blob: 34e26834ad28bd15ab18a216ed324bf0aab090dc [file] [log] [blame]
Thomas Gleixneraf1a8892019-05-20 19:08:12 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * raid1.c : Multiple Devices driver for Linux
4 *
5 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
6 *
7 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
8 *
9 * RAID-1 management functions.
10 *
11 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
12 *
Jan Engelhardt96de0e22007-10-19 23:21:04 +020013 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
15 *
NeilBrown191ea9b2005-06-21 17:17:23 -070016 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
17 * bitmapped intelligence in resync:
18 *
19 * - bitmap marked during normal i/o
20 * - bitmap used to skip nondirty blocks during sync
21 *
22 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
23 * - persistent bitmap code
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 */
25
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090026#include <linux/slab.h>
Stephen Rothwell25570722008-10-15 09:09:21 +110027#include <linux/delay.h>
NeilBrownbff61972009-03-31 14:33:13 +110028#include <linux/blkdev.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040029#include <linux/module.h>
NeilBrownbff61972009-03-31 14:33:13 +110030#include <linux/seq_file.h>
Christian Dietrich8bda4702011-07-27 11:00:36 +100031#include <linux/ratelimit.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010032
NeilBrown109e3762016-11-18 13:22:04 +110033#include <trace/events/block.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010034
NeilBrown43b2e5d2009-03-31 14:33:13 +110035#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110036#include "raid1.h"
Mike Snitzer935fe092017-10-10 17:02:41 -040037#include "md-bitmap.h"
NeilBrown191ea9b2005-06-21 17:17:23 -070038
Shaohua Li394ed8e2017-01-04 16:10:19 -080039#define UNSUPPORTED_MDDEV_FLAGS \
40 ((1L << MD_HAS_JOURNAL) | \
Artur Paszkiewiczea0213e2017-03-09 09:59:57 +010041 (1L << MD_JOURNAL_CLEAN) | \
Pawel Baldysiakddc08822017-08-16 17:13:45 +020042 (1L << MD_HAS_PPL) | \
43 (1L << MD_HAS_MULTIPLE_PPLS))
Shaohua Li394ed8e2017-01-04 16:10:19 -080044
colyli@suse.defd768632017-02-18 03:05:56 +080045static void allow_barrier(struct r1conf *conf, sector_t sector_nr);
46static void lower_barrier(struct r1conf *conf, sector_t sector_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
NeilBrown578b54a2016-11-14 16:30:21 +110048#define raid1_log(md, fmt, args...) \
49 do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0)
50
Ming Leifb0eb5d2017-07-14 16:14:43 +080051#include "raid1-10.c"
52
Guoqing Jiang3e148a32019-06-19 17:30:46 +080053static int check_and_add_wb(struct md_rdev *rdev, sector_t lo, sector_t hi)
54{
55 struct wb_info *wi, *temp_wi;
56 unsigned long flags;
57 int ret = 0;
58 struct mddev *mddev = rdev->mddev;
59
60 wi = mempool_alloc(mddev->wb_info_pool, GFP_NOIO);
61
62 spin_lock_irqsave(&rdev->wb_list_lock, flags);
63 list_for_each_entry(temp_wi, &rdev->wb_list, list) {
64 /* collision happened */
65 if (hi > temp_wi->lo && lo < temp_wi->hi) {
66 ret = -EBUSY;
67 break;
68 }
69 }
70
71 if (!ret) {
72 wi->lo = lo;
73 wi->hi = hi;
74 list_add(&wi->list, &rdev->wb_list);
75 } else
76 mempool_free(wi, mddev->wb_info_pool);
77 spin_unlock_irqrestore(&rdev->wb_list_lock, flags);
78
79 return ret;
80}
81
82static void remove_wb(struct md_rdev *rdev, sector_t lo, sector_t hi)
83{
84 struct wb_info *wi;
85 unsigned long flags;
86 int found = 0;
87 struct mddev *mddev = rdev->mddev;
88
89 spin_lock_irqsave(&rdev->wb_list_lock, flags);
90 list_for_each_entry(wi, &rdev->wb_list, list)
91 if (hi == wi->hi && lo == wi->lo) {
92 list_del(&wi->list);
93 mempool_free(wi, mddev->wb_info_pool);
94 found = 1;
95 break;
96 }
97
98 if (!found)
Dan Carpenter16d4b742019-06-26 12:42:51 +030099 WARN(1, "The write behind IO is not recorded\n");
Guoqing Jiang3e148a32019-06-19 17:30:46 +0800100 spin_unlock_irqrestore(&rdev->wb_list_lock, flags);
101 wake_up(&rdev->wb_io_wait);
102}
103
Ming Lei98d30c52017-03-17 00:12:26 +0800104/*
Ming Lei98d30c52017-03-17 00:12:26 +0800105 * for resync bio, r1bio pointer can be retrieved from the per-bio
106 * 'struct resync_pages'.
107 */
108static inline struct r1bio *get_resync_r1bio(struct bio *bio)
109{
110 return get_resync_pages(bio)->raid_bio;
111}
112
Al Virodd0fc662005-10-07 07:46:04 +0100113static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
115 struct pool_info *pi = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100116 int size = offsetof(struct r1bio, bios[pi->raid_disks]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
118 /* allocate a r1bio with room for raid_disks entries in the bios array */
Jens Axboe7eaceac2011-03-10 08:52:07 +0100119 return kzalloc(size, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120}
121
majianpeng8e005f72013-11-14 15:16:18 +1100122#define RESYNC_DEPTH 32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
majianpeng8e005f72013-11-14 15:16:18 +1100124#define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
125#define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +1000126#define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)
127#define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Al Virodd0fc662005-10-07 07:46:04 +0100129static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130{
131 struct pool_info *pi = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100132 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 struct bio *bio;
NeilBrownda1aab32014-04-09 12:25:43 +1000134 int need_pages;
Ming Lei98d30c52017-03-17 00:12:26 +0800135 int j;
136 struct resync_pages *rps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100139 if (!r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Kees Cook6da2ec52018-06-12 13:55:00 -0700142 rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages),
143 gfp_flags);
Ming Lei98d30c52017-03-17 00:12:26 +0800144 if (!rps)
145 goto out_free_r1bio;
146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 /*
148 * Allocate bios : 1 for reading, n-1 for writing
149 */
150 for (j = pi->raid_disks ; j-- ; ) {
NeilBrown67465572010-10-26 17:33:54 +1100151 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 if (!bio)
153 goto out_free_bio;
154 r1_bio->bios[j] = bio;
155 }
156 /*
157 * Allocate RESYNC_PAGES data pages and attach them to
NeilBrownd11c1712006-01-06 00:20:26 -0800158 * the first bio.
159 * If this is a user-requested check/repair, allocate
160 * RESYNC_PAGES for each bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 */
NeilBrownd11c1712006-01-06 00:20:26 -0800162 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
NeilBrownda1aab32014-04-09 12:25:43 +1000163 need_pages = pi->raid_disks;
NeilBrownd11c1712006-01-06 00:20:26 -0800164 else
NeilBrownda1aab32014-04-09 12:25:43 +1000165 need_pages = 1;
Ming Lei98d30c52017-03-17 00:12:26 +0800166 for (j = 0; j < pi->raid_disks; j++) {
167 struct resync_pages *rp = &rps[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Ming Lei98d30c52017-03-17 00:12:26 +0800169 bio = r1_bio->bios[j];
170
171 if (j < need_pages) {
172 if (resync_alloc_pages(rp, gfp_flags))
173 goto out_free_pages;
174 } else {
175 memcpy(rp, &rps[0], sizeof(*rp));
176 resync_get_all_pages(rp);
177 }
178
Ming Lei98d30c52017-03-17 00:12:26 +0800179 rp->raid_bio = r1_bio;
180 bio->bi_private = rp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 }
182
183 r1_bio->master_bio = NULL;
184
185 return r1_bio;
186
NeilBrownda1aab32014-04-09 12:25:43 +1000187out_free_pages:
Guoqing Jiang491221f2016-09-22 03:10:01 -0400188 while (--j >= 0)
Ming Lei98d30c52017-03-17 00:12:26 +0800189 resync_free_pages(&rps[j]);
NeilBrownda1aab32014-04-09 12:25:43 +1000190
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191out_free_bio:
NeilBrown8f19ccb2011-12-23 10:17:56 +1100192 while (++j < pi->raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 bio_put(r1_bio->bios[j]);
Ming Lei98d30c52017-03-17 00:12:26 +0800194 kfree(rps);
195
196out_free_r1bio:
Marcos Paulo de Souzac7afa802019-06-14 15:41:10 -0700197 rbio_pool_free(r1_bio, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 return NULL;
199}
200
201static void r1buf_pool_free(void *__r1_bio, void *data)
202{
203 struct pool_info *pi = data;
Ming Lei98d30c52017-03-17 00:12:26 +0800204 int i;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100205 struct r1bio *r1bio = __r1_bio;
Ming Lei98d30c52017-03-17 00:12:26 +0800206 struct resync_pages *rp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Ming Lei98d30c52017-03-17 00:12:26 +0800208 for (i = pi->raid_disks; i--; ) {
209 rp = get_resync_pages(r1bio->bios[i]);
210 resync_free_pages(rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 bio_put(r1bio->bios[i]);
Ming Lei98d30c52017-03-17 00:12:26 +0800212 }
213
214 /* resync pages array stored in the 1st bio's .bi_private */
215 kfree(rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Marcos Paulo de Souzac7afa802019-06-14 15:41:10 -0700217 rbio_pool_free(r1bio, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
219
NeilBrowne8096362011-10-11 16:49:05 +1100220static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
222 int i;
223
NeilBrown8f19ccb2011-12-23 10:17:56 +1100224 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 struct bio **bio = r1_bio->bios + i;
NeilBrown4367af52011-07-28 11:31:49 +1000226 if (!BIO_SPECIAL(*bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 bio_put(*bio);
228 *bio = NULL;
229 }
230}
231
NeilBrown9f2c9d12011-10-11 16:48:43 +1100232static void free_r1bio(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
NeilBrowne8096362011-10-11 16:49:05 +1100234 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 put_all_bios(conf, r1_bio);
Kent Overstreetafeee512018-05-20 18:25:52 -0400237 mempool_free(r1_bio, &conf->r1bio_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
NeilBrown9f2c9d12011-10-11 16:48:43 +1100240static void put_buf(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
NeilBrowne8096362011-10-11 16:49:05 +1100242 struct r1conf *conf = r1_bio->mddev->private;
Shaohua Liaf5f42a2017-02-19 22:41:27 -0800243 sector_t sect = r1_bio->sector;
NeilBrown3e198f72006-01-06 00:20:21 -0800244 int i;
245
NeilBrown8f19ccb2011-12-23 10:17:56 +1100246 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3e198f72006-01-06 00:20:21 -0800247 struct bio *bio = r1_bio->bios[i];
248 if (bio->bi_end_io)
249 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
Kent Overstreetafeee512018-05-20 18:25:52 -0400252 mempool_free(r1_bio, &conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Shaohua Liaf5f42a2017-02-19 22:41:27 -0800254 lower_barrier(conf, sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
256
NeilBrown9f2c9d12011-10-11 16:48:43 +1100257static void reschedule_retry(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
259 unsigned long flags;
NeilBrownfd01b882011-10-11 16:47:53 +1100260 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +1100261 struct r1conf *conf = mddev->private;
colyli@suse.defd768632017-02-18 03:05:56 +0800262 int idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
colyli@suse.defd768632017-02-18 03:05:56 +0800264 idx = sector_to_idx(r1_bio->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 spin_lock_irqsave(&conf->device_lock, flags);
266 list_add(&r1_bio->retry_list, &conf->retry_list);
colyli@suse.de824e47d2017-02-18 03:05:57 +0800267 atomic_inc(&conf->nr_queued[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 spin_unlock_irqrestore(&conf->device_lock, flags);
269
NeilBrown17999be2006-01-06 00:20:12 -0800270 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 md_wakeup_thread(mddev->thread);
272}
273
274/*
275 * raid_end_bio_io() is called when we have finished servicing a mirrored
276 * operation and are ready to return a success/failure code to the buffer
277 * cache layer.
278 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100279static void call_bio_endio(struct r1bio *r1_bio)
NeilBrownd2eb35a2011-07-28 11:31:48 +1000280{
281 struct bio *bio = r1_bio->master_bio;
NeilBrowne8096362011-10-11 16:49:05 +1100282 struct r1conf *conf = r1_bio->mddev->private;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000283
284 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200285 bio->bi_status = BLK_STS_IOERR;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200286
NeilBrown37011e32017-03-15 14:05:14 +1100287 bio_endio(bio);
288 /*
289 * Wake up any possible resync thread that waits for the device
290 * to go idle.
291 */
292 allow_barrier(conf, r1_bio->sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000293}
294
NeilBrown9f2c9d12011-10-11 16:48:43 +1100295static void raid_end_bio_io(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
297 struct bio *bio = r1_bio->master_bio;
298
NeilBrown4b6d2872005-09-09 16:23:47 -0700299 /* if nobody has done the final endio yet, do it now */
300 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
NeilBrown36a4e1f2011-10-07 14:23:17 +1100301 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
302 (bio_data_dir(bio) == WRITE) ? "write" : "read",
Kent Overstreet4f024f32013-10-11 15:44:27 -0700303 (unsigned long long) bio->bi_iter.bi_sector,
304 (unsigned long long) bio_end_sector(bio) - 1);
NeilBrown4b6d2872005-09-09 16:23:47 -0700305
NeilBrownd2eb35a2011-07-28 11:31:48 +1000306 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 free_r1bio(r1_bio);
309}
310
311/*
312 * Update disk head position estimator based on IRQ completion info.
313 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100314static inline void update_head_pos(int disk, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
NeilBrowne8096362011-10-11 16:49:05 +1100316 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
318 conf->mirrors[disk].head_position =
319 r1_bio->sector + (r1_bio->sectors);
320}
321
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100322/*
323 * Find the disk number which triggered given bio
324 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100325static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100326{
327 int mirror;
NeilBrown30194632011-12-23 10:17:56 +1100328 struct r1conf *conf = r1_bio->mddev->private;
329 int raid_disks = conf->raid_disks;
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100330
NeilBrown8f19ccb2011-12-23 10:17:56 +1100331 for (mirror = 0; mirror < raid_disks * 2; mirror++)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100332 if (r1_bio->bios[mirror] == bio)
333 break;
334
NeilBrown8f19ccb2011-12-23 10:17:56 +1100335 BUG_ON(mirror == raid_disks * 2);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100336 update_head_pos(mirror, r1_bio);
337
338 return mirror;
339}
340
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200341static void raid1_end_read_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200343 int uptodate = !bio->bi_status;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100344 struct r1bio *r1_bio = bio->bi_private;
NeilBrowne8096362011-10-11 16:49:05 +1100345 struct r1conf *conf = r1_bio->mddev->private;
NeilBrowne5872d52016-06-02 16:19:52 +1000346 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 /*
349 * this branch is our 'one mirror IO has finished' event handler:
350 */
NeilBrowne5872d52016-06-02 16:19:52 +1000351 update_head_pos(r1_bio->read_disk, r1_bio);
NeilBrownddaf22a2006-01-06 00:20:19 -0800352
NeilBrowndd00a992007-05-10 03:15:50 -0700353 if (uptodate)
354 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrown2e52d442016-11-18 16:16:12 +1100355 else if (test_bit(FailFast, &rdev->flags) &&
356 test_bit(R1BIO_FailFast, &r1_bio->state))
357 /* This was a fail-fast read so we definitely
358 * want to retry */
359 ;
NeilBrowndd00a992007-05-10 03:15:50 -0700360 else {
361 /* If all other devices have failed, we want to return
362 * the error upwards rather than fail the last device.
363 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 */
NeilBrowndd00a992007-05-10 03:15:50 -0700365 unsigned long flags;
366 spin_lock_irqsave(&conf->device_lock, flags);
367 if (r1_bio->mddev->degraded == conf->raid_disks ||
368 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
NeilBrowne5872d52016-06-02 16:19:52 +1000369 test_bit(In_sync, &rdev->flags)))
NeilBrowndd00a992007-05-10 03:15:50 -0700370 uptodate = 1;
371 spin_unlock_irqrestore(&conf->device_lock, flags);
372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100374 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 raid_end_bio_io(r1_bio);
NeilBrowne5872d52016-06-02 16:19:52 +1000376 rdev_dec_pending(rdev, conf->mddev);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100377 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 /*
379 * oops, read error:
380 */
381 char b[BDEVNAME_SIZE];
NeilBrown1d41c212016-11-02 14:16:50 +1100382 pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n",
383 mdname(conf->mddev),
384 bdevname(rdev->bdev, b),
385 (unsigned long long)r1_bio->sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000386 set_bit(R1BIO_ReadError, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 reschedule_retry(r1_bio);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100388 /* don't drop the reference on read_disk yet */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391
NeilBrown9f2c9d12011-10-11 16:48:43 +1100392static void close_write(struct r1bio *r1_bio)
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000393{
394 /* it really is the end of this request */
395 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
Ming Lei841c1312017-03-17 00:12:31 +0800396 bio_free_pages(r1_bio->behind_master_bio);
397 bio_put(r1_bio->behind_master_bio);
398 r1_bio->behind_master_bio = NULL;
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000399 }
400 /* clear the bitmap if all writes complete successfully */
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700401 md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
402 r1_bio->sectors,
403 !test_bit(R1BIO_Degraded, &r1_bio->state),
404 test_bit(R1BIO_BehindIO, &r1_bio->state));
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000405 md_write_end(r1_bio->mddev);
406}
407
NeilBrown9f2c9d12011-10-11 16:48:43 +1100408static void r1_bio_write_done(struct r1bio *r1_bio)
NeilBrown4e780642010-10-19 12:54:01 +1100409{
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000410 if (!atomic_dec_and_test(&r1_bio->remaining))
411 return;
412
413 if (test_bit(R1BIO_WriteError, &r1_bio->state))
414 reschedule_retry(r1_bio);
415 else {
416 close_write(r1_bio);
NeilBrown4367af52011-07-28 11:31:49 +1000417 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
418 reschedule_retry(r1_bio);
419 else
420 raid_end_bio_io(r1_bio);
NeilBrown4e780642010-10-19 12:54:01 +1100421 }
422}
423
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200424static void raid1_end_write_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
NeilBrown9f2c9d12011-10-11 16:48:43 +1100426 struct r1bio *r1_bio = bio->bi_private;
NeilBrowne5872d52016-06-02 16:19:52 +1000427 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
NeilBrowne8096362011-10-11 16:49:05 +1100428 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown04b857f2006-03-09 17:33:46 -0800429 struct bio *to_put = NULL;
NeilBrowne5872d52016-06-02 16:19:52 +1000430 int mirror = find_bio_disk(r1_bio, bio);
431 struct md_rdev *rdev = conf->mirrors[mirror].rdev;
Shaohua Lie3f948c2016-10-06 14:09:16 -0700432 bool discard_error;
433
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200434 discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Tejun Heoe9c74692010-09-03 11:56:18 +0200436 /*
437 * 'one mirror IO has finished' event handler:
438 */
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200439 if (bio->bi_status && !discard_error) {
NeilBrowne5872d52016-06-02 16:19:52 +1000440 set_bit(WriteErrorSeen, &rdev->flags);
441 if (!test_and_set_bit(WantReplacement, &rdev->flags))
NeilBrown19d67162011-12-23 10:17:57 +1100442 set_bit(MD_RECOVERY_NEEDED, &
443 conf->mddev->recovery);
444
NeilBrown212e7eb2016-11-18 16:16:12 +1100445 if (test_bit(FailFast, &rdev->flags) &&
446 (bio->bi_opf & MD_FAILFAST) &&
447 /* We never try FailFast to WriteMostly devices */
448 !test_bit(WriteMostly, &rdev->flags)) {
449 md_error(r1_bio->mddev, rdev);
450 if (!test_bit(Faulty, &rdev->flags))
451 /* This is the only remaining device,
452 * We need to retry the write without
453 * FailFast
454 */
455 set_bit(R1BIO_WriteError, &r1_bio->state);
456 else {
457 /* Finished with this branch */
458 r1_bio->bios[mirror] = NULL;
459 to_put = bio;
460 }
461 } else
462 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +1000463 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 /*
Tejun Heoe9c74692010-09-03 11:56:18 +0200465 * Set R1BIO_Uptodate in our master bio, so that we
466 * will return a good error code for to the higher
467 * levels even if IO on some other mirrored buffer
468 * fails.
469 *
470 * The 'master' represents the composite IO operation
471 * to user-side. So if something waits for IO, then it
472 * will wait for the 'master' bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 */
NeilBrown4367af52011-07-28 11:31:49 +1000474 sector_t first_bad;
475 int bad_sectors;
476
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000477 r1_bio->bios[mirror] = NULL;
478 to_put = bio;
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300479 /*
480 * Do not set R1BIO_Uptodate if the current device is
481 * rebuilding or Faulty. This is because we cannot use
482 * such device for properly reading the data back (we could
483 * potentially use it, if the current write would have felt
484 * before rdev->recovery_offset, but for simplicity we don't
485 * check this here.
486 */
NeilBrowne5872d52016-06-02 16:19:52 +1000487 if (test_bit(In_sync, &rdev->flags) &&
488 !test_bit(Faulty, &rdev->flags))
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300489 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
NeilBrown4367af52011-07-28 11:31:49 +1000491 /* Maybe we can clear some bad blocks. */
NeilBrowne5872d52016-06-02 16:19:52 +1000492 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
Shaohua Lie3f948c2016-10-06 14:09:16 -0700493 &first_bad, &bad_sectors) && !discard_error) {
NeilBrown4367af52011-07-28 11:31:49 +1000494 r1_bio->bios[mirror] = IO_MADE_GOOD;
495 set_bit(R1BIO_MadeGood, &r1_bio->state);
496 }
497 }
498
Tejun Heoe9c74692010-09-03 11:56:18 +0200499 if (behind) {
Guoqing Jiang3e148a32019-06-19 17:30:46 +0800500 if (test_bit(WBCollisionCheck, &rdev->flags)) {
501 sector_t lo = r1_bio->sector;
502 sector_t hi = r1_bio->sector + r1_bio->sectors;
503
504 remove_wb(rdev, lo, hi);
505 }
NeilBrowne5872d52016-06-02 16:19:52 +1000506 if (test_bit(WriteMostly, &rdev->flags))
Tejun Heoe9c74692010-09-03 11:56:18 +0200507 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700508
Tejun Heoe9c74692010-09-03 11:56:18 +0200509 /*
510 * In behind mode, we ACK the master bio once the I/O
511 * has safely reached all non-writemostly
512 * disks. Setting the Returned bit ensures that this
513 * gets done only once -- we don't ever want to return
514 * -EIO here, instead we'll wait
515 */
516 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
517 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
518 /* Maybe we can return now */
519 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
520 struct bio *mbio = r1_bio->master_bio;
NeilBrown36a4e1f2011-10-07 14:23:17 +1100521 pr_debug("raid1: behind end write sectors"
522 " %llu-%llu\n",
Kent Overstreet4f024f32013-10-11 15:44:27 -0700523 (unsigned long long) mbio->bi_iter.bi_sector,
524 (unsigned long long) bio_end_sector(mbio) - 1);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000525 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700526 }
527 }
528 }
NeilBrown4367af52011-07-28 11:31:49 +1000529 if (r1_bio->bios[mirror] == NULL)
NeilBrowne5872d52016-06-02 16:19:52 +1000530 rdev_dec_pending(rdev, conf->mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 * Let's see if all mirrored write operations have finished
534 * already.
535 */
NeilBrownaf6d7b72011-05-11 14:51:19 +1000536 r1_bio_write_done(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700537
NeilBrown04b857f2006-03-09 17:33:46 -0800538 if (to_put)
539 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
541
colyli@suse.defd768632017-02-18 03:05:56 +0800542static sector_t align_to_barrier_unit_end(sector_t start_sector,
543 sector_t sectors)
544{
545 sector_t len;
546
547 WARN_ON(sectors == 0);
548 /*
549 * len is the number of sectors from start_sector to end of the
550 * barrier unit which start_sector belongs to.
551 */
552 len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) -
553 start_sector;
554
555 if (len > sectors)
556 len = sectors;
557
558 return len;
559}
560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561/*
562 * This routine returns the disk from which the requested read should
563 * be done. There is a per-array 'next expected sequential IO' sector
564 * number - if this matches on the next IO then we use the last disk.
565 * There is also a per-disk 'last know head position' sector that is
566 * maintained from IRQ contexts, both the normal and the resync IO
567 * completion handlers update this position correctly. If there is no
568 * perfect sequential match then we pick the disk whose head is closest.
569 *
570 * If there are 2 mirrors in the same 2 devices, performance degrades
571 * because position is mirror, not device based.
572 *
573 * The rdev for the device selected will have nr_pending incremented.
574 */
NeilBrowne8096362011-10-11 16:49:05 +1100575static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000577 const sector_t this_sector = r1_bio->sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000578 int sectors;
579 int best_good_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000580 int best_disk, best_dist_disk, best_pending_disk;
581 int has_nonrot_disk;
Shaohua Libe4d3282012-07-31 10:03:53 +1000582 int disk;
NeilBrown76073052011-05-11 14:34:56 +1000583 sector_t best_dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000584 unsigned int min_pending;
NeilBrown3cb03002011-10-11 16:45:26 +1100585 struct md_rdev *rdev;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000586 int choose_first;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000587 int choose_next_idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589 rcu_read_lock();
590 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700591 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 * device if no resync is going on, or below the resync window.
593 * We take the first readable disk when above the resync window.
594 */
595 retry:
NeilBrownd2eb35a2011-07-28 11:31:48 +1000596 sectors = r1_bio->sectors;
NeilBrown76073052011-05-11 14:34:56 +1000597 best_disk = -1;
Shaohua Li9dedf602012-07-31 10:03:53 +1000598 best_dist_disk = -1;
NeilBrown76073052011-05-11 14:34:56 +1000599 best_dist = MaxSector;
Shaohua Li9dedf602012-07-31 10:03:53 +1000600 best_pending_disk = -1;
601 min_pending = UINT_MAX;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000602 best_good_sectors = 0;
Shaohua Li9dedf602012-07-31 10:03:53 +1000603 has_nonrot_disk = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000604 choose_next_idle = 0;
NeilBrown2e52d442016-11-18 16:16:12 +1100605 clear_bit(R1BIO_FailFast, &r1_bio->state);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000606
Goldwyn Rodrigues7d49ffc2014-08-12 10:13:19 -0500607 if ((conf->mddev->recovery_cp < this_sector + sectors) ||
608 (mddev_is_clustered(conf->mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -0500609 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
Goldwyn Rodrigues7d49ffc2014-08-12 10:13:19 -0500610 this_sector + sectors)))
611 choose_first = 1;
612 else
613 choose_first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Shaohua Libe4d3282012-07-31 10:03:53 +1000615 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
NeilBrown76073052011-05-11 14:34:56 +1000616 sector_t dist;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000617 sector_t first_bad;
618 int bad_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000619 unsigned int pending;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000620 bool nonrot;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000621
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000622 rdev = rcu_dereference(conf->mirrors[disk].rdev);
623 if (r1_bio->bios[disk] == IO_BLOCKED
624 || rdev == NULL
NeilBrown76073052011-05-11 14:34:56 +1000625 || test_bit(Faulty, &rdev->flags))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000626 continue;
NeilBrown76073052011-05-11 14:34:56 +1000627 if (!test_bit(In_sync, &rdev->flags) &&
628 rdev->recovery_offset < this_sector + sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 continue;
NeilBrown76073052011-05-11 14:34:56 +1000630 if (test_bit(WriteMostly, &rdev->flags)) {
631 /* Don't balance among write-mostly, just
632 * use the first as a last resort */
Tomáš Hodekd1901ef2015-02-23 11:00:38 +1100633 if (best_dist_disk < 0) {
NeilBrown307729c2012-01-09 01:41:51 +1100634 if (is_badblock(rdev, this_sector, sectors,
635 &first_bad, &bad_sectors)) {
Wei Fang816b0ac2016-03-21 19:18:32 +0800636 if (first_bad <= this_sector)
NeilBrown307729c2012-01-09 01:41:51 +1100637 /* Cannot use this */
638 continue;
639 best_good_sectors = first_bad - this_sector;
640 } else
641 best_good_sectors = sectors;
Tomáš Hodekd1901ef2015-02-23 11:00:38 +1100642 best_dist_disk = disk;
643 best_pending_disk = disk;
NeilBrown307729c2012-01-09 01:41:51 +1100644 }
NeilBrown76073052011-05-11 14:34:56 +1000645 continue;
646 }
647 /* This is a reasonable device to use. It might
648 * even be best.
649 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000650 if (is_badblock(rdev, this_sector, sectors,
651 &first_bad, &bad_sectors)) {
652 if (best_dist < MaxSector)
653 /* already have a better device */
654 continue;
655 if (first_bad <= this_sector) {
656 /* cannot read here. If this is the 'primary'
657 * device, then we must not read beyond
658 * bad_sectors from another device..
659 */
660 bad_sectors -= (this_sector - first_bad);
661 if (choose_first && sectors > bad_sectors)
662 sectors = bad_sectors;
663 if (best_good_sectors > sectors)
664 best_good_sectors = sectors;
665
666 } else {
667 sector_t good_sectors = first_bad - this_sector;
668 if (good_sectors > best_good_sectors) {
669 best_good_sectors = good_sectors;
670 best_disk = disk;
671 }
672 if (choose_first)
673 break;
674 }
675 continue;
Tomasz Majchrzakd82dd0e2017-05-12 14:26:10 +0200676 } else {
677 if ((sectors > best_good_sectors) && (best_disk >= 0))
678 best_disk = -1;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000679 best_good_sectors = sectors;
Tomasz Majchrzakd82dd0e2017-05-12 14:26:10 +0200680 }
NeilBrownd2eb35a2011-07-28 11:31:48 +1000681
NeilBrown2e52d442016-11-18 16:16:12 +1100682 if (best_disk >= 0)
683 /* At least two disks to choose from so failfast is OK */
684 set_bit(R1BIO_FailFast, &r1_bio->state);
685
Shaohua Li12cee5a2012-07-31 10:03:53 +1000686 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
687 has_nonrot_disk |= nonrot;
Shaohua Li9dedf602012-07-31 10:03:53 +1000688 pending = atomic_read(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000689 dist = abs(this_sector - conf->mirrors[disk].head_position);
Shaohua Li12cee5a2012-07-31 10:03:53 +1000690 if (choose_first) {
NeilBrown76073052011-05-11 14:34:56 +1000691 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 break;
693 }
Shaohua Li12cee5a2012-07-31 10:03:53 +1000694 /* Don't change to another disk for sequential reads */
695 if (conf->mirrors[disk].next_seq_sect == this_sector
696 || dist == 0) {
697 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
698 struct raid1_info *mirror = &conf->mirrors[disk];
699
700 best_disk = disk;
701 /*
702 * If buffered sequential IO size exceeds optimal
703 * iosize, check if there is idle disk. If yes, choose
704 * the idle disk. read_balance could already choose an
705 * idle disk before noticing it's a sequential IO in
706 * this disk. This doesn't matter because this disk
707 * will idle, next time it will be utilized after the
708 * first disk has IO size exceeds optimal iosize. In
709 * this way, iosize of the first disk will be optimal
710 * iosize at least. iosize of the second disk might be
711 * small, but not a big deal since when the second disk
712 * starts IO, the first disk is likely still busy.
713 */
714 if (nonrot && opt_iosize > 0 &&
715 mirror->seq_start != MaxSector &&
716 mirror->next_seq_sect > opt_iosize &&
717 mirror->next_seq_sect - opt_iosize >=
718 mirror->seq_start) {
719 choose_next_idle = 1;
720 continue;
721 }
722 break;
723 }
Shaohua Li12cee5a2012-07-31 10:03:53 +1000724
725 if (choose_next_idle)
726 continue;
Shaohua Li9dedf602012-07-31 10:03:53 +1000727
728 if (min_pending > pending) {
729 min_pending = pending;
730 best_pending_disk = disk;
731 }
732
NeilBrown76073052011-05-11 14:34:56 +1000733 if (dist < best_dist) {
734 best_dist = dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000735 best_dist_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
Shaohua Li9dedf602012-07-31 10:03:53 +1000739 /*
740 * If all disks are rotational, choose the closest disk. If any disk is
741 * non-rotational, choose the disk with less pending request even the
742 * disk is rotational, which might/might not be optimal for raids with
743 * mixed ratation/non-rotational disks depending on workload.
744 */
745 if (best_disk == -1) {
NeilBrown2e52d442016-11-18 16:16:12 +1100746 if (has_nonrot_disk || min_pending == 0)
Shaohua Li9dedf602012-07-31 10:03:53 +1000747 best_disk = best_pending_disk;
748 else
749 best_disk = best_dist_disk;
750 }
751
NeilBrown76073052011-05-11 14:34:56 +1000752 if (best_disk >= 0) {
753 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700754 if (!rdev)
755 goto retry;
756 atomic_inc(&rdev->nr_pending);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000757 sectors = best_good_sectors;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000758
759 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
760 conf->mirrors[best_disk].seq_start = this_sector;
761
Shaohua Libe4d3282012-07-31 10:03:53 +1000762 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 }
764 rcu_read_unlock();
NeilBrownd2eb35a2011-07-28 11:31:48 +1000765 *max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
NeilBrown76073052011-05-11 14:34:56 +1000767 return best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768}
769
NeilBrown5c675f82014-12-15 12:56:56 +1100770static int raid1_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700771{
NeilBrowne8096362011-10-11 16:49:05 +1100772 struct r1conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700773 int i, ret = 0;
774
Tejun Heo44522262015-05-22 17:13:26 -0400775 if ((bits & (1 << WB_async_congested)) &&
NeilBrown34db0cd2011-10-11 16:50:01 +1100776 conf->pending_count >= max_queued_requests)
777 return 1;
778
NeilBrown0d129222006-10-03 01:15:54 -0700779 rcu_read_lock();
NeilBrownf53e29f2012-02-13 14:24:05 +1100780 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100781 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700782 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200783 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700784
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500785 BUG_ON(!q);
786
NeilBrown0d129222006-10-03 01:15:54 -0700787 /* Note the '|| 1' - when read_balance prefers
788 * non-congested targets, it can be removed
789 */
Tejun Heo44522262015-05-22 17:13:26 -0400790 if ((bits & (1 << WB_async_congested)) || 1)
Jan Karadc3b17c2017-02-02 15:56:50 +0100791 ret |= bdi_congested(q->backing_dev_info, bits);
NeilBrown0d129222006-10-03 01:15:54 -0700792 else
Jan Karadc3b17c2017-02-02 15:56:50 +0100793 ret &= bdi_congested(q->backing_dev_info, bits);
NeilBrown0d129222006-10-03 01:15:54 -0700794 }
795 }
796 rcu_read_unlock();
797 return ret;
798}
NeilBrown0d129222006-10-03 01:15:54 -0700799
NeilBrown673ca682017-04-05 14:05:51 +1000800static void flush_bio_list(struct r1conf *conf, struct bio *bio)
801{
802 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700803 md_bitmap_unplug(conf->mddev->bitmap);
NeilBrown673ca682017-04-05 14:05:51 +1000804 wake_up(&conf->wait_barrier);
805
806 while (bio) { /* submit pending writes */
807 struct bio *next = bio->bi_next;
Christoph Hellwig74d46992017-08-23 19:10:32 +0200808 struct md_rdev *rdev = (void *)bio->bi_disk;
NeilBrown673ca682017-04-05 14:05:51 +1000809 bio->bi_next = NULL;
Christoph Hellwig74d46992017-08-23 19:10:32 +0200810 bio_set_dev(bio, rdev->bdev);
NeilBrown673ca682017-04-05 14:05:51 +1000811 if (test_bit(Faulty, &rdev->flags)) {
Guoqing Jiang6308d8e2017-07-21 16:33:44 +0800812 bio_io_error(bio);
NeilBrown673ca682017-04-05 14:05:51 +1000813 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
Christoph Hellwig74d46992017-08-23 19:10:32 +0200814 !blk_queue_discard(bio->bi_disk->queue)))
NeilBrown673ca682017-04-05 14:05:51 +1000815 /* Just ignore it */
816 bio_endio(bio);
817 else
818 generic_make_request(bio);
819 bio = next;
820 }
821}
822
NeilBrowne8096362011-10-11 16:49:05 +1100823static void flush_pending_writes(struct r1conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800824{
825 /* Any writes that have been queued but are awaiting
826 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800827 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800828 spin_lock_irq(&conf->device_lock);
829
830 if (conf->pending_bio_list.head) {
Shaohua Li18022a12017-12-01 12:12:34 -0800831 struct blk_plug plug;
NeilBrowna35e63e2008-03-04 14:29:29 -0800832 struct bio *bio;
Shaohua Li18022a12017-12-01 12:12:34 -0800833
NeilBrowna35e63e2008-03-04 14:29:29 -0800834 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100835 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800836 spin_unlock_irq(&conf->device_lock);
NeilBrown474beb52017-12-04 08:21:04 +1100837
838 /*
839 * As this is called in a wait_event() loop (see freeze_array),
840 * current->state might be TASK_UNINTERRUPTIBLE which will
841 * cause a warning when we prepare to wait again. As it is
842 * rare that this path is taken, it is perfectly safe to force
843 * us to go around the wait_event() loop again, so the warning
844 * is a false-positive. Silence the warning by resetting
845 * thread state
846 */
847 __set_current_state(TASK_RUNNING);
Shaohua Li18022a12017-12-01 12:12:34 -0800848 blk_start_plug(&plug);
NeilBrown673ca682017-04-05 14:05:51 +1000849 flush_bio_list(conf, bio);
Shaohua Li18022a12017-12-01 12:12:34 -0800850 blk_finish_plug(&plug);
NeilBrowna35e63e2008-03-04 14:29:29 -0800851 } else
852 spin_unlock_irq(&conf->device_lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100853}
854
NeilBrown17999be2006-01-06 00:20:12 -0800855/* Barriers....
856 * Sometimes we need to suspend IO while we do something else,
857 * either some resync/recovery, or reconfigure the array.
858 * To do this we raise a 'barrier'.
859 * The 'barrier' is a counter that can be raised multiple times
860 * to count how many activities are happening which preclude
861 * normal IO.
862 * We can only raise the barrier if there is no pending IO.
863 * i.e. if nr_pending == 0.
864 * We choose only to raise the barrier if no-one is waiting for the
865 * barrier to go down. This means that as soon as an IO request
866 * is ready, no other operations which require a barrier will start
867 * until the IO request has had a chance.
868 *
869 * So: regular IO calls 'wait_barrier'. When that returns there
870 * is no backgroup IO happening, It must arrange to call
871 * allow_barrier when it has finished its IO.
872 * backgroup IO calls must call raise_barrier. Once that returns
873 * there is no normal IO happeing. It must arrange to call
874 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 */
Yufen Yu8c242592018-04-09 09:50:44 +0800876static sector_t raise_barrier(struct r1conf *conf, sector_t sector_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
colyli@suse.defd768632017-02-18 03:05:56 +0800878 int idx = sector_to_idx(sector_nr);
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800881
882 /* Wait until no block IO is waiting */
colyli@suse.de824e47d2017-02-18 03:05:57 +0800883 wait_event_lock_irq(conf->wait_barrier,
884 !atomic_read(&conf->nr_waiting[idx]),
Lukas Czernereed8c022012-11-30 11:42:40 +0100885 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800886
887 /* block any new IO from starting */
colyli@suse.de824e47d2017-02-18 03:05:57 +0800888 atomic_inc(&conf->barrier[idx]);
889 /*
890 * In raise_barrier() we firstly increase conf->barrier[idx] then
891 * check conf->nr_pending[idx]. In _wait_barrier() we firstly
892 * increase conf->nr_pending[idx] then check conf->barrier[idx].
893 * A memory barrier here to make sure conf->nr_pending[idx] won't
894 * be fetched before conf->barrier[idx] is increased. Otherwise
895 * there will be a race between raise_barrier() and _wait_barrier().
896 */
897 smp_mb__after_atomic();
NeilBrown17999be2006-01-06 00:20:12 -0800898
majianpeng79ef3a82013-11-15 14:55:02 +0800899 /* For these conditions we must wait:
900 * A: while the array is in frozen state
colyli@suse.defd768632017-02-18 03:05:56 +0800901 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O
902 * existing in corresponding I/O barrier bucket.
903 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches
904 * max resync count which allowed on current I/O barrier bucket.
majianpeng79ef3a82013-11-15 14:55:02 +0800905 */
NeilBrown17999be2006-01-06 00:20:12 -0800906 wait_event_lock_irq(conf->wait_barrier,
Yufen Yu8c242592018-04-09 09:50:44 +0800907 (!conf->array_frozen &&
colyli@suse.de824e47d2017-02-18 03:05:57 +0800908 !atomic_read(&conf->nr_pending[idx]) &&
Yufen Yu8c242592018-04-09 09:50:44 +0800909 atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) ||
910 test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery),
Lukas Czernereed8c022012-11-30 11:42:40 +0100911 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800912
Yufen Yu8c242592018-04-09 09:50:44 +0800913 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
914 atomic_dec(&conf->barrier[idx]);
915 spin_unlock_irq(&conf->resync_lock);
916 wake_up(&conf->wait_barrier);
917 return -EINTR;
918 }
919
Xiao Ni43ac9b82017-04-27 16:28:49 +0800920 atomic_inc(&conf->nr_sync_pending);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 spin_unlock_irq(&conf->resync_lock);
Yufen Yu8c242592018-04-09 09:50:44 +0800922
923 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924}
925
colyli@suse.defd768632017-02-18 03:05:56 +0800926static void lower_barrier(struct r1conf *conf, sector_t sector_nr)
NeilBrown17999be2006-01-06 00:20:12 -0800927{
colyli@suse.defd768632017-02-18 03:05:56 +0800928 int idx = sector_to_idx(sector_nr);
929
colyli@suse.de824e47d2017-02-18 03:05:57 +0800930 BUG_ON(atomic_read(&conf->barrier[idx]) <= 0);
colyli@suse.defd768632017-02-18 03:05:56 +0800931
colyli@suse.de824e47d2017-02-18 03:05:57 +0800932 atomic_dec(&conf->barrier[idx]);
Xiao Ni43ac9b82017-04-27 16:28:49 +0800933 atomic_dec(&conf->nr_sync_pending);
NeilBrown17999be2006-01-06 00:20:12 -0800934 wake_up(&conf->wait_barrier);
935}
936
colyli@suse.defd768632017-02-18 03:05:56 +0800937static void _wait_barrier(struct r1conf *conf, int idx)
NeilBrown17999be2006-01-06 00:20:12 -0800938{
colyli@suse.de824e47d2017-02-18 03:05:57 +0800939 /*
940 * We need to increase conf->nr_pending[idx] very early here,
941 * then raise_barrier() can be blocked when it waits for
942 * conf->nr_pending[idx] to be 0. Then we can avoid holding
943 * conf->resync_lock when there is no barrier raised in same
944 * barrier unit bucket. Also if the array is frozen, I/O
945 * should be blocked until array is unfrozen.
946 */
947 atomic_inc(&conf->nr_pending[idx]);
948 /*
949 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then
950 * check conf->barrier[idx]. In raise_barrier() we firstly increase
951 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory
952 * barrier is necessary here to make sure conf->barrier[idx] won't be
953 * fetched before conf->nr_pending[idx] is increased. Otherwise there
954 * will be a race between _wait_barrier() and raise_barrier().
955 */
956 smp_mb__after_atomic();
majianpeng79ef3a82013-11-15 14:55:02 +0800957
colyli@suse.de824e47d2017-02-18 03:05:57 +0800958 /*
959 * Don't worry about checking two atomic_t variables at same time
960 * here. If during we check conf->barrier[idx], the array is
961 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is
962 * 0, it is safe to return and make the I/O continue. Because the
963 * array is frozen, all I/O returned here will eventually complete
964 * or be queued, no race will happen. See code comment in
965 * frozen_array().
966 */
967 if (!READ_ONCE(conf->array_frozen) &&
968 !atomic_read(&conf->barrier[idx]))
969 return;
majianpeng79ef3a82013-11-15 14:55:02 +0800970
colyli@suse.de824e47d2017-02-18 03:05:57 +0800971 /*
972 * After holding conf->resync_lock, conf->nr_pending[idx]
973 * should be decreased before waiting for barrier to drop.
974 * Otherwise, we may encounter a race condition because
975 * raise_barrer() might be waiting for conf->nr_pending[idx]
976 * to be 0 at same time.
977 */
978 spin_lock_irq(&conf->resync_lock);
979 atomic_inc(&conf->nr_waiting[idx]);
980 atomic_dec(&conf->nr_pending[idx]);
981 /*
982 * In case freeze_array() is waiting for
983 * get_unqueued_pending() == extra
984 */
985 wake_up(&conf->wait_barrier);
986 /* Wait for the barrier in same barrier unit bucket to drop. */
987 wait_event_lock_irq(conf->wait_barrier,
988 !conf->array_frozen &&
989 !atomic_read(&conf->barrier[idx]),
990 conf->resync_lock);
991 atomic_inc(&conf->nr_pending[idx]);
992 atomic_dec(&conf->nr_waiting[idx]);
colyli@suse.defd768632017-02-18 03:05:56 +0800993 spin_unlock_irq(&conf->resync_lock);
majianpeng79ef3a82013-11-15 14:55:02 +0800994}
995
colyli@suse.defd768632017-02-18 03:05:56 +0800996static void wait_read_barrier(struct r1conf *conf, sector_t sector_nr)
majianpeng79ef3a82013-11-15 14:55:02 +0800997{
colyli@suse.defd768632017-02-18 03:05:56 +0800998 int idx = sector_to_idx(sector_nr);
majianpeng79ef3a82013-11-15 14:55:02 +0800999
colyli@suse.de824e47d2017-02-18 03:05:57 +08001000 /*
1001 * Very similar to _wait_barrier(). The difference is, for read
1002 * I/O we don't need wait for sync I/O, but if the whole array
1003 * is frozen, the read I/O still has to wait until the array is
1004 * unfrozen. Since there is no ordering requirement with
1005 * conf->barrier[idx] here, memory barrier is unnecessary as well.
1006 */
1007 atomic_inc(&conf->nr_pending[idx]);
majianpeng79ef3a82013-11-15 14:55:02 +08001008
colyli@suse.de824e47d2017-02-18 03:05:57 +08001009 if (!READ_ONCE(conf->array_frozen))
1010 return;
NeilBrown17999be2006-01-06 00:20:12 -08001011
1012 spin_lock_irq(&conf->resync_lock);
colyli@suse.de824e47d2017-02-18 03:05:57 +08001013 atomic_inc(&conf->nr_waiting[idx]);
1014 atomic_dec(&conf->nr_pending[idx]);
1015 /*
1016 * In case freeze_array() is waiting for
1017 * get_unqueued_pending() == extra
1018 */
1019 wake_up(&conf->wait_barrier);
1020 /* Wait for array to be unfrozen */
1021 wait_event_lock_irq(conf->wait_barrier,
1022 !conf->array_frozen,
1023 conf->resync_lock);
1024 atomic_inc(&conf->nr_pending[idx]);
1025 atomic_dec(&conf->nr_waiting[idx]);
NeilBrown17999be2006-01-06 00:20:12 -08001026 spin_unlock_irq(&conf->resync_lock);
1027}
1028
colyli@suse.defd768632017-02-18 03:05:56 +08001029static void wait_barrier(struct r1conf *conf, sector_t sector_nr)
NeilBrown17999be2006-01-06 00:20:12 -08001030{
colyli@suse.defd768632017-02-18 03:05:56 +08001031 int idx = sector_to_idx(sector_nr);
majianpeng79ef3a82013-11-15 14:55:02 +08001032
colyli@suse.defd768632017-02-18 03:05:56 +08001033 _wait_barrier(conf, idx);
1034}
majianpeng79ef3a82013-11-15 14:55:02 +08001035
colyli@suse.defd768632017-02-18 03:05:56 +08001036static void _allow_barrier(struct r1conf *conf, int idx)
NeilBrown17999be2006-01-06 00:20:12 -08001037{
colyli@suse.de824e47d2017-02-18 03:05:57 +08001038 atomic_dec(&conf->nr_pending[idx]);
NeilBrown17999be2006-01-06 00:20:12 -08001039 wake_up(&conf->wait_barrier);
1040}
1041
colyli@suse.defd768632017-02-18 03:05:56 +08001042static void allow_barrier(struct r1conf *conf, sector_t sector_nr)
1043{
1044 int idx = sector_to_idx(sector_nr);
1045
1046 _allow_barrier(conf, idx);
1047}
1048
colyli@suse.defd768632017-02-18 03:05:56 +08001049/* conf->resync_lock should be held */
1050static int get_unqueued_pending(struct r1conf *conf)
1051{
1052 int idx, ret;
1053
Xiao Ni43ac9b82017-04-27 16:28:49 +08001054 ret = atomic_read(&conf->nr_sync_pending);
1055 for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++)
colyli@suse.de824e47d2017-02-18 03:05:57 +08001056 ret += atomic_read(&conf->nr_pending[idx]) -
1057 atomic_read(&conf->nr_queued[idx]);
colyli@suse.defd768632017-02-18 03:05:56 +08001058
1059 return ret;
1060}
1061
NeilBrowne2d59922013-06-12 11:01:22 +10001062static void freeze_array(struct r1conf *conf, int extra)
NeilBrownddaf22a2006-01-06 00:20:19 -08001063{
colyli@suse.defd768632017-02-18 03:05:56 +08001064 /* Stop sync I/O and normal I/O and wait for everything to
Zhilong Liu11353b92017-03-14 15:52:26 +08001065 * go quiet.
colyli@suse.defd768632017-02-18 03:05:56 +08001066 * This is called in two situations:
1067 * 1) management command handlers (reshape, remove disk, quiesce).
1068 * 2) one normal I/O request failed.
1069
1070 * After array_frozen is set to 1, new sync IO will be blocked at
1071 * raise_barrier(), and new normal I/O will blocked at _wait_barrier()
1072 * or wait_read_barrier(). The flying I/Os will either complete or be
1073 * queued. When everything goes quite, there are only queued I/Os left.
1074
1075 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the
1076 * barrier bucket index which this I/O request hits. When all sync and
1077 * normal I/O are queued, sum of all conf->nr_pending[] will match sum
1078 * of all conf->nr_queued[]. But normal I/O failure is an exception,
1079 * in handle_read_error(), we may call freeze_array() before trying to
1080 * fix the read error. In this case, the error read I/O is not queued,
1081 * so get_unqueued_pending() == 1.
1082 *
1083 * Therefore before this function returns, we need to wait until
1084 * get_unqueued_pendings(conf) gets equal to extra. For
1085 * normal I/O context, extra is 1, in rested situations extra is 0.
NeilBrownddaf22a2006-01-06 00:20:19 -08001086 */
1087 spin_lock_irq(&conf->resync_lock);
majianpengb364e3d2013-11-14 15:16:18 +11001088 conf->array_frozen = 1;
NeilBrown578b54a2016-11-14 16:30:21 +11001089 raid1_log(conf->mddev, "wait freeze");
colyli@suse.defd768632017-02-18 03:05:56 +08001090 wait_event_lock_irq_cmd(
1091 conf->wait_barrier,
1092 get_unqueued_pending(conf) == extra,
1093 conf->resync_lock,
1094 flush_pending_writes(conf));
NeilBrownddaf22a2006-01-06 00:20:19 -08001095 spin_unlock_irq(&conf->resync_lock);
1096}
NeilBrowne8096362011-10-11 16:49:05 +11001097static void unfreeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -08001098{
1099 /* reverse the effect of the freeze */
1100 spin_lock_irq(&conf->resync_lock);
majianpengb364e3d2013-11-14 15:16:18 +11001101 conf->array_frozen = 0;
NeilBrownddaf22a2006-01-06 00:20:19 -08001102 spin_unlock_irq(&conf->resync_lock);
colyli@suse.de824e47d2017-02-18 03:05:57 +08001103 wake_up(&conf->wait_barrier);
NeilBrownddaf22a2006-01-06 00:20:19 -08001104}
1105
Shaohua Li16d56e22017-07-17 14:33:48 -07001106static void alloc_behind_master_bio(struct r1bio *r1_bio,
NeilBrowncb83efc2017-04-05 14:05:50 +10001107 struct bio *bio)
NeilBrown4b6d2872005-09-09 16:23:47 -07001108{
NeilBrowncb83efc2017-04-05 14:05:50 +10001109 int size = bio->bi_iter.bi_size;
Ming Lei841c1312017-03-17 00:12:31 +08001110 unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1111 int i = 0;
1112 struct bio *behind_bio = NULL;
NeilBrown4b6d2872005-09-09 16:23:47 -07001113
Ming Lei841c1312017-03-17 00:12:31 +08001114 behind_bio = bio_alloc_mddev(GFP_NOIO, vcnt, r1_bio->mddev);
1115 if (!behind_bio)
Shaohua Li16d56e22017-07-17 14:33:48 -07001116 return;
Ming Lei841c1312017-03-17 00:12:31 +08001117
Shaohua Li41743c12017-03-24 15:20:47 -07001118 /* discard op, we don't support writezero/writesame yet */
Shaohua Li16d56e22017-07-17 14:33:48 -07001119 if (!bio_has_data(bio)) {
1120 behind_bio->bi_iter.bi_size = size;
Shaohua Li41743c12017-03-24 15:20:47 -07001121 goto skip_copy;
Shaohua Li16d56e22017-07-17 14:33:48 -07001122 }
Shaohua Li41743c12017-03-24 15:20:47 -07001123
Mariusz Dabrowskidba40d42018-02-14 14:23:30 +01001124 behind_bio->bi_write_hint = bio->bi_write_hint;
1125
Ming Lei841c1312017-03-17 00:12:31 +08001126 while (i < vcnt && size) {
1127 struct page *page;
1128 int len = min_t(int, PAGE_SIZE, size);
1129
1130 page = alloc_page(GFP_NOIO);
1131 if (unlikely(!page))
1132 goto free_pages;
1133
1134 bio_add_page(behind_bio, page, len, 0);
1135
1136 size -= len;
1137 i++;
NeilBrown4b6d2872005-09-09 16:23:47 -07001138 }
NeilBrown4b6d2872005-09-09 16:23:47 -07001139
NeilBrowncb83efc2017-04-05 14:05:50 +10001140 bio_copy_data(behind_bio, bio);
Shaohua Li41743c12017-03-24 15:20:47 -07001141skip_copy:
Luis de Bethencourt56a64c12018-01-17 13:38:02 +00001142 r1_bio->behind_master_bio = behind_bio;
Ming Lei841c1312017-03-17 00:12:31 +08001143 set_bit(R1BIO_BehindIO, &r1_bio->state);
1144
Shaohua Li16d56e22017-07-17 14:33:48 -07001145 return;
Ming Lei841c1312017-03-17 00:12:31 +08001146
1147free_pages:
Kent Overstreet4f024f32013-10-11 15:44:27 -07001148 pr_debug("%dB behind alloc failed, doing sync I/O\n",
1149 bio->bi_iter.bi_size);
Ming Lei841c1312017-03-17 00:12:31 +08001150 bio_free_pages(behind_bio);
Shaohua Li16d56e22017-07-17 14:33:48 -07001151 bio_put(behind_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -07001152}
1153
NeilBrownf54a9d02012-08-02 08:33:20 +10001154struct raid1_plug_cb {
1155 struct blk_plug_cb cb;
1156 struct bio_list pending;
1157 int pending_cnt;
1158};
1159
1160static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
1161{
1162 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1163 cb);
1164 struct mddev *mddev = plug->cb.data;
1165 struct r1conf *conf = mddev->private;
1166 struct bio *bio;
1167
NeilBrown874807a2012-11-27 12:14:40 +11001168 if (from_schedule || current->bio_list) {
NeilBrownf54a9d02012-08-02 08:33:20 +10001169 spin_lock_irq(&conf->device_lock);
1170 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1171 conf->pending_count += plug->pending_cnt;
1172 spin_unlock_irq(&conf->device_lock);
NeilBrownee0b0242013-02-25 12:38:29 +11001173 wake_up(&conf->wait_barrier);
NeilBrownf54a9d02012-08-02 08:33:20 +10001174 md_wakeup_thread(mddev->thread);
1175 kfree(plug);
1176 return;
1177 }
1178
1179 /* we aren't scheduling, so we can do the write-out directly. */
1180 bio = bio_list_get(&plug->pending);
NeilBrown673ca682017-04-05 14:05:51 +10001181 flush_bio_list(conf, bio);
NeilBrownf54a9d02012-08-02 08:33:20 +10001182 kfree(plug);
1183}
1184
NeilBrown689389a2017-04-05 14:05:50 +10001185static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio)
1186{
1187 r1_bio->master_bio = bio;
1188 r1_bio->sectors = bio_sectors(bio);
1189 r1_bio->state = 0;
1190 r1_bio->mddev = mddev;
1191 r1_bio->sector = bio->bi_iter.bi_sector;
1192}
1193
colyli@suse.defd768632017-02-18 03:05:56 +08001194static inline struct r1bio *
NeilBrown689389a2017-04-05 14:05:50 +10001195alloc_r1bio(struct mddev *mddev, struct bio *bio)
colyli@suse.defd768632017-02-18 03:05:56 +08001196{
1197 struct r1conf *conf = mddev->private;
1198 struct r1bio *r1_bio;
1199
Kent Overstreetafeee512018-05-20 18:25:52 -04001200 r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO);
NeilBrown689389a2017-04-05 14:05:50 +10001201 /* Ensure no bio records IO_BLOCKED */
1202 memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0]));
1203 init_r1bio(r1_bio, mddev, bio);
colyli@suse.defd768632017-02-18 03:05:56 +08001204 return r1_bio;
1205}
1206
NeilBrownc230e7e2017-04-05 14:05:50 +10001207static void raid1_read_request(struct mddev *mddev, struct bio *bio,
NeilBrown689389a2017-04-05 14:05:50 +10001208 int max_read_sectors, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
NeilBrowne8096362011-10-11 16:49:05 +11001210 struct r1conf *conf = mddev->private;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001211 struct raid1_info *mirror;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 struct bio *read_bio;
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001213 struct bitmap *bitmap = mddev->bitmap;
1214 const int op = bio_op(bio);
1215 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001216 int max_sectors;
1217 int rdisk;
NeilBrown689389a2017-04-05 14:05:50 +10001218 bool print_msg = !!r1_bio;
1219 char b[BDEVNAME_SIZE];
1220
1221 /*
1222 * If r1_bio is set, we are blocking the raid1d thread
1223 * so there is a tiny risk of deadlock. So ask for
1224 * emergency memory if needed.
1225 */
1226 gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO;
1227
1228 if (print_msg) {
1229 /* Need to get the block device name carefully */
1230 struct md_rdev *rdev;
1231 rcu_read_lock();
1232 rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev);
1233 if (rdev)
1234 bdevname(rdev->bdev, b);
1235 else
1236 strcpy(b, "???");
1237 rcu_read_unlock();
1238 }
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001239
colyli@suse.defd768632017-02-18 03:05:56 +08001240 /*
1241 * Still need barrier for READ in case that whole
1242 * array is frozen.
1243 */
1244 wait_read_barrier(conf, bio->bi_iter.bi_sector);
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001245
NeilBrown689389a2017-04-05 14:05:50 +10001246 if (!r1_bio)
1247 r1_bio = alloc_r1bio(mddev, bio);
1248 else
1249 init_r1bio(r1_bio, mddev, bio);
NeilBrownc230e7e2017-04-05 14:05:50 +10001250 r1_bio->sectors = max_read_sectors;
colyli@suse.defd768632017-02-18 03:05:56 +08001251
1252 /*
1253 * make_request() can abort the operation when read-ahead is being
1254 * used and no empty request is available.
1255 */
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001256 rdisk = read_balance(conf, r1_bio, &max_sectors);
1257
1258 if (rdisk < 0) {
1259 /* couldn't find anywhere to read from */
NeilBrown689389a2017-04-05 14:05:50 +10001260 if (print_msg) {
1261 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
1262 mdname(mddev),
1263 b,
1264 (unsigned long long)r1_bio->sector);
1265 }
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001266 raid_end_bio_io(r1_bio);
1267 return;
1268 }
1269 mirror = conf->mirrors + rdisk;
1270
NeilBrown689389a2017-04-05 14:05:50 +10001271 if (print_msg)
1272 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n",
1273 mdname(mddev),
1274 (unsigned long long)r1_bio->sector,
1275 bdevname(mirror->rdev->bdev, b));
1276
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001277 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1278 bitmap) {
1279 /*
1280 * Reading from a write-mostly device must take care not to
1281 * over-take any writes that are 'behind'
1282 */
1283 raid1_log(mddev, "wait behind writes");
1284 wait_event(bitmap->behind_wait,
1285 atomic_read(&bitmap->behind_writes) == 0);
1286 }
NeilBrownc230e7e2017-04-05 14:05:50 +10001287
1288 if (max_sectors < bio_sectors(bio)) {
1289 struct bio *split = bio_split(bio, max_sectors,
Kent Overstreetafeee512018-05-20 18:25:52 -04001290 gfp, &conf->bio_split);
NeilBrownc230e7e2017-04-05 14:05:50 +10001291 bio_chain(split, bio);
1292 generic_make_request(bio);
1293 bio = split;
1294 r1_bio->master_bio = bio;
1295 r1_bio->sectors = max_sectors;
1296 }
1297
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001298 r1_bio->read_disk = rdisk;
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001299
Kent Overstreetafeee512018-05-20 18:25:52 -04001300 read_bio = bio_clone_fast(bio, gfp, &mddev->bio_set);
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001301
1302 r1_bio->bios[rdisk] = read_bio;
1303
1304 read_bio->bi_iter.bi_sector = r1_bio->sector +
1305 mirror->rdev->data_offset;
Christoph Hellwig74d46992017-08-23 19:10:32 +02001306 bio_set_dev(read_bio, mirror->rdev->bdev);
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001307 read_bio->bi_end_io = raid1_end_read_request;
1308 bio_set_op_attrs(read_bio, op, do_sync);
1309 if (test_bit(FailFast, &mirror->rdev->flags) &&
1310 test_bit(R1BIO_FailFast, &r1_bio->state))
1311 read_bio->bi_opf |= MD_FAILFAST;
1312 read_bio->bi_private = r1_bio;
1313
1314 if (mddev->gendisk)
Christoph Hellwig74d46992017-08-23 19:10:32 +02001315 trace_block_bio_remap(read_bio->bi_disk->queue, read_bio,
1316 disk_devt(mddev->gendisk), r1_bio->sector);
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001317
NeilBrownc230e7e2017-04-05 14:05:50 +10001318 generic_make_request(read_bio);
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001319}
1320
NeilBrownc230e7e2017-04-05 14:05:50 +10001321static void raid1_write_request(struct mddev *mddev, struct bio *bio,
1322 int max_write_sectors)
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001323{
1324 struct r1conf *conf = mddev->private;
colyli@suse.defd768632017-02-18 03:05:56 +08001325 struct r1bio *r1_bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001326 int i, disks;
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001327 struct bitmap *bitmap = mddev->bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -07001328 unsigned long flags;
NeilBrown3cb03002011-10-11 16:45:26 +11001329 struct md_rdev *blocked_rdev;
NeilBrownf54a9d02012-08-02 08:33:20 +10001330 struct blk_plug_cb *cb;
1331 struct raid1_plug_cb *plug = NULL;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001332 int first_clone;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001333 int max_sectors;
NeilBrown191ea9b2005-06-21 17:17:23 -07001334
NeilBrownb3143b92017-10-17 13:46:43 +11001335 if (mddev_is_clustered(mddev) &&
Goldwyn Rodrigues90382ed2015-06-24 09:30:32 -05001336 md_cluster_ops->area_resyncing(mddev, WRITE,
NeilBrownb3143b92017-10-17 13:46:43 +11001337 bio->bi_iter.bi_sector, bio_end_sector(bio))) {
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001338
NeilBrown6eef4b22009-12-14 12:49:51 +11001339 DEFINE_WAIT(w);
1340 for (;;) {
NeilBrown6eef4b22009-12-14 12:49:51 +11001341 prepare_to_wait(&conf->wait_barrier,
Mikulas Patockaae89fd32017-10-18 19:01:11 -04001342 &w, TASK_IDLE);
Guoqing Jiangf81f7302017-10-24 15:33:33 +08001343 if (!md_cluster_ops->area_resyncing(mddev, WRITE,
Guoqing Jiang385f4d72017-09-29 09:16:43 +08001344 bio->bi_iter.bi_sector,
NeilBrownb3143b92017-10-17 13:46:43 +11001345 bio_end_sector(bio)))
NeilBrown6eef4b22009-12-14 12:49:51 +11001346 break;
1347 schedule();
1348 }
1349 finish_wait(&conf->wait_barrier, &w);
1350 }
Guoqing Jiangf81f7302017-10-24 15:33:33 +08001351
1352 /*
1353 * Register the new request and wait if the reconstruction
1354 * thread has put up a bar for new requests.
1355 * Continue immediately if no resync is active currently.
1356 */
colyli@suse.defd768632017-02-18 03:05:56 +08001357 wait_barrier(conf, bio->bi_iter.bi_sector);
1358
NeilBrown689389a2017-04-05 14:05:50 +10001359 r1_bio = alloc_r1bio(mddev, bio);
NeilBrownc230e7e2017-04-05 14:05:50 +10001360 r1_bio->sectors = max_write_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
NeilBrown34db0cd2011-10-11 16:50:01 +11001362 if (conf->pending_count >= max_queued_requests) {
1363 md_wakeup_thread(mddev->thread);
NeilBrown578b54a2016-11-14 16:30:21 +11001364 raid1_log(mddev, "wait queued");
NeilBrown34db0cd2011-10-11 16:50:01 +11001365 wait_event(conf->wait_barrier,
1366 conf->pending_count < max_queued_requests);
1367 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001368 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 * inc refcount on their rdev. Record them by setting
1370 * bios[x] to bio
NeilBrown1f68f0c2011-07-28 11:31:48 +10001371 * If there are known/acknowledged bad blocks on any device on
1372 * which we have seen a write error, we want to avoid writing those
1373 * blocks.
1374 * This potentially requires several writes to write around
1375 * the bad blocks. Each set of writes gets it's own r1bio
1376 * with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001378
NeilBrown8f19ccb2011-12-23 10:17:56 +11001379 disks = conf->raid_disks * 2;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001380 retry_write:
1381 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 rcu_read_lock();
NeilBrown1f68f0c2011-07-28 11:31:48 +10001383 max_sectors = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001385 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001386 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1387 atomic_inc(&rdev->nr_pending);
1388 blocked_rdev = rdev;
1389 break;
1390 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001391 r1_bio->bios[i] = NULL;
Kent Overstreet8ae12662015-04-27 23:48:34 -07001392 if (!rdev || test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11001393 if (i < conf->raid_disks)
1394 set_bit(R1BIO_Degraded, &r1_bio->state);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001395 continue;
1396 }
1397
1398 atomic_inc(&rdev->nr_pending);
1399 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1400 sector_t first_bad;
1401 int bad_sectors;
1402 int is_bad;
1403
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001404 is_bad = is_badblock(rdev, r1_bio->sector, max_sectors,
NeilBrown1f68f0c2011-07-28 11:31:48 +10001405 &first_bad, &bad_sectors);
1406 if (is_bad < 0) {
1407 /* mustn't write here until the bad block is
1408 * acknowledged*/
1409 set_bit(BlockedBadBlocks, &rdev->flags);
1410 blocked_rdev = rdev;
1411 break;
NeilBrown964147d2010-05-18 15:27:13 +10001412 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001413 if (is_bad && first_bad <= r1_bio->sector) {
1414 /* Cannot write here at all */
1415 bad_sectors -= (r1_bio->sector - first_bad);
1416 if (bad_sectors < max_sectors)
1417 /* mustn't write more than bad_sectors
1418 * to other devices yet
1419 */
1420 max_sectors = bad_sectors;
1421 rdev_dec_pending(rdev, mddev);
1422 /* We don't set R1BIO_Degraded as that
1423 * only applies if the disk is
1424 * missing, so it might be re-added,
1425 * and we want to know to recover this
1426 * chunk.
1427 * In this case the device is here,
1428 * and the fact that this chunk is not
1429 * in-sync is recorded in the bad
1430 * block log
1431 */
1432 continue;
1433 }
1434 if (is_bad) {
1435 int good_sectors = first_bad - r1_bio->sector;
1436 if (good_sectors < max_sectors)
1437 max_sectors = good_sectors;
1438 }
1439 }
1440 r1_bio->bios[i] = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 }
1442 rcu_read_unlock();
1443
Dan Williams6bfe0b42008-04-30 00:52:32 -07001444 if (unlikely(blocked_rdev)) {
1445 /* Wait for this device to become unblocked */
1446 int j;
1447
1448 for (j = 0; j < i; j++)
1449 if (r1_bio->bios[j])
1450 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001451 r1_bio->state = 0;
colyli@suse.defd768632017-02-18 03:05:56 +08001452 allow_barrier(conf, bio->bi_iter.bi_sector);
NeilBrown578b54a2016-11-14 16:30:21 +11001453 raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001454 md_wait_for_blocked_rdev(blocked_rdev, mddev);
colyli@suse.defd768632017-02-18 03:05:56 +08001455 wait_barrier(conf, bio->bi_iter.bi_sector);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001456 goto retry_write;
1457 }
1458
NeilBrownc230e7e2017-04-05 14:05:50 +10001459 if (max_sectors < bio_sectors(bio)) {
1460 struct bio *split = bio_split(bio, max_sectors,
Kent Overstreetafeee512018-05-20 18:25:52 -04001461 GFP_NOIO, &conf->bio_split);
NeilBrownc230e7e2017-04-05 14:05:50 +10001462 bio_chain(split, bio);
1463 generic_make_request(bio);
1464 bio = split;
1465 r1_bio->master_bio = bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001466 r1_bio->sectors = max_sectors;
NeilBrown191ea9b2005-06-21 17:17:23 -07001467 }
NeilBrown4b6d2872005-09-09 16:23:47 -07001468
NeilBrown4e780642010-10-19 12:54:01 +11001469 atomic_set(&r1_bio->remaining, 1);
NeilBrown4b6d2872005-09-09 16:23:47 -07001470 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -07001471
NeilBrown1f68f0c2011-07-28 11:31:48 +10001472 first_clone = 1;
Ming Leid8c84c42017-03-17 00:12:30 +08001473
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 for (i = 0; i < disks; i++) {
Ming Lei8e58e322017-02-14 23:29:01 +08001475 struct bio *mbio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (!r1_bio->bios[i])
1477 continue;
1478
NeilBrown1f68f0c2011-07-28 11:31:48 +10001479 if (first_clone) {
1480 /* do behind I/O ?
1481 * Not if there are too many, or cannot
1482 * allocate memory, or a reader on WriteMostly
1483 * is waiting for behind writes to flush */
1484 if (bitmap &&
1485 (atomic_read(&bitmap->behind_writes)
1486 < mddev->bitmap_info.max_write_behind) &&
Ming Lei8e58e322017-02-14 23:29:01 +08001487 !waitqueue_active(&bitmap->behind_wait)) {
Shaohua Li16d56e22017-07-17 14:33:48 -07001488 alloc_behind_master_bio(r1_bio, bio);
Ming Lei8e58e322017-02-14 23:29:01 +08001489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
Andy Shevchenkoe64e40182018-08-01 15:20:50 -07001491 md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors,
1492 test_bit(R1BIO_BehindIO, &r1_bio->state));
NeilBrown1f68f0c2011-07-28 11:31:48 +10001493 first_clone = 0;
1494 }
Ming Lei8e58e322017-02-14 23:29:01 +08001495
Shaohua Li16d56e22017-07-17 14:33:48 -07001496 if (r1_bio->behind_master_bio)
1497 mbio = bio_clone_fast(r1_bio->behind_master_bio,
Kent Overstreetafeee512018-05-20 18:25:52 -04001498 GFP_NOIO, &mddev->bio_set);
Shaohua Li16d56e22017-07-17 14:33:48 -07001499 else
Kent Overstreetafeee512018-05-20 18:25:52 -04001500 mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set);
Ming Lei8e58e322017-02-14 23:29:01 +08001501
Ming Lei841c1312017-03-17 00:12:31 +08001502 if (r1_bio->behind_master_bio) {
Guoqing Jiang3e148a32019-06-19 17:30:46 +08001503 struct md_rdev *rdev = conf->mirrors[i].rdev;
1504
1505 if (test_bit(WBCollisionCheck, &rdev->flags)) {
1506 sector_t lo = r1_bio->sector;
1507 sector_t hi = r1_bio->sector + r1_bio->sectors;
1508
1509 wait_event(rdev->wb_io_wait,
1510 check_and_add_wb(rdev, lo, hi) == 0);
1511 }
1512 if (test_bit(WriteMostly, &rdev->flags))
NeilBrown4b6d2872005-09-09 16:23:47 -07001513 atomic_inc(&r1_bio->behind_remaining);
1514 }
1515
NeilBrown1f68f0c2011-07-28 11:31:48 +10001516 r1_bio->bios[i] = mbio;
1517
Kent Overstreet4f024f32013-10-11 15:44:27 -07001518 mbio->bi_iter.bi_sector = (r1_bio->sector +
NeilBrown1f68f0c2011-07-28 11:31:48 +10001519 conf->mirrors[i].rdev->data_offset);
Christoph Hellwig74d46992017-08-23 19:10:32 +02001520 bio_set_dev(mbio, conf->mirrors[i].rdev->bdev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001521 mbio->bi_end_io = raid1_end_write_request;
Linus Torvaldsa682e002017-02-24 14:42:19 -08001522 mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA));
NeilBrown212e7eb2016-11-18 16:16:12 +11001523 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) &&
1524 !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) &&
1525 conf->raid_disks - mddev->degraded > 1)
1526 mbio->bi_opf |= MD_FAILFAST;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001527 mbio->bi_private = r1_bio;
1528
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 atomic_inc(&r1_bio->remaining);
NeilBrownf54a9d02012-08-02 08:33:20 +10001530
NeilBrown109e3762016-11-18 13:22:04 +11001531 if (mddev->gendisk)
Christoph Hellwig74d46992017-08-23 19:10:32 +02001532 trace_block_bio_remap(mbio->bi_disk->queue,
NeilBrown109e3762016-11-18 13:22:04 +11001533 mbio, disk_devt(mddev->gendisk),
1534 r1_bio->sector);
1535 /* flush_pending_writes() needs access to the rdev so...*/
Christoph Hellwig74d46992017-08-23 19:10:32 +02001536 mbio->bi_disk = (void *)conf->mirrors[i].rdev;
NeilBrown109e3762016-11-18 13:22:04 +11001537
NeilBrownf54a9d02012-08-02 08:33:20 +10001538 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1539 if (cb)
1540 plug = container_of(cb, struct raid1_plug_cb, cb);
1541 else
1542 plug = NULL;
NeilBrownf54a9d02012-08-02 08:33:20 +10001543 if (plug) {
1544 bio_list_add(&plug->pending, mbio);
1545 plug->pending_cnt++;
1546 } else {
Shaohua Li23b245c2017-05-10 08:47:11 -07001547 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001548 bio_list_add(&conf->pending_bio_list, mbio);
1549 conf->pending_count++;
Shaohua Li23b245c2017-05-10 08:47:11 -07001550 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrownb357f042012-07-03 17:45:31 +10001551 md_wakeup_thread(mddev->thread);
Shaohua Li23b245c2017-05-10 08:47:11 -07001552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001554
NeilBrown079fa162011-09-10 17:21:23 +10001555 r1_bio_write_done(r1_bio);
1556
1557 /* In case raid1d snuck in to freeze_array */
1558 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
1560
NeilBrowncc27b0c2017-06-05 16:49:39 +10001561static bool raid1_make_request(struct mddev *mddev, struct bio *bio)
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001562{
colyli@suse.defd768632017-02-18 03:05:56 +08001563 sector_t sectors;
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001564
Shaohua Liaff8da02017-02-21 10:56:19 -08001565 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1566 md_flush_request(mddev, bio);
NeilBrowncc27b0c2017-06-05 16:49:39 +10001567 return true;
Shaohua Liaff8da02017-02-21 10:56:19 -08001568 }
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001569
NeilBrownc230e7e2017-04-05 14:05:50 +10001570 /*
1571 * There is a limit to the maximum size, but
1572 * the read/write handler might find a lower limit
1573 * due to bad blocks. To avoid multiple splits,
1574 * we pass the maximum number of sectors down
1575 * and let the lower level perform the split.
1576 */
1577 sectors = align_to_barrier_unit_end(
1578 bio->bi_iter.bi_sector, bio_sectors(bio));
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001579
NeilBrownc230e7e2017-04-05 14:05:50 +10001580 if (bio_data_dir(bio) == READ)
NeilBrown689389a2017-04-05 14:05:50 +10001581 raid1_read_request(mddev, bio, sectors, NULL);
NeilBrowncc27b0c2017-06-05 16:49:39 +10001582 else {
1583 if (!md_write_start(mddev,bio))
1584 return false;
NeilBrownc230e7e2017-04-05 14:05:50 +10001585 raid1_write_request(mddev, bio, sectors);
NeilBrowncc27b0c2017-06-05 16:49:39 +10001586 }
1587 return true;
Robert LeBlanc3b046a92016-12-05 13:02:57 -07001588}
1589
Shaohua Li849674e2016-01-20 13:52:20 -08001590static void raid1_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591{
NeilBrowne8096362011-10-11 16:49:05 +11001592 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 int i;
1594
1595 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001596 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001597 rcu_read_lock();
1598 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001599 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001601 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1602 }
1603 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 seq_printf(seq, "]");
1605}
1606
Shaohua Li849674e2016-01-20 13:52:20 -08001607static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608{
1609 char b[BDEVNAME_SIZE];
NeilBrowne8096362011-10-11 16:49:05 +11001610 struct r1conf *conf = mddev->private;
NeilBrown423f04d2015-07-27 11:48:52 +10001611 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
1613 /*
1614 * If it is not operational, then we have already marked it as dead
1615 * else if it is the last working disks, ignore the error, let the
1616 * next level up know.
1617 * else mark the drive as failed
1618 */
NeilBrown2e52d442016-11-18 16:16:12 +11001619 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownb2d444d2005-11-08 21:39:31 -08001620 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001621 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 /*
1623 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001624 * normal single drive.
1625 * However don't try a recovery from this drive as
1626 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 */
NeilBrown53890422011-07-27 11:00:36 +10001628 conf->recovery_disabled = mddev->recovery_disabled;
NeilBrown2e52d442016-11-18 16:16:12 +11001629 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001631 }
NeilBrownde393cd2011-07-28 11:31:48 +10001632 set_bit(Blocked, &rdev->flags);
Yufen Yuebda52f2019-02-01 10:45:01 +08001633 if (test_and_clear_bit(In_sync, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 mddev->degraded++;
Yufen Yuebda52f2019-02-01 10:45:01 +08001635 set_bit(Faulty, &rdev->flags);
NeilBrown423f04d2015-07-27 11:48:52 +10001636 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrown2446dba2014-07-31 10:16:29 +10001637 /*
1638 * if recovery is running, make sure it aborts.
1639 */
1640 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Shaohua Li29530792016-12-08 15:48:19 -08001641 set_mask_bits(&mddev->sb_flags, 0,
1642 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
NeilBrown1d41c212016-11-02 14:16:50 +11001643 pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n"
1644 "md/raid1:%s: Operation continuing on %d devices.\n",
1645 mdname(mddev), bdevname(rdev->bdev, b),
1646 mdname(mddev), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647}
1648
NeilBrowne8096362011-10-11 16:49:05 +11001649static void print_conf(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650{
1651 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
NeilBrown1d41c212016-11-02 14:16:50 +11001653 pr_debug("RAID1 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (!conf) {
NeilBrown1d41c212016-11-02 14:16:50 +11001655 pr_debug("(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 return;
1657 }
NeilBrown1d41c212016-11-02 14:16:50 +11001658 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1659 conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
NeilBrownddac7c72006-08-31 21:27:36 -07001661 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 for (i = 0; i < conf->raid_disks; i++) {
1663 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001664 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownddac7c72006-08-31 21:27:36 -07001665 if (rdev)
NeilBrown1d41c212016-11-02 14:16:50 +11001666 pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n",
1667 i, !test_bit(In_sync, &rdev->flags),
1668 !test_bit(Faulty, &rdev->flags),
1669 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
NeilBrownddac7c72006-08-31 21:27:36 -07001671 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672}
1673
NeilBrowne8096362011-10-11 16:49:05 +11001674static void close_sync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Nate Daileyf6eca2d2017-10-17 08:17:03 -04001676 int idx;
1677
1678 for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) {
1679 _wait_barrier(conf, idx);
1680 _allow_barrier(conf, idx);
1681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
Kent Overstreetafeee512018-05-20 18:25:52 -04001683 mempool_exit(&conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684}
1685
NeilBrownfd01b882011-10-11 16:47:53 +11001686static int raid1_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687{
1688 int i;
NeilBrowne8096362011-10-11 16:49:05 +11001689 struct r1conf *conf = mddev->private;
NeilBrown6b965622010-08-18 11:56:59 +10001690 int count = 0;
1691 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
1693 /*
NeilBrownf72ffdd2014-09-30 14:23:59 +10001694 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001695 * and mark them readable.
1696 * Called under mddev lock, so rcu protection not needed.
NeilBrown423f04d2015-07-27 11:48:52 +10001697 * device_lock used to avoid races with raid1_end_read_request
1698 * which expects 'In_sync' flags and ->degraded to be consistent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 */
NeilBrown423f04d2015-07-27 11:48:52 +10001700 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001702 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001703 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1704 if (repl
Goldwyn Rodrigues1aee41f2014-10-29 18:51:31 -05001705 && !test_bit(Candidate, &repl->flags)
NeilBrown8c7a2c22011-12-23 10:17:57 +11001706 && repl->recovery_offset == MaxSector
1707 && !test_bit(Faulty, &repl->flags)
1708 && !test_and_set_bit(In_sync, &repl->flags)) {
1709 /* replacement has just become active */
1710 if (!rdev ||
1711 !test_and_clear_bit(In_sync, &rdev->flags))
1712 count++;
1713 if (rdev) {
1714 /* Replaced device not technically
1715 * faulty, but we need to be sure
1716 * it gets removed and never re-added
1717 */
1718 set_bit(Faulty, &rdev->flags);
1719 sysfs_notify_dirent_safe(
1720 rdev->sysfs_state);
1721 }
1722 }
NeilBrownddac7c72006-08-31 21:27:36 -07001723 if (rdev
Lukasz Dorau61e49472013-10-24 12:55:17 +11001724 && rdev->recovery_offset == MaxSector
NeilBrownddac7c72006-08-31 21:27:36 -07001725 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001726 && !test_and_set_bit(In_sync, &rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001727 count++;
Jonathan Brassow654e8b52011-07-27 11:00:36 +10001728 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 }
1730 }
NeilBrown6b965622010-08-18 11:56:59 +10001731 mddev->degraded -= count;
1732 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001735 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736}
1737
NeilBrownfd01b882011-10-11 16:47:53 +11001738static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
NeilBrowne8096362011-10-11 16:49:05 +11001740 struct r1conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001741 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001742 int mirror = 0;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001743 struct raid1_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001744 int first = 0;
NeilBrown30194632011-12-23 10:17:56 +11001745 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
NeilBrown53890422011-07-27 11:00:36 +10001747 if (mddev->recovery_disabled == conf->recovery_disabled)
1748 return -EBUSY;
1749
Dan Williams1501efa2016-01-13 16:00:07 -08001750 if (md_integrity_add_rdev(rdev, mddev))
1751 return -ENXIO;
1752
Neil Brown6c2fce22008-06-28 08:31:31 +10001753 if (rdev->raid_disk >= 0)
1754 first = last = rdev->raid_disk;
1755
Goldwyn Rodrigues70bcecd2015-08-21 10:33:39 -05001756 /*
1757 * find the disk ... but prefer rdev->saved_raid_disk
1758 * if possible.
1759 */
1760 if (rdev->saved_raid_disk >= 0 &&
1761 rdev->saved_raid_disk >= first &&
Shaohua Li9e753ba2018-10-14 17:05:07 -07001762 rdev->saved_raid_disk < conf->raid_disks &&
Goldwyn Rodrigues70bcecd2015-08-21 10:33:39 -05001763 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1764 first = last = rdev->saved_raid_disk;
1765
NeilBrown7ef449d2011-12-23 10:17:57 +11001766 for (mirror = first; mirror <= last; mirror++) {
Yufen Yuebfeb442019-06-14 15:41:08 -07001767 p = conf->mirrors + mirror;
NeilBrown7ef449d2011-12-23 10:17:57 +11001768 if (!p->rdev) {
Jonathan Brassow9092c022013-05-02 14:19:24 -05001769 if (mddev->gendisk)
1770 disk_stack_limits(mddev->gendisk, rdev->bdev,
1771 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
1773 p->head_position = 0;
1774 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001775 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001776 /* As all devices are equivalent, we don't need a full recovery
1777 * if this was recently any drive of the array
1778 */
1779 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001780 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001781 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 break;
1783 }
NeilBrown7ef449d2011-12-23 10:17:57 +11001784 if (test_bit(WantReplacement, &p->rdev->flags) &&
1785 p[conf->raid_disks].rdev == NULL) {
1786 /* Add this device as a replacement */
1787 clear_bit(In_sync, &rdev->flags);
1788 set_bit(Replacement, &rdev->flags);
1789 rdev->raid_disk = mirror;
1790 err = 0;
1791 conf->fullsync = 1;
1792 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1793 break;
1794 }
1795 }
Jonathan Brassow9092c022013-05-02 14:19:24 -05001796 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
Bart Van Assche8b904b52018-03-07 17:10:10 -08001797 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001799 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800}
1801
NeilBrownb8321b62011-12-23 10:17:51 +11001802static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
NeilBrowne8096362011-10-11 16:49:05 +11001804 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001806 int number = rdev->raid_disk;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001807 struct raid1_info *p = conf->mirrors + number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
NeilBrownb014f142011-12-23 10:17:56 +11001809 if (rdev != p->rdev)
1810 p = conf->mirrors + conf->raid_disks + number;
1811
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 print_conf(conf);
NeilBrownb8321b62011-12-23 10:17:51 +11001813 if (rdev == p->rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001814 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 atomic_read(&rdev->nr_pending)) {
1816 err = -EBUSY;
1817 goto abort;
1818 }
NeilBrown046abee2010-10-26 15:46:20 +11001819 /* Only remove non-faulty devices if recovery
NeilBrowndfc70642008-05-23 13:04:39 -07001820 * is not possible.
1821 */
1822 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown53890422011-07-27 11:00:36 +10001823 mddev->recovery_disabled != conf->recovery_disabled &&
NeilBrowndfc70642008-05-23 13:04:39 -07001824 mddev->degraded < conf->raid_disks) {
1825 err = -EBUSY;
1826 goto abort;
1827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 p->rdev = NULL;
NeilBrownd787be42016-06-02 16:19:53 +10001829 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
1830 synchronize_rcu();
1831 if (atomic_read(&rdev->nr_pending)) {
1832 /* lost the race, try later */
1833 err = -EBUSY;
1834 p->rdev = rdev;
1835 goto abort;
1836 }
1837 }
1838 if (conf->mirrors[conf->raid_disks + number].rdev) {
NeilBrown8c7a2c22011-12-23 10:17:57 +11001839 /* We just removed a device that is being replaced.
1840 * Move down the replacement. We drain all IO before
1841 * doing this to avoid confusion.
1842 */
1843 struct md_rdev *repl =
1844 conf->mirrors[conf->raid_disks + number].rdev;
NeilBrowne2d59922013-06-12 11:01:22 +10001845 freeze_array(conf, 0);
Yufen Yu3de59bb2018-02-24 12:05:56 +08001846 if (atomic_read(&repl->nr_pending)) {
1847 /* It means that some queued IO of retry_list
1848 * hold repl. Thus, we cannot set replacement
1849 * as NULL, avoiding rdev NULL pointer
1850 * dereference in sync_request_write and
1851 * handle_write_finished.
1852 */
1853 err = -EBUSY;
1854 unfreeze_array(conf);
1855 goto abort;
1856 }
NeilBrown8c7a2c22011-12-23 10:17:57 +11001857 clear_bit(Replacement, &repl->flags);
1858 p->rdev = repl;
1859 conf->mirrors[conf->raid_disks + number].rdev = NULL;
NeilBrowne2d59922013-06-12 11:01:22 +10001860 unfreeze_array(conf);
Guoqing Jiange5bc9c32017-04-24 15:58:04 +08001861 }
1862
1863 clear_bit(WantReplacement, &rdev->flags);
Martin K. Petersena91a2782011-03-17 11:11:05 +01001864 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 }
1866abort:
1867
1868 print_conf(conf);
1869 return err;
1870}
1871
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001872static void end_sync_read(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873{
Ming Lei98d30c52017-03-17 00:12:26 +08001874 struct r1bio *r1_bio = get_resync_r1bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
NeilBrown0fc280f2011-10-07 14:22:55 +11001876 update_head_pos(r1_bio->read_disk, r1_bio);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001877
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 /*
1879 * we have read a block, now it needs to be re-written,
1880 * or re-read if the read failed.
1881 * We don't do much here, just schedule handling by raid1d
1882 */
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001883 if (!bio->bi_status)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001885
1886 if (atomic_dec_and_test(&r1_bio->remaining))
1887 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888}
1889
Nate Daileydfcc34c2019-02-07 14:19:01 -05001890static void abort_sync_write(struct mddev *mddev, struct r1bio *r1_bio)
1891{
1892 sector_t sync_blocks = 0;
1893 sector_t s = r1_bio->sector;
1894 long sectors_to_go = r1_bio->sectors;
1895
1896 /* make sure these bits don't get cleared. */
1897 do {
1898 md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1);
1899 s += sync_blocks;
1900 sectors_to_go -= sync_blocks;
1901 } while (sectors_to_go > 0);
1902}
1903
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001904static void end_sync_write(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905{
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001906 int uptodate = !bio->bi_status;
Ming Lei98d30c52017-03-17 00:12:26 +08001907 struct r1bio *r1_bio = get_resync_r1bio(bio);
NeilBrownfd01b882011-10-11 16:47:53 +11001908 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001909 struct r1conf *conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10001910 sector_t first_bad;
1911 int bad_sectors;
NeilBrown854abd72016-06-02 16:19:52 +10001912 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001913
NeilBrown6b1117d2006-03-31 02:31:57 -08001914 if (!uptodate) {
Nate Daileydfcc34c2019-02-07 14:19:01 -05001915 abort_sync_write(mddev, r1_bio);
NeilBrown854abd72016-06-02 16:19:52 +10001916 set_bit(WriteErrorSeen, &rdev->flags);
1917 if (!test_and_set_bit(WantReplacement, &rdev->flags))
NeilBrown19d67162011-12-23 10:17:57 +11001918 set_bit(MD_RECOVERY_NEEDED, &
1919 mddev->recovery);
NeilBrownd8f05d22011-07-28 11:33:00 +10001920 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown854abd72016-06-02 16:19:52 +10001921 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
NeilBrown3a9f28a2011-07-28 11:33:42 +10001922 &first_bad, &bad_sectors) &&
1923 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1924 r1_bio->sector,
1925 r1_bio->sectors,
1926 &first_bad, &bad_sectors)
1927 )
NeilBrown4367af52011-07-28 11:31:49 +10001928 set_bit(R1BIO_MadeGood, &r1_bio->state);
NeilBrowne3b97032005-08-04 12:53:34 -07001929
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown4367af52011-07-28 11:31:49 +10001931 int s = r1_bio->sectors;
NeilBrownd8f05d22011-07-28 11:33:00 +10001932 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1933 test_bit(R1BIO_WriteError, &r1_bio->state))
NeilBrown4367af52011-07-28 11:31:49 +10001934 reschedule_retry(r1_bio);
1935 else {
1936 put_buf(r1_bio);
1937 md_done_sync(mddev, s, uptodate);
1938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940}
1941
NeilBrown3cb03002011-10-11 16:45:26 +11001942static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrownd8f05d22011-07-28 11:33:00 +10001943 int sectors, struct page *page, int rw)
1944{
Mike Christie796a5cf2016-06-05 14:32:07 -05001945 if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
NeilBrownd8f05d22011-07-28 11:33:00 +10001946 /* success */
1947 return 1;
NeilBrown19d67162011-12-23 10:17:57 +11001948 if (rw == WRITE) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001949 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001950 if (!test_and_set_bit(WantReplacement,
1951 &rdev->flags))
1952 set_bit(MD_RECOVERY_NEEDED, &
1953 rdev->mddev->recovery);
1954 }
NeilBrownd8f05d22011-07-28 11:33:00 +10001955 /* need to record an error - either for the block or the device */
1956 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1957 md_error(rdev->mddev, rdev);
1958 return 0;
1959}
1960
NeilBrown9f2c9d12011-10-11 16:48:43 +11001961static int fix_sync_read_error(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001962{
1963 /* Try some synchronous reads of other devices to get
1964 * good data, much like with normal read errors. Only
1965 * read into the pages we already have so we don't
1966 * need to re-issue the read request.
1967 * We don't need to freeze the array, because being in an
1968 * active sync request, there is no normal IO, and
1969 * no overlapping syncs.
NeilBrown06f60382011-07-28 11:31:48 +10001970 * We don't need to check is_badblock() again as we
1971 * made sure that anything with a bad block in range
1972 * will have bi_end_io clear.
NeilBrowna68e5872011-05-11 14:40:44 +10001973 */
NeilBrownfd01b882011-10-11 16:47:53 +11001974 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001975 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001976 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
Ming Lei44cf0f42017-03-17 00:12:27 +08001977 struct page **pages = get_resync_pages(bio)->pages;
NeilBrowna68e5872011-05-11 14:40:44 +10001978 sector_t sect = r1_bio->sector;
1979 int sectors = r1_bio->sectors;
1980 int idx = 0;
NeilBrown2e52d442016-11-18 16:16:12 +11001981 struct md_rdev *rdev;
1982
1983 rdev = conf->mirrors[r1_bio->read_disk].rdev;
1984 if (test_bit(FailFast, &rdev->flags)) {
1985 /* Don't try recovering from here - just fail it
1986 * ... unless it is the last working device of course */
1987 md_error(mddev, rdev);
1988 if (test_bit(Faulty, &rdev->flags))
1989 /* Don't try to read from here, but make sure
1990 * put_buf does it's thing
1991 */
1992 bio->bi_end_io = end_sync_write;
1993 }
NeilBrowna68e5872011-05-11 14:40:44 +10001994
1995 while(sectors) {
1996 int s = sectors;
1997 int d = r1_bio->read_disk;
1998 int success = 0;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001999 int start;
NeilBrowna68e5872011-05-11 14:40:44 +10002000
2001 if (s > (PAGE_SIZE>>9))
2002 s = PAGE_SIZE >> 9;
2003 do {
2004 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
2005 /* No rcu protection needed here devices
2006 * can only be removed when no resync is
2007 * active, and resync is currently active
2008 */
2009 rdev = conf->mirrors[d].rdev;
Namhyung Kim9d3d8012011-07-27 11:00:36 +10002010 if (sync_page_io(rdev, sect, s<<9,
Ming Lei44cf0f42017-03-17 00:12:27 +08002011 pages[idx],
Mike Christie796a5cf2016-06-05 14:32:07 -05002012 REQ_OP_READ, 0, false)) {
NeilBrowna68e5872011-05-11 14:40:44 +10002013 success = 1;
2014 break;
2015 }
2016 }
2017 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002018 if (d == conf->raid_disks * 2)
NeilBrowna68e5872011-05-11 14:40:44 +10002019 d = 0;
2020 } while (!success && d != r1_bio->read_disk);
2021
NeilBrown78d7f5f2011-05-11 14:48:56 +10002022 if (!success) {
NeilBrowna68e5872011-05-11 14:40:44 +10002023 char b[BDEVNAME_SIZE];
NeilBrown3a9f28a2011-07-28 11:33:42 +10002024 int abort = 0;
2025 /* Cannot read from anywhere, this block is lost.
2026 * Record a bad block on each device. If that doesn't
2027 * work just disable and interrupt the recovery.
2028 * Don't fail devices as that won't really help.
2029 */
NeilBrown1d41c212016-11-02 14:16:50 +11002030 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
Christoph Hellwig74d46992017-08-23 19:10:32 +02002031 mdname(mddev), bio_devname(bio, b),
NeilBrown1d41c212016-11-02 14:16:50 +11002032 (unsigned long long)r1_bio->sector);
NeilBrown8f19ccb2011-12-23 10:17:56 +11002033 for (d = 0; d < conf->raid_disks * 2; d++) {
NeilBrown3a9f28a2011-07-28 11:33:42 +10002034 rdev = conf->mirrors[d].rdev;
2035 if (!rdev || test_bit(Faulty, &rdev->flags))
2036 continue;
2037 if (!rdev_set_badblocks(rdev, sect, s, 0))
2038 abort = 1;
2039 }
2040 if (abort) {
NeilBrownd890fa22011-10-26 11:54:39 +11002041 conf->recovery_disabled =
2042 mddev->recovery_disabled;
NeilBrown3a9f28a2011-07-28 11:33:42 +10002043 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2044 md_done_sync(mddev, r1_bio->sectors, 0);
2045 put_buf(r1_bio);
2046 return 0;
2047 }
2048 /* Try next page */
2049 sectors -= s;
2050 sect += s;
2051 idx++;
2052 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10002053 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10002054
2055 start = d;
2056 /* write it back and re-read */
2057 while (d != r1_bio->read_disk) {
2058 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002059 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10002060 d--;
2061 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
2062 continue;
2063 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10002064 if (r1_sync_page_io(rdev, sect, s,
Ming Lei44cf0f42017-03-17 00:12:27 +08002065 pages[idx],
NeilBrownd8f05d22011-07-28 11:33:00 +10002066 WRITE) == 0) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10002067 r1_bio->bios[d]->bi_end_io = NULL;
2068 rdev_dec_pending(rdev, mddev);
Namhyung Kim9d3d8012011-07-27 11:00:36 +10002069 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10002070 }
2071 d = start;
2072 while (d != r1_bio->read_disk) {
2073 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002074 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10002075 d--;
2076 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
2077 continue;
2078 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10002079 if (r1_sync_page_io(rdev, sect, s,
Ming Lei44cf0f42017-03-17 00:12:27 +08002080 pages[idx],
NeilBrownd8f05d22011-07-28 11:33:00 +10002081 READ) != 0)
Namhyung Kim9d3d8012011-07-27 11:00:36 +10002082 atomic_add(s, &rdev->corrected_errors);
NeilBrown78d7f5f2011-05-11 14:48:56 +10002083 }
NeilBrowna68e5872011-05-11 14:40:44 +10002084 sectors -= s;
2085 sect += s;
2086 idx ++;
2087 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10002088 set_bit(R1BIO_Uptodate, &r1_bio->state);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002089 bio->bi_status = 0;
NeilBrowna68e5872011-05-11 14:40:44 +10002090 return 1;
2091}
2092
NeilBrownc95e6382014-09-09 13:54:11 +10002093static void process_checks(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10002094{
2095 /* We have read all readable devices. If we haven't
2096 * got the block, then there is no hope left.
2097 * If we have, then we want to do a comparison
2098 * and skip the write if everything is the same.
2099 * If any blocks failed to read, then we need to
2100 * attempt an over-write
2101 */
NeilBrownfd01b882011-10-11 16:47:53 +11002102 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11002103 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10002104 int primary;
2105 int i;
majianpengf4380a92012-04-12 16:04:47 +10002106 int vcnt;
NeilBrowna68e5872011-05-11 14:40:44 +10002107
NeilBrown30bc9b52013-07-17 15:19:29 +10002108 /* Fix variable parts of all bios */
2109 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
2110 for (i = 0; i < conf->raid_disks * 2; i++) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002111 blk_status_t status;
NeilBrown30bc9b52013-07-17 15:19:29 +10002112 struct bio *b = r1_bio->bios[i];
Ming Lei98d30c52017-03-17 00:12:26 +08002113 struct resync_pages *rp = get_resync_pages(b);
NeilBrown30bc9b52013-07-17 15:19:29 +10002114 if (b->bi_end_io != end_sync_read)
2115 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002116 /* fixup the bio for reuse, but preserve errno */
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002117 status = b->bi_status;
NeilBrown30bc9b52013-07-17 15:19:29 +10002118 bio_reset(b);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002119 b->bi_status = status;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002120 b->bi_iter.bi_sector = r1_bio->sector +
NeilBrown30bc9b52013-07-17 15:19:29 +10002121 conf->mirrors[i].rdev->data_offset;
Christoph Hellwig74d46992017-08-23 19:10:32 +02002122 bio_set_dev(b, conf->mirrors[i].rdev->bdev);
NeilBrown30bc9b52013-07-17 15:19:29 +10002123 b->bi_end_io = end_sync_read;
Ming Lei98d30c52017-03-17 00:12:26 +08002124 rp->raid_bio = r1_bio;
2125 b->bi_private = rp;
NeilBrown30bc9b52013-07-17 15:19:29 +10002126
Ming Leifb0eb5d2017-07-14 16:14:43 +08002127 /* initialize bvec table again */
2128 md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9);
NeilBrown30bc9b52013-07-17 15:19:29 +10002129 }
NeilBrown8f19ccb2011-12-23 10:17:56 +11002130 for (primary = 0; primary < conf->raid_disks * 2; primary++)
NeilBrowna68e5872011-05-11 14:40:44 +10002131 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002132 !r1_bio->bios[primary]->bi_status) {
NeilBrowna68e5872011-05-11 14:40:44 +10002133 r1_bio->bios[primary]->bi_end_io = NULL;
2134 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
2135 break;
2136 }
2137 r1_bio->read_disk = primary;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002138 for (i = 0; i < conf->raid_disks * 2; i++) {
Christoph Hellwig2b070cf2019-04-25 09:03:00 +02002139 int j = 0;
NeilBrown78d7f5f2011-05-11 14:48:56 +10002140 struct bio *pbio = r1_bio->bios[primary];
2141 struct bio *sbio = r1_bio->bios[i];
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002142 blk_status_t status = sbio->bi_status;
Ming Lei44cf0f42017-03-17 00:12:27 +08002143 struct page **ppages = get_resync_pages(pbio)->pages;
2144 struct page **spages = get_resync_pages(sbio)->pages;
Ming Lei60928a92017-03-17 00:12:28 +08002145 struct bio_vec *bi;
Ming Lei8fc04e6e2017-03-28 16:17:55 +08002146 int page_len[RESYNC_PAGES] = { 0 };
Ming Lei6dc4f102019-02-15 19:13:19 +08002147 struct bvec_iter_all iter_all;
NeilBrowna68e5872011-05-11 14:40:44 +10002148
Kent Overstreet2aabaa62012-09-11 11:26:12 -07002149 if (sbio->bi_end_io != end_sync_read)
NeilBrown78d7f5f2011-05-11 14:48:56 +10002150 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002151 /* Now we can 'fixup' the error value */
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002152 sbio->bi_status = 0;
NeilBrowna68e5872011-05-11 14:40:44 +10002153
Christoph Hellwig2b070cf2019-04-25 09:03:00 +02002154 bio_for_each_segment_all(bi, sbio, iter_all)
2155 page_len[j++] = bi->bv_len;
Ming Lei60928a92017-03-17 00:12:28 +08002156
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002157 if (!status) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10002158 for (j = vcnt; j-- ; ) {
Ming Lei44cf0f42017-03-17 00:12:27 +08002159 if (memcmp(page_address(ppages[j]),
2160 page_address(spages[j]),
Ming Lei60928a92017-03-17 00:12:28 +08002161 page_len[j]))
NeilBrown78d7f5f2011-05-11 14:48:56 +10002162 break;
NeilBrowna68e5872011-05-11 14:40:44 +10002163 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10002164 } else
2165 j = 0;
2166 if (j >= 0)
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11002167 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
NeilBrown78d7f5f2011-05-11 14:48:56 +10002168 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002169 && !status)) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10002170 /* No need to write to this device. */
2171 sbio->bi_end_io = NULL;
2172 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
2173 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10002174 }
Kent Overstreetd3b45c22012-09-10 13:49:33 -07002175
2176 bio_copy_data(sbio, pbio);
NeilBrown78d7f5f2011-05-11 14:48:56 +10002177 }
NeilBrowna68e5872011-05-11 14:40:44 +10002178}
2179
NeilBrown9f2c9d12011-10-11 16:48:43 +11002180static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181{
NeilBrowne8096362011-10-11 16:49:05 +11002182 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 int i;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002184 int disks = conf->raid_disks * 2;
Guoqing Jiang037d2ff2017-06-15 17:00:25 +08002185 struct bio *wbio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
NeilBrowna68e5872011-05-11 14:40:44 +10002187 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
2188 /* ouch - failed to read all of that. */
2189 if (!fix_sync_read_error(r1_bio))
2190 return;
NeilBrown7ca78d52011-05-11 14:50:37 +10002191
2192 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrownc95e6382014-09-09 13:54:11 +10002193 process_checks(r1_bio);
2194
NeilBrownd11c1712006-01-06 00:20:26 -08002195 /*
2196 * schedule writes
2197 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 atomic_set(&r1_bio->remaining, 1);
2199 for (i = 0; i < disks ; i++) {
2200 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08002201 if (wbio->bi_end_io == NULL ||
2202 (wbio->bi_end_io == end_sync_read &&
2203 (i == r1_bio->read_disk ||
2204 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 continue;
Nate Daileydfcc34c2019-02-07 14:19:01 -05002206 if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) {
2207 abort_sync_write(mddev, r1_bio);
NeilBrown0c9d5b12017-04-06 12:06:37 +10002208 continue;
Nate Daileydfcc34c2019-02-07 14:19:01 -05002209 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
Mike Christie796a5cf2016-06-05 14:32:07 -05002211 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
NeilBrown212e7eb2016-11-18 16:16:12 +11002212 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags))
2213 wbio->bi_opf |= MD_FAILFAST;
2214
NeilBrown3e198f72006-01-06 00:20:21 -08002215 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 atomic_inc(&r1_bio->remaining);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002217 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
NeilBrown191ea9b2005-06-21 17:17:23 -07002218
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 generic_make_request(wbio);
2220 }
2221
2222 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002223 /* if we're here, all write(s) have completed, so clean up */
NeilBrown58e94ae2012-07-19 15:59:18 +10002224 int s = r1_bio->sectors;
2225 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2226 test_bit(R1BIO_WriteError, &r1_bio->state))
2227 reschedule_retry(r1_bio);
2228 else {
2229 put_buf(r1_bio);
2230 md_done_sync(mddev, s, 1);
2231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 }
2233}
2234
2235/*
2236 * This is a kernel thread which:
2237 *
2238 * 1. Retries failed read operations on working mirrors.
2239 * 2. Updates the raid superblock when problems encounter.
NeilBrownd2eb35a2011-07-28 11:31:48 +10002240 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 */
2242
NeilBrowne8096362011-10-11 16:49:05 +11002243static void fix_read_error(struct r1conf *conf, int read_disk,
NeilBrown867868f2006-10-03 01:15:51 -07002244 sector_t sect, int sectors)
2245{
NeilBrownfd01b882011-10-11 16:47:53 +11002246 struct mddev *mddev = conf->mddev;
NeilBrown867868f2006-10-03 01:15:51 -07002247 while(sectors) {
2248 int s = sectors;
2249 int d = read_disk;
2250 int success = 0;
2251 int start;
NeilBrown3cb03002011-10-11 16:45:26 +11002252 struct md_rdev *rdev;
NeilBrown867868f2006-10-03 01:15:51 -07002253
2254 if (s > (PAGE_SIZE>>9))
2255 s = PAGE_SIZE >> 9;
2256
2257 do {
NeilBrownd2eb35a2011-07-28 11:31:48 +10002258 sector_t first_bad;
2259 int bad_sectors;
2260
NeilBrown707a6a42016-06-02 16:19:52 +10002261 rcu_read_lock();
2262 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002263 if (rdev &&
majianpengda8840a2012-05-22 13:55:03 +10002264 (test_bit(In_sync, &rdev->flags) ||
2265 (!test_bit(Faulty, &rdev->flags) &&
2266 rdev->recovery_offset >= sect + s)) &&
NeilBrownd2eb35a2011-07-28 11:31:48 +10002267 is_badblock(rdev, sect, s,
NeilBrown707a6a42016-06-02 16:19:52 +10002268 &first_bad, &bad_sectors) == 0) {
2269 atomic_inc(&rdev->nr_pending);
2270 rcu_read_unlock();
2271 if (sync_page_io(rdev, sect, s<<9,
Mike Christie796a5cf2016-06-05 14:32:07 -05002272 conf->tmppage, REQ_OP_READ, 0, false))
NeilBrown707a6a42016-06-02 16:19:52 +10002273 success = 1;
2274 rdev_dec_pending(rdev, mddev);
2275 if (success)
2276 break;
2277 } else
2278 rcu_read_unlock();
2279 d++;
2280 if (d == conf->raid_disks * 2)
2281 d = 0;
NeilBrown867868f2006-10-03 01:15:51 -07002282 } while (!success && d != read_disk);
2283
2284 if (!success) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002285 /* Cannot read from anywhere - mark it bad */
NeilBrown3cb03002011-10-11 16:45:26 +11002286 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10002287 if (!rdev_set_badblocks(rdev, sect, s, 0))
2288 md_error(mddev, rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002289 break;
2290 }
2291 /* write it back and re-read */
2292 start = d;
2293 while (d != read_disk) {
2294 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002295 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002296 d--;
NeilBrown707a6a42016-06-02 16:19:52 +10002297 rcu_read_lock();
2298 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002299 if (rdev &&
NeilBrown707a6a42016-06-02 16:19:52 +10002300 !test_bit(Faulty, &rdev->flags)) {
2301 atomic_inc(&rdev->nr_pending);
2302 rcu_read_unlock();
NeilBrownd8f05d22011-07-28 11:33:00 +10002303 r1_sync_page_io(rdev, sect, s,
2304 conf->tmppage, WRITE);
NeilBrown707a6a42016-06-02 16:19:52 +10002305 rdev_dec_pending(rdev, mddev);
2306 } else
2307 rcu_read_unlock();
NeilBrown867868f2006-10-03 01:15:51 -07002308 }
2309 d = start;
2310 while (d != read_disk) {
2311 char b[BDEVNAME_SIZE];
2312 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002313 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002314 d--;
NeilBrown707a6a42016-06-02 16:19:52 +10002315 rcu_read_lock();
2316 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002317 if (rdev &&
NeilBrownb8cb6b42014-09-18 11:09:04 +10002318 !test_bit(Faulty, &rdev->flags)) {
NeilBrown707a6a42016-06-02 16:19:52 +10002319 atomic_inc(&rdev->nr_pending);
2320 rcu_read_unlock();
NeilBrownd8f05d22011-07-28 11:33:00 +10002321 if (r1_sync_page_io(rdev, sect, s,
2322 conf->tmppage, READ)) {
NeilBrown867868f2006-10-03 01:15:51 -07002323 atomic_add(s, &rdev->corrected_errors);
NeilBrown1d41c212016-11-02 14:16:50 +11002324 pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n",
2325 mdname(mddev), s,
2326 (unsigned long long)(sect +
2327 rdev->data_offset),
2328 bdevname(rdev->bdev, b));
NeilBrown867868f2006-10-03 01:15:51 -07002329 }
NeilBrown707a6a42016-06-02 16:19:52 +10002330 rdev_dec_pending(rdev, mddev);
2331 } else
2332 rcu_read_unlock();
NeilBrown867868f2006-10-03 01:15:51 -07002333 }
2334 sectors -= s;
2335 sect += s;
2336 }
2337}
2338
NeilBrown9f2c9d12011-10-11 16:48:43 +11002339static int narrow_write_error(struct r1bio *r1_bio, int i)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002340{
NeilBrownfd01b882011-10-11 16:47:53 +11002341 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11002342 struct r1conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11002343 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002344
2345 /* bio has the data to be written to device 'i' where
2346 * we just recently had a write error.
2347 * We repeatedly clone the bio and trim down to one block,
2348 * then try the write. Where the write fails we record
2349 * a bad block.
2350 * It is conceivable that the bio doesn't exactly align with
2351 * blocks. We must handle this somehow.
2352 *
2353 * We currently own a reference on the rdev.
2354 */
2355
2356 int block_sectors;
2357 sector_t sector;
2358 int sectors;
2359 int sect_to_write = r1_bio->sectors;
2360 int ok = 1;
2361
2362 if (rdev->badblocks.shift < 0)
2363 return 0;
2364
Nate Daileyab713cd2015-02-12 12:02:09 -05002365 block_sectors = roundup(1 << rdev->badblocks.shift,
2366 bdev_logical_block_size(rdev->bdev) >> 9);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002367 sector = r1_bio->sector;
2368 sectors = ((sector + block_sectors)
2369 & ~(sector_t)(block_sectors - 1))
2370 - sector;
2371
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002372 while (sect_to_write) {
2373 struct bio *wbio;
2374 if (sectors > sect_to_write)
2375 sectors = sect_to_write;
2376 /* Write at 'sector' for 'sectors'*/
2377
Kent Overstreetb7838632012-09-10 15:17:11 -07002378 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
Ming Lei841c1312017-03-17 00:12:31 +08002379 wbio = bio_clone_fast(r1_bio->behind_master_bio,
2380 GFP_NOIO,
Kent Overstreetafeee512018-05-20 18:25:52 -04002381 &mddev->bio_set);
Kent Overstreetb7838632012-09-10 15:17:11 -07002382 } else {
Ming Leid7a10302017-02-14 23:29:03 +08002383 wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO,
Kent Overstreetafeee512018-05-20 18:25:52 -04002384 &mddev->bio_set);
Kent Overstreetb7838632012-09-10 15:17:11 -07002385 }
2386
Mike Christie796a5cf2016-06-05 14:32:07 -05002387 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002388 wbio->bi_iter.bi_sector = r1_bio->sector;
2389 wbio->bi_iter.bi_size = r1_bio->sectors << 9;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002390
Kent Overstreet6678d832013-08-07 11:14:32 -07002391 bio_trim(wbio, sector - r1_bio->sector, sectors);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002392 wbio->bi_iter.bi_sector += rdev->data_offset;
Christoph Hellwig74d46992017-08-23 19:10:32 +02002393 bio_set_dev(wbio, rdev->bdev);
Mike Christie4e49ea42016-06-05 14:31:41 -05002394
2395 if (submit_bio_wait(wbio) < 0)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002396 /* failure! */
2397 ok = rdev_set_badblocks(rdev, sector,
2398 sectors, 0)
2399 && ok;
2400
2401 bio_put(wbio);
2402 sect_to_write -= sectors;
2403 sector += sectors;
2404 sectors = block_sectors;
2405 }
2406 return ok;
2407}
2408
NeilBrowne8096362011-10-11 16:49:05 +11002409static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002410{
2411 int m;
2412 int s = r1_bio->sectors;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002413 for (m = 0; m < conf->raid_disks * 2 ; m++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002414 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002415 struct bio *bio = r1_bio->bios[m];
2416 if (bio->bi_end_io == NULL)
2417 continue;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002418 if (!bio->bi_status &&
NeilBrown62096bc2011-07-28 11:38:13 +10002419 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
NeilBrownc6563a82012-05-21 09:27:00 +10002420 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002421 }
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002422 if (bio->bi_status &&
NeilBrown62096bc2011-07-28 11:38:13 +10002423 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2424 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2425 md_error(conf->mddev, rdev);
2426 }
2427 }
2428 put_buf(r1_bio);
2429 md_done_sync(conf->mddev, s, 1);
2430}
2431
NeilBrowne8096362011-10-11 16:49:05 +11002432static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002433{
colyli@suse.defd768632017-02-18 03:05:56 +08002434 int m, idx;
NeilBrown55ce74d2015-08-14 11:11:10 +10002435 bool fail = false;
colyli@suse.defd768632017-02-18 03:05:56 +08002436
NeilBrown8f19ccb2011-12-23 10:17:56 +11002437 for (m = 0; m < conf->raid_disks * 2 ; m++)
NeilBrown62096bc2011-07-28 11:38:13 +10002438 if (r1_bio->bios[m] == IO_MADE_GOOD) {
NeilBrown3cb03002011-10-11 16:45:26 +11002439 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002440 rdev_clear_badblocks(rdev,
2441 r1_bio->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10002442 r1_bio->sectors, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002443 rdev_dec_pending(rdev, conf->mddev);
2444 } else if (r1_bio->bios[m] != NULL) {
2445 /* This drive got a write error. We need to
2446 * narrow down and record precise write
2447 * errors.
2448 */
NeilBrown55ce74d2015-08-14 11:11:10 +10002449 fail = true;
NeilBrown62096bc2011-07-28 11:38:13 +10002450 if (!narrow_write_error(r1_bio, m)) {
2451 md_error(conf->mddev,
2452 conf->mirrors[m].rdev);
2453 /* an I/O failed, we can't clear the bitmap */
2454 set_bit(R1BIO_Degraded, &r1_bio->state);
2455 }
2456 rdev_dec_pending(conf->mirrors[m].rdev,
2457 conf->mddev);
2458 }
NeilBrown55ce74d2015-08-14 11:11:10 +10002459 if (fail) {
2460 spin_lock_irq(&conf->device_lock);
2461 list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
colyli@suse.defd768632017-02-18 03:05:56 +08002462 idx = sector_to_idx(r1_bio->sector);
colyli@suse.de824e47d2017-02-18 03:05:57 +08002463 atomic_inc(&conf->nr_queued[idx]);
NeilBrown55ce74d2015-08-14 11:11:10 +10002464 spin_unlock_irq(&conf->device_lock);
colyli@suse.de824e47d2017-02-18 03:05:57 +08002465 /*
2466 * In case freeze_array() is waiting for condition
2467 * get_unqueued_pending() == extra to be true.
2468 */
2469 wake_up(&conf->wait_barrier);
NeilBrown55ce74d2015-08-14 11:11:10 +10002470 md_wakeup_thread(conf->mddev->thread);
NeilBrownbd8688a2015-10-24 16:02:16 +11002471 } else {
2472 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2473 close_write(r1_bio);
NeilBrown55ce74d2015-08-14 11:11:10 +10002474 raid_end_bio_io(r1_bio);
NeilBrownbd8688a2015-10-24 16:02:16 +11002475 }
NeilBrown62096bc2011-07-28 11:38:13 +10002476}
2477
NeilBrowne8096362011-10-11 16:49:05 +11002478static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002479{
NeilBrownfd01b882011-10-11 16:47:53 +11002480 struct mddev *mddev = conf->mddev;
NeilBrown62096bc2011-07-28 11:38:13 +10002481 struct bio *bio;
NeilBrown3cb03002011-10-11 16:45:26 +11002482 struct md_rdev *rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002483
2484 clear_bit(R1BIO_ReadError, &r1_bio->state);
2485 /* we got a read error. Maybe the drive is bad. Maybe just
2486 * the block and we can fix it.
2487 * We freeze all other IO, and try reading the block from
2488 * other devices. When we find one, we re-write
2489 * and check it that fixes the read error.
2490 * This is all done synchronously while the array is
2491 * frozen
2492 */
Tomasz Majchrzak7449f692016-10-28 14:45:58 +02002493
2494 bio = r1_bio->bios[r1_bio->read_disk];
Tomasz Majchrzak7449f692016-10-28 14:45:58 +02002495 bio_put(bio);
2496 r1_bio->bios[r1_bio->read_disk] = NULL;
2497
NeilBrown2e52d442016-11-18 16:16:12 +11002498 rdev = conf->mirrors[r1_bio->read_disk].rdev;
2499 if (mddev->ro == 0
2500 && !test_bit(FailFast, &rdev->flags)) {
NeilBrowne2d59922013-06-12 11:01:22 +10002501 freeze_array(conf, 1);
NeilBrown62096bc2011-07-28 11:38:13 +10002502 fix_read_error(conf, r1_bio->read_disk,
2503 r1_bio->sector, r1_bio->sectors);
2504 unfreeze_array(conf);
Gioh Kimb33d1062018-05-02 13:08:11 +02002505 } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) {
2506 md_error(mddev, rdev);
Tomasz Majchrzak7449f692016-10-28 14:45:58 +02002507 } else {
2508 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;
2509 }
2510
NeilBrown2e52d442016-11-18 16:16:12 +11002511 rdev_dec_pending(rdev, conf->mddev);
NeilBrown689389a2017-04-05 14:05:50 +10002512 allow_barrier(conf, r1_bio->sector);
2513 bio = r1_bio->master_bio;
NeilBrown62096bc2011-07-28 11:38:13 +10002514
NeilBrown689389a2017-04-05 14:05:50 +10002515 /* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */
2516 r1_bio->state = 0;
2517 raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio);
NeilBrown62096bc2011-07-28 11:38:13 +10002518}
2519
Shaohua Li4ed87312012-10-11 13:34:00 +11002520static void raid1d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521{
Shaohua Li4ed87312012-10-11 13:34:00 +11002522 struct mddev *mddev = thread->mddev;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002523 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 unsigned long flags;
NeilBrowne8096362011-10-11 16:49:05 +11002525 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002527 struct blk_plug plug;
colyli@suse.defd768632017-02-18 03:05:56 +08002528 int idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529
2530 md_check_recovery(mddev);
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002531
NeilBrown55ce74d2015-08-14 11:11:10 +10002532 if (!list_empty_careful(&conf->bio_end_io_list) &&
Shaohua Li29530792016-12-08 15:48:19 -08002533 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
NeilBrown55ce74d2015-08-14 11:11:10 +10002534 LIST_HEAD(tmp);
2535 spin_lock_irqsave(&conf->device_lock, flags);
colyli@suse.defd768632017-02-18 03:05:56 +08002536 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
2537 list_splice_init(&conf->bio_end_io_list, &tmp);
NeilBrown55ce74d2015-08-14 11:11:10 +10002538 spin_unlock_irqrestore(&conf->device_lock, flags);
2539 while (!list_empty(&tmp)) {
Mikulas Patockaa4527442015-10-01 15:17:43 -04002540 r1_bio = list_first_entry(&tmp, struct r1bio,
2541 retry_list);
NeilBrown55ce74d2015-08-14 11:11:10 +10002542 list_del(&r1_bio->retry_list);
colyli@suse.defd768632017-02-18 03:05:56 +08002543 idx = sector_to_idx(r1_bio->sector);
colyli@suse.de824e47d2017-02-18 03:05:57 +08002544 atomic_dec(&conf->nr_queued[idx]);
NeilBrownbd8688a2015-10-24 16:02:16 +11002545 if (mddev->degraded)
2546 set_bit(R1BIO_Degraded, &r1_bio->state);
2547 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2548 close_write(r1_bio);
NeilBrown55ce74d2015-08-14 11:11:10 +10002549 raid_end_bio_io(r1_bio);
2550 }
2551 }
2552
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002553 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002555
NeilBrown0021b7b2012-07-31 09:08:14 +02002556 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002557
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002559 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002560 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002562 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002563 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 list_del(head->prev);
colyli@suse.defd768632017-02-18 03:05:56 +08002565 idx = sector_to_idx(r1_bio->sector);
colyli@suse.de824e47d2017-02-18 03:05:57 +08002566 atomic_dec(&conf->nr_queued[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 spin_unlock_irqrestore(&conf->device_lock, flags);
2568
2569 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002570 conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10002571 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002572 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002573 test_bit(R1BIO_WriteError, &r1_bio->state))
2574 handle_sync_write_finished(conf, r1_bio);
2575 else
NeilBrown4367af52011-07-28 11:31:49 +10002576 sync_request_write(mddev, r1_bio);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002577 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002578 test_bit(R1BIO_WriteError, &r1_bio->state))
2579 handle_write_finished(conf, r1_bio);
2580 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2581 handle_read_error(conf, r1_bio);
2582 else
NeilBrownc230e7e2017-04-05 14:05:50 +10002583 WARN_ON_ONCE(1);
NeilBrown62096bc2011-07-28 11:38:13 +10002584
NeilBrown1d9d5242009-10-16 15:55:32 +11002585 cond_resched();
Shaohua Li29530792016-12-08 15:48:19 -08002586 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING))
NeilBrownde393cd2011-07-28 11:31:48 +10002587 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002589 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590}
2591
NeilBrowne8096362011-10-11 16:49:05 +11002592static int init_resync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593{
2594 int buffs;
2595
2596 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Kent Overstreetafeee512018-05-20 18:25:52 -04002597 BUG_ON(mempool_initialized(&conf->r1buf_pool));
2598
2599 return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc,
2600 r1buf_pool_free, conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601}
2602
Shaohua Li208410b2017-08-24 17:50:40 -07002603static struct r1bio *raid1_alloc_init_r1buf(struct r1conf *conf)
2604{
Kent Overstreetafeee512018-05-20 18:25:52 -04002605 struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO);
Shaohua Li208410b2017-08-24 17:50:40 -07002606 struct resync_pages *rps;
2607 struct bio *bio;
2608 int i;
2609
2610 for (i = conf->poolinfo->raid_disks; i--; ) {
2611 bio = r1bio->bios[i];
2612 rps = bio->bi_private;
2613 bio_reset(bio);
2614 bio->bi_private = rps;
2615 }
2616 r1bio->master_bio = NULL;
2617 return r1bio;
2618}
2619
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620/*
2621 * perform a "sync" on one "block"
2622 *
2623 * We need to make sure that no normal I/O request - particularly write
2624 * requests - conflict with active sync requests.
2625 *
2626 * This is achieved by tracking pending requests and a 'barrier' concept
2627 * that can be installed to exclude normal IO requests.
2628 */
2629
Shaohua Li849674e2016-01-20 13:52:20 -08002630static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2631 int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632{
NeilBrowne8096362011-10-11 16:49:05 +11002633 struct r1conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002634 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 struct bio *bio;
2636 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08002637 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08002639 int wonly = -1;
2640 int write_targets = 0, read_targets = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11002641 sector_t sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07002642 int still_degraded = 0;
NeilBrown06f60382011-07-28 11:31:48 +10002643 int good_sectors = RESYNC_SECTORS;
2644 int min_bad = 0; /* number of sectors that are bad in all devices */
colyli@suse.defd768632017-02-18 03:05:56 +08002645 int idx = sector_to_idx(sector_nr);
Ming Lei022e5102017-07-14 16:14:42 +08002646 int page_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
Kent Overstreetafeee512018-05-20 18:25:52 -04002648 if (!mempool_initialized(&conf->r1buf_pool))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002650 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651
Andre Noll58c0fed2009-03-31 14:33:13 +11002652 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002654 /* If we aborted, we need to abort the
2655 * sync on the 'current' bitmap chunk (there will
2656 * only be one in raid1 resync.
2657 * We can find the current addess in mddev->curr_resync
2658 */
NeilBrown6a806c52005-07-15 03:56:35 -07002659 if (mddev->curr_resync < max_sector) /* aborted */
Andy Shevchenkoe64e40182018-08-01 15:20:50 -07002660 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2661 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07002662 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07002663 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07002664
Andy Shevchenkoe64e40182018-08-01 15:20:50 -07002665 md_bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 close_sync(conf);
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002667
2668 if (mddev_is_clustered(mddev)) {
2669 conf->cluster_sync_low = 0;
2670 conf->cluster_sync_high = 0;
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 return 0;
2673 }
2674
NeilBrown07d84d102006-06-26 00:27:56 -07002675 if (mddev->bitmap == NULL &&
2676 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07002677 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07002678 conf->fullsync == 0) {
2679 *skipped = 1;
2680 return max_sector - sector_nr;
2681 }
NeilBrown6394cca2006-08-27 01:23:50 -07002682 /* before building a request, check if we can skip these blocks..
2683 * This call the bitmap_start_sync doesn't actually record anything
2684 */
Andy Shevchenkoe64e40182018-08-01 15:20:50 -07002685 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de485f2005-11-08 21:39:38 -08002686 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002687 /* We can skip this block, and probably several more */
2688 *skipped = 1;
2689 return sync_blocks;
2690 }
NeilBrown17999be2006-01-06 00:20:12 -08002691
Tomasz Majchrzak7ac50442016-06-13 15:51:19 +02002692 /*
2693 * If there is non-resync activity waiting for a turn, then let it
2694 * though before starting on this new sync request.
2695 */
colyli@suse.de824e47d2017-02-18 03:05:57 +08002696 if (atomic_read(&conf->nr_waiting[idx]))
Tomasz Majchrzak7ac50442016-06-13 15:51:19 +02002697 schedule_timeout_uninterruptible(1);
2698
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002699 /* we are incrementing sector_nr below. To be safe, we check against
2700 * sector_nr + two times RESYNC_SECTORS
2701 */
2702
Andy Shevchenkoe64e40182018-08-01 15:20:50 -07002703 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr,
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002704 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
NeilBrown17999be2006-01-06 00:20:12 -08002705
Yufen Yu8c242592018-04-09 09:50:44 +08002706
2707 if (raise_barrier(conf, sector_nr))
2708 return 0;
2709
2710 r1_bio = raid1_alloc_init_r1buf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
NeilBrown3e198f72006-01-06 00:20:21 -08002712 rcu_read_lock();
2713 /*
2714 * If we get a correctably read error during resync or recovery,
2715 * we might want to read from a different device. So we
2716 * flag all drives that could conceivably be read from for READ,
2717 * and any others (which will be non-In_sync devices) for WRITE.
2718 * If a read fails, we try reading from something else for which READ
2719 * is OK.
2720 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 r1_bio->mddev = mddev;
2723 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07002724 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 set_bit(R1BIO_IsSync, &r1_bio->state);
colyli@suse.defd768632017-02-18 03:05:56 +08002726 /* make sure good_sectors won't go across barrier unit boundary */
2727 good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
NeilBrown8f19ccb2011-12-23 10:17:56 +11002729 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002730 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 bio = r1_bio->bios[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732
NeilBrown3e198f72006-01-06 00:20:21 -08002733 rdev = rcu_dereference(conf->mirrors[i].rdev);
2734 if (rdev == NULL ||
NeilBrown06f60382011-07-28 11:31:48 +10002735 test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11002736 if (i < conf->raid_disks)
2737 still_degraded = 1;
NeilBrown3e198f72006-01-06 00:20:21 -08002738 } else if (!test_bit(In_sync, &rdev->flags)) {
Mike Christie796a5cf2016-06-05 14:32:07 -05002739 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 bio->bi_end_io = end_sync_write;
2741 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08002742 } else {
2743 /* may need to read from here */
NeilBrown06f60382011-07-28 11:31:48 +10002744 sector_t first_bad = MaxSector;
2745 int bad_sectors;
2746
2747 if (is_badblock(rdev, sector_nr, good_sectors,
2748 &first_bad, &bad_sectors)) {
2749 if (first_bad > sector_nr)
2750 good_sectors = first_bad - sector_nr;
2751 else {
2752 bad_sectors -= (sector_nr - first_bad);
2753 if (min_bad == 0 ||
2754 min_bad > bad_sectors)
2755 min_bad = bad_sectors;
2756 }
NeilBrown3e198f72006-01-06 00:20:21 -08002757 }
NeilBrown06f60382011-07-28 11:31:48 +10002758 if (sector_nr < first_bad) {
2759 if (test_bit(WriteMostly, &rdev->flags)) {
2760 if (wonly < 0)
2761 wonly = i;
2762 } else {
2763 if (disk < 0)
2764 disk = i;
2765 }
Mike Christie796a5cf2016-06-05 14:32:07 -05002766 bio_set_op_attrs(bio, REQ_OP_READ, 0);
NeilBrown06f60382011-07-28 11:31:48 +10002767 bio->bi_end_io = end_sync_read;
2768 read_targets++;
Alexander Lyakasd57368a2012-07-17 13:17:55 +03002769 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2770 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2771 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2772 /*
2773 * The device is suitable for reading (InSync),
2774 * but has bad block(s) here. Let's try to correct them,
2775 * if we are doing resync or repair. Otherwise, leave
2776 * this device alone for this sync request.
2777 */
Mike Christie796a5cf2016-06-05 14:32:07 -05002778 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Alexander Lyakasd57368a2012-07-17 13:17:55 +03002779 bio->bi_end_io = end_sync_write;
2780 write_targets++;
NeilBrown06f60382011-07-28 11:31:48 +10002781 }
NeilBrown3e198f72006-01-06 00:20:21 -08002782 }
NeilBrown06f60382011-07-28 11:31:48 +10002783 if (bio->bi_end_io) {
2784 atomic_inc(&rdev->nr_pending);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002785 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
Christoph Hellwig74d46992017-08-23 19:10:32 +02002786 bio_set_dev(bio, rdev->bdev);
NeilBrown2e52d442016-11-18 16:16:12 +11002787 if (test_bit(FailFast, &rdev->flags))
2788 bio->bi_opf |= MD_FAILFAST;
NeilBrown06f60382011-07-28 11:31:48 +10002789 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 }
NeilBrown3e198f72006-01-06 00:20:21 -08002791 rcu_read_unlock();
2792 if (disk < 0)
2793 disk = wonly;
2794 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07002795
NeilBrown06f60382011-07-28 11:31:48 +10002796 if (read_targets == 0 && min_bad > 0) {
2797 /* These sectors are bad on all InSync devices, so we
2798 * need to mark them bad on all write targets
2799 */
2800 int ok = 1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002801 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
NeilBrown06f60382011-07-28 11:31:48 +10002802 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
majianpenga42f9d82012-04-02 01:04:19 +10002803 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown06f60382011-07-28 11:31:48 +10002804 ok = rdev_set_badblocks(rdev, sector_nr,
2805 min_bad, 0
2806 ) && ok;
2807 }
Shaohua Li29530792016-12-08 15:48:19 -08002808 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
NeilBrown06f60382011-07-28 11:31:48 +10002809 *skipped = 1;
2810 put_buf(r1_bio);
2811
2812 if (!ok) {
2813 /* Cannot record the badblocks, so need to
2814 * abort the resync.
2815 * If there are multiple read targets, could just
2816 * fail the really bad ones ???
2817 */
2818 conf->recovery_disabled = mddev->recovery_disabled;
2819 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2820 return 0;
2821 } else
2822 return min_bad;
2823
2824 }
2825 if (min_bad > 0 && min_bad < good_sectors) {
2826 /* only resync enough to reach the next bad->good
2827 * transition */
2828 good_sectors = min_bad;
2829 }
2830
NeilBrown3e198f72006-01-06 00:20:21 -08002831 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2832 /* extra read targets are also write targets */
2833 write_targets += read_targets-1;
2834
2835 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 /* There is nowhere to write, so all non-sync
2837 * drives must be failed - so we are finished
2838 */
NeilBrownb7219cc2012-07-31 10:05:34 +10002839 sector_t rv;
2840 if (min_bad > 0)
2841 max_sector = sector_nr + min_bad;
2842 rv = max_sector - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07002843 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 return rv;
2846 }
2847
NeilBrownc6207272008-02-06 01:39:52 -08002848 if (max_sector > mddev->resync_max)
2849 max_sector = mddev->resync_max; /* Don't do IO beyond here */
NeilBrown06f60382011-07-28 11:31:48 +10002850 if (max_sector > sector_nr + good_sectors)
2851 max_sector = sector_nr + good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07002853 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 do {
2855 struct page *page;
2856 int len = PAGE_SIZE;
2857 if (sector_nr + (len>>9) > max_sector)
2858 len = (max_sector - sector_nr) << 9;
2859 if (len == 0)
2860 break;
NeilBrown6a806c52005-07-15 03:56:35 -07002861 if (sync_blocks == 0) {
Andy Shevchenkoe64e40182018-08-01 15:20:50 -07002862 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr,
2863 &sync_blocks, still_degraded) &&
NeilBrowne5de485f2005-11-08 21:39:38 -08002864 !conf->fullsync &&
2865 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07002866 break;
NeilBrown7571ae82010-10-07 11:54:46 +11002867 if ((len >> 9) > sync_blocks)
NeilBrown6a806c52005-07-15 03:56:35 -07002868 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07002869 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002870
NeilBrown8f19ccb2011-12-23 10:17:56 +11002871 for (i = 0 ; i < conf->raid_disks * 2; i++) {
Ming Lei98d30c52017-03-17 00:12:26 +08002872 struct resync_pages *rp;
2873
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 bio = r1_bio->bios[i];
Ming Lei98d30c52017-03-17 00:12:26 +08002875 rp = get_resync_pages(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 if (bio->bi_end_io) {
Ming Lei022e5102017-07-14 16:14:42 +08002877 page = resync_fetch_page(rp, page_idx);
Ming Leic85ba142017-03-17 00:12:22 +08002878
2879 /*
2880 * won't fail because the vec table is big
2881 * enough to hold all these pages
2882 */
2883 bio_add_page(bio, page, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 }
2885 }
2886 nr_sectors += len>>9;
2887 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07002888 sync_blocks -= (len>>9);
Ming Lei022e5102017-07-14 16:14:42 +08002889 } while (++page_idx < RESYNC_PAGES);
Ming Lei98d30c52017-03-17 00:12:26 +08002890
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 r1_bio->sectors = nr_sectors;
2892
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10002893 if (mddev_is_clustered(mddev) &&
2894 conf->cluster_sync_high < sector_nr + nr_sectors) {
2895 conf->cluster_sync_low = mddev->curr_resync_completed;
2896 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2897 /* Send resync message */
2898 md_cluster_ops->resync_info_update(mddev,
2899 conf->cluster_sync_low,
2900 conf->cluster_sync_high);
2901 }
2902
NeilBrownd11c1712006-01-06 00:20:26 -08002903 /* For a user-requested sync, we read all readable devices and do a
2904 * compare
2905 */
2906 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2907 atomic_set(&r1_bio->remaining, read_targets);
NeilBrown2d4f4f32012-07-09 11:34:13 +10002908 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
NeilBrownd11c1712006-01-06 00:20:26 -08002909 bio = r1_bio->bios[i];
2910 if (bio->bi_end_io == end_sync_read) {
NeilBrown2d4f4f32012-07-09 11:34:13 +10002911 read_targets--;
Christoph Hellwig74d46992017-08-23 19:10:32 +02002912 md_sync_acct_bio(bio, nr_sectors);
NeilBrown2e52d442016-11-18 16:16:12 +11002913 if (read_targets == 1)
2914 bio->bi_opf &= ~MD_FAILFAST;
NeilBrownd11c1712006-01-06 00:20:26 -08002915 generic_make_request(bio);
2916 }
2917 }
2918 } else {
2919 atomic_set(&r1_bio->remaining, 1);
2920 bio = r1_bio->bios[r1_bio->read_disk];
Christoph Hellwig74d46992017-08-23 19:10:32 +02002921 md_sync_acct_bio(bio, nr_sectors);
NeilBrown2e52d442016-11-18 16:16:12 +11002922 if (read_targets == 1)
2923 bio->bi_opf &= ~MD_FAILFAST;
NeilBrownd11c1712006-01-06 00:20:26 -08002924 generic_make_request(bio);
NeilBrownd11c1712006-01-06 00:20:26 -08002925 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 return nr_sectors;
2927}
2928
NeilBrownfd01b882011-10-11 16:47:53 +11002929static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07002930{
2931 if (sectors)
2932 return sectors;
2933
2934 return mddev->dev_sectors;
2935}
2936
NeilBrowne8096362011-10-11 16:49:05 +11002937static struct r1conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938{
NeilBrowne8096362011-10-11 16:49:05 +11002939 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002940 int i;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002941 struct raid1_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11002942 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002943 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944
NeilBrowne8096362011-10-11 16:49:05 +11002945 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11002947 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948
colyli@suse.defd768632017-02-18 03:05:56 +08002949 conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR,
colyli@suse.de824e47d2017-02-18 03:05:57 +08002950 sizeof(atomic_t), GFP_KERNEL);
colyli@suse.defd768632017-02-18 03:05:56 +08002951 if (!conf->nr_pending)
2952 goto abort;
2953
2954 conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR,
colyli@suse.de824e47d2017-02-18 03:05:57 +08002955 sizeof(atomic_t), GFP_KERNEL);
colyli@suse.defd768632017-02-18 03:05:56 +08002956 if (!conf->nr_waiting)
2957 goto abort;
2958
2959 conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR,
colyli@suse.de824e47d2017-02-18 03:05:57 +08002960 sizeof(atomic_t), GFP_KERNEL);
colyli@suse.defd768632017-02-18 03:05:56 +08002961 if (!conf->nr_queued)
2962 goto abort;
2963
2964 conf->barrier = kcalloc(BARRIER_BUCKETS_NR,
colyli@suse.de824e47d2017-02-18 03:05:57 +08002965 sizeof(atomic_t), GFP_KERNEL);
colyli@suse.defd768632017-02-18 03:05:56 +08002966 if (!conf->barrier)
2967 goto abort;
2968
Kees Cook6396bb22018-06-12 14:03:40 -07002969 conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info),
2970 mddev->raid_disks, 2),
2971 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11002973 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974
NeilBrownddaf22a2006-01-06 00:20:19 -08002975 conf->tmppage = alloc_page(GFP_KERNEL);
2976 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11002977 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08002978
NeilBrown709ae482009-12-14 12:49:51 +11002979 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11002981 goto abort;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002982 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
Marcos Paulo de Souza3f677f92019-06-14 15:41:04 -07002983 err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc,
Marcos Paulo de Souzac7afa802019-06-14 15:41:10 -07002984 rbio_pool_free, conf->poolinfo);
Kent Overstreetafeee512018-05-20 18:25:52 -04002985 if (err)
NeilBrown709ae482009-12-14 12:49:51 +11002986 goto abort;
2987
Kent Overstreetafeee512018-05-20 18:25:52 -04002988 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0);
2989 if (err)
NeilBrownc230e7e2017-04-05 14:05:50 +10002990 goto abort;
2991
NeilBrowned9bfdf2009-10-16 15:55:44 +11002992 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993
NeilBrownc19d5792011-12-23 10:17:57 +11002994 err = -EINVAL;
Neil Browne7e72bf2008-05-14 16:05:54 -07002995 spin_lock_init(&conf->device_lock);
NeilBrowndafb20f2012-03-19 12:46:39 +11002996 rdev_for_each(rdev, mddev) {
NeilBrown709ae482009-12-14 12:49:51 +11002997 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 if (disk_idx >= mddev->raid_disks
2999 || disk_idx < 0)
3000 continue;
NeilBrownc19d5792011-12-23 10:17:57 +11003001 if (test_bit(Replacement, &rdev->flags))
NeilBrown02b898f2012-10-31 11:42:03 +11003002 disk = conf->mirrors + mddev->raid_disks + disk_idx;
NeilBrownc19d5792011-12-23 10:17:57 +11003003 else
3004 disk = conf->mirrors + disk_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005
NeilBrownc19d5792011-12-23 10:17:57 +11003006 if (disk->rdev)
3007 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 disk->rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 disk->head_position = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +10003010 disk->seq_start = MaxSector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 }
3012 conf->raid_disks = mddev->raid_disks;
3013 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 INIT_LIST_HEAD(&conf->retry_list);
NeilBrown55ce74d2015-08-14 11:11:10 +10003015 INIT_LIST_HEAD(&conf->bio_end_io_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016
3017 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08003018 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019
NeilBrown191ea9b2005-06-21 17:17:23 -07003020 bio_list_init(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +11003021 conf->pending_count = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11003022 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown191ea9b2005-06-21 17:17:23 -07003023
NeilBrownc19d5792011-12-23 10:17:57 +11003024 err = -EIO;
NeilBrown8f19ccb2011-12-23 10:17:56 +11003025 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026
3027 disk = conf->mirrors + i;
3028
NeilBrownc19d5792011-12-23 10:17:57 +11003029 if (i < conf->raid_disks &&
3030 disk[conf->raid_disks].rdev) {
3031 /* This slot has a replacement. */
3032 if (!disk->rdev) {
3033 /* No original, just make the replacement
3034 * a recovering spare
3035 */
3036 disk->rdev =
3037 disk[conf->raid_disks].rdev;
3038 disk[conf->raid_disks].rdev = NULL;
3039 } else if (!test_bit(In_sync, &disk->rdev->flags))
3040 /* Original is not in_sync - bad */
3041 goto abort;
3042 }
3043
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003044 if (!disk->rdev ||
3045 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 disk->head_position = 0;
Jonathan Brassow4f0a5e02012-05-22 13:55:31 +10003047 if (disk->rdev &&
3048 (disk->rdev->saved_raid_disk < 0))
NeilBrown918f0232007-08-22 14:01:52 -07003049 conf->fullsync = 1;
Shaohua Libe4d3282012-07-31 10:03:53 +10003050 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 }
NeilBrown709ae482009-12-14 12:49:51 +11003052
NeilBrown709ae482009-12-14 12:49:51 +11003053 err = -ENOMEM;
NeilBrown02326052012-07-03 15:56:52 +10003054 conf->thread = md_register_thread(raid1d, mddev, "raid1");
NeilBrown1d41c212016-11-02 14:16:50 +11003055 if (!conf->thread)
NeilBrown709ae482009-12-14 12:49:51 +11003056 goto abort;
NeilBrown191ea9b2005-06-21 17:17:23 -07003057
NeilBrown709ae482009-12-14 12:49:51 +11003058 return conf;
3059
3060 abort:
3061 if (conf) {
Kent Overstreetafeee512018-05-20 18:25:52 -04003062 mempool_exit(&conf->r1bio_pool);
NeilBrown709ae482009-12-14 12:49:51 +11003063 kfree(conf->mirrors);
3064 safe_put_page(conf->tmppage);
3065 kfree(conf->poolinfo);
colyli@suse.defd768632017-02-18 03:05:56 +08003066 kfree(conf->nr_pending);
3067 kfree(conf->nr_waiting);
3068 kfree(conf->nr_queued);
3069 kfree(conf->barrier);
Kent Overstreetafeee512018-05-20 18:25:52 -04003070 bioset_exit(&conf->bio_split);
NeilBrown709ae482009-12-14 12:49:51 +11003071 kfree(conf);
3072 }
3073 return ERR_PTR(err);
3074}
3075
NeilBrownafa0f552014-12-15 12:56:58 +11003076static void raid1_free(struct mddev *mddev, void *priv);
Shaohua Li849674e2016-01-20 13:52:20 -08003077static int raid1_run(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11003078{
NeilBrowne8096362011-10-11 16:49:05 +11003079 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11003080 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11003081 struct md_rdev *rdev;
majianpeng5220ea12012-04-02 09:48:38 +10003082 int ret;
Shaohua Li2ff8cc22012-10-11 13:28:54 +11003083 bool discard_supported = false;
NeilBrown709ae482009-12-14 12:49:51 +11003084
3085 if (mddev->level != 1) {
NeilBrown1d41c212016-11-02 14:16:50 +11003086 pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n",
3087 mdname(mddev), mddev->level);
NeilBrown709ae482009-12-14 12:49:51 +11003088 return -EIO;
3089 }
3090 if (mddev->reshape_position != MaxSector) {
NeilBrown1d41c212016-11-02 14:16:50 +11003091 pr_warn("md/raid1:%s: reshape_position set but not supported\n",
3092 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11003093 return -EIO;
3094 }
NeilBrowna415c0f2017-06-05 16:05:13 +10003095 if (mddev_init_writes_pending(mddev) < 0)
3096 return -ENOMEM;
NeilBrown709ae482009-12-14 12:49:51 +11003097 /*
3098 * copy the already verified devices into our private RAID1
3099 * bookkeeping area. [whatever we allocate in run(),
NeilBrownafa0f552014-12-15 12:56:58 +11003100 * should be freed in raid1_free()]
NeilBrown709ae482009-12-14 12:49:51 +11003101 */
3102 if (mddev->private == NULL)
3103 conf = setup_conf(mddev);
3104 else
3105 conf = mddev->private;
3106
3107 if (IS_ERR(conf))
3108 return PTR_ERR(conf);
3109
Christoph Hellwig3deff1a2017-04-05 19:21:03 +02003110 if (mddev->queue) {
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07003111 blk_queue_max_write_same_sectors(mddev->queue, 0);
Christoph Hellwig3deff1a2017-04-05 19:21:03 +02003112 blk_queue_max_write_zeroes_sectors(mddev->queue, 0);
3113 }
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07003114
NeilBrowndafb20f2012-03-19 12:46:39 +11003115 rdev_for_each(rdev, mddev) {
Jonathan Brassow1ed72422011-06-07 17:50:35 -05003116 if (!mddev->gendisk)
3117 continue;
NeilBrown709ae482009-12-14 12:49:51 +11003118 disk_stack_limits(mddev->gendisk, rdev->bdev,
3119 rdev->data_offset << 9);
Shaohua Li2ff8cc22012-10-11 13:28:54 +11003120 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
3121 discard_supported = true;
NeilBrown709ae482009-12-14 12:49:51 +11003122 }
3123
3124 mddev->degraded = 0;
Yufen Yuebfeb442019-06-14 15:41:08 -07003125 for (i = 0; i < conf->raid_disks; i++)
NeilBrown709ae482009-12-14 12:49:51 +11003126 if (conf->mirrors[i].rdev == NULL ||
3127 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
3128 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
3129 mddev->degraded++;
3130
3131 if (conf->raid_disks - mddev->degraded == 1)
3132 mddev->recovery_cp = MaxSector;
3133
Andre Noll8c6ac8682009-06-18 08:48:06 +10003134 if (mddev->recovery_cp != MaxSector)
NeilBrown1d41c212016-11-02 14:16:50 +11003135 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n",
3136 mdname(mddev));
3137 pr_info("md/raid1:%s: active with %d out of %d mirrors\n",
NeilBrownf72ffdd2014-09-30 14:23:59 +10003138 mdname(mddev), mddev->raid_disks - mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11003140
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 /*
3142 * Ok, everything is just fine now
3143 */
NeilBrown709ae482009-12-14 12:49:51 +11003144 mddev->thread = conf->thread;
3145 conf->thread = NULL;
3146 mddev->private = conf;
NeilBrown46533ff2016-11-18 16:16:11 +11003147 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);
NeilBrown709ae482009-12-14 12:49:51 +11003148
Dan Williams1f403622009-03-31 14:59:03 +11003149 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150
Jonathan Brassow1ed72422011-06-07 17:50:35 -05003151 if (mddev->queue) {
Shaohua Li2ff8cc22012-10-11 13:28:54 +11003152 if (discard_supported)
Bart Van Assche8b904b52018-03-07 17:10:10 -08003153 blk_queue_flag_set(QUEUE_FLAG_DISCARD,
Shaohua Li2ff8cc22012-10-11 13:28:54 +11003154 mddev->queue);
3155 else
Bart Van Assche8b904b52018-03-07 17:10:10 -08003156 blk_queue_flag_clear(QUEUE_FLAG_DISCARD,
Shaohua Li2ff8cc22012-10-11 13:28:54 +11003157 mddev->queue);
Jonathan Brassow1ed72422011-06-07 17:50:35 -05003158 }
majianpeng5220ea12012-04-02 09:48:38 +10003159
Yufen Yuebfeb442019-06-14 15:41:08 -07003160 ret = md_integrity_register(mddev);
NeilBrown5aa61f42014-12-15 12:56:57 +11003161 if (ret) {
3162 md_unregister_thread(&mddev->thread);
NeilBrownafa0f552014-12-15 12:56:58 +11003163 raid1_free(mddev, conf);
NeilBrown5aa61f42014-12-15 12:56:57 +11003164 }
majianpeng5220ea12012-04-02 09:48:38 +10003165 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166}
3167
NeilBrownafa0f552014-12-15 12:56:58 +11003168static void raid1_free(struct mddev *mddev, void *priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169{
NeilBrownafa0f552014-12-15 12:56:58 +11003170 struct r1conf *conf = priv;
NeilBrown4b6d2872005-09-09 16:23:47 -07003171
Kent Overstreetafeee512018-05-20 18:25:52 -04003172 mempool_exit(&conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07003173 kfree(conf->mirrors);
Hirokazu Takahashi0fea7ed2013-04-24 11:42:44 +10003174 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07003175 kfree(conf->poolinfo);
colyli@suse.defd768632017-02-18 03:05:56 +08003176 kfree(conf->nr_pending);
3177 kfree(conf->nr_waiting);
3178 kfree(conf->nr_queued);
3179 kfree(conf->barrier);
Kent Overstreetafeee512018-05-20 18:25:52 -04003180 bioset_exit(&conf->bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182}
3183
NeilBrownfd01b882011-10-11 16:47:53 +11003184static int raid1_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185{
3186 /* no resync is happening, and there is enough space
3187 * on all devices, so we can resize.
3188 * We need to make sure resync covers any new space.
3189 * If the array is shrinking we should possibly wait until
3190 * any io in the removed space completes, but it hardly seems
3191 * worth it.
3192 */
NeilBrowna4a61252012-05-22 13:55:27 +10003193 sector_t newsize = raid1_size(mddev, sectors, 0);
3194 if (mddev->external_size &&
3195 mddev->array_sectors > newsize)
Dan Williamsb522adc2009-03-31 15:00:31 +11003196 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10003197 if (mddev->bitmap) {
Andy Shevchenkoe64e40182018-08-01 15:20:50 -07003198 int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0);
NeilBrowna4a61252012-05-22 13:55:27 +10003199 if (ret)
3200 return ret;
3201 }
3202 md_set_array_sectors(mddev, newsize);
Dan Williamsb522adc2009-03-31 15:00:31 +11003203 if (sectors > mddev->dev_sectors &&
NeilBrownb0986362011-05-11 15:52:21 +10003204 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11003205 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3207 }
Dan Williamsb522adc2009-03-31 15:00:31 +11003208 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07003209 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 return 0;
3211}
3212
NeilBrownfd01b882011-10-11 16:47:53 +11003213static int raid1_reshape(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214{
3215 /* We need to:
3216 * 1/ resize the r1bio_pool
3217 * 2/ resize conf->mirrors
3218 *
3219 * We allocate a new r1bio_pool if we can.
3220 * Then raise a device barrier and wait until all IO stops.
3221 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07003222 *
3223 * At the same time, we "pack" the devices so that all the missing
3224 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 */
Kent Overstreetafeee512018-05-20 18:25:52 -04003226 mempool_t newpool, oldpool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 struct pool_info *newpoolinfo;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10003228 struct raid1_info *newmirrors;
NeilBrowne8096362011-10-11 16:49:05 +11003229 struct r1conf *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08003230 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07003231 unsigned long flags;
Artur Paszkiewicz2214c262017-05-08 11:56:55 +02003232 int d, d2;
Kent Overstreetafeee512018-05-20 18:25:52 -04003233 int ret;
3234
3235 memset(&newpool, 0, sizeof(newpool));
3236 memset(&oldpool, 0, sizeof(oldpool));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237
NeilBrown63c70c42006-03-27 01:18:13 -08003238 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10003239 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08003240 mddev->layout != mddev->new_layout ||
3241 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10003242 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08003243 mddev->new_layout = mddev->layout;
3244 mddev->new_level = mddev->level;
3245 return -EINVAL;
3246 }
3247
Artur Paszkiewicz2214c262017-05-08 11:56:55 +02003248 if (!mddev_is_clustered(mddev))
3249 md_allow_write(mddev);
NeilBrown2a2275d2007-01-26 00:57:11 -08003250
NeilBrown63c70c42006-03-27 01:18:13 -08003251 raid_disks = mddev->raid_disks + mddev->delta_disks;
3252
NeilBrown6ea9c072005-06-21 17:17:09 -07003253 if (raid_disks < conf->raid_disks) {
3254 cnt=0;
3255 for (d= 0; d < conf->raid_disks; d++)
3256 if (conf->mirrors[d].rdev)
3257 cnt++;
3258 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07003260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261
3262 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3263 if (!newpoolinfo)
3264 return -ENOMEM;
3265 newpoolinfo->mddev = mddev;
NeilBrown8f19ccb2011-12-23 10:17:56 +11003266 newpoolinfo->raid_disks = raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267
Marcos Paulo de Souza3f677f92019-06-14 15:41:04 -07003268 ret = mempool_init(&newpool, NR_RAID_BIOS, r1bio_pool_alloc,
Marcos Paulo de Souzac7afa802019-06-14 15:41:10 -07003269 rbio_pool_free, newpoolinfo);
Kent Overstreetafeee512018-05-20 18:25:52 -04003270 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 kfree(newpoolinfo);
Kent Overstreetafeee512018-05-20 18:25:52 -04003272 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 }
Kees Cook6396bb22018-06-12 14:03:40 -07003274 newmirrors = kzalloc(array3_size(sizeof(struct raid1_info),
3275 raid_disks, 2),
NeilBrown8f19ccb2011-12-23 10:17:56 +11003276 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277 if (!newmirrors) {
3278 kfree(newpoolinfo);
Kent Overstreetafeee512018-05-20 18:25:52 -04003279 mempool_exit(&newpool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 return -ENOMEM;
3281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282
NeilBrowne2d59922013-06-12 11:01:22 +10003283 freeze_array(conf, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284
3285 /* ok, everything is stopped */
3286 oldpool = conf->r1bio_pool;
3287 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07003288
NeilBrowna88aa782007-08-22 14:01:53 -07003289 for (d = d2 = 0; d < conf->raid_disks; d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11003290 struct md_rdev *rdev = conf->mirrors[d].rdev;
NeilBrowna88aa782007-08-22 14:01:53 -07003291 if (rdev && rdev->raid_disk != d2) {
Namhyung Kim36fad852011-07-27 11:00:36 +10003292 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna88aa782007-08-22 14:01:53 -07003293 rdev->raid_disk = d2;
Namhyung Kim36fad852011-07-27 11:00:36 +10003294 sysfs_unlink_rdev(mddev, rdev);
3295 if (sysfs_link_rdev(mddev, rdev))
NeilBrown1d41c212016-11-02 14:16:50 +11003296 pr_warn("md/raid1:%s: cannot register rd%d\n",
3297 mdname(mddev), rdev->raid_disk);
NeilBrown6ea9c072005-06-21 17:17:09 -07003298 }
NeilBrowna88aa782007-08-22 14:01:53 -07003299 if (rdev)
3300 newmirrors[d2++].rdev = rdev;
3301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 kfree(conf->mirrors);
3303 conf->mirrors = newmirrors;
3304 kfree(conf->poolinfo);
3305 conf->poolinfo = newpoolinfo;
3306
NeilBrownc04be0a2006-10-03 01:15:53 -07003307 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07003309 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08003311 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312
NeilBrowne2d59922013-06-12 11:01:22 +10003313 unfreeze_array(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314
NeilBrown985ca972015-07-06 12:26:57 +10003315 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3317 md_wakeup_thread(mddev->thread);
3318
Kent Overstreetafeee512018-05-20 18:25:52 -04003319 mempool_exit(&oldpool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 return 0;
3321}
3322
NeilBrownb03e0cc2017-10-19 12:49:15 +11003323static void raid1_quiesce(struct mddev *mddev, int quiesce)
NeilBrown36fa3062005-09-09 16:23:45 -07003324{
NeilBrowne8096362011-10-11 16:49:05 +11003325 struct r1conf *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07003326
NeilBrownb03e0cc2017-10-19 12:49:15 +11003327 if (quiesce)
majianpeng07169fd2013-11-14 15:16:18 +11003328 freeze_array(conf, 0);
NeilBrownb03e0cc2017-10-19 12:49:15 +11003329 else
majianpeng07169fd2013-11-14 15:16:18 +11003330 unfreeze_array(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07003331}
3332
NeilBrownfd01b882011-10-11 16:47:53 +11003333static void *raid1_takeover(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11003334{
3335 /* raid1 can take over:
3336 * raid5 with 2 devices, any layout or chunk size
3337 */
3338 if (mddev->level == 5 && mddev->raid_disks == 2) {
NeilBrowne8096362011-10-11 16:49:05 +11003339 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11003340 mddev->new_level = 1;
3341 mddev->new_layout = 0;
3342 mddev->new_chunk_sectors = 0;
3343 conf = setup_conf(mddev);
Shaohua Li6995f0b2016-12-08 15:48:17 -08003344 if (!IS_ERR(conf)) {
majianpeng07169fd2013-11-14 15:16:18 +11003345 /* Array must appear to be quiesced */
3346 conf->array_frozen = 1;
Shaohua Li394ed8e2017-01-04 16:10:19 -08003347 mddev_clear_unsupported_flags(mddev,
3348 UNSUPPORTED_MDDEV_FLAGS);
Shaohua Li6995f0b2016-12-08 15:48:17 -08003349 }
NeilBrown709ae482009-12-14 12:49:51 +11003350 return conf;
3351 }
3352 return ERR_PTR(-EINVAL);
3353}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354
NeilBrown84fc4b52011-10-11 16:49:58 +11003355static struct md_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356{
3357 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08003358 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359 .owner = THIS_MODULE,
Shaohua Li849674e2016-01-20 13:52:20 -08003360 .make_request = raid1_make_request,
3361 .run = raid1_run,
NeilBrownafa0f552014-12-15 12:56:58 +11003362 .free = raid1_free,
Shaohua Li849674e2016-01-20 13:52:20 -08003363 .status = raid1_status,
3364 .error_handler = raid1_error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 .hot_add_disk = raid1_add_disk,
3366 .hot_remove_disk= raid1_remove_disk,
3367 .spare_active = raid1_spare_active,
Shaohua Li849674e2016-01-20 13:52:20 -08003368 .sync_request = raid1_sync_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07003370 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08003371 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07003372 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11003373 .takeover = raid1_takeover,
NeilBrown5c675f82014-12-15 12:56:56 +11003374 .congested = raid1_congested,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375};
3376
3377static int __init raid_init(void)
3378{
NeilBrown2604b702006-01-06 00:20:36 -08003379 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380}
3381
3382static void raid_exit(void)
3383{
NeilBrown2604b702006-01-06 00:20:36 -08003384 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385}
3386
3387module_init(raid_init);
3388module_exit(raid_exit);
3389MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11003390MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08003392MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08003393MODULE_ALIAS("md-level-1");
NeilBrown34db0cd2011-10-11 16:50:01 +11003394
3395module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);