 best_disk = disk;
648 break;
649 }
650
651 if (choose_next_idle)
652 continue;
Shaohua Li9dedf602012-07-31 10:03:53 +1000653
654 if (min_pending > pending) {
655 min_pending = pending;
656 best_pending_disk = disk;
657 }
658
NeilBrown76073052011-05-11 14:34:56 +1000659 if (dist < best_dist) {
660 best_dist = dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000661 best_dist_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 }
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Shaohua Li9dedf602012-07-31 10:03:53 +1000665 /*
666 * If all disks are rotational, choose the closest disk. If any disk is
667 * non-rotational, choose the disk with less pending request even the
668 * disk is rotational, which might/might not be optimal for raids with
669 * mixed ratation/non-rotational disks depending on workload.
670 */
671 if (best_disk == -1) {
672 if (has_nonrot_disk)
673 best_disk = best_pending_disk;
674 else
675 best_disk = best_dist_disk;
676 }
677
NeilBrown76073052011-05-11 14:34:56 +1000678 if (best_disk >= 0) {
679 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700680 if (!rdev)
681 goto retry;
682 atomic_inc(&rdev->nr_pending);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000683 sectors = best_good_sectors;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000684
685 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
686 conf->mirrors[best_disk].seq_start = this_sector;
687
Shaohua Libe4d3282012-07-31 10:03:53 +1000688 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 }
690 rcu_read_unlock();
NeilBrownd2eb35a2011-07-28 11:31:48 +1000691 *max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
NeilBrown76073052011-05-11 14:34:56 +1000693 return best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
NeilBrown5c675f82014-12-15 12:56:56 +1100696static int raid1_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700697{
NeilBrowne8096362011-10-11 16:49:05 +1100698 struct r1conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700699 int i, ret = 0;
700
Tejun Heo44522262015-05-22 17:13:26 -0400701 if ((bits & (1 << WB_async_congested)) &&
NeilBrown34db0cd2011-10-11 16:50:01 +1100702 conf->pending_count >= max_queued_requests)
703 return 1;
704
NeilBrown0d129222006-10-03 01:15:54 -0700705 rcu_read_lock();
NeilBrownf53e29f2012-02-13 14:24:05 +1100706 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100707 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700708 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200709 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700710
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500711 BUG_ON(!q);
712
NeilBrown0d129222006-10-03 01:15:54 -0700713 /* Note the '|| 1' - when read_balance prefers
714 * non-congested targets, it can be removed
715 */
Tejun Heo44522262015-05-22 17:13:26 -0400716 if ((bits & (1 << WB_async_congested)) || 1)
NeilBrown0d129222006-10-03 01:15:54 -0700717 ret |= bdi_congested(&q->backing_dev_info, bits);
718 else
719 ret &= bdi_congested(&q->backing_dev_info, bits);
720 }
721 }
722 rcu_read_unlock();
723 return ret;
724}
NeilBrown0d129222006-10-03 01:15:54 -0700725
NeilBrowne8096362011-10-11 16:49:05 +1100726static void flush_pending_writes(struct r1conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800727{
728 /* Any writes that have been queued but are awaiting
729 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800730 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800731 spin_lock_irq(&conf->device_lock);
732
733 if (conf->pending_bio_list.head) {
734 struct bio *bio;
735 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100736 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800737 spin_unlock_irq(&conf->device_lock);
738 /* flush any pending bitmap writes to
739 * disk before proceeding w/ I/O */
740 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100741 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800742
743 while (bio) { /* submit pending writes */
744 struct bio *next = bio->bi_next;
745 bio->bi_next = NULL;
Mike Christie796a5cf2016-06-05 14:32:07 -0500746 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
Shaohua Li2ff8cc22012-10-11 13:28:54 +1100747 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
748 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200749 bio_endio(bio);
Shaohua Li2ff8cc22012-10-11 13:28:54 +1100750 else
751 generic_make_request(bio);
NeilBrowna35e63e2008-03-04 14:29:29 -0800752 bio = next;
753 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800754 } else
755 spin_unlock_irq(&conf->device_lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100756}
757
NeilBrown17999be2006-01-06 00:20:12 -0800758/* Barriers....
759 * Sometimes we need to suspend IO while we do something else,
760 * either some resync/recovery, or reconfigure the array.
761 * To do this we raise a 'barrier'.
762 * The 'barrier' is a counter that can be raised multiple times
763 * to count how many activities are happening which preclude
764 * normal IO.
765 * We can only raise the barrier if there is no pending IO.
766 * i.e. if nr_pending == 0.
767 * We choose only to raise the barrier if no-one is waiting for the
768 * barrier to go down. This means that as soon as an IO request
769 * is ready, no other operations which require a barrier will start
770 * until the IO request has had a chance.
771 *
772 * So: regular IO calls 'wait_barrier'. When that returns there
773 * is no backgroup IO happening, It must arrange to call
774 * allow_barrier when it has finished its IO.
775 * backgroup IO calls must call raise_barrier. Once that returns
776 * there is no normal IO happeing. It must arrange to call
777 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
NeilBrownc2fd4c92014-09-10 16:01:24 +1000779static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
781 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800782
783 /* Wait until no block IO is waiting */
784 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
Lukas Czernereed8c022012-11-30 11:42:40 +0100785 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800786
787 /* block any new IO from starting */
788 conf->barrier++;
NeilBrownc2fd4c92014-09-10 16:01:24 +1000789 conf->next_resync = sector_nr;
NeilBrown17999be2006-01-06 00:20:12 -0800790
majianpeng79ef3a82013-11-15 14:55:02 +0800791 /* For these conditions we must wait:
792 * A: while the array is in frozen state
793 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
794 * the max count which allowed.
795 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
796 * next resync will reach to the window which normal bios are
797 * handling.
NeilBrown2f73d3c2014-09-10 15:01:49 +1000798 * D: while there are any active requests in the current window.
majianpeng79ef3a82013-11-15 14:55:02 +0800799 */
NeilBrown17999be2006-01-06 00:20:12 -0800800 wait_event_lock_irq(conf->wait_barrier,
majianpengb364e3d2013-11-14 15:16:18 +1100801 !conf->array_frozen &&
majianpeng79ef3a82013-11-15 14:55:02 +0800802 conf->barrier < RESYNC_DEPTH &&
NeilBrown2f73d3c2014-09-10 15:01:49 +1000803 conf->current_window_requests == 0 &&
majianpeng79ef3a82013-11-15 14:55:02 +0800804 (conf->start_next_window >=
805 conf->next_resync + RESYNC_SECTORS),
Lukas Czernereed8c022012-11-30 11:42:40 +0100806 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800807
NeilBrown34e97f12014-09-16 12:14:14 +1000808 conf->nr_pending++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 spin_unlock_irq(&conf->resync_lock);
810}
811
NeilBrowne8096362011-10-11 16:49:05 +1100812static void lower_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800813{
814 unsigned long flags;
NeilBrown709ae482009-12-14 12:49:51 +1100815 BUG_ON(conf->barrier <= 0);
NeilBrown17999be2006-01-06 00:20:12 -0800816 spin_lock_irqsave(&conf->resync_lock, flags);
817 conf->barrier--;
NeilBrown34e97f12014-09-16 12:14:14 +1000818 conf->nr_pending--;
NeilBrown17999be2006-01-06 00:20:12 -0800819 spin_unlock_irqrestore(&conf->resync_lock, flags);
820 wake_up(&conf->wait_barrier);
821}
822
majianpeng79ef3a82013-11-15 14:55:02 +0800823static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
NeilBrown17999be2006-01-06 00:20:12 -0800824{
majianpeng79ef3a82013-11-15 14:55:02 +0800825 bool wait = false;
826
827 if (conf->array_frozen || !bio)
828 wait = true;
829 else if (conf->barrier && bio_data_dir(bio) == WRITE) {
NeilBrown23554962014-09-10 15:56:57 +1000830 if ((conf->mddev->curr_resync_completed
831 >= bio_end_sector(bio)) ||
832 (conf->next_resync + NEXT_NORMALIO_DISTANCE
833 <= bio->bi_iter.bi_sector))
majianpeng79ef3a82013-11-15 14:55:02 +0800834 wait = false;
835 else
836 wait = true;
837 }
838
839 return wait;
840}
841
842static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
843{
844 sector_t sector = 0;
845
NeilBrown17999be2006-01-06 00:20:12 -0800846 spin_lock_irq(&conf->resync_lock);
majianpeng79ef3a82013-11-15 14:55:02 +0800847 if (need_to_wait_for_sync(conf, bio)) {
NeilBrown17999be2006-01-06 00:20:12 -0800848 conf->nr_waiting++;
NeilBrownd6b42dc2012-03-19 12:46:38 +1100849 /* Wait for the barrier to drop.
850 * However if there are already pending
851 * requests (preventing the barrier from
852 * rising completely), and the
NeilBrown5965b642014-09-04 15:51:44 +1000853 * per-process bio queue isn't empty,
NeilBrownd6b42dc2012-03-19 12:46:38 +1100854 * then don't wait, as we need to empty
NeilBrown5965b642014-09-04 15:51:44 +1000855 * that queue to allow conf->start_next_window
856 * to increase.
NeilBrownd6b42dc2012-03-19 12:46:38 +1100857 */
858 wait_event_lock_irq(conf->wait_barrier,
majianpengb364e3d2013-11-14 15:16:18 +1100859 !conf->array_frozen &&
860 (!conf->barrier ||
NeilBrown5965b642014-09-04 15:51:44 +1000861 ((conf->start_next_window <
862 conf->next_resync + RESYNC_SECTORS) &&
863 current->bio_list &&
864 !bio_list_empty(current->bio_list))),
Lukas Czernereed8c022012-11-30 11:42:40 +0100865 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800866 conf->nr_waiting--;
867 }
majianpeng79ef3a82013-11-15 14:55:02 +0800868
869 if (bio && bio_data_dir(bio) == WRITE) {
Jes Sorensene8ff8bf2015-09-16 10:20:05 -0400870 if (bio->bi_iter.bi_sector >= conf->next_resync) {
majianpeng79ef3a82013-11-15 14:55:02 +0800871 if (conf->start_next_window == MaxSector)
872 conf->start_next_window =
873 conf->next_resync +
874 NEXT_NORMALIO_DISTANCE;
875
876 if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
Kent Overstreet4f024f32013-10-11 15:44:27 -0700877 <= bio->bi_iter.bi_sector)
majianpeng79ef3a82013-11-15 14:55:02 +0800878 conf->next_window_requests++;
879 else
880 conf->current_window_requests++;
majianpeng79ef3a82013-11-15 14:55:02 +0800881 sector = conf->start_next_window;
NeilBrown41a336e2014-01-14 11:56:14 +1100882 }
majianpeng79ef3a82013-11-15 14:55:02 +0800883 }
884
NeilBrown17999be2006-01-06 00:20:12 -0800885 conf->nr_pending++;
886 spin_unlock_irq(&conf->resync_lock);
majianpeng79ef3a82013-11-15 14:55:02 +0800887 return sector;
NeilBrown17999be2006-01-06 00:20:12 -0800888}
889
majianpeng79ef3a82013-11-15 14:55:02 +0800890static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
891 sector_t bi_sector)
NeilBrown17999be2006-01-06 00:20:12 -0800892{
893 unsigned long flags;
majianpeng79ef3a82013-11-15 14:55:02 +0800894
NeilBrown17999be2006-01-06 00:20:12 -0800895 spin_lock_irqsave(&conf->resync_lock, flags);
896 conf->nr_pending--;
majianpeng79ef3a82013-11-15 14:55:02 +0800897 if (start_next_window) {
898 if (start_next_window == conf->start_next_window) {
899 if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
900 <= bi_sector)
901 conf->next_window_requests--;
902 else
903 conf->current_window_requests--;
904 } else
905 conf->current_window_requests--;
906
907 if (!conf->current_window_requests) {
908 if (conf->next_window_requests) {
909 conf->current_window_requests =
910 conf->next_window_requests;
911 conf->next_window_requests = 0;
912 conf->start_next_window +=
913 NEXT_NORMALIO_DISTANCE;
914 } else
915 conf->start_next_window = MaxSector;
916 }
917 }
NeilBrown17999be2006-01-06 00:20:12 -0800918 spin_unlock_irqrestore(&conf->resync_lock, flags);
919 wake_up(&conf->wait_barrier);
920}
921
NeilBrowne2d59922013-06-12 11:01:22 +1000922static void freeze_array(struct r1conf *conf, int extra)
NeilBrownddaf22a2006-01-06 00:20:19 -0800923{
924 /* stop syncio and normal IO and wait for everything to
925 * go quite.
majianpengb364e3d2013-11-14 15:16:18 +1100926 * We wait until nr_pending match nr_queued+extra
NeilBrown1c830532008-03-04 14:29:35 -0800927 * This is called in the context of one normal IO request
928 * that has failed. Thus any sync request that might be pending
929 * will be blocked by nr_pending, and we need to wait for
930 * pending IO requests to complete or be queued for re-try.
NeilBrowne2d59922013-06-12 11:01:22 +1000931 * Thus the number queued (nr_queued) plus this request (extra)
NeilBrown1c830532008-03-04 14:29:35 -0800932 * must match the number of pending IOs (nr_pending) before
933 * we continue.
NeilBrownddaf22a2006-01-06 00:20:19 -0800934 */
935 spin_lock_irq(&conf->resync_lock);
majianpengb364e3d2013-11-14 15:16:18 +1100936 conf->array_frozen = 1;
Lukas Czernereed8c022012-11-30 11:42:40 +0100937 wait_event_lock_irq_cmd(conf->wait_barrier,
NeilBrowne2d59922013-06-12 11:01:22 +1000938 conf->nr_pending == conf->nr_queued+extra,
Lukas Czernereed8c022012-11-30 11:42:40 +0100939 conf->resync_lock,
940 flush_pending_writes(conf));
NeilBrownddaf22a2006-01-06 00:20:19 -0800941 spin_unlock_irq(&conf->resync_lock);
942}
NeilBrowne8096362011-10-11 16:49:05 +1100943static void unfreeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800944{
945 /* reverse the effect of the freeze */
946 spin_lock_irq(&conf->resync_lock);
majianpengb364e3d2013-11-14 15:16:18 +1100947 conf->array_frozen = 0;
NeilBrownddaf22a2006-01-06 00:20:19 -0800948 wake_up(&conf->wait_barrier);
949 spin_unlock_irq(&conf->resync_lock);
950}
951
NeilBrownf72ffdd2014-09-30 14:23:59 +1000952/* duplicate the data pages for behind I/O
NeilBrown4e780642010-10-19 12:54:01 +1100953 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100954static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
NeilBrown4b6d2872005-09-09 16:23:47 -0700955{
956 int i;
957 struct bio_vec *bvec;
NeilBrown2ca68f52011-07-28 11:32:10 +1000958 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
NeilBrown4b6d2872005-09-09 16:23:47 -0700959 GFP_NOIO);
NeilBrown2ca68f52011-07-28 11:32:10 +1000960 if (unlikely(!bvecs))
NeilBrownaf6d7b72011-05-11 14:51:19 +1000961 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700962
Kent Overstreetcb34e052012-09-05 15:22:02 -0700963 bio_for_each_segment_all(bvec, bio, i) {
NeilBrown2ca68f52011-07-28 11:32:10 +1000964 bvecs[i] = *bvec;
965 bvecs[i].bv_page = alloc_page(GFP_NOIO);
966 if (unlikely(!bvecs[i].bv_page))
NeilBrown4b6d2872005-09-09 16:23:47 -0700967 goto do_sync_io;
NeilBrown2ca68f52011-07-28 11:32:10 +1000968 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
969 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
970 kunmap(bvecs[i].bv_page);
NeilBrown4b6d2872005-09-09 16:23:47 -0700971 kunmap(bvec->bv_page);
972 }
NeilBrown2ca68f52011-07-28 11:32:10 +1000973 r1_bio->behind_bvecs = bvecs;
NeilBrownaf6d7b72011-05-11 14:51:19 +1000974 r1_bio->behind_page_count = bio->bi_vcnt;
975 set_bit(R1BIO_BehindIO, &r1_bio->state);
976 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700977
978do_sync_io:
NeilBrownaf6d7b72011-05-11 14:51:19 +1000979 for (i = 0; i < bio->bi_vcnt; i++)
NeilBrown2ca68f52011-07-28 11:32:10 +1000980 if (bvecs[i].bv_page)
981 put_page(bvecs[i].bv_page);
982 kfree(bvecs);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700983 pr_debug("%dB behind alloc failed, doing sync I/O\n",
984 bio->bi_iter.bi_size);
NeilBrown4b6d2872005-09-09 16:23:47 -0700985}
986
NeilBrownf54a9d02012-08-02 08:33:20 +1000987struct raid1_plug_cb {
988 struct blk_plug_cb cb;
989 struct bio_list pending;
990 int pending_cnt;
991};
992
993static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
994{
995 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
996 cb);
997 struct mddev *mddev = plug->cb.data;
998 struct r1conf *conf = mddev->private;
999 struct bio *bio;
1000
NeilBrown874807a2012-11-27 12:14:40 +11001001 if (from_schedule || current->bio_list) {
NeilBrownf54a9d02012-08-02 08:33:20 +10001002 spin_lock_irq(&conf->device_lock);
1003 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1004 conf->pending_count += plug->pending_cnt;
1005 spin_unlock_irq(&conf->device_lock);
NeilBrownee0b0242013-02-25 12:38:29 +11001006 wake_up(&conf->wait_barrier);
NeilBrownf54a9d02012-08-02 08:33:20 +10001007 md_wakeup_thread(mddev->thread);
1008 kfree(plug);
1009 return;
1010 }
1011
1012 /* we aren't scheduling, so we can do the write-out directly. */
1013 bio = bio_list_get(&plug->pending);
1014 bitmap_unplug(mddev->bitmap);
1015 wake_up(&conf->wait_barrier);
1016
1017 while (bio) { /* submit pending writes */
1018 struct bio *next = bio->bi_next;
1019 bio->bi_next = NULL;
Mike Christie796a5cf2016-06-05 14:32:07 -05001020 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
Shaohua Li32f9f572013-04-28 18:26:38 +08001021 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1022 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001023 bio_endio(bio);
Shaohua Li32f9f572013-04-28 18:26:38 +08001024 else
1025 generic_make_request(bio);
NeilBrownf54a9d02012-08-02 08:33:20 +10001026 bio = next;
1027 }
1028 kfree(plug);
1029}
1030
Shaohua Li849674e2016-01-20 13:52:20 -08001031static void raid1_make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032{
NeilBrowne8096362011-10-11 16:49:05 +11001033 struct r1conf *conf = mddev->private;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001034 struct raid1_info *mirror;
NeilBrown9f2c9d12011-10-11 16:48:43 +11001035 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 struct bio *read_bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001037 int i, disks;
NeilBrown84255d12008-05-23 13:04:32 -07001038 struct bitmap *bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -07001039 unsigned long flags;
Mike Christie796a5cf2016-06-05 14:32:07 -05001040 const int op = bio_op(bio);
Jens Axboea3623572005-11-01 09:26:16 +01001041 const int rw = bio_data_dir(bio);
Jens Axboe1eff9d32016-08-05 15:35:16 -06001042 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1043 const unsigned long do_flush_fua = (bio->bi_opf &
Mike Christie28a8f0d2016-06-05 14:32:25 -05001044 (REQ_PREFLUSH | REQ_FUA));
NeilBrown3cb03002011-10-11 16:45:26 +11001045 struct md_rdev *blocked_rdev;
NeilBrownf54a9d02012-08-02 08:33:20 +10001046 struct blk_plug_cb *cb;
1047 struct raid1_plug_cb *plug = NULL;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001048 int first_clone;
1049 int sectors_handled;
1050 int max_sectors;
majianpeng79ef3a82013-11-15 14:55:02 +08001051 sector_t start_next_window;
NeilBrown191ea9b2005-06-21 17:17:23 -07001052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 /*
1054 * Register the new request and wait if the reconstruction
1055 * thread has put up a bar for new requests.
1056 * Continue immediately if no resync is active currently.
1057 */
NeilBrown62de6082006-05-01 12:15:47 -07001058
NeilBrown3d310eb2005-06-21 17:17:26 -07001059 md_write_start(mddev, bio); /* wait on superblock update early */
1060
NeilBrown6eef4b22009-12-14 12:49:51 +11001061 if (bio_data_dir(bio) == WRITE &&
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001062 ((bio_end_sector(bio) > mddev->suspend_lo &&
1063 bio->bi_iter.bi_sector < mddev->suspend_hi) ||
1064 (mddev_is_clustered(mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -05001065 md_cluster_ops->area_resyncing(mddev, WRITE,
1066 bio->bi_iter.bi_sector, bio_end_sector(bio))))) {
NeilBrown6eef4b22009-12-14 12:49:51 +11001067 /* As the suspend_* range is controlled by
1068 * userspace, we want an interruptible
1069 * wait.
1070 */
1071 DEFINE_WAIT(w);
1072 for (;;) {
1073 flush_signals(current);
1074 prepare_to_wait(&conf->wait_barrier,
1075 &w, TASK_INTERRUPTIBLE);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001076 if (bio_end_sector(bio) <= mddev->suspend_lo ||
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001077 bio->bi_iter.bi_sector >= mddev->suspend_hi ||
1078 (mddev_is_clustered(mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -05001079 !md_cluster_ops->area_resyncing(mddev, WRITE,
Goldwyn Rodrigues589a1c42014-06-07 02:39:37 -05001080 bio->bi_iter.bi_sector, bio_end_sector(bio))))
NeilBrown6eef4b22009-12-14 12:49:51 +11001081 break;
1082 schedule();
1083 }
1084 finish_wait(&conf->wait_barrier, &w);
1085 }
NeilBrown62de6082006-05-01 12:15:47 -07001086
majianpeng79ef3a82013-11-15 14:55:02 +08001087 start_next_window = wait_barrier(conf, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
NeilBrown84255d12008-05-23 13:04:32 -07001089 bitmap = mddev->bitmap;
1090
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 /*
Christoph Hellwig70246282016-07-19 11:28:41 +02001092 * make_request() can abort the operation when read-ahead is being
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 * used and no empty request is available.
1094 *
1095 */
1096 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1097
1098 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001099 r1_bio->sectors = bio_sectors(bio);
NeilBrown191ea9b2005-06-21 17:17:23 -07001100 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001102 r1_bio->sector = bio->bi_iter.bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
NeilBrownd2eb35a2011-07-28 11:31:48 +10001104 /* We might need to issue multiple reads to different
1105 * devices if there are bad blocks around, so we keep
1106 * track of the number of reads in bio->bi_phys_segments.
1107 * If this is 0, there is only one r1_bio and no locking
1108 * will be needed when requests complete. If it is
1109 * non-zero, then it is the number of not-completed requests.
1110 */
1111 bio->bi_phys_segments = 0;
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001112 bio_clear_flag(bio, BIO_SEG_VALID);
NeilBrownd2eb35a2011-07-28 11:31:48 +10001113
Jens Axboea3623572005-11-01 09:26:16 +01001114 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 /*
1116 * read balancing logic:
1117 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001118 int rdisk;
1119
1120read_again:
1121 rdisk = read_balance(conf, r1_bio, &max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
1123 if (rdisk < 0) {
1124 /* couldn't find anywhere to read from */
1125 raid_end_bio_io(r1_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001126 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
1128 mirror = conf->mirrors + rdisk;
1129
NeilBrowne5551902010-03-31 11:21:44 +11001130 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1131 bitmap) {
1132 /* Reading from a write-mostly device must
1133 * take care not to over-take any writes
1134 * that are 'behind'
1135 */
1136 wait_event(bitmap->behind_wait,
1137 atomic_read(&bitmap->behind_writes) == 0);
1138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 r1_bio->read_disk = rdisk;
NeilBrownf0cc9a02014-09-22 10:06:23 +10001140 r1_bio->start_next_window = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
NeilBrowna167f662010-10-26 18:31:13 +11001142 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001143 bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
Kent Overstreet6678d832013-08-07 11:14:32 -07001144 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 r1_bio->bios[rdisk] = read_bio;
1147
Kent Overstreet4f024f32013-10-11 15:44:27 -07001148 read_bio->bi_iter.bi_sector = r1_bio->sector +
1149 mirror->rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 read_bio->bi_bdev = mirror->rdev->bdev;
1151 read_bio->bi_end_io = raid1_end_read_request;
Mike Christie796a5cf2016-06-05 14:32:07 -05001152 bio_set_op_attrs(read_bio, op, do_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 read_bio->bi_private = r1_bio;
1154
NeilBrownd2eb35a2011-07-28 11:31:48 +10001155 if (max_sectors < r1_bio->sectors) {
1156 /* could not read all from this device, so we will
1157 * need another r1_bio.
1158 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001159
1160 sectors_handled = (r1_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07001161 - bio->bi_iter.bi_sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +10001162 r1_bio->sectors = max_sectors;
1163 spin_lock_irq(&conf->device_lock);
1164 if (bio->bi_phys_segments == 0)
1165 bio->bi_phys_segments = 2;
1166 else
1167 bio->bi_phys_segments++;
1168 spin_unlock_irq(&conf->device_lock);
1169 /* Cannot call generic_make_request directly
1170 * as that will be queued in __make_request
1171 * and subsequent mempool_alloc might block waiting
1172 * for it. So hand bio over to raid1d.
1173 */
1174 reschedule_retry(r1_bio);
1175
1176 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1177
1178 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001179 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrownd2eb35a2011-07-28 11:31:48 +10001180 r1_bio->state = 0;
1181 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001182 r1_bio->sector = bio->bi_iter.bi_sector +
1183 sectors_handled;
NeilBrownd2eb35a2011-07-28 11:31:48 +10001184 goto read_again;
1185 } else
1186 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001187 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 }
1189
1190 /*
1191 * WRITE:
1192 */
NeilBrown34db0cd2011-10-11 16:50:01 +11001193 if (conf->pending_count >= max_queued_requests) {
1194 md_wakeup_thread(mddev->thread);
1195 wait_event(conf->wait_barrier,
1196 conf->pending_count < max_queued_requests);
1197 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001198 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 * inc refcount on their rdev. Record them by setting
1200 * bios[x] to bio
NeilBrown1f68f0c2011-07-28 11:31:48 +10001201 * If there are known/acknowledged bad blocks on any device on
1202 * which we have seen a write error, we want to avoid writing those
1203 * blocks.
1204 * This potentially requires several writes to write around
1205 * the bad blocks. Each set of writes gets it's own r1bio
1206 * with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001208
NeilBrown8f19ccb2011-12-23 10:17:56 +11001209 disks = conf->raid_disks * 2;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001210 retry_write:
majianpeng79ef3a82013-11-15 14:55:02 +08001211 r1_bio->start_next_window = start_next_window;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001212 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 rcu_read_lock();
NeilBrown1f68f0c2011-07-28 11:31:48 +10001214 max_sectors = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001216 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001217 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1218 atomic_inc(&rdev->nr_pending);
1219 blocked_rdev = rdev;
1220 break;
1221 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001222 r1_bio->bios[i] = NULL;
Kent Overstreet8ae12662015-04-27 23:48:34 -07001223 if (!rdev || test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11001224 if (i < conf->raid_disks)
1225 set_bit(R1BIO_Degraded, &r1_bio->state);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001226 continue;
1227 }
1228
1229 atomic_inc(&rdev->nr_pending);
1230 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1231 sector_t first_bad;
1232 int bad_sectors;
1233 int is_bad;
1234
1235 is_bad = is_badblock(rdev, r1_bio->sector,
1236 max_sectors,
1237 &first_bad, &bad_sectors);
1238 if (is_bad < 0) {
1239 /* mustn't write here until the bad block is
1240 * acknowledged*/
1241 set_bit(BlockedBadBlocks, &rdev->flags);
1242 blocked_rdev = rdev;
1243 break;
NeilBrown964147d2010-05-18 15:27:13 +10001244 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001245 if (is_bad && first_bad <= r1_bio->sector) {
1246 /* Cannot write here at all */
1247 bad_sectors -= (r1_bio->sector - first_bad);
1248 if (bad_sectors < max_sectors)
1249 /* mustn't write more than bad_sectors
1250 * to other devices yet
1251 */
1252 max_sectors = bad_sectors;
1253 rdev_dec_pending(rdev, mddev);
1254 /* We don't set R1BIO_Degraded as that
1255 * only applies if the disk is
1256 * missing, so it might be re-added,
1257 * and we want to know to recover this
1258 * chunk.
1259 * In this case the device is here,
1260 * and the fact that this chunk is not
1261 * in-sync is recorded in the bad
1262 * block log
1263 */
1264 continue;
1265 }
1266 if (is_bad) {
1267 int good_sectors = first_bad - r1_bio->sector;
1268 if (good_sectors < max_sectors)
1269 max_sectors = good_sectors;
1270 }
1271 }
1272 r1_bio->bios[i] = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
1274 rcu_read_unlock();
1275
Dan Williams6bfe0b42008-04-30 00:52:32 -07001276 if (unlikely(blocked_rdev)) {
1277 /* Wait for this device to become unblocked */
1278 int j;
majianpeng79ef3a82013-11-15 14:55:02 +08001279 sector_t old = start_next_window;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001280
1281 for (j = 0; j < i; j++)
1282 if (r1_bio->bios[j])
1283 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001284 r1_bio->state = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001285 allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001286 md_wait_for_blocked_rdev(blocked_rdev, mddev);
majianpeng79ef3a82013-11-15 14:55:02 +08001287 start_next_window = wait_barrier(conf, bio);
1288 /*
1289 * We must make sure the multi r1bios of bio have
1290 * the same value of bi_phys_segments
1291 */
1292 if (bio->bi_phys_segments && old &&
1293 old != start_next_window)
1294 /* Wait for the former r1bio(s) to complete */
1295 wait_event(conf->wait_barrier,
1296 bio->bi_phys_segments == 1);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001297 goto retry_write;
1298 }
1299
NeilBrown1f68f0c2011-07-28 11:31:48 +10001300 if (max_sectors < r1_bio->sectors) {
1301 /* We are splitting this write into multiple parts, so
1302 * we need to prepare for allocating another r1_bio.
1303 */
1304 r1_bio->sectors = max_sectors;
1305 spin_lock_irq(&conf->device_lock);
1306 if (bio->bi_phys_segments == 0)
1307 bio->bi_phys_segments = 2;
1308 else
1309 bio->bi_phys_segments++;
1310 spin_unlock_irq(&conf->device_lock);
NeilBrown191ea9b2005-06-21 17:17:23 -07001311 }
Kent Overstreet4f024f32013-10-11 15:44:27 -07001312 sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
NeilBrown4b6d2872005-09-09 16:23:47 -07001313
NeilBrown4e780642010-10-19 12:54:01 +11001314 atomic_set(&r1_bio->remaining, 1);
NeilBrown4b6d2872005-09-09 16:23:47 -07001315 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -07001316
NeilBrown1f68f0c2011-07-28 11:31:48 +10001317 first_clone = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 for (i = 0; i < disks; i++) {
1319 struct bio *mbio;
1320 if (!r1_bio->bios[i])
1321 continue;
1322
NeilBrowna167f662010-10-26 18:31:13 +11001323 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001324 bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
NeilBrown1f68f0c2011-07-28 11:31:48 +10001326 if (first_clone) {
1327 /* do behind I/O ?
1328 * Not if there are too many, or cannot
1329 * allocate memory, or a reader on WriteMostly
1330 * is waiting for behind writes to flush */
1331 if (bitmap &&
1332 (atomic_read(&bitmap->behind_writes)
1333 < mddev->bitmap_info.max_write_behind) &&
1334 !waitqueue_active(&bitmap->behind_wait))
1335 alloc_behind_pages(mbio, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
NeilBrown1f68f0c2011-07-28 11:31:48 +10001337 bitmap_startwrite(bitmap, r1_bio->sector,
1338 r1_bio->sectors,
1339 test_bit(R1BIO_BehindIO,
1340 &r1_bio->state));
1341 first_clone = 0;
1342 }
NeilBrown2ca68f52011-07-28 11:32:10 +10001343 if (r1_bio->behind_bvecs) {
NeilBrown4b6d2872005-09-09 16:23:47 -07001344 struct bio_vec *bvec;
1345 int j;
1346
Kent Overstreetcb34e052012-09-05 15:22:02 -07001347 /*
1348 * We trimmed the bio, so _all is legit
NeilBrown4b6d2872005-09-09 16:23:47 -07001349 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001350 bio_for_each_segment_all(bvec, mbio, j)
NeilBrown2ca68f52011-07-28 11:32:10 +10001351 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
NeilBrown4b6d2872005-09-09 16:23:47 -07001352 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1353 atomic_inc(&r1_bio->behind_remaining);
1354 }
1355
NeilBrown1f68f0c2011-07-28 11:31:48 +10001356 r1_bio->bios[i] = mbio;
1357
Kent Overstreet4f024f32013-10-11 15:44:27 -07001358 mbio->bi_iter.bi_sector = (r1_bio->sector +
NeilBrown1f68f0c2011-07-28 11:31:48 +10001359 conf->mirrors[i].rdev->data_offset);
1360 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1361 mbio->bi_end_io = raid1_end_write_request;
Christoph Hellwig288dab82016-06-09 16:00:36 +02001362 bio_set_op_attrs(mbio, op, do_flush_fua | do_sync);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001363 mbio->bi_private = r1_bio;
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 atomic_inc(&r1_bio->remaining);
NeilBrownf54a9d02012-08-02 08:33:20 +10001366
1367 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1368 if (cb)
1369 plug = container_of(cb, struct raid1_plug_cb, cb);
1370 else
1371 plug = NULL;
NeilBrown4e780642010-10-19 12:54:01 +11001372 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001373 if (plug) {
1374 bio_list_add(&plug->pending, mbio);
1375 plug->pending_cnt++;
1376 } else {
1377 bio_list_add(&conf->pending_bio_list, mbio);
1378 conf->pending_count++;
1379 }
NeilBrown4e780642010-10-19 12:54:01 +11001380 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001381 if (!plug)
NeilBrownb357f042012-07-03 17:45:31 +10001382 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 }
NeilBrown079fa162011-09-10 17:21:23 +10001384 /* Mustn't call r1_bio_write_done before this next test,
1385 * as it could result in the bio being freed.
1386 */
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001387 if (sectors_handled < bio_sectors(bio)) {
NeilBrown079fa162011-09-10 17:21:23 +10001388 r1_bio_write_done(r1_bio);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001389 /* We need another r1_bio. It has already been counted
1390 * in bio->bi_phys_segments
1391 */
1392 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1393 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001394 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001395 r1_bio->state = 0;
1396 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001397 r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001398 goto retry_write;
1399 }
1400
NeilBrown079fa162011-09-10 17:21:23 +10001401 r1_bio_write_done(r1_bio);
1402
1403 /* In case raid1d snuck in to freeze_array */
1404 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405}
1406
Shaohua Li849674e2016-01-20 13:52:20 -08001407static void raid1_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
NeilBrowne8096362011-10-11 16:49:05 +11001409 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 int i;
1411
1412 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001413 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001414 rcu_read_lock();
1415 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001416 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001418 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1419 }
1420 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 seq_printf(seq, "]");
1422}
1423
Shaohua Li849674e2016-01-20 13:52:20 -08001424static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 char b[BDEVNAME_SIZE];
NeilBrowne8096362011-10-11 16:49:05 +11001427 struct r1conf *conf = mddev->private;
NeilBrown423f04d2015-07-27 11:48:52 +10001428 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 /*
1431 * If it is not operational, then we have already marked it as dead
1432 * else if it is the last working disks, ignore the error, let the
1433 * next level up know.
1434 * else mark the drive as failed
1435 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001436 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001437 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 /*
1439 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001440 * normal single drive.
1441 * However don't try a recovery from this drive as
1442 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 */
NeilBrown53890422011-07-27 11:00:36 +10001444 conf->recovery_disabled = mddev->recovery_disabled;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001446 }
NeilBrownde393cd2011-07-28 11:31:48 +10001447 set_bit(Blocked, &rdev->flags);
NeilBrown423f04d2015-07-27 11:48:52 +10001448 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001449 if (test_and_clear_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -07001451 set_bit(Faulty, &rdev->flags);
NeilBrowndd00a992007-05-10 03:15:50 -07001452 } else
1453 set_bit(Faulty, &rdev->flags);
NeilBrown423f04d2015-07-27 11:48:52 +10001454 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrown2446dba2014-07-31 10:16:29 +10001455 /*
1456 * if recovery is running, make sure it aborts.
1457 */
1458 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Guoqing Jiang85ad1d12016-05-03 22:22:13 -04001459 set_mask_bits(&mddev->flags, 0,
1460 BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING));
NeilBrown1d41c212016-11-02 14:16:50 +11001461 pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n"
1462 "md/raid1:%s: Operation continuing on %d devices.\n",
1463 mdname(mddev), bdevname(rdev->bdev, b),
1464 mdname(mddev), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465}
1466
NeilBrowne8096362011-10-11 16:49:05 +11001467static void print_conf(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
1469 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
NeilBrown1d41c212016-11-02 14:16:50 +11001471 pr_debug("RAID1 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 if (!conf) {
NeilBrown1d41c212016-11-02 14:16:50 +11001473 pr_debug("(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 return;
1475 }
NeilBrown1d41c212016-11-02 14:16:50 +11001476 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1477 conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
NeilBrownddac7c72006-08-31 21:27:36 -07001479 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 for (i = 0; i < conf->raid_disks; i++) {
1481 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001482 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownddac7c72006-08-31 21:27:36 -07001483 if (rdev)
NeilBrown1d41c212016-11-02 14:16:50 +11001484 pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n",
1485 i, !test_bit(In_sync, &rdev->flags),
1486 !test_bit(Faulty, &rdev->flags),
1487 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 }
NeilBrownddac7c72006-08-31 21:27:36 -07001489 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490}
1491
NeilBrowne8096362011-10-11 16:49:05 +11001492static void close_sync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493{
majianpeng79ef3a82013-11-15 14:55:02 +08001494 wait_barrier(conf, NULL);
1495 allow_barrier(conf, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
1497 mempool_destroy(conf->r1buf_pool);
1498 conf->r1buf_pool = NULL;
majianpeng79ef3a82013-11-15 14:55:02 +08001499
NeilBrown669cc7b2014-09-04 16:30:38 +10001500 spin_lock_irq(&conf->resync_lock);
Jes Sorensene8ff8bf2015-09-16 10:20:05 -04001501 conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE;
majianpeng79ef3a82013-11-15 14:55:02 +08001502 conf->start_next_window = MaxSector;
NeilBrown669cc7b2014-09-04 16:30:38 +10001503 conf->current_window_requests +=
1504 conf->next_window_requests;
1505 conf->next_window_requests = 0;
1506 spin_unlock_irq(&conf->resync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
1508
NeilBrownfd01b882011-10-11 16:47:53 +11001509static int raid1_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510{
1511 int i;
NeilBrowne8096362011-10-11 16:49:05 +11001512 struct r1conf *conf = mddev->private;
NeilBrown6b965622010-08-18 11:56:59 +10001513 int count = 0;
1514 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 /*
NeilBrownf72ffdd2014-09-30 14:23:59 +10001517 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001518 * and mark them readable.
1519 * Called under mddev lock, so rcu protection not needed.
NeilBrown423f04d2015-07-27 11:48:52 +10001520 * device_lock used to avoid races with raid1_end_read_request
1521 * which expects 'In_sync' flags and ->degraded to be consistent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 */
NeilBrown423f04d2015-07-27 11:48:52 +10001523 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001525 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001526 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1527 if (repl
Goldwyn Rodrigues1aee41f2014-10-29 18:51:31 -05001528 && !test_bit(Candidate, &repl->flags)
NeilBrown8c7a2c22011-12-23 10:17:57 +11001529 && repl->recovery_offset == MaxSector
1530 && !test_bit(Faulty, &repl->flags)
1531 && !test_and_set_bit(In_sync, &repl->flags)) {
1532 /* replacement has just become active */
1533 if (!rdev ||
1534 !test_and_clear_bit(In_sync, &rdev->flags))
1535 count++;
1536 if (rdev) {
1537 /* Replaced device not technically
1538 * faulty, but we need to be sure
1539 * it gets removed and never re-added
1540 */
1541 set_bit(Faulty, &rdev->flags);
1542 sysfs_notify_dirent_safe(
1543 rdev->sysfs_state);
1544 }
1545 }
NeilBrownddac7c72006-08-31 21:27:36 -07001546 if (rdev
Lukasz Dorau61e49472013-10-24 12:55:17 +11001547 && rdev->recovery_offset == MaxSector
NeilBrownddac7c72006-08-31 21:27:36 -07001548 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001549 && !test_and_set_bit(In_sync, &rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001550 count++;
Jonathan Brassow654e8b52011-07-27 11:00:36 +10001551 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 }
1553 }
NeilBrown6b965622010-08-18 11:56:59 +10001554 mddev->degraded -= count;
1555 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
1557 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001558 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
1560
NeilBrownfd01b882011-10-11 16:47:53 +11001561static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562{
NeilBrowne8096362011-10-11 16:49:05 +11001563 struct r1conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001564 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001565 int mirror = 0;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001566 struct raid1_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001567 int first = 0;
NeilBrown30194632011-12-23 10:17:56 +11001568 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
NeilBrown53890422011-07-27 11:00:36 +10001570 if (mddev->recovery_disabled == conf->recovery_disabled)
1571 return -EBUSY;
1572
Dan Williams1501efa2016-01-13 16:00:07 -08001573 if (md_integrity_add_rdev(rdev, mddev))
1574 return -ENXIO;
1575
Neil Brown6c2fce22008-06-28 08:31:31 +10001576 if (rdev->raid_disk >= 0)
1577 first = last = rdev->raid_disk;
1578
Goldwyn Rodrigues70bcecd2015-08-21 10:33:39 -05001579 /*
1580 * find the disk ... but prefer rdev->saved_raid_disk
1581 * if possible.
1582 */
1583 if (rdev->saved_raid_disk >= 0 &&
1584 rdev->saved_raid_disk >= first &&
1585 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1586 first = last = rdev->saved_raid_disk;
1587
NeilBrown7ef449d2011-12-23 10:17:57 +11001588 for (mirror = first; mirror <= last; mirror++) {
1589 p = conf->mirrors+mirror;
1590 if (!p->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
Jonathan Brassow9092c022013-05-02 14:19:24 -05001592 if (mddev->gendisk)
1593 disk_stack_limits(mddev->gendisk, rdev->bdev,
1594 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
1596 p->head_position = 0;
1597 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001598 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001599 /* As all devices are equivalent, we don't need a full recovery
1600 * if this was recently any drive of the array
1601 */
1602 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001603 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001604 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 break;
1606 }
NeilBrown7ef449d2011-12-23 10:17:57 +11001607 if (test_bit(WantReplacement, &p->rdev->flags) &&
1608 p[conf->raid_disks].rdev == NULL) {
1609 /* Add this device as a replacement */
1610 clear_bit(In_sync, &rdev->flags);
1611 set_bit(Replacement, &rdev->flags);
1612 rdev->raid_disk = mirror;
1613 err = 0;
1614 conf->fullsync = 1;
1615 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1616 break;
1617 }
1618 }
Jonathan Brassow9092c022013-05-02 14:19:24 -05001619 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
Shaohua Li2ff8cc22012-10-11 13:28:54 +11001620 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001622 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623}
1624
NeilBrownb8321b62011-12-23 10:17:51 +11001625static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
NeilBrowne8096362011-10-11 16:49:05 +11001627 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001629 int number = rdev->raid_disk;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001630 struct raid1_info *p = conf->mirrors + number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
NeilBrownb014f142011-12-23 10:17:56 +11001632 if (rdev != p->rdev)
1633 p = conf->mirrors + conf->raid_disks + number;
1634
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 print_conf(conf);
NeilBrownb8321b62011-12-23 10:17:51 +11001636 if (rdev == p->rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001637 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 atomic_read(&rdev->nr_pending)) {
1639 err = -EBUSY;
1640 goto abort;
1641 }
NeilBrown046abee2010-10-26 15:46:20 +11001642 /* Only remove non-faulty devices if recovery
NeilBrowndfc70642008-05-23 13:04:39 -07001643 * is not possible.
1644 */
1645 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown53890422011-07-27 11:00:36 +10001646 mddev->recovery_disabled != conf->recovery_disabled &&
NeilBrowndfc70642008-05-23 13:04:39 -07001647 mddev->degraded < conf->raid_disks) {
1648 err = -EBUSY;
1649 goto abort;
1650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 p->rdev = NULL;
NeilBrownd787be42016-06-02 16:19:53 +10001652 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
1653 synchronize_rcu();
1654 if (atomic_read(&rdev->nr_pending)) {
1655 /* lost the race, try later */
1656 err = -EBUSY;
1657 p->rdev = rdev;
1658 goto abort;
1659 }
1660 }
1661 if (conf->mirrors[conf->raid_disks + number].rdev) {
NeilBrown8c7a2c22011-12-23 10:17:57 +11001662 /* We just removed a device that is being replaced.
1663 * Move down the replacement. We drain all IO before
1664 * doing this to avoid confusion.
1665 */
1666 struct md_rdev *repl =
1667 conf->mirrors[conf->raid_disks + number].rdev;
NeilBrowne2d59922013-06-12 11:01:22 +10001668 freeze_array(conf, 0);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001669 clear_bit(Replacement, &repl->flags);
1670 p->rdev = repl;
1671 conf->mirrors[conf->raid_disks + number].rdev = NULL;
NeilBrowne2d59922013-06-12 11:01:22 +10001672 unfreeze_array(conf);
NeilBrownb014f142011-12-23 10:17:56 +11001673 clear_bit(WantReplacement, &rdev->flags);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001674 } else
1675 clear_bit(WantReplacement, &rdev->flags);
Martin K. Petersena91a2782011-03-17 11:11:05 +01001676 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 }
1678abort:
1679
1680 print_conf(conf);
1681 return err;
1682}
1683
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001684static void end_sync_read(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001686 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
NeilBrown0fc280f2011-10-07 14:22:55 +11001688 update_head_pos(r1_bio->read_disk, r1_bio);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001689
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 /*
1691 * we have read a block, now it needs to be re-written,
1692 * or re-read if the read failed.
1693 * We don't do much here, just schedule handling by raid1d
1694 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001695 if (!bio->bi_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001697
1698 if (atomic_dec_and_test(&r1_bio->remaining))
1699 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700}
1701
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001702static void end_sync_write(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001704 int uptodate = !bio->bi_error;
NeilBrown9f2c9d12011-10-11 16:48:43 +11001705 struct r1bio *r1_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11001706 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001707 struct r1conf *conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10001708 sector_t first_bad;
1709 int bad_sectors;
NeilBrown854abd72016-06-02 16:19:52 +10001710 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001711
NeilBrown6b1117d2006-03-31 02:31:57 -08001712 if (!uptodate) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001713 sector_t sync_blocks = 0;
NeilBrown6b1117d2006-03-31 02:31:57 -08001714 sector_t s = r1_bio->sector;
1715 long sectors_to_go = r1_bio->sectors;
1716 /* make sure these bits doesn't get cleared. */
1717 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001718 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001719 &sync_blocks, 1);
1720 s += sync_blocks;
1721 sectors_to_go -= sync_blocks;
1722 } while (sectors_to_go > 0);
NeilBrown854abd72016-06-02 16:19:52 +10001723 set_bit(WriteErrorSeen, &rdev->flags);
1724 if (!test_and_set_bit(WantReplacement, &rdev->flags))
NeilBrown19d67162011-12-23 10:17:57 +11001725 set_bit(MD_RECOVERY_NEEDED, &
1726 mddev->recovery);
NeilBrownd8f05d22011-07-28 11:33:00 +10001727 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown854abd72016-06-02 16:19:52 +10001728 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
NeilBrown3a9f28a2011-07-28 11:33:42 +10001729 &first_bad, &bad_sectors) &&
1730 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1731 r1_bio->sector,
1732 r1_bio->sectors,
1733 &first_bad, &bad_sectors)
1734 )
NeilBrown4367af52011-07-28 11:31:49 +10001735 set_bit(R1BIO_MadeGood, &r1_bio->state);
NeilBrowne3b97032005-08-04 12:53:34 -07001736
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown4367af52011-07-28 11:31:49 +10001738 int s = r1_bio->sectors;
NeilBrownd8f05d22011-07-28 11:33:00 +10001739 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1740 test_bit(R1BIO_WriteError, &r1_bio->state))
NeilBrown4367af52011-07-28 11:31:49 +10001741 reschedule_retry(r1_bio);
1742 else {
1743 put_buf(r1_bio);
1744 md_done_sync(mddev, s, uptodate);
1745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747}
1748
NeilBrown3cb03002011-10-11 16:45:26 +11001749static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrownd8f05d22011-07-28 11:33:00 +10001750 int sectors, struct page *page, int rw)
1751{
Mike Christie796a5cf2016-06-05 14:32:07 -05001752 if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
NeilBrownd8f05d22011-07-28 11:33:00 +10001753 /* success */
1754 return 1;
NeilBrown19d67162011-12-23 10:17:57 +11001755 if (rw == WRITE) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001756 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001757 if (!test_and_set_bit(WantReplacement,
1758 &rdev->flags))
1759 set_bit(MD_RECOVERY_NEEDED, &
1760 rdev->mddev->recovery);
1761 }
NeilBrownd8f05d22011-07-28 11:33:00 +10001762 /* need to record an error - either for the block or the device */
1763 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1764 md_error(rdev->mddev, rdev);
1765 return 0;
1766}
1767
NeilBrown9f2c9d12011-10-11 16:48:43 +11001768static int fix_sync_read_error(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001769{
1770 /* Try some synchronous reads of other devices to get
1771 * good data, much like with normal read errors. Only
1772 * read into the pages we already have so we don't
1773 * need to re-issue the read request.
1774 * We don't need to freeze the array, because being in an
1775 * active sync request, there is no normal IO, and
1776 * no overlapping syncs.
NeilBrown06f60382011-07-28 11:31:48 +10001777 * We don't need to check is_badblock() again as we
1778 * made sure that anything with a bad block in range
1779 * will have bi_end_io clear.
NeilBrowna68e5872011-05-11 14:40:44 +10001780 */
NeilBrownfd01b882011-10-11 16:47:53 +11001781 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001782 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001783 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1784 sector_t sect = r1_bio->sector;
1785 int sectors = r1_bio->sectors;
1786 int idx = 0;
1787
1788 while(sectors) {
1789 int s = sectors;
1790 int d = r1_bio->read_disk;
1791 int success = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11001792 struct md_rdev *rdev;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001793 int start;
NeilBrowna68e5872011-05-11 14:40:44 +10001794
1795 if (s > (PAGE_SIZE>>9))
1796 s = PAGE_SIZE >> 9;
1797 do {
1798 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1799 /* No rcu protection needed here devices
1800 * can only be removed when no resync is
1801 * active, and resync is currently active
1802 */
1803 rdev = conf->mirrors[d].rdev;
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001804 if (sync_page_io(rdev, sect, s<<9,
NeilBrowna68e5872011-05-11 14:40:44 +10001805 bio->bi_io_vec[idx].bv_page,
Mike Christie796a5cf2016-06-05 14:32:07 -05001806 REQ_OP_READ, 0, false)) {
NeilBrowna68e5872011-05-11 14:40:44 +10001807 success = 1;
1808 break;
1809 }
1810 }
1811 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001812 if (d == conf->raid_disks * 2)
NeilBrowna68e5872011-05-11 14:40:44 +10001813 d = 0;
1814 } while (!success && d != r1_bio->read_disk);
1815
NeilBrown78d7f5f2011-05-11 14:48:56 +10001816 if (!success) {
NeilBrowna68e5872011-05-11 14:40:44 +10001817 char b[BDEVNAME_SIZE];
NeilBrown3a9f28a2011-07-28 11:33:42 +10001818 int abort = 0;
1819 /* Cannot read from anywhere, this block is lost.
1820 * Record a bad block on each device. If that doesn't
1821 * work just disable and interrupt the recovery.
1822 * Don't fail devices as that won't really help.
1823 */
NeilBrown1d41c212016-11-02 14:16:50 +11001824 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
1825 mdname(mddev),
1826 bdevname(bio->bi_bdev, b),
1827 (unsigned long long)r1_bio->sector);
NeilBrown8f19ccb2011-12-23 10:17:56 +11001828 for (d = 0; d < conf->raid_disks * 2; d++) {
NeilBrown3a9f28a2011-07-28 11:33:42 +10001829 rdev = conf->mirrors[d].rdev;
1830 if (!rdev || test_bit(Faulty, &rdev->flags))
1831 continue;
1832 if (!rdev_set_badblocks(rdev, sect, s, 0))
1833 abort = 1;
1834 }
1835 if (abort) {
NeilBrownd890fa22011-10-26 11:54:39 +11001836 conf->recovery_disabled =
1837 mddev->recovery_disabled;
NeilBrown3a9f28a2011-07-28 11:33:42 +10001838 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1839 md_done_sync(mddev, r1_bio->sectors, 0);
1840 put_buf(r1_bio);
1841 return 0;
1842 }
1843 /* Try next page */
1844 sectors -= s;
1845 sect += s;
1846 idx++;
1847 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001848 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001849
1850 start = d;
1851 /* write it back and re-read */
1852 while (d != r1_bio->read_disk) {
1853 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001854 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001855 d--;
1856 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1857 continue;
1858 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001859 if (r1_sync_page_io(rdev, sect, s,
1860 bio->bi_io_vec[idx].bv_page,
1861 WRITE) == 0) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001862 r1_bio->bios[d]->bi_end_io = NULL;
1863 rdev_dec_pending(rdev, mddev);
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001864 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001865 }
1866 d = start;
1867 while (d != r1_bio->read_disk) {
1868 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001869 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001870 d--;
1871 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1872 continue;
1873 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001874 if (r1_sync_page_io(rdev, sect, s,
1875 bio->bi_io_vec[idx].bv_page,
1876 READ) != 0)
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001877 atomic_add(s, &rdev->corrected_errors);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001878 }
NeilBrowna68e5872011-05-11 14:40:44 +10001879 sectors -= s;
1880 sect += s;
1881 idx ++;
1882 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001883 set_bit(R1BIO_Uptodate, &r1_bio->state);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001884 bio->bi_error = 0;
NeilBrowna68e5872011-05-11 14:40:44 +10001885 return 1;
1886}
1887
NeilBrownc95e6382014-09-09 13:54:11 +10001888static void process_checks(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001889{
1890 /* We have read all readable devices. If we haven't
1891 * got the block, then there is no hope left.
1892 * If we have, then we want to do a comparison
1893 * and skip the write if everything is the same.
1894 * If any blocks failed to read, then we need to
1895 * attempt an over-write
1896 */
NeilBrownfd01b882011-10-11 16:47:53 +11001897 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001898 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001899 int primary;
1900 int i;
majianpengf4380a92012-04-12 16:04:47 +10001901 int vcnt;
NeilBrowna68e5872011-05-11 14:40:44 +10001902
NeilBrown30bc9b52013-07-17 15:19:29 +10001903 /* Fix variable parts of all bios */
1904 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1905 for (i = 0; i < conf->raid_disks * 2; i++) {
1906 int j;
1907 int size;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001908 int error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001909 struct bio *b = r1_bio->bios[i];
1910 if (b->bi_end_io != end_sync_read)
1911 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001912 /* fixup the bio for reuse, but preserve errno */
1913 error = b->bi_error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001914 bio_reset(b);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001915 b->bi_error = error;
NeilBrown30bc9b52013-07-17 15:19:29 +10001916 b->bi_vcnt = vcnt;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001917 b->bi_iter.bi_size = r1_bio->sectors << 9;
1918 b->bi_iter.bi_sector = r1_bio->sector +
NeilBrown30bc9b52013-07-17 15:19:29 +10001919 conf->mirrors[i].rdev->data_offset;
1920 b->bi_bdev = conf->mirrors[i].rdev->bdev;
1921 b->bi_end_io = end_sync_read;
1922 b->bi_private = r1_bio;
1923
Kent Overstreet4f024f32013-10-11 15:44:27 -07001924 size = b->bi_iter.bi_size;
NeilBrown30bc9b52013-07-17 15:19:29 +10001925 for (j = 0; j < vcnt ; j++) {
1926 struct bio_vec *bi;
1927 bi = &b->bi_io_vec[j];
1928 bi->bv_offset = 0;
1929 if (size > PAGE_SIZE)
1930 bi->bv_len = PAGE_SIZE;
1931 else
1932 bi->bv_len = size;
1933 size -= PAGE_SIZE;
1934 }
1935 }
NeilBrown8f19ccb2011-12-23 10:17:56 +11001936 for (primary = 0; primary < conf->raid_disks * 2; primary++)
NeilBrowna68e5872011-05-11 14:40:44 +10001937 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001938 !r1_bio->bios[primary]->bi_error) {
NeilBrowna68e5872011-05-11 14:40:44 +10001939 r1_bio->bios[primary]->bi_end_io = NULL;
1940 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1941 break;
1942 }
1943 r1_bio->read_disk = primary;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001944 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001945 int j;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001946 struct bio *pbio = r1_bio->bios[primary];
1947 struct bio *sbio = r1_bio->bios[i];
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001948 int error = sbio->bi_error;
NeilBrowna68e5872011-05-11 14:40:44 +10001949
Kent Overstreet2aabaa62012-09-11 11:26:12 -07001950 if (sbio->bi_end_io != end_sync_read)
NeilBrown78d7f5f2011-05-11 14:48:56 +10001951 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001952 /* Now we can 'fixup' the error value */
1953 sbio->bi_error = 0;
NeilBrowna68e5872011-05-11 14:40:44 +10001954
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001955 if (!error) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001956 for (j = vcnt; j-- ; ) {
1957 struct page *p, *s;
1958 p = pbio->bi_io_vec[j].bv_page;
1959 s = sbio->bi_io_vec[j].bv_page;
1960 if (memcmp(page_address(p),
1961 page_address(s),
NeilBrown5020ad72012-04-02 01:39:05 +10001962 sbio->bi_io_vec[j].bv_len))
NeilBrown78d7f5f2011-05-11 14:48:56 +10001963 break;
NeilBrowna68e5872011-05-11 14:40:44 +10001964 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001965 } else
1966 j = 0;
1967 if (j >= 0)
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11001968 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001969 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001970 && !error)) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001971 /* No need to write to this device. */
1972 sbio->bi_end_io = NULL;
1973 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1974 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001975 }
Kent Overstreetd3b45c22012-09-10 13:49:33 -07001976
1977 bio_copy_data(sbio, pbio);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001978 }
NeilBrowna68e5872011-05-11 14:40:44 +10001979}
1980
NeilBrown9f2c9d12011-10-11 16:48:43 +11001981static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982{
NeilBrowne8096362011-10-11 16:49:05 +11001983 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 int i;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001985 int disks = conf->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 struct bio *bio, *wbio;
1987
1988 bio = r1_bio->bios[r1_bio->read_disk];
1989
NeilBrowna68e5872011-05-11 14:40:44 +10001990 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
1991 /* ouch - failed to read all of that. */
1992 if (!fix_sync_read_error(r1_bio))
1993 return;
NeilBrown7ca78d52011-05-11 14:50:37 +10001994
1995 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrownc95e6382014-09-09 13:54:11 +10001996 process_checks(r1_bio);
1997
NeilBrownd11c1712006-01-06 00:20:26 -08001998 /*
1999 * schedule writes
2000 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 atomic_set(&r1_bio->remaining, 1);
2002 for (i = 0; i < disks ; i++) {
2003 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08002004 if (wbio->bi_end_io == NULL ||
2005 (wbio->bi_end_io == end_sync_read &&
2006 (i == r1_bio->read_disk ||
2007 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 continue;
2009
Mike Christie796a5cf2016-06-05 14:32:07 -05002010 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
NeilBrown3e198f72006-01-06 00:20:21 -08002011 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 atomic_inc(&r1_bio->remaining);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002013 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
NeilBrown191ea9b2005-06-21 17:17:23 -07002014
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 generic_make_request(wbio);
2016 }
2017
2018 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002019 /* if we're here, all write(s) have completed, so clean up */
NeilBrown58e94ae2012-07-19 15:59:18 +10002020 int s = r1_bio->sectors;
2021 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2022 test_bit(R1BIO_WriteError, &r1_bio->state))
2023 reschedule_retry(r1_bio);
2024 else {
2025 put_buf(r1_bio);
2026 md_done_sync(mddev, s, 1);
2027 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 }
2029}
2030
2031/*
2032 * This is a kernel thread which:
2033 *
2034 * 1. Retries failed read operations on working mirrors.
2035 * 2. Updates the raid superblock when problems encounter.
NeilBrownd2eb35a2011-07-28 11:31:48 +10002036 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 */
2038
NeilBrowne8096362011-10-11 16:49:05 +11002039static void fix_read_error(struct r1conf *conf, int read_disk,
NeilBrown867868f2006-10-03 01:15:51 -07002040 sector_t sect, int sectors)
2041{
NeilBrownfd01b882011-10-11 16:47:53 +11002042 struct mddev *mddev = conf->mddev;
NeilBrown867868f2006-10-03 01:15:51 -07002043 while(sectors) {
2044 int s = sectors;
2045 int d = read_disk;
2046 int success = 0;
2047 int start;
NeilBrown3cb03002011-10-11 16:45:26 +11002048 struct md_rdev *rdev;
NeilBrown867868f2006-10-03 01:15:51 -07002049
2050 if (s > (PAGE_SIZE>>9))
2051 s = PAGE_SIZE >> 9;
2052
2053 do {
NeilBrownd2eb35a2011-07-28 11:31:48 +10002054 sector_t first_bad;
2055 int bad_sectors;
2056
NeilBrown707a6a42016-06-02 16:19:52 +10002057 rcu_read_lock();
2058 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002059 if (rdev &&
majianpengda8840a2012-05-22 13:55:03 +10002060 (test_bit(In_sync, &rdev->flags) ||
2061 (!test_bit(Faulty, &rdev->flags) &&
2062 rdev->recovery_offset >= sect + s)) &&
NeilBrownd2eb35a2011-07-28 11:31:48 +10002063 is_badblock(rdev, sect, s,
NeilBrown707a6a42016-06-02 16:19:52 +10002064 &first_bad, &bad_sectors) == 0) {
2065 atomic_inc(&rdev->nr_pending);
2066 rcu_read_unlock();
2067 if (sync_page_io(rdev, sect, s<<9,
Mike Christie796a5cf2016-06-05 14:32:07 -05002068 conf->tmppage, REQ_OP_READ, 0, false))
NeilBrown707a6a42016-06-02 16:19:52 +10002069 success = 1;
2070 rdev_dec_pending(rdev, mddev);
2071 if (success)
2072 break;
2073 } else
2074 rcu_read_unlock();
2075 d++;
2076 if (d == conf->raid_disks * 2)
2077 d = 0;
NeilBrown867868f2006-10-03 01:15:51 -07002078 } while (!success && d != read_disk);
2079
2080 if (!success) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002081 /* Cannot read from anywhere - mark it bad */
NeilBrown3cb03002011-10-11 16:45:26 +11002082 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10002083 if (!rdev_set_badblocks(rdev, sect, s, 0))
2084 md_error(mddev, rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002085 break;
2086 }
2087 /* write it back and re-read */
2088 start = d;
2089 while (d != read_disk) {
2090 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002091 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002092 d--;
NeilBrown707a6a42016-06-02 16:19:52 +10002093 rcu_read_lock();
2094 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002095 if (rdev &&
NeilBrown707a6a42016-06-02 16:19:52 +10002096 !test_bit(Faulty, &rdev->flags)) {
2097 atomic_inc(&rdev->nr_pending);
2098 rcu_read_unlock();
NeilBrownd8f05d22011-07-28 11:33:00 +10002099 r1_sync_page_io(rdev, sect, s,
2100 conf->tmppage, WRITE);
NeilBrown707a6a42016-06-02 16:19:52 +10002101 rdev_dec_pending(rdev, mddev);
2102 } else
2103 rcu_read_unlock();
NeilBrown867868f2006-10-03 01:15:51 -07002104 }
2105 d = start;
2106 while (d != read_disk) {
2107 char b[BDEVNAME_SIZE];
2108 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002109 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002110 d--;
NeilBrown707a6a42016-06-02 16:19:52 +10002111 rcu_read_lock();
2112 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002113 if (rdev &&
NeilBrownb8cb6b42014-09-18 11:09:04 +10002114 !test_bit(Faulty, &rdev->flags)) {
NeilBrown707a6a42016-06-02 16:19:52 +10002115 atomic_inc(&rdev->nr_pending);
2116 rcu_read_unlock();
NeilBrownd8f05d22011-07-28 11:33:00 +10002117 if (r1_sync_page_io(rdev, sect, s,
2118 conf->tmppage, READ)) {
NeilBrown867868f2006-10-03 01:15:51 -07002119 atomic_add(s, &rdev->corrected_errors);
NeilBrown1d41c212016-11-02 14:16:50 +11002120 pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n",
2121 mdname(mddev), s,
2122 (unsigned long long)(sect +
2123 rdev->data_offset),
2124 bdevname(rdev->bdev, b));
NeilBrown867868f2006-10-03 01:15:51 -07002125 }
NeilBrown707a6a42016-06-02 16:19:52 +10002126 rdev_dec_pending(rdev, mddev);
2127 } else
2128 rcu_read_unlock();
NeilBrown867868f2006-10-03 01:15:51 -07002129 }
2130 sectors -= s;
2131 sect += s;
2132 }
2133}
2134
NeilBrown9f2c9d12011-10-11 16:48:43 +11002135static int narrow_write_error(struct r1bio *r1_bio, int i)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002136{
NeilBrownfd01b882011-10-11 16:47:53 +11002137 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11002138 struct r1conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11002139 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002140
2141 /* bio has the data to be written to device 'i' where
2142 * we just recently had a write error.
2143 * We repeatedly clone the bio and trim down to one block,
2144 * then try the write. Where the write fails we record
2145 * a bad block.
2146 * It is conceivable that the bio doesn't exactly align with
2147 * blocks. We must handle this somehow.
2148 *
2149 * We currently own a reference on the rdev.
2150 */
2151
2152 int block_sectors;
2153 sector_t sector;
2154 int sectors;
2155 int sect_to_write = r1_bio->sectors;
2156 int ok = 1;
2157
2158 if (rdev->badblocks.shift < 0)
2159 return 0;
2160
Nate Daileyab713cd2015-02-12 12:02:09 -05002161 block_sectors = roundup(1 << rdev->badblocks.shift,
2162 bdev_logical_block_size(rdev->bdev) >> 9);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002163 sector = r1_bio->sector;
2164 sectors = ((sector + block_sectors)
2165 & ~(sector_t)(block_sectors - 1))
2166 - sector;
2167
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002168 while (sect_to_write) {
2169 struct bio *wbio;
2170 if (sectors > sect_to_write)
2171 sectors = sect_to_write;
2172 /* Write at 'sector' for 'sectors'*/
2173
Kent Overstreetb7838632012-09-10 15:17:11 -07002174 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2175 unsigned vcnt = r1_bio->behind_page_count;
2176 struct bio_vec *vec = r1_bio->behind_bvecs;
2177
2178 while (!vec->bv_page) {
2179 vec++;
2180 vcnt--;
2181 }
2182
2183 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2184 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2185
2186 wbio->bi_vcnt = vcnt;
2187 } else {
2188 wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2189 }
2190
Mike Christie796a5cf2016-06-05 14:32:07 -05002191 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002192 wbio->bi_iter.bi_sector = r1_bio->sector;
2193 wbio->bi_iter.bi_size = r1_bio->sectors << 9;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002194
Kent Overstreet6678d832013-08-07 11:14:32 -07002195 bio_trim(wbio, sector - r1_bio->sector, sectors);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002196 wbio->bi_iter.bi_sector += rdev->data_offset;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002197 wbio->bi_bdev = rdev->bdev;
Mike Christie4e49ea42016-06-05 14:31:41 -05002198
2199 if (submit_bio_wait(wbio) < 0)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002200 /* failure! */
2201 ok = rdev_set_badblocks(rdev, sector,
2202 sectors, 0)
2203 && ok;
2204
2205 bio_put(wbio);
2206 sect_to_write -= sectors;
2207 sector += sectors;
2208 sectors = block_sectors;
2209 }
2210 return ok;
2211}
2212
NeilBrowne8096362011-10-11 16:49:05 +11002213static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002214{
2215 int m;
2216 int s = r1_bio->sectors;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002217 for (m = 0; m < conf->raid_disks * 2 ; m++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002218 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002219 struct bio *bio = r1_bio->bios[m];
2220 if (bio->bi_end_io == NULL)
2221 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002222 if (!bio->bi_error &&
NeilBrown62096bc2011-07-28 11:38:13 +10002223 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
NeilBrownc6563a82012-05-21 09:27:00 +10002224 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002225 }
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002226 if (bio->bi_error &&
NeilBrown62096bc2011-07-28 11:38:13 +10002227 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2228 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2229 md_error(conf->mddev, rdev);
2230 }
2231 }
2232 put_buf(r1_bio);
2233 md_done_sync(conf->mddev, s, 1);
2234}
2235
NeilBrowne8096362011-10-11 16:49:05 +11002236static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002237{
2238 int m;
NeilBrown55ce74d2015-08-14 11:11:10 +10002239 bool fail = false;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002240 for (m = 0; m < conf->raid_disks * 2 ; m++)
NeilBrown62096bc2011-07-28 11:38:13 +10002241 if (r1_bio->bios[m] == IO_MADE_GOOD) {
NeilBrown3cb03002011-10-11 16:45:26 +11002242 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002243 rdev_clear_badblocks(rdev,
2244 r1_bio->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10002245 r1_bio->sectors, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002246 rdev_dec_pending(rdev, conf->mddev);
2247 } else if (r1_bio->bios[m] != NULL) {
2248 /* This drive got a write error. We need to
2249 * narrow down and record precise write
2250 * errors.
2251 */
NeilBrown55ce74d2015-08-14 11:11:10 +10002252 fail = true;
NeilBrown62096bc2011-07-28 11:38:13 +10002253 if (!narrow_write_error(r1_bio, m)) {
2254 md_error(conf->mddev,
2255 conf->mirrors[m].rdev);
2256 /* an I/O failed, we can't clear the bitmap */
2257 set_bit(R1BIO_Degraded, &r1_bio->state);
2258 }
2259 rdev_dec_pending(conf->mirrors[m].rdev,
2260 conf->mddev);
2261 }
NeilBrown55ce74d2015-08-14 11:11:10 +10002262 if (fail) {
2263 spin_lock_irq(&conf->device_lock);
2264 list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
Nate Daileyccfc7bf2016-02-29 10:43:58 -05002265 conf->nr_queued++;
NeilBrown55ce74d2015-08-14 11:11:10 +10002266 spin_unlock_irq(&conf->device_lock);
2267 md_wakeup_thread(conf->mddev->thread);
NeilBrownbd8688a2015-10-24 16:02:16 +11002268 } else {
2269 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2270 close_write(r1_bio);
NeilBrown55ce74d2015-08-14 11:11:10 +10002271 raid_end_bio_io(r1_bio);
NeilBrownbd8688a2015-10-24 16:02:16 +11002272 }
NeilBrown62096bc2011-07-28 11:38:13 +10002273}
2274
NeilBrowne8096362011-10-11 16:49:05 +11002275static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002276{
2277 int disk;
2278 int max_sectors;
NeilBrownfd01b882011-10-11 16:47:53 +11002279 struct mddev *mddev = conf->mddev;
NeilBrown62096bc2011-07-28 11:38:13 +10002280 struct bio *bio;
2281 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11002282 struct md_rdev *rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002283
2284 clear_bit(R1BIO_ReadError, &r1_bio->state);
2285 /* we got a read error. Maybe the drive is bad. Maybe just
2286 * the block and we can fix it.
2287 * We freeze all other IO, and try reading the block from
2288 * other devices. When we find one, we re-write
2289 * and check it that fixes the read error.
2290 * This is all done synchronously while the array is
2291 * frozen
2292 */
Tomasz Majchrzak7449f692016-10-28 14:45:58 +02002293
2294 bio = r1_bio->bios[r1_bio->read_disk];
2295 bdevname(bio->bi_bdev, b);
2296 bio_put(bio);
2297 r1_bio->bios[r1_bio->read_disk] = NULL;
2298
NeilBrown62096bc2011-07-28 11:38:13 +10002299 if (mddev->ro == 0) {
NeilBrowne2d59922013-06-12 11:01:22 +10002300 freeze_array(conf, 1);
NeilBrown62096bc2011-07-28 11:38:13 +10002301 fix_read_error(conf, r1_bio->read_disk,
2302 r1_bio->sector, r1_bio->sectors);
2303 unfreeze_array(conf);
Tomasz Majchrzak7449f692016-10-28 14:45:58 +02002304 } else {
2305 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;
2306 }
2307
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
NeilBrown62096bc2011-07-28 11:38:13 +10002310read_more:
2311 disk = read_balance(conf, r1_bio, &max_sectors);
2312 if (disk == -1) {
NeilBrown1d41c212016-11-02 14:16:50 +11002313 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
2314 mdname(mddev), b, (unsigned long long)r1_bio->sector);
NeilBrown62096bc2011-07-28 11:38:13 +10002315 raid_end_bio_io(r1_bio);
2316 } else {
2317 const unsigned long do_sync
Jens Axboe1eff9d32016-08-05 15:35:16 -06002318 = r1_bio->master_bio->bi_opf & REQ_SYNC;
NeilBrown62096bc2011-07-28 11:38:13 +10002319 r1_bio->read_disk = disk;
2320 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002321 bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
2322 max_sectors);
NeilBrown62096bc2011-07-28 11:38:13 +10002323 r1_bio->bios[r1_bio->read_disk] = bio;
2324 rdev = conf->mirrors[disk].rdev;
NeilBrown1d41c212016-11-02 14:16:50 +11002325 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n",
2326 mdname(mddev),
2327 (unsigned long long)r1_bio->sector,
2328 bdevname(rdev->bdev, b));
Kent Overstreet4f024f32013-10-11 15:44:27 -07002329 bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
NeilBrown62096bc2011-07-28 11:38:13 +10002330 bio->bi_bdev = rdev->bdev;
2331 bio->bi_end_io = raid1_end_read_request;
Mike Christie796a5cf2016-06-05 14:32:07 -05002332 bio_set_op_attrs(bio, REQ_OP_READ, do_sync);
NeilBrown62096bc2011-07-28 11:38:13 +10002333 bio->bi_private = r1_bio;
2334 if (max_sectors < r1_bio->sectors) {
2335 /* Drat - have to split this up more */
2336 struct bio *mbio = r1_bio->master_bio;
2337 int sectors_handled = (r1_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07002338 - mbio->bi_iter.bi_sector);
NeilBrown62096bc2011-07-28 11:38:13 +10002339 r1_bio->sectors = max_sectors;
2340 spin_lock_irq(&conf->device_lock);
2341 if (mbio->bi_phys_segments == 0)
2342 mbio->bi_phys_segments = 2;
2343 else
2344 mbio->bi_phys_segments++;
2345 spin_unlock_irq(&conf->device_lock);
2346 generic_make_request(bio);
2347 bio = NULL;
2348
2349 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2350
2351 r1_bio->master_bio = mbio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002352 r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
NeilBrown62096bc2011-07-28 11:38:13 +10002353 r1_bio->state = 0;
2354 set_bit(R1BIO_ReadError, &r1_bio->state);
2355 r1_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002356 r1_bio->sector = mbio->bi_iter.bi_sector +
2357 sectors_handled;
NeilBrown62096bc2011-07-28 11:38:13 +10002358
2359 goto read_more;
2360 } else
2361 generic_make_request(bio);
2362 }
2363}
2364
Shaohua Li4ed87312012-10-11 13:34:00 +11002365static void raid1d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366{
Shaohua Li4ed87312012-10-11 13:34:00 +11002367 struct mddev *mddev = thread->mddev;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002368 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 unsigned long flags;
NeilBrowne8096362011-10-11 16:49:05 +11002370 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002372 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
2374 md_check_recovery(mddev);
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002375
NeilBrown55ce74d2015-08-14 11:11:10 +10002376 if (!list_empty_careful(&conf->bio_end_io_list) &&
2377 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2378 LIST_HEAD(tmp);
2379 spin_lock_irqsave(&conf->device_lock, flags);
2380 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
Nate Daileyccfc7bf2016-02-29 10:43:58 -05002381 while (!list_empty(&conf->bio_end_io_list)) {
2382 list_move(conf->bio_end_io_list.prev, &tmp);
2383 conf->nr_queued--;
2384 }
NeilBrown55ce74d2015-08-14 11:11:10 +10002385 }
2386 spin_unlock_irqrestore(&conf->device_lock, flags);
2387 while (!list_empty(&tmp)) {
Mikulas Patockaa4527442015-10-01 15:17:43 -04002388 r1_bio = list_first_entry(&tmp, struct r1bio,
2389 retry_list);
NeilBrown55ce74d2015-08-14 11:11:10 +10002390 list_del(&r1_bio->retry_list);
NeilBrownbd8688a2015-10-24 16:02:16 +11002391 if (mddev->degraded)
2392 set_bit(R1BIO_Degraded, &r1_bio->state);
2393 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2394 close_write(r1_bio);
NeilBrown55ce74d2015-08-14 11:11:10 +10002395 raid_end_bio_io(r1_bio);
2396 }
2397 }
2398
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002399 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002401
NeilBrown0021b7b2012-07-31 09:08:14 +02002402 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002403
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002405 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002406 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002408 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002409 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08002411 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 spin_unlock_irqrestore(&conf->device_lock, flags);
2413
2414 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002415 conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10002416 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002417 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002418 test_bit(R1BIO_WriteError, &r1_bio->state))
2419 handle_sync_write_finished(conf, r1_bio);
2420 else
NeilBrown4367af52011-07-28 11:31:49 +10002421 sync_request_write(mddev, r1_bio);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002422 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002423 test_bit(R1BIO_WriteError, &r1_bio->state))
2424 handle_write_finished(conf, r1_bio);
2425 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2426 handle_read_error(conf, r1_bio);
2427 else
NeilBrownd2eb35a2011-07-28 11:31:48 +10002428 /* just a partial read to be scheduled from separate
2429 * context
2430 */
2431 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
NeilBrown62096bc2011-07-28 11:38:13 +10002432
NeilBrown1d9d5242009-10-16 15:55:32 +11002433 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002434 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2435 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002437 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438}
2439
NeilBrowne8096362011-10-11 16:49:05 +11002440static int init_resync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441{
2442 int buffs;
2443
2444 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002445 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2447 conf->poolinfo);
2448 if (!conf->r1buf_pool)
2449 return -ENOMEM;
2450 conf->next_resync = 0;
2451 return 0;
2452}
2453
2454/*
2455 * perform a "sync" on one "block"
2456 *
2457 * We need to make sure that no normal I/O request - particularly write
2458 * requests - conflict with active sync requests.
2459 *
2460 * This is achieved by tracking pending requests and a 'barrier' concept
2461 * that can be installed to exclude normal IO requests.
2462 */
2463
Shaohua Li849674e2016-01-20 13:52:20 -08002464static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2465 int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466{
NeilBrowne8096362011-10-11 16:49:05 +11002467 struct r1conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002468 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 struct bio *bio;
2470 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08002471 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08002473 int wonly = -1;
2474 int write_targets = 0, read_targets = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11002475 sector_t sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07002476 int still_degraded = 0;
NeilBrown06f60382011-07-28 11:31:48 +10002477 int good_sectors = RESYNC_SECTORS;
2478 int min_bad = 0; /* number of sectors that are bad in all devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
2480 if (!conf->r1buf_pool)
2481 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002482 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
Andre Noll58c0fed2009-03-31 14:33:13 +11002484 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002486 /* If we aborted, we need to abort the
2487 * sync on the 'current' bitmap chunk (there will
2488 * only be one in raid1 resync.
2489 * We can find the current addess in mddev->curr_resync
2490 */
NeilBrown6a806c52005-07-15 03:56:35 -07002491 if (mddev->curr_resync < max_sector) /* aborted */
2492 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07002493 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07002494 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07002495 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07002496
2497 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 close_sync(conf);
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002499
2500 if (mddev_is_clustered(mddev)) {
2501 conf->cluster_sync_low = 0;
2502 conf->cluster_sync_high = 0;
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 return 0;
2505 }
2506
NeilBrown07d84d102006-06-26 00:27:56 -07002507 if (mddev->bitmap == NULL &&
2508 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07002509 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07002510 conf->fullsync == 0) {
2511 *skipped = 1;
2512 return max_sector - sector_nr;
2513 }
NeilBrown6394cca2006-08-27 01:23:50 -07002514 /* before building a request, check if we can skip these blocks..
2515 * This call the bitmap_start_sync doesn't actually record anything
2516 */
NeilBrowne3b97032005-08-04 12:53:34 -07002517 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de485f2005-11-08 21:39:38 -08002518 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002519 /* We can skip this block, and probably several more */
2520 *skipped = 1;
2521 return sync_blocks;
2522 }
NeilBrown17999be2006-01-06 00:20:12 -08002523
Tomasz Majchrzak7ac50442016-06-13 15:51:19 +02002524 /*
2525 * If there is non-resync activity waiting for a turn, then let it
2526 * though before starting on this new sync request.
2527 */
2528 if (conf->nr_waiting)
2529 schedule_timeout_uninterruptible(1);
2530
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002531 /* we are incrementing sector_nr below. To be safe, we check against
2532 * sector_nr + two times RESYNC_SECTORS
2533 */
2534
2535 bitmap_cond_end_sync(mddev->bitmap, sector_nr,
2536 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
NeilBrown1c4588e2010-10-26 17:41:22 +11002537 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown17999be2006-01-06 00:20:12 -08002538
NeilBrownc2fd4c92014-09-10 16:01:24 +10002539 raise_barrier(conf, sector_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540
NeilBrown3e198f72006-01-06 00:20:21 -08002541 rcu_read_lock();
2542 /*
2543 * If we get a correctably read error during resync or recovery,
2544 * we might want to read from a different device. So we
2545 * flag all drives that could conceivably be read from for READ,
2546 * and any others (which will be non-In_sync devices) for WRITE.
2547 * If a read fails, we try reading from something else for which READ
2548 * is OK.
2549 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 r1_bio->mddev = mddev;
2552 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07002553 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555
NeilBrown8f19ccb2011-12-23 10:17:56 +11002556 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002557 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 bio = r1_bio->bios[i];
Kent Overstreet2aabaa62012-09-11 11:26:12 -07002559 bio_reset(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
NeilBrown3e198f72006-01-06 00:20:21 -08002561 rdev = rcu_dereference(conf->mirrors[i].rdev);
2562 if (rdev == NULL ||
NeilBrown06f60382011-07-28 11:31:48 +10002563 test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11002564 if (i < conf->raid_disks)
2565 still_degraded = 1;
NeilBrown3e198f72006-01-06 00:20:21 -08002566 } else if (!test_bit(In_sync, &rdev->flags)) {
Mike Christie796a5cf2016-06-05 14:32:07 -05002567 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 bio->bi_end_io = end_sync_write;
2569 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08002570 } else {
2571 /* may need to read from here */
NeilBrown06f60382011-07-28 11:31:48 +10002572 sector_t first_bad = MaxSector;
2573 int bad_sectors;
2574
2575 if (is_badblock(rdev, sector_nr, good_sectors,
2576 &first_bad, &bad_sectors)) {
2577 if (first_bad > sector_nr)
2578 good_sectors = first_bad - sector_nr;
2579 else {
2580 bad_sectors -= (sector_nr - first_bad);
2581 if (min_bad == 0 ||
2582 min_bad > bad_sectors)
2583 min_bad = bad_sectors;
2584 }
NeilBrown3e198f72006-01-06 00:20:21 -08002585 }
NeilBrown06f60382011-07-28 11:31:48 +10002586 if (sector_nr < first_bad) {
2587 if (test_bit(WriteMostly, &rdev->flags)) {
2588 if (wonly < 0)
2589 wonly = i;
2590 } else {
2591 if (disk < 0)
2592 disk = i;
2593 }
Mike Christie796a5cf2016-06-05 14:32:07 -05002594 bio_set_op_attrs(bio, REQ_OP_READ, 0);
NeilBrown06f60382011-07-28 11:31:48 +10002595 bio->bi_end_io = end_sync_read;
2596 read_targets++;
Alexander Lyakasd57368a2012-07-17 13:17:55 +03002597 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2598 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2599 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2600 /*
2601 * The device is suitable for reading (InSync),
2602 * but has bad block(s) here. Let's try to correct them,
2603 * if we are doing resync or repair. Otherwise, leave
2604 * this device alone for this sync request.
2605 */
Mike Christie796a5cf2016-06-05 14:32:07 -05002606 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Alexander Lyakasd57368a2012-07-17 13:17:55 +03002607 bio->bi_end_io = end_sync_write;
2608 write_targets++;
NeilBrown06f60382011-07-28 11:31:48 +10002609 }
NeilBrown3e198f72006-01-06 00:20:21 -08002610 }
NeilBrown06f60382011-07-28 11:31:48 +10002611 if (bio->bi_end_io) {
2612 atomic_inc(&rdev->nr_pending);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002613 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
NeilBrown06f60382011-07-28 11:31:48 +10002614 bio->bi_bdev = rdev->bdev;
2615 bio->bi_private = r1_bio;
2616 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 }
NeilBrown3e198f72006-01-06 00:20:21 -08002618 rcu_read_unlock();
2619 if (disk < 0)
2620 disk = wonly;
2621 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07002622
NeilBrown06f60382011-07-28 11:31:48 +10002623 if (read_targets == 0 && min_bad > 0) {
2624 /* These sectors are bad on all InSync devices, so we
2625 * need to mark them bad on all write targets
2626 */
2627 int ok = 1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002628 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
NeilBrown06f60382011-07-28 11:31:48 +10002629 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
majianpenga42f9d82012-04-02 01:04:19 +10002630 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown06f60382011-07-28 11:31:48 +10002631 ok = rdev_set_badblocks(rdev, sector_nr,
2632 min_bad, 0
2633 ) && ok;
2634 }
2635 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2636 *skipped = 1;
2637 put_buf(r1_bio);
2638
2639 if (!ok) {
2640 /* Cannot record the badblocks, so need to
2641 * abort the resync.
2642 * If there are multiple read targets, could just
2643 * fail the really bad ones ???
2644 */
2645 conf->recovery_disabled = mddev->recovery_disabled;
2646 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2647 return 0;
2648 } else
2649 return min_bad;
2650
2651 }
2652 if (min_bad > 0 && min_bad < good_sectors) {
2653 /* only resync enough to reach the next bad->good
2654 * transition */
2655 good_sectors = min_bad;
2656 }
2657
NeilBrown3e198f72006-01-06 00:20:21 -08002658 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2659 /* extra read targets are also write targets */
2660 write_targets += read_targets-1;
2661
2662 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 /* There is nowhere to write, so all non-sync
2664 * drives must be failed - so we are finished
2665 */
NeilBrownb7219cc2012-07-31 10:05:34 +10002666 sector_t rv;
2667 if (min_bad > 0)
2668 max_sector = sector_nr + min_bad;
2669 rv = max_sector - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07002670 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 return rv;
2673 }
2674
NeilBrownc6207272008-02-06 01:39:52 -08002675 if (max_sector > mddev->resync_max)
2676 max_sector = mddev->resync_max; /* Don't do IO beyond here */
NeilBrown06f60382011-07-28 11:31:48 +10002677 if (max_sector > sector_nr + good_sectors)
2678 max_sector = sector_nr + good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07002680 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 do {
2682 struct page *page;
2683 int len = PAGE_SIZE;
2684 if (sector_nr + (len>>9) > max_sector)
2685 len = (max_sector - sector_nr) << 9;
2686 if (len == 0)
2687 break;
NeilBrown6a806c52005-07-15 03:56:35 -07002688 if (sync_blocks == 0) {
2689 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de485f2005-11-08 21:39:38 -08002690 &sync_blocks, still_degraded) &&
2691 !conf->fullsync &&
2692 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07002693 break;
NeilBrown7571ae82010-10-07 11:54:46 +11002694 if ((len >> 9) > sync_blocks)
NeilBrown6a806c52005-07-15 03:56:35 -07002695 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07002696 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002697
NeilBrown8f19ccb2011-12-23 10:17:56 +11002698 for (i = 0 ; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 bio = r1_bio->bios[i];
2700 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08002701 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 if (bio_add_page(bio, page, len, 0) == 0) {
2703 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08002704 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 while (i > 0) {
2706 i--;
2707 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07002708 if (bio->bi_end_io==NULL)
2709 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 /* remove last page from this bio */
2711 bio->bi_vcnt--;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002712 bio->bi_iter.bi_size -= len;
Jens Axboeb7c44ed2015-07-24 12:37:59 -06002713 bio_clear_flag(bio, BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 }
2715 goto bio_full;
2716 }
2717 }
2718 }
2719 nr_sectors += len>>9;
2720 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07002721 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2723 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 r1_bio->sectors = nr_sectors;
2725
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002726 if (mddev_is_clustered(mddev) &&
2727 conf->cluster_sync_high < sector_nr + nr_sectors) {
2728 conf->cluster_sync_low = mddev->curr_resync_completed;
2729 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2730 /* Send resync message */
2731 md_cluster_ops->resync_info_update(mddev,
2732 conf->cluster_sync_low,
2733 conf->cluster_sync_high);
2734 }
2735
NeilBrownd11c1712006-01-06 00:20:26 -08002736 /* For a user-requested sync, we read all readable devices and do a
2737 * compare
2738 */
2739 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2740 atomic_set(&r1_bio->remaining, read_targets);
NeilBrown2d4f4f32012-07-09 11:34:13 +10002741 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
NeilBrownd11c1712006-01-06 00:20:26 -08002742 bio = r1_bio->bios[i];
2743 if (bio->bi_end_io == end_sync_read) {
NeilBrown2d4f4f32012-07-09 11:34:13 +10002744 read_targets--;
NeilBrownddac7c72006-08-31 21:27:36 -07002745 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002746 generic_make_request(bio);
2747 }
2748 }
2749 } else {
2750 atomic_set(&r1_bio->remaining, 1);
2751 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07002752 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002753 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754
NeilBrownd11c1712006-01-06 00:20:26 -08002755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 return nr_sectors;
2757}
2758
NeilBrownfd01b882011-10-11 16:47:53 +11002759static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07002760{
2761 if (sectors)
2762 return sectors;
2763
2764 return mddev->dev_sectors;
2765}
2766
NeilBrowne8096362011-10-11 16:49:05 +11002767static struct r1conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768{
NeilBrowne8096362011-10-11 16:49:05 +11002769 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002770 int i;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002771 struct raid1_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11002772 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002773 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
NeilBrowne8096362011-10-11 16:49:05 +11002775 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11002777 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002779 conf->mirrors = kzalloc(sizeof(struct raid1_info)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002780 * mddev->raid_disks * 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 GFP_KERNEL);
2782 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11002783 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784
NeilBrownddaf22a2006-01-06 00:20:19 -08002785 conf->tmppage = alloc_page(GFP_KERNEL);
2786 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11002787 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08002788
NeilBrown709ae482009-12-14 12:49:51 +11002789 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11002791 goto abort;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002792 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2794 r1bio_pool_free,
2795 conf->poolinfo);
2796 if (!conf->r1bio_pool)
NeilBrown709ae482009-12-14 12:49:51 +11002797 goto abort;
2798
NeilBrowned9bfdf2009-10-16 15:55:44 +11002799 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800
NeilBrownc19d5792011-12-23 10:17:57 +11002801 err = -EINVAL;
Neil Browne7e72bf2008-05-14 16:05:54 -07002802 spin_lock_init(&conf->device_lock);
NeilBrowndafb20f2012-03-19 12:46:39 +11002803 rdev_for_each(rdev, mddev) {
NeilBrownaba336b2012-05-31 15:39:11 +10002804 struct request_queue *q;
NeilBrown709ae482009-12-14 12:49:51 +11002805 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 if (disk_idx >= mddev->raid_disks
2807 || disk_idx < 0)
2808 continue;
NeilBrownc19d5792011-12-23 10:17:57 +11002809 if (test_bit(Replacement, &rdev->flags))
NeilBrown02b898f2012-10-31 11:42:03 +11002810 disk = conf->mirrors + mddev->raid_disks + disk_idx;
NeilBrownc19d5792011-12-23 10:17:57 +11002811 else
2812 disk = conf->mirrors + disk_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813
NeilBrownc19d5792011-12-23 10:17:57 +11002814 if (disk->rdev)
2815 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 disk->rdev = rdev;
NeilBrownaba336b2012-05-31 15:39:11 +10002817 q = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818
2819 disk->head_position = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +10002820 disk->seq_start = MaxSector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 }
2822 conf->raid_disks = mddev->raid_disks;
2823 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 INIT_LIST_HEAD(&conf->retry_list);
NeilBrown55ce74d2015-08-14 11:11:10 +10002825 INIT_LIST_HEAD(&conf->bio_end_io_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826
2827 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08002828 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829
NeilBrown191ea9b2005-06-21 17:17:23 -07002830 bio_list_init(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +11002831 conf->pending_count = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11002832 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown191ea9b2005-06-21 17:17:23 -07002833
majianpeng79ef3a82013-11-15 14:55:02 +08002834 conf->start_next_window = MaxSector;
2835 conf->current_window_requests = conf->next_window_requests = 0;
2836
NeilBrownc19d5792011-12-23 10:17:57 +11002837 err = -EIO;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002838 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839
2840 disk = conf->mirrors + i;
2841
NeilBrownc19d5792011-12-23 10:17:57 +11002842 if (i < conf->raid_disks &&
2843 disk[conf->raid_disks].rdev) {
2844 /* This slot has a replacement. */
2845 if (!disk->rdev) {
2846 /* No original, just make the replacement
2847 * a recovering spare
2848 */
2849 disk->rdev =
2850 disk[conf->raid_disks].rdev;
2851 disk[conf->raid_disks].rdev = NULL;
2852 } else if (!test_bit(In_sync, &disk->rdev->flags))
2853 /* Original is not in_sync - bad */
2854 goto abort;
2855 }
2856
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002857 if (!disk->rdev ||
2858 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 disk->head_position = 0;
Jonathan Brassow4f0a5e02012-05-22 13:55:31 +10002860 if (disk->rdev &&
2861 (disk->rdev->saved_raid_disk < 0))
NeilBrown918f0232007-08-22 14:01:52 -07002862 conf->fullsync = 1;
Shaohua Libe4d3282012-07-31 10:03:53 +10002863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 }
NeilBrown709ae482009-12-14 12:49:51 +11002865
NeilBrown709ae482009-12-14 12:49:51 +11002866 err = -ENOMEM;
NeilBrown02326052012-07-03 15:56:52 +10002867 conf->thread = md_register_thread(raid1d, mddev, "raid1");
NeilBrown1d41c212016-11-02 14:16:50 +11002868 if (!conf->thread)
NeilBrown709ae482009-12-14 12:49:51 +11002869 goto abort;
NeilBrown191ea9b2005-06-21 17:17:23 -07002870
NeilBrown709ae482009-12-14 12:49:51 +11002871 return conf;
2872
2873 abort:
2874 if (conf) {
Julia Lawall644df1a2015-09-13 14:15:10 +02002875 mempool_destroy(conf->r1bio_pool);
NeilBrown709ae482009-12-14 12:49:51 +11002876 kfree(conf->mirrors);
2877 safe_put_page(conf->tmppage);
2878 kfree(conf->poolinfo);
2879 kfree(conf);
2880 }
2881 return ERR_PTR(err);
2882}
2883
NeilBrownafa0f552014-12-15 12:56:58 +11002884static void raid1_free(struct mddev *mddev, void *priv);
Shaohua Li849674e2016-01-20 13:52:20 -08002885static int raid1_run(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002886{
NeilBrowne8096362011-10-11 16:49:05 +11002887 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002888 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11002889 struct md_rdev *rdev;
majianpeng5220ea12012-04-02 09:48:38 +10002890 int ret;
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002891 bool discard_supported = false;
NeilBrown709ae482009-12-14 12:49:51 +11002892
2893 if (mddev->level != 1) {
NeilBrown1d41c212016-11-02 14:16:50 +11002894 pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n",
2895 mdname(mddev), mddev->level);
NeilBrown709ae482009-12-14 12:49:51 +11002896 return -EIO;
2897 }
2898 if (mddev->reshape_position != MaxSector) {
NeilBrown1d41c212016-11-02 14:16:50 +11002899 pr_warn("md/raid1:%s: reshape_position set but not supported\n",
2900 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11002901 return -EIO;
2902 }
2903 /*
2904 * copy the already verified devices into our private RAID1
2905 * bookkeeping area. [whatever we allocate in run(),
NeilBrownafa0f552014-12-15 12:56:58 +11002906 * should be freed in raid1_free()]
NeilBrown709ae482009-12-14 12:49:51 +11002907 */
2908 if (mddev->private == NULL)
2909 conf = setup_conf(mddev);
2910 else
2911 conf = mddev->private;
2912
2913 if (IS_ERR(conf))
2914 return PTR_ERR(conf);
2915
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11002916 if (mddev->queue)
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07002917 blk_queue_max_write_same_sectors(mddev->queue, 0);
2918
NeilBrowndafb20f2012-03-19 12:46:39 +11002919 rdev_for_each(rdev, mddev) {
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002920 if (!mddev->gendisk)
2921 continue;
NeilBrown709ae482009-12-14 12:49:51 +11002922 disk_stack_limits(mddev->gendisk, rdev->bdev,
2923 rdev->data_offset << 9);
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002924 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
2925 discard_supported = true;
NeilBrown709ae482009-12-14 12:49:51 +11002926 }
2927
2928 mddev->degraded = 0;
2929 for (i=0; i < conf->raid_disks; i++)
2930 if (conf->mirrors[i].rdev == NULL ||
2931 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2932 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2933 mddev->degraded++;
2934
2935 if (conf->raid_disks - mddev->degraded == 1)
2936 mddev->recovery_cp = MaxSector;
2937
Andre Noll8c6ac862009-06-18 08:48:06 +10002938 if (mddev->recovery_cp != MaxSector)
NeilBrown1d41c212016-11-02 14:16:50 +11002939 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n",
2940 mdname(mddev));
2941 pr_info("md/raid1:%s: active with %d out of %d mirrors\n",
NeilBrownf72ffdd2014-09-30 14:23:59 +10002942 mdname(mddev), mddev->raid_disks - mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11002944
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 /*
2946 * Ok, everything is just fine now
2947 */
NeilBrown709ae482009-12-14 12:49:51 +11002948 mddev->thread = conf->thread;
2949 conf->thread = NULL;
2950 mddev->private = conf;
2951
Dan Williams1f403622009-03-31 14:59:03 +11002952 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002954 if (mddev->queue) {
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002955 if (discard_supported)
2956 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
2957 mddev->queue);
2958 else
2959 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
2960 mddev->queue);
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002961 }
majianpeng5220ea12012-04-02 09:48:38 +10002962
2963 ret = md_integrity_register(mddev);
NeilBrown5aa61f42014-12-15 12:56:57 +11002964 if (ret) {
2965 md_unregister_thread(&mddev->thread);
NeilBrownafa0f552014-12-15 12:56:58 +11002966 raid1_free(mddev, conf);
NeilBrown5aa61f42014-12-15 12:56:57 +11002967 }
majianpeng5220ea12012-04-02 09:48:38 +10002968 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969}
2970
NeilBrownafa0f552014-12-15 12:56:58 +11002971static void raid1_free(struct mddev *mddev, void *priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972{
NeilBrownafa0f552014-12-15 12:56:58 +11002973 struct r1conf *conf = priv;
NeilBrown4b6d2872005-09-09 16:23:47 -07002974
Julia Lawall644df1a2015-09-13 14:15:10 +02002975 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002976 kfree(conf->mirrors);
Hirokazu Takahashi0fea7ed2013-04-24 11:42:44 +10002977 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002978 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980}
2981
NeilBrownfd01b882011-10-11 16:47:53 +11002982static int raid1_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983{
2984 /* no resync is happening, and there is enough space
2985 * on all devices, so we can resize.
2986 * We need to make sure resync covers any new space.
2987 * If the array is shrinking we should possibly wait until
2988 * any io in the removed space completes, but it hardly seems
2989 * worth it.
2990 */
NeilBrowna4a61252012-05-22 13:55:27 +10002991 sector_t newsize = raid1_size(mddev, sectors, 0);
2992 if (mddev->external_size &&
2993 mddev->array_sectors > newsize)
Dan Williamsb522adc2009-03-31 15:00:31 +11002994 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10002995 if (mddev->bitmap) {
2996 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
2997 if (ret)
2998 return ret;
2999 }
3000 md_set_array_sectors(mddev, newsize);
Andre Nollf233ea52008-07-21 17:05:22 +10003001 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10003002 revalidate_disk(mddev->gendisk);
Dan Williamsb522adc2009-03-31 15:00:31 +11003003 if (sectors > mddev->dev_sectors &&
NeilBrownb0986362011-05-11 15:52:21 +10003004 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11003005 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3007 }
Dan Williamsb522adc2009-03-31 15:00:31 +11003008 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07003009 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 return 0;
3011}
3012
NeilBrownfd01b882011-10-11 16:47:53 +11003013static int raid1_reshape(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014{
3015 /* We need to:
3016 * 1/ resize the r1bio_pool
3017 * 2/ resize conf->mirrors
3018 *
3019 * We allocate a new r1bio_pool if we can.
3020 * Then raise a device barrier and wait until all IO stops.
3021 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07003022 *
3023 * At the same time, we "pack" the devices so that all the missing
3024 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 */
3026 mempool_t *newpool, *oldpool;
3027 struct pool_info *newpoolinfo;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10003028 struct raid1_info *newmirrors;
NeilBrowne8096362011-10-11 16:49:05 +11003029 struct r1conf *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08003030 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07003031 unsigned long flags;
Dan Williamsb5470dc2008-06-27 21:44:04 -07003032 int d, d2, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033
NeilBrown63c70c42006-03-27 01:18:13 -08003034 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10003035 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08003036 mddev->layout != mddev->new_layout ||
3037 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10003038 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08003039 mddev->new_layout = mddev->layout;
3040 mddev->new_level = mddev->level;
3041 return -EINVAL;
3042 }
3043
Goldwyn Rodrigues28c1b9f2015-10-22 16:01:25 +11003044 if (!mddev_is_clustered(mddev)) {
3045 err = md_allow_write(mddev);
3046 if (err)
3047 return err;
3048 }
NeilBrown2a2275d2007-01-26 00:57:11 -08003049
NeilBrown63c70c42006-03-27 01:18:13 -08003050 raid_disks = mddev->raid_disks + mddev->delta_disks;
3051
NeilBrown6ea9c072005-06-21 17:17:09 -07003052 if (raid_disks < conf->raid_disks) {
3053 cnt=0;
3054 for (d= 0; d < conf->raid_disks; d++)
3055 if (conf->mirrors[d].rdev)
3056 cnt++;
3057 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07003059 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060
3061 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3062 if (!newpoolinfo)
3063 return -ENOMEM;
3064 newpoolinfo->mddev = mddev;
NeilBrown8f19ccb2011-12-23 10:17:56 +11003065 newpoolinfo->raid_disks = raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066
3067 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3068 r1bio_pool_free, newpoolinfo);
3069 if (!newpool) {
3070 kfree(newpoolinfo);
3071 return -ENOMEM;
3072 }
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10003073 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
NeilBrown8f19ccb2011-12-23 10:17:56 +11003074 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 if (!newmirrors) {
3076 kfree(newpoolinfo);
3077 mempool_destroy(newpool);
3078 return -ENOMEM;
3079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
NeilBrowne2d59922013-06-12 11:01:22 +10003081 freeze_array(conf, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082
3083 /* ok, everything is stopped */
3084 oldpool = conf->r1bio_pool;
3085 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07003086
NeilBrowna88aa782007-08-22 14:01:53 -07003087 for (d = d2 = 0; d < conf->raid_disks; d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11003088 struct md_rdev *rdev = conf->mirrors[d].rdev;
NeilBrowna88aa782007-08-22 14:01:53 -07003089 if (rdev && rdev->raid_disk != d2) {
Namhyung Kim36fad852011-07-27 11:00:36 +10003090 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna88aa782007-08-22 14:01:53 -07003091 rdev->raid_disk = d2;
Namhyung Kim36fad852011-07-27 11:00:36 +10003092 sysfs_unlink_rdev(mddev, rdev);
3093 if (sysfs_link_rdev(mddev, rdev))
NeilBrown1d41c212016-11-02 14:16:50 +11003094 pr_warn("md/raid1:%s: cannot register rd%d\n",
3095 mdname(mddev), rdev->raid_disk);
NeilBrown6ea9c072005-06-21 17:17:09 -07003096 }
NeilBrowna88aa782007-08-22 14:01:53 -07003097 if (rdev)
3098 newmirrors[d2++].rdev = rdev;
3099 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 kfree(conf->mirrors);
3101 conf->mirrors = newmirrors;
3102 kfree(conf->poolinfo);
3103 conf->poolinfo = newpoolinfo;
3104
NeilBrownc04be0a2006-10-03 01:15:53 -07003105 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07003107 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08003109 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
NeilBrowne2d59922013-06-12 11:01:22 +10003111 unfreeze_array(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
NeilBrown985ca972015-07-06 12:26:57 +10003113 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3115 md_wakeup_thread(mddev->thread);
3116
3117 mempool_destroy(oldpool);
3118 return 0;
3119}
3120
NeilBrownfd01b882011-10-11 16:47:53 +11003121static void raid1_quiesce(struct mddev *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07003122{
NeilBrowne8096362011-10-11 16:49:05 +11003123 struct r1conf *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07003124
3125 switch(state) {
NeilBrown6eef4b22009-12-14 12:49:51 +11003126 case 2: /* wake for suspend */
3127 wake_up(&conf->wait_barrier);
3128 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07003129 case 1:
majianpeng07169fd2013-11-14 15:16:18 +11003130 freeze_array(conf, 0);
NeilBrown36fa3062005-09-09 16:23:45 -07003131 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07003132 case 0:
majianpeng07169fd2013-11-14 15:16:18 +11003133 unfreeze_array(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07003134 break;
3135 }
NeilBrown36fa3062005-09-09 16:23:45 -07003136}
3137
NeilBrownfd01b882011-10-11 16:47:53 +11003138static void *raid1_takeover(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11003139{
3140 /* raid1 can take over:
3141 * raid5 with 2 devices, any layout or chunk size
3142 */
3143 if (mddev->level == 5 && mddev->raid_disks == 2) {
NeilBrowne8096362011-10-11 16:49:05 +11003144 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11003145 mddev->new_level = 1;
3146 mddev->new_layout = 0;
3147 mddev->new_chunk_sectors = 0;
3148 conf = setup_conf(mddev);
3149 if (!IS_ERR(conf))
majianpeng07169fd2013-11-14 15:16:18 +11003150 /* Array must appear to be quiesced */
3151 conf->array_frozen = 1;
NeilBrown709ae482009-12-14 12:49:51 +11003152 return conf;
3153 }
3154 return ERR_PTR(-EINVAL);
3155}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156
NeilBrown84fc4b52011-10-11 16:49:58 +11003157static struct md_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158{
3159 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08003160 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 .owner = THIS_MODULE,
Shaohua Li849674e2016-01-20 13:52:20 -08003162 .make_request = raid1_make_request,
3163 .run = raid1_run,
NeilBrownafa0f552014-12-15 12:56:58 +11003164 .free = raid1_free,
Shaohua Li849674e2016-01-20 13:52:20 -08003165 .status = raid1_status,
3166 .error_handler = raid1_error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 .hot_add_disk = raid1_add_disk,
3168 .hot_remove_disk= raid1_remove_disk,
3169 .spare_active = raid1_spare_active,
Shaohua Li849674e2016-01-20 13:52:20 -08003170 .sync_request = raid1_sync_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07003172 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08003173 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07003174 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11003175 .takeover = raid1_takeover,
NeilBrown5c675f82014-12-15 12:56:56 +11003176 .congested = raid1_congested,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177};
3178
3179static int __init raid_init(void)
3180{
NeilBrown2604b702006-01-06 00:20:36 -08003181 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182}
3183
3184static void raid_exit(void)
3185{
NeilBrown2604b702006-01-06 00:20:36 -08003186 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187}
3188
3189module_init(raid_init);
3190module_exit(raid_exit);
3191MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11003192MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08003194MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08003195MODULE_ALIAS("md-level-1");
NeilBrown34db0cd2011-10-11 16:50:01 +11003196
3197module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);