Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * raid1.c : Multiple Devices driver for Linux |
| 3 | * |
| 4 | * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat |
| 5 | * |
| 6 | * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman |
| 7 | * |
| 8 | * RAID-1 management functions. |
| 9 | * |
| 10 | * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 |
| 11 | * |
Jan Engelhardt | 96de0e2 | 2007-10-19 23:21:04 +0200 | [diff] [blame] | 12 | * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> |
| 14 | * |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 15 | * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support |
| 16 | * bitmapped intelligence in resync: |
| 17 | * |
| 18 | * - bitmap marked during normal i/o |
| 19 | * - bitmap used to skip nondirty blocks during sync |
| 20 | * |
| 21 | * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: |
| 22 | * - persistent bitmap code |
| 23 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | * This program is free software; you can redistribute it and/or modify |
| 25 | * it under the terms of the GNU General Public License as published by |
| 26 | * the Free Software Foundation; either version 2, or (at your option) |
| 27 | * any later version. |
| 28 | * |
| 29 | * You should have received a copy of the GNU General Public License |
| 30 | * (for example /usr/src/linux/COPYING); if not, write to the Free |
| 31 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
| 32 | */ |
| 33 | |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 34 | #include <linux/slab.h> |
Stephen Rothwell | 2557072 | 2008-10-15 09:09:21 +1100 | [diff] [blame] | 35 | #include <linux/delay.h> |
NeilBrown | bff6197 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 36 | #include <linux/blkdev.h> |
Paul Gortmaker | 056075c | 2011-07-03 13:58:33 -0400 | [diff] [blame] | 37 | #include <linux/module.h> |
NeilBrown | bff6197 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 38 | #include <linux/seq_file.h> |
Christian Dietrich | 8bda470 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 39 | #include <linux/ratelimit.h> |
NeilBrown | 43b2e5d | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 40 | #include "md.h" |
Christoph Hellwig | ef740c3 | 2009-03-31 14:27:03 +1100 | [diff] [blame] | 41 | #include "raid1.h" |
| 42 | #include "bitmap.h" |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 43 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | /* |
| 45 | * Number of guaranteed r1bios in case of extreme VM load: |
| 46 | */ |
| 47 | #define NR_RAID1_BIOS 256 |
| 48 | |
Jonathan Brassow | 473e87c | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 49 | /* when we get a read error on a read-only array, we redirect to another |
| 50 | * device without failing the first device, or trying to over-write to |
| 51 | * correct the read error. To keep track of bad blocks on a per-bio |
| 52 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer |
| 53 | */ |
| 54 | #define IO_BLOCKED ((struct bio *)1) |
| 55 | /* When we successfully write to a known bad-block, we need to remove the |
| 56 | * bad-block marking which must be done from process context. So we record |
| 57 | * the success by setting devs[n].bio to IO_MADE_GOOD |
| 58 | */ |
| 59 | #define IO_MADE_GOOD ((struct bio *)2) |
| 60 | |
| 61 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) |
| 62 | |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 63 | /* When there are this many requests queue to be written by |
| 64 | * the raid1 thread, we become 'congested' to provide back-pressure |
| 65 | * for writeback. |
| 66 | */ |
| 67 | static int max_queued_requests = 1024; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 69 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
| 70 | sector_t bi_sector); |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 71 | static void lower_barrier(struct r1conf *conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 73 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | { |
| 75 | struct pool_info *pi = data; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 76 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | |
| 78 | /* allocate a r1bio with room for raid_disks entries in the bios array */ |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 79 | return kzalloc(size, gfp_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | } |
| 81 | |
| 82 | static void r1bio_pool_free(void *r1_bio, void *data) |
| 83 | { |
| 84 | kfree(r1_bio); |
| 85 | } |
| 86 | |
| 87 | #define RESYNC_BLOCK_SIZE (64*1024) |
majianpeng | 8e005f7 | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 88 | #define RESYNC_DEPTH 32 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
| 90 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
majianpeng | 8e005f7 | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 91 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
| 92 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) |
Goldwyn Rodrigues | c40f341 | 2015-08-19 08:14:42 +1000 | [diff] [blame] | 93 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
| 94 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) |
majianpeng | 8e005f7 | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 95 | #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 97 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | { |
| 99 | struct pool_info *pi = data; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 100 | struct r1bio *r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | struct bio *bio; |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 102 | int need_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | int i, j; |
| 104 | |
| 105 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 106 | if (!r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | |
| 109 | /* |
| 110 | * Allocate bios : 1 for reading, n-1 for writing |
| 111 | */ |
| 112 | for (j = pi->raid_disks ; j-- ; ) { |
NeilBrown | 6746557 | 2010-10-26 17:33:54 +1100 | [diff] [blame] | 113 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | if (!bio) |
| 115 | goto out_free_bio; |
| 116 | r1_bio->bios[j] = bio; |
| 117 | } |
| 118 | /* |
| 119 | * Allocate RESYNC_PAGES data pages and attach them to |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 120 | * the first bio. |
| 121 | * If this is a user-requested check/repair, allocate |
| 122 | * RESYNC_PAGES for each bio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | */ |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 124 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 125 | need_pages = pi->raid_disks; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 126 | else |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 127 | need_pages = 1; |
| 128 | for (j = 0; j < need_pages; j++) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 129 | bio = r1_bio->bios[j]; |
Kent Overstreet | a078760 | 2012-09-10 14:03:28 -0700 | [diff] [blame] | 130 | bio->bi_vcnt = RESYNC_PAGES; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 131 | |
Kent Overstreet | a078760 | 2012-09-10 14:03:28 -0700 | [diff] [blame] | 132 | if (bio_alloc_pages(bio, gfp_flags)) |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 133 | goto out_free_pages; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 134 | } |
| 135 | /* If not user-requests, copy the page pointers to all bios */ |
| 136 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { |
| 137 | for (i=0; i<RESYNC_PAGES ; i++) |
| 138 | for (j=1; j<pi->raid_disks; j++) |
| 139 | r1_bio->bios[j]->bi_io_vec[i].bv_page = |
| 140 | r1_bio->bios[0]->bi_io_vec[i].bv_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | } |
| 142 | |
| 143 | r1_bio->master_bio = NULL; |
| 144 | |
| 145 | return r1_bio; |
| 146 | |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 147 | out_free_pages: |
| 148 | while (--j >= 0) { |
| 149 | struct bio_vec *bv; |
| 150 | |
| 151 | bio_for_each_segment_all(bv, r1_bio->bios[j], i) |
| 152 | __free_page(bv->bv_page); |
| 153 | } |
| 154 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | out_free_bio: |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 156 | while (++j < pi->raid_disks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | bio_put(r1_bio->bios[j]); |
| 158 | r1bio_pool_free(r1_bio, data); |
| 159 | return NULL; |
| 160 | } |
| 161 | |
| 162 | static void r1buf_pool_free(void *__r1_bio, void *data) |
| 163 | { |
| 164 | struct pool_info *pi = data; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 165 | int i,j; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 166 | struct r1bio *r1bio = __r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 168 | for (i = 0; i < RESYNC_PAGES; i++) |
| 169 | for (j = pi->raid_disks; j-- ;) { |
| 170 | if (j == 0 || |
| 171 | r1bio->bios[j]->bi_io_vec[i].bv_page != |
| 172 | r1bio->bios[0]->bi_io_vec[i].bv_page) |
NeilBrown | 1345b1d | 2006-01-06 00:20:40 -0800 | [diff] [blame] | 173 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 174 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | for (i=0 ; i < pi->raid_disks; i++) |
| 176 | bio_put(r1bio->bios[i]); |
| 177 | |
| 178 | r1bio_pool_free(r1bio, data); |
| 179 | } |
| 180 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 181 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | { |
| 183 | int i; |
| 184 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 185 | for (i = 0; i < conf->raid_disks * 2; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | struct bio **bio = r1_bio->bios + i; |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 187 | if (!BIO_SPECIAL(*bio)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | bio_put(*bio); |
| 189 | *bio = NULL; |
| 190 | } |
| 191 | } |
| 192 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 193 | static void free_r1bio(struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 195 | struct r1conf *conf = r1_bio->mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | put_all_bios(conf, r1_bio); |
| 198 | mempool_free(r1_bio, conf->r1bio_pool); |
| 199 | } |
| 200 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 201 | static void put_buf(struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 203 | struct r1conf *conf = r1_bio->mddev->private; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 204 | int i; |
| 205 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 206 | for (i = 0; i < conf->raid_disks * 2; i++) { |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 207 | struct bio *bio = r1_bio->bios[i]; |
| 208 | if (bio->bi_end_io) |
| 209 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); |
| 210 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | |
| 212 | mempool_free(r1_bio, conf->r1buf_pool); |
| 213 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 214 | lower_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | } |
| 216 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 217 | static void reschedule_retry(struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 218 | { |
| 219 | unsigned long flags; |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 220 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 221 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 222 | |
| 223 | spin_lock_irqsave(&conf->device_lock, flags); |
| 224 | list_add(&r1_bio->retry_list, &conf->retry_list); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 225 | conf->nr_queued ++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 227 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 228 | wake_up(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 229 | md_wakeup_thread(mddev->thread); |
| 230 | } |
| 231 | |
| 232 | /* |
| 233 | * raid_end_bio_io() is called when we have finished servicing a mirrored |
| 234 | * operation and are ready to return a success/failure code to the buffer |
| 235 | * cache layer. |
| 236 | */ |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 237 | static void call_bio_endio(struct r1bio *r1_bio) |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 238 | { |
| 239 | struct bio *bio = r1_bio->master_bio; |
| 240 | int done; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 241 | struct r1conf *conf = r1_bio->mddev->private; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 242 | sector_t start_next_window = r1_bio->start_next_window; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 243 | sector_t bi_sector = bio->bi_iter.bi_sector; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 244 | |
| 245 | if (bio->bi_phys_segments) { |
| 246 | unsigned long flags; |
| 247 | spin_lock_irqsave(&conf->device_lock, flags); |
| 248 | bio->bi_phys_segments--; |
| 249 | done = (bio->bi_phys_segments == 0); |
| 250 | spin_unlock_irqrestore(&conf->device_lock, flags); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 251 | /* |
| 252 | * make_request() might be waiting for |
| 253 | * bi_phys_segments to decrease |
| 254 | */ |
| 255 | wake_up(&conf->wait_barrier); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 256 | } else |
| 257 | done = 1; |
| 258 | |
| 259 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 260 | bio->bi_error = -EIO; |
| 261 | |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 262 | if (done) { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 263 | bio_endio(bio); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 264 | /* |
| 265 | * Wake up any possible resync thread that waits for the device |
| 266 | * to go idle. |
| 267 | */ |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 268 | allow_barrier(conf, start_next_window, bi_sector); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 269 | } |
| 270 | } |
| 271 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 272 | static void raid_end_bio_io(struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | { |
| 274 | struct bio *bio = r1_bio->master_bio; |
| 275 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 276 | /* if nobody has done the final endio yet, do it now */ |
| 277 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { |
NeilBrown | 36a4e1f | 2011-10-07 14:23:17 +1100 | [diff] [blame] | 278 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
| 279 | (bio_data_dir(bio) == WRITE) ? "write" : "read", |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 280 | (unsigned long long) bio->bi_iter.bi_sector, |
| 281 | (unsigned long long) bio_end_sector(bio) - 1); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 282 | |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 283 | call_bio_endio(r1_bio); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 284 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | free_r1bio(r1_bio); |
| 286 | } |
| 287 | |
| 288 | /* |
| 289 | * Update disk head position estimator based on IRQ completion info. |
| 290 | */ |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 291 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 293 | struct r1conf *conf = r1_bio->mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | |
| 295 | conf->mirrors[disk].head_position = |
| 296 | r1_bio->sector + (r1_bio->sectors); |
| 297 | } |
| 298 | |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 299 | /* |
| 300 | * Find the disk number which triggered given bio |
| 301 | */ |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 302 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 303 | { |
| 304 | int mirror; |
NeilBrown | 3019463 | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 305 | struct r1conf *conf = r1_bio->mddev->private; |
| 306 | int raid_disks = conf->raid_disks; |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 307 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 308 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 309 | if (r1_bio->bios[mirror] == bio) |
| 310 | break; |
| 311 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 312 | BUG_ON(mirror == raid_disks * 2); |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 313 | update_head_pos(mirror, r1_bio); |
| 314 | |
| 315 | return mirror; |
| 316 | } |
| 317 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 318 | static void raid1_end_read_request(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 320 | int uptodate = !bio->bi_error; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 321 | struct r1bio *r1_bio = bio->bi_private; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 322 | struct r1conf *conf = r1_bio->mddev->private; |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 323 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 324 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 325 | /* |
| 326 | * this branch is our 'one mirror IO has finished' event handler: |
| 327 | */ |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 328 | update_head_pos(r1_bio->read_disk, r1_bio); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 329 | |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 330 | if (uptodate) |
| 331 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
| 332 | else { |
| 333 | /* If all other devices have failed, we want to return |
| 334 | * the error upwards rather than fail the last device. |
| 335 | * Here we redefine "uptodate" to mean "Don't want to retry" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | */ |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 337 | unsigned long flags; |
| 338 | spin_lock_irqsave(&conf->device_lock, flags); |
| 339 | if (r1_bio->mddev->degraded == conf->raid_disks || |
| 340 | (r1_bio->mddev->degraded == conf->raid_disks-1 && |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 341 | test_bit(In_sync, &rdev->flags))) |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 342 | uptodate = 1; |
| 343 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 344 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | |
NeilBrown | 7ad4d4a | 2012-10-11 13:44:30 +1100 | [diff] [blame] | 346 | if (uptodate) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 347 | raid_end_bio_io(r1_bio); |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 348 | rdev_dec_pending(rdev, conf->mddev); |
NeilBrown | 7ad4d4a | 2012-10-11 13:44:30 +1100 | [diff] [blame] | 349 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 350 | /* |
| 351 | * oops, read error: |
| 352 | */ |
| 353 | char b[BDEVNAME_SIZE]; |
Christian Dietrich | 8bda470 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 354 | printk_ratelimited( |
| 355 | KERN_ERR "md/raid1:%s: %s: " |
| 356 | "rescheduling sector %llu\n", |
| 357 | mdname(conf->mddev), |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 358 | bdevname(rdev->bdev, |
Christian Dietrich | 8bda470 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 359 | b), |
| 360 | (unsigned long long)r1_bio->sector); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 361 | set_bit(R1BIO_ReadError, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | reschedule_retry(r1_bio); |
NeilBrown | 7ad4d4a | 2012-10-11 13:44:30 +1100 | [diff] [blame] | 363 | /* don't drop the reference on read_disk yet */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | } |
| 366 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 367 | static void close_write(struct r1bio *r1_bio) |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 368 | { |
| 369 | /* it really is the end of this request */ |
| 370 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
| 371 | /* free extra copy of the data pages */ |
| 372 | int i = r1_bio->behind_page_count; |
| 373 | while (i--) |
| 374 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); |
| 375 | kfree(r1_bio->behind_bvecs); |
| 376 | r1_bio->behind_bvecs = NULL; |
| 377 | } |
| 378 | /* clear the bitmap if all writes complete successfully */ |
| 379 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
| 380 | r1_bio->sectors, |
| 381 | !test_bit(R1BIO_Degraded, &r1_bio->state), |
| 382 | test_bit(R1BIO_BehindIO, &r1_bio->state)); |
| 383 | md_write_end(r1_bio->mddev); |
| 384 | } |
| 385 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 386 | static void r1_bio_write_done(struct r1bio *r1_bio) |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 387 | { |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 388 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
| 389 | return; |
| 390 | |
| 391 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 392 | reschedule_retry(r1_bio); |
| 393 | else { |
| 394 | close_write(r1_bio); |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 395 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
| 396 | reschedule_retry(r1_bio); |
| 397 | else |
| 398 | raid_end_bio_io(r1_bio); |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 399 | } |
| 400 | } |
| 401 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 402 | static void raid1_end_write_request(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 403 | { |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 404 | struct r1bio *r1_bio = bio->bi_private; |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 405 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 406 | struct r1conf *conf = r1_bio->mddev->private; |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 407 | struct bio *to_put = NULL; |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 408 | int mirror = find_bio_disk(r1_bio, bio); |
| 409 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 410 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 411 | /* |
| 412 | * 'one mirror IO has finished' event handler: |
| 413 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 414 | if (bio->bi_error) { |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 415 | set_bit(WriteErrorSeen, &rdev->flags); |
| 416 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
NeilBrown | 19d6716 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 417 | set_bit(MD_RECOVERY_NEEDED, & |
| 418 | conf->mddev->recovery); |
| 419 | |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 420 | set_bit(R1BIO_WriteError, &r1_bio->state); |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 421 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | /* |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 423 | * Set R1BIO_Uptodate in our master bio, so that we |
| 424 | * will return a good error code for to the higher |
| 425 | * levels even if IO on some other mirrored buffer |
| 426 | * fails. |
| 427 | * |
| 428 | * The 'master' represents the composite IO operation |
| 429 | * to user-side. So if something waits for IO, then it |
| 430 | * will wait for the 'master' bio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 431 | */ |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 432 | sector_t first_bad; |
| 433 | int bad_sectors; |
| 434 | |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 435 | r1_bio->bios[mirror] = NULL; |
| 436 | to_put = bio; |
Alex Lyakas | 3056e3a | 2013-06-04 20:42:21 +0300 | [diff] [blame] | 437 | /* |
| 438 | * Do not set R1BIO_Uptodate if the current device is |
| 439 | * rebuilding or Faulty. This is because we cannot use |
| 440 | * such device for properly reading the data back (we could |
| 441 | * potentially use it, if the current write would have felt |
| 442 | * before rdev->recovery_offset, but for simplicity we don't |
| 443 | * check this here. |
| 444 | */ |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 445 | if (test_bit(In_sync, &rdev->flags) && |
| 446 | !test_bit(Faulty, &rdev->flags)) |
Alex Lyakas | 3056e3a | 2013-06-04 20:42:21 +0300 | [diff] [blame] | 447 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 448 | |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 449 | /* Maybe we can clear some bad blocks. */ |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 450 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 451 | &first_bad, &bad_sectors)) { |
| 452 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
| 453 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
| 454 | } |
| 455 | } |
| 456 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 457 | if (behind) { |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 458 | if (test_bit(WriteMostly, &rdev->flags)) |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 459 | atomic_dec(&r1_bio->behind_remaining); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 460 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 461 | /* |
| 462 | * In behind mode, we ACK the master bio once the I/O |
| 463 | * has safely reached all non-writemostly |
| 464 | * disks. Setting the Returned bit ensures that this |
| 465 | * gets done only once -- we don't ever want to return |
| 466 | * -EIO here, instead we'll wait |
| 467 | */ |
| 468 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && |
| 469 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
| 470 | /* Maybe we can return now */ |
| 471 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { |
| 472 | struct bio *mbio = r1_bio->master_bio; |
NeilBrown | 36a4e1f | 2011-10-07 14:23:17 +1100 | [diff] [blame] | 473 | pr_debug("raid1: behind end write sectors" |
| 474 | " %llu-%llu\n", |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 475 | (unsigned long long) mbio->bi_iter.bi_sector, |
| 476 | (unsigned long long) bio_end_sector(mbio) - 1); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 477 | call_bio_endio(r1_bio); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 478 | } |
| 479 | } |
| 480 | } |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 481 | if (r1_bio->bios[mirror] == NULL) |
NeilBrown | e5872d5 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 482 | rdev_dec_pending(rdev, conf->mddev); |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 483 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 485 | * Let's see if all mirrored write operations have finished |
| 486 | * already. |
| 487 | */ |
NeilBrown | af6d7b7 | 2011-05-11 14:51:19 +1000 | [diff] [blame] | 488 | r1_bio_write_done(r1_bio); |
NeilBrown | c70810b | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 489 | |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 490 | if (to_put) |
| 491 | bio_put(to_put); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | } |
| 493 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | /* |
| 495 | * This routine returns the disk from which the requested read should |
| 496 | * be done. There is a per-array 'next expected sequential IO' sector |
| 497 | * number - if this matches on the next IO then we use the last disk. |
| 498 | * There is also a per-disk 'last know head position' sector that is |
| 499 | * maintained from IRQ contexts, both the normal and the resync IO |
| 500 | * completion handlers update this position correctly. If there is no |
| 501 | * perfect sequential match then we pick the disk whose head is closest. |
| 502 | * |
| 503 | * If there are 2 mirrors in the same 2 devices, performance degrades |
| 504 | * because position is mirror, not device based. |
| 505 | * |
| 506 | * The rdev for the device selected will have nr_pending incremented. |
| 507 | */ |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 508 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | { |
NeilBrown | af3a2cd | 2010-05-08 08:20:17 +1000 | [diff] [blame] | 510 | const sector_t this_sector = r1_bio->sector; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 511 | int sectors; |
| 512 | int best_good_sectors; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 513 | int best_disk, best_dist_disk, best_pending_disk; |
| 514 | int has_nonrot_disk; |
Shaohua Li | be4d328 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 515 | int disk; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 516 | sector_t best_dist; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 517 | unsigned int min_pending; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 518 | struct md_rdev *rdev; |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 519 | int choose_first; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 520 | int choose_next_idle; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | |
| 522 | rcu_read_lock(); |
| 523 | /* |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 524 | * Check if we can balance. We can balance on the whole |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 525 | * device if no resync is going on, or below the resync window. |
| 526 | * We take the first readable disk when above the resync window. |
| 527 | */ |
| 528 | retry: |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 529 | sectors = r1_bio->sectors; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 530 | best_disk = -1; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 531 | best_dist_disk = -1; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 532 | best_dist = MaxSector; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 533 | best_pending_disk = -1; |
| 534 | min_pending = UINT_MAX; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 535 | best_good_sectors = 0; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 536 | has_nonrot_disk = 0; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 537 | choose_next_idle = 0; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 538 | |
Goldwyn Rodrigues | 7d49ffc | 2014-08-12 10:13:19 -0500 | [diff] [blame] | 539 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
| 540 | (mddev_is_clustered(conf->mddev) && |
Goldwyn Rodrigues | 90382ed | 2015-06-24 09:30:32 -0500 | [diff] [blame] | 541 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
Goldwyn Rodrigues | 7d49ffc | 2014-08-12 10:13:19 -0500 | [diff] [blame] | 542 | this_sector + sectors))) |
| 543 | choose_first = 1; |
| 544 | else |
| 545 | choose_first = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | |
Shaohua Li | be4d328 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 547 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 548 | sector_t dist; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 549 | sector_t first_bad; |
| 550 | int bad_sectors; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 551 | unsigned int pending; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 552 | bool nonrot; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 553 | |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 554 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
| 555 | if (r1_bio->bios[disk] == IO_BLOCKED |
| 556 | || rdev == NULL |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 557 | || test_bit(Faulty, &rdev->flags)) |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 558 | continue; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 559 | if (!test_bit(In_sync, &rdev->flags) && |
| 560 | rdev->recovery_offset < this_sector + sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | continue; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 562 | if (test_bit(WriteMostly, &rdev->flags)) { |
| 563 | /* Don't balance among write-mostly, just |
| 564 | * use the first as a last resort */ |
Tomáš Hodek | d1901ef | 2015-02-23 11:00:38 +1100 | [diff] [blame] | 565 | if (best_dist_disk < 0) { |
NeilBrown | 307729c | 2012-01-09 01:41:51 +1100 | [diff] [blame] | 566 | if (is_badblock(rdev, this_sector, sectors, |
| 567 | &first_bad, &bad_sectors)) { |
Wei Fang | 816b0ac | 2016-03-21 19:18:32 +0800 | [diff] [blame] | 568 | if (first_bad <= this_sector) |
NeilBrown | 307729c | 2012-01-09 01:41:51 +1100 | [diff] [blame] | 569 | /* Cannot use this */ |
| 570 | continue; |
| 571 | best_good_sectors = first_bad - this_sector; |
| 572 | } else |
| 573 | best_good_sectors = sectors; |
Tomáš Hodek | d1901ef | 2015-02-23 11:00:38 +1100 | [diff] [blame] | 574 | best_dist_disk = disk; |
| 575 | best_pending_disk = disk; |
NeilBrown | 307729c | 2012-01-09 01:41:51 +1100 | [diff] [blame] | 576 | } |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 577 | continue; |
| 578 | } |
| 579 | /* This is a reasonable device to use. It might |
| 580 | * even be best. |
| 581 | */ |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 582 | if (is_badblock(rdev, this_sector, sectors, |
| 583 | &first_bad, &bad_sectors)) { |
| 584 | if (best_dist < MaxSector) |
| 585 | /* already have a better device */ |
| 586 | continue; |
| 587 | if (first_bad <= this_sector) { |
| 588 | /* cannot read here. If this is the 'primary' |
| 589 | * device, then we must not read beyond |
| 590 | * bad_sectors from another device.. |
| 591 | */ |
| 592 | bad_sectors -= (this_sector - first_bad); |
| 593 | if (choose_first && sectors > bad_sectors) |
| 594 | sectors = bad_sectors; |
| 595 | if (best_good_sectors > sectors) |
| 596 | best_good_sectors = sectors; |
| 597 | |
| 598 | } else { |
| 599 | sector_t good_sectors = first_bad - this_sector; |
| 600 | if (good_sectors > best_good_sectors) { |
| 601 | best_good_sectors = good_sectors; |
| 602 | best_disk = disk; |
| 603 | } |
| 604 | if (choose_first) |
| 605 | break; |
| 606 | } |
| 607 | continue; |
| 608 | } else |
| 609 | best_good_sectors = sectors; |
| 610 | |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 611 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
| 612 | has_nonrot_disk |= nonrot; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 613 | pending = atomic_read(&rdev->nr_pending); |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 614 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 615 | if (choose_first) { |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 616 | best_disk = disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | break; |
| 618 | } |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 619 | /* Don't change to another disk for sequential reads */ |
| 620 | if (conf->mirrors[disk].next_seq_sect == this_sector |
| 621 | || dist == 0) { |
| 622 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; |
| 623 | struct raid1_info *mirror = &conf->mirrors[disk]; |
| 624 | |
| 625 | best_disk = disk; |
| 626 | /* |
| 627 | * If buffered sequential IO size exceeds optimal |
| 628 | * iosize, check if there is idle disk. If yes, choose |
| 629 | * the idle disk. read_balance could already choose an |
| 630 | * idle disk before noticing it's a sequential IO in |
| 631 | * this disk. This doesn't matter because this disk |
| 632 | * will idle, next time it will be utilized after the |
| 633 | * first disk has IO size exceeds optimal iosize. In |
| 634 | * this way, iosize of the first disk will be optimal |
| 635 | * iosize at least. iosize of the second disk might be |
| 636 | * small, but not a big deal since when the second disk |
| 637 | * starts IO, the first disk is likely still busy. |
| 638 | */ |
| 639 | if (nonrot && opt_iosize > 0 && |
| 640 | mirror->seq_start != MaxSector && |
| 641 | mirror->next_seq_sect > opt_iosize && |
| 642 | mirror->next_seq_sect - opt_iosize >= |
| 643 | mirror->seq_start) { |
| 644 | choose_next_idle = 1; |
| 645 | continue; |
| 646 | } |
| 647 | break; |
| 648 | } |
| 649 | /* If device is idle, use it */ |
| 650 | if (pending == 0) { |
| 651 | best_disk = disk; |
| 652 | break; |
| 653 | } |
| 654 | |
| 655 | if (choose_next_idle) |
| 656 | continue; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 657 | |
| 658 | if (min_pending > pending) { |
| 659 | min_pending = pending; |
| 660 | best_pending_disk = disk; |
| 661 | } |
| 662 | |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 663 | if (dist < best_dist) { |
| 664 | best_dist = dist; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 665 | best_dist_disk = disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | } |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 667 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 669 | /* |
| 670 | * If all disks are rotational, choose the closest disk. If any disk is |
| 671 | * non-rotational, choose the disk with less pending request even the |
| 672 | * disk is rotational, which might/might not be optimal for raids with |
| 673 | * mixed ratation/non-rotational disks depending on workload. |
| 674 | */ |
| 675 | if (best_disk == -1) { |
| 676 | if (has_nonrot_disk) |
| 677 | best_disk = best_pending_disk; |
| 678 | else |
| 679 | best_disk = best_dist_disk; |
| 680 | } |
| 681 | |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 682 | if (best_disk >= 0) { |
| 683 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 684 | if (!rdev) |
| 685 | goto retry; |
| 686 | atomic_inc(&rdev->nr_pending); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 687 | sectors = best_good_sectors; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 688 | |
| 689 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) |
| 690 | conf->mirrors[best_disk].seq_start = this_sector; |
| 691 | |
Shaohua Li | be4d328 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 692 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | } |
| 694 | rcu_read_unlock(); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 695 | *max_sectors = sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 697 | return best_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 698 | } |
| 699 | |
NeilBrown | 5c675f8 | 2014-12-15 12:56:56 +1100 | [diff] [blame] | 700 | static int raid1_congested(struct mddev *mddev, int bits) |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 701 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 702 | struct r1conf *conf = mddev->private; |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 703 | int i, ret = 0; |
| 704 | |
Tejun Heo | 4452226 | 2015-05-22 17:13:26 -0400 | [diff] [blame] | 705 | if ((bits & (1 << WB_async_congested)) && |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 706 | conf->pending_count >= max_queued_requests) |
| 707 | return 1; |
| 708 | |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 709 | rcu_read_lock(); |
NeilBrown | f53e29f | 2012-02-13 14:24:05 +1100 | [diff] [blame] | 710 | for (i = 0; i < conf->raid_disks * 2; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 711 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 712 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 713 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 714 | |
Jonathan Brassow | 1ed7242 | 2011-06-07 17:50:35 -0500 | [diff] [blame] | 715 | BUG_ON(!q); |
| 716 | |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 717 | /* Note the '|| 1' - when read_balance prefers |
| 718 | * non-congested targets, it can be removed |
| 719 | */ |
Tejun Heo | 4452226 | 2015-05-22 17:13:26 -0400 | [diff] [blame] | 720 | if ((bits & (1 << WB_async_congested)) || 1) |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 721 | ret |= bdi_congested(&q->backing_dev_info, bits); |
| 722 | else |
| 723 | ret &= bdi_congested(&q->backing_dev_info, bits); |
| 724 | } |
| 725 | } |
| 726 | rcu_read_unlock(); |
| 727 | return ret; |
| 728 | } |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 729 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 730 | static void flush_pending_writes(struct r1conf *conf) |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 731 | { |
| 732 | /* Any writes that have been queued but are awaiting |
| 733 | * bitmap updates get flushed here. |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 734 | */ |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 735 | spin_lock_irq(&conf->device_lock); |
| 736 | |
| 737 | if (conf->pending_bio_list.head) { |
| 738 | struct bio *bio; |
| 739 | bio = bio_list_get(&conf->pending_bio_list); |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 740 | conf->pending_count = 0; |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 741 | spin_unlock_irq(&conf->device_lock); |
| 742 | /* flush any pending bitmap writes to |
| 743 | * disk before proceeding w/ I/O */ |
| 744 | bitmap_unplug(conf->mddev->bitmap); |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 745 | wake_up(&conf->wait_barrier); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 746 | |
| 747 | while (bio) { /* submit pending writes */ |
| 748 | struct bio *next = bio->bi_next; |
| 749 | bio->bi_next = NULL; |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 750 | if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 751 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) |
| 752 | /* Just ignore it */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 753 | bio_endio(bio); |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 754 | else |
| 755 | generic_make_request(bio); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 756 | bio = next; |
| 757 | } |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 758 | } else |
| 759 | spin_unlock_irq(&conf->device_lock); |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 760 | } |
| 761 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 762 | /* Barriers.... |
| 763 | * Sometimes we need to suspend IO while we do something else, |
| 764 | * either some resync/recovery, or reconfigure the array. |
| 765 | * To do this we raise a 'barrier'. |
| 766 | * The 'barrier' is a counter that can be raised multiple times |
| 767 | * to count how many activities are happening which preclude |
| 768 | * normal IO. |
| 769 | * We can only raise the barrier if there is no pending IO. |
| 770 | * i.e. if nr_pending == 0. |
| 771 | * We choose only to raise the barrier if no-one is waiting for the |
| 772 | * barrier to go down. This means that as soon as an IO request |
| 773 | * is ready, no other operations which require a barrier will start |
| 774 | * until the IO request has had a chance. |
| 775 | * |
| 776 | * So: regular IO calls 'wait_barrier'. When that returns there |
| 777 | * is no backgroup IO happening, It must arrange to call |
| 778 | * allow_barrier when it has finished its IO. |
| 779 | * backgroup IO calls must call raise_barrier. Once that returns |
| 780 | * there is no normal IO happeing. It must arrange to call |
| 781 | * lower_barrier when the particular background IO completes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | */ |
NeilBrown | c2fd4c9 | 2014-09-10 16:01:24 +1000 | [diff] [blame] | 783 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 784 | { |
| 785 | spin_lock_irq(&conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 786 | |
| 787 | /* Wait until no block IO is waiting */ |
| 788 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 789 | conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 790 | |
| 791 | /* block any new IO from starting */ |
| 792 | conf->barrier++; |
NeilBrown | c2fd4c9 | 2014-09-10 16:01:24 +1000 | [diff] [blame] | 793 | conf->next_resync = sector_nr; |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 794 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 795 | /* For these conditions we must wait: |
| 796 | * A: while the array is in frozen state |
| 797 | * B: while barrier >= RESYNC_DEPTH, meaning resync reach |
| 798 | * the max count which allowed. |
| 799 | * C: next_resync + RESYNC_SECTORS > start_next_window, meaning |
| 800 | * next resync will reach to the window which normal bios are |
| 801 | * handling. |
NeilBrown | 2f73d3c | 2014-09-10 15:01:49 +1000 | [diff] [blame] | 802 | * D: while there are any active requests in the current window. |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 803 | */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 804 | wait_event_lock_irq(conf->wait_barrier, |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 805 | !conf->array_frozen && |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 806 | conf->barrier < RESYNC_DEPTH && |
NeilBrown | 2f73d3c | 2014-09-10 15:01:49 +1000 | [diff] [blame] | 807 | conf->current_window_requests == 0 && |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 808 | (conf->start_next_window >= |
| 809 | conf->next_resync + RESYNC_SECTORS), |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 810 | conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 811 | |
NeilBrown | 34e97f1 | 2014-09-16 12:14:14 +1000 | [diff] [blame] | 812 | conf->nr_pending++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | spin_unlock_irq(&conf->resync_lock); |
| 814 | } |
| 815 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 816 | static void lower_barrier(struct r1conf *conf) |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 817 | { |
| 818 | unsigned long flags; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 819 | BUG_ON(conf->barrier <= 0); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 820 | spin_lock_irqsave(&conf->resync_lock, flags); |
| 821 | conf->barrier--; |
NeilBrown | 34e97f1 | 2014-09-16 12:14:14 +1000 | [diff] [blame] | 822 | conf->nr_pending--; |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 823 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
| 824 | wake_up(&conf->wait_barrier); |
| 825 | } |
| 826 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 827 | static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 828 | { |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 829 | bool wait = false; |
| 830 | |
| 831 | if (conf->array_frozen || !bio) |
| 832 | wait = true; |
| 833 | else if (conf->barrier && bio_data_dir(bio) == WRITE) { |
NeilBrown | 2355496 | 2014-09-10 15:56:57 +1000 | [diff] [blame] | 834 | if ((conf->mddev->curr_resync_completed |
| 835 | >= bio_end_sector(bio)) || |
| 836 | (conf->next_resync + NEXT_NORMALIO_DISTANCE |
| 837 | <= bio->bi_iter.bi_sector)) |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 838 | wait = false; |
| 839 | else |
| 840 | wait = true; |
| 841 | } |
| 842 | |
| 843 | return wait; |
| 844 | } |
| 845 | |
| 846 | static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) |
| 847 | { |
| 848 | sector_t sector = 0; |
| 849 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 850 | spin_lock_irq(&conf->resync_lock); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 851 | if (need_to_wait_for_sync(conf, bio)) { |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 852 | conf->nr_waiting++; |
NeilBrown | d6b42dc | 2012-03-19 12:46:38 +1100 | [diff] [blame] | 853 | /* Wait for the barrier to drop. |
| 854 | * However if there are already pending |
| 855 | * requests (preventing the barrier from |
| 856 | * rising completely), and the |
NeilBrown | 5965b64 | 2014-09-04 15:51:44 +1000 | [diff] [blame] | 857 | * per-process bio queue isn't empty, |
NeilBrown | d6b42dc | 2012-03-19 12:46:38 +1100 | [diff] [blame] | 858 | * then don't wait, as we need to empty |
NeilBrown | 5965b64 | 2014-09-04 15:51:44 +1000 | [diff] [blame] | 859 | * that queue to allow conf->start_next_window |
| 860 | * to increase. |
NeilBrown | d6b42dc | 2012-03-19 12:46:38 +1100 | [diff] [blame] | 861 | */ |
| 862 | wait_event_lock_irq(conf->wait_barrier, |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 863 | !conf->array_frozen && |
| 864 | (!conf->barrier || |
NeilBrown | 5965b64 | 2014-09-04 15:51:44 +1000 | [diff] [blame] | 865 | ((conf->start_next_window < |
| 866 | conf->next_resync + RESYNC_SECTORS) && |
| 867 | current->bio_list && |
| 868 | !bio_list_empty(current->bio_list))), |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 869 | conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 870 | conf->nr_waiting--; |
| 871 | } |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 872 | |
| 873 | if (bio && bio_data_dir(bio) == WRITE) { |
Jes Sorensen | e8ff8bf | 2015-09-16 10:20:05 -0400 | [diff] [blame] | 874 | if (bio->bi_iter.bi_sector >= conf->next_resync) { |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 875 | if (conf->start_next_window == MaxSector) |
| 876 | conf->start_next_window = |
| 877 | conf->next_resync + |
| 878 | NEXT_NORMALIO_DISTANCE; |
| 879 | |
| 880 | if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 881 | <= bio->bi_iter.bi_sector) |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 882 | conf->next_window_requests++; |
| 883 | else |
| 884 | conf->current_window_requests++; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 885 | sector = conf->start_next_window; |
NeilBrown | 41a336e | 2014-01-14 11:56:14 +1100 | [diff] [blame] | 886 | } |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 887 | } |
| 888 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 889 | conf->nr_pending++; |
| 890 | spin_unlock_irq(&conf->resync_lock); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 891 | return sector; |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 892 | } |
| 893 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 894 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
| 895 | sector_t bi_sector) |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 896 | { |
| 897 | unsigned long flags; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 898 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 899 | spin_lock_irqsave(&conf->resync_lock, flags); |
| 900 | conf->nr_pending--; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 901 | if (start_next_window) { |
| 902 | if (start_next_window == conf->start_next_window) { |
| 903 | if (conf->start_next_window + NEXT_NORMALIO_DISTANCE |
| 904 | <= bi_sector) |
| 905 | conf->next_window_requests--; |
| 906 | else |
| 907 | conf->current_window_requests--; |
| 908 | } else |
| 909 | conf->current_window_requests--; |
| 910 | |
| 911 | if (!conf->current_window_requests) { |
| 912 | if (conf->next_window_requests) { |
| 913 | conf->current_window_requests = |
| 914 | conf->next_window_requests; |
| 915 | conf->next_window_requests = 0; |
| 916 | conf->start_next_window += |
| 917 | NEXT_NORMALIO_DISTANCE; |
| 918 | } else |
| 919 | conf->start_next_window = MaxSector; |
| 920 | } |
| 921 | } |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 922 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
| 923 | wake_up(&conf->wait_barrier); |
| 924 | } |
| 925 | |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 926 | static void freeze_array(struct r1conf *conf, int extra) |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 927 | { |
| 928 | /* stop syncio and normal IO and wait for everything to |
| 929 | * go quite. |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 930 | * We wait until nr_pending match nr_queued+extra |
NeilBrown | 1c83053 | 2008-03-04 14:29:35 -0800 | [diff] [blame] | 931 | * This is called in the context of one normal IO request |
| 932 | * that has failed. Thus any sync request that might be pending |
| 933 | * will be blocked by nr_pending, and we need to wait for |
| 934 | * pending IO requests to complete or be queued for re-try. |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 935 | * Thus the number queued (nr_queued) plus this request (extra) |
NeilBrown | 1c83053 | 2008-03-04 14:29:35 -0800 | [diff] [blame] | 936 | * must match the number of pending IOs (nr_pending) before |
| 937 | * we continue. |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 938 | */ |
| 939 | spin_lock_irq(&conf->resync_lock); |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 940 | conf->array_frozen = 1; |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 941 | wait_event_lock_irq_cmd(conf->wait_barrier, |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 942 | conf->nr_pending == conf->nr_queued+extra, |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 943 | conf->resync_lock, |
| 944 | flush_pending_writes(conf)); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 945 | spin_unlock_irq(&conf->resync_lock); |
| 946 | } |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 947 | static void unfreeze_array(struct r1conf *conf) |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 948 | { |
| 949 | /* reverse the effect of the freeze */ |
| 950 | spin_lock_irq(&conf->resync_lock); |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 951 | conf->array_frozen = 0; |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 952 | wake_up(&conf->wait_barrier); |
| 953 | spin_unlock_irq(&conf->resync_lock); |
| 954 | } |
| 955 | |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 956 | /* duplicate the data pages for behind I/O |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 957 | */ |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 958 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 959 | { |
| 960 | int i; |
| 961 | struct bio_vec *bvec; |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 962 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 963 | GFP_NOIO); |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 964 | if (unlikely(!bvecs)) |
NeilBrown | af6d7b7 | 2011-05-11 14:51:19 +1000 | [diff] [blame] | 965 | return; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 966 | |
Kent Overstreet | cb34e05 | 2012-09-05 15:22:02 -0700 | [diff] [blame] | 967 | bio_for_each_segment_all(bvec, bio, i) { |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 968 | bvecs[i] = *bvec; |
| 969 | bvecs[i].bv_page = alloc_page(GFP_NOIO); |
| 970 | if (unlikely(!bvecs[i].bv_page)) |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 971 | goto do_sync_io; |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 972 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
| 973 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); |
| 974 | kunmap(bvecs[i].bv_page); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 975 | kunmap(bvec->bv_page); |
| 976 | } |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 977 | r1_bio->behind_bvecs = bvecs; |
NeilBrown | af6d7b7 | 2011-05-11 14:51:19 +1000 | [diff] [blame] | 978 | r1_bio->behind_page_count = bio->bi_vcnt; |
| 979 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
| 980 | return; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 981 | |
| 982 | do_sync_io: |
NeilBrown | af6d7b7 | 2011-05-11 14:51:19 +1000 | [diff] [blame] | 983 | for (i = 0; i < bio->bi_vcnt; i++) |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 984 | if (bvecs[i].bv_page) |
| 985 | put_page(bvecs[i].bv_page); |
| 986 | kfree(bvecs); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 987 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
| 988 | bio->bi_iter.bi_size); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 989 | } |
| 990 | |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 991 | struct raid1_plug_cb { |
| 992 | struct blk_plug_cb cb; |
| 993 | struct bio_list pending; |
| 994 | int pending_cnt; |
| 995 | }; |
| 996 | |
| 997 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) |
| 998 | { |
| 999 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, |
| 1000 | cb); |
| 1001 | struct mddev *mddev = plug->cb.data; |
| 1002 | struct r1conf *conf = mddev->private; |
| 1003 | struct bio *bio; |
| 1004 | |
NeilBrown | 874807a | 2012-11-27 12:14:40 +1100 | [diff] [blame] | 1005 | if (from_schedule || current->bio_list) { |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1006 | spin_lock_irq(&conf->device_lock); |
| 1007 | bio_list_merge(&conf->pending_bio_list, &plug->pending); |
| 1008 | conf->pending_count += plug->pending_cnt; |
| 1009 | spin_unlock_irq(&conf->device_lock); |
NeilBrown | ee0b024 | 2013-02-25 12:38:29 +1100 | [diff] [blame] | 1010 | wake_up(&conf->wait_barrier); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1011 | md_wakeup_thread(mddev->thread); |
| 1012 | kfree(plug); |
| 1013 | return; |
| 1014 | } |
| 1015 | |
| 1016 | /* we aren't scheduling, so we can do the write-out directly. */ |
| 1017 | bio = bio_list_get(&plug->pending); |
| 1018 | bitmap_unplug(mddev->bitmap); |
| 1019 | wake_up(&conf->wait_barrier); |
| 1020 | |
| 1021 | while (bio) { /* submit pending writes */ |
| 1022 | struct bio *next = bio->bi_next; |
| 1023 | bio->bi_next = NULL; |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 1024 | if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
Shaohua Li | 32f9f57 | 2013-04-28 18:26:38 +0800 | [diff] [blame] | 1025 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) |
| 1026 | /* Just ignore it */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1027 | bio_endio(bio); |
Shaohua Li | 32f9f57 | 2013-04-28 18:26:38 +0800 | [diff] [blame] | 1028 | else |
| 1029 | generic_make_request(bio); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1030 | bio = next; |
| 1031 | } |
| 1032 | kfree(plug); |
| 1033 | } |
| 1034 | |
Shaohua Li | 849674e | 2016-01-20 13:52:20 -0800 | [diff] [blame] | 1035 | static void raid1_make_request(struct mddev *mddev, struct bio * bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1037 | struct r1conf *conf = mddev->private; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 1038 | struct raid1_info *mirror; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1039 | struct r1bio *r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1040 | struct bio *read_bio; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1041 | int i, disks; |
NeilBrown | 84255d1 | 2008-05-23 13:04:32 -0700 | [diff] [blame] | 1042 | struct bitmap *bitmap; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1043 | unsigned long flags; |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 1044 | const int op = bio_op(bio); |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1045 | const int rw = bio_data_dir(bio); |
Jens Axboe | 1eff9d3 | 2016-08-05 15:35:16 -0600 | [diff] [blame] | 1046 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
| 1047 | const unsigned long do_flush_fua = (bio->bi_opf & |
Mike Christie | 28a8f0d | 2016-06-05 14:32:25 -0500 | [diff] [blame] | 1048 | (REQ_PREFLUSH | REQ_FUA)); |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1049 | struct md_rdev *blocked_rdev; |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1050 | struct blk_plug_cb *cb; |
| 1051 | struct raid1_plug_cb *plug = NULL; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1052 | int first_clone; |
| 1053 | int sectors_handled; |
| 1054 | int max_sectors; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1055 | sector_t start_next_window; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1056 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | /* |
| 1058 | * Register the new request and wait if the reconstruction |
| 1059 | * thread has put up a bar for new requests. |
| 1060 | * Continue immediately if no resync is active currently. |
| 1061 | */ |
NeilBrown | 62de608 | 2006-05-01 12:15:47 -0700 | [diff] [blame] | 1062 | |
NeilBrown | 3d310eb | 2005-06-21 17:17:26 -0700 | [diff] [blame] | 1063 | md_write_start(mddev, bio); /* wait on superblock update early */ |
| 1064 | |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1065 | if (bio_data_dir(bio) == WRITE && |
Goldwyn Rodrigues | 589a1c4 | 2014-06-07 02:39:37 -0500 | [diff] [blame] | 1066 | ((bio_end_sector(bio) > mddev->suspend_lo && |
| 1067 | bio->bi_iter.bi_sector < mddev->suspend_hi) || |
| 1068 | (mddev_is_clustered(mddev) && |
Goldwyn Rodrigues | 90382ed | 2015-06-24 09:30:32 -0500 | [diff] [blame] | 1069 | md_cluster_ops->area_resyncing(mddev, WRITE, |
| 1070 | bio->bi_iter.bi_sector, bio_end_sector(bio))))) { |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1071 | /* As the suspend_* range is controlled by |
| 1072 | * userspace, we want an interruptible |
| 1073 | * wait. |
| 1074 | */ |
| 1075 | DEFINE_WAIT(w); |
| 1076 | for (;;) { |
| 1077 | flush_signals(current); |
| 1078 | prepare_to_wait(&conf->wait_barrier, |
| 1079 | &w, TASK_INTERRUPTIBLE); |
Kent Overstreet | f73a1c7 | 2012-09-25 15:05:12 -0700 | [diff] [blame] | 1080 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
Goldwyn Rodrigues | 589a1c4 | 2014-06-07 02:39:37 -0500 | [diff] [blame] | 1081 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
| 1082 | (mddev_is_clustered(mddev) && |
Goldwyn Rodrigues | 90382ed | 2015-06-24 09:30:32 -0500 | [diff] [blame] | 1083 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
Goldwyn Rodrigues | 589a1c4 | 2014-06-07 02:39:37 -0500 | [diff] [blame] | 1084 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1085 | break; |
| 1086 | schedule(); |
| 1087 | } |
| 1088 | finish_wait(&conf->wait_barrier, &w); |
| 1089 | } |
NeilBrown | 62de608 | 2006-05-01 12:15:47 -0700 | [diff] [blame] | 1090 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1091 | start_next_window = wait_barrier(conf, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1092 | |
NeilBrown | 84255d1 | 2008-05-23 13:04:32 -0700 | [diff] [blame] | 1093 | bitmap = mddev->bitmap; |
| 1094 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1095 | /* |
Christoph Hellwig | 7024628 | 2016-07-19 11:28:41 +0200 | [diff] [blame] | 1096 | * make_request() can abort the operation when read-ahead is being |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | * used and no empty request is available. |
| 1098 | * |
| 1099 | */ |
| 1100 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 1101 | |
| 1102 | r1_bio->master_bio = bio; |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 1103 | r1_bio->sectors = bio_sectors(bio); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1104 | r1_bio->state = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | r1_bio->mddev = mddev; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1106 | r1_bio->sector = bio->bi_iter.bi_sector; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1107 | |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1108 | /* We might need to issue multiple reads to different |
| 1109 | * devices if there are bad blocks around, so we keep |
| 1110 | * track of the number of reads in bio->bi_phys_segments. |
| 1111 | * If this is 0, there is only one r1_bio and no locking |
| 1112 | * will be needed when requests complete. If it is |
| 1113 | * non-zero, then it is the number of not-completed requests. |
| 1114 | */ |
| 1115 | bio->bi_phys_segments = 0; |
Jens Axboe | b7c44ed | 2015-07-24 12:37:59 -0600 | [diff] [blame] | 1116 | bio_clear_flag(bio, BIO_SEG_VALID); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1117 | |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1118 | if (rw == READ) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | /* |
| 1120 | * read balancing logic: |
| 1121 | */ |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1122 | int rdisk; |
| 1123 | |
| 1124 | read_again: |
| 1125 | rdisk = read_balance(conf, r1_bio, &max_sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1126 | |
| 1127 | if (rdisk < 0) { |
| 1128 | /* couldn't find anywhere to read from */ |
| 1129 | raid_end_bio_io(r1_bio); |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1130 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1131 | } |
| 1132 | mirror = conf->mirrors + rdisk; |
| 1133 | |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 1134 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
| 1135 | bitmap) { |
| 1136 | /* Reading from a write-mostly device must |
| 1137 | * take care not to over-take any writes |
| 1138 | * that are 'behind' |
| 1139 | */ |
| 1140 | wait_event(bitmap->behind_wait, |
| 1141 | atomic_read(&bitmap->behind_writes) == 0); |
| 1142 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | r1_bio->read_disk = rdisk; |
NeilBrown | f0cc9a0 | 2014-09-22 10:06:23 +1000 | [diff] [blame] | 1144 | r1_bio->start_next_window = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1145 | |
NeilBrown | a167f66 | 2010-10-26 18:31:13 +1100 | [diff] [blame] | 1146 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1147 | bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, |
Kent Overstreet | 6678d83 | 2013-08-07 11:14:32 -0700 | [diff] [blame] | 1148 | max_sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | |
| 1150 | r1_bio->bios[rdisk] = read_bio; |
| 1151 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1152 | read_bio->bi_iter.bi_sector = r1_bio->sector + |
| 1153 | mirror->rdev->data_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1154 | read_bio->bi_bdev = mirror->rdev->bdev; |
| 1155 | read_bio->bi_end_io = raid1_end_read_request; |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 1156 | bio_set_op_attrs(read_bio, op, do_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1157 | read_bio->bi_private = r1_bio; |
| 1158 | |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1159 | if (max_sectors < r1_bio->sectors) { |
| 1160 | /* could not read all from this device, so we will |
| 1161 | * need another r1_bio. |
| 1162 | */ |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1163 | |
| 1164 | sectors_handled = (r1_bio->sector + max_sectors |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1165 | - bio->bi_iter.bi_sector); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1166 | r1_bio->sectors = max_sectors; |
| 1167 | spin_lock_irq(&conf->device_lock); |
| 1168 | if (bio->bi_phys_segments == 0) |
| 1169 | bio->bi_phys_segments = 2; |
| 1170 | else |
| 1171 | bio->bi_phys_segments++; |
| 1172 | spin_unlock_irq(&conf->device_lock); |
| 1173 | /* Cannot call generic_make_request directly |
| 1174 | * as that will be queued in __make_request |
| 1175 | * and subsequent mempool_alloc might block waiting |
| 1176 | * for it. So hand bio over to raid1d. |
| 1177 | */ |
| 1178 | reschedule_retry(r1_bio); |
| 1179 | |
| 1180 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 1181 | |
| 1182 | r1_bio->master_bio = bio; |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 1183 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1184 | r1_bio->state = 0; |
| 1185 | r1_bio->mddev = mddev; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1186 | r1_bio->sector = bio->bi_iter.bi_sector + |
| 1187 | sectors_handled; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1188 | goto read_again; |
| 1189 | } else |
| 1190 | generic_make_request(read_bio); |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1191 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1192 | } |
| 1193 | |
| 1194 | /* |
| 1195 | * WRITE: |
| 1196 | */ |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 1197 | if (conf->pending_count >= max_queued_requests) { |
| 1198 | md_wakeup_thread(mddev->thread); |
| 1199 | wait_event(conf->wait_barrier, |
| 1200 | conf->pending_count < max_queued_requests); |
| 1201 | } |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1202 | /* first select target devices under rcu_lock and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1203 | * inc refcount on their rdev. Record them by setting |
| 1204 | * bios[x] to bio |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1205 | * If there are known/acknowledged bad blocks on any device on |
| 1206 | * which we have seen a write error, we want to avoid writing those |
| 1207 | * blocks. |
| 1208 | * This potentially requires several writes to write around |
| 1209 | * the bad blocks. Each set of writes gets it's own r1bio |
| 1210 | * with a set of bios attached. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1211 | */ |
NeilBrown | c3b328a | 2011-04-18 18:25:43 +1000 | [diff] [blame] | 1212 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1213 | disks = conf->raid_disks * 2; |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1214 | retry_write: |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1215 | r1_bio->start_next_window = start_next_window; |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1216 | blocked_rdev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1217 | rcu_read_lock(); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1218 | max_sectors = r1_bio->sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1219 | for (i = 0; i < disks; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1220 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1221 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
| 1222 | atomic_inc(&rdev->nr_pending); |
| 1223 | blocked_rdev = rdev; |
| 1224 | break; |
| 1225 | } |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1226 | r1_bio->bios[i] = NULL; |
Kent Overstreet | 8ae1266 | 2015-04-27 23:48:34 -0700 | [diff] [blame] | 1227 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1228 | if (i < conf->raid_disks) |
| 1229 | set_bit(R1BIO_Degraded, &r1_bio->state); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1230 | continue; |
| 1231 | } |
| 1232 | |
| 1233 | atomic_inc(&rdev->nr_pending); |
| 1234 | if (test_bit(WriteErrorSeen, &rdev->flags)) { |
| 1235 | sector_t first_bad; |
| 1236 | int bad_sectors; |
| 1237 | int is_bad; |
| 1238 | |
| 1239 | is_bad = is_badblock(rdev, r1_bio->sector, |
| 1240 | max_sectors, |
| 1241 | &first_bad, &bad_sectors); |
| 1242 | if (is_bad < 0) { |
| 1243 | /* mustn't write here until the bad block is |
| 1244 | * acknowledged*/ |
| 1245 | set_bit(BlockedBadBlocks, &rdev->flags); |
| 1246 | blocked_rdev = rdev; |
| 1247 | break; |
NeilBrown | 964147d | 2010-05-18 15:27:13 +1000 | [diff] [blame] | 1248 | } |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1249 | if (is_bad && first_bad <= r1_bio->sector) { |
| 1250 | /* Cannot write here at all */ |
| 1251 | bad_sectors -= (r1_bio->sector - first_bad); |
| 1252 | if (bad_sectors < max_sectors) |
| 1253 | /* mustn't write more than bad_sectors |
| 1254 | * to other devices yet |
| 1255 | */ |
| 1256 | max_sectors = bad_sectors; |
| 1257 | rdev_dec_pending(rdev, mddev); |
| 1258 | /* We don't set R1BIO_Degraded as that |
| 1259 | * only applies if the disk is |
| 1260 | * missing, so it might be re-added, |
| 1261 | * and we want to know to recover this |
| 1262 | * chunk. |
| 1263 | * In this case the device is here, |
| 1264 | * and the fact that this chunk is not |
| 1265 | * in-sync is recorded in the bad |
| 1266 | * block log |
| 1267 | */ |
| 1268 | continue; |
| 1269 | } |
| 1270 | if (is_bad) { |
| 1271 | int good_sectors = first_bad - r1_bio->sector; |
| 1272 | if (good_sectors < max_sectors) |
| 1273 | max_sectors = good_sectors; |
| 1274 | } |
| 1275 | } |
| 1276 | r1_bio->bios[i] = bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | } |
| 1278 | rcu_read_unlock(); |
| 1279 | |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1280 | if (unlikely(blocked_rdev)) { |
| 1281 | /* Wait for this device to become unblocked */ |
| 1282 | int j; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1283 | sector_t old = start_next_window; |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1284 | |
| 1285 | for (j = 0; j < i; j++) |
| 1286 | if (r1_bio->bios[j]) |
| 1287 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1288 | r1_bio->state = 0; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1289 | allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector); |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1290 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1291 | start_next_window = wait_barrier(conf, bio); |
| 1292 | /* |
| 1293 | * We must make sure the multi r1bios of bio have |
| 1294 | * the same value of bi_phys_segments |
| 1295 | */ |
| 1296 | if (bio->bi_phys_segments && old && |
| 1297 | old != start_next_window) |
| 1298 | /* Wait for the former r1bio(s) to complete */ |
| 1299 | wait_event(conf->wait_barrier, |
| 1300 | bio->bi_phys_segments == 1); |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1301 | goto retry_write; |
| 1302 | } |
| 1303 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1304 | if (max_sectors < r1_bio->sectors) { |
| 1305 | /* We are splitting this write into multiple parts, so |
| 1306 | * we need to prepare for allocating another r1_bio. |
| 1307 | */ |
| 1308 | r1_bio->sectors = max_sectors; |
| 1309 | spin_lock_irq(&conf->device_lock); |
| 1310 | if (bio->bi_phys_segments == 0) |
| 1311 | bio->bi_phys_segments = 2; |
| 1312 | else |
| 1313 | bio->bi_phys_segments++; |
| 1314 | spin_unlock_irq(&conf->device_lock); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1315 | } |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1316 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1317 | |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 1318 | atomic_set(&r1_bio->remaining, 1); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1319 | atomic_set(&r1_bio->behind_remaining, 0); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1320 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1321 | first_clone = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1322 | for (i = 0; i < disks; i++) { |
| 1323 | struct bio *mbio; |
| 1324 | if (!r1_bio->bios[i]) |
| 1325 | continue; |
| 1326 | |
NeilBrown | a167f66 | 2010-10-26 18:31:13 +1100 | [diff] [blame] | 1327 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1328 | bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1330 | if (first_clone) { |
| 1331 | /* do behind I/O ? |
| 1332 | * Not if there are too many, or cannot |
| 1333 | * allocate memory, or a reader on WriteMostly |
| 1334 | * is waiting for behind writes to flush */ |
| 1335 | if (bitmap && |
| 1336 | (atomic_read(&bitmap->behind_writes) |
| 1337 | < mddev->bitmap_info.max_write_behind) && |
| 1338 | !waitqueue_active(&bitmap->behind_wait)) |
| 1339 | alloc_behind_pages(mbio, r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1340 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1341 | bitmap_startwrite(bitmap, r1_bio->sector, |
| 1342 | r1_bio->sectors, |
| 1343 | test_bit(R1BIO_BehindIO, |
| 1344 | &r1_bio->state)); |
| 1345 | first_clone = 0; |
| 1346 | } |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 1347 | if (r1_bio->behind_bvecs) { |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1348 | struct bio_vec *bvec; |
| 1349 | int j; |
| 1350 | |
Kent Overstreet | cb34e05 | 2012-09-05 15:22:02 -0700 | [diff] [blame] | 1351 | /* |
| 1352 | * We trimmed the bio, so _all is legit |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1353 | */ |
Kent Overstreet | d74c6d5 | 2013-02-06 12:23:11 -0800 | [diff] [blame] | 1354 | bio_for_each_segment_all(bvec, mbio, j) |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 1355 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1356 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
| 1357 | atomic_inc(&r1_bio->behind_remaining); |
| 1358 | } |
| 1359 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1360 | r1_bio->bios[i] = mbio; |
| 1361 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1362 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1363 | conf->mirrors[i].rdev->data_offset); |
| 1364 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
| 1365 | mbio->bi_end_io = raid1_end_write_request; |
Christoph Hellwig | 288dab8 | 2016-06-09 16:00:36 +0200 | [diff] [blame] | 1366 | bio_set_op_attrs(mbio, op, do_flush_fua | do_sync); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1367 | mbio->bi_private = r1_bio; |
| 1368 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1369 | atomic_inc(&r1_bio->remaining); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1370 | |
| 1371 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
| 1372 | if (cb) |
| 1373 | plug = container_of(cb, struct raid1_plug_cb, cb); |
| 1374 | else |
| 1375 | plug = NULL; |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 1376 | spin_lock_irqsave(&conf->device_lock, flags); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1377 | if (plug) { |
| 1378 | bio_list_add(&plug->pending, mbio); |
| 1379 | plug->pending_cnt++; |
| 1380 | } else { |
| 1381 | bio_list_add(&conf->pending_bio_list, mbio); |
| 1382 | conf->pending_count++; |
| 1383 | } |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 1384 | spin_unlock_irqrestore(&conf->device_lock, flags); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1385 | if (!plug) |
NeilBrown | b357f04 | 2012-07-03 17:45:31 +1000 | [diff] [blame] | 1386 | md_wakeup_thread(mddev->thread); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1387 | } |
NeilBrown | 079fa16 | 2011-09-10 17:21:23 +1000 | [diff] [blame] | 1388 | /* Mustn't call r1_bio_write_done before this next test, |
| 1389 | * as it could result in the bio being freed. |
| 1390 | */ |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 1391 | if (sectors_handled < bio_sectors(bio)) { |
NeilBrown | 079fa16 | 2011-09-10 17:21:23 +1000 | [diff] [blame] | 1392 | r1_bio_write_done(r1_bio); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1393 | /* We need another r1_bio. It has already been counted |
| 1394 | * in bio->bi_phys_segments |
| 1395 | */ |
| 1396 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 1397 | r1_bio->master_bio = bio; |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 1398 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1399 | r1_bio->state = 0; |
| 1400 | r1_bio->mddev = mddev; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1401 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1402 | goto retry_write; |
| 1403 | } |
| 1404 | |
NeilBrown | 079fa16 | 2011-09-10 17:21:23 +1000 | [diff] [blame] | 1405 | r1_bio_write_done(r1_bio); |
| 1406 | |
| 1407 | /* In case raid1d snuck in to freeze_array */ |
| 1408 | wake_up(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1409 | } |
| 1410 | |
Shaohua Li | 849674e | 2016-01-20 13:52:20 -0800 | [diff] [blame] | 1411 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1412 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1413 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1414 | int i; |
| 1415 | |
| 1416 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
NeilBrown | 11ce99e | 2006-10-03 01:15:52 -0700 | [diff] [blame] | 1417 | conf->raid_disks - mddev->degraded); |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1418 | rcu_read_lock(); |
| 1419 | for (i = 0; i < conf->raid_disks; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1420 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1421 | seq_printf(seq, "%s", |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1422 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
| 1423 | } |
| 1424 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1425 | seq_printf(seq, "]"); |
| 1426 | } |
| 1427 | |
Shaohua Li | 849674e | 2016-01-20 13:52:20 -0800 | [diff] [blame] | 1428 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1429 | { |
| 1430 | char b[BDEVNAME_SIZE]; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1431 | struct r1conf *conf = mddev->private; |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1432 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1433 | |
| 1434 | /* |
| 1435 | * If it is not operational, then we have already marked it as dead |
| 1436 | * else if it is the last working disks, ignore the error, let the |
| 1437 | * next level up know. |
| 1438 | * else mark the drive as failed |
| 1439 | */ |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1440 | if (test_bit(In_sync, &rdev->flags) |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1441 | && (conf->raid_disks - mddev->degraded) == 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1442 | /* |
| 1443 | * Don't fail the drive, act as though we were just a |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1444 | * normal single drive. |
| 1445 | * However don't try a recovery from this drive as |
| 1446 | * it is very likely to fail. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1447 | */ |
NeilBrown | 5389042 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1448 | conf->recovery_disabled = mddev->recovery_disabled; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1449 | return; |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1450 | } |
NeilBrown | de393cd | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1451 | set_bit(Blocked, &rdev->flags); |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1452 | spin_lock_irqsave(&conf->device_lock, flags); |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 1453 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | mddev->degraded++; |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 1455 | set_bit(Faulty, &rdev->flags); |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 1456 | } else |
| 1457 | set_bit(Faulty, &rdev->flags); |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1458 | spin_unlock_irqrestore(&conf->device_lock, flags); |
NeilBrown | 2446dba | 2014-07-31 10:16:29 +1000 | [diff] [blame] | 1459 | /* |
| 1460 | * if recovery is running, make sure it aborts. |
| 1461 | */ |
| 1462 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
Guoqing Jiang | 85ad1d1 | 2016-05-03 22:22:13 -0400 | [diff] [blame] | 1463 | set_mask_bits(&mddev->flags, 0, |
| 1464 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); |
Joe Perches | 067032b | 2011-01-14 09:14:33 +1100 | [diff] [blame] | 1465 | printk(KERN_ALERT |
| 1466 | "md/raid1:%s: Disk failure on %s, disabling device.\n" |
| 1467 | "md/raid1:%s: Operation continuing on %d devices.\n", |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1468 | mdname(mddev), bdevname(rdev->bdev, b), |
| 1469 | mdname(mddev), conf->raid_disks - mddev->degraded); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1470 | } |
| 1471 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1472 | static void print_conf(struct r1conf *conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | { |
| 1474 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1475 | |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1476 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | if (!conf) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1478 | printk(KERN_DEBUG "(!conf)\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1479 | return; |
| 1480 | } |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1481 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1482 | conf->raid_disks); |
| 1483 | |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1484 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1485 | for (i = 0; i < conf->raid_disks; i++) { |
| 1486 | char b[BDEVNAME_SIZE]; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1487 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1488 | if (rdev) |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1489 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1490 | i, !test_bit(In_sync, &rdev->flags), |
| 1491 | !test_bit(Faulty, &rdev->flags), |
| 1492 | bdevname(rdev->bdev,b)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1493 | } |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1494 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | } |
| 1496 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1497 | static void close_sync(struct r1conf *conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | { |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1499 | wait_barrier(conf, NULL); |
| 1500 | allow_barrier(conf, 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1501 | |
| 1502 | mempool_destroy(conf->r1buf_pool); |
| 1503 | conf->r1buf_pool = NULL; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1504 | |
NeilBrown | 669cc7b | 2014-09-04 16:30:38 +1000 | [diff] [blame] | 1505 | spin_lock_irq(&conf->resync_lock); |
Jes Sorensen | e8ff8bf | 2015-09-16 10:20:05 -0400 | [diff] [blame] | 1506 | conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1507 | conf->start_next_window = MaxSector; |
NeilBrown | 669cc7b | 2014-09-04 16:30:38 +1000 | [diff] [blame] | 1508 | conf->current_window_requests += |
| 1509 | conf->next_window_requests; |
| 1510 | conf->next_window_requests = 0; |
| 1511 | spin_unlock_irq(&conf->resync_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1512 | } |
| 1513 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1514 | static int raid1_spare_active(struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1515 | { |
| 1516 | int i; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1517 | struct r1conf *conf = mddev->private; |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1518 | int count = 0; |
| 1519 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1520 | |
| 1521 | /* |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 1522 | * Find all failed disks within the RAID1 configuration |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1523 | * and mark them readable. |
| 1524 | * Called under mddev lock, so rcu protection not needed. |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1525 | * device_lock used to avoid races with raid1_end_read_request |
| 1526 | * which expects 'In_sync' flags and ->degraded to be consistent. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | */ |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1528 | spin_lock_irqsave(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1529 | for (i = 0; i < conf->raid_disks; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1530 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1531 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
| 1532 | if (repl |
Goldwyn Rodrigues | 1aee41f | 2014-10-29 18:51:31 -0500 | [diff] [blame] | 1533 | && !test_bit(Candidate, &repl->flags) |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1534 | && repl->recovery_offset == MaxSector |
| 1535 | && !test_bit(Faulty, &repl->flags) |
| 1536 | && !test_and_set_bit(In_sync, &repl->flags)) { |
| 1537 | /* replacement has just become active */ |
| 1538 | if (!rdev || |
| 1539 | !test_and_clear_bit(In_sync, &rdev->flags)) |
| 1540 | count++; |
| 1541 | if (rdev) { |
| 1542 | /* Replaced device not technically |
| 1543 | * faulty, but we need to be sure |
| 1544 | * it gets removed and never re-added |
| 1545 | */ |
| 1546 | set_bit(Faulty, &rdev->flags); |
| 1547 | sysfs_notify_dirent_safe( |
| 1548 | rdev->sysfs_state); |
| 1549 | } |
| 1550 | } |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1551 | if (rdev |
Lukasz Dorau | 61e4947 | 2013-10-24 12:55:17 +1100 | [diff] [blame] | 1552 | && rdev->recovery_offset == MaxSector |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1553 | && !test_bit(Faulty, &rdev->flags) |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 1554 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1555 | count++; |
Jonathan Brassow | 654e8b5 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1556 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | } |
| 1558 | } |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1559 | mddev->degraded -= count; |
| 1560 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1561 | |
| 1562 | print_conf(conf); |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1563 | return count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1564 | } |
| 1565 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1566 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1567 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1568 | struct r1conf *conf = mddev->private; |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1569 | int err = -EEXIST; |
NeilBrown | 41158c7 | 2005-06-21 17:17:25 -0700 | [diff] [blame] | 1570 | int mirror = 0; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 1571 | struct raid1_info *p; |
Neil Brown | 6c2fce2 | 2008-06-28 08:31:31 +1000 | [diff] [blame] | 1572 | int first = 0; |
NeilBrown | 3019463 | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1573 | int last = conf->raid_disks - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1574 | |
NeilBrown | 5389042 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1575 | if (mddev->recovery_disabled == conf->recovery_disabled) |
| 1576 | return -EBUSY; |
| 1577 | |
Dan Williams | 1501efa | 2016-01-13 16:00:07 -0800 | [diff] [blame] | 1578 | if (md_integrity_add_rdev(rdev, mddev)) |
| 1579 | return -ENXIO; |
| 1580 | |
Neil Brown | 6c2fce2 | 2008-06-28 08:31:31 +1000 | [diff] [blame] | 1581 | if (rdev->raid_disk >= 0) |
| 1582 | first = last = rdev->raid_disk; |
| 1583 | |
Goldwyn Rodrigues | 70bcecd | 2015-08-21 10:33:39 -0500 | [diff] [blame] | 1584 | /* |
| 1585 | * find the disk ... but prefer rdev->saved_raid_disk |
| 1586 | * if possible. |
| 1587 | */ |
| 1588 | if (rdev->saved_raid_disk >= 0 && |
| 1589 | rdev->saved_raid_disk >= first && |
| 1590 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
| 1591 | first = last = rdev->saved_raid_disk; |
| 1592 | |
NeilBrown | 7ef449d | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1593 | for (mirror = first; mirror <= last; mirror++) { |
| 1594 | p = conf->mirrors+mirror; |
| 1595 | if (!p->rdev) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1596 | |
Jonathan Brassow | 9092c02 | 2013-05-02 14:19:24 -0500 | [diff] [blame] | 1597 | if (mddev->gendisk) |
| 1598 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 1599 | rdev->data_offset << 9); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1600 | |
| 1601 | p->head_position = 0; |
| 1602 | rdev->raid_disk = mirror; |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1603 | err = 0; |
NeilBrown | 6aea114a | 2005-11-28 13:44:13 -0800 | [diff] [blame] | 1604 | /* As all devices are equivalent, we don't need a full recovery |
| 1605 | * if this was recently any drive of the array |
| 1606 | */ |
| 1607 | if (rdev->saved_raid_disk < 0) |
NeilBrown | 41158c7 | 2005-06-21 17:17:25 -0700 | [diff] [blame] | 1608 | conf->fullsync = 1; |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 1609 | rcu_assign_pointer(p->rdev, rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1610 | break; |
| 1611 | } |
NeilBrown | 7ef449d | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1612 | if (test_bit(WantReplacement, &p->rdev->flags) && |
| 1613 | p[conf->raid_disks].rdev == NULL) { |
| 1614 | /* Add this device as a replacement */ |
| 1615 | clear_bit(In_sync, &rdev->flags); |
| 1616 | set_bit(Replacement, &rdev->flags); |
| 1617 | rdev->raid_disk = mirror; |
| 1618 | err = 0; |
| 1619 | conf->fullsync = 1; |
| 1620 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); |
| 1621 | break; |
| 1622 | } |
| 1623 | } |
Jonathan Brassow | 9092c02 | 2013-05-02 14:19:24 -0500 | [diff] [blame] | 1624 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 1625 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | print_conf(conf); |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1627 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | } |
| 1629 | |
NeilBrown | b8321b6 | 2011-12-23 10:17:51 +1100 | [diff] [blame] | 1630 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1631 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1632 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1633 | int err = 0; |
NeilBrown | b8321b6 | 2011-12-23 10:17:51 +1100 | [diff] [blame] | 1634 | int number = rdev->raid_disk; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 1635 | struct raid1_info *p = conf->mirrors + number; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1636 | |
NeilBrown | b014f14 | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1637 | if (rdev != p->rdev) |
| 1638 | p = conf->mirrors + conf->raid_disks + number; |
| 1639 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1640 | print_conf(conf); |
NeilBrown | b8321b6 | 2011-12-23 10:17:51 +1100 | [diff] [blame] | 1641 | if (rdev == p->rdev) { |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1642 | if (test_bit(In_sync, &rdev->flags) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1643 | atomic_read(&rdev->nr_pending)) { |
| 1644 | err = -EBUSY; |
| 1645 | goto abort; |
| 1646 | } |
NeilBrown | 046abee | 2010-10-26 15:46:20 +1100 | [diff] [blame] | 1647 | /* Only remove non-faulty devices if recovery |
NeilBrown | dfc7064 | 2008-05-23 13:04:39 -0700 | [diff] [blame] | 1648 | * is not possible. |
| 1649 | */ |
| 1650 | if (!test_bit(Faulty, &rdev->flags) && |
NeilBrown | 5389042 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1651 | mddev->recovery_disabled != conf->recovery_disabled && |
NeilBrown | dfc7064 | 2008-05-23 13:04:39 -0700 | [diff] [blame] | 1652 | mddev->degraded < conf->raid_disks) { |
| 1653 | err = -EBUSY; |
| 1654 | goto abort; |
| 1655 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1656 | p->rdev = NULL; |
NeilBrown | d787be4 | 2016-06-02 16:19:53 +1000 | [diff] [blame] | 1657 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
| 1658 | synchronize_rcu(); |
| 1659 | if (atomic_read(&rdev->nr_pending)) { |
| 1660 | /* lost the race, try later */ |
| 1661 | err = -EBUSY; |
| 1662 | p->rdev = rdev; |
| 1663 | goto abort; |
| 1664 | } |
| 1665 | } |
| 1666 | if (conf->mirrors[conf->raid_disks + number].rdev) { |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1667 | /* We just removed a device that is being replaced. |
| 1668 | * Move down the replacement. We drain all IO before |
| 1669 | * doing this to avoid confusion. |
| 1670 | */ |
| 1671 | struct md_rdev *repl = |
| 1672 | conf->mirrors[conf->raid_disks + number].rdev; |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 1673 | freeze_array(conf, 0); |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1674 | clear_bit(Replacement, &repl->flags); |
| 1675 | p->rdev = repl; |
| 1676 | conf->mirrors[conf->raid_disks + number].rdev = NULL; |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 1677 | unfreeze_array(conf); |
NeilBrown | b014f14 | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1678 | clear_bit(WantReplacement, &rdev->flags); |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1679 | } else |
| 1680 | clear_bit(WantReplacement, &rdev->flags); |
Martin K. Petersen | a91a278 | 2011-03-17 11:11:05 +0100 | [diff] [blame] | 1681 | err = md_integrity_register(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1682 | } |
| 1683 | abort: |
| 1684 | |
| 1685 | print_conf(conf); |
| 1686 | return err; |
| 1687 | } |
| 1688 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1689 | static void end_sync_read(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1690 | { |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1691 | struct r1bio *r1_bio = bio->bi_private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | |
NeilBrown | 0fc280f | 2011-10-07 14:22:55 +1100 | [diff] [blame] | 1693 | update_head_pos(r1_bio->read_disk, r1_bio); |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 1694 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1695 | /* |
| 1696 | * we have read a block, now it needs to be re-written, |
| 1697 | * or re-read if the read failed. |
| 1698 | * We don't do much here, just schedule handling by raid1d |
| 1699 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1700 | if (!bio->bi_error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1701 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1702 | |
| 1703 | if (atomic_dec_and_test(&r1_bio->remaining)) |
| 1704 | reschedule_retry(r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1705 | } |
| 1706 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1707 | static void end_sync_write(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1708 | { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1709 | int uptodate = !bio->bi_error; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1710 | struct r1bio *r1_bio = bio->bi_private; |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1711 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1712 | struct r1conf *conf = mddev->private; |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1713 | sector_t first_bad; |
| 1714 | int bad_sectors; |
NeilBrown | 854abd7 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 1715 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 1716 | |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1717 | if (!uptodate) { |
NeilBrown | 57dab0b | 2010-10-19 10:03:39 +1100 | [diff] [blame] | 1718 | sector_t sync_blocks = 0; |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1719 | sector_t s = r1_bio->sector; |
| 1720 | long sectors_to_go = r1_bio->sectors; |
| 1721 | /* make sure these bits doesn't get cleared. */ |
| 1722 | do { |
NeilBrown | 5e3db64 | 2006-07-10 04:44:18 -0700 | [diff] [blame] | 1723 | bitmap_end_sync(mddev->bitmap, s, |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1724 | &sync_blocks, 1); |
| 1725 | s += sync_blocks; |
| 1726 | sectors_to_go -= sync_blocks; |
| 1727 | } while (sectors_to_go > 0); |
NeilBrown | 854abd7 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 1728 | set_bit(WriteErrorSeen, &rdev->flags); |
| 1729 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
NeilBrown | 19d6716 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1730 | set_bit(MD_RECOVERY_NEEDED, & |
| 1731 | mddev->recovery); |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1732 | set_bit(R1BIO_WriteError, &r1_bio->state); |
NeilBrown | 854abd7 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 1733 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
NeilBrown | 3a9f28a | 2011-07-28 11:33:42 +1000 | [diff] [blame] | 1734 | &first_bad, &bad_sectors) && |
| 1735 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, |
| 1736 | r1_bio->sector, |
| 1737 | r1_bio->sectors, |
| 1738 | &first_bad, &bad_sectors) |
| 1739 | ) |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1740 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1741 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1742 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1743 | int s = r1_bio->sectors; |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1744 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
| 1745 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1746 | reschedule_retry(r1_bio); |
| 1747 | else { |
| 1748 | put_buf(r1_bio); |
| 1749 | md_done_sync(mddev, s, uptodate); |
| 1750 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1751 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1752 | } |
| 1753 | |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1754 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1755 | int sectors, struct page *page, int rw) |
| 1756 | { |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 1757 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1758 | /* success */ |
| 1759 | return 1; |
NeilBrown | 19d6716 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1760 | if (rw == WRITE) { |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1761 | set_bit(WriteErrorSeen, &rdev->flags); |
NeilBrown | 19d6716 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1762 | if (!test_and_set_bit(WantReplacement, |
| 1763 | &rdev->flags)) |
| 1764 | set_bit(MD_RECOVERY_NEEDED, & |
| 1765 | rdev->mddev->recovery); |
| 1766 | } |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1767 | /* need to record an error - either for the block or the device */ |
| 1768 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) |
| 1769 | md_error(rdev->mddev, rdev); |
| 1770 | return 0; |
| 1771 | } |
| 1772 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1773 | static int fix_sync_read_error(struct r1bio *r1_bio) |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1774 | { |
| 1775 | /* Try some synchronous reads of other devices to get |
| 1776 | * good data, much like with normal read errors. Only |
| 1777 | * read into the pages we already have so we don't |
| 1778 | * need to re-issue the read request. |
| 1779 | * We don't need to freeze the array, because being in an |
| 1780 | * active sync request, there is no normal IO, and |
| 1781 | * no overlapping syncs. |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1782 | * We don't need to check is_badblock() again as we |
| 1783 | * made sure that anything with a bad block in range |
| 1784 | * will have bi_end_io clear. |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1785 | */ |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1786 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1787 | struct r1conf *conf = mddev->private; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1788 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
| 1789 | sector_t sect = r1_bio->sector; |
| 1790 | int sectors = r1_bio->sectors; |
| 1791 | int idx = 0; |
| 1792 | |
| 1793 | while(sectors) { |
| 1794 | int s = sectors; |
| 1795 | int d = r1_bio->read_disk; |
| 1796 | int success = 0; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1797 | struct md_rdev *rdev; |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1798 | int start; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1799 | |
| 1800 | if (s > (PAGE_SIZE>>9)) |
| 1801 | s = PAGE_SIZE >> 9; |
| 1802 | do { |
| 1803 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { |
| 1804 | /* No rcu protection needed here devices |
| 1805 | * can only be removed when no resync is |
| 1806 | * active, and resync is currently active |
| 1807 | */ |
| 1808 | rdev = conf->mirrors[d].rdev; |
Namhyung Kim | 9d3d801 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1809 | if (sync_page_io(rdev, sect, s<<9, |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1810 | bio->bi_io_vec[idx].bv_page, |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 1811 | REQ_OP_READ, 0, false)) { |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1812 | success = 1; |
| 1813 | break; |
| 1814 | } |
| 1815 | } |
| 1816 | d++; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1817 | if (d == conf->raid_disks * 2) |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1818 | d = 0; |
| 1819 | } while (!success && d != r1_bio->read_disk); |
| 1820 | |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1821 | if (!success) { |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1822 | char b[BDEVNAME_SIZE]; |
NeilBrown | 3a9f28a | 2011-07-28 11:33:42 +1000 | [diff] [blame] | 1823 | int abort = 0; |
| 1824 | /* Cannot read from anywhere, this block is lost. |
| 1825 | * Record a bad block on each device. If that doesn't |
| 1826 | * work just disable and interrupt the recovery. |
| 1827 | * Don't fail devices as that won't really help. |
| 1828 | */ |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1829 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
| 1830 | " for block %llu\n", |
| 1831 | mdname(mddev), |
| 1832 | bdevname(bio->bi_bdev, b), |
| 1833 | (unsigned long long)r1_bio->sector); |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1834 | for (d = 0; d < conf->raid_disks * 2; d++) { |
NeilBrown | 3a9f28a | 2011-07-28 11:33:42 +1000 | [diff] [blame] | 1835 | rdev = conf->mirrors[d].rdev; |
| 1836 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
| 1837 | continue; |
| 1838 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
| 1839 | abort = 1; |
| 1840 | } |
| 1841 | if (abort) { |
NeilBrown | d890fa2 | 2011-10-26 11:54:39 +1100 | [diff] [blame] | 1842 | conf->recovery_disabled = |
| 1843 | mddev->recovery_disabled; |
NeilBrown | 3a9f28a | 2011-07-28 11:33:42 +1000 | [diff] [blame] | 1844 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
| 1845 | md_done_sync(mddev, r1_bio->sectors, 0); |
| 1846 | put_buf(r1_bio); |
| 1847 | return 0; |
| 1848 | } |
| 1849 | /* Try next page */ |
| 1850 | sectors -= s; |
| 1851 | sect += s; |
| 1852 | idx++; |
| 1853 | continue; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1854 | } |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1855 | |
| 1856 | start = d; |
| 1857 | /* write it back and re-read */ |
| 1858 | while (d != r1_bio->read_disk) { |
| 1859 | if (d == 0) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1860 | d = conf->raid_disks * 2; |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1861 | d--; |
| 1862 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) |
| 1863 | continue; |
| 1864 | rdev = conf->mirrors[d].rdev; |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1865 | if (r1_sync_page_io(rdev, sect, s, |
| 1866 | bio->bi_io_vec[idx].bv_page, |
| 1867 | WRITE) == 0) { |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1868 | r1_bio->bios[d]->bi_end_io = NULL; |
| 1869 | rdev_dec_pending(rdev, mddev); |
Namhyung Kim | 9d3d801 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1870 | } |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1871 | } |
| 1872 | d = start; |
| 1873 | while (d != r1_bio->read_disk) { |
| 1874 | if (d == 0) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1875 | d = conf->raid_disks * 2; |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1876 | d--; |
| 1877 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) |
| 1878 | continue; |
| 1879 | rdev = conf->mirrors[d].rdev; |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1880 | if (r1_sync_page_io(rdev, sect, s, |
| 1881 | bio->bi_io_vec[idx].bv_page, |
| 1882 | READ) != 0) |
Namhyung Kim | 9d3d801 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1883 | atomic_add(s, &rdev->corrected_errors); |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1884 | } |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1885 | sectors -= s; |
| 1886 | sect += s; |
| 1887 | idx ++; |
| 1888 | } |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1889 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1890 | bio->bi_error = 0; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1891 | return 1; |
| 1892 | } |
| 1893 | |
NeilBrown | c95e638 | 2014-09-09 13:54:11 +1000 | [diff] [blame] | 1894 | static void process_checks(struct r1bio *r1_bio) |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1895 | { |
| 1896 | /* We have read all readable devices. If we haven't |
| 1897 | * got the block, then there is no hope left. |
| 1898 | * If we have, then we want to do a comparison |
| 1899 | * and skip the write if everything is the same. |
| 1900 | * If any blocks failed to read, then we need to |
| 1901 | * attempt an over-write |
| 1902 | */ |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1903 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1904 | struct r1conf *conf = mddev->private; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1905 | int primary; |
| 1906 | int i; |
majianpeng | f4380a9 | 2012-04-12 16:04:47 +1000 | [diff] [blame] | 1907 | int vcnt; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1908 | |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1909 | /* Fix variable parts of all bios */ |
| 1910 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); |
| 1911 | for (i = 0; i < conf->raid_disks * 2; i++) { |
| 1912 | int j; |
| 1913 | int size; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1914 | int error; |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1915 | struct bio *b = r1_bio->bios[i]; |
| 1916 | if (b->bi_end_io != end_sync_read) |
| 1917 | continue; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1918 | /* fixup the bio for reuse, but preserve errno */ |
| 1919 | error = b->bi_error; |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1920 | bio_reset(b); |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1921 | b->bi_error = error; |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1922 | b->bi_vcnt = vcnt; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1923 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
| 1924 | b->bi_iter.bi_sector = r1_bio->sector + |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1925 | conf->mirrors[i].rdev->data_offset; |
| 1926 | b->bi_bdev = conf->mirrors[i].rdev->bdev; |
| 1927 | b->bi_end_io = end_sync_read; |
| 1928 | b->bi_private = r1_bio; |
| 1929 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1930 | size = b->bi_iter.bi_size; |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1931 | for (j = 0; j < vcnt ; j++) { |
| 1932 | struct bio_vec *bi; |
| 1933 | bi = &b->bi_io_vec[j]; |
| 1934 | bi->bv_offset = 0; |
| 1935 | if (size > PAGE_SIZE) |
| 1936 | bi->bv_len = PAGE_SIZE; |
| 1937 | else |
| 1938 | bi->bv_len = size; |
| 1939 | size -= PAGE_SIZE; |
| 1940 | } |
| 1941 | } |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1942 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1943 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1944 | !r1_bio->bios[primary]->bi_error) { |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1945 | r1_bio->bios[primary]->bi_end_io = NULL; |
| 1946 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); |
| 1947 | break; |
| 1948 | } |
| 1949 | r1_bio->read_disk = primary; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1950 | for (i = 0; i < conf->raid_disks * 2; i++) { |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1951 | int j; |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1952 | struct bio *pbio = r1_bio->bios[primary]; |
| 1953 | struct bio *sbio = r1_bio->bios[i]; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1954 | int error = sbio->bi_error; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1955 | |
Kent Overstreet | 2aabaa6 | 2012-09-11 11:26:12 -0700 | [diff] [blame] | 1956 | if (sbio->bi_end_io != end_sync_read) |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1957 | continue; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1958 | /* Now we can 'fixup' the error value */ |
| 1959 | sbio->bi_error = 0; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1960 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1961 | if (!error) { |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1962 | for (j = vcnt; j-- ; ) { |
| 1963 | struct page *p, *s; |
| 1964 | p = pbio->bi_io_vec[j].bv_page; |
| 1965 | s = sbio->bi_io_vec[j].bv_page; |
| 1966 | if (memcmp(page_address(p), |
| 1967 | page_address(s), |
NeilBrown | 5020ad7 | 2012-04-02 01:39:05 +1000 | [diff] [blame] | 1968 | sbio->bi_io_vec[j].bv_len)) |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1969 | break; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1970 | } |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1971 | } else |
| 1972 | j = 0; |
| 1973 | if (j >= 0) |
Jianpeng Ma | 7f7583d | 2012-10-11 14:17:59 +1100 | [diff] [blame] | 1974 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1975 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1976 | && !error)) { |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1977 | /* No need to write to this device. */ |
| 1978 | sbio->bi_end_io = NULL; |
| 1979 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); |
| 1980 | continue; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1981 | } |
Kent Overstreet | d3b45c2 | 2012-09-10 13:49:33 -0700 | [diff] [blame] | 1982 | |
| 1983 | bio_copy_data(sbio, pbio); |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1984 | } |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1985 | } |
| 1986 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1987 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1988 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1989 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1990 | int i; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1991 | int disks = conf->raid_disks * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1992 | struct bio *bio, *wbio; |
| 1993 | |
| 1994 | bio = r1_bio->bios[r1_bio->read_disk]; |
| 1995 | |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1996 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
| 1997 | /* ouch - failed to read all of that. */ |
| 1998 | if (!fix_sync_read_error(r1_bio)) |
| 1999 | return; |
NeilBrown | 7ca78d5 | 2011-05-11 14:50:37 +1000 | [diff] [blame] | 2000 | |
| 2001 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
NeilBrown | c95e638 | 2014-09-09 13:54:11 +1000 | [diff] [blame] | 2002 | process_checks(r1_bio); |
| 2003 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2004 | /* |
| 2005 | * schedule writes |
| 2006 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2007 | atomic_set(&r1_bio->remaining, 1); |
| 2008 | for (i = 0; i < disks ; i++) { |
| 2009 | wbio = r1_bio->bios[i]; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2010 | if (wbio->bi_end_io == NULL || |
| 2011 | (wbio->bi_end_io == end_sync_read && |
| 2012 | (i == r1_bio->read_disk || |
| 2013 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2014 | continue; |
| 2015 | |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 2016 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2017 | wbio->bi_end_io = end_sync_write; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2018 | atomic_inc(&r1_bio->remaining); |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 2019 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2020 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2021 | generic_make_request(wbio); |
| 2022 | } |
| 2023 | |
| 2024 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2025 | /* if we're here, all write(s) have completed, so clean up */ |
NeilBrown | 58e94ae | 2012-07-19 15:59:18 +1000 | [diff] [blame] | 2026 | int s = r1_bio->sectors; |
| 2027 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
| 2028 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2029 | reschedule_retry(r1_bio); |
| 2030 | else { |
| 2031 | put_buf(r1_bio); |
| 2032 | md_done_sync(mddev, s, 1); |
| 2033 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2034 | } |
| 2035 | } |
| 2036 | |
| 2037 | /* |
| 2038 | * This is a kernel thread which: |
| 2039 | * |
| 2040 | * 1. Retries failed read operations on working mirrors. |
| 2041 | * 2. Updates the raid superblock when problems encounter. |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2042 | * 3. Performs writes following reads for array synchronising. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2043 | */ |
| 2044 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2045 | static void fix_read_error(struct r1conf *conf, int read_disk, |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2046 | sector_t sect, int sectors) |
| 2047 | { |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2048 | struct mddev *mddev = conf->mddev; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2049 | while(sectors) { |
| 2050 | int s = sectors; |
| 2051 | int d = read_disk; |
| 2052 | int success = 0; |
| 2053 | int start; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2054 | struct md_rdev *rdev; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2055 | |
| 2056 | if (s > (PAGE_SIZE>>9)) |
| 2057 | s = PAGE_SIZE >> 9; |
| 2058 | |
| 2059 | do { |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2060 | sector_t first_bad; |
| 2061 | int bad_sectors; |
| 2062 | |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2063 | rcu_read_lock(); |
| 2064 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2065 | if (rdev && |
majianpeng | da8840a | 2012-05-22 13:55:03 +1000 | [diff] [blame] | 2066 | (test_bit(In_sync, &rdev->flags) || |
| 2067 | (!test_bit(Faulty, &rdev->flags) && |
| 2068 | rdev->recovery_offset >= sect + s)) && |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2069 | is_badblock(rdev, sect, s, |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2070 | &first_bad, &bad_sectors) == 0) { |
| 2071 | atomic_inc(&rdev->nr_pending); |
| 2072 | rcu_read_unlock(); |
| 2073 | if (sync_page_io(rdev, sect, s<<9, |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 2074 | conf->tmppage, REQ_OP_READ, 0, false)) |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2075 | success = 1; |
| 2076 | rdev_dec_pending(rdev, mddev); |
| 2077 | if (success) |
| 2078 | break; |
| 2079 | } else |
| 2080 | rcu_read_unlock(); |
| 2081 | d++; |
| 2082 | if (d == conf->raid_disks * 2) |
| 2083 | d = 0; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2084 | } while (!success && d != read_disk); |
| 2085 | |
| 2086 | if (!success) { |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2087 | /* Cannot read from anywhere - mark it bad */ |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2088 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2089 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
| 2090 | md_error(mddev, rdev); |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2091 | break; |
| 2092 | } |
| 2093 | /* write it back and re-read */ |
| 2094 | start = d; |
| 2095 | while (d != read_disk) { |
| 2096 | if (d==0) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2097 | d = conf->raid_disks * 2; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2098 | d--; |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2099 | rcu_read_lock(); |
| 2100 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2101 | if (rdev && |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2102 | !test_bit(Faulty, &rdev->flags)) { |
| 2103 | atomic_inc(&rdev->nr_pending); |
| 2104 | rcu_read_unlock(); |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2105 | r1_sync_page_io(rdev, sect, s, |
| 2106 | conf->tmppage, WRITE); |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2107 | rdev_dec_pending(rdev, mddev); |
| 2108 | } else |
| 2109 | rcu_read_unlock(); |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2110 | } |
| 2111 | d = start; |
| 2112 | while (d != read_disk) { |
| 2113 | char b[BDEVNAME_SIZE]; |
| 2114 | if (d==0) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2115 | d = conf->raid_disks * 2; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2116 | d--; |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2117 | rcu_read_lock(); |
| 2118 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2119 | if (rdev && |
NeilBrown | b8cb6b4 | 2014-09-18 11:09:04 +1000 | [diff] [blame] | 2120 | !test_bit(Faulty, &rdev->flags)) { |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2121 | atomic_inc(&rdev->nr_pending); |
| 2122 | rcu_read_unlock(); |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2123 | if (r1_sync_page_io(rdev, sect, s, |
| 2124 | conf->tmppage, READ)) { |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2125 | atomic_add(s, &rdev->corrected_errors); |
| 2126 | printk(KERN_INFO |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2127 | "md/raid1:%s: read error corrected " |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2128 | "(%d sectors at %llu on %s)\n", |
| 2129 | mdname(mddev), s, |
Randy Dunlap | 969b755 | 2006-10-28 10:38:32 -0700 | [diff] [blame] | 2130 | (unsigned long long)(sect + |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2131 | rdev->data_offset), |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2132 | bdevname(rdev->bdev, b)); |
| 2133 | } |
NeilBrown | 707a6a4 | 2016-06-02 16:19:52 +1000 | [diff] [blame] | 2134 | rdev_dec_pending(rdev, mddev); |
| 2135 | } else |
| 2136 | rcu_read_unlock(); |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2137 | } |
| 2138 | sectors -= s; |
| 2139 | sect += s; |
| 2140 | } |
| 2141 | } |
| 2142 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 2143 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2144 | { |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2145 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2146 | struct r1conf *conf = mddev->private; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2147 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2148 | |
| 2149 | /* bio has the data to be written to device 'i' where |
| 2150 | * we just recently had a write error. |
| 2151 | * We repeatedly clone the bio and trim down to one block, |
| 2152 | * then try the write. Where the write fails we record |
| 2153 | * a bad block. |
| 2154 | * It is conceivable that the bio doesn't exactly align with |
| 2155 | * blocks. We must handle this somehow. |
| 2156 | * |
| 2157 | * We currently own a reference on the rdev. |
| 2158 | */ |
| 2159 | |
| 2160 | int block_sectors; |
| 2161 | sector_t sector; |
| 2162 | int sectors; |
| 2163 | int sect_to_write = r1_bio->sectors; |
| 2164 | int ok = 1; |
| 2165 | |
| 2166 | if (rdev->badblocks.shift < 0) |
| 2167 | return 0; |
| 2168 | |
Nate Dailey | ab713cd | 2015-02-12 12:02:09 -0500 | [diff] [blame] | 2169 | block_sectors = roundup(1 << rdev->badblocks.shift, |
| 2170 | bdev_logical_block_size(rdev->bdev) >> 9); |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2171 | sector = r1_bio->sector; |
| 2172 | sectors = ((sector + block_sectors) |
| 2173 | & ~(sector_t)(block_sectors - 1)) |
| 2174 | - sector; |
| 2175 | |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2176 | while (sect_to_write) { |
| 2177 | struct bio *wbio; |
| 2178 | if (sectors > sect_to_write) |
| 2179 | sectors = sect_to_write; |
| 2180 | /* Write at 'sector' for 'sectors'*/ |
| 2181 | |
Kent Overstreet | b783863 | 2012-09-10 15:17:11 -0700 | [diff] [blame] | 2182 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
| 2183 | unsigned vcnt = r1_bio->behind_page_count; |
| 2184 | struct bio_vec *vec = r1_bio->behind_bvecs; |
| 2185 | |
| 2186 | while (!vec->bv_page) { |
| 2187 | vec++; |
| 2188 | vcnt--; |
| 2189 | } |
| 2190 | |
| 2191 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); |
| 2192 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); |
| 2193 | |
| 2194 | wbio->bi_vcnt = vcnt; |
| 2195 | } else { |
| 2196 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); |
| 2197 | } |
| 2198 | |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 2199 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2200 | wbio->bi_iter.bi_sector = r1_bio->sector; |
| 2201 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2202 | |
Kent Overstreet | 6678d83 | 2013-08-07 11:14:32 -0700 | [diff] [blame] | 2203 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2204 | wbio->bi_iter.bi_sector += rdev->data_offset; |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2205 | wbio->bi_bdev = rdev->bdev; |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 2206 | |
| 2207 | if (submit_bio_wait(wbio) < 0) |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2208 | /* failure! */ |
| 2209 | ok = rdev_set_badblocks(rdev, sector, |
| 2210 | sectors, 0) |
| 2211 | && ok; |
| 2212 | |
| 2213 | bio_put(wbio); |
| 2214 | sect_to_write -= sectors; |
| 2215 | sector += sectors; |
| 2216 | sectors = block_sectors; |
| 2217 | } |
| 2218 | return ok; |
| 2219 | } |
| 2220 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2221 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2222 | { |
| 2223 | int m; |
| 2224 | int s = r1_bio->sectors; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2225 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2226 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2227 | struct bio *bio = r1_bio->bios[m]; |
| 2228 | if (bio->bi_end_io == NULL) |
| 2229 | continue; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2230 | if (!bio->bi_error && |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2231 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
NeilBrown | c6563a8 | 2012-05-21 09:27:00 +1000 | [diff] [blame] | 2232 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2233 | } |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2234 | if (bio->bi_error && |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2235 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
| 2236 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) |
| 2237 | md_error(conf->mddev, rdev); |
| 2238 | } |
| 2239 | } |
| 2240 | put_buf(r1_bio); |
| 2241 | md_done_sync(conf->mddev, s, 1); |
| 2242 | } |
| 2243 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2244 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2245 | { |
| 2246 | int m; |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2247 | bool fail = false; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2248 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2249 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2250 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2251 | rdev_clear_badblocks(rdev, |
| 2252 | r1_bio->sector, |
NeilBrown | c6563a8 | 2012-05-21 09:27:00 +1000 | [diff] [blame] | 2253 | r1_bio->sectors, 0); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2254 | rdev_dec_pending(rdev, conf->mddev); |
| 2255 | } else if (r1_bio->bios[m] != NULL) { |
| 2256 | /* This drive got a write error. We need to |
| 2257 | * narrow down and record precise write |
| 2258 | * errors. |
| 2259 | */ |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2260 | fail = true; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2261 | if (!narrow_write_error(r1_bio, m)) { |
| 2262 | md_error(conf->mddev, |
| 2263 | conf->mirrors[m].rdev); |
| 2264 | /* an I/O failed, we can't clear the bitmap */ |
| 2265 | set_bit(R1BIO_Degraded, &r1_bio->state); |
| 2266 | } |
| 2267 | rdev_dec_pending(conf->mirrors[m].rdev, |
| 2268 | conf->mddev); |
| 2269 | } |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2270 | if (fail) { |
| 2271 | spin_lock_irq(&conf->device_lock); |
| 2272 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); |
Nate Dailey | ccfc7bf | 2016-02-29 10:43:58 -0500 | [diff] [blame] | 2273 | conf->nr_queued++; |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2274 | spin_unlock_irq(&conf->device_lock); |
| 2275 | md_wakeup_thread(conf->mddev->thread); |
NeilBrown | bd8688a | 2015-10-24 16:02:16 +1100 | [diff] [blame] | 2276 | } else { |
| 2277 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2278 | close_write(r1_bio); |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2279 | raid_end_bio_io(r1_bio); |
NeilBrown | bd8688a | 2015-10-24 16:02:16 +1100 | [diff] [blame] | 2280 | } |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2281 | } |
| 2282 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2283 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2284 | { |
| 2285 | int disk; |
| 2286 | int max_sectors; |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2287 | struct mddev *mddev = conf->mddev; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2288 | struct bio *bio; |
| 2289 | char b[BDEVNAME_SIZE]; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2290 | struct md_rdev *rdev; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2291 | |
| 2292 | clear_bit(R1BIO_ReadError, &r1_bio->state); |
| 2293 | /* we got a read error. Maybe the drive is bad. Maybe just |
| 2294 | * the block and we can fix it. |
| 2295 | * We freeze all other IO, and try reading the block from |
| 2296 | * other devices. When we find one, we re-write |
| 2297 | * and check it that fixes the read error. |
| 2298 | * This is all done synchronously while the array is |
| 2299 | * frozen |
| 2300 | */ |
| 2301 | if (mddev->ro == 0) { |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 2302 | freeze_array(conf, 1); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2303 | fix_read_error(conf, r1_bio->read_disk, |
| 2304 | r1_bio->sector, r1_bio->sectors); |
| 2305 | unfreeze_array(conf); |
| 2306 | } else |
| 2307 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); |
NeilBrown | 7ad4d4a | 2012-10-11 13:44:30 +1100 | [diff] [blame] | 2308 | rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2309 | |
| 2310 | bio = r1_bio->bios[r1_bio->read_disk]; |
| 2311 | bdevname(bio->bi_bdev, b); |
| 2312 | read_more: |
| 2313 | disk = read_balance(conf, r1_bio, &max_sectors); |
| 2314 | if (disk == -1) { |
| 2315 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" |
| 2316 | " read error for block %llu\n", |
| 2317 | mdname(mddev), b, (unsigned long long)r1_bio->sector); |
| 2318 | raid_end_bio_io(r1_bio); |
| 2319 | } else { |
| 2320 | const unsigned long do_sync |
Jens Axboe | 1eff9d3 | 2016-08-05 15:35:16 -0600 | [diff] [blame] | 2321 | = r1_bio->master_bio->bi_opf & REQ_SYNC; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2322 | if (bio) { |
| 2323 | r1_bio->bios[r1_bio->read_disk] = |
| 2324 | mddev->ro ? IO_BLOCKED : NULL; |
| 2325 | bio_put(bio); |
| 2326 | } |
| 2327 | r1_bio->read_disk = disk; |
| 2328 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2329 | bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, |
| 2330 | max_sectors); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2331 | r1_bio->bios[r1_bio->read_disk] = bio; |
| 2332 | rdev = conf->mirrors[disk].rdev; |
| 2333 | printk_ratelimited(KERN_ERR |
| 2334 | "md/raid1:%s: redirecting sector %llu" |
| 2335 | " to other mirror: %s\n", |
| 2336 | mdname(mddev), |
| 2337 | (unsigned long long)r1_bio->sector, |
| 2338 | bdevname(rdev->bdev, b)); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2339 | bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2340 | bio->bi_bdev = rdev->bdev; |
| 2341 | bio->bi_end_io = raid1_end_read_request; |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 2342 | bio_set_op_attrs(bio, REQ_OP_READ, do_sync); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2343 | bio->bi_private = r1_bio; |
| 2344 | if (max_sectors < r1_bio->sectors) { |
| 2345 | /* Drat - have to split this up more */ |
| 2346 | struct bio *mbio = r1_bio->master_bio; |
| 2347 | int sectors_handled = (r1_bio->sector + max_sectors |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2348 | - mbio->bi_iter.bi_sector); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2349 | r1_bio->sectors = max_sectors; |
| 2350 | spin_lock_irq(&conf->device_lock); |
| 2351 | if (mbio->bi_phys_segments == 0) |
| 2352 | mbio->bi_phys_segments = 2; |
| 2353 | else |
| 2354 | mbio->bi_phys_segments++; |
| 2355 | spin_unlock_irq(&conf->device_lock); |
| 2356 | generic_make_request(bio); |
| 2357 | bio = NULL; |
| 2358 | |
| 2359 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 2360 | |
| 2361 | r1_bio->master_bio = mbio; |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 2362 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2363 | r1_bio->state = 0; |
| 2364 | set_bit(R1BIO_ReadError, &r1_bio->state); |
| 2365 | r1_bio->mddev = mddev; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2366 | r1_bio->sector = mbio->bi_iter.bi_sector + |
| 2367 | sectors_handled; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2368 | |
| 2369 | goto read_more; |
| 2370 | } else |
| 2371 | generic_make_request(bio); |
| 2372 | } |
| 2373 | } |
| 2374 | |
Shaohua Li | 4ed8731 | 2012-10-11 13:34:00 +1100 | [diff] [blame] | 2375 | static void raid1d(struct md_thread *thread) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2376 | { |
Shaohua Li | 4ed8731 | 2012-10-11 13:34:00 +1100 | [diff] [blame] | 2377 | struct mddev *mddev = thread->mddev; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 2378 | struct r1bio *r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2379 | unsigned long flags; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2380 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2381 | struct list_head *head = &conf->retry_list; |
NeilBrown | e1dfa0a | 2011-04-18 18:25:41 +1000 | [diff] [blame] | 2382 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2383 | |
| 2384 | md_check_recovery(mddev); |
NeilBrown | e1dfa0a | 2011-04-18 18:25:41 +1000 | [diff] [blame] | 2385 | |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2386 | if (!list_empty_careful(&conf->bio_end_io_list) && |
| 2387 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { |
| 2388 | LIST_HEAD(tmp); |
| 2389 | spin_lock_irqsave(&conf->device_lock, flags); |
| 2390 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { |
Nate Dailey | ccfc7bf | 2016-02-29 10:43:58 -0500 | [diff] [blame] | 2391 | while (!list_empty(&conf->bio_end_io_list)) { |
| 2392 | list_move(conf->bio_end_io_list.prev, &tmp); |
| 2393 | conf->nr_queued--; |
| 2394 | } |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2395 | } |
| 2396 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 2397 | while (!list_empty(&tmp)) { |
Mikulas Patocka | a452744 | 2015-10-01 15:17:43 -0400 | [diff] [blame] | 2398 | r1_bio = list_first_entry(&tmp, struct r1bio, |
| 2399 | retry_list); |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2400 | list_del(&r1_bio->retry_list); |
NeilBrown | bd8688a | 2015-10-24 16:02:16 +1100 | [diff] [blame] | 2401 | if (mddev->degraded) |
| 2402 | set_bit(R1BIO_Degraded, &r1_bio->state); |
| 2403 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2404 | close_write(r1_bio); |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2405 | raid_end_bio_io(r1_bio); |
| 2406 | } |
| 2407 | } |
| 2408 | |
NeilBrown | e1dfa0a | 2011-04-18 18:25:41 +1000 | [diff] [blame] | 2409 | blk_start_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2410 | for (;;) { |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 2411 | |
NeilBrown | 0021b7b | 2012-07-31 09:08:14 +0200 | [diff] [blame] | 2412 | flush_pending_writes(conf); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 2413 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2414 | spin_lock_irqsave(&conf->device_lock, flags); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 2415 | if (list_empty(head)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2416 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2417 | break; |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 2418 | } |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 2419 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2420 | list_del(head->prev); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 2421 | conf->nr_queued--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2422 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 2423 | |
| 2424 | mddev = r1_bio->mddev; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 2425 | conf = mddev->private; |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 2426 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2427 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2428 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2429 | handle_sync_write_finished(conf, r1_bio); |
| 2430 | else |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 2431 | sync_request_write(mddev, r1_bio); |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2432 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2433 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2434 | handle_write_finished(conf, r1_bio); |
| 2435 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) |
| 2436 | handle_read_error(conf, r1_bio); |
| 2437 | else |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2438 | /* just a partial read to be scheduled from separate |
| 2439 | * context |
| 2440 | */ |
| 2441 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2442 | |
NeilBrown | 1d9d524 | 2009-10-16 15:55:32 +1100 | [diff] [blame] | 2443 | cond_resched(); |
NeilBrown | de393cd | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2444 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
| 2445 | md_check_recovery(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2446 | } |
NeilBrown | e1dfa0a | 2011-04-18 18:25:41 +1000 | [diff] [blame] | 2447 | blk_finish_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2448 | } |
| 2449 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2450 | static int init_resync(struct r1conf *conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2451 | { |
| 2452 | int buffs; |
| 2453 | |
| 2454 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; |
Eric Sesterhenn | 9e77c48 | 2006-04-01 01:08:49 +0200 | [diff] [blame] | 2455 | BUG_ON(conf->r1buf_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2456 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
| 2457 | conf->poolinfo); |
| 2458 | if (!conf->r1buf_pool) |
| 2459 | return -ENOMEM; |
| 2460 | conf->next_resync = 0; |
| 2461 | return 0; |
| 2462 | } |
| 2463 | |
| 2464 | /* |
| 2465 | * perform a "sync" on one "block" |
| 2466 | * |
| 2467 | * We need to make sure that no normal I/O request - particularly write |
| 2468 | * requests - conflict with active sync requests. |
| 2469 | * |
| 2470 | * This is achieved by tracking pending requests and a 'barrier' concept |
| 2471 | * that can be installed to exclude normal IO requests. |
| 2472 | */ |
| 2473 | |
Shaohua Li | 849674e | 2016-01-20 13:52:20 -0800 | [diff] [blame] | 2474 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
| 2475 | int *skipped) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2476 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2477 | struct r1conf *conf = mddev->private; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 2478 | struct r1bio *r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2479 | struct bio *bio; |
| 2480 | sector_t max_sector, nr_sectors; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2481 | int disk = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2482 | int i; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2483 | int wonly = -1; |
| 2484 | int write_targets = 0, read_targets = 0; |
NeilBrown | 57dab0b | 2010-10-19 10:03:39 +1100 | [diff] [blame] | 2485 | sector_t sync_blocks; |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 2486 | int still_degraded = 0; |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2487 | int good_sectors = RESYNC_SECTORS; |
| 2488 | int min_bad = 0; /* number of sectors that are bad in all devices */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2489 | |
| 2490 | if (!conf->r1buf_pool) |
| 2491 | if (init_resync(conf)) |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 2492 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2493 | |
Andre Noll | 58c0fed | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 2494 | max_sector = mddev->dev_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2495 | if (sector_nr >= max_sector) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2496 | /* If we aborted, we need to abort the |
| 2497 | * sync on the 'current' bitmap chunk (there will |
| 2498 | * only be one in raid1 resync. |
| 2499 | * We can find the current addess in mddev->curr_resync |
| 2500 | */ |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2501 | if (mddev->curr_resync < max_sector) /* aborted */ |
| 2502 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2503 | &sync_blocks, 1); |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2504 | else /* completed sync */ |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2505 | conf->fullsync = 0; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2506 | |
| 2507 | bitmap_close_sync(mddev->bitmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2508 | close_sync(conf); |
Goldwyn Rodrigues | c40f341 | 2015-08-19 08:14:42 +1000 | [diff] [blame] | 2509 | |
| 2510 | if (mddev_is_clustered(mddev)) { |
| 2511 | conf->cluster_sync_low = 0; |
| 2512 | conf->cluster_sync_high = 0; |
Goldwyn Rodrigues | c40f341 | 2015-08-19 08:14:42 +1000 | [diff] [blame] | 2513 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2514 | return 0; |
| 2515 | } |
| 2516 | |
NeilBrown | 07d84d10 | 2006-06-26 00:27:56 -0700 | [diff] [blame] | 2517 | if (mddev->bitmap == NULL && |
| 2518 | mddev->recovery_cp == MaxSector && |
NeilBrown | 6394cca | 2006-08-27 01:23:50 -0700 | [diff] [blame] | 2519 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
NeilBrown | 07d84d10 | 2006-06-26 00:27:56 -0700 | [diff] [blame] | 2520 | conf->fullsync == 0) { |
| 2521 | *skipped = 1; |
| 2522 | return max_sector - sector_nr; |
| 2523 | } |
NeilBrown | 6394cca | 2006-08-27 01:23:50 -0700 | [diff] [blame] | 2524 | /* before building a request, check if we can skip these blocks.. |
| 2525 | * This call the bitmap_start_sync doesn't actually record anything |
| 2526 | */ |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 2527 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
NeilBrown | e5de485f | 2005-11-08 21:39:38 -0800 | [diff] [blame] | 2528 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2529 | /* We can skip this block, and probably several more */ |
| 2530 | *skipped = 1; |
| 2531 | return sync_blocks; |
| 2532 | } |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2533 | |
Tomasz Majchrzak | 7ac5044 | 2016-06-13 15:51:19 +0200 | [diff] [blame] | 2534 | /* |
| 2535 | * If there is non-resync activity waiting for a turn, then let it |
| 2536 | * though before starting on this new sync request. |
| 2537 | */ |
| 2538 | if (conf->nr_waiting) |
| 2539 | schedule_timeout_uninterruptible(1); |
| 2540 | |
Goldwyn Rodrigues | c40f341 | 2015-08-19 08:14:42 +1000 | [diff] [blame] | 2541 | /* we are incrementing sector_nr below. To be safe, we check against |
| 2542 | * sector_nr + two times RESYNC_SECTORS |
| 2543 | */ |
| 2544 | |
| 2545 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
| 2546 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); |
NeilBrown | 1c4588e | 2010-10-26 17:41:22 +1100 | [diff] [blame] | 2547 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2548 | |
NeilBrown | c2fd4c9 | 2014-09-10 16:01:24 +1000 | [diff] [blame] | 2549 | raise_barrier(conf, sector_nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2550 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2551 | rcu_read_lock(); |
| 2552 | /* |
| 2553 | * If we get a correctably read error during resync or recovery, |
| 2554 | * we might want to read from a different device. So we |
| 2555 | * flag all drives that could conceivably be read from for READ, |
| 2556 | * and any others (which will be non-In_sync devices) for WRITE. |
| 2557 | * If a read fails, we try reading from something else for which READ |
| 2558 | * is OK. |
| 2559 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2560 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2561 | r1_bio->mddev = mddev; |
| 2562 | r1_bio->sector = sector_nr; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2563 | r1_bio->state = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2564 | set_bit(R1BIO_IsSync, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2565 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2566 | for (i = 0; i < conf->raid_disks * 2; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2567 | struct md_rdev *rdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2568 | bio = r1_bio->bios[i]; |
Kent Overstreet | 2aabaa6 | 2012-09-11 11:26:12 -0700 | [diff] [blame] | 2569 | bio_reset(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2570 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2571 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
| 2572 | if (rdev == NULL || |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2573 | test_bit(Faulty, &rdev->flags)) { |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2574 | if (i < conf->raid_disks) |
| 2575 | still_degraded = 1; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2576 | } else if (!test_bit(In_sync, &rdev->flags)) { |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 2577 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2578 | bio->bi_end_io = end_sync_write; |
| 2579 | write_targets ++; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2580 | } else { |
| 2581 | /* may need to read from here */ |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2582 | sector_t first_bad = MaxSector; |
| 2583 | int bad_sectors; |
| 2584 | |
| 2585 | if (is_badblock(rdev, sector_nr, good_sectors, |
| 2586 | &first_bad, &bad_sectors)) { |
| 2587 | if (first_bad > sector_nr) |
| 2588 | good_sectors = first_bad - sector_nr; |
| 2589 | else { |
| 2590 | bad_sectors -= (sector_nr - first_bad); |
| 2591 | if (min_bad == 0 || |
| 2592 | min_bad > bad_sectors) |
| 2593 | min_bad = bad_sectors; |
| 2594 | } |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2595 | } |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2596 | if (sector_nr < first_bad) { |
| 2597 | if (test_bit(WriteMostly, &rdev->flags)) { |
| 2598 | if (wonly < 0) |
| 2599 | wonly = i; |
| 2600 | } else { |
| 2601 | if (disk < 0) |
| 2602 | disk = i; |
| 2603 | } |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 2604 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2605 | bio->bi_end_io = end_sync_read; |
| 2606 | read_targets++; |
Alexander Lyakas | d57368a | 2012-07-17 13:17:55 +0300 | [diff] [blame] | 2607 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
| 2608 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
| 2609 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
| 2610 | /* |
| 2611 | * The device is suitable for reading (InSync), |
| 2612 | * but has bad block(s) here. Let's try to correct them, |
| 2613 | * if we are doing resync or repair. Otherwise, leave |
| 2614 | * this device alone for this sync request. |
| 2615 | */ |
Mike Christie | 796a5cf | 2016-06-05 14:32:07 -0500 | [diff] [blame] | 2616 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
Alexander Lyakas | d57368a | 2012-07-17 13:17:55 +0300 | [diff] [blame] | 2617 | bio->bi_end_io = end_sync_write; |
| 2618 | write_targets++; |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2619 | } |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2620 | } |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2621 | if (bio->bi_end_io) { |
| 2622 | atomic_inc(&rdev->nr_pending); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2623 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2624 | bio->bi_bdev = rdev->bdev; |
| 2625 | bio->bi_private = r1_bio; |
| 2626 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2627 | } |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2628 | rcu_read_unlock(); |
| 2629 | if (disk < 0) |
| 2630 | disk = wonly; |
| 2631 | r1_bio->read_disk = disk; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2632 | |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2633 | if (read_targets == 0 && min_bad > 0) { |
| 2634 | /* These sectors are bad on all InSync devices, so we |
| 2635 | * need to mark them bad on all write targets |
| 2636 | */ |
| 2637 | int ok = 1; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2638 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2639 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
majianpeng | a42f9d8 | 2012-04-02 01:04:19 +1000 | [diff] [blame] | 2640 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2641 | ok = rdev_set_badblocks(rdev, sector_nr, |
| 2642 | min_bad, 0 |
| 2643 | ) && ok; |
| 2644 | } |
| 2645 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
| 2646 | *skipped = 1; |
| 2647 | put_buf(r1_bio); |
| 2648 | |
| 2649 | if (!ok) { |
| 2650 | /* Cannot record the badblocks, so need to |
| 2651 | * abort the resync. |
| 2652 | * If there are multiple read targets, could just |
| 2653 | * fail the really bad ones ??? |
| 2654 | */ |
| 2655 | conf->recovery_disabled = mddev->recovery_disabled; |
| 2656 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
| 2657 | return 0; |
| 2658 | } else |
| 2659 | return min_bad; |
| 2660 | |
| 2661 | } |
| 2662 | if (min_bad > 0 && min_bad < good_sectors) { |
| 2663 | /* only resync enough to reach the next bad->good |
| 2664 | * transition */ |
| 2665 | good_sectors = min_bad; |
| 2666 | } |
| 2667 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2668 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
| 2669 | /* extra read targets are also write targets */ |
| 2670 | write_targets += read_targets-1; |
| 2671 | |
| 2672 | if (write_targets == 0 || read_targets == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2673 | /* There is nowhere to write, so all non-sync |
| 2674 | * drives must be failed - so we are finished |
| 2675 | */ |
NeilBrown | b7219cc | 2012-07-31 10:05:34 +1000 | [diff] [blame] | 2676 | sector_t rv; |
| 2677 | if (min_bad > 0) |
| 2678 | max_sector = sector_nr + min_bad; |
| 2679 | rv = max_sector - sector_nr; |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 2680 | *skipped = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2681 | put_buf(r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2682 | return rv; |
| 2683 | } |
| 2684 | |
NeilBrown | c620727 | 2008-02-06 01:39:52 -0800 | [diff] [blame] | 2685 | if (max_sector > mddev->resync_max) |
| 2686 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2687 | if (max_sector > sector_nr + good_sectors) |
| 2688 | max_sector = sector_nr + good_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2689 | nr_sectors = 0; |
NeilBrown | 289e99e | 2005-06-21 17:17:24 -0700 | [diff] [blame] | 2690 | sync_blocks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2691 | do { |
| 2692 | struct page *page; |
| 2693 | int len = PAGE_SIZE; |
| 2694 | if (sector_nr + (len>>9) > max_sector) |
| 2695 | len = (max_sector - sector_nr) << 9; |
| 2696 | if (len == 0) |
| 2697 | break; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2698 | if (sync_blocks == 0) { |
| 2699 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
NeilBrown | e5de485f | 2005-11-08 21:39:38 -0800 | [diff] [blame] | 2700 | &sync_blocks, still_degraded) && |
| 2701 | !conf->fullsync && |
| 2702 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2703 | break; |
NeilBrown | 7571ae8 | 2010-10-07 11:54:46 +1100 | [diff] [blame] | 2704 | if ((len >> 9) > sync_blocks) |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2705 | len = sync_blocks<<9; |
NeilBrown | ab7a30c | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2706 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2707 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2708 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2709 | bio = r1_bio->bios[i]; |
| 2710 | if (bio->bi_end_io) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2711 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2712 | if (bio_add_page(bio, page, len, 0) == 0) { |
| 2713 | /* stop here */ |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2714 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2715 | while (i > 0) { |
| 2716 | i--; |
| 2717 | bio = r1_bio->bios[i]; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2718 | if (bio->bi_end_io==NULL) |
| 2719 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2720 | /* remove last page from this bio */ |
| 2721 | bio->bi_vcnt--; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2722 | bio->bi_iter.bi_size -= len; |
Jens Axboe | b7c44ed | 2015-07-24 12:37:59 -0600 | [diff] [blame] | 2723 | bio_clear_flag(bio, BIO_SEG_VALID); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2724 | } |
| 2725 | goto bio_full; |
| 2726 | } |
| 2727 | } |
| 2728 | } |
| 2729 | nr_sectors += len>>9; |
| 2730 | sector_nr += len>>9; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2731 | sync_blocks -= (len>>9); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2732 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
| 2733 | bio_full: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2734 | r1_bio->sectors = nr_sectors; |
| 2735 | |
Goldwyn Rodrigues | c40f341 | 2015-08-19 08:14:42 +1000 | [diff] [blame] | 2736 | if (mddev_is_clustered(mddev) && |
| 2737 | conf->cluster_sync_high < sector_nr + nr_sectors) { |
| 2738 | conf->cluster_sync_low = mddev->curr_resync_completed; |
| 2739 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; |
| 2740 | /* Send resync message */ |
| 2741 | md_cluster_ops->resync_info_update(mddev, |
| 2742 | conf->cluster_sync_low, |
| 2743 | conf->cluster_sync_high); |
| 2744 | } |
| 2745 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2746 | /* For a user-requested sync, we read all readable devices and do a |
| 2747 | * compare |
| 2748 | */ |
| 2749 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
| 2750 | atomic_set(&r1_bio->remaining, read_targets); |
NeilBrown | 2d4f4f3 | 2012-07-09 11:34:13 +1000 | [diff] [blame] | 2751 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2752 | bio = r1_bio->bios[i]; |
| 2753 | if (bio->bi_end_io == end_sync_read) { |
NeilBrown | 2d4f4f3 | 2012-07-09 11:34:13 +1000 | [diff] [blame] | 2754 | read_targets--; |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 2755 | md_sync_acct(bio->bi_bdev, nr_sectors); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2756 | generic_make_request(bio); |
| 2757 | } |
| 2758 | } |
| 2759 | } else { |
| 2760 | atomic_set(&r1_bio->remaining, 1); |
| 2761 | bio = r1_bio->bios[r1_bio->read_disk]; |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 2762 | md_sync_acct(bio->bi_bdev, nr_sectors); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2763 | generic_make_request(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2764 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2765 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2766 | return nr_sectors; |
| 2767 | } |
| 2768 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2769 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 2770 | { |
| 2771 | if (sectors) |
| 2772 | return sectors; |
| 2773 | |
| 2774 | return mddev->dev_sectors; |
| 2775 | } |
| 2776 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2777 | static struct r1conf *setup_conf(struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2778 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2779 | struct r1conf *conf; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2780 | int i; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 2781 | struct raid1_info *disk; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2782 | struct md_rdev *rdev; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2783 | int err = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2784 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2785 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2786 | if (!conf) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2787 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2788 | |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 2789 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2790 | * mddev->raid_disks * 2, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2791 | GFP_KERNEL); |
| 2792 | if (!conf->mirrors) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2793 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2794 | |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 2795 | conf->tmppage = alloc_page(GFP_KERNEL); |
| 2796 | if (!conf->tmppage) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2797 | goto abort; |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 2798 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2799 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2800 | if (!conf->poolinfo) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2801 | goto abort; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2802 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2803 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
| 2804 | r1bio_pool_free, |
| 2805 | conf->poolinfo); |
| 2806 | if (!conf->r1bio_pool) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2807 | goto abort; |
| 2808 | |
NeilBrown | ed9bfdf | 2009-10-16 15:55:44 +1100 | [diff] [blame] | 2809 | conf->poolinfo->mddev = mddev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2810 | |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2811 | err = -EINVAL; |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 2812 | spin_lock_init(&conf->device_lock); |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 2813 | rdev_for_each(rdev, mddev) { |
NeilBrown | aba336b | 2012-05-31 15:39:11 +1000 | [diff] [blame] | 2814 | struct request_queue *q; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2815 | int disk_idx = rdev->raid_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2816 | if (disk_idx >= mddev->raid_disks |
| 2817 | || disk_idx < 0) |
| 2818 | continue; |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2819 | if (test_bit(Replacement, &rdev->flags)) |
NeilBrown | 02b898f | 2012-10-31 11:42:03 +1100 | [diff] [blame] | 2820 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2821 | else |
| 2822 | disk = conf->mirrors + disk_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2823 | |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2824 | if (disk->rdev) |
| 2825 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2826 | disk->rdev = rdev; |
NeilBrown | aba336b | 2012-05-31 15:39:11 +1000 | [diff] [blame] | 2827 | q = bdev_get_queue(rdev->bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2828 | |
| 2829 | disk->head_position = 0; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 2830 | disk->seq_start = MaxSector; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2831 | } |
| 2832 | conf->raid_disks = mddev->raid_disks; |
| 2833 | conf->mddev = mddev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2834 | INIT_LIST_HEAD(&conf->retry_list); |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2835 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2836 | |
| 2837 | spin_lock_init(&conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2838 | init_waitqueue_head(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2839 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2840 | bio_list_init(&conf->pending_bio_list); |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 2841 | conf->pending_count = 0; |
NeilBrown | d890fa2 | 2011-10-26 11:54:39 +1100 | [diff] [blame] | 2842 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2843 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 2844 | conf->start_next_window = MaxSector; |
| 2845 | conf->current_window_requests = conf->next_window_requests = 0; |
| 2846 | |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2847 | err = -EIO; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2848 | for (i = 0; i < conf->raid_disks * 2; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2849 | |
| 2850 | disk = conf->mirrors + i; |
| 2851 | |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2852 | if (i < conf->raid_disks && |
| 2853 | disk[conf->raid_disks].rdev) { |
| 2854 | /* This slot has a replacement. */ |
| 2855 | if (!disk->rdev) { |
| 2856 | /* No original, just make the replacement |
| 2857 | * a recovering spare |
| 2858 | */ |
| 2859 | disk->rdev = |
| 2860 | disk[conf->raid_disks].rdev; |
| 2861 | disk[conf->raid_disks].rdev = NULL; |
| 2862 | } else if (!test_bit(In_sync, &disk->rdev->flags)) |
| 2863 | /* Original is not in_sync - bad */ |
| 2864 | goto abort; |
| 2865 | } |
| 2866 | |
NeilBrown | 5fd6c1d | 2006-06-26 00:27:40 -0700 | [diff] [blame] | 2867 | if (!disk->rdev || |
| 2868 | !test_bit(In_sync, &disk->rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2869 | disk->head_position = 0; |
Jonathan Brassow | 4f0a5e0 | 2012-05-22 13:55:31 +1000 | [diff] [blame] | 2870 | if (disk->rdev && |
| 2871 | (disk->rdev->saved_raid_disk < 0)) |
NeilBrown | 918f023 | 2007-08-22 14:01:52 -0700 | [diff] [blame] | 2872 | conf->fullsync = 1; |
Shaohua Li | be4d328 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 2873 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2874 | } |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2875 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2876 | err = -ENOMEM; |
NeilBrown | 0232605 | 2012-07-03 15:56:52 +1000 | [diff] [blame] | 2877 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2878 | if (!conf->thread) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2879 | printk(KERN_ERR |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2880 | "md/raid1:%s: couldn't allocate thread\n", |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2881 | mdname(mddev)); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2882 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2883 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2884 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2885 | return conf; |
| 2886 | |
| 2887 | abort: |
| 2888 | if (conf) { |
Julia Lawall | 644df1a | 2015-09-13 14:15:10 +0200 | [diff] [blame] | 2889 | mempool_destroy(conf->r1bio_pool); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2890 | kfree(conf->mirrors); |
| 2891 | safe_put_page(conf->tmppage); |
| 2892 | kfree(conf->poolinfo); |
| 2893 | kfree(conf); |
| 2894 | } |
| 2895 | return ERR_PTR(err); |
| 2896 | } |
| 2897 | |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2898 | static void raid1_free(struct mddev *mddev, void *priv); |
Shaohua Li | 849674e | 2016-01-20 13:52:20 -0800 | [diff] [blame] | 2899 | static int raid1_run(struct mddev *mddev) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2900 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2901 | struct r1conf *conf; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2902 | int i; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2903 | struct md_rdev *rdev; |
majianpeng | 5220ea1 | 2012-04-02 09:48:38 +1000 | [diff] [blame] | 2904 | int ret; |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 2905 | bool discard_supported = false; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2906 | |
| 2907 | if (mddev->level != 1) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2908 | printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n", |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2909 | mdname(mddev), mddev->level); |
| 2910 | return -EIO; |
| 2911 | } |
| 2912 | if (mddev->reshape_position != MaxSector) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2913 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2914 | mdname(mddev)); |
| 2915 | return -EIO; |
| 2916 | } |
| 2917 | /* |
| 2918 | * copy the already verified devices into our private RAID1 |
| 2919 | * bookkeeping area. [whatever we allocate in run(), |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2920 | * should be freed in raid1_free()] |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2921 | */ |
| 2922 | if (mddev->private == NULL) |
| 2923 | conf = setup_conf(mddev); |
| 2924 | else |
| 2925 | conf = mddev->private; |
| 2926 | |
| 2927 | if (IS_ERR(conf)) |
| 2928 | return PTR_ERR(conf); |
| 2929 | |
Joe Lawrence | c8dc9c6 | 2013-02-21 13:28:09 +1100 | [diff] [blame] | 2930 | if (mddev->queue) |
H. Peter Anvin | 5026d7a | 2013-06-12 07:37:43 -0700 | [diff] [blame] | 2931 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
| 2932 | |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 2933 | rdev_for_each(rdev, mddev) { |
Jonathan Brassow | 1ed7242 | 2011-06-07 17:50:35 -0500 | [diff] [blame] | 2934 | if (!mddev->gendisk) |
| 2935 | continue; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2936 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 2937 | rdev->data_offset << 9); |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 2938 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
| 2939 | discard_supported = true; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2940 | } |
| 2941 | |
| 2942 | mddev->degraded = 0; |
| 2943 | for (i=0; i < conf->raid_disks; i++) |
| 2944 | if (conf->mirrors[i].rdev == NULL || |
| 2945 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || |
| 2946 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) |
| 2947 | mddev->degraded++; |
| 2948 | |
| 2949 | if (conf->raid_disks - mddev->degraded == 1) |
| 2950 | mddev->recovery_cp = MaxSector; |
| 2951 | |
Andre Noll | 8c6ac86 | 2009-06-18 08:48:06 +1000 | [diff] [blame] | 2952 | if (mddev->recovery_cp != MaxSector) |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2953 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
Andre Noll | 8c6ac86 | 2009-06-18 08:48:06 +1000 | [diff] [blame] | 2954 | " -- starting background reconstruction\n", |
| 2955 | mdname(mddev)); |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 2956 | printk(KERN_INFO |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2957 | "md/raid1:%s: active with %d out of %d mirrors\n", |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 2958 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2959 | mddev->raid_disks); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2960 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2961 | /* |
| 2962 | * Ok, everything is just fine now |
| 2963 | */ |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2964 | mddev->thread = conf->thread; |
| 2965 | conf->thread = NULL; |
| 2966 | mddev->private = conf; |
| 2967 | |
Dan Williams | 1f40362 | 2009-03-31 14:59:03 +1100 | [diff] [blame] | 2968 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2969 | |
Jonathan Brassow | 1ed7242 | 2011-06-07 17:50:35 -0500 | [diff] [blame] | 2970 | if (mddev->queue) { |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 2971 | if (discard_supported) |
| 2972 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, |
| 2973 | mddev->queue); |
| 2974 | else |
| 2975 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, |
| 2976 | mddev->queue); |
Jonathan Brassow | 1ed7242 | 2011-06-07 17:50:35 -0500 | [diff] [blame] | 2977 | } |
majianpeng | 5220ea1 | 2012-04-02 09:48:38 +1000 | [diff] [blame] | 2978 | |
| 2979 | ret = md_integrity_register(mddev); |
NeilBrown | 5aa61f4 | 2014-12-15 12:56:57 +1100 | [diff] [blame] | 2980 | if (ret) { |
| 2981 | md_unregister_thread(&mddev->thread); |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2982 | raid1_free(mddev, conf); |
NeilBrown | 5aa61f4 | 2014-12-15 12:56:57 +1100 | [diff] [blame] | 2983 | } |
majianpeng | 5220ea1 | 2012-04-02 09:48:38 +1000 | [diff] [blame] | 2984 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2985 | } |
| 2986 | |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2987 | static void raid1_free(struct mddev *mddev, void *priv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2988 | { |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2989 | struct r1conf *conf = priv; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 2990 | |
Julia Lawall | 644df1a | 2015-09-13 14:15:10 +0200 | [diff] [blame] | 2991 | mempool_destroy(conf->r1bio_pool); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 2992 | kfree(conf->mirrors); |
Hirokazu Takahashi | 0fea7ed | 2013-04-24 11:42:44 +1000 | [diff] [blame] | 2993 | safe_put_page(conf->tmppage); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 2994 | kfree(conf->poolinfo); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2995 | kfree(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2996 | } |
| 2997 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2998 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2999 | { |
| 3000 | /* no resync is happening, and there is enough space |
| 3001 | * on all devices, so we can resize. |
| 3002 | * We need to make sure resync covers any new space. |
| 3003 | * If the array is shrinking we should possibly wait until |
| 3004 | * any io in the removed space completes, but it hardly seems |
| 3005 | * worth it. |
| 3006 | */ |
NeilBrown | a4a6125 | 2012-05-22 13:55:27 +1000 | [diff] [blame] | 3007 | sector_t newsize = raid1_size(mddev, sectors, 0); |
| 3008 | if (mddev->external_size && |
| 3009 | mddev->array_sectors > newsize) |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 3010 | return -EINVAL; |
NeilBrown | a4a6125 | 2012-05-22 13:55:27 +1000 | [diff] [blame] | 3011 | if (mddev->bitmap) { |
| 3012 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); |
| 3013 | if (ret) |
| 3014 | return ret; |
| 3015 | } |
| 3016 | md_set_array_sectors(mddev, newsize); |
Andre Noll | f233ea5 | 2008-07-21 17:05:22 +1000 | [diff] [blame] | 3017 | set_capacity(mddev->gendisk, mddev->array_sectors); |
NeilBrown | 449aad3 | 2009-08-03 10:59:58 +1000 | [diff] [blame] | 3018 | revalidate_disk(mddev->gendisk); |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 3019 | if (sectors > mddev->dev_sectors && |
NeilBrown | b098636 | 2011-05-11 15:52:21 +1000 | [diff] [blame] | 3020 | mddev->recovery_cp > mddev->dev_sectors) { |
Andre Noll | 58c0fed | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 3021 | mddev->recovery_cp = mddev->dev_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3022 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
| 3023 | } |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 3024 | mddev->dev_sectors = sectors; |
NeilBrown | 4b5c7ae | 2005-07-27 11:43:28 -0700 | [diff] [blame] | 3025 | mddev->resync_max_sectors = sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3026 | return 0; |
| 3027 | } |
| 3028 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 3029 | static int raid1_reshape(struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3030 | { |
| 3031 | /* We need to: |
| 3032 | * 1/ resize the r1bio_pool |
| 3033 | * 2/ resize conf->mirrors |
| 3034 | * |
| 3035 | * We allocate a new r1bio_pool if we can. |
| 3036 | * Then raise a device barrier and wait until all IO stops. |
| 3037 | * Then resize conf->mirrors and swap in the new r1bio pool. |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3038 | * |
| 3039 | * At the same time, we "pack" the devices so that all the missing |
| 3040 | * devices have the higher raid_disk numbers. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3041 | */ |
| 3042 | mempool_t *newpool, *oldpool; |
| 3043 | struct pool_info *newpoolinfo; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 3044 | struct raid1_info *newmirrors; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 3045 | struct r1conf *conf = mddev->private; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3046 | int cnt, raid_disks; |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 3047 | unsigned long flags; |
Dan Williams | b5470dc | 2008-06-27 21:44:04 -0700 | [diff] [blame] | 3048 | int d, d2, err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3049 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3050 | /* Cannot change chunk_size, layout, or level */ |
Andre Noll | 664e7c4 | 2009-06-18 08:45:27 +1000 | [diff] [blame] | 3051 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3052 | mddev->layout != mddev->new_layout || |
| 3053 | mddev->level != mddev->new_level) { |
Andre Noll | 664e7c4 | 2009-06-18 08:45:27 +1000 | [diff] [blame] | 3054 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3055 | mddev->new_layout = mddev->layout; |
| 3056 | mddev->new_level = mddev->level; |
| 3057 | return -EINVAL; |
| 3058 | } |
| 3059 | |
Goldwyn Rodrigues | 28c1b9f | 2015-10-22 16:01:25 +1100 | [diff] [blame] | 3060 | if (!mddev_is_clustered(mddev)) { |
| 3061 | err = md_allow_write(mddev); |
| 3062 | if (err) |
| 3063 | return err; |
| 3064 | } |
NeilBrown | 2a2275d | 2007-01-26 00:57:11 -0800 | [diff] [blame] | 3065 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3066 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
| 3067 | |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3068 | if (raid_disks < conf->raid_disks) { |
| 3069 | cnt=0; |
| 3070 | for (d= 0; d < conf->raid_disks; d++) |
| 3071 | if (conf->mirrors[d].rdev) |
| 3072 | cnt++; |
| 3073 | if (cnt > raid_disks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3074 | return -EBUSY; |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3075 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3076 | |
| 3077 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); |
| 3078 | if (!newpoolinfo) |
| 3079 | return -ENOMEM; |
| 3080 | newpoolinfo->mddev = mddev; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 3081 | newpoolinfo->raid_disks = raid_disks * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | |
| 3083 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
| 3084 | r1bio_pool_free, newpoolinfo); |
| 3085 | if (!newpool) { |
| 3086 | kfree(newpoolinfo); |
| 3087 | return -ENOMEM; |
| 3088 | } |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 3089 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 3090 | GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3091 | if (!newmirrors) { |
| 3092 | kfree(newpoolinfo); |
| 3093 | mempool_destroy(newpool); |
| 3094 | return -ENOMEM; |
| 3095 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3096 | |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 3097 | freeze_array(conf, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3098 | |
| 3099 | /* ok, everything is stopped */ |
| 3100 | oldpool = conf->r1bio_pool; |
| 3101 | conf->r1bio_pool = newpool; |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3102 | |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3103 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 3104 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3105 | if (rdev && rdev->raid_disk != d2) { |
Namhyung Kim | 36fad85 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 3106 | sysfs_unlink_rdev(mddev, rdev); |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3107 | rdev->raid_disk = d2; |
Namhyung Kim | 36fad85 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 3108 | sysfs_unlink_rdev(mddev, rdev); |
| 3109 | if (sysfs_link_rdev(mddev, rdev)) |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3110 | printk(KERN_WARNING |
Namhyung Kim | 36fad85 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 3111 | "md/raid1:%s: cannot register rd%d\n", |
| 3112 | mdname(mddev), rdev->raid_disk); |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3113 | } |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3114 | if (rdev) |
| 3115 | newmirrors[d2++].rdev = rdev; |
| 3116 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3117 | kfree(conf->mirrors); |
| 3118 | conf->mirrors = newmirrors; |
| 3119 | kfree(conf->poolinfo); |
| 3120 | conf->poolinfo = newpoolinfo; |
| 3121 | |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 3122 | spin_lock_irqsave(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3123 | mddev->degraded += (raid_disks - conf->raid_disks); |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 3124 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3125 | conf->raid_disks = mddev->raid_disks = raid_disks; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3126 | mddev->delta_disks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3127 | |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 3128 | unfreeze_array(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3129 | |
NeilBrown | 985ca97 | 2015-07-06 12:26:57 +1000 | [diff] [blame] | 3130 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3131 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
| 3132 | md_wakeup_thread(mddev->thread); |
| 3133 | |
| 3134 | mempool_destroy(oldpool); |
| 3135 | return 0; |
| 3136 | } |
| 3137 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 3138 | static void raid1_quiesce(struct mddev *mddev, int state) |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3139 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 3140 | struct r1conf *conf = mddev->private; |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3141 | |
| 3142 | switch(state) { |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3143 | case 2: /* wake for suspend */ |
| 3144 | wake_up(&conf->wait_barrier); |
| 3145 | break; |
NeilBrown | 9e6603d | 2005-09-09 16:23:48 -0700 | [diff] [blame] | 3146 | case 1: |
majianpeng | 07169fd | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 3147 | freeze_array(conf, 0); |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3148 | break; |
NeilBrown | 9e6603d | 2005-09-09 16:23:48 -0700 | [diff] [blame] | 3149 | case 0: |
majianpeng | 07169fd | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 3150 | unfreeze_array(conf); |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3151 | break; |
| 3152 | } |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3153 | } |
| 3154 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 3155 | static void *raid1_takeover(struct mddev *mddev) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3156 | { |
| 3157 | /* raid1 can take over: |
| 3158 | * raid5 with 2 devices, any layout or chunk size |
| 3159 | */ |
| 3160 | if (mddev->level == 5 && mddev->raid_disks == 2) { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 3161 | struct r1conf *conf; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3162 | mddev->new_level = 1; |
| 3163 | mddev->new_layout = 0; |
| 3164 | mddev->new_chunk_sectors = 0; |
| 3165 | conf = setup_conf(mddev); |
| 3166 | if (!IS_ERR(conf)) |
majianpeng | 07169fd | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 3167 | /* Array must appear to be quiesced */ |
| 3168 | conf->array_frozen = 1; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3169 | return conf; |
| 3170 | } |
| 3171 | return ERR_PTR(-EINVAL); |
| 3172 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3173 | |
NeilBrown | 84fc4b5 | 2011-10-11 16:49:58 +1100 | [diff] [blame] | 3174 | static struct md_personality raid1_personality = |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3175 | { |
| 3176 | .name = "raid1", |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 3177 | .level = 1, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3178 | .owner = THIS_MODULE, |
Shaohua Li | 849674e | 2016-01-20 13:52:20 -0800 | [diff] [blame] | 3179 | .make_request = raid1_make_request, |
| 3180 | .run = raid1_run, |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 3181 | .free = raid1_free, |
Shaohua Li | 849674e | 2016-01-20 13:52:20 -0800 | [diff] [blame] | 3182 | .status = raid1_status, |
| 3183 | .error_handler = raid1_error, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3184 | .hot_add_disk = raid1_add_disk, |
| 3185 | .hot_remove_disk= raid1_remove_disk, |
| 3186 | .spare_active = raid1_spare_active, |
Shaohua Li | 849674e | 2016-01-20 13:52:20 -0800 | [diff] [blame] | 3187 | .sync_request = raid1_sync_request, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3188 | .resize = raid1_resize, |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 3189 | .size = raid1_size, |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3190 | .check_reshape = raid1_reshape, |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3191 | .quiesce = raid1_quiesce, |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3192 | .takeover = raid1_takeover, |
NeilBrown | 5c675f8 | 2014-12-15 12:56:56 +1100 | [diff] [blame] | 3193 | .congested = raid1_congested, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3194 | }; |
| 3195 | |
| 3196 | static int __init raid_init(void) |
| 3197 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 3198 | return register_md_personality(&raid1_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3199 | } |
| 3200 | |
| 3201 | static void raid_exit(void) |
| 3202 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 3203 | unregister_md_personality(&raid1_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3204 | } |
| 3205 | |
| 3206 | module_init(raid_init); |
| 3207 | module_exit(raid_exit); |
| 3208 | MODULE_LICENSE("GPL"); |
NeilBrown | 0efb9e6 | 2009-12-14 12:49:58 +1100 | [diff] [blame] | 3209 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3210 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
NeilBrown | d9d166c | 2006-01-06 00:20:51 -0800 | [diff] [blame] | 3211 | MODULE_ALIAS("md-raid1"); |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 3212 | MODULE_ALIAS("md-level-1"); |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 3213 | |
| 3214 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |