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) |
| 93 | #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 95 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | { |
| 97 | struct pool_info *pi = data; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 98 | struct r1bio *r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | struct bio *bio; |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 100 | int need_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | int i, j; |
| 102 | |
| 103 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 104 | if (!r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | |
| 107 | /* |
| 108 | * Allocate bios : 1 for reading, n-1 for writing |
| 109 | */ |
| 110 | for (j = pi->raid_disks ; j-- ; ) { |
NeilBrown | 6746557 | 2010-10-26 17:33:54 +1100 | [diff] [blame] | 111 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | if (!bio) |
| 113 | goto out_free_bio; |
| 114 | r1_bio->bios[j] = bio; |
| 115 | } |
| 116 | /* |
| 117 | * Allocate RESYNC_PAGES data pages and attach them to |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 118 | * the first bio. |
| 119 | * If this is a user-requested check/repair, allocate |
| 120 | * RESYNC_PAGES for each bio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | */ |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 122 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 123 | need_pages = pi->raid_disks; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 124 | else |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 125 | need_pages = 1; |
| 126 | for (j = 0; j < need_pages; j++) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 127 | bio = r1_bio->bios[j]; |
Kent Overstreet | a078760 | 2012-09-10 14:03:28 -0700 | [diff] [blame] | 128 | bio->bi_vcnt = RESYNC_PAGES; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 129 | |
Kent Overstreet | a078760 | 2012-09-10 14:03:28 -0700 | [diff] [blame] | 130 | if (bio_alloc_pages(bio, gfp_flags)) |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 131 | goto out_free_pages; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 132 | } |
| 133 | /* If not user-requests, copy the page pointers to all bios */ |
| 134 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { |
| 135 | for (i=0; i<RESYNC_PAGES ; i++) |
| 136 | for (j=1; j<pi->raid_disks; j++) |
| 137 | r1_bio->bios[j]->bi_io_vec[i].bv_page = |
| 138 | r1_bio->bios[0]->bi_io_vec[i].bv_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 139 | } |
| 140 | |
| 141 | r1_bio->master_bio = NULL; |
| 142 | |
| 143 | return r1_bio; |
| 144 | |
NeilBrown | da1aab3 | 2014-04-09 12:25:43 +1000 | [diff] [blame] | 145 | out_free_pages: |
| 146 | while (--j >= 0) { |
| 147 | struct bio_vec *bv; |
| 148 | |
| 149 | bio_for_each_segment_all(bv, r1_bio->bios[j], i) |
| 150 | __free_page(bv->bv_page); |
| 151 | } |
| 152 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | out_free_bio: |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 154 | while (++j < pi->raid_disks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | bio_put(r1_bio->bios[j]); |
| 156 | r1bio_pool_free(r1_bio, data); |
| 157 | return NULL; |
| 158 | } |
| 159 | |
| 160 | static void r1buf_pool_free(void *__r1_bio, void *data) |
| 161 | { |
| 162 | struct pool_info *pi = data; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 163 | int i,j; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 164 | struct r1bio *r1bio = __r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 166 | for (i = 0; i < RESYNC_PAGES; i++) |
| 167 | for (j = pi->raid_disks; j-- ;) { |
| 168 | if (j == 0 || |
| 169 | r1bio->bios[j]->bi_io_vec[i].bv_page != |
| 170 | r1bio->bios[0]->bi_io_vec[i].bv_page) |
NeilBrown | 1345b1d | 2006-01-06 00:20:40 -0800 | [diff] [blame] | 171 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 172 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | for (i=0 ; i < pi->raid_disks; i++) |
| 174 | bio_put(r1bio->bios[i]); |
| 175 | |
| 176 | r1bio_pool_free(r1bio, data); |
| 177 | } |
| 178 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 179 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | { |
| 181 | int i; |
| 182 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 183 | for (i = 0; i < conf->raid_disks * 2; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | struct bio **bio = r1_bio->bios + i; |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 185 | if (!BIO_SPECIAL(*bio)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | bio_put(*bio); |
| 187 | *bio = NULL; |
| 188 | } |
| 189 | } |
| 190 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 191 | static void free_r1bio(struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 193 | struct r1conf *conf = r1_bio->mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | put_all_bios(conf, r1_bio); |
| 196 | mempool_free(r1_bio, conf->r1bio_pool); |
| 197 | } |
| 198 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 199 | static void put_buf(struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 201 | struct r1conf *conf = r1_bio->mddev->private; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 202 | int i; |
| 203 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 204 | for (i = 0; i < conf->raid_disks * 2; i++) { |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 205 | struct bio *bio = r1_bio->bios[i]; |
| 206 | if (bio->bi_end_io) |
| 207 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); |
| 208 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | |
| 210 | mempool_free(r1_bio, conf->r1buf_pool); |
| 211 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 212 | lower_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | } |
| 214 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 215 | static void reschedule_retry(struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | { |
| 217 | unsigned long flags; |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 218 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 219 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | |
| 221 | spin_lock_irqsave(&conf->device_lock, flags); |
| 222 | list_add(&r1_bio->retry_list, &conf->retry_list); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 223 | conf->nr_queued ++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 225 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 226 | wake_up(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | md_wakeup_thread(mddev->thread); |
| 228 | } |
| 229 | |
| 230 | /* |
| 231 | * raid_end_bio_io() is called when we have finished servicing a mirrored |
| 232 | * operation and are ready to return a success/failure code to the buffer |
| 233 | * cache layer. |
| 234 | */ |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 235 | static void call_bio_endio(struct r1bio *r1_bio) |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 236 | { |
| 237 | struct bio *bio = r1_bio->master_bio; |
| 238 | int done; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 239 | struct r1conf *conf = r1_bio->mddev->private; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 240 | sector_t start_next_window = r1_bio->start_next_window; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 241 | sector_t bi_sector = bio->bi_iter.bi_sector; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 242 | |
| 243 | if (bio->bi_phys_segments) { |
| 244 | unsigned long flags; |
| 245 | spin_lock_irqsave(&conf->device_lock, flags); |
| 246 | bio->bi_phys_segments--; |
| 247 | done = (bio->bi_phys_segments == 0); |
| 248 | spin_unlock_irqrestore(&conf->device_lock, flags); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 249 | /* |
| 250 | * make_request() might be waiting for |
| 251 | * bi_phys_segments to decrease |
| 252 | */ |
| 253 | wake_up(&conf->wait_barrier); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 254 | } else |
| 255 | done = 1; |
| 256 | |
| 257 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 258 | bio->bi_error = -EIO; |
| 259 | |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 260 | if (done) { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 261 | bio_endio(bio); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 262 | /* |
| 263 | * Wake up any possible resync thread that waits for the device |
| 264 | * to go idle. |
| 265 | */ |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 266 | allow_barrier(conf, start_next_window, bi_sector); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 267 | } |
| 268 | } |
| 269 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 270 | static void raid_end_bio_io(struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | { |
| 272 | struct bio *bio = r1_bio->master_bio; |
| 273 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 274 | /* if nobody has done the final endio yet, do it now */ |
| 275 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { |
NeilBrown | 36a4e1f | 2011-10-07 14:23:17 +1100 | [diff] [blame] | 276 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
| 277 | (bio_data_dir(bio) == WRITE) ? "write" : "read", |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 278 | (unsigned long long) bio->bi_iter.bi_sector, |
| 279 | (unsigned long long) bio_end_sector(bio) - 1); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 280 | |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 281 | call_bio_endio(r1_bio); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 282 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 283 | free_r1bio(r1_bio); |
| 284 | } |
| 285 | |
| 286 | /* |
| 287 | * Update disk head position estimator based on IRQ completion info. |
| 288 | */ |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 289 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 291 | struct r1conf *conf = r1_bio->mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | |
| 293 | conf->mirrors[disk].head_position = |
| 294 | r1_bio->sector + (r1_bio->sectors); |
| 295 | } |
| 296 | |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 297 | /* |
| 298 | * Find the disk number which triggered given bio |
| 299 | */ |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 300 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 301 | { |
| 302 | int mirror; |
NeilBrown | 3019463 | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 303 | struct r1conf *conf = r1_bio->mddev->private; |
| 304 | int raid_disks = conf->raid_disks; |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 305 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 306 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 307 | if (r1_bio->bios[mirror] == bio) |
| 308 | break; |
| 309 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 310 | BUG_ON(mirror == raid_disks * 2); |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 311 | update_head_pos(mirror, r1_bio); |
| 312 | |
| 313 | return mirror; |
| 314 | } |
| 315 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 316 | static void raid1_end_read_request(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 318 | int uptodate = !bio->bi_error; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 319 | struct r1bio *r1_bio = bio->bi_private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | int mirror; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 321 | struct r1conf *conf = r1_bio->mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | mirror = r1_bio->read_disk; |
| 324 | /* |
| 325 | * this branch is our 'one mirror IO has finished' event handler: |
| 326 | */ |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 327 | update_head_pos(mirror, r1_bio); |
| 328 | |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 329 | if (uptodate) |
| 330 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
| 331 | else { |
| 332 | /* If all other devices have failed, we want to return |
| 333 | * the error upwards rather than fail the last device. |
| 334 | * Here we redefine "uptodate" to mean "Don't want to retry" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | */ |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 336 | unsigned long flags; |
| 337 | spin_lock_irqsave(&conf->device_lock, flags); |
| 338 | if (r1_bio->mddev->degraded == conf->raid_disks || |
| 339 | (r1_bio->mddev->degraded == conf->raid_disks-1 && |
NeilBrown | 34cab6f | 2015-07-24 09:22:16 +1000 | [diff] [blame] | 340 | test_bit(In_sync, &conf->mirrors[mirror].rdev->flags))) |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 341 | uptodate = 1; |
| 342 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 343 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | |
NeilBrown | 7ad4d4a | 2012-10-11 13:44:30 +1100 | [diff] [blame] | 345 | if (uptodate) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | raid_end_bio_io(r1_bio); |
NeilBrown | 7ad4d4a | 2012-10-11 13:44:30 +1100 | [diff] [blame] | 347 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
| 348 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | /* |
| 350 | * oops, read error: |
| 351 | */ |
| 352 | char b[BDEVNAME_SIZE]; |
Christian Dietrich | 8bda470 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 353 | printk_ratelimited( |
| 354 | KERN_ERR "md/raid1:%s: %s: " |
| 355 | "rescheduling sector %llu\n", |
| 356 | mdname(conf->mddev), |
| 357 | bdevname(conf->mirrors[mirror].rdev->bdev, |
| 358 | b), |
| 359 | (unsigned long long)r1_bio->sector); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 360 | set_bit(R1BIO_ReadError, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | reschedule_retry(r1_bio); |
NeilBrown | 7ad4d4a | 2012-10-11 13:44:30 +1100 | [diff] [blame] | 362 | /* don't drop the reference on read_disk yet */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 363 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | } |
| 365 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 366 | static void close_write(struct r1bio *r1_bio) |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 367 | { |
| 368 | /* it really is the end of this request */ |
| 369 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
| 370 | /* free extra copy of the data pages */ |
| 371 | int i = r1_bio->behind_page_count; |
| 372 | while (i--) |
| 373 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); |
| 374 | kfree(r1_bio->behind_bvecs); |
| 375 | r1_bio->behind_bvecs = NULL; |
| 376 | } |
| 377 | /* clear the bitmap if all writes complete successfully */ |
| 378 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
| 379 | r1_bio->sectors, |
| 380 | !test_bit(R1BIO_Degraded, &r1_bio->state), |
| 381 | test_bit(R1BIO_BehindIO, &r1_bio->state)); |
| 382 | md_write_end(r1_bio->mddev); |
| 383 | } |
| 384 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 385 | static void r1_bio_write_done(struct r1bio *r1_bio) |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 386 | { |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 387 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
| 388 | return; |
| 389 | |
| 390 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 391 | reschedule_retry(r1_bio); |
| 392 | else { |
| 393 | close_write(r1_bio); |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 394 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
| 395 | reschedule_retry(r1_bio); |
| 396 | else |
| 397 | raid_end_bio_io(r1_bio); |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 398 | } |
| 399 | } |
| 400 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 401 | static void raid1_end_write_request(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | { |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 403 | struct r1bio *r1_bio = bio->bi_private; |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 404 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 405 | struct r1conf *conf = r1_bio->mddev->private; |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 406 | struct bio *to_put = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 408 | mirror = find_bio_disk(r1_bio, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 410 | /* |
| 411 | * 'one mirror IO has finished' event handler: |
| 412 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 413 | if (bio->bi_error) { |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 414 | set_bit(WriteErrorSeen, |
| 415 | &conf->mirrors[mirror].rdev->flags); |
NeilBrown | 19d6716 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 416 | if (!test_and_set_bit(WantReplacement, |
| 417 | &conf->mirrors[mirror].rdev->flags)) |
| 418 | set_bit(MD_RECOVERY_NEEDED, & |
| 419 | conf->mddev->recovery); |
| 420 | |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 421 | set_bit(R1BIO_WriteError, &r1_bio->state); |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 422 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | /* |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 424 | * Set R1BIO_Uptodate in our master bio, so that we |
| 425 | * will return a good error code for to the higher |
| 426 | * levels even if IO on some other mirrored buffer |
| 427 | * fails. |
| 428 | * |
| 429 | * The 'master' represents the composite IO operation |
| 430 | * to user-side. So if something waits for IO, then it |
| 431 | * will wait for the 'master' bio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | */ |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 433 | sector_t first_bad; |
| 434 | int bad_sectors; |
| 435 | |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 436 | r1_bio->bios[mirror] = NULL; |
| 437 | to_put = bio; |
Alex Lyakas | 3056e3a | 2013-06-04 20:42:21 +0300 | [diff] [blame] | 438 | /* |
| 439 | * Do not set R1BIO_Uptodate if the current device is |
| 440 | * rebuilding or Faulty. This is because we cannot use |
| 441 | * such device for properly reading the data back (we could |
| 442 | * potentially use it, if the current write would have felt |
| 443 | * before rdev->recovery_offset, but for simplicity we don't |
| 444 | * check this here. |
| 445 | */ |
| 446 | if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) && |
| 447 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)) |
| 448 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 449 | |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 450 | /* Maybe we can clear some bad blocks. */ |
| 451 | if (is_badblock(conf->mirrors[mirror].rdev, |
| 452 | r1_bio->sector, r1_bio->sectors, |
| 453 | &first_bad, &bad_sectors)) { |
| 454 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
| 455 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
| 456 | } |
| 457 | } |
| 458 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 459 | if (behind) { |
| 460 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) |
| 461 | atomic_dec(&r1_bio->behind_remaining); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 462 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 463 | /* |
| 464 | * In behind mode, we ACK the master bio once the I/O |
| 465 | * has safely reached all non-writemostly |
| 466 | * disks. Setting the Returned bit ensures that this |
| 467 | * gets done only once -- we don't ever want to return |
| 468 | * -EIO here, instead we'll wait |
| 469 | */ |
| 470 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && |
| 471 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
| 472 | /* Maybe we can return now */ |
| 473 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { |
| 474 | struct bio *mbio = r1_bio->master_bio; |
NeilBrown | 36a4e1f | 2011-10-07 14:23:17 +1100 | [diff] [blame] | 475 | pr_debug("raid1: behind end write sectors" |
| 476 | " %llu-%llu\n", |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 477 | (unsigned long long) mbio->bi_iter.bi_sector, |
| 478 | (unsigned long long) bio_end_sector(mbio) - 1); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 479 | call_bio_endio(r1_bio); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 480 | } |
| 481 | } |
| 482 | } |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 483 | if (r1_bio->bios[mirror] == NULL) |
| 484 | rdev_dec_pending(conf->mirrors[mirror].rdev, |
| 485 | conf->mddev); |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 486 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 487 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | * Let's see if all mirrored write operations have finished |
| 489 | * already. |
| 490 | */ |
NeilBrown | af6d7b7 | 2011-05-11 14:51:19 +1000 | [diff] [blame] | 491 | r1_bio_write_done(r1_bio); |
NeilBrown | c70810b | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 492 | |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 493 | if (to_put) |
| 494 | bio_put(to_put); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | } |
| 496 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 497 | /* |
| 498 | * This routine returns the disk from which the requested read should |
| 499 | * be done. There is a per-array 'next expected sequential IO' sector |
| 500 | * number - if this matches on the next IO then we use the last disk. |
| 501 | * There is also a per-disk 'last know head position' sector that is |
| 502 | * maintained from IRQ contexts, both the normal and the resync IO |
| 503 | * completion handlers update this position correctly. If there is no |
| 504 | * perfect sequential match then we pick the disk whose head is closest. |
| 505 | * |
| 506 | * If there are 2 mirrors in the same 2 devices, performance degrades |
| 507 | * because position is mirror, not device based. |
| 508 | * |
| 509 | * The rdev for the device selected will have nr_pending incremented. |
| 510 | */ |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 511 | 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] | 512 | { |
NeilBrown | af3a2cd | 2010-05-08 08:20:17 +1000 | [diff] [blame] | 513 | const sector_t this_sector = r1_bio->sector; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 514 | int sectors; |
| 515 | int best_good_sectors; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 516 | int best_disk, best_dist_disk, best_pending_disk; |
| 517 | int has_nonrot_disk; |
Shaohua Li | be4d328 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 518 | int disk; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 519 | sector_t best_dist; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 520 | unsigned int min_pending; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 521 | struct md_rdev *rdev; |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 522 | int choose_first; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 523 | int choose_next_idle; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | |
| 525 | rcu_read_lock(); |
| 526 | /* |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 527 | * Check if we can balance. We can balance on the whole |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 528 | * device if no resync is going on, or below the resync window. |
| 529 | * We take the first readable disk when above the resync window. |
| 530 | */ |
| 531 | retry: |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 532 | sectors = r1_bio->sectors; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 533 | best_disk = -1; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 534 | best_dist_disk = -1; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 535 | best_dist = MaxSector; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 536 | best_pending_disk = -1; |
| 537 | min_pending = UINT_MAX; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 538 | best_good_sectors = 0; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 539 | has_nonrot_disk = 0; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 540 | choose_next_idle = 0; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 541 | |
Goldwyn Rodrigues | 7d49ffc | 2014-08-12 10:13:19 -0500 | [diff] [blame] | 542 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
| 543 | (mddev_is_clustered(conf->mddev) && |
Goldwyn Rodrigues | 90382ed | 2015-06-24 09:30:32 -0500 | [diff] [blame] | 544 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
Goldwyn Rodrigues | 7d49ffc | 2014-08-12 10:13:19 -0500 | [diff] [blame] | 545 | this_sector + sectors))) |
| 546 | choose_first = 1; |
| 547 | else |
| 548 | choose_first = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 549 | |
Shaohua Li | be4d328 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 550 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 551 | sector_t dist; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 552 | sector_t first_bad; |
| 553 | int bad_sectors; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 554 | unsigned int pending; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 555 | bool nonrot; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 556 | |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 557 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
| 558 | if (r1_bio->bios[disk] == IO_BLOCKED |
| 559 | || rdev == NULL |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 560 | || test_bit(Faulty, &rdev->flags)) |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 561 | continue; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 562 | if (!test_bit(In_sync, &rdev->flags) && |
| 563 | rdev->recovery_offset < this_sector + sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 564 | continue; |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 565 | if (test_bit(WriteMostly, &rdev->flags)) { |
| 566 | /* Don't balance among write-mostly, just |
| 567 | * use the first as a last resort */ |
Tomáš Hodek | d1901ef | 2015-02-23 11:00:38 +1100 | [diff] [blame] | 568 | if (best_dist_disk < 0) { |
NeilBrown | 307729c | 2012-01-09 01:41:51 +1100 | [diff] [blame] | 569 | if (is_badblock(rdev, this_sector, sectors, |
| 570 | &first_bad, &bad_sectors)) { |
| 571 | if (first_bad < this_sector) |
| 572 | /* Cannot use this */ |
| 573 | continue; |
| 574 | best_good_sectors = first_bad - this_sector; |
| 575 | } else |
| 576 | best_good_sectors = sectors; |
Tomáš Hodek | d1901ef | 2015-02-23 11:00:38 +1100 | [diff] [blame] | 577 | best_dist_disk = disk; |
| 578 | best_pending_disk = disk; |
NeilBrown | 307729c | 2012-01-09 01:41:51 +1100 | [diff] [blame] | 579 | } |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 580 | continue; |
| 581 | } |
| 582 | /* This is a reasonable device to use. It might |
| 583 | * even be best. |
| 584 | */ |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 585 | if (is_badblock(rdev, this_sector, sectors, |
| 586 | &first_bad, &bad_sectors)) { |
| 587 | if (best_dist < MaxSector) |
| 588 | /* already have a better device */ |
| 589 | continue; |
| 590 | if (first_bad <= this_sector) { |
| 591 | /* cannot read here. If this is the 'primary' |
| 592 | * device, then we must not read beyond |
| 593 | * bad_sectors from another device.. |
| 594 | */ |
| 595 | bad_sectors -= (this_sector - first_bad); |
| 596 | if (choose_first && sectors > bad_sectors) |
| 597 | sectors = bad_sectors; |
| 598 | if (best_good_sectors > sectors) |
| 599 | best_good_sectors = sectors; |
| 600 | |
| 601 | } else { |
| 602 | sector_t good_sectors = first_bad - this_sector; |
| 603 | if (good_sectors > best_good_sectors) { |
| 604 | best_good_sectors = good_sectors; |
| 605 | best_disk = disk; |
| 606 | } |
| 607 | if (choose_first) |
| 608 | break; |
| 609 | } |
| 610 | continue; |
| 611 | } else |
| 612 | best_good_sectors = sectors; |
| 613 | |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 614 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
| 615 | has_nonrot_disk |= nonrot; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 616 | pending = atomic_read(&rdev->nr_pending); |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 617 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 618 | if (choose_first) { |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 619 | best_disk = disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | break; |
| 621 | } |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 622 | /* Don't change to another disk for sequential reads */ |
| 623 | if (conf->mirrors[disk].next_seq_sect == this_sector |
| 624 | || dist == 0) { |
| 625 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; |
| 626 | struct raid1_info *mirror = &conf->mirrors[disk]; |
| 627 | |
| 628 | best_disk = disk; |
| 629 | /* |
| 630 | * If buffered sequential IO size exceeds optimal |
| 631 | * iosize, check if there is idle disk. If yes, choose |
| 632 | * the idle disk. read_balance could already choose an |
| 633 | * idle disk before noticing it's a sequential IO in |
| 634 | * this disk. This doesn't matter because this disk |
| 635 | * will idle, next time it will be utilized after the |
| 636 | * first disk has IO size exceeds optimal iosize. In |
| 637 | * this way, iosize of the first disk will be optimal |
| 638 | * iosize at least. iosize of the second disk might be |
| 639 | * small, but not a big deal since when the second disk |
| 640 | * starts IO, the first disk is likely still busy. |
| 641 | */ |
| 642 | if (nonrot && opt_iosize > 0 && |
| 643 | mirror->seq_start != MaxSector && |
| 644 | mirror->next_seq_sect > opt_iosize && |
| 645 | mirror->next_seq_sect - opt_iosize >= |
| 646 | mirror->seq_start) { |
| 647 | choose_next_idle = 1; |
| 648 | continue; |
| 649 | } |
| 650 | break; |
| 651 | } |
| 652 | /* If device is idle, use it */ |
| 653 | if (pending == 0) { |
| 654 | best_disk = disk; |
| 655 | break; |
| 656 | } |
| 657 | |
| 658 | if (choose_next_idle) |
| 659 | continue; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 660 | |
| 661 | if (min_pending > pending) { |
| 662 | min_pending = pending; |
| 663 | best_pending_disk = disk; |
| 664 | } |
| 665 | |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 666 | if (dist < best_dist) { |
| 667 | best_dist = dist; |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 668 | best_dist_disk = disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | } |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 670 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | |
Shaohua Li | 9dedf60 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 672 | /* |
| 673 | * If all disks are rotational, choose the closest disk. If any disk is |
| 674 | * non-rotational, choose the disk with less pending request even the |
| 675 | * disk is rotational, which might/might not be optimal for raids with |
| 676 | * mixed ratation/non-rotational disks depending on workload. |
| 677 | */ |
| 678 | if (best_disk == -1) { |
| 679 | if (has_nonrot_disk) |
| 680 | best_disk = best_pending_disk; |
| 681 | else |
| 682 | best_disk = best_dist_disk; |
| 683 | } |
| 684 | |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 685 | if (best_disk >= 0) { |
| 686 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 687 | if (!rdev) |
| 688 | goto retry; |
| 689 | atomic_inc(&rdev->nr_pending); |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 690 | if (test_bit(Faulty, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 691 | /* cannot risk returning a device that failed |
| 692 | * before we inc'ed nr_pending |
| 693 | */ |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 694 | rdev_dec_pending(rdev, conf->mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | goto retry; |
| 696 | } |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 697 | sectors = best_good_sectors; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 698 | |
| 699 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) |
| 700 | conf->mirrors[best_disk].seq_start = this_sector; |
| 701 | |
Shaohua Li | be4d328 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 702 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 703 | } |
| 704 | rcu_read_unlock(); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 705 | *max_sectors = sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | |
NeilBrown | 7607305 | 2011-05-11 14:34:56 +1000 | [diff] [blame] | 707 | return best_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 708 | } |
| 709 | |
NeilBrown | 5c675f8 | 2014-12-15 12:56:56 +1100 | [diff] [blame] | 710 | static int raid1_congested(struct mddev *mddev, int bits) |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 711 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 712 | struct r1conf *conf = mddev->private; |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 713 | int i, ret = 0; |
| 714 | |
Tejun Heo | 4452226 | 2015-05-22 17:13:26 -0400 | [diff] [blame] | 715 | if ((bits & (1 << WB_async_congested)) && |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 716 | conf->pending_count >= max_queued_requests) |
| 717 | return 1; |
| 718 | |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 719 | rcu_read_lock(); |
NeilBrown | f53e29f | 2012-02-13 14:24:05 +1100 | [diff] [blame] | 720 | for (i = 0; i < conf->raid_disks * 2; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 721 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 722 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 723 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 724 | |
Jonathan Brassow | 1ed7242 | 2011-06-07 17:50:35 -0500 | [diff] [blame] | 725 | BUG_ON(!q); |
| 726 | |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 727 | /* Note the '|| 1' - when read_balance prefers |
| 728 | * non-congested targets, it can be removed |
| 729 | */ |
Tejun Heo | 4452226 | 2015-05-22 17:13:26 -0400 | [diff] [blame] | 730 | if ((bits & (1 << WB_async_congested)) || 1) |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 731 | ret |= bdi_congested(&q->backing_dev_info, bits); |
| 732 | else |
| 733 | ret &= bdi_congested(&q->backing_dev_info, bits); |
| 734 | } |
| 735 | } |
| 736 | rcu_read_unlock(); |
| 737 | return ret; |
| 738 | } |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 739 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 740 | static void flush_pending_writes(struct r1conf *conf) |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 741 | { |
| 742 | /* Any writes that have been queued but are awaiting |
| 743 | * bitmap updates get flushed here. |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 744 | */ |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 745 | spin_lock_irq(&conf->device_lock); |
| 746 | |
| 747 | if (conf->pending_bio_list.head) { |
| 748 | struct bio *bio; |
| 749 | bio = bio_list_get(&conf->pending_bio_list); |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 750 | conf->pending_count = 0; |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 751 | spin_unlock_irq(&conf->device_lock); |
| 752 | /* flush any pending bitmap writes to |
| 753 | * disk before proceeding w/ I/O */ |
| 754 | bitmap_unplug(conf->mddev->bitmap); |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 755 | wake_up(&conf->wait_barrier); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 756 | |
| 757 | while (bio) { /* submit pending writes */ |
| 758 | struct bio *next = bio->bi_next; |
| 759 | bio->bi_next = NULL; |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 760 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
| 761 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) |
| 762 | /* Just ignore it */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 763 | bio_endio(bio); |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 764 | else |
| 765 | generic_make_request(bio); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 766 | bio = next; |
| 767 | } |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 768 | } else |
| 769 | spin_unlock_irq(&conf->device_lock); |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 770 | } |
| 771 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 772 | /* Barriers.... |
| 773 | * Sometimes we need to suspend IO while we do something else, |
| 774 | * either some resync/recovery, or reconfigure the array. |
| 775 | * To do this we raise a 'barrier'. |
| 776 | * The 'barrier' is a counter that can be raised multiple times |
| 777 | * to count how many activities are happening which preclude |
| 778 | * normal IO. |
| 779 | * We can only raise the barrier if there is no pending IO. |
| 780 | * i.e. if nr_pending == 0. |
| 781 | * We choose only to raise the barrier if no-one is waiting for the |
| 782 | * barrier to go down. This means that as soon as an IO request |
| 783 | * is ready, no other operations which require a barrier will start |
| 784 | * until the IO request has had a chance. |
| 785 | * |
| 786 | * So: regular IO calls 'wait_barrier'. When that returns there |
| 787 | * is no backgroup IO happening, It must arrange to call |
| 788 | * allow_barrier when it has finished its IO. |
| 789 | * backgroup IO calls must call raise_barrier. Once that returns |
| 790 | * there is no normal IO happeing. It must arrange to call |
| 791 | * lower_barrier when the particular background IO completes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | */ |
NeilBrown | c2fd4c9 | 2014-09-10 16:01:24 +1000 | [diff] [blame] | 793 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 794 | { |
| 795 | spin_lock_irq(&conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 796 | |
| 797 | /* Wait until no block IO is waiting */ |
| 798 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 799 | conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 800 | |
| 801 | /* block any new IO from starting */ |
| 802 | conf->barrier++; |
NeilBrown | c2fd4c9 | 2014-09-10 16:01:24 +1000 | [diff] [blame] | 803 | conf->next_resync = sector_nr; |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 804 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 805 | /* For these conditions we must wait: |
| 806 | * A: while the array is in frozen state |
| 807 | * B: while barrier >= RESYNC_DEPTH, meaning resync reach |
| 808 | * the max count which allowed. |
| 809 | * C: next_resync + RESYNC_SECTORS > start_next_window, meaning |
| 810 | * next resync will reach to the window which normal bios are |
| 811 | * handling. |
NeilBrown | 2f73d3c | 2014-09-10 15:01:49 +1000 | [diff] [blame] | 812 | * D: while there are any active requests in the current window. |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 813 | */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 814 | wait_event_lock_irq(conf->wait_barrier, |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 815 | !conf->array_frozen && |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 816 | conf->barrier < RESYNC_DEPTH && |
NeilBrown | 2f73d3c | 2014-09-10 15:01:49 +1000 | [diff] [blame] | 817 | conf->current_window_requests == 0 && |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 818 | (conf->start_next_window >= |
| 819 | conf->next_resync + RESYNC_SECTORS), |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 820 | conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 821 | |
NeilBrown | 34e97f1 | 2014-09-16 12:14:14 +1000 | [diff] [blame] | 822 | conf->nr_pending++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 823 | spin_unlock_irq(&conf->resync_lock); |
| 824 | } |
| 825 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 826 | static void lower_barrier(struct r1conf *conf) |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 827 | { |
| 828 | unsigned long flags; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 829 | BUG_ON(conf->barrier <= 0); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 830 | spin_lock_irqsave(&conf->resync_lock, flags); |
| 831 | conf->barrier--; |
NeilBrown | 34e97f1 | 2014-09-16 12:14:14 +1000 | [diff] [blame] | 832 | conf->nr_pending--; |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 833 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
| 834 | wake_up(&conf->wait_barrier); |
| 835 | } |
| 836 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 837 | static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 838 | { |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 839 | bool wait = false; |
| 840 | |
| 841 | if (conf->array_frozen || !bio) |
| 842 | wait = true; |
| 843 | else if (conf->barrier && bio_data_dir(bio) == WRITE) { |
NeilBrown | 2355496 | 2014-09-10 15:56:57 +1000 | [diff] [blame] | 844 | if ((conf->mddev->curr_resync_completed |
| 845 | >= bio_end_sector(bio)) || |
| 846 | (conf->next_resync + NEXT_NORMALIO_DISTANCE |
| 847 | <= bio->bi_iter.bi_sector)) |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 848 | wait = false; |
| 849 | else |
| 850 | wait = true; |
| 851 | } |
| 852 | |
| 853 | return wait; |
| 854 | } |
| 855 | |
| 856 | static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) |
| 857 | { |
| 858 | sector_t sector = 0; |
| 859 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 860 | spin_lock_irq(&conf->resync_lock); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 861 | if (need_to_wait_for_sync(conf, bio)) { |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 862 | conf->nr_waiting++; |
NeilBrown | d6b42dc | 2012-03-19 12:46:38 +1100 | [diff] [blame] | 863 | /* Wait for the barrier to drop. |
| 864 | * However if there are already pending |
| 865 | * requests (preventing the barrier from |
| 866 | * rising completely), and the |
NeilBrown | 5965b64 | 2014-09-04 15:51:44 +1000 | [diff] [blame] | 867 | * per-process bio queue isn't empty, |
NeilBrown | d6b42dc | 2012-03-19 12:46:38 +1100 | [diff] [blame] | 868 | * then don't wait, as we need to empty |
NeilBrown | 5965b64 | 2014-09-04 15:51:44 +1000 | [diff] [blame] | 869 | * that queue to allow conf->start_next_window |
| 870 | * to increase. |
NeilBrown | d6b42dc | 2012-03-19 12:46:38 +1100 | [diff] [blame] | 871 | */ |
| 872 | wait_event_lock_irq(conf->wait_barrier, |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 873 | !conf->array_frozen && |
| 874 | (!conf->barrier || |
NeilBrown | 5965b64 | 2014-09-04 15:51:44 +1000 | [diff] [blame] | 875 | ((conf->start_next_window < |
| 876 | conf->next_resync + RESYNC_SECTORS) && |
| 877 | current->bio_list && |
| 878 | !bio_list_empty(current->bio_list))), |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 879 | conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 880 | conf->nr_waiting--; |
| 881 | } |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 882 | |
| 883 | if (bio && bio_data_dir(bio) == WRITE) { |
NeilBrown | 2f73d3c | 2014-09-10 15:01:49 +1000 | [diff] [blame] | 884 | if (bio->bi_iter.bi_sector >= |
NeilBrown | 2355496 | 2014-09-10 15:56:57 +1000 | [diff] [blame] | 885 | conf->mddev->curr_resync_completed) { |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 886 | if (conf->start_next_window == MaxSector) |
| 887 | conf->start_next_window = |
| 888 | conf->next_resync + |
| 889 | NEXT_NORMALIO_DISTANCE; |
| 890 | |
| 891 | if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 892 | <= bio->bi_iter.bi_sector) |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 893 | conf->next_window_requests++; |
| 894 | else |
| 895 | conf->current_window_requests++; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 896 | sector = conf->start_next_window; |
NeilBrown | 41a336e | 2014-01-14 11:56:14 +1100 | [diff] [blame] | 897 | } |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 898 | } |
| 899 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 900 | conf->nr_pending++; |
| 901 | spin_unlock_irq(&conf->resync_lock); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 902 | return sector; |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 903 | } |
| 904 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 905 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
| 906 | sector_t bi_sector) |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 907 | { |
| 908 | unsigned long flags; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 909 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 910 | spin_lock_irqsave(&conf->resync_lock, flags); |
| 911 | conf->nr_pending--; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 912 | if (start_next_window) { |
| 913 | if (start_next_window == conf->start_next_window) { |
| 914 | if (conf->start_next_window + NEXT_NORMALIO_DISTANCE |
| 915 | <= bi_sector) |
| 916 | conf->next_window_requests--; |
| 917 | else |
| 918 | conf->current_window_requests--; |
| 919 | } else |
| 920 | conf->current_window_requests--; |
| 921 | |
| 922 | if (!conf->current_window_requests) { |
| 923 | if (conf->next_window_requests) { |
| 924 | conf->current_window_requests = |
| 925 | conf->next_window_requests; |
| 926 | conf->next_window_requests = 0; |
| 927 | conf->start_next_window += |
| 928 | NEXT_NORMALIO_DISTANCE; |
| 929 | } else |
| 930 | conf->start_next_window = MaxSector; |
| 931 | } |
| 932 | } |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 933 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
| 934 | wake_up(&conf->wait_barrier); |
| 935 | } |
| 936 | |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 937 | static void freeze_array(struct r1conf *conf, int extra) |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 938 | { |
| 939 | /* stop syncio and normal IO and wait for everything to |
| 940 | * go quite. |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 941 | * We wait until nr_pending match nr_queued+extra |
NeilBrown | 1c83053 | 2008-03-04 14:29:35 -0800 | [diff] [blame] | 942 | * This is called in the context of one normal IO request |
| 943 | * that has failed. Thus any sync request that might be pending |
| 944 | * will be blocked by nr_pending, and we need to wait for |
| 945 | * pending IO requests to complete or be queued for re-try. |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 946 | * Thus the number queued (nr_queued) plus this request (extra) |
NeilBrown | 1c83053 | 2008-03-04 14:29:35 -0800 | [diff] [blame] | 947 | * must match the number of pending IOs (nr_pending) before |
| 948 | * we continue. |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 949 | */ |
| 950 | spin_lock_irq(&conf->resync_lock); |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 951 | conf->array_frozen = 1; |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 952 | wait_event_lock_irq_cmd(conf->wait_barrier, |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 953 | conf->nr_pending == conf->nr_queued+extra, |
Lukas Czerner | eed8c02 | 2012-11-30 11:42:40 +0100 | [diff] [blame] | 954 | conf->resync_lock, |
| 955 | flush_pending_writes(conf)); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 956 | spin_unlock_irq(&conf->resync_lock); |
| 957 | } |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 958 | static void unfreeze_array(struct r1conf *conf) |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 959 | { |
| 960 | /* reverse the effect of the freeze */ |
| 961 | spin_lock_irq(&conf->resync_lock); |
majianpeng | b364e3d | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 962 | conf->array_frozen = 0; |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 963 | wake_up(&conf->wait_barrier); |
| 964 | spin_unlock_irq(&conf->resync_lock); |
| 965 | } |
| 966 | |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 967 | /* duplicate the data pages for behind I/O |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 968 | */ |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 969 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 970 | { |
| 971 | int i; |
| 972 | struct bio_vec *bvec; |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 973 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 974 | GFP_NOIO); |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 975 | if (unlikely(!bvecs)) |
NeilBrown | af6d7b7 | 2011-05-11 14:51:19 +1000 | [diff] [blame] | 976 | return; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 977 | |
Kent Overstreet | cb34e05 | 2012-09-05 15:22:02 -0700 | [diff] [blame] | 978 | bio_for_each_segment_all(bvec, bio, i) { |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 979 | bvecs[i] = *bvec; |
| 980 | bvecs[i].bv_page = alloc_page(GFP_NOIO); |
| 981 | if (unlikely(!bvecs[i].bv_page)) |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 982 | goto do_sync_io; |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 983 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
| 984 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); |
| 985 | kunmap(bvecs[i].bv_page); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 986 | kunmap(bvec->bv_page); |
| 987 | } |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 988 | r1_bio->behind_bvecs = bvecs; |
NeilBrown | af6d7b7 | 2011-05-11 14:51:19 +1000 | [diff] [blame] | 989 | r1_bio->behind_page_count = bio->bi_vcnt; |
| 990 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
| 991 | return; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 992 | |
| 993 | do_sync_io: |
NeilBrown | af6d7b7 | 2011-05-11 14:51:19 +1000 | [diff] [blame] | 994 | for (i = 0; i < bio->bi_vcnt; i++) |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 995 | if (bvecs[i].bv_page) |
| 996 | put_page(bvecs[i].bv_page); |
| 997 | kfree(bvecs); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 998 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
| 999 | bio->bi_iter.bi_size); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1000 | } |
| 1001 | |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1002 | struct raid1_plug_cb { |
| 1003 | struct blk_plug_cb cb; |
| 1004 | struct bio_list pending; |
| 1005 | int pending_cnt; |
| 1006 | }; |
| 1007 | |
| 1008 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) |
| 1009 | { |
| 1010 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, |
| 1011 | cb); |
| 1012 | struct mddev *mddev = plug->cb.data; |
| 1013 | struct r1conf *conf = mddev->private; |
| 1014 | struct bio *bio; |
| 1015 | |
NeilBrown | 874807a | 2012-11-27 12:14:40 +1100 | [diff] [blame] | 1016 | if (from_schedule || current->bio_list) { |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1017 | spin_lock_irq(&conf->device_lock); |
| 1018 | bio_list_merge(&conf->pending_bio_list, &plug->pending); |
| 1019 | conf->pending_count += plug->pending_cnt; |
| 1020 | spin_unlock_irq(&conf->device_lock); |
NeilBrown | ee0b024 | 2013-02-25 12:38:29 +1100 | [diff] [blame] | 1021 | wake_up(&conf->wait_barrier); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1022 | md_wakeup_thread(mddev->thread); |
| 1023 | kfree(plug); |
| 1024 | return; |
| 1025 | } |
| 1026 | |
| 1027 | /* we aren't scheduling, so we can do the write-out directly. */ |
| 1028 | bio = bio_list_get(&plug->pending); |
| 1029 | bitmap_unplug(mddev->bitmap); |
| 1030 | wake_up(&conf->wait_barrier); |
| 1031 | |
| 1032 | while (bio) { /* submit pending writes */ |
| 1033 | struct bio *next = bio->bi_next; |
| 1034 | bio->bi_next = NULL; |
Shaohua Li | 32f9f57 | 2013-04-28 18:26:38 +0800 | [diff] [blame] | 1035 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
| 1036 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) |
| 1037 | /* Just ignore it */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1038 | bio_endio(bio); |
Shaohua Li | 32f9f57 | 2013-04-28 18:26:38 +0800 | [diff] [blame] | 1039 | else |
| 1040 | generic_make_request(bio); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1041 | bio = next; |
| 1042 | } |
| 1043 | kfree(plug); |
| 1044 | } |
| 1045 | |
Linus Torvalds | b4fdcb0 | 2011-11-04 17:06:58 -0700 | [diff] [blame] | 1046 | static void make_request(struct mddev *mddev, struct bio * bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1047 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1048 | struct r1conf *conf = mddev->private; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 1049 | struct raid1_info *mirror; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1050 | struct r1bio *r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1051 | struct bio *read_bio; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1052 | int i, disks; |
NeilBrown | 84255d1 | 2008-05-23 13:04:32 -0700 | [diff] [blame] | 1053 | struct bitmap *bitmap; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1054 | unsigned long flags; |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1055 | const int rw = bio_data_dir(bio); |
NeilBrown | 2c7d46e | 2010-08-18 16:16:05 +1000 | [diff] [blame] | 1056 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 1057 | const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 1058 | const unsigned long do_discard = (bio->bi_rw |
| 1059 | & (REQ_DISCARD | REQ_SECURE)); |
Joe Lawrence | c8dc9c6 | 2013-02-21 13:28:09 +1100 | [diff] [blame] | 1060 | const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME); |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1061 | struct md_rdev *blocked_rdev; |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1062 | struct blk_plug_cb *cb; |
| 1063 | struct raid1_plug_cb *plug = NULL; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1064 | int first_clone; |
| 1065 | int sectors_handled; |
| 1066 | int max_sectors; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1067 | sector_t start_next_window; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1068 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | /* |
| 1070 | * Register the new request and wait if the reconstruction |
| 1071 | * thread has put up a bar for new requests. |
| 1072 | * Continue immediately if no resync is active currently. |
| 1073 | */ |
NeilBrown | 62de608 | 2006-05-01 12:15:47 -0700 | [diff] [blame] | 1074 | |
NeilBrown | 3d310eb | 2005-06-21 17:17:26 -0700 | [diff] [blame] | 1075 | md_write_start(mddev, bio); /* wait on superblock update early */ |
| 1076 | |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1077 | if (bio_data_dir(bio) == WRITE && |
Goldwyn Rodrigues | 589a1c4 | 2014-06-07 02:39:37 -0500 | [diff] [blame] | 1078 | ((bio_end_sector(bio) > mddev->suspend_lo && |
| 1079 | bio->bi_iter.bi_sector < mddev->suspend_hi) || |
| 1080 | (mddev_is_clustered(mddev) && |
Goldwyn Rodrigues | 90382ed | 2015-06-24 09:30:32 -0500 | [diff] [blame] | 1081 | md_cluster_ops->area_resyncing(mddev, WRITE, |
| 1082 | bio->bi_iter.bi_sector, bio_end_sector(bio))))) { |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1083 | /* As the suspend_* range is controlled by |
| 1084 | * userspace, we want an interruptible |
| 1085 | * wait. |
| 1086 | */ |
| 1087 | DEFINE_WAIT(w); |
| 1088 | for (;;) { |
| 1089 | flush_signals(current); |
| 1090 | prepare_to_wait(&conf->wait_barrier, |
| 1091 | &w, TASK_INTERRUPTIBLE); |
Kent Overstreet | f73a1c7 | 2012-09-25 15:05:12 -0700 | [diff] [blame] | 1092 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
Goldwyn Rodrigues | 589a1c4 | 2014-06-07 02:39:37 -0500 | [diff] [blame] | 1093 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
| 1094 | (mddev_is_clustered(mddev) && |
Goldwyn Rodrigues | 90382ed | 2015-06-24 09:30:32 -0500 | [diff] [blame] | 1095 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
Goldwyn Rodrigues | 589a1c4 | 2014-06-07 02:39:37 -0500 | [diff] [blame] | 1096 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1097 | break; |
| 1098 | schedule(); |
| 1099 | } |
| 1100 | finish_wait(&conf->wait_barrier, &w); |
| 1101 | } |
NeilBrown | 62de608 | 2006-05-01 12:15:47 -0700 | [diff] [blame] | 1102 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1103 | start_next_window = wait_barrier(conf, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1104 | |
NeilBrown | 84255d1 | 2008-05-23 13:04:32 -0700 | [diff] [blame] | 1105 | bitmap = mddev->bitmap; |
| 1106 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1107 | /* |
| 1108 | * make_request() can abort the operation when READA is being |
| 1109 | * used and no empty request is available. |
| 1110 | * |
| 1111 | */ |
| 1112 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 1113 | |
| 1114 | r1_bio->master_bio = bio; |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 1115 | r1_bio->sectors = bio_sectors(bio); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1116 | r1_bio->state = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1117 | r1_bio->mddev = mddev; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1118 | r1_bio->sector = bio->bi_iter.bi_sector; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1120 | /* We might need to issue multiple reads to different |
| 1121 | * devices if there are bad blocks around, so we keep |
| 1122 | * track of the number of reads in bio->bi_phys_segments. |
| 1123 | * If this is 0, there is only one r1_bio and no locking |
| 1124 | * will be needed when requests complete. If it is |
| 1125 | * non-zero, then it is the number of not-completed requests. |
| 1126 | */ |
| 1127 | bio->bi_phys_segments = 0; |
Jens Axboe | b7c44ed | 2015-07-24 12:37:59 -0600 | [diff] [blame] | 1128 | bio_clear_flag(bio, BIO_SEG_VALID); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1129 | |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1130 | if (rw == READ) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1131 | /* |
| 1132 | * read balancing logic: |
| 1133 | */ |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1134 | int rdisk; |
| 1135 | |
| 1136 | read_again: |
| 1137 | rdisk = read_balance(conf, r1_bio, &max_sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1138 | |
| 1139 | if (rdisk < 0) { |
| 1140 | /* couldn't find anywhere to read from */ |
| 1141 | raid_end_bio_io(r1_bio); |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1142 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | } |
| 1144 | mirror = conf->mirrors + rdisk; |
| 1145 | |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 1146 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
| 1147 | bitmap) { |
| 1148 | /* Reading from a write-mostly device must |
| 1149 | * take care not to over-take any writes |
| 1150 | * that are 'behind' |
| 1151 | */ |
| 1152 | wait_event(bitmap->behind_wait, |
| 1153 | atomic_read(&bitmap->behind_writes) == 0); |
| 1154 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1155 | r1_bio->read_disk = rdisk; |
NeilBrown | f0cc9a0 | 2014-09-22 10:06:23 +1000 | [diff] [blame] | 1156 | r1_bio->start_next_window = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1157 | |
NeilBrown | a167f66 | 2010-10-26 18:31:13 +1100 | [diff] [blame] | 1158 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1159 | bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, |
Kent Overstreet | 6678d83 | 2013-08-07 11:14:32 -0700 | [diff] [blame] | 1160 | max_sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1161 | |
| 1162 | r1_bio->bios[rdisk] = read_bio; |
| 1163 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1164 | read_bio->bi_iter.bi_sector = r1_bio->sector + |
| 1165 | mirror->rdev->data_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1166 | read_bio->bi_bdev = mirror->rdev->bdev; |
| 1167 | read_bio->bi_end_io = raid1_end_read_request; |
Christoph Hellwig | 7b6d91d | 2010-08-07 18:20:39 +0200 | [diff] [blame] | 1168 | read_bio->bi_rw = READ | do_sync; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | read_bio->bi_private = r1_bio; |
| 1170 | |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1171 | if (max_sectors < r1_bio->sectors) { |
| 1172 | /* could not read all from this device, so we will |
| 1173 | * need another r1_bio. |
| 1174 | */ |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1175 | |
| 1176 | sectors_handled = (r1_bio->sector + max_sectors |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1177 | - bio->bi_iter.bi_sector); |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1178 | r1_bio->sectors = max_sectors; |
| 1179 | spin_lock_irq(&conf->device_lock); |
| 1180 | if (bio->bi_phys_segments == 0) |
| 1181 | bio->bi_phys_segments = 2; |
| 1182 | else |
| 1183 | bio->bi_phys_segments++; |
| 1184 | spin_unlock_irq(&conf->device_lock); |
| 1185 | /* Cannot call generic_make_request directly |
| 1186 | * as that will be queued in __make_request |
| 1187 | * and subsequent mempool_alloc might block waiting |
| 1188 | * for it. So hand bio over to raid1d. |
| 1189 | */ |
| 1190 | reschedule_retry(r1_bio); |
| 1191 | |
| 1192 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 1193 | |
| 1194 | r1_bio->master_bio = bio; |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 1195 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1196 | r1_bio->state = 0; |
| 1197 | r1_bio->mddev = mddev; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1198 | r1_bio->sector = bio->bi_iter.bi_sector + |
| 1199 | sectors_handled; |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1200 | goto read_again; |
| 1201 | } else |
| 1202 | generic_make_request(read_bio); |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1203 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1204 | } |
| 1205 | |
| 1206 | /* |
| 1207 | * WRITE: |
| 1208 | */ |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 1209 | if (conf->pending_count >= max_queued_requests) { |
| 1210 | md_wakeup_thread(mddev->thread); |
| 1211 | wait_event(conf->wait_barrier, |
| 1212 | conf->pending_count < max_queued_requests); |
| 1213 | } |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1214 | /* first select target devices under rcu_lock and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | * inc refcount on their rdev. Record them by setting |
| 1216 | * bios[x] to bio |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1217 | * If there are known/acknowledged bad blocks on any device on |
| 1218 | * which we have seen a write error, we want to avoid writing those |
| 1219 | * blocks. |
| 1220 | * This potentially requires several writes to write around |
| 1221 | * the bad blocks. Each set of writes gets it's own r1bio |
| 1222 | * with a set of bios attached. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1223 | */ |
NeilBrown | c3b328a | 2011-04-18 18:25:43 +1000 | [diff] [blame] | 1224 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1225 | disks = conf->raid_disks * 2; |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1226 | retry_write: |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1227 | r1_bio->start_next_window = start_next_window; |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1228 | blocked_rdev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | rcu_read_lock(); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1230 | max_sectors = r1_bio->sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1231 | for (i = 0; i < disks; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1232 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1233 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
| 1234 | atomic_inc(&rdev->nr_pending); |
| 1235 | blocked_rdev = rdev; |
| 1236 | break; |
| 1237 | } |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1238 | r1_bio->bios[i] = NULL; |
Kent Overstreet | 8ae1266 | 2015-04-27 23:48:34 -0700 | [diff] [blame] | 1239 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1240 | if (i < conf->raid_disks) |
| 1241 | set_bit(R1BIO_Degraded, &r1_bio->state); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1242 | continue; |
| 1243 | } |
| 1244 | |
| 1245 | atomic_inc(&rdev->nr_pending); |
| 1246 | if (test_bit(WriteErrorSeen, &rdev->flags)) { |
| 1247 | sector_t first_bad; |
| 1248 | int bad_sectors; |
| 1249 | int is_bad; |
| 1250 | |
| 1251 | is_bad = is_badblock(rdev, r1_bio->sector, |
| 1252 | max_sectors, |
| 1253 | &first_bad, &bad_sectors); |
| 1254 | if (is_bad < 0) { |
| 1255 | /* mustn't write here until the bad block is |
| 1256 | * acknowledged*/ |
| 1257 | set_bit(BlockedBadBlocks, &rdev->flags); |
| 1258 | blocked_rdev = rdev; |
| 1259 | break; |
NeilBrown | 964147d | 2010-05-18 15:27:13 +1000 | [diff] [blame] | 1260 | } |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1261 | if (is_bad && first_bad <= r1_bio->sector) { |
| 1262 | /* Cannot write here at all */ |
| 1263 | bad_sectors -= (r1_bio->sector - first_bad); |
| 1264 | if (bad_sectors < max_sectors) |
| 1265 | /* mustn't write more than bad_sectors |
| 1266 | * to other devices yet |
| 1267 | */ |
| 1268 | max_sectors = bad_sectors; |
| 1269 | rdev_dec_pending(rdev, mddev); |
| 1270 | /* We don't set R1BIO_Degraded as that |
| 1271 | * only applies if the disk is |
| 1272 | * missing, so it might be re-added, |
| 1273 | * and we want to know to recover this |
| 1274 | * chunk. |
| 1275 | * In this case the device is here, |
| 1276 | * and the fact that this chunk is not |
| 1277 | * in-sync is recorded in the bad |
| 1278 | * block log |
| 1279 | */ |
| 1280 | continue; |
| 1281 | } |
| 1282 | if (is_bad) { |
| 1283 | int good_sectors = first_bad - r1_bio->sector; |
| 1284 | if (good_sectors < max_sectors) |
| 1285 | max_sectors = good_sectors; |
| 1286 | } |
| 1287 | } |
| 1288 | r1_bio->bios[i] = bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1289 | } |
| 1290 | rcu_read_unlock(); |
| 1291 | |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1292 | if (unlikely(blocked_rdev)) { |
| 1293 | /* Wait for this device to become unblocked */ |
| 1294 | int j; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1295 | sector_t old = start_next_window; |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1296 | |
| 1297 | for (j = 0; j < i; j++) |
| 1298 | if (r1_bio->bios[j]) |
| 1299 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1300 | r1_bio->state = 0; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1301 | allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector); |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1302 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1303 | start_next_window = wait_barrier(conf, bio); |
| 1304 | /* |
| 1305 | * We must make sure the multi r1bios of bio have |
| 1306 | * the same value of bi_phys_segments |
| 1307 | */ |
| 1308 | if (bio->bi_phys_segments && old && |
| 1309 | old != start_next_window) |
| 1310 | /* Wait for the former r1bio(s) to complete */ |
| 1311 | wait_event(conf->wait_barrier, |
| 1312 | bio->bi_phys_segments == 1); |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 1313 | goto retry_write; |
| 1314 | } |
| 1315 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1316 | if (max_sectors < r1_bio->sectors) { |
| 1317 | /* We are splitting this write into multiple parts, so |
| 1318 | * we need to prepare for allocating another r1_bio. |
| 1319 | */ |
| 1320 | r1_bio->sectors = max_sectors; |
| 1321 | spin_lock_irq(&conf->device_lock); |
| 1322 | if (bio->bi_phys_segments == 0) |
| 1323 | bio->bi_phys_segments = 2; |
| 1324 | else |
| 1325 | bio->bi_phys_segments++; |
| 1326 | spin_unlock_irq(&conf->device_lock); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1327 | } |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1328 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1329 | |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 1330 | atomic_set(&r1_bio->remaining, 1); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1331 | atomic_set(&r1_bio->behind_remaining, 0); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1332 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1333 | first_clone = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1334 | for (i = 0; i < disks; i++) { |
| 1335 | struct bio *mbio; |
| 1336 | if (!r1_bio->bios[i]) |
| 1337 | continue; |
| 1338 | |
NeilBrown | a167f66 | 2010-10-26 18:31:13 +1100 | [diff] [blame] | 1339 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1340 | 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] | 1341 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1342 | if (first_clone) { |
| 1343 | /* do behind I/O ? |
| 1344 | * Not if there are too many, or cannot |
| 1345 | * allocate memory, or a reader on WriteMostly |
| 1346 | * is waiting for behind writes to flush */ |
| 1347 | if (bitmap && |
| 1348 | (atomic_read(&bitmap->behind_writes) |
| 1349 | < mddev->bitmap_info.max_write_behind) && |
| 1350 | !waitqueue_active(&bitmap->behind_wait)) |
| 1351 | alloc_behind_pages(mbio, r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1353 | bitmap_startwrite(bitmap, r1_bio->sector, |
| 1354 | r1_bio->sectors, |
| 1355 | test_bit(R1BIO_BehindIO, |
| 1356 | &r1_bio->state)); |
| 1357 | first_clone = 0; |
| 1358 | } |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 1359 | if (r1_bio->behind_bvecs) { |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1360 | struct bio_vec *bvec; |
| 1361 | int j; |
| 1362 | |
Kent Overstreet | cb34e05 | 2012-09-05 15:22:02 -0700 | [diff] [blame] | 1363 | /* |
| 1364 | * We trimmed the bio, so _all is legit |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1365 | */ |
Kent Overstreet | d74c6d5 | 2013-02-06 12:23:11 -0800 | [diff] [blame] | 1366 | bio_for_each_segment_all(bvec, mbio, j) |
NeilBrown | 2ca68f5 | 2011-07-28 11:32:10 +1000 | [diff] [blame] | 1367 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1368 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
| 1369 | atomic_inc(&r1_bio->behind_remaining); |
| 1370 | } |
| 1371 | |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1372 | r1_bio->bios[i] = mbio; |
| 1373 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1374 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1375 | conf->mirrors[i].rdev->data_offset); |
| 1376 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
| 1377 | mbio->bi_end_io = raid1_end_write_request; |
Joe Lawrence | c8dc9c6 | 2013-02-21 13:28:09 +1100 | [diff] [blame] | 1378 | mbio->bi_rw = |
| 1379 | WRITE | do_flush_fua | do_sync | do_discard | do_same; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1380 | mbio->bi_private = r1_bio; |
| 1381 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | atomic_inc(&r1_bio->remaining); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1383 | |
| 1384 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
| 1385 | if (cb) |
| 1386 | plug = container_of(cb, struct raid1_plug_cb, cb); |
| 1387 | else |
| 1388 | plug = NULL; |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 1389 | spin_lock_irqsave(&conf->device_lock, flags); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1390 | if (plug) { |
| 1391 | bio_list_add(&plug->pending, mbio); |
| 1392 | plug->pending_cnt++; |
| 1393 | } else { |
| 1394 | bio_list_add(&conf->pending_bio_list, mbio); |
| 1395 | conf->pending_count++; |
| 1396 | } |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 1397 | spin_unlock_irqrestore(&conf->device_lock, flags); |
NeilBrown | f54a9d0 | 2012-08-02 08:33:20 +1000 | [diff] [blame] | 1398 | if (!plug) |
NeilBrown | b357f04 | 2012-07-03 17:45:31 +1000 | [diff] [blame] | 1399 | md_wakeup_thread(mddev->thread); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1400 | } |
NeilBrown | 079fa16 | 2011-09-10 17:21:23 +1000 | [diff] [blame] | 1401 | /* Mustn't call r1_bio_write_done before this next test, |
| 1402 | * as it could result in the bio being freed. |
| 1403 | */ |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 1404 | if (sectors_handled < bio_sectors(bio)) { |
NeilBrown | 079fa16 | 2011-09-10 17:21:23 +1000 | [diff] [blame] | 1405 | r1_bio_write_done(r1_bio); |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1406 | /* We need another r1_bio. It has already been counted |
| 1407 | * in bio->bi_phys_segments |
| 1408 | */ |
| 1409 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 1410 | r1_bio->master_bio = bio; |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 1411 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1412 | r1_bio->state = 0; |
| 1413 | r1_bio->mddev = mddev; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1414 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
NeilBrown | 1f68f0c | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1415 | goto retry_write; |
| 1416 | } |
| 1417 | |
NeilBrown | 079fa16 | 2011-09-10 17:21:23 +1000 | [diff] [blame] | 1418 | r1_bio_write_done(r1_bio); |
| 1419 | |
| 1420 | /* In case raid1d snuck in to freeze_array */ |
| 1421 | wake_up(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1422 | } |
| 1423 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1424 | static void status(struct seq_file *seq, struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1425 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1426 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1427 | int i; |
| 1428 | |
| 1429 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
NeilBrown | 11ce99e | 2006-10-03 01:15:52 -0700 | [diff] [blame] | 1430 | conf->raid_disks - mddev->degraded); |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1431 | rcu_read_lock(); |
| 1432 | for (i = 0; i < conf->raid_disks; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1433 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1434 | seq_printf(seq, "%s", |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1435 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
| 1436 | } |
| 1437 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1438 | seq_printf(seq, "]"); |
| 1439 | } |
| 1440 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1441 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1442 | { |
| 1443 | char b[BDEVNAME_SIZE]; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1444 | struct r1conf *conf = mddev->private; |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1445 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1446 | |
| 1447 | /* |
| 1448 | * If it is not operational, then we have already marked it as dead |
| 1449 | * else if it is the last working disks, ignore the error, let the |
| 1450 | * next level up know. |
| 1451 | * else mark the drive as failed |
| 1452 | */ |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1453 | if (test_bit(In_sync, &rdev->flags) |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1454 | && (conf->raid_disks - mddev->degraded) == 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1455 | /* |
| 1456 | * Don't fail the drive, act as though we were just a |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1457 | * normal single drive. |
| 1458 | * However don't try a recovery from this drive as |
| 1459 | * it is very likely to fail. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1460 | */ |
NeilBrown | 5389042 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1461 | conf->recovery_disabled = mddev->recovery_disabled; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1462 | return; |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1463 | } |
NeilBrown | de393cd | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1464 | set_bit(Blocked, &rdev->flags); |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1465 | spin_lock_irqsave(&conf->device_lock, flags); |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 1466 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1467 | mddev->degraded++; |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 1468 | set_bit(Faulty, &rdev->flags); |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 1469 | } else |
| 1470 | set_bit(Faulty, &rdev->flags); |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1471 | spin_unlock_irqrestore(&conf->device_lock, flags); |
NeilBrown | 2446dba | 2014-07-31 10:16:29 +1000 | [diff] [blame] | 1472 | /* |
| 1473 | * if recovery is running, make sure it aborts. |
| 1474 | */ |
| 1475 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
NeilBrown | 850b2b4 | 2006-10-03 01:15:46 -0700 | [diff] [blame] | 1476 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 1477 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
Joe Perches | 067032b | 2011-01-14 09:14:33 +1100 | [diff] [blame] | 1478 | printk(KERN_ALERT |
| 1479 | "md/raid1:%s: Disk failure on %s, disabling device.\n" |
| 1480 | "md/raid1:%s: Operation continuing on %d devices.\n", |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1481 | mdname(mddev), bdevname(rdev->bdev, b), |
| 1482 | mdname(mddev), conf->raid_disks - mddev->degraded); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1483 | } |
| 1484 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1485 | static void print_conf(struct r1conf *conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | { |
| 1487 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1488 | |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1489 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1490 | if (!conf) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1491 | printk(KERN_DEBUG "(!conf)\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | return; |
| 1493 | } |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1494 | 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] | 1495 | conf->raid_disks); |
| 1496 | |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1497 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | for (i = 0; i < conf->raid_disks; i++) { |
| 1499 | char b[BDEVNAME_SIZE]; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1500 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1501 | if (rdev) |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1502 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1503 | i, !test_bit(In_sync, &rdev->flags), |
| 1504 | !test_bit(Faulty, &rdev->flags), |
| 1505 | bdevname(rdev->bdev,b)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | } |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1507 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | } |
| 1509 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1510 | static void close_sync(struct r1conf *conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1511 | { |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1512 | wait_barrier(conf, NULL); |
| 1513 | allow_barrier(conf, 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | |
| 1515 | mempool_destroy(conf->r1buf_pool); |
| 1516 | conf->r1buf_pool = NULL; |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1517 | |
NeilBrown | 669cc7b | 2014-09-04 16:30:38 +1000 | [diff] [blame] | 1518 | spin_lock_irq(&conf->resync_lock); |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 1519 | conf->next_resync = 0; |
| 1520 | conf->start_next_window = MaxSector; |
NeilBrown | 669cc7b | 2014-09-04 16:30:38 +1000 | [diff] [blame] | 1521 | conf->current_window_requests += |
| 1522 | conf->next_window_requests; |
| 1523 | conf->next_window_requests = 0; |
| 1524 | spin_unlock_irq(&conf->resync_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1525 | } |
| 1526 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1527 | static int raid1_spare_active(struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1528 | { |
| 1529 | int i; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1530 | struct r1conf *conf = mddev->private; |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1531 | int count = 0; |
| 1532 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1533 | |
| 1534 | /* |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 1535 | * Find all failed disks within the RAID1 configuration |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1536 | * and mark them readable. |
| 1537 | * Called under mddev lock, so rcu protection not needed. |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1538 | * device_lock used to avoid races with raid1_end_read_request |
| 1539 | * which expects 'In_sync' flags and ->degraded to be consistent. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1540 | */ |
NeilBrown | 423f04d | 2015-07-27 11:48:52 +1000 | [diff] [blame] | 1541 | spin_lock_irqsave(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1542 | for (i = 0; i < conf->raid_disks; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1543 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1544 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
| 1545 | if (repl |
Goldwyn Rodrigues | 1aee41f | 2014-10-29 18:51:31 -0500 | [diff] [blame] | 1546 | && !test_bit(Candidate, &repl->flags) |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1547 | && repl->recovery_offset == MaxSector |
| 1548 | && !test_bit(Faulty, &repl->flags) |
| 1549 | && !test_and_set_bit(In_sync, &repl->flags)) { |
| 1550 | /* replacement has just become active */ |
| 1551 | if (!rdev || |
| 1552 | !test_and_clear_bit(In_sync, &rdev->flags)) |
| 1553 | count++; |
| 1554 | if (rdev) { |
| 1555 | /* Replaced device not technically |
| 1556 | * faulty, but we need to be sure |
| 1557 | * it gets removed and never re-added |
| 1558 | */ |
| 1559 | set_bit(Faulty, &rdev->flags); |
| 1560 | sysfs_notify_dirent_safe( |
| 1561 | rdev->sysfs_state); |
| 1562 | } |
| 1563 | } |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1564 | if (rdev |
Lukasz Dorau | 61e4947 | 2013-10-24 12:55:17 +1100 | [diff] [blame] | 1565 | && rdev->recovery_offset == MaxSector |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1566 | && !test_bit(Faulty, &rdev->flags) |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 1567 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1568 | count++; |
Jonathan Brassow | 654e8b5 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1569 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1570 | } |
| 1571 | } |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1572 | mddev->degraded -= count; |
| 1573 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1574 | |
| 1575 | print_conf(conf); |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1576 | return count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1577 | } |
| 1578 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1579 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1580 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1581 | struct r1conf *conf = mddev->private; |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1582 | int err = -EEXIST; |
NeilBrown | 41158c7 | 2005-06-21 17:17:25 -0700 | [diff] [blame] | 1583 | int mirror = 0; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 1584 | struct raid1_info *p; |
Neil Brown | 6c2fce2 | 2008-06-28 08:31:31 +1000 | [diff] [blame] | 1585 | int first = 0; |
NeilBrown | 3019463 | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1586 | int last = conf->raid_disks - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1587 | |
NeilBrown | 5389042 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1588 | if (mddev->recovery_disabled == conf->recovery_disabled) |
| 1589 | return -EBUSY; |
| 1590 | |
Neil Brown | 6c2fce2 | 2008-06-28 08:31:31 +1000 | [diff] [blame] | 1591 | if (rdev->raid_disk >= 0) |
| 1592 | first = last = rdev->raid_disk; |
| 1593 | |
NeilBrown | 7ef449d | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1594 | for (mirror = first; mirror <= last; mirror++) { |
| 1595 | p = conf->mirrors+mirror; |
| 1596 | if (!p->rdev) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1597 | |
Jonathan Brassow | 9092c02 | 2013-05-02 14:19:24 -0500 | [diff] [blame] | 1598 | if (mddev->gendisk) |
| 1599 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 1600 | rdev->data_offset << 9); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1601 | |
| 1602 | p->head_position = 0; |
| 1603 | rdev->raid_disk = mirror; |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1604 | err = 0; |
NeilBrown | 6aea114a | 2005-11-28 13:44:13 -0800 | [diff] [blame] | 1605 | /* As all devices are equivalent, we don't need a full recovery |
| 1606 | * if this was recently any drive of the array |
| 1607 | */ |
| 1608 | if (rdev->saved_raid_disk < 0) |
NeilBrown | 41158c7 | 2005-06-21 17:17:25 -0700 | [diff] [blame] | 1609 | conf->fullsync = 1; |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 1610 | rcu_assign_pointer(p->rdev, rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1611 | break; |
| 1612 | } |
NeilBrown | 7ef449d | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1613 | if (test_bit(WantReplacement, &p->rdev->flags) && |
| 1614 | p[conf->raid_disks].rdev == NULL) { |
| 1615 | /* Add this device as a replacement */ |
| 1616 | clear_bit(In_sync, &rdev->flags); |
| 1617 | set_bit(Replacement, &rdev->flags); |
| 1618 | rdev->raid_disk = mirror; |
| 1619 | err = 0; |
| 1620 | conf->fullsync = 1; |
| 1621 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); |
| 1622 | break; |
| 1623 | } |
| 1624 | } |
Andre Noll | ac5e711 | 2009-08-03 10:59:47 +1000 | [diff] [blame] | 1625 | md_integrity_add_rdev(rdev, mddev); |
Jonathan Brassow | 9092c02 | 2013-05-02 14:19:24 -0500 | [diff] [blame] | 1626 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 1627 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | print_conf(conf); |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1629 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | } |
| 1631 | |
NeilBrown | b8321b6 | 2011-12-23 10:17:51 +1100 | [diff] [blame] | 1632 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1633 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1634 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1635 | int err = 0; |
NeilBrown | b8321b6 | 2011-12-23 10:17:51 +1100 | [diff] [blame] | 1636 | int number = rdev->raid_disk; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 1637 | struct raid1_info *p = conf->mirrors + number; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | |
NeilBrown | b014f14 | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1639 | if (rdev != p->rdev) |
| 1640 | p = conf->mirrors + conf->raid_disks + number; |
| 1641 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | print_conf(conf); |
NeilBrown | b8321b6 | 2011-12-23 10:17:51 +1100 | [diff] [blame] | 1643 | if (rdev == p->rdev) { |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1644 | if (test_bit(In_sync, &rdev->flags) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1645 | atomic_read(&rdev->nr_pending)) { |
| 1646 | err = -EBUSY; |
| 1647 | goto abort; |
| 1648 | } |
NeilBrown | 046abee | 2010-10-26 15:46:20 +1100 | [diff] [blame] | 1649 | /* Only remove non-faulty devices if recovery |
NeilBrown | dfc7064 | 2008-05-23 13:04:39 -0700 | [diff] [blame] | 1650 | * is not possible. |
| 1651 | */ |
| 1652 | if (!test_bit(Faulty, &rdev->flags) && |
NeilBrown | 5389042 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1653 | mddev->recovery_disabled != conf->recovery_disabled && |
NeilBrown | dfc7064 | 2008-05-23 13:04:39 -0700 | [diff] [blame] | 1654 | mddev->degraded < conf->raid_disks) { |
| 1655 | err = -EBUSY; |
| 1656 | goto abort; |
| 1657 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1658 | p->rdev = NULL; |
Paul E. McKenney | fbd568a3e | 2005-05-01 08:59:04 -0700 | [diff] [blame] | 1659 | synchronize_rcu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1660 | if (atomic_read(&rdev->nr_pending)) { |
| 1661 | /* lost the race, try later */ |
| 1662 | err = -EBUSY; |
| 1663 | p->rdev = rdev; |
Andre Noll | ac5e711 | 2009-08-03 10:59:47 +1000 | [diff] [blame] | 1664 | goto abort; |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1665 | } else if (conf->mirrors[conf->raid_disks + number].rdev) { |
| 1666 | /* We just removed a device that is being replaced. |
| 1667 | * Move down the replacement. We drain all IO before |
| 1668 | * doing this to avoid confusion. |
| 1669 | */ |
| 1670 | struct md_rdev *repl = |
| 1671 | conf->mirrors[conf->raid_disks + number].rdev; |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 1672 | freeze_array(conf, 0); |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1673 | clear_bit(Replacement, &repl->flags); |
| 1674 | p->rdev = repl; |
| 1675 | conf->mirrors[conf->raid_disks + number].rdev = NULL; |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 1676 | unfreeze_array(conf); |
NeilBrown | b014f14 | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1677 | clear_bit(WantReplacement, &rdev->flags); |
NeilBrown | 8c7a2c2 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1678 | } else |
| 1679 | clear_bit(WantReplacement, &rdev->flags); |
Martin K. Petersen | a91a278 | 2011-03-17 11:11:05 +0100 | [diff] [blame] | 1680 | err = md_integrity_register(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1681 | } |
| 1682 | abort: |
| 1683 | |
| 1684 | print_conf(conf); |
| 1685 | return err; |
| 1686 | } |
| 1687 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1688 | static void end_sync_read(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1689 | { |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1690 | struct r1bio *r1_bio = bio->bi_private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1691 | |
NeilBrown | 0fc280f | 2011-10-07 14:22:55 +1100 | [diff] [blame] | 1692 | update_head_pos(r1_bio->read_disk, r1_bio); |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 1693 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1694 | /* |
| 1695 | * we have read a block, now it needs to be re-written, |
| 1696 | * or re-read if the read failed. |
| 1697 | * We don't do much here, just schedule handling by raid1d |
| 1698 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1699 | if (!bio->bi_error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1700 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1701 | |
| 1702 | if (atomic_dec_and_test(&r1_bio->remaining)) |
| 1703 | reschedule_retry(r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1704 | } |
| 1705 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1706 | static void end_sync_write(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1707 | { |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1708 | int uptodate = !bio->bi_error; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1709 | struct r1bio *r1_bio = bio->bi_private; |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1710 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1711 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1712 | int mirror=0; |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1713 | sector_t first_bad; |
| 1714 | int bad_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1715 | |
Namhyung Kim | ba3ae3b | 2011-10-07 14:22:53 +1100 | [diff] [blame] | 1716 | mirror = find_bio_disk(r1_bio, bio); |
| 1717 | |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1718 | if (!uptodate) { |
NeilBrown | 57dab0b | 2010-10-19 10:03:39 +1100 | [diff] [blame] | 1719 | sector_t sync_blocks = 0; |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1720 | sector_t s = r1_bio->sector; |
| 1721 | long sectors_to_go = r1_bio->sectors; |
| 1722 | /* make sure these bits doesn't get cleared. */ |
| 1723 | do { |
NeilBrown | 5e3db64 | 2006-07-10 04:44:18 -0700 | [diff] [blame] | 1724 | bitmap_end_sync(mddev->bitmap, s, |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1725 | &sync_blocks, 1); |
| 1726 | s += sync_blocks; |
| 1727 | sectors_to_go -= sync_blocks; |
| 1728 | } while (sectors_to_go > 0); |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1729 | set_bit(WriteErrorSeen, |
| 1730 | &conf->mirrors[mirror].rdev->flags); |
NeilBrown | 19d6716 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1731 | if (!test_and_set_bit(WantReplacement, |
| 1732 | &conf->mirrors[mirror].rdev->flags)) |
| 1733 | set_bit(MD_RECOVERY_NEEDED, & |
| 1734 | mddev->recovery); |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1735 | set_bit(R1BIO_WriteError, &r1_bio->state); |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1736 | } else if (is_badblock(conf->mirrors[mirror].rdev, |
| 1737 | r1_bio->sector, |
| 1738 | r1_bio->sectors, |
NeilBrown | 3a9f28a | 2011-07-28 11:33:42 +1000 | [diff] [blame] | 1739 | &first_bad, &bad_sectors) && |
| 1740 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, |
| 1741 | r1_bio->sector, |
| 1742 | r1_bio->sectors, |
| 1743 | &first_bad, &bad_sectors) |
| 1744 | ) |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1745 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1746 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1747 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1748 | int s = r1_bio->sectors; |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1749 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
| 1750 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 1751 | reschedule_retry(r1_bio); |
| 1752 | else { |
| 1753 | put_buf(r1_bio); |
| 1754 | md_done_sync(mddev, s, uptodate); |
| 1755 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1756 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1757 | } |
| 1758 | |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1759 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1760 | int sectors, struct page *page, int rw) |
| 1761 | { |
| 1762 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) |
| 1763 | /* success */ |
| 1764 | return 1; |
NeilBrown | 19d6716 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1765 | if (rw == WRITE) { |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1766 | set_bit(WriteErrorSeen, &rdev->flags); |
NeilBrown | 19d6716 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 1767 | if (!test_and_set_bit(WantReplacement, |
| 1768 | &rdev->flags)) |
| 1769 | set_bit(MD_RECOVERY_NEEDED, & |
| 1770 | rdev->mddev->recovery); |
| 1771 | } |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1772 | /* need to record an error - either for the block or the device */ |
| 1773 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) |
| 1774 | md_error(rdev->mddev, rdev); |
| 1775 | return 0; |
| 1776 | } |
| 1777 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1778 | static int fix_sync_read_error(struct r1bio *r1_bio) |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1779 | { |
| 1780 | /* Try some synchronous reads of other devices to get |
| 1781 | * good data, much like with normal read errors. Only |
| 1782 | * read into the pages we already have so we don't |
| 1783 | * need to re-issue the read request. |
| 1784 | * We don't need to freeze the array, because being in an |
| 1785 | * active sync request, there is no normal IO, and |
| 1786 | * no overlapping syncs. |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 1787 | * We don't need to check is_badblock() again as we |
| 1788 | * made sure that anything with a bad block in range |
| 1789 | * will have bi_end_io clear. |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1790 | */ |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1791 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1792 | struct r1conf *conf = mddev->private; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1793 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
| 1794 | sector_t sect = r1_bio->sector; |
| 1795 | int sectors = r1_bio->sectors; |
| 1796 | int idx = 0; |
| 1797 | |
| 1798 | while(sectors) { |
| 1799 | int s = sectors; |
| 1800 | int d = r1_bio->read_disk; |
| 1801 | int success = 0; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 1802 | struct md_rdev *rdev; |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1803 | int start; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1804 | |
| 1805 | if (s > (PAGE_SIZE>>9)) |
| 1806 | s = PAGE_SIZE >> 9; |
| 1807 | do { |
| 1808 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { |
| 1809 | /* No rcu protection needed here devices |
| 1810 | * can only be removed when no resync is |
| 1811 | * active, and resync is currently active |
| 1812 | */ |
| 1813 | rdev = conf->mirrors[d].rdev; |
Namhyung Kim | 9d3d801 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1814 | if (sync_page_io(rdev, sect, s<<9, |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1815 | bio->bi_io_vec[idx].bv_page, |
| 1816 | READ, false)) { |
| 1817 | success = 1; |
| 1818 | break; |
| 1819 | } |
| 1820 | } |
| 1821 | d++; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1822 | if (d == conf->raid_disks * 2) |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1823 | d = 0; |
| 1824 | } while (!success && d != r1_bio->read_disk); |
| 1825 | |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1826 | if (!success) { |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1827 | char b[BDEVNAME_SIZE]; |
NeilBrown | 3a9f28a | 2011-07-28 11:33:42 +1000 | [diff] [blame] | 1828 | int abort = 0; |
| 1829 | /* Cannot read from anywhere, this block is lost. |
| 1830 | * Record a bad block on each device. If that doesn't |
| 1831 | * work just disable and interrupt the recovery. |
| 1832 | * Don't fail devices as that won't really help. |
| 1833 | */ |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1834 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
| 1835 | " for block %llu\n", |
| 1836 | mdname(mddev), |
| 1837 | bdevname(bio->bi_bdev, b), |
| 1838 | (unsigned long long)r1_bio->sector); |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1839 | for (d = 0; d < conf->raid_disks * 2; d++) { |
NeilBrown | 3a9f28a | 2011-07-28 11:33:42 +1000 | [diff] [blame] | 1840 | rdev = conf->mirrors[d].rdev; |
| 1841 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
| 1842 | continue; |
| 1843 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
| 1844 | abort = 1; |
| 1845 | } |
| 1846 | if (abort) { |
NeilBrown | d890fa2 | 2011-10-26 11:54:39 +1100 | [diff] [blame] | 1847 | conf->recovery_disabled = |
| 1848 | mddev->recovery_disabled; |
NeilBrown | 3a9f28a | 2011-07-28 11:33:42 +1000 | [diff] [blame] | 1849 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
| 1850 | md_done_sync(mddev, r1_bio->sectors, 0); |
| 1851 | put_buf(r1_bio); |
| 1852 | return 0; |
| 1853 | } |
| 1854 | /* Try next page */ |
| 1855 | sectors -= s; |
| 1856 | sect += s; |
| 1857 | idx++; |
| 1858 | continue; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1859 | } |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1860 | |
| 1861 | start = d; |
| 1862 | /* write it back and re-read */ |
| 1863 | while (d != r1_bio->read_disk) { |
| 1864 | if (d == 0) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1865 | d = conf->raid_disks * 2; |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1866 | d--; |
| 1867 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) |
| 1868 | continue; |
| 1869 | rdev = conf->mirrors[d].rdev; |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1870 | if (r1_sync_page_io(rdev, sect, s, |
| 1871 | bio->bi_io_vec[idx].bv_page, |
| 1872 | WRITE) == 0) { |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1873 | r1_bio->bios[d]->bi_end_io = NULL; |
| 1874 | rdev_dec_pending(rdev, mddev); |
Namhyung Kim | 9d3d801 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1875 | } |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1876 | } |
| 1877 | d = start; |
| 1878 | while (d != r1_bio->read_disk) { |
| 1879 | if (d == 0) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1880 | d = conf->raid_disks * 2; |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1881 | d--; |
| 1882 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) |
| 1883 | continue; |
| 1884 | rdev = conf->mirrors[d].rdev; |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 1885 | if (r1_sync_page_io(rdev, sect, s, |
| 1886 | bio->bi_io_vec[idx].bv_page, |
| 1887 | READ) != 0) |
Namhyung Kim | 9d3d801 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 1888 | atomic_add(s, &rdev->corrected_errors); |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1889 | } |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1890 | sectors -= s; |
| 1891 | sect += s; |
| 1892 | idx ++; |
| 1893 | } |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1894 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1895 | bio->bi_error = 0; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1896 | return 1; |
| 1897 | } |
| 1898 | |
NeilBrown | c95e638 | 2014-09-09 13:54:11 +1000 | [diff] [blame] | 1899 | static void process_checks(struct r1bio *r1_bio) |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1900 | { |
| 1901 | /* We have read all readable devices. If we haven't |
| 1902 | * got the block, then there is no hope left. |
| 1903 | * If we have, then we want to do a comparison |
| 1904 | * and skip the write if everything is the same. |
| 1905 | * If any blocks failed to read, then we need to |
| 1906 | * attempt an over-write |
| 1907 | */ |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 1908 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1909 | struct r1conf *conf = mddev->private; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1910 | int primary; |
| 1911 | int i; |
majianpeng | f4380a9 | 2012-04-12 16:04:47 +1000 | [diff] [blame] | 1912 | int vcnt; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1913 | |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1914 | /* Fix variable parts of all bios */ |
| 1915 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); |
| 1916 | for (i = 0; i < conf->raid_disks * 2; i++) { |
| 1917 | int j; |
| 1918 | int size; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1919 | int error; |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1920 | struct bio *b = r1_bio->bios[i]; |
| 1921 | if (b->bi_end_io != end_sync_read) |
| 1922 | continue; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1923 | /* fixup the bio for reuse, but preserve errno */ |
| 1924 | error = b->bi_error; |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1925 | bio_reset(b); |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1926 | b->bi_error = error; |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1927 | b->bi_vcnt = vcnt; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1928 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
| 1929 | b->bi_iter.bi_sector = r1_bio->sector + |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1930 | conf->mirrors[i].rdev->data_offset; |
| 1931 | b->bi_bdev = conf->mirrors[i].rdev->bdev; |
| 1932 | b->bi_end_io = end_sync_read; |
| 1933 | b->bi_private = r1_bio; |
| 1934 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 1935 | size = b->bi_iter.bi_size; |
NeilBrown | 30bc9b5 | 2013-07-17 15:19:29 +1000 | [diff] [blame] | 1936 | for (j = 0; j < vcnt ; j++) { |
| 1937 | struct bio_vec *bi; |
| 1938 | bi = &b->bi_io_vec[j]; |
| 1939 | bi->bv_offset = 0; |
| 1940 | if (size > PAGE_SIZE) |
| 1941 | bi->bv_len = PAGE_SIZE; |
| 1942 | else |
| 1943 | bi->bv_len = size; |
| 1944 | size -= PAGE_SIZE; |
| 1945 | } |
| 1946 | } |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1947 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1948 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1949 | !r1_bio->bios[primary]->bi_error) { |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1950 | r1_bio->bios[primary]->bi_end_io = NULL; |
| 1951 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); |
| 1952 | break; |
| 1953 | } |
| 1954 | r1_bio->read_disk = primary; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1955 | for (i = 0; i < conf->raid_disks * 2; i++) { |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1956 | int j; |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1957 | struct bio *pbio = r1_bio->bios[primary]; |
| 1958 | struct bio *sbio = r1_bio->bios[i]; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1959 | int error = sbio->bi_error; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1960 | |
Kent Overstreet | 2aabaa6 | 2012-09-11 11:26:12 -0700 | [diff] [blame] | 1961 | if (sbio->bi_end_io != end_sync_read) |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1962 | continue; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1963 | /* Now we can 'fixup' the error value */ |
| 1964 | sbio->bi_error = 0; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1965 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1966 | if (!error) { |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1967 | for (j = vcnt; j-- ; ) { |
| 1968 | struct page *p, *s; |
| 1969 | p = pbio->bi_io_vec[j].bv_page; |
| 1970 | s = sbio->bi_io_vec[j].bv_page; |
| 1971 | if (memcmp(page_address(p), |
| 1972 | page_address(s), |
NeilBrown | 5020ad7 | 2012-04-02 01:39:05 +1000 | [diff] [blame] | 1973 | sbio->bi_io_vec[j].bv_len)) |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1974 | break; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1975 | } |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1976 | } else |
| 1977 | j = 0; |
| 1978 | if (j >= 0) |
Jianpeng Ma | 7f7583d | 2012-10-11 14:17:59 +1100 | [diff] [blame] | 1979 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1980 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 1981 | && !error)) { |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1982 | /* No need to write to this device. */ |
| 1983 | sbio->bi_end_io = NULL; |
| 1984 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); |
| 1985 | continue; |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1986 | } |
Kent Overstreet | d3b45c2 | 2012-09-10 13:49:33 -0700 | [diff] [blame] | 1987 | |
| 1988 | bio_copy_data(sbio, pbio); |
NeilBrown | 78d7f5f | 2011-05-11 14:48:56 +1000 | [diff] [blame] | 1989 | } |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 1990 | } |
| 1991 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 1992 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1993 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 1994 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1995 | int i; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 1996 | int disks = conf->raid_disks * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1997 | struct bio *bio, *wbio; |
| 1998 | |
| 1999 | bio = r1_bio->bios[r1_bio->read_disk]; |
| 2000 | |
NeilBrown | a68e587 | 2011-05-11 14:40:44 +1000 | [diff] [blame] | 2001 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
| 2002 | /* ouch - failed to read all of that. */ |
| 2003 | if (!fix_sync_read_error(r1_bio)) |
| 2004 | return; |
NeilBrown | 7ca78d5 | 2011-05-11 14:50:37 +1000 | [diff] [blame] | 2005 | |
| 2006 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
NeilBrown | c95e638 | 2014-09-09 13:54:11 +1000 | [diff] [blame] | 2007 | process_checks(r1_bio); |
| 2008 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2009 | /* |
| 2010 | * schedule writes |
| 2011 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2012 | atomic_set(&r1_bio->remaining, 1); |
| 2013 | for (i = 0; i < disks ; i++) { |
| 2014 | wbio = r1_bio->bios[i]; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2015 | if (wbio->bi_end_io == NULL || |
| 2016 | (wbio->bi_end_io == end_sync_read && |
| 2017 | (i == r1_bio->read_disk || |
| 2018 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2019 | continue; |
| 2020 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2021 | wbio->bi_rw = WRITE; |
| 2022 | wbio->bi_end_io = end_sync_write; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2023 | atomic_inc(&r1_bio->remaining); |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 2024 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2025 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2026 | generic_make_request(wbio); |
| 2027 | } |
| 2028 | |
| 2029 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2030 | /* if we're here, all write(s) have completed, so clean up */ |
NeilBrown | 58e94ae | 2012-07-19 15:59:18 +1000 | [diff] [blame] | 2031 | int s = r1_bio->sectors; |
| 2032 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
| 2033 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2034 | reschedule_retry(r1_bio); |
| 2035 | else { |
| 2036 | put_buf(r1_bio); |
| 2037 | md_done_sync(mddev, s, 1); |
| 2038 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2039 | } |
| 2040 | } |
| 2041 | |
| 2042 | /* |
| 2043 | * This is a kernel thread which: |
| 2044 | * |
| 2045 | * 1. Retries failed read operations on working mirrors. |
| 2046 | * 2. Updates the raid superblock when problems encounter. |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2047 | * 3. Performs writes following reads for array synchronising. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2048 | */ |
| 2049 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2050 | static void fix_read_error(struct r1conf *conf, int read_disk, |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2051 | sector_t sect, int sectors) |
| 2052 | { |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2053 | struct mddev *mddev = conf->mddev; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2054 | while(sectors) { |
| 2055 | int s = sectors; |
| 2056 | int d = read_disk; |
| 2057 | int success = 0; |
| 2058 | int start; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2059 | struct md_rdev *rdev; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2060 | |
| 2061 | if (s > (PAGE_SIZE>>9)) |
| 2062 | s = PAGE_SIZE >> 9; |
| 2063 | |
| 2064 | do { |
| 2065 | /* Note: no rcu protection needed here |
| 2066 | * as this is synchronous in the raid1d thread |
| 2067 | * which is the thread that might remove |
| 2068 | * a device. If raid1d ever becomes multi-threaded.... |
| 2069 | */ |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2070 | sector_t first_bad; |
| 2071 | int bad_sectors; |
| 2072 | |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2073 | rdev = conf->mirrors[d].rdev; |
| 2074 | if (rdev && |
majianpeng | da8840a | 2012-05-22 13:55:03 +1000 | [diff] [blame] | 2075 | (test_bit(In_sync, &rdev->flags) || |
| 2076 | (!test_bit(Faulty, &rdev->flags) && |
| 2077 | rdev->recovery_offset >= sect + s)) && |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2078 | is_badblock(rdev, sect, s, |
| 2079 | &first_bad, &bad_sectors) == 0 && |
Jonathan Brassow | ccebd4c | 2011-01-14 09:14:33 +1100 | [diff] [blame] | 2080 | sync_page_io(rdev, sect, s<<9, |
| 2081 | conf->tmppage, READ, false)) |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2082 | success = 1; |
| 2083 | else { |
| 2084 | d++; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2085 | if (d == conf->raid_disks * 2) |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2086 | d = 0; |
| 2087 | } |
| 2088 | } while (!success && d != read_disk); |
| 2089 | |
| 2090 | if (!success) { |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2091 | /* Cannot read from anywhere - mark it bad */ |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2092 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2093 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
| 2094 | md_error(mddev, rdev); |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2095 | break; |
| 2096 | } |
| 2097 | /* write it back and re-read */ |
| 2098 | start = d; |
| 2099 | while (d != read_disk) { |
| 2100 | if (d==0) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2101 | d = conf->raid_disks * 2; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2102 | d--; |
| 2103 | rdev = conf->mirrors[d].rdev; |
| 2104 | if (rdev && |
NeilBrown | b8cb6b4 | 2014-09-18 11:09:04 +1000 | [diff] [blame] | 2105 | !test_bit(Faulty, &rdev->flags)) |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2106 | r1_sync_page_io(rdev, sect, s, |
| 2107 | conf->tmppage, WRITE); |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2108 | } |
| 2109 | d = start; |
| 2110 | while (d != read_disk) { |
| 2111 | char b[BDEVNAME_SIZE]; |
| 2112 | if (d==0) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2113 | d = conf->raid_disks * 2; |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2114 | d--; |
| 2115 | rdev = conf->mirrors[d].rdev; |
| 2116 | if (rdev && |
NeilBrown | b8cb6b4 | 2014-09-18 11:09:04 +1000 | [diff] [blame] | 2117 | !test_bit(Faulty, &rdev->flags)) { |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2118 | if (r1_sync_page_io(rdev, sect, s, |
| 2119 | conf->tmppage, READ)) { |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2120 | atomic_add(s, &rdev->corrected_errors); |
| 2121 | printk(KERN_INFO |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2122 | "md/raid1:%s: read error corrected " |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2123 | "(%d sectors at %llu on %s)\n", |
| 2124 | mdname(mddev), s, |
Randy Dunlap | 969b755 | 2006-10-28 10:38:32 -0700 | [diff] [blame] | 2125 | (unsigned long long)(sect + |
| 2126 | rdev->data_offset), |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 2127 | bdevname(rdev->bdev, b)); |
| 2128 | } |
| 2129 | } |
| 2130 | } |
| 2131 | sectors -= s; |
| 2132 | sect += s; |
| 2133 | } |
| 2134 | } |
| 2135 | |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 2136 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2137 | { |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2138 | struct mddev *mddev = r1_bio->mddev; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2139 | struct r1conf *conf = mddev->private; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2140 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2141 | |
| 2142 | /* bio has the data to be written to device 'i' where |
| 2143 | * we just recently had a write error. |
| 2144 | * We repeatedly clone the bio and trim down to one block, |
| 2145 | * then try the write. Where the write fails we record |
| 2146 | * a bad block. |
| 2147 | * It is conceivable that the bio doesn't exactly align with |
| 2148 | * blocks. We must handle this somehow. |
| 2149 | * |
| 2150 | * We currently own a reference on the rdev. |
| 2151 | */ |
| 2152 | |
| 2153 | int block_sectors; |
| 2154 | sector_t sector; |
| 2155 | int sectors; |
| 2156 | int sect_to_write = r1_bio->sectors; |
| 2157 | int ok = 1; |
| 2158 | |
| 2159 | if (rdev->badblocks.shift < 0) |
| 2160 | return 0; |
| 2161 | |
Nate Dailey | ab713cd | 2015-02-12 12:02:09 -0500 | [diff] [blame] | 2162 | block_sectors = roundup(1 << rdev->badblocks.shift, |
| 2163 | bdev_logical_block_size(rdev->bdev) >> 9); |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2164 | sector = r1_bio->sector; |
| 2165 | sectors = ((sector + block_sectors) |
| 2166 | & ~(sector_t)(block_sectors - 1)) |
| 2167 | - sector; |
| 2168 | |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2169 | while (sect_to_write) { |
| 2170 | struct bio *wbio; |
| 2171 | if (sectors > sect_to_write) |
| 2172 | sectors = sect_to_write; |
| 2173 | /* Write at 'sector' for 'sectors'*/ |
| 2174 | |
Kent Overstreet | b783863 | 2012-09-10 15:17:11 -0700 | [diff] [blame] | 2175 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
| 2176 | unsigned vcnt = r1_bio->behind_page_count; |
| 2177 | struct bio_vec *vec = r1_bio->behind_bvecs; |
| 2178 | |
| 2179 | while (!vec->bv_page) { |
| 2180 | vec++; |
| 2181 | vcnt--; |
| 2182 | } |
| 2183 | |
| 2184 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); |
| 2185 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); |
| 2186 | |
| 2187 | wbio->bi_vcnt = vcnt; |
| 2188 | } else { |
| 2189 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); |
| 2190 | } |
| 2191 | |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2192 | wbio->bi_rw = WRITE; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2193 | wbio->bi_iter.bi_sector = r1_bio->sector; |
| 2194 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2195 | |
Kent Overstreet | 6678d83 | 2013-08-07 11:14:32 -0700 | [diff] [blame] | 2196 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2197 | wbio->bi_iter.bi_sector += rdev->data_offset; |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2198 | wbio->bi_bdev = rdev->bdev; |
| 2199 | if (submit_bio_wait(WRITE, wbio) == 0) |
| 2200 | /* failure! */ |
| 2201 | ok = rdev_set_badblocks(rdev, sector, |
| 2202 | sectors, 0) |
| 2203 | && ok; |
| 2204 | |
| 2205 | bio_put(wbio); |
| 2206 | sect_to_write -= sectors; |
| 2207 | sector += sectors; |
| 2208 | sectors = block_sectors; |
| 2209 | } |
| 2210 | return ok; |
| 2211 | } |
| 2212 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2213 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2214 | { |
| 2215 | int m; |
| 2216 | int s = r1_bio->sectors; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2217 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2218 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2219 | struct bio *bio = r1_bio->bios[m]; |
| 2220 | if (bio->bi_end_io == NULL) |
| 2221 | continue; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2222 | if (!bio->bi_error && |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2223 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
NeilBrown | c6563a8 | 2012-05-21 09:27:00 +1000 | [diff] [blame] | 2224 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2225 | } |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2226 | if (bio->bi_error && |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2227 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
| 2228 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) |
| 2229 | md_error(conf->mddev, rdev); |
| 2230 | } |
| 2231 | } |
| 2232 | put_buf(r1_bio); |
| 2233 | md_done_sync(conf->mddev, s, 1); |
| 2234 | } |
| 2235 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2236 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2237 | { |
| 2238 | int m; |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2239 | bool fail = false; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2240 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2241 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2242 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2243 | rdev_clear_badblocks(rdev, |
| 2244 | r1_bio->sector, |
NeilBrown | c6563a8 | 2012-05-21 09:27:00 +1000 | [diff] [blame] | 2245 | r1_bio->sectors, 0); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2246 | rdev_dec_pending(rdev, conf->mddev); |
| 2247 | } else if (r1_bio->bios[m] != NULL) { |
| 2248 | /* This drive got a write error. We need to |
| 2249 | * narrow down and record precise write |
| 2250 | * errors. |
| 2251 | */ |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2252 | fail = true; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2253 | if (!narrow_write_error(r1_bio, m)) { |
| 2254 | md_error(conf->mddev, |
| 2255 | conf->mirrors[m].rdev); |
| 2256 | /* an I/O failed, we can't clear the bitmap */ |
| 2257 | set_bit(R1BIO_Degraded, &r1_bio->state); |
| 2258 | } |
| 2259 | rdev_dec_pending(conf->mirrors[m].rdev, |
| 2260 | conf->mddev); |
| 2261 | } |
| 2262 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2263 | close_write(r1_bio); |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2264 | if (fail) { |
| 2265 | spin_lock_irq(&conf->device_lock); |
| 2266 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); |
| 2267 | spin_unlock_irq(&conf->device_lock); |
| 2268 | md_wakeup_thread(conf->mddev->thread); |
| 2269 | } else |
| 2270 | raid_end_bio_io(r1_bio); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2271 | } |
| 2272 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2273 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2274 | { |
| 2275 | int disk; |
| 2276 | int max_sectors; |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2277 | struct mddev *mddev = conf->mddev; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2278 | struct bio *bio; |
| 2279 | char b[BDEVNAME_SIZE]; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2280 | struct md_rdev *rdev; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2281 | |
| 2282 | clear_bit(R1BIO_ReadError, &r1_bio->state); |
| 2283 | /* we got a read error. Maybe the drive is bad. Maybe just |
| 2284 | * the block and we can fix it. |
| 2285 | * We freeze all other IO, and try reading the block from |
| 2286 | * other devices. When we find one, we re-write |
| 2287 | * and check it that fixes the read error. |
| 2288 | * This is all done synchronously while the array is |
| 2289 | * frozen |
| 2290 | */ |
| 2291 | if (mddev->ro == 0) { |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 2292 | freeze_array(conf, 1); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2293 | fix_read_error(conf, r1_bio->read_disk, |
| 2294 | r1_bio->sector, r1_bio->sectors); |
| 2295 | unfreeze_array(conf); |
| 2296 | } else |
| 2297 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); |
NeilBrown | 7ad4d4a | 2012-10-11 13:44:30 +1100 | [diff] [blame] | 2298 | rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2299 | |
| 2300 | bio = r1_bio->bios[r1_bio->read_disk]; |
| 2301 | bdevname(bio->bi_bdev, b); |
| 2302 | read_more: |
| 2303 | disk = read_balance(conf, r1_bio, &max_sectors); |
| 2304 | if (disk == -1) { |
| 2305 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" |
| 2306 | " read error for block %llu\n", |
| 2307 | mdname(mddev), b, (unsigned long long)r1_bio->sector); |
| 2308 | raid_end_bio_io(r1_bio); |
| 2309 | } else { |
| 2310 | const unsigned long do_sync |
| 2311 | = r1_bio->master_bio->bi_rw & REQ_SYNC; |
| 2312 | if (bio) { |
| 2313 | r1_bio->bios[r1_bio->read_disk] = |
| 2314 | mddev->ro ? IO_BLOCKED : NULL; |
| 2315 | bio_put(bio); |
| 2316 | } |
| 2317 | r1_bio->read_disk = disk; |
| 2318 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2319 | bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, |
| 2320 | max_sectors); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2321 | r1_bio->bios[r1_bio->read_disk] = bio; |
| 2322 | rdev = conf->mirrors[disk].rdev; |
| 2323 | printk_ratelimited(KERN_ERR |
| 2324 | "md/raid1:%s: redirecting sector %llu" |
| 2325 | " to other mirror: %s\n", |
| 2326 | mdname(mddev), |
| 2327 | (unsigned long long)r1_bio->sector, |
| 2328 | bdevname(rdev->bdev, b)); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2329 | bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2330 | bio->bi_bdev = rdev->bdev; |
| 2331 | bio->bi_end_io = raid1_end_read_request; |
| 2332 | bio->bi_rw = READ | do_sync; |
| 2333 | bio->bi_private = r1_bio; |
| 2334 | if (max_sectors < r1_bio->sectors) { |
| 2335 | /* Drat - have to split this up more */ |
| 2336 | struct bio *mbio = r1_bio->master_bio; |
| 2337 | int sectors_handled = (r1_bio->sector + max_sectors |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2338 | - mbio->bi_iter.bi_sector); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2339 | r1_bio->sectors = max_sectors; |
| 2340 | spin_lock_irq(&conf->device_lock); |
| 2341 | if (mbio->bi_phys_segments == 0) |
| 2342 | mbio->bi_phys_segments = 2; |
| 2343 | else |
| 2344 | mbio->bi_phys_segments++; |
| 2345 | spin_unlock_irq(&conf->device_lock); |
| 2346 | generic_make_request(bio); |
| 2347 | bio = NULL; |
| 2348 | |
| 2349 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 2350 | |
| 2351 | r1_bio->master_bio = mbio; |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 2352 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2353 | r1_bio->state = 0; |
| 2354 | set_bit(R1BIO_ReadError, &r1_bio->state); |
| 2355 | r1_bio->mddev = mddev; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2356 | r1_bio->sector = mbio->bi_iter.bi_sector + |
| 2357 | sectors_handled; |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2358 | |
| 2359 | goto read_more; |
| 2360 | } else |
| 2361 | generic_make_request(bio); |
| 2362 | } |
| 2363 | } |
| 2364 | |
Shaohua Li | 4ed8731 | 2012-10-11 13:34:00 +1100 | [diff] [blame] | 2365 | static void raid1d(struct md_thread *thread) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2366 | { |
Shaohua Li | 4ed8731 | 2012-10-11 13:34:00 +1100 | [diff] [blame] | 2367 | struct mddev *mddev = thread->mddev; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 2368 | struct r1bio *r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2369 | unsigned long flags; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2370 | struct r1conf *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2371 | struct list_head *head = &conf->retry_list; |
NeilBrown | e1dfa0a | 2011-04-18 18:25:41 +1000 | [diff] [blame] | 2372 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2373 | |
| 2374 | md_check_recovery(mddev); |
NeilBrown | e1dfa0a | 2011-04-18 18:25:41 +1000 | [diff] [blame] | 2375 | |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2376 | if (!list_empty_careful(&conf->bio_end_io_list) && |
| 2377 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { |
| 2378 | LIST_HEAD(tmp); |
| 2379 | spin_lock_irqsave(&conf->device_lock, flags); |
| 2380 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { |
| 2381 | list_add(&tmp, &conf->bio_end_io_list); |
| 2382 | list_del_init(&conf->bio_end_io_list); |
| 2383 | } |
| 2384 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 2385 | while (!list_empty(&tmp)) { |
| 2386 | r1_bio = list_first_entry(&conf->bio_end_io_list, |
| 2387 | struct r1bio, retry_list); |
| 2388 | list_del(&r1_bio->retry_list); |
| 2389 | raid_end_bio_io(r1_bio); |
| 2390 | } |
| 2391 | } |
| 2392 | |
NeilBrown | e1dfa0a | 2011-04-18 18:25:41 +1000 | [diff] [blame] | 2393 | blk_start_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2394 | for (;;) { |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 2395 | |
NeilBrown | 0021b7b | 2012-07-31 09:08:14 +0200 | [diff] [blame] | 2396 | flush_pending_writes(conf); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 2397 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2398 | spin_lock_irqsave(&conf->device_lock, flags); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 2399 | if (list_empty(head)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2400 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2401 | break; |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 2402 | } |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 2403 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | list_del(head->prev); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 2405 | conf->nr_queued--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2406 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 2407 | |
| 2408 | mddev = r1_bio->mddev; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 2409 | conf = mddev->private; |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 2410 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
NeilBrown | d8f05d2 | 2011-07-28 11:33:00 +1000 | [diff] [blame] | 2411 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2412 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2413 | handle_sync_write_finished(conf, r1_bio); |
| 2414 | else |
NeilBrown | 4367af5 | 2011-07-28 11:31:49 +1000 | [diff] [blame] | 2415 | sync_request_write(mddev, r1_bio); |
NeilBrown | cd5ff9a1 | 2011-07-28 11:32:41 +1000 | [diff] [blame] | 2416 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2417 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
| 2418 | handle_write_finished(conf, r1_bio); |
| 2419 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) |
| 2420 | handle_read_error(conf, r1_bio); |
| 2421 | else |
NeilBrown | d2eb35a | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2422 | /* just a partial read to be scheduled from separate |
| 2423 | * context |
| 2424 | */ |
| 2425 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); |
NeilBrown | 62096bc | 2011-07-28 11:38:13 +1000 | [diff] [blame] | 2426 | |
NeilBrown | 1d9d524 | 2009-10-16 15:55:32 +1100 | [diff] [blame] | 2427 | cond_resched(); |
NeilBrown | de393cd | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2428 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
| 2429 | md_check_recovery(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2430 | } |
NeilBrown | e1dfa0a | 2011-04-18 18:25:41 +1000 | [diff] [blame] | 2431 | blk_finish_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2432 | } |
| 2433 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2434 | static int init_resync(struct r1conf *conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2435 | { |
| 2436 | int buffs; |
| 2437 | |
| 2438 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; |
Eric Sesterhenn | 9e77c48 | 2006-04-01 01:08:49 +0200 | [diff] [blame] | 2439 | BUG_ON(conf->r1buf_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2440 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
| 2441 | conf->poolinfo); |
| 2442 | if (!conf->r1buf_pool) |
| 2443 | return -ENOMEM; |
| 2444 | conf->next_resync = 0; |
| 2445 | return 0; |
| 2446 | } |
| 2447 | |
| 2448 | /* |
| 2449 | * perform a "sync" on one "block" |
| 2450 | * |
| 2451 | * We need to make sure that no normal I/O request - particularly write |
| 2452 | * requests - conflict with active sync requests. |
| 2453 | * |
| 2454 | * This is achieved by tracking pending requests and a 'barrier' concept |
| 2455 | * that can be installed to exclude normal IO requests. |
| 2456 | */ |
| 2457 | |
NeilBrown | 0931479 | 2015-02-19 16:04:40 +1100 | [diff] [blame] | 2458 | static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2459 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2460 | struct r1conf *conf = mddev->private; |
NeilBrown | 9f2c9d1 | 2011-10-11 16:48:43 +1100 | [diff] [blame] | 2461 | struct r1bio *r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2462 | struct bio *bio; |
| 2463 | sector_t max_sector, nr_sectors; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2464 | int disk = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2465 | int i; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2466 | int wonly = -1; |
| 2467 | int write_targets = 0, read_targets = 0; |
NeilBrown | 57dab0b | 2010-10-19 10:03:39 +1100 | [diff] [blame] | 2468 | sector_t sync_blocks; |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 2469 | int still_degraded = 0; |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2470 | int good_sectors = RESYNC_SECTORS; |
| 2471 | 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] | 2472 | |
| 2473 | if (!conf->r1buf_pool) |
| 2474 | if (init_resync(conf)) |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 2475 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2476 | |
Andre Noll | 58c0fed | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 2477 | max_sector = mddev->dev_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2478 | if (sector_nr >= max_sector) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2479 | /* If we aborted, we need to abort the |
| 2480 | * sync on the 'current' bitmap chunk (there will |
| 2481 | * only be one in raid1 resync. |
| 2482 | * We can find the current addess in mddev->curr_resync |
| 2483 | */ |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2484 | if (mddev->curr_resync < max_sector) /* aborted */ |
| 2485 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2486 | &sync_blocks, 1); |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2487 | else /* completed sync */ |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2488 | conf->fullsync = 0; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2489 | |
| 2490 | bitmap_close_sync(mddev->bitmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2491 | close_sync(conf); |
| 2492 | return 0; |
| 2493 | } |
| 2494 | |
NeilBrown | 07d84d10 | 2006-06-26 00:27:56 -0700 | [diff] [blame] | 2495 | if (mddev->bitmap == NULL && |
| 2496 | mddev->recovery_cp == MaxSector && |
NeilBrown | 6394cca | 2006-08-27 01:23:50 -0700 | [diff] [blame] | 2497 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
NeilBrown | 07d84d10 | 2006-06-26 00:27:56 -0700 | [diff] [blame] | 2498 | conf->fullsync == 0) { |
| 2499 | *skipped = 1; |
| 2500 | return max_sector - sector_nr; |
| 2501 | } |
NeilBrown | 6394cca | 2006-08-27 01:23:50 -0700 | [diff] [blame] | 2502 | /* before building a request, check if we can skip these blocks.. |
| 2503 | * This call the bitmap_start_sync doesn't actually record anything |
| 2504 | */ |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 2505 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
NeilBrown | e5de485f | 2005-11-08 21:39:38 -0800 | [diff] [blame] | 2506 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2507 | /* We can skip this block, and probably several more */ |
| 2508 | *skipped = 1; |
| 2509 | return sync_blocks; |
| 2510 | } |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2511 | |
NeilBrown | b47490c | 2008-02-06 01:39:50 -0800 | [diff] [blame] | 2512 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
NeilBrown | 1c4588e | 2010-10-26 17:41:22 +1100 | [diff] [blame] | 2513 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2514 | |
NeilBrown | c2fd4c9 | 2014-09-10 16:01:24 +1000 | [diff] [blame] | 2515 | raise_barrier(conf, sector_nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2516 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2517 | rcu_read_lock(); |
| 2518 | /* |
| 2519 | * If we get a correctably read error during resync or recovery, |
| 2520 | * we might want to read from a different device. So we |
| 2521 | * flag all drives that could conceivably be read from for READ, |
| 2522 | * and any others (which will be non-In_sync devices) for WRITE. |
| 2523 | * If a read fails, we try reading from something else for which READ |
| 2524 | * is OK. |
| 2525 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2526 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2527 | r1_bio->mddev = mddev; |
| 2528 | r1_bio->sector = sector_nr; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2529 | r1_bio->state = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2530 | set_bit(R1BIO_IsSync, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2531 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2532 | for (i = 0; i < conf->raid_disks * 2; i++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2533 | struct md_rdev *rdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2534 | bio = r1_bio->bios[i]; |
Kent Overstreet | 2aabaa6 | 2012-09-11 11:26:12 -0700 | [diff] [blame] | 2535 | bio_reset(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2536 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2537 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
| 2538 | if (rdev == NULL || |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2539 | test_bit(Faulty, &rdev->flags)) { |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2540 | if (i < conf->raid_disks) |
| 2541 | still_degraded = 1; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2542 | } else if (!test_bit(In_sync, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2543 | bio->bi_rw = WRITE; |
| 2544 | bio->bi_end_io = end_sync_write; |
| 2545 | write_targets ++; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2546 | } else { |
| 2547 | /* may need to read from here */ |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2548 | sector_t first_bad = MaxSector; |
| 2549 | int bad_sectors; |
| 2550 | |
| 2551 | if (is_badblock(rdev, sector_nr, good_sectors, |
| 2552 | &first_bad, &bad_sectors)) { |
| 2553 | if (first_bad > sector_nr) |
| 2554 | good_sectors = first_bad - sector_nr; |
| 2555 | else { |
| 2556 | bad_sectors -= (sector_nr - first_bad); |
| 2557 | if (min_bad == 0 || |
| 2558 | min_bad > bad_sectors) |
| 2559 | min_bad = bad_sectors; |
| 2560 | } |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2561 | } |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2562 | if (sector_nr < first_bad) { |
| 2563 | if (test_bit(WriteMostly, &rdev->flags)) { |
| 2564 | if (wonly < 0) |
| 2565 | wonly = i; |
| 2566 | } else { |
| 2567 | if (disk < 0) |
| 2568 | disk = i; |
| 2569 | } |
| 2570 | bio->bi_rw = READ; |
| 2571 | bio->bi_end_io = end_sync_read; |
| 2572 | read_targets++; |
Alexander Lyakas | d57368a | 2012-07-17 13:17:55 +0300 | [diff] [blame] | 2573 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
| 2574 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
| 2575 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
| 2576 | /* |
| 2577 | * The device is suitable for reading (InSync), |
| 2578 | * but has bad block(s) here. Let's try to correct them, |
| 2579 | * if we are doing resync or repair. Otherwise, leave |
| 2580 | * this device alone for this sync request. |
| 2581 | */ |
| 2582 | bio->bi_rw = WRITE; |
| 2583 | bio->bi_end_io = end_sync_write; |
| 2584 | write_targets++; |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2585 | } |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2586 | } |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2587 | if (bio->bi_end_io) { |
| 2588 | atomic_inc(&rdev->nr_pending); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2589 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2590 | bio->bi_bdev = rdev->bdev; |
| 2591 | bio->bi_private = r1_bio; |
| 2592 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2593 | } |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2594 | rcu_read_unlock(); |
| 2595 | if (disk < 0) |
| 2596 | disk = wonly; |
| 2597 | r1_bio->read_disk = disk; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2598 | |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2599 | if (read_targets == 0 && min_bad > 0) { |
| 2600 | /* These sectors are bad on all InSync devices, so we |
| 2601 | * need to mark them bad on all write targets |
| 2602 | */ |
| 2603 | int ok = 1; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2604 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2605 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
majianpeng | a42f9d8 | 2012-04-02 01:04:19 +1000 | [diff] [blame] | 2606 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2607 | ok = rdev_set_badblocks(rdev, sector_nr, |
| 2608 | min_bad, 0 |
| 2609 | ) && ok; |
| 2610 | } |
| 2611 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
| 2612 | *skipped = 1; |
| 2613 | put_buf(r1_bio); |
| 2614 | |
| 2615 | if (!ok) { |
| 2616 | /* Cannot record the badblocks, so need to |
| 2617 | * abort the resync. |
| 2618 | * If there are multiple read targets, could just |
| 2619 | * fail the really bad ones ??? |
| 2620 | */ |
| 2621 | conf->recovery_disabled = mddev->recovery_disabled; |
| 2622 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
| 2623 | return 0; |
| 2624 | } else |
| 2625 | return min_bad; |
| 2626 | |
| 2627 | } |
| 2628 | if (min_bad > 0 && min_bad < good_sectors) { |
| 2629 | /* only resync enough to reach the next bad->good |
| 2630 | * transition */ |
| 2631 | good_sectors = min_bad; |
| 2632 | } |
| 2633 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 2634 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
| 2635 | /* extra read targets are also write targets */ |
| 2636 | write_targets += read_targets-1; |
| 2637 | |
| 2638 | if (write_targets == 0 || read_targets == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2639 | /* There is nowhere to write, so all non-sync |
| 2640 | * drives must be failed - so we are finished |
| 2641 | */ |
NeilBrown | b7219cc | 2012-07-31 10:05:34 +1000 | [diff] [blame] | 2642 | sector_t rv; |
| 2643 | if (min_bad > 0) |
| 2644 | max_sector = sector_nr + min_bad; |
| 2645 | rv = max_sector - sector_nr; |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 2646 | *skipped = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2647 | put_buf(r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2648 | return rv; |
| 2649 | } |
| 2650 | |
NeilBrown | c620727 | 2008-02-06 01:39:52 -0800 | [diff] [blame] | 2651 | if (max_sector > mddev->resync_max) |
| 2652 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ |
NeilBrown | 06f6038 | 2011-07-28 11:31:48 +1000 | [diff] [blame] | 2653 | if (max_sector > sector_nr + good_sectors) |
| 2654 | max_sector = sector_nr + good_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2655 | nr_sectors = 0; |
NeilBrown | 289e99e | 2005-06-21 17:17:24 -0700 | [diff] [blame] | 2656 | sync_blocks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2657 | do { |
| 2658 | struct page *page; |
| 2659 | int len = PAGE_SIZE; |
| 2660 | if (sector_nr + (len>>9) > max_sector) |
| 2661 | len = (max_sector - sector_nr) << 9; |
| 2662 | if (len == 0) |
| 2663 | break; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2664 | if (sync_blocks == 0) { |
| 2665 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
NeilBrown | e5de485f | 2005-11-08 21:39:38 -0800 | [diff] [blame] | 2666 | &sync_blocks, still_degraded) && |
| 2667 | !conf->fullsync && |
| 2668 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2669 | break; |
Eric Sesterhenn | 9e77c48 | 2006-04-01 01:08:49 +0200 | [diff] [blame] | 2670 | BUG_ON(sync_blocks < (PAGE_SIZE>>9)); |
NeilBrown | 7571ae8 | 2010-10-07 11:54:46 +1100 | [diff] [blame] | 2671 | if ((len >> 9) > sync_blocks) |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2672 | len = sync_blocks<<9; |
NeilBrown | ab7a30c | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2673 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2674 | |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2675 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2676 | bio = r1_bio->bios[i]; |
| 2677 | if (bio->bi_end_io) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2678 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2679 | if (bio_add_page(bio, page, len, 0) == 0) { |
| 2680 | /* stop here */ |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2681 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2682 | while (i > 0) { |
| 2683 | i--; |
| 2684 | bio = r1_bio->bios[i]; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 2685 | if (bio->bi_end_io==NULL) |
| 2686 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2687 | /* remove last page from this bio */ |
| 2688 | bio->bi_vcnt--; |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2689 | bio->bi_iter.bi_size -= len; |
Jens Axboe | b7c44ed | 2015-07-24 12:37:59 -0600 | [diff] [blame] | 2690 | bio_clear_flag(bio, BIO_SEG_VALID); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2691 | } |
| 2692 | goto bio_full; |
| 2693 | } |
| 2694 | } |
| 2695 | } |
| 2696 | nr_sectors += len>>9; |
| 2697 | sector_nr += len>>9; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2698 | sync_blocks -= (len>>9); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2699 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
| 2700 | bio_full: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2701 | r1_bio->sectors = nr_sectors; |
| 2702 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2703 | /* For a user-requested sync, we read all readable devices and do a |
| 2704 | * compare |
| 2705 | */ |
| 2706 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
| 2707 | atomic_set(&r1_bio->remaining, read_targets); |
NeilBrown | 2d4f4f3 | 2012-07-09 11:34:13 +1000 | [diff] [blame] | 2708 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2709 | bio = r1_bio->bios[i]; |
| 2710 | if (bio->bi_end_io == end_sync_read) { |
NeilBrown | 2d4f4f3 | 2012-07-09 11:34:13 +1000 | [diff] [blame] | 2711 | read_targets--; |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 2712 | md_sync_acct(bio->bi_bdev, nr_sectors); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2713 | generic_make_request(bio); |
| 2714 | } |
| 2715 | } |
| 2716 | } else { |
| 2717 | atomic_set(&r1_bio->remaining, 1); |
| 2718 | bio = r1_bio->bios[r1_bio->read_disk]; |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 2719 | md_sync_acct(bio->bi_bdev, nr_sectors); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2720 | generic_make_request(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2721 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 2722 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2723 | return nr_sectors; |
| 2724 | } |
| 2725 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2726 | 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] | 2727 | { |
| 2728 | if (sectors) |
| 2729 | return sectors; |
| 2730 | |
| 2731 | return mddev->dev_sectors; |
| 2732 | } |
| 2733 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2734 | static struct r1conf *setup_conf(struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2735 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2736 | struct r1conf *conf; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2737 | int i; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 2738 | struct raid1_info *disk; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2739 | struct md_rdev *rdev; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2740 | int err = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2741 | |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2742 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2743 | if (!conf) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2744 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2745 | |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 2746 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2747 | * mddev->raid_disks * 2, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2748 | GFP_KERNEL); |
| 2749 | if (!conf->mirrors) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2750 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2751 | |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 2752 | conf->tmppage = alloc_page(GFP_KERNEL); |
| 2753 | if (!conf->tmppage) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2754 | goto abort; |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 2755 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2756 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2757 | if (!conf->poolinfo) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2758 | goto abort; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2759 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2760 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
| 2761 | r1bio_pool_free, |
| 2762 | conf->poolinfo); |
| 2763 | if (!conf->r1bio_pool) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2764 | goto abort; |
| 2765 | |
NeilBrown | ed9bfdf | 2009-10-16 15:55:44 +1100 | [diff] [blame] | 2766 | conf->poolinfo->mddev = mddev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2767 | |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2768 | err = -EINVAL; |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 2769 | spin_lock_init(&conf->device_lock); |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 2770 | rdev_for_each(rdev, mddev) { |
NeilBrown | aba336b | 2012-05-31 15:39:11 +1000 | [diff] [blame] | 2771 | struct request_queue *q; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2772 | int disk_idx = rdev->raid_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2773 | if (disk_idx >= mddev->raid_disks |
| 2774 | || disk_idx < 0) |
| 2775 | continue; |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2776 | if (test_bit(Replacement, &rdev->flags)) |
NeilBrown | 02b898f | 2012-10-31 11:42:03 +1100 | [diff] [blame] | 2777 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2778 | else |
| 2779 | disk = conf->mirrors + disk_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2780 | |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2781 | if (disk->rdev) |
| 2782 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2783 | disk->rdev = rdev; |
NeilBrown | aba336b | 2012-05-31 15:39:11 +1000 | [diff] [blame] | 2784 | q = bdev_get_queue(rdev->bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2785 | |
| 2786 | disk->head_position = 0; |
Shaohua Li | 12cee5a | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 2787 | disk->seq_start = MaxSector; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2788 | } |
| 2789 | conf->raid_disks = mddev->raid_disks; |
| 2790 | conf->mddev = mddev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2791 | INIT_LIST_HEAD(&conf->retry_list); |
NeilBrown | 55ce74d | 2015-08-14 11:11:10 +1000 | [diff] [blame] | 2792 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2793 | |
| 2794 | spin_lock_init(&conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2795 | init_waitqueue_head(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2796 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2797 | bio_list_init(&conf->pending_bio_list); |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 2798 | conf->pending_count = 0; |
NeilBrown | d890fa2 | 2011-10-26 11:54:39 +1100 | [diff] [blame] | 2799 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2800 | |
majianpeng | 79ef3a8 | 2013-11-15 14:55:02 +0800 | [diff] [blame] | 2801 | conf->start_next_window = MaxSector; |
| 2802 | conf->current_window_requests = conf->next_window_requests = 0; |
| 2803 | |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2804 | err = -EIO; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 2805 | for (i = 0; i < conf->raid_disks * 2; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2806 | |
| 2807 | disk = conf->mirrors + i; |
| 2808 | |
NeilBrown | c19d579 | 2011-12-23 10:17:57 +1100 | [diff] [blame] | 2809 | if (i < conf->raid_disks && |
| 2810 | disk[conf->raid_disks].rdev) { |
| 2811 | /* This slot has a replacement. */ |
| 2812 | if (!disk->rdev) { |
| 2813 | /* No original, just make the replacement |
| 2814 | * a recovering spare |
| 2815 | */ |
| 2816 | disk->rdev = |
| 2817 | disk[conf->raid_disks].rdev; |
| 2818 | disk[conf->raid_disks].rdev = NULL; |
| 2819 | } else if (!test_bit(In_sync, &disk->rdev->flags)) |
| 2820 | /* Original is not in_sync - bad */ |
| 2821 | goto abort; |
| 2822 | } |
| 2823 | |
NeilBrown | 5fd6c1d | 2006-06-26 00:27:40 -0700 | [diff] [blame] | 2824 | if (!disk->rdev || |
| 2825 | !test_bit(In_sync, &disk->rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2826 | disk->head_position = 0; |
Jonathan Brassow | 4f0a5e0 | 2012-05-22 13:55:31 +1000 | [diff] [blame] | 2827 | if (disk->rdev && |
| 2828 | (disk->rdev->saved_raid_disk < 0)) |
NeilBrown | 918f023 | 2007-08-22 14:01:52 -0700 | [diff] [blame] | 2829 | conf->fullsync = 1; |
Shaohua Li | be4d328 | 2012-07-31 10:03:53 +1000 | [diff] [blame] | 2830 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2831 | } |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2832 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2833 | err = -ENOMEM; |
NeilBrown | 0232605 | 2012-07-03 15:56:52 +1000 | [diff] [blame] | 2834 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2835 | if (!conf->thread) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2836 | printk(KERN_ERR |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2837 | "md/raid1:%s: couldn't allocate thread\n", |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2838 | mdname(mddev)); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2839 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2840 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 2841 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2842 | return conf; |
| 2843 | |
| 2844 | abort: |
| 2845 | if (conf) { |
| 2846 | if (conf->r1bio_pool) |
| 2847 | mempool_destroy(conf->r1bio_pool); |
| 2848 | kfree(conf->mirrors); |
| 2849 | safe_put_page(conf->tmppage); |
| 2850 | kfree(conf->poolinfo); |
| 2851 | kfree(conf); |
| 2852 | } |
| 2853 | return ERR_PTR(err); |
| 2854 | } |
| 2855 | |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2856 | static void raid1_free(struct mddev *mddev, void *priv); |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2857 | static int run(struct mddev *mddev) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2858 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 2859 | struct r1conf *conf; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2860 | int i; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 2861 | struct md_rdev *rdev; |
majianpeng | 5220ea1 | 2012-04-02 09:48:38 +1000 | [diff] [blame] | 2862 | int ret; |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 2863 | bool discard_supported = false; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2864 | |
| 2865 | if (mddev->level != 1) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2866 | 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] | 2867 | mdname(mddev), mddev->level); |
| 2868 | return -EIO; |
| 2869 | } |
| 2870 | if (mddev->reshape_position != MaxSector) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2871 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2872 | mdname(mddev)); |
| 2873 | return -EIO; |
| 2874 | } |
| 2875 | /* |
| 2876 | * copy the already verified devices into our private RAID1 |
| 2877 | * bookkeeping area. [whatever we allocate in run(), |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2878 | * should be freed in raid1_free()] |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2879 | */ |
| 2880 | if (mddev->private == NULL) |
| 2881 | conf = setup_conf(mddev); |
| 2882 | else |
| 2883 | conf = mddev->private; |
| 2884 | |
| 2885 | if (IS_ERR(conf)) |
| 2886 | return PTR_ERR(conf); |
| 2887 | |
Joe Lawrence | c8dc9c6 | 2013-02-21 13:28:09 +1100 | [diff] [blame] | 2888 | if (mddev->queue) |
H. Peter Anvin | 5026d7a | 2013-06-12 07:37:43 -0700 | [diff] [blame] | 2889 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
| 2890 | |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 2891 | rdev_for_each(rdev, mddev) { |
Jonathan Brassow | 1ed7242 | 2011-06-07 17:50:35 -0500 | [diff] [blame] | 2892 | if (!mddev->gendisk) |
| 2893 | continue; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2894 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 2895 | rdev->data_offset << 9); |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 2896 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
| 2897 | discard_supported = true; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2898 | } |
| 2899 | |
| 2900 | mddev->degraded = 0; |
| 2901 | for (i=0; i < conf->raid_disks; i++) |
| 2902 | if (conf->mirrors[i].rdev == NULL || |
| 2903 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || |
| 2904 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) |
| 2905 | mddev->degraded++; |
| 2906 | |
| 2907 | if (conf->raid_disks - mddev->degraded == 1) |
| 2908 | mddev->recovery_cp = MaxSector; |
| 2909 | |
Andre Noll | 8c6ac86 | 2009-06-18 08:48:06 +1000 | [diff] [blame] | 2910 | if (mddev->recovery_cp != MaxSector) |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2911 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
Andre Noll | 8c6ac86 | 2009-06-18 08:48:06 +1000 | [diff] [blame] | 2912 | " -- starting background reconstruction\n", |
| 2913 | mdname(mddev)); |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 2914 | printk(KERN_INFO |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2915 | "md/raid1:%s: active with %d out of %d mirrors\n", |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 2916 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2917 | mddev->raid_disks); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2918 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2919 | /* |
| 2920 | * Ok, everything is just fine now |
| 2921 | */ |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2922 | mddev->thread = conf->thread; |
| 2923 | conf->thread = NULL; |
| 2924 | mddev->private = conf; |
| 2925 | |
Dan Williams | 1f40362 | 2009-03-31 14:59:03 +1100 | [diff] [blame] | 2926 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2927 | |
Jonathan Brassow | 1ed7242 | 2011-06-07 17:50:35 -0500 | [diff] [blame] | 2928 | if (mddev->queue) { |
Shaohua Li | 2ff8cc2 | 2012-10-11 13:28:54 +1100 | [diff] [blame] | 2929 | if (discard_supported) |
| 2930 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, |
| 2931 | mddev->queue); |
| 2932 | else |
| 2933 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, |
| 2934 | mddev->queue); |
Jonathan Brassow | 1ed7242 | 2011-06-07 17:50:35 -0500 | [diff] [blame] | 2935 | } |
majianpeng | 5220ea1 | 2012-04-02 09:48:38 +1000 | [diff] [blame] | 2936 | |
| 2937 | ret = md_integrity_register(mddev); |
NeilBrown | 5aa61f4 | 2014-12-15 12:56:57 +1100 | [diff] [blame] | 2938 | if (ret) { |
| 2939 | md_unregister_thread(&mddev->thread); |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2940 | raid1_free(mddev, conf); |
NeilBrown | 5aa61f4 | 2014-12-15 12:56:57 +1100 | [diff] [blame] | 2941 | } |
majianpeng | 5220ea1 | 2012-04-02 09:48:38 +1000 | [diff] [blame] | 2942 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2943 | } |
| 2944 | |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2945 | static void raid1_free(struct mddev *mddev, void *priv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2946 | { |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 2947 | struct r1conf *conf = priv; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 2948 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2949 | if (conf->r1bio_pool) |
| 2950 | mempool_destroy(conf->r1bio_pool); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 2951 | kfree(conf->mirrors); |
Hirokazu Takahashi | 0fea7ed | 2013-04-24 11:42:44 +1000 | [diff] [blame] | 2952 | safe_put_page(conf->tmppage); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 2953 | kfree(conf->poolinfo); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2954 | kfree(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2955 | } |
| 2956 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2957 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2958 | { |
| 2959 | /* no resync is happening, and there is enough space |
| 2960 | * on all devices, so we can resize. |
| 2961 | * We need to make sure resync covers any new space. |
| 2962 | * If the array is shrinking we should possibly wait until |
| 2963 | * any io in the removed space completes, but it hardly seems |
| 2964 | * worth it. |
| 2965 | */ |
NeilBrown | a4a6125 | 2012-05-22 13:55:27 +1000 | [diff] [blame] | 2966 | sector_t newsize = raid1_size(mddev, sectors, 0); |
| 2967 | if (mddev->external_size && |
| 2968 | mddev->array_sectors > newsize) |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 2969 | return -EINVAL; |
NeilBrown | a4a6125 | 2012-05-22 13:55:27 +1000 | [diff] [blame] | 2970 | if (mddev->bitmap) { |
| 2971 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); |
| 2972 | if (ret) |
| 2973 | return ret; |
| 2974 | } |
| 2975 | md_set_array_sectors(mddev, newsize); |
Andre Noll | f233ea5 | 2008-07-21 17:05:22 +1000 | [diff] [blame] | 2976 | set_capacity(mddev->gendisk, mddev->array_sectors); |
NeilBrown | 449aad3 | 2009-08-03 10:59:58 +1000 | [diff] [blame] | 2977 | revalidate_disk(mddev->gendisk); |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 2978 | if (sectors > mddev->dev_sectors && |
NeilBrown | b098636 | 2011-05-11 15:52:21 +1000 | [diff] [blame] | 2979 | mddev->recovery_cp > mddev->dev_sectors) { |
Andre Noll | 58c0fed | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 2980 | mddev->recovery_cp = mddev->dev_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2981 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
| 2982 | } |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 2983 | mddev->dev_sectors = sectors; |
NeilBrown | 4b5c7ae | 2005-07-27 11:43:28 -0700 | [diff] [blame] | 2984 | mddev->resync_max_sectors = sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2985 | return 0; |
| 2986 | } |
| 2987 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 2988 | static int raid1_reshape(struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2989 | { |
| 2990 | /* We need to: |
| 2991 | * 1/ resize the r1bio_pool |
| 2992 | * 2/ resize conf->mirrors |
| 2993 | * |
| 2994 | * We allocate a new r1bio_pool if we can. |
| 2995 | * Then raise a device barrier and wait until all IO stops. |
| 2996 | * Then resize conf->mirrors and swap in the new r1bio pool. |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2997 | * |
| 2998 | * At the same time, we "pack" the devices so that all the missing |
| 2999 | * devices have the higher raid_disk numbers. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3000 | */ |
| 3001 | mempool_t *newpool, *oldpool; |
| 3002 | struct pool_info *newpoolinfo; |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 3003 | struct raid1_info *newmirrors; |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 3004 | struct r1conf *conf = mddev->private; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3005 | int cnt, raid_disks; |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 3006 | unsigned long flags; |
Dan Williams | b5470dc | 2008-06-27 21:44:04 -0700 | [diff] [blame] | 3007 | int d, d2, err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3008 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3009 | /* Cannot change chunk_size, layout, or level */ |
Andre Noll | 664e7c4 | 2009-06-18 08:45:27 +1000 | [diff] [blame] | 3010 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3011 | mddev->layout != mddev->new_layout || |
| 3012 | mddev->level != mddev->new_level) { |
Andre Noll | 664e7c4 | 2009-06-18 08:45:27 +1000 | [diff] [blame] | 3013 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3014 | mddev->new_layout = mddev->layout; |
| 3015 | mddev->new_level = mddev->level; |
| 3016 | return -EINVAL; |
| 3017 | } |
| 3018 | |
Dan Williams | b5470dc | 2008-06-27 21:44:04 -0700 | [diff] [blame] | 3019 | err = md_allow_write(mddev); |
| 3020 | if (err) |
| 3021 | return err; |
NeilBrown | 2a2275d | 2007-01-26 00:57:11 -0800 | [diff] [blame] | 3022 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3023 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
| 3024 | |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3025 | if (raid_disks < conf->raid_disks) { |
| 3026 | cnt=0; |
| 3027 | for (d= 0; d < conf->raid_disks; d++) |
| 3028 | if (conf->mirrors[d].rdev) |
| 3029 | cnt++; |
| 3030 | if (cnt > raid_disks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3031 | return -EBUSY; |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3032 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3033 | |
| 3034 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); |
| 3035 | if (!newpoolinfo) |
| 3036 | return -ENOMEM; |
| 3037 | newpoolinfo->mddev = mddev; |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 3038 | newpoolinfo->raid_disks = raid_disks * 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3039 | |
| 3040 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
| 3041 | r1bio_pool_free, newpoolinfo); |
| 3042 | if (!newpool) { |
| 3043 | kfree(newpoolinfo); |
| 3044 | return -ENOMEM; |
| 3045 | } |
Jonathan Brassow | 0eaf822 | 2012-07-31 10:03:52 +1000 | [diff] [blame] | 3046 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
NeilBrown | 8f19ccb | 2011-12-23 10:17:56 +1100 | [diff] [blame] | 3047 | GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3048 | if (!newmirrors) { |
| 3049 | kfree(newpoolinfo); |
| 3050 | mempool_destroy(newpool); |
| 3051 | return -ENOMEM; |
| 3052 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3053 | |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 3054 | freeze_array(conf, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3055 | |
| 3056 | /* ok, everything is stopped */ |
| 3057 | oldpool = conf->r1bio_pool; |
| 3058 | conf->r1bio_pool = newpool; |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3059 | |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3060 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 3061 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3062 | if (rdev && rdev->raid_disk != d2) { |
Namhyung Kim | 36fad85 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 3063 | sysfs_unlink_rdev(mddev, rdev); |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3064 | rdev->raid_disk = d2; |
Namhyung Kim | 36fad85 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 3065 | sysfs_unlink_rdev(mddev, rdev); |
| 3066 | if (sysfs_link_rdev(mddev, rdev)) |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3067 | printk(KERN_WARNING |
Namhyung Kim | 36fad85 | 2011-07-27 11:00:36 +1000 | [diff] [blame] | 3068 | "md/raid1:%s: cannot register rd%d\n", |
| 3069 | mdname(mddev), rdev->raid_disk); |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 3070 | } |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 3071 | if (rdev) |
| 3072 | newmirrors[d2++].rdev = rdev; |
| 3073 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3074 | kfree(conf->mirrors); |
| 3075 | conf->mirrors = newmirrors; |
| 3076 | kfree(conf->poolinfo); |
| 3077 | conf->poolinfo = newpoolinfo; |
| 3078 | |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 3079 | spin_lock_irqsave(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3080 | mddev->degraded += (raid_disks - conf->raid_disks); |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 3081 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | conf->raid_disks = mddev->raid_disks = raid_disks; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3083 | mddev->delta_disks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3084 | |
NeilBrown | e2d5992 | 2013-06-12 11:01:22 +1000 | [diff] [blame] | 3085 | unfreeze_array(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3086 | |
NeilBrown | 985ca97 | 2015-07-06 12:26:57 +1000 | [diff] [blame] | 3087 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3088 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
| 3089 | md_wakeup_thread(mddev->thread); |
| 3090 | |
| 3091 | mempool_destroy(oldpool); |
| 3092 | return 0; |
| 3093 | } |
| 3094 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 3095 | static void raid1_quiesce(struct mddev *mddev, int state) |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3096 | { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 3097 | struct r1conf *conf = mddev->private; |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3098 | |
| 3099 | switch(state) { |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3100 | case 2: /* wake for suspend */ |
| 3101 | wake_up(&conf->wait_barrier); |
| 3102 | break; |
NeilBrown | 9e6603d | 2005-09-09 16:23:48 -0700 | [diff] [blame] | 3103 | case 1: |
majianpeng | 07169fd4 | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 3104 | freeze_array(conf, 0); |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3105 | break; |
NeilBrown | 9e6603d | 2005-09-09 16:23:48 -0700 | [diff] [blame] | 3106 | case 0: |
majianpeng | 07169fd4 | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 3107 | unfreeze_array(conf); |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3108 | break; |
| 3109 | } |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3110 | } |
| 3111 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 3112 | static void *raid1_takeover(struct mddev *mddev) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3113 | { |
| 3114 | /* raid1 can take over: |
| 3115 | * raid5 with 2 devices, any layout or chunk size |
| 3116 | */ |
| 3117 | if (mddev->level == 5 && mddev->raid_disks == 2) { |
NeilBrown | e809636 | 2011-10-11 16:49:05 +1100 | [diff] [blame] | 3118 | struct r1conf *conf; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3119 | mddev->new_level = 1; |
| 3120 | mddev->new_layout = 0; |
| 3121 | mddev->new_chunk_sectors = 0; |
| 3122 | conf = setup_conf(mddev); |
| 3123 | if (!IS_ERR(conf)) |
majianpeng | 07169fd4 | 2013-11-14 15:16:18 +1100 | [diff] [blame] | 3124 | /* Array must appear to be quiesced */ |
| 3125 | conf->array_frozen = 1; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3126 | return conf; |
| 3127 | } |
| 3128 | return ERR_PTR(-EINVAL); |
| 3129 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3130 | |
NeilBrown | 84fc4b5 | 2011-10-11 16:49:58 +1100 | [diff] [blame] | 3131 | static struct md_personality raid1_personality = |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3132 | { |
| 3133 | .name = "raid1", |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 3134 | .level = 1, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3135 | .owner = THIS_MODULE, |
| 3136 | .make_request = make_request, |
| 3137 | .run = run, |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 3138 | .free = raid1_free, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3139 | .status = status, |
| 3140 | .error_handler = error, |
| 3141 | .hot_add_disk = raid1_add_disk, |
| 3142 | .hot_remove_disk= raid1_remove_disk, |
| 3143 | .spare_active = raid1_spare_active, |
| 3144 | .sync_request = sync_request, |
| 3145 | .resize = raid1_resize, |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 3146 | .size = raid1_size, |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 3147 | .check_reshape = raid1_reshape, |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 3148 | .quiesce = raid1_quiesce, |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 3149 | .takeover = raid1_takeover, |
NeilBrown | 5c675f8 | 2014-12-15 12:56:56 +1100 | [diff] [blame] | 3150 | .congested = raid1_congested, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3151 | }; |
| 3152 | |
| 3153 | static int __init raid_init(void) |
| 3154 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 3155 | return register_md_personality(&raid1_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3156 | } |
| 3157 | |
| 3158 | static void raid_exit(void) |
| 3159 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 3160 | unregister_md_personality(&raid1_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3161 | } |
| 3162 | |
| 3163 | module_init(raid_init); |
| 3164 | module_exit(raid_exit); |
| 3165 | MODULE_LICENSE("GPL"); |
NeilBrown | 0efb9e6 | 2009-12-14 12:49:58 +1100 | [diff] [blame] | 3166 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3167 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
NeilBrown | d9d166c | 2006-01-06 00:20:51 -0800 | [diff] [blame] | 3168 | MODULE_ALIAS("md-raid1"); |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 3169 | MODULE_ALIAS("md-level-1"); |
NeilBrown | 34db0cd | 2011-10-11 16:50:01 +1100 | [diff] [blame] | 3170 | |
| 3171 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |