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> |
NeilBrown | bff6197 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 37 | #include <linux/seq_file.h> |
NeilBrown | 43b2e5d | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 38 | #include "md.h" |
Christoph Hellwig | ef740c3 | 2009-03-31 14:27:03 +1100 | [diff] [blame] | 39 | #include "raid1.h" |
| 40 | #include "bitmap.h" |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 41 | |
| 42 | #define DEBUG 0 |
| 43 | #if DEBUG |
| 44 | #define PRINTK(x...) printk(x) |
| 45 | #else |
| 46 | #define PRINTK(x...) |
| 47 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | |
| 49 | /* |
| 50 | * Number of guaranteed r1bios in case of extreme VM load: |
| 51 | */ |
| 52 | #define NR_RAID1_BIOS 256 |
| 53 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | |
| 55 | static void unplug_slaves(mddev_t *mddev); |
| 56 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 57 | static void allow_barrier(conf_t *conf); |
| 58 | static void lower_barrier(conf_t *conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 60 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | { |
| 62 | struct pool_info *pi = data; |
| 63 | r1bio_t *r1_bio; |
| 64 | int size = offsetof(r1bio_t, bios[pi->raid_disks]); |
| 65 | |
| 66 | /* allocate a r1bio with room for raid_disks entries in the bios array */ |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 67 | r1_bio = kzalloc(size, gfp_flags); |
NeilBrown | ed9bfdf | 2009-10-16 15:55:44 +1100 | [diff] [blame] | 68 | if (!r1_bio && pi->mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | unplug_slaves(pi->mddev); |
| 70 | |
| 71 | return r1_bio; |
| 72 | } |
| 73 | |
| 74 | static void r1bio_pool_free(void *r1_bio, void *data) |
| 75 | { |
| 76 | kfree(r1_bio); |
| 77 | } |
| 78 | |
| 79 | #define RESYNC_BLOCK_SIZE (64*1024) |
| 80 | //#define RESYNC_BLOCK_SIZE PAGE_SIZE |
| 81 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
| 82 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
| 83 | #define RESYNC_WINDOW (2048*1024) |
| 84 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 85 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | { |
| 87 | struct pool_info *pi = data; |
| 88 | struct page *page; |
| 89 | r1bio_t *r1_bio; |
| 90 | struct bio *bio; |
| 91 | int i, j; |
| 92 | |
| 93 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); |
| 94 | if (!r1_bio) { |
| 95 | unplug_slaves(pi->mddev); |
| 96 | return NULL; |
| 97 | } |
| 98 | |
| 99 | /* |
| 100 | * Allocate bios : 1 for reading, n-1 for writing |
| 101 | */ |
| 102 | for (j = pi->raid_disks ; j-- ; ) { |
NeilBrown | 6746557 | 2010-10-26 17:33:54 +1100 | [diff] [blame] | 103 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | if (!bio) |
| 105 | goto out_free_bio; |
| 106 | r1_bio->bios[j] = bio; |
| 107 | } |
| 108 | /* |
| 109 | * Allocate RESYNC_PAGES data pages and attach them to |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 110 | * the first bio. |
| 111 | * If this is a user-requested check/repair, allocate |
| 112 | * RESYNC_PAGES for each bio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | */ |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 114 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
| 115 | j = pi->raid_disks; |
| 116 | else |
| 117 | j = 1; |
| 118 | while(j--) { |
| 119 | bio = r1_bio->bios[j]; |
| 120 | for (i = 0; i < RESYNC_PAGES; i++) { |
| 121 | page = alloc_page(gfp_flags); |
| 122 | if (unlikely(!page)) |
| 123 | goto out_free_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 125 | bio->bi_io_vec[i].bv_page = page; |
NeilBrown | 303a0e1 | 2009-04-06 14:40:38 +1000 | [diff] [blame] | 126 | bio->bi_vcnt = i+1; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 127 | } |
| 128 | } |
| 129 | /* If not user-requests, copy the page pointers to all bios */ |
| 130 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { |
| 131 | for (i=0; i<RESYNC_PAGES ; i++) |
| 132 | for (j=1; j<pi->raid_disks; j++) |
| 133 | r1_bio->bios[j]->bi_io_vec[i].bv_page = |
| 134 | r1_bio->bios[0]->bi_io_vec[i].bv_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | } |
| 136 | |
| 137 | r1_bio->master_bio = NULL; |
| 138 | |
| 139 | return r1_bio; |
| 140 | |
| 141 | out_free_pages: |
NeilBrown | 303a0e1 | 2009-04-06 14:40:38 +1000 | [diff] [blame] | 142 | for (j=0 ; j < pi->raid_disks; j++) |
| 143 | for (i=0; i < r1_bio->bios[j]->bi_vcnt ; i++) |
| 144 | put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 145 | j = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | out_free_bio: |
| 147 | while ( ++j < pi->raid_disks ) |
| 148 | bio_put(r1_bio->bios[j]); |
| 149 | r1bio_pool_free(r1_bio, data); |
| 150 | return NULL; |
| 151 | } |
| 152 | |
| 153 | static void r1buf_pool_free(void *__r1_bio, void *data) |
| 154 | { |
| 155 | struct pool_info *pi = data; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 156 | int i,j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | r1bio_t *r1bio = __r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 159 | for (i = 0; i < RESYNC_PAGES; i++) |
| 160 | for (j = pi->raid_disks; j-- ;) { |
| 161 | if (j == 0 || |
| 162 | r1bio->bios[j]->bi_io_vec[i].bv_page != |
| 163 | r1bio->bios[0]->bi_io_vec[i].bv_page) |
NeilBrown | 1345b1d | 2006-01-06 00:20:40 -0800 | [diff] [blame] | 164 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 165 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | for (i=0 ; i < pi->raid_disks; i++) |
| 167 | bio_put(r1bio->bios[i]); |
| 168 | |
| 169 | r1bio_pool_free(r1bio, data); |
| 170 | } |
| 171 | |
| 172 | static void put_all_bios(conf_t *conf, r1bio_t *r1_bio) |
| 173 | { |
| 174 | int i; |
| 175 | |
| 176 | for (i = 0; i < conf->raid_disks; i++) { |
| 177 | struct bio **bio = r1_bio->bios + i; |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 178 | if (*bio && *bio != IO_BLOCKED) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | bio_put(*bio); |
| 180 | *bio = NULL; |
| 181 | } |
| 182 | } |
| 183 | |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 184 | static void free_r1bio(r1bio_t *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 186 | conf_t *conf = r1_bio->mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | |
| 188 | /* |
| 189 | * Wake up any possible resync thread that waits for the device |
| 190 | * to go idle. |
| 191 | */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 192 | allow_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | |
| 194 | put_all_bios(conf, r1_bio); |
| 195 | mempool_free(r1_bio, conf->r1bio_pool); |
| 196 | } |
| 197 | |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 198 | static void put_buf(r1bio_t *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 200 | conf_t *conf = r1_bio->mddev->private; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 201 | int i; |
| 202 | |
| 203 | for (i=0; i<conf->raid_disks; i++) { |
| 204 | struct bio *bio = r1_bio->bios[i]; |
| 205 | if (bio->bi_end_io) |
| 206 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); |
| 207 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | |
| 209 | mempool_free(r1_bio, conf->r1buf_pool); |
| 210 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 211 | lower_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | } |
| 213 | |
| 214 | static void reschedule_retry(r1bio_t *r1_bio) |
| 215 | { |
| 216 | unsigned long flags; |
| 217 | mddev_t *mddev = r1_bio->mddev; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 218 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | |
| 220 | spin_lock_irqsave(&conf->device_lock, flags); |
| 221 | list_add(&r1_bio->retry_list, &conf->retry_list); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 222 | conf->nr_queued ++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 224 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 225 | wake_up(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | md_wakeup_thread(mddev->thread); |
| 227 | } |
| 228 | |
| 229 | /* |
| 230 | * raid_end_bio_io() is called when we have finished servicing a mirrored |
| 231 | * operation and are ready to return a success/failure code to the buffer |
| 232 | * cache layer. |
| 233 | */ |
| 234 | static void raid_end_bio_io(r1bio_t *r1_bio) |
| 235 | { |
| 236 | struct bio *bio = r1_bio->master_bio; |
| 237 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 238 | /* if nobody has done the final endio yet, do it now */ |
| 239 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { |
| 240 | PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n", |
| 241 | (bio_data_dir(bio) == WRITE) ? "write" : "read", |
| 242 | (unsigned long long) bio->bi_sector, |
| 243 | (unsigned long long) bio->bi_sector + |
| 244 | (bio->bi_size >> 9) - 1); |
| 245 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 246 | bio_endio(bio, |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 247 | test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO); |
| 248 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 249 | free_r1bio(r1_bio); |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * Update disk head position estimator based on IRQ completion info. |
| 254 | */ |
| 255 | static inline void update_head_pos(int disk, r1bio_t *r1_bio) |
| 256 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 257 | conf_t *conf = r1_bio->mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | |
| 259 | conf->mirrors[disk].head_position = |
| 260 | r1_bio->sector + (r1_bio->sectors); |
| 261 | } |
| 262 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 263 | static void raid1_end_read_request(struct bio *bio, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 264 | { |
| 265 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
H Hartley Sweeten | 7b92813 | 2010-03-08 16:02:40 +1100 | [diff] [blame] | 266 | r1bio_t *r1_bio = bio->bi_private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | int mirror; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 268 | conf_t *conf = r1_bio->mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | mirror = r1_bio->read_disk; |
| 271 | /* |
| 272 | * this branch is our 'one mirror IO has finished' event handler: |
| 273 | */ |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 274 | update_head_pos(mirror, r1_bio); |
| 275 | |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 276 | if (uptodate) |
| 277 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
| 278 | else { |
| 279 | /* If all other devices have failed, we want to return |
| 280 | * the error upwards rather than fail the last device. |
| 281 | * Here we redefine "uptodate" to mean "Don't want to retry" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | */ |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 283 | unsigned long flags; |
| 284 | spin_lock_irqsave(&conf->device_lock, flags); |
| 285 | if (r1_bio->mddev->degraded == conf->raid_disks || |
| 286 | (r1_bio->mddev->degraded == conf->raid_disks-1 && |
| 287 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))) |
| 288 | uptodate = 1; |
| 289 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 290 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 292 | if (uptodate) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | raid_end_bio_io(r1_bio); |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 294 | else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | /* |
| 296 | * oops, read error: |
| 297 | */ |
| 298 | char b[BDEVNAME_SIZE]; |
| 299 | if (printk_ratelimit()) |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 300 | printk(KERN_ERR "md/raid1:%s: %s: rescheduling sector %llu\n", |
| 301 | mdname(conf->mddev), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 302 | bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector); |
| 303 | reschedule_retry(r1_bio); |
| 304 | } |
| 305 | |
| 306 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | } |
| 308 | |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 309 | static void r1_bio_write_done(r1bio_t *r1_bio, int vcnt, struct bio_vec *bv, |
| 310 | int behind) |
| 311 | { |
| 312 | if (atomic_dec_and_test(&r1_bio->remaining)) |
| 313 | { |
| 314 | /* it really is the end of this request */ |
| 315 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
| 316 | /* free extra copy of the data pages */ |
| 317 | int i = vcnt; |
| 318 | while (i--) |
| 319 | safe_put_page(bv[i].bv_page); |
| 320 | } |
| 321 | /* clear the bitmap if all writes complete successfully */ |
| 322 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
| 323 | r1_bio->sectors, |
| 324 | !test_bit(R1BIO_Degraded, &r1_bio->state), |
| 325 | behind); |
| 326 | md_write_end(r1_bio->mddev); |
| 327 | raid_end_bio_io(r1_bio); |
| 328 | } |
| 329 | } |
| 330 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 331 | static void raid1_end_write_request(struct bio *bio, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | { |
| 333 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
H Hartley Sweeten | 7b92813 | 2010-03-08 16:02:40 +1100 | [diff] [blame] | 334 | r1bio_t *r1_bio = bio->bi_private; |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 335 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 336 | conf_t *conf = r1_bio->mddev->private; |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 337 | struct bio *to_put = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 338 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | |
| 340 | for (mirror = 0; mirror < conf->raid_disks; mirror++) |
| 341 | if (r1_bio->bios[mirror] == bio) |
| 342 | break; |
| 343 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 344 | /* |
| 345 | * 'one mirror IO has finished' event handler: |
| 346 | */ |
| 347 | r1_bio->bios[mirror] = NULL; |
| 348 | to_put = bio; |
| 349 | if (!uptodate) { |
| 350 | md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); |
| 351 | /* an I/O failed, we can't clear the bitmap */ |
| 352 | set_bit(R1BIO_Degraded, &r1_bio->state); |
| 353 | } else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | /* |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 355 | * Set R1BIO_Uptodate in our master bio, so that we |
| 356 | * will return a good error code for to the higher |
| 357 | * levels even if IO on some other mirrored buffer |
| 358 | * fails. |
| 359 | * |
| 360 | * The 'master' represents the composite IO operation |
| 361 | * to user-side. So if something waits for IO, then it |
| 362 | * will wait for the 'master' bio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 363 | */ |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 364 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 366 | update_head_pos(mirror, r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 368 | if (behind) { |
| 369 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) |
| 370 | atomic_dec(&r1_bio->behind_remaining); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 371 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 372 | /* |
| 373 | * In behind mode, we ACK the master bio once the I/O |
| 374 | * has safely reached all non-writemostly |
| 375 | * disks. Setting the Returned bit ensures that this |
| 376 | * gets done only once -- we don't ever want to return |
| 377 | * -EIO here, instead we'll wait |
| 378 | */ |
| 379 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && |
| 380 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
| 381 | /* Maybe we can return now */ |
| 382 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { |
| 383 | struct bio *mbio = r1_bio->master_bio; |
| 384 | PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n", |
| 385 | (unsigned long long) mbio->bi_sector, |
| 386 | (unsigned long long) mbio->bi_sector + |
| 387 | (mbio->bi_size >> 9) - 1); |
| 388 | bio_endio(mbio, 0); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 389 | } |
| 390 | } |
| 391 | } |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 392 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
| 393 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | * Let's see if all mirrored write operations have finished |
| 396 | * already. |
| 397 | */ |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 398 | r1_bio_write_done(r1_bio, bio->bi_vcnt, bio->bi_io_vec, behind); |
NeilBrown | c70810b | 2006-06-26 00:27:35 -0700 | [diff] [blame] | 399 | |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 400 | if (to_put) |
| 401 | bio_put(to_put); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | } |
| 403 | |
| 404 | |
| 405 | /* |
| 406 | * This routine returns the disk from which the requested read should |
| 407 | * be done. There is a per-array 'next expected sequential IO' sector |
| 408 | * number - if this matches on the next IO then we use the last disk. |
| 409 | * There is also a per-disk 'last know head position' sector that is |
| 410 | * maintained from IRQ contexts, both the normal and the resync IO |
| 411 | * completion handlers update this position correctly. If there is no |
| 412 | * perfect sequential match then we pick the disk whose head is closest. |
| 413 | * |
| 414 | * If there are 2 mirrors in the same 2 devices, performance degrades |
| 415 | * because position is mirror, not device based. |
| 416 | * |
| 417 | * The rdev for the device selected will have nr_pending incremented. |
| 418 | */ |
| 419 | static int read_balance(conf_t *conf, r1bio_t *r1_bio) |
| 420 | { |
NeilBrown | af3a2cd | 2010-05-08 08:20:17 +1000 | [diff] [blame] | 421 | const sector_t this_sector = r1_bio->sector; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | const int sectors = r1_bio->sectors; |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 423 | int new_disk = -1; |
| 424 | int start_disk; |
| 425 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | sector_t new_distance, current_distance; |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 427 | mdk_rdev_t *rdev; |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 428 | int choose_first; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 429 | |
| 430 | rcu_read_lock(); |
| 431 | /* |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 432 | * Check if we can balance. We can balance on the whole |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | * device if no resync is going on, or below the resync window. |
| 434 | * We take the first readable disk when above the resync window. |
| 435 | */ |
| 436 | retry: |
| 437 | if (conf->mddev->recovery_cp < MaxSector && |
| 438 | (this_sector + sectors >= conf->next_resync)) { |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 439 | choose_first = 1; |
| 440 | start_disk = 0; |
| 441 | } else { |
| 442 | choose_first = 0; |
| 443 | start_disk = conf->last_used; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | } |
| 445 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | /* make sure the disk is operational */ |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 447 | for (i = 0 ; i < conf->raid_disks ; i++) { |
| 448 | int disk = start_disk + i; |
| 449 | if (disk >= conf->raid_disks) |
| 450 | disk -= conf->raid_disks; |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 451 | |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 452 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
| 453 | if (r1_bio->bios[disk] == IO_BLOCKED |
| 454 | || rdev == NULL |
| 455 | || !test_bit(In_sync, &rdev->flags)) |
| 456 | continue; |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 457 | |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 458 | new_disk = disk; |
| 459 | if (!test_bit(WriteMostly, &rdev->flags)) |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 460 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 461 | } |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 462 | |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 463 | if (new_disk < 0 || choose_first) |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 464 | goto rb_out; |
| 465 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 466 | /* |
| 467 | * Don't change to another disk for sequential reads: |
| 468 | */ |
| 469 | if (conf->next_seq_sect == this_sector) |
| 470 | goto rb_out; |
| 471 | if (this_sector == conf->mirrors[new_disk].head_position) |
| 472 | goto rb_out; |
| 473 | |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 474 | current_distance = abs(this_sector |
| 475 | - conf->mirrors[new_disk].head_position); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 477 | /* look for a better disk - i.e. head is closer */ |
| 478 | start_disk = new_disk; |
| 479 | for (i = 1; i < conf->raid_disks; i++) { |
| 480 | int disk = start_disk + 1; |
| 481 | if (disk >= conf->raid_disks) |
| 482 | disk -= conf->raid_disks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 484 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 485 | if (r1_bio->bios[disk] == IO_BLOCKED |
| 486 | || rdev == NULL |
| 487 | || !test_bit(In_sync, &rdev->flags) |
| 488 | || test_bit(WriteMostly, &rdev->flags)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | continue; |
| 490 | |
| 491 | if (!atomic_read(&rdev->nr_pending)) { |
| 492 | new_disk = disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | break; |
| 494 | } |
| 495 | new_distance = abs(this_sector - conf->mirrors[disk].head_position); |
| 496 | if (new_distance < current_distance) { |
| 497 | current_distance = new_distance; |
| 498 | new_disk = disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 499 | } |
NeilBrown | f3ac8bf | 2010-09-06 14:10:08 +1000 | [diff] [blame] | 500 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 502 | rb_out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 503 | if (new_disk >= 0) { |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 504 | rdev = rcu_dereference(conf->mirrors[new_disk].rdev); |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 505 | if (!rdev) |
| 506 | goto retry; |
| 507 | atomic_inc(&rdev->nr_pending); |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 508 | if (!test_bit(In_sync, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | /* cannot risk returning a device that failed |
| 510 | * before we inc'ed nr_pending |
| 511 | */ |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 512 | rdev_dec_pending(rdev, conf->mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | goto retry; |
| 514 | } |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 515 | conf->next_seq_sect = this_sector + sectors; |
| 516 | conf->last_used = new_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | } |
| 518 | rcu_read_unlock(); |
| 519 | |
| 520 | return new_disk; |
| 521 | } |
| 522 | |
| 523 | static void unplug_slaves(mddev_t *mddev) |
| 524 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 525 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 526 | int i; |
| 527 | |
| 528 | rcu_read_lock(); |
| 529 | for (i=0; i<mddev->raid_disks; i++) { |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 530 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 531 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 532 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 533 | |
| 534 | atomic_inc(&rdev->nr_pending); |
| 535 | rcu_read_unlock(); |
| 536 | |
Alan D. Brunelle | 2ad8b1e | 2007-11-07 14:26:56 -0500 | [diff] [blame] | 537 | blk_unplug(r_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | |
| 539 | rdev_dec_pending(rdev, mddev); |
| 540 | rcu_read_lock(); |
| 541 | } |
| 542 | } |
| 543 | rcu_read_unlock(); |
| 544 | } |
| 545 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 546 | static void raid1_unplug(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 548 | mddev_t *mddev = q->queuedata; |
| 549 | |
| 550 | unplug_slaves(mddev); |
| 551 | md_wakeup_thread(mddev->thread); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | } |
| 553 | |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 554 | static int raid1_congested(void *data, int bits) |
| 555 | { |
| 556 | mddev_t *mddev = data; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 557 | conf_t *conf = mddev->private; |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 558 | int i, ret = 0; |
| 559 | |
NeilBrown | 3fa841d | 2009-09-23 18:10:29 +1000 | [diff] [blame] | 560 | if (mddev_congested(mddev, bits)) |
| 561 | return 1; |
| 562 | |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 563 | rcu_read_lock(); |
| 564 | for (i = 0; i < mddev->raid_disks; i++) { |
| 565 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
| 566 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 567 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 568 | |
| 569 | /* Note the '|| 1' - when read_balance prefers |
| 570 | * non-congested targets, it can be removed |
| 571 | */ |
Alexander Beregalov | 91a9e99 | 2009-04-07 01:35:56 +0400 | [diff] [blame] | 572 | if ((bits & (1<<BDI_async_congested)) || 1) |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 573 | ret |= bdi_congested(&q->backing_dev_info, bits); |
| 574 | else |
| 575 | ret &= bdi_congested(&q->backing_dev_info, bits); |
| 576 | } |
| 577 | } |
| 578 | rcu_read_unlock(); |
| 579 | return ret; |
| 580 | } |
| 581 | |
| 582 | |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 583 | static int flush_pending_writes(conf_t *conf) |
| 584 | { |
| 585 | /* Any writes that have been queued but are awaiting |
| 586 | * bitmap updates get flushed here. |
| 587 | * We return 1 if any requests were actually submitted. |
| 588 | */ |
| 589 | int rv = 0; |
| 590 | |
| 591 | spin_lock_irq(&conf->device_lock); |
| 592 | |
| 593 | if (conf->pending_bio_list.head) { |
| 594 | struct bio *bio; |
| 595 | bio = bio_list_get(&conf->pending_bio_list); |
| 596 | blk_remove_plug(conf->mddev->queue); |
| 597 | spin_unlock_irq(&conf->device_lock); |
| 598 | /* flush any pending bitmap writes to |
| 599 | * disk before proceeding w/ I/O */ |
| 600 | bitmap_unplug(conf->mddev->bitmap); |
| 601 | |
| 602 | while (bio) { /* submit pending writes */ |
| 603 | struct bio *next = bio->bi_next; |
| 604 | bio->bi_next = NULL; |
| 605 | generic_make_request(bio); |
| 606 | bio = next; |
| 607 | } |
| 608 | rv = 1; |
| 609 | } else |
| 610 | spin_unlock_irq(&conf->device_lock); |
| 611 | return rv; |
| 612 | } |
| 613 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 614 | /* Barriers.... |
| 615 | * Sometimes we need to suspend IO while we do something else, |
| 616 | * either some resync/recovery, or reconfigure the array. |
| 617 | * To do this we raise a 'barrier'. |
| 618 | * The 'barrier' is a counter that can be raised multiple times |
| 619 | * to count how many activities are happening which preclude |
| 620 | * normal IO. |
| 621 | * We can only raise the barrier if there is no pending IO. |
| 622 | * i.e. if nr_pending == 0. |
| 623 | * We choose only to raise the barrier if no-one is waiting for the |
| 624 | * barrier to go down. This means that as soon as an IO request |
| 625 | * is ready, no other operations which require a barrier will start |
| 626 | * until the IO request has had a chance. |
| 627 | * |
| 628 | * So: regular IO calls 'wait_barrier'. When that returns there |
| 629 | * is no backgroup IO happening, It must arrange to call |
| 630 | * allow_barrier when it has finished its IO. |
| 631 | * backgroup IO calls must call raise_barrier. Once that returns |
| 632 | * there is no normal IO happeing. It must arrange to call |
| 633 | * lower_barrier when the particular background IO completes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 634 | */ |
| 635 | #define RESYNC_DEPTH 32 |
| 636 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 637 | static void raise_barrier(conf_t *conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | { |
| 639 | spin_lock_irq(&conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 640 | |
| 641 | /* Wait until no block IO is waiting */ |
| 642 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, |
| 643 | conf->resync_lock, |
| 644 | raid1_unplug(conf->mddev->queue)); |
| 645 | |
| 646 | /* block any new IO from starting */ |
| 647 | conf->barrier++; |
| 648 | |
NeilBrown | 046abee | 2010-10-26 15:46:20 +1100 | [diff] [blame] | 649 | /* Now wait for all pending IO to complete */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 650 | wait_event_lock_irq(conf->wait_barrier, |
| 651 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, |
| 652 | conf->resync_lock, |
| 653 | raid1_unplug(conf->mddev->queue)); |
| 654 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | spin_unlock_irq(&conf->resync_lock); |
| 656 | } |
| 657 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 658 | static void lower_barrier(conf_t *conf) |
| 659 | { |
| 660 | unsigned long flags; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 661 | BUG_ON(conf->barrier <= 0); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 662 | spin_lock_irqsave(&conf->resync_lock, flags); |
| 663 | conf->barrier--; |
| 664 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
| 665 | wake_up(&conf->wait_barrier); |
| 666 | } |
| 667 | |
| 668 | static void wait_barrier(conf_t *conf) |
| 669 | { |
| 670 | spin_lock_irq(&conf->resync_lock); |
| 671 | if (conf->barrier) { |
| 672 | conf->nr_waiting++; |
| 673 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, |
| 674 | conf->resync_lock, |
| 675 | raid1_unplug(conf->mddev->queue)); |
| 676 | conf->nr_waiting--; |
| 677 | } |
| 678 | conf->nr_pending++; |
| 679 | spin_unlock_irq(&conf->resync_lock); |
| 680 | } |
| 681 | |
| 682 | static void allow_barrier(conf_t *conf) |
| 683 | { |
| 684 | unsigned long flags; |
| 685 | spin_lock_irqsave(&conf->resync_lock, flags); |
| 686 | conf->nr_pending--; |
| 687 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
| 688 | wake_up(&conf->wait_barrier); |
| 689 | } |
| 690 | |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 691 | static void freeze_array(conf_t *conf) |
| 692 | { |
| 693 | /* stop syncio and normal IO and wait for everything to |
| 694 | * go quite. |
| 695 | * We increment barrier and nr_waiting, and then |
NeilBrown | 1c83053 | 2008-03-04 14:29:35 -0800 | [diff] [blame] | 696 | * wait until nr_pending match nr_queued+1 |
| 697 | * This is called in the context of one normal IO request |
| 698 | * that has failed. Thus any sync request that might be pending |
| 699 | * will be blocked by nr_pending, and we need to wait for |
| 700 | * pending IO requests to complete or be queued for re-try. |
| 701 | * Thus the number queued (nr_queued) plus this request (1) |
| 702 | * must match the number of pending IOs (nr_pending) before |
| 703 | * we continue. |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 704 | */ |
| 705 | spin_lock_irq(&conf->resync_lock); |
| 706 | conf->barrier++; |
| 707 | conf->nr_waiting++; |
| 708 | wait_event_lock_irq(conf->wait_barrier, |
NeilBrown | 1c83053 | 2008-03-04 14:29:35 -0800 | [diff] [blame] | 709 | conf->nr_pending == conf->nr_queued+1, |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 710 | conf->resync_lock, |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 711 | ({ flush_pending_writes(conf); |
| 712 | raid1_unplug(conf->mddev->queue); })); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 713 | spin_unlock_irq(&conf->resync_lock); |
| 714 | } |
| 715 | static void unfreeze_array(conf_t *conf) |
| 716 | { |
| 717 | /* reverse the effect of the freeze */ |
| 718 | spin_lock_irq(&conf->resync_lock); |
| 719 | conf->barrier--; |
| 720 | conf->nr_waiting--; |
| 721 | wake_up(&conf->wait_barrier); |
| 722 | spin_unlock_irq(&conf->resync_lock); |
| 723 | } |
| 724 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 725 | |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 726 | /* duplicate the data pages for behind I/O |
| 727 | * We return a list of bio_vec rather than just page pointers |
| 728 | * as it makes freeing easier |
| 729 | */ |
| 730 | static struct bio_vec *alloc_behind_pages(struct bio *bio) |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 731 | { |
| 732 | int i; |
| 733 | struct bio_vec *bvec; |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 734 | struct bio_vec *pages = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 735 | GFP_NOIO); |
| 736 | if (unlikely(!pages)) |
| 737 | goto do_sync_io; |
| 738 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 739 | bio_for_each_segment(bvec, bio, i) { |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 740 | pages[i].bv_page = alloc_page(GFP_NOIO); |
| 741 | if (unlikely(!pages[i].bv_page)) |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 742 | goto do_sync_io; |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 743 | memcpy(kmap(pages[i].bv_page) + bvec->bv_offset, |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 744 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 745 | kunmap(pages[i].bv_page); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 746 | kunmap(bvec->bv_page); |
| 747 | } |
| 748 | |
| 749 | return pages; |
| 750 | |
| 751 | do_sync_io: |
| 752 | if (pages) |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 753 | for (i = 0; i < bio->bi_vcnt && pages[i].bv_page; i++) |
| 754 | put_page(pages[i].bv_page); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 755 | kfree(pages); |
| 756 | PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); |
| 757 | return NULL; |
| 758 | } |
| 759 | |
NeilBrown | 21a52c6 | 2010-04-01 15:02:13 +1100 | [diff] [blame] | 760 | static int make_request(mddev_t *mddev, struct bio * bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 761 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 762 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | mirror_info_t *mirror; |
| 764 | r1bio_t *r1_bio; |
| 765 | struct bio *read_bio; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 766 | int i, targets = 0, disks; |
NeilBrown | 84255d1 | 2008-05-23 13:04:32 -0700 | [diff] [blame] | 767 | struct bitmap *bitmap; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 768 | unsigned long flags; |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 769 | struct bio_vec *behind_pages = NULL; |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 770 | const int rw = bio_data_dir(bio); |
NeilBrown | 2c7d46e | 2010-08-18 16:16:05 +1000 | [diff] [blame] | 771 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 772 | const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 773 | mdk_rdev_t *blocked_rdev; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 774 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | /* |
| 776 | * Register the new request and wait if the reconstruction |
| 777 | * thread has put up a bar for new requests. |
| 778 | * Continue immediately if no resync is active currently. |
| 779 | */ |
NeilBrown | 62de608 | 2006-05-01 12:15:47 -0700 | [diff] [blame] | 780 | |
NeilBrown | 3d310eb | 2005-06-21 17:17:26 -0700 | [diff] [blame] | 781 | md_write_start(mddev, bio); /* wait on superblock update early */ |
| 782 | |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 783 | if (bio_data_dir(bio) == WRITE && |
| 784 | bio->bi_sector + bio->bi_size/512 > mddev->suspend_lo && |
| 785 | bio->bi_sector < mddev->suspend_hi) { |
| 786 | /* As the suspend_* range is controlled by |
| 787 | * userspace, we want an interruptible |
| 788 | * wait. |
| 789 | */ |
| 790 | DEFINE_WAIT(w); |
| 791 | for (;;) { |
| 792 | flush_signals(current); |
| 793 | prepare_to_wait(&conf->wait_barrier, |
| 794 | &w, TASK_INTERRUPTIBLE); |
| 795 | if (bio->bi_sector + bio->bi_size/512 <= mddev->suspend_lo || |
| 796 | bio->bi_sector >= mddev->suspend_hi) |
| 797 | break; |
| 798 | schedule(); |
| 799 | } |
| 800 | finish_wait(&conf->wait_barrier, &w); |
| 801 | } |
NeilBrown | 62de608 | 2006-05-01 12:15:47 -0700 | [diff] [blame] | 802 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 803 | wait_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 804 | |
NeilBrown | 84255d1 | 2008-05-23 13:04:32 -0700 | [diff] [blame] | 805 | bitmap = mddev->bitmap; |
| 806 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | /* |
| 808 | * make_request() can abort the operation when READA is being |
| 809 | * used and no empty request is available. |
| 810 | * |
| 811 | */ |
| 812 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 813 | |
| 814 | r1_bio->master_bio = bio; |
| 815 | r1_bio->sectors = bio->bi_size >> 9; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 816 | r1_bio->state = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 817 | r1_bio->mddev = mddev; |
| 818 | r1_bio->sector = bio->bi_sector; |
| 819 | |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 820 | if (rw == READ) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 821 | /* |
| 822 | * read balancing logic: |
| 823 | */ |
| 824 | int rdisk = read_balance(conf, r1_bio); |
| 825 | |
| 826 | if (rdisk < 0) { |
| 827 | /* couldn't find anywhere to read from */ |
| 828 | raid_end_bio_io(r1_bio); |
| 829 | return 0; |
| 830 | } |
| 831 | mirror = conf->mirrors + rdisk; |
| 832 | |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 833 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
| 834 | bitmap) { |
| 835 | /* Reading from a write-mostly device must |
| 836 | * take care not to over-take any writes |
| 837 | * that are 'behind' |
| 838 | */ |
| 839 | wait_event(bitmap->behind_wait, |
| 840 | atomic_read(&bitmap->behind_writes) == 0); |
| 841 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 842 | r1_bio->read_disk = rdisk; |
| 843 | |
NeilBrown | a167f66 | 2010-10-26 18:31:13 +1100 | [diff] [blame] | 844 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 845 | |
| 846 | r1_bio->bios[rdisk] = read_bio; |
| 847 | |
| 848 | read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; |
| 849 | read_bio->bi_bdev = mirror->rdev->bdev; |
| 850 | read_bio->bi_end_io = raid1_end_read_request; |
Christoph Hellwig | 7b6d91d | 2010-08-07 18:20:39 +0200 | [diff] [blame] | 851 | read_bio->bi_rw = READ | do_sync; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | read_bio->bi_private = r1_bio; |
| 853 | |
| 854 | generic_make_request(read_bio); |
| 855 | return 0; |
| 856 | } |
| 857 | |
| 858 | /* |
| 859 | * WRITE: |
| 860 | */ |
| 861 | /* first select target devices under spinlock and |
| 862 | * inc refcount on their rdev. Record them by setting |
| 863 | * bios[x] to bio |
| 864 | */ |
| 865 | disks = conf->raid_disks; |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 866 | retry_write: |
| 867 | blocked_rdev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 868 | rcu_read_lock(); |
| 869 | for (i = 0; i < disks; i++) { |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 870 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
| 871 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
| 872 | atomic_inc(&rdev->nr_pending); |
| 873 | blocked_rdev = rdev; |
| 874 | break; |
| 875 | } |
| 876 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 877 | atomic_inc(&rdev->nr_pending); |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 878 | if (test_bit(Faulty, &rdev->flags)) { |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 879 | rdev_dec_pending(rdev, mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 880 | r1_bio->bios[i] = NULL; |
NeilBrown | 964147d | 2010-05-18 15:27:13 +1000 | [diff] [blame] | 881 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | r1_bio->bios[i] = bio; |
NeilBrown | 964147d | 2010-05-18 15:27:13 +1000 | [diff] [blame] | 883 | targets++; |
| 884 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | } else |
| 886 | r1_bio->bios[i] = NULL; |
| 887 | } |
| 888 | rcu_read_unlock(); |
| 889 | |
Dan Williams | 6bfe0b4 | 2008-04-30 00:52:32 -0700 | [diff] [blame] | 890 | if (unlikely(blocked_rdev)) { |
| 891 | /* Wait for this device to become unblocked */ |
| 892 | int j; |
| 893 | |
| 894 | for (j = 0; j < i; j++) |
| 895 | if (r1_bio->bios[j]) |
| 896 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); |
| 897 | |
| 898 | allow_barrier(conf); |
| 899 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
| 900 | wait_barrier(conf); |
| 901 | goto retry_write; |
| 902 | } |
| 903 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 904 | BUG_ON(targets == 0); /* we never fail the last device */ |
| 905 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 906 | if (targets < conf->raid_disks) { |
| 907 | /* array is degraded, we will not clear the bitmap |
| 908 | * on I/O completion (see raid1_end_write_request) */ |
| 909 | set_bit(R1BIO_Degraded, &r1_bio->state); |
| 910 | } |
NeilBrown | 06d91a5 | 2005-06-21 17:17:12 -0700 | [diff] [blame] | 911 | |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 912 | /* do behind I/O ? |
| 913 | * Not if there are too many, or cannot allocate memory, |
| 914 | * or a reader on WriteMostly is waiting for behind writes |
| 915 | * to flush */ |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 916 | if (bitmap && |
NeilBrown | 42a04b5 | 2009-12-14 12:49:53 +1100 | [diff] [blame] | 917 | (atomic_read(&bitmap->behind_writes) |
| 918 | < mddev->bitmap_info.max_write_behind) && |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 919 | !waitqueue_active(&bitmap->behind_wait) && |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 920 | (behind_pages = alloc_behind_pages(bio)) != NULL) |
| 921 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
| 922 | |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 923 | atomic_set(&r1_bio->remaining, 1); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 924 | atomic_set(&r1_bio->behind_remaining, 0); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 925 | |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 926 | bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors, |
| 927 | test_bit(R1BIO_BehindIO, &r1_bio->state)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | for (i = 0; i < disks; i++) { |
| 929 | struct bio *mbio; |
| 930 | if (!r1_bio->bios[i]) |
| 931 | continue; |
| 932 | |
NeilBrown | a167f66 | 2010-10-26 18:31:13 +1100 | [diff] [blame] | 933 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 934 | r1_bio->bios[i] = mbio; |
| 935 | |
| 936 | mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset; |
| 937 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
| 938 | mbio->bi_end_io = raid1_end_write_request; |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 939 | mbio->bi_rw = WRITE | do_flush_fua | do_sync; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 940 | mbio->bi_private = r1_bio; |
| 941 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 942 | if (behind_pages) { |
| 943 | struct bio_vec *bvec; |
| 944 | int j; |
| 945 | |
| 946 | /* Yes, I really want the '__' version so that |
| 947 | * we clear any unused pointer in the io_vec, rather |
| 948 | * than leave them unchanged. This is important |
| 949 | * because when we come to free the pages, we won't |
NeilBrown | 046abee | 2010-10-26 15:46:20 +1100 | [diff] [blame] | 950 | * know the original bi_idx, so we just free |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 951 | * them all |
| 952 | */ |
| 953 | __bio_for_each_segment(bvec, mbio, j, 0) |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 954 | bvec->bv_page = behind_pages[j].bv_page; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 955 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
| 956 | atomic_inc(&r1_bio->behind_remaining); |
| 957 | } |
| 958 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | atomic_inc(&r1_bio->remaining); |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 960 | spin_lock_irqsave(&conf->device_lock, flags); |
| 961 | bio_list_add(&conf->pending_bio_list, mbio); |
| 962 | blk_plug_device(mddev->queue); |
| 963 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | } |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 965 | r1_bio_write_done(r1_bio, bio->bi_vcnt, behind_pages, behind_pages != NULL); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 966 | kfree(behind_pages); /* the behind pages are attached to the bios now */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 967 | |
NeilBrown | 4e78064 | 2010-10-19 12:54:01 +1100 | [diff] [blame] | 968 | /* In case raid1d snuck in to freeze_array */ |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 969 | wake_up(&conf->wait_barrier); |
| 970 | |
Lars Ellenberg | e3881a6 | 2007-01-10 23:15:37 -0800 | [diff] [blame] | 971 | if (do_sync) |
| 972 | md_wakeup_thread(mddev->thread); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | |
| 974 | return 0; |
| 975 | } |
| 976 | |
| 977 | static void status(struct seq_file *seq, mddev_t *mddev) |
| 978 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 979 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | int i; |
| 981 | |
| 982 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
NeilBrown | 11ce99e | 2006-10-03 01:15:52 -0700 | [diff] [blame] | 983 | conf->raid_disks - mddev->degraded); |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 984 | rcu_read_lock(); |
| 985 | for (i = 0; i < conf->raid_disks; i++) { |
| 986 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 987 | seq_printf(seq, "%s", |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 988 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
| 989 | } |
| 990 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 991 | seq_printf(seq, "]"); |
| 992 | } |
| 993 | |
| 994 | |
| 995 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) |
| 996 | { |
| 997 | char b[BDEVNAME_SIZE]; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 998 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 999 | |
| 1000 | /* |
| 1001 | * If it is not operational, then we have already marked it as dead |
| 1002 | * else if it is the last working disks, ignore the error, let the |
| 1003 | * next level up know. |
| 1004 | * else mark the drive as failed |
| 1005 | */ |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1006 | if (test_bit(In_sync, &rdev->flags) |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1007 | && (conf->raid_disks - mddev->degraded) == 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1008 | /* |
| 1009 | * Don't fail the drive, act as though we were just a |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1010 | * normal single drive. |
| 1011 | * However don't try a recovery from this drive as |
| 1012 | * it is very likely to fail. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1013 | */ |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1014 | mddev->recovery_disabled = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | return; |
NeilBrown | 4044ba5 | 2009-01-09 08:31:11 +1100 | [diff] [blame] | 1016 | } |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 1017 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
| 1018 | unsigned long flags; |
| 1019 | spin_lock_irqsave(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1020 | mddev->degraded++; |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 1021 | set_bit(Faulty, &rdev->flags); |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 1022 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1023 | /* |
| 1024 | * if recovery is running, make sure it aborts. |
| 1025 | */ |
NeilBrown | dfc7064 | 2008-05-23 13:04:39 -0700 | [diff] [blame] | 1026 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
NeilBrown | dd00a99 | 2007-05-10 03:15:50 -0700 | [diff] [blame] | 1027 | } else |
| 1028 | set_bit(Faulty, &rdev->flags); |
NeilBrown | 850b2b4 | 2006-10-03 01:15:46 -0700 | [diff] [blame] | 1029 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1030 | printk(KERN_ALERT "md/raid1:%s: Disk failure on %s, disabling device.\n" |
| 1031 | KERN_ALERT "md/raid1:%s: Operation continuing on %d devices.\n", |
| 1032 | mdname(mddev), bdevname(rdev->bdev, b), |
| 1033 | mdname(mddev), conf->raid_disks - mddev->degraded); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | } |
| 1035 | |
| 1036 | static void print_conf(conf_t *conf) |
| 1037 | { |
| 1038 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1039 | |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1040 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1041 | if (!conf) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1042 | printk(KERN_DEBUG "(!conf)\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1043 | return; |
| 1044 | } |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1045 | 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] | 1046 | conf->raid_disks); |
| 1047 | |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1048 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1049 | for (i = 0; i < conf->raid_disks; i++) { |
| 1050 | char b[BDEVNAME_SIZE]; |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1051 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
| 1052 | if (rdev) |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1053 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1054 | i, !test_bit(In_sync, &rdev->flags), |
| 1055 | !test_bit(Faulty, &rdev->flags), |
| 1056 | bdevname(rdev->bdev,b)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | } |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1058 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | } |
| 1060 | |
| 1061 | static void close_sync(conf_t *conf) |
| 1062 | { |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1063 | wait_barrier(conf); |
| 1064 | allow_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1065 | |
| 1066 | mempool_destroy(conf->r1buf_pool); |
| 1067 | conf->r1buf_pool = NULL; |
| 1068 | } |
| 1069 | |
| 1070 | static int raid1_spare_active(mddev_t *mddev) |
| 1071 | { |
| 1072 | int i; |
| 1073 | conf_t *conf = mddev->private; |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1074 | int count = 0; |
| 1075 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1076 | |
| 1077 | /* |
| 1078 | * Find all failed disks within the RAID1 configuration |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1079 | * and mark them readable. |
| 1080 | * Called under mddev lock, so rcu protection not needed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1081 | */ |
| 1082 | for (i = 0; i < conf->raid_disks; i++) { |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1083 | mdk_rdev_t *rdev = conf->mirrors[i].rdev; |
| 1084 | if (rdev |
| 1085 | && !test_bit(Faulty, &rdev->flags) |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 1086 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1087 | count++; |
Adrian Drzewiecki | e6ffbcb | 2010-08-18 11:49:02 +1000 | [diff] [blame] | 1088 | sysfs_notify_dirent(rdev->sysfs_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1089 | } |
| 1090 | } |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1091 | spin_lock_irqsave(&conf->device_lock, flags); |
| 1092 | mddev->degraded -= count; |
| 1093 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1094 | |
| 1095 | print_conf(conf); |
NeilBrown | 6b96562 | 2010-08-18 11:56:59 +1000 | [diff] [blame] | 1096 | return count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | } |
| 1098 | |
| 1099 | |
| 1100 | static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) |
| 1101 | { |
| 1102 | conf_t *conf = mddev->private; |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1103 | int err = -EEXIST; |
NeilBrown | 41158c7 | 2005-06-21 17:17:25 -0700 | [diff] [blame] | 1104 | int mirror = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | mirror_info_t *p; |
Neil Brown | 6c2fce2 | 2008-06-28 08:31:31 +1000 | [diff] [blame] | 1106 | int first = 0; |
| 1107 | int last = mddev->raid_disks - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1108 | |
Neil Brown | 6c2fce2 | 2008-06-28 08:31:31 +1000 | [diff] [blame] | 1109 | if (rdev->raid_disk >= 0) |
| 1110 | first = last = rdev->raid_disk; |
| 1111 | |
| 1112 | for (mirror = first; mirror <= last; mirror++) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | if ( !(p=conf->mirrors+mirror)->rdev) { |
| 1114 | |
Martin K. Petersen | 8f6c2e4 | 2009-07-01 11:13:45 +1000 | [diff] [blame] | 1115 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 1116 | rdev->data_offset << 9); |
NeilBrown | 627a2d3 | 2010-03-08 16:44:38 +1100 | [diff] [blame] | 1117 | /* as we don't honour merge_bvec_fn, we must |
| 1118 | * never risk violating it, so limit |
| 1119 | * ->max_segments to one lying with a single |
| 1120 | * page, as a one page request is never in |
| 1121 | * violation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1122 | */ |
NeilBrown | 627a2d3 | 2010-03-08 16:44:38 +1100 | [diff] [blame] | 1123 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
| 1124 | blk_queue_max_segments(mddev->queue, 1); |
| 1125 | blk_queue_segment_boundary(mddev->queue, |
| 1126 | PAGE_CACHE_SIZE - 1); |
| 1127 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1128 | |
| 1129 | p->head_position = 0; |
| 1130 | rdev->raid_disk = mirror; |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1131 | err = 0; |
NeilBrown | 6aea114a | 2005-11-28 13:44:13 -0800 | [diff] [blame] | 1132 | /* As all devices are equivalent, we don't need a full recovery |
| 1133 | * if this was recently any drive of the array |
| 1134 | */ |
| 1135 | if (rdev->saved_raid_disk < 0) |
NeilBrown | 41158c7 | 2005-06-21 17:17:25 -0700 | [diff] [blame] | 1136 | conf->fullsync = 1; |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 1137 | rcu_assign_pointer(p->rdev, rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1138 | break; |
| 1139 | } |
Andre Noll | ac5e711 | 2009-08-03 10:59:47 +1000 | [diff] [blame] | 1140 | md_integrity_add_rdev(rdev, mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1141 | print_conf(conf); |
Neil Brown | 199050e | 2008-06-28 08:31:33 +1000 | [diff] [blame] | 1142 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | } |
| 1144 | |
| 1145 | static int raid1_remove_disk(mddev_t *mddev, int number) |
| 1146 | { |
| 1147 | conf_t *conf = mddev->private; |
| 1148 | int err = 0; |
| 1149 | mdk_rdev_t *rdev; |
| 1150 | mirror_info_t *p = conf->mirrors+ number; |
| 1151 | |
| 1152 | print_conf(conf); |
| 1153 | rdev = p->rdev; |
| 1154 | if (rdev) { |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1155 | if (test_bit(In_sync, &rdev->flags) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1156 | atomic_read(&rdev->nr_pending)) { |
| 1157 | err = -EBUSY; |
| 1158 | goto abort; |
| 1159 | } |
NeilBrown | 046abee | 2010-10-26 15:46:20 +1100 | [diff] [blame] | 1160 | /* Only remove non-faulty devices if recovery |
NeilBrown | dfc7064 | 2008-05-23 13:04:39 -0700 | [diff] [blame] | 1161 | * is not possible. |
| 1162 | */ |
| 1163 | if (!test_bit(Faulty, &rdev->flags) && |
NeilBrown | 8f9e0ee | 2010-11-24 16:39:46 +1100 | [diff] [blame] | 1164 | !mddev->recovery_disabled && |
NeilBrown | dfc7064 | 2008-05-23 13:04:39 -0700 | [diff] [blame] | 1165 | mddev->degraded < conf->raid_disks) { |
| 1166 | err = -EBUSY; |
| 1167 | goto abort; |
| 1168 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | p->rdev = NULL; |
Paul E. McKenney | fbd568a3e | 2005-05-01 08:59:04 -0700 | [diff] [blame] | 1170 | synchronize_rcu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1171 | if (atomic_read(&rdev->nr_pending)) { |
| 1172 | /* lost the race, try later */ |
| 1173 | err = -EBUSY; |
| 1174 | p->rdev = rdev; |
Andre Noll | ac5e711 | 2009-08-03 10:59:47 +1000 | [diff] [blame] | 1175 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1176 | } |
Andre Noll | ac5e711 | 2009-08-03 10:59:47 +1000 | [diff] [blame] | 1177 | md_integrity_register(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | } |
| 1179 | abort: |
| 1180 | |
| 1181 | print_conf(conf); |
| 1182 | return err; |
| 1183 | } |
| 1184 | |
| 1185 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1186 | static void end_sync_read(struct bio *bio, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1187 | { |
H Hartley Sweeten | 7b92813 | 2010-03-08 16:02:40 +1100 | [diff] [blame] | 1188 | r1bio_t *r1_bio = bio->bi_private; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1189 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1190 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1191 | for (i=r1_bio->mddev->raid_disks; i--; ) |
| 1192 | if (r1_bio->bios[i] == bio) |
| 1193 | break; |
| 1194 | BUG_ON(i < 0); |
| 1195 | update_head_pos(i, r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1196 | /* |
| 1197 | * we have read a block, now it needs to be re-written, |
| 1198 | * or re-read if the read failed. |
| 1199 | * We don't do much here, just schedule handling by raid1d |
| 1200 | */ |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1201 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1202 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1203 | |
| 1204 | if (atomic_dec_and_test(&r1_bio->remaining)) |
| 1205 | reschedule_retry(r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1206 | } |
| 1207 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1208 | static void end_sync_write(struct bio *bio, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1209 | { |
| 1210 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
H Hartley Sweeten | 7b92813 | 2010-03-08 16:02:40 +1100 | [diff] [blame] | 1211 | r1bio_t *r1_bio = bio->bi_private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1212 | mddev_t *mddev = r1_bio->mddev; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 1213 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1214 | int i; |
| 1215 | int mirror=0; |
| 1216 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1217 | for (i = 0; i < conf->raid_disks; i++) |
| 1218 | if (r1_bio->bios[i] == bio) { |
| 1219 | mirror = i; |
| 1220 | break; |
| 1221 | } |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1222 | if (!uptodate) { |
NeilBrown | 57dab0b | 2010-10-19 10:03:39 +1100 | [diff] [blame] | 1223 | sector_t sync_blocks = 0; |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1224 | sector_t s = r1_bio->sector; |
| 1225 | long sectors_to_go = r1_bio->sectors; |
| 1226 | /* make sure these bits doesn't get cleared. */ |
| 1227 | do { |
NeilBrown | 5e3db64 | 2006-07-10 04:44:18 -0700 | [diff] [blame] | 1228 | bitmap_end_sync(mddev->bitmap, s, |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1229 | &sync_blocks, 1); |
| 1230 | s += sync_blocks; |
| 1231 | sectors_to_go -= sync_blocks; |
| 1232 | } while (sectors_to_go > 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1233 | md_error(mddev, conf->mirrors[mirror].rdev); |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1234 | } |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1235 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | update_head_pos(mirror, r1_bio); |
| 1237 | |
| 1238 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
NeilBrown | 73d5c38 | 2009-02-25 13:18:47 +1100 | [diff] [blame] | 1239 | sector_t s = r1_bio->sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1240 | put_buf(r1_bio); |
NeilBrown | 73d5c38 | 2009-02-25 13:18:47 +1100 | [diff] [blame] | 1241 | md_done_sync(mddev, s, uptodate); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1243 | } |
| 1244 | |
| 1245 | static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) |
| 1246 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 1247 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1248 | int i; |
| 1249 | int disks = conf->raid_disks; |
| 1250 | struct bio *bio, *wbio; |
| 1251 | |
| 1252 | bio = r1_bio->bios[r1_bio->read_disk]; |
| 1253 | |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1254 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1255 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
| 1256 | /* We have read all readable devices. If we haven't |
| 1257 | * got the block, then there is no hope left. |
| 1258 | * If we have, then we want to do a comparison |
| 1259 | * and skip the write if everything is the same. |
| 1260 | * If any blocks failed to read, then we need to |
| 1261 | * attempt an over-write |
| 1262 | */ |
| 1263 | int primary; |
| 1264 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
| 1265 | for (i=0; i<mddev->raid_disks; i++) |
| 1266 | if (r1_bio->bios[i]->bi_end_io == end_sync_read) |
| 1267 | md_error(mddev, conf->mirrors[i].rdev); |
| 1268 | |
| 1269 | md_done_sync(mddev, r1_bio->sectors, 1); |
| 1270 | put_buf(r1_bio); |
| 1271 | return; |
| 1272 | } |
| 1273 | for (primary=0; primary<mddev->raid_disks; primary++) |
| 1274 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
| 1275 | test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { |
| 1276 | r1_bio->bios[primary]->bi_end_io = NULL; |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 1277 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1278 | break; |
| 1279 | } |
| 1280 | r1_bio->read_disk = primary; |
| 1281 | for (i=0; i<mddev->raid_disks; i++) |
Mike Accetta | ed45666 | 2007-06-16 10:16:07 -0700 | [diff] [blame] | 1282 | if (r1_bio->bios[i]->bi_end_io == end_sync_read) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1283 | int j; |
| 1284 | int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9); |
| 1285 | struct bio *pbio = r1_bio->bios[primary]; |
| 1286 | struct bio *sbio = r1_bio->bios[i]; |
Mike Accetta | ed45666 | 2007-06-16 10:16:07 -0700 | [diff] [blame] | 1287 | |
| 1288 | if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) { |
| 1289 | for (j = vcnt; j-- ; ) { |
| 1290 | struct page *p, *s; |
| 1291 | p = pbio->bi_io_vec[j].bv_page; |
| 1292 | s = sbio->bi_io_vec[j].bv_page; |
| 1293 | if (memcmp(page_address(p), |
| 1294 | page_address(s), |
| 1295 | PAGE_SIZE)) |
| 1296 | break; |
| 1297 | } |
| 1298 | } else |
| 1299 | j = 0; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1300 | if (j >= 0) |
| 1301 | mddev->resync_mismatches += r1_bio->sectors; |
NeilBrown | cf7a441 | 2007-10-16 23:30:55 -0700 | [diff] [blame] | 1302 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
| 1303 | && test_bit(BIO_UPTODATE, &sbio->bi_flags))) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1304 | sbio->bi_end_io = NULL; |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 1305 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); |
| 1306 | } else { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1307 | /* fixup the bio for reuse */ |
NeilBrown | 698b18c | 2008-05-23 13:04:35 -0700 | [diff] [blame] | 1308 | int size; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1309 | sbio->bi_vcnt = vcnt; |
| 1310 | sbio->bi_size = r1_bio->sectors << 9; |
| 1311 | sbio->bi_idx = 0; |
| 1312 | sbio->bi_phys_segments = 0; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1313 | sbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
| 1314 | sbio->bi_flags |= 1 << BIO_UPTODATE; |
| 1315 | sbio->bi_next = NULL; |
| 1316 | sbio->bi_sector = r1_bio->sector + |
| 1317 | conf->mirrors[i].rdev->data_offset; |
| 1318 | sbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
NeilBrown | 698b18c | 2008-05-23 13:04:35 -0700 | [diff] [blame] | 1319 | size = sbio->bi_size; |
| 1320 | for (j = 0; j < vcnt ; j++) { |
| 1321 | struct bio_vec *bi; |
| 1322 | bi = &sbio->bi_io_vec[j]; |
| 1323 | bi->bv_offset = 0; |
| 1324 | if (size > PAGE_SIZE) |
| 1325 | bi->bv_len = PAGE_SIZE; |
| 1326 | else |
| 1327 | bi->bv_len = size; |
| 1328 | size -= PAGE_SIZE; |
| 1329 | memcpy(page_address(bi->bv_page), |
NeilBrown | 3eda22d | 2007-01-26 00:57:01 -0800 | [diff] [blame] | 1330 | page_address(pbio->bi_io_vec[j].bv_page), |
| 1331 | PAGE_SIZE); |
NeilBrown | 698b18c | 2008-05-23 13:04:35 -0700 | [diff] [blame] | 1332 | } |
NeilBrown | 3eda22d | 2007-01-26 00:57:01 -0800 | [diff] [blame] | 1333 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1334 | } |
| 1335 | } |
| 1336 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1338 | /* ouch - failed to read all of that. |
| 1339 | * Try some synchronous reads of other devices to get |
| 1340 | * good data, much like with normal read errors. Only |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1341 | * read into the pages we already have so we don't |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1342 | * need to re-issue the read request. |
| 1343 | * We don't need to freeze the array, because being in an |
| 1344 | * active sync request, there is no normal IO, and |
| 1345 | * no overlapping syncs. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1346 | */ |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1347 | sector_t sect = r1_bio->sector; |
| 1348 | int sectors = r1_bio->sectors; |
| 1349 | int idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1350 | |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1351 | while(sectors) { |
| 1352 | int s = sectors; |
| 1353 | int d = r1_bio->read_disk; |
| 1354 | int success = 0; |
| 1355 | mdk_rdev_t *rdev; |
| 1356 | |
| 1357 | if (s > (PAGE_SIZE>>9)) |
| 1358 | s = PAGE_SIZE >> 9; |
| 1359 | do { |
| 1360 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1361 | /* No rcu protection needed here devices |
| 1362 | * can only be removed when no resync is |
| 1363 | * active, and resync is currently active |
| 1364 | */ |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1365 | rdev = conf->mirrors[d].rdev; |
NeilBrown | 2b19336 | 2010-10-27 15:16:40 +1100 | [diff] [blame] | 1366 | if (sync_page_io(rdev, |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1367 | sect + rdev->data_offset, |
| 1368 | s<<9, |
| 1369 | bio->bi_io_vec[idx].bv_page, |
| 1370 | READ)) { |
| 1371 | success = 1; |
| 1372 | break; |
| 1373 | } |
| 1374 | } |
| 1375 | d++; |
| 1376 | if (d == conf->raid_disks) |
| 1377 | d = 0; |
| 1378 | } while (!success && d != r1_bio->read_disk); |
| 1379 | |
| 1380 | if (success) { |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1381 | int start = d; |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1382 | /* write it back and re-read */ |
| 1383 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
| 1384 | while (d != r1_bio->read_disk) { |
| 1385 | if (d == 0) |
| 1386 | d = conf->raid_disks; |
| 1387 | d--; |
| 1388 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) |
| 1389 | continue; |
| 1390 | rdev = conf->mirrors[d].rdev; |
NeilBrown | 4dbcdc7 | 2006-01-06 00:20:52 -0800 | [diff] [blame] | 1391 | atomic_add(s, &rdev->corrected_errors); |
NeilBrown | 2b19336 | 2010-10-27 15:16:40 +1100 | [diff] [blame] | 1392 | if (sync_page_io(rdev, |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1393 | sect + rdev->data_offset, |
| 1394 | s<<9, |
| 1395 | bio->bi_io_vec[idx].bv_page, |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1396 | WRITE) == 0) |
| 1397 | md_error(mddev, rdev); |
| 1398 | } |
| 1399 | d = start; |
| 1400 | while (d != r1_bio->read_disk) { |
| 1401 | if (d == 0) |
| 1402 | d = conf->raid_disks; |
| 1403 | d--; |
| 1404 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) |
| 1405 | continue; |
| 1406 | rdev = conf->mirrors[d].rdev; |
NeilBrown | 2b19336 | 2010-10-27 15:16:40 +1100 | [diff] [blame] | 1407 | if (sync_page_io(rdev, |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1408 | sect + rdev->data_offset, |
| 1409 | s<<9, |
| 1410 | bio->bi_io_vec[idx].bv_page, |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1411 | READ) == 0) |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1412 | md_error(mddev, rdev); |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1413 | } |
| 1414 | } else { |
| 1415 | char b[BDEVNAME_SIZE]; |
| 1416 | /* Cannot read from anywhere, array is toast */ |
| 1417 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1418 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1419 | " for block %llu\n", |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1420 | mdname(mddev), |
| 1421 | bdevname(bio->bi_bdev, b), |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1422 | (unsigned long long)r1_bio->sector); |
| 1423 | md_done_sync(mddev, r1_bio->sectors, 0); |
| 1424 | put_buf(r1_bio); |
| 1425 | return; |
| 1426 | } |
| 1427 | sectors -= s; |
| 1428 | sect += s; |
| 1429 | idx ++; |
| 1430 | } |
| 1431 | } |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1432 | |
| 1433 | /* |
| 1434 | * schedule writes |
| 1435 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1436 | atomic_set(&r1_bio->remaining, 1); |
| 1437 | for (i = 0; i < disks ; i++) { |
| 1438 | wbio = r1_bio->bios[i]; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1439 | if (wbio->bi_end_io == NULL || |
| 1440 | (wbio->bi_end_io == end_sync_read && |
| 1441 | (i == r1_bio->read_disk || |
| 1442 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | continue; |
| 1444 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1445 | wbio->bi_rw = WRITE; |
| 1446 | wbio->bi_end_io = end_sync_write; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1447 | atomic_inc(&r1_bio->remaining); |
| 1448 | md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1449 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | generic_make_request(wbio); |
| 1451 | } |
| 1452 | |
| 1453 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1454 | /* if we're here, all write(s) have completed, so clean up */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1455 | md_done_sync(mddev, r1_bio->sectors, 1); |
| 1456 | put_buf(r1_bio); |
| 1457 | } |
| 1458 | } |
| 1459 | |
| 1460 | /* |
| 1461 | * This is a kernel thread which: |
| 1462 | * |
| 1463 | * 1. Retries failed read operations on working mirrors. |
| 1464 | * 2. Updates the raid superblock when problems encounter. |
| 1465 | * 3. Performs writes following reads for array syncronising. |
| 1466 | */ |
| 1467 | |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 1468 | static void fix_read_error(conf_t *conf, int read_disk, |
| 1469 | sector_t sect, int sectors) |
| 1470 | { |
| 1471 | mddev_t *mddev = conf->mddev; |
| 1472 | while(sectors) { |
| 1473 | int s = sectors; |
| 1474 | int d = read_disk; |
| 1475 | int success = 0; |
| 1476 | int start; |
| 1477 | mdk_rdev_t *rdev; |
| 1478 | |
| 1479 | if (s > (PAGE_SIZE>>9)) |
| 1480 | s = PAGE_SIZE >> 9; |
| 1481 | |
| 1482 | do { |
| 1483 | /* Note: no rcu protection needed here |
| 1484 | * as this is synchronous in the raid1d thread |
| 1485 | * which is the thread that might remove |
| 1486 | * a device. If raid1d ever becomes multi-threaded.... |
| 1487 | */ |
| 1488 | rdev = conf->mirrors[d].rdev; |
| 1489 | if (rdev && |
| 1490 | test_bit(In_sync, &rdev->flags) && |
NeilBrown | 2b19336 | 2010-10-27 15:16:40 +1100 | [diff] [blame] | 1491 | sync_page_io(rdev, |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 1492 | sect + rdev->data_offset, |
| 1493 | s<<9, |
| 1494 | conf->tmppage, READ)) |
| 1495 | success = 1; |
| 1496 | else { |
| 1497 | d++; |
| 1498 | if (d == conf->raid_disks) |
| 1499 | d = 0; |
| 1500 | } |
| 1501 | } while (!success && d != read_disk); |
| 1502 | |
| 1503 | if (!success) { |
| 1504 | /* Cannot read from anywhere -- bye bye array */ |
| 1505 | md_error(mddev, conf->mirrors[read_disk].rdev); |
| 1506 | break; |
| 1507 | } |
| 1508 | /* write it back and re-read */ |
| 1509 | start = d; |
| 1510 | while (d != read_disk) { |
| 1511 | if (d==0) |
| 1512 | d = conf->raid_disks; |
| 1513 | d--; |
| 1514 | rdev = conf->mirrors[d].rdev; |
| 1515 | if (rdev && |
| 1516 | test_bit(In_sync, &rdev->flags)) { |
NeilBrown | 2b19336 | 2010-10-27 15:16:40 +1100 | [diff] [blame] | 1517 | if (sync_page_io(rdev, |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 1518 | sect + rdev->data_offset, |
| 1519 | s<<9, conf->tmppage, WRITE) |
| 1520 | == 0) |
| 1521 | /* Well, this device is dead */ |
| 1522 | md_error(mddev, rdev); |
| 1523 | } |
| 1524 | } |
| 1525 | d = start; |
| 1526 | while (d != read_disk) { |
| 1527 | char b[BDEVNAME_SIZE]; |
| 1528 | if (d==0) |
| 1529 | d = conf->raid_disks; |
| 1530 | d--; |
| 1531 | rdev = conf->mirrors[d].rdev; |
| 1532 | if (rdev && |
| 1533 | test_bit(In_sync, &rdev->flags)) { |
NeilBrown | 2b19336 | 2010-10-27 15:16:40 +1100 | [diff] [blame] | 1534 | if (sync_page_io(rdev, |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 1535 | sect + rdev->data_offset, |
| 1536 | s<<9, conf->tmppage, READ) |
| 1537 | == 0) |
| 1538 | /* Well, this device is dead */ |
| 1539 | md_error(mddev, rdev); |
| 1540 | else { |
| 1541 | atomic_add(s, &rdev->corrected_errors); |
| 1542 | printk(KERN_INFO |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1543 | "md/raid1:%s: read error corrected " |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 1544 | "(%d sectors at %llu on %s)\n", |
| 1545 | mdname(mddev), s, |
Randy Dunlap | 969b755 | 2006-10-28 10:38:32 -0700 | [diff] [blame] | 1546 | (unsigned long long)(sect + |
| 1547 | rdev->data_offset), |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 1548 | bdevname(rdev->bdev, b)); |
| 1549 | } |
| 1550 | } |
| 1551 | } |
| 1552 | sectors -= s; |
| 1553 | sect += s; |
| 1554 | } |
| 1555 | } |
| 1556 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | static void raid1d(mddev_t *mddev) |
| 1558 | { |
| 1559 | r1bio_t *r1_bio; |
| 1560 | struct bio *bio; |
| 1561 | unsigned long flags; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 1562 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1563 | struct list_head *head = &conf->retry_list; |
| 1564 | int unplug=0; |
| 1565 | mdk_rdev_t *rdev; |
| 1566 | |
| 1567 | md_check_recovery(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1568 | |
| 1569 | for (;;) { |
| 1570 | char b[BDEVNAME_SIZE]; |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 1571 | |
| 1572 | unplug += flush_pending_writes(conf); |
| 1573 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1574 | spin_lock_irqsave(&conf->device_lock, flags); |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 1575 | if (list_empty(head)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1576 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1577 | break; |
NeilBrown | a35e63e | 2008-03-04 14:29:29 -0800 | [diff] [blame] | 1578 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1579 | r1_bio = list_entry(head->prev, r1bio_t, retry_list); |
| 1580 | list_del(head->prev); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1581 | conf->nr_queued--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1582 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 1583 | |
| 1584 | mddev = r1_bio->mddev; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 1585 | conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
| 1587 | sync_request_write(mddev, r1_bio); |
| 1588 | unplug = 1; |
| 1589 | } else { |
| 1590 | int disk; |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1591 | |
| 1592 | /* we got a read error. Maybe the drive is bad. Maybe just |
| 1593 | * the block and we can fix it. |
| 1594 | * We freeze all other IO, and try reading the block from |
| 1595 | * other devices. When we find one, we re-write |
| 1596 | * and check it that fixes the read error. |
| 1597 | * This is all done synchronously while the array is |
| 1598 | * frozen |
| 1599 | */ |
NeilBrown | 867868f | 2006-10-03 01:15:51 -0700 | [diff] [blame] | 1600 | if (mddev->ro == 0) { |
| 1601 | freeze_array(conf); |
| 1602 | fix_read_error(conf, r1_bio->read_disk, |
| 1603 | r1_bio->sector, |
| 1604 | r1_bio->sectors); |
| 1605 | unfreeze_array(conf); |
NeilBrown | d0e2607 | 2009-12-01 17:30:59 +1100 | [diff] [blame] | 1606 | } else |
| 1607 | md_error(mddev, |
| 1608 | conf->mirrors[r1_bio->read_disk].rdev); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1609 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1610 | bio = r1_bio->bios[r1_bio->read_disk]; |
NeilBrown | d0e2607 | 2009-12-01 17:30:59 +1100 | [diff] [blame] | 1611 | if ((disk=read_balance(conf, r1_bio)) == -1) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1612 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1613 | " read error for block %llu\n", |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1614 | mdname(mddev), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1615 | bdevname(bio->bi_bdev,b), |
| 1616 | (unsigned long long)r1_bio->sector); |
| 1617 | raid_end_bio_io(r1_bio); |
| 1618 | } else { |
NeilBrown | 2c7d46e | 2010-08-18 16:16:05 +1000 | [diff] [blame] | 1619 | const unsigned long do_sync = r1_bio->master_bio->bi_rw & REQ_SYNC; |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 1620 | r1_bio->bios[r1_bio->read_disk] = |
| 1621 | mddev->ro ? IO_BLOCKED : NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1622 | r1_bio->read_disk = disk; |
| 1623 | bio_put(bio); |
NeilBrown | a167f66 | 2010-10-26 18:31:13 +1100 | [diff] [blame] | 1624 | bio = bio_clone_mddev(r1_bio->master_bio, |
| 1625 | GFP_NOIO, mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | r1_bio->bios[r1_bio->read_disk] = bio; |
| 1627 | rdev = conf->mirrors[disk].rdev; |
| 1628 | if (printk_ratelimit()) |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1629 | printk(KERN_ERR "md/raid1:%s: redirecting sector %llu to" |
NeilBrown | d754c5a | 2010-04-07 12:14:43 +1000 | [diff] [blame] | 1630 | " other mirror: %s\n", |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1631 | mdname(mddev), |
NeilBrown | d754c5a | 2010-04-07 12:14:43 +1000 | [diff] [blame] | 1632 | (unsigned long long)r1_bio->sector, |
| 1633 | bdevname(rdev->bdev,b)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1634 | bio->bi_sector = r1_bio->sector + rdev->data_offset; |
| 1635 | bio->bi_bdev = rdev->bdev; |
| 1636 | bio->bi_end_io = raid1_end_read_request; |
Christoph Hellwig | 7b6d91d | 2010-08-07 18:20:39 +0200 | [diff] [blame] | 1637 | bio->bi_rw = READ | do_sync; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | bio->bi_private = r1_bio; |
| 1639 | unplug = 1; |
| 1640 | generic_make_request(bio); |
| 1641 | } |
| 1642 | } |
NeilBrown | 1d9d524 | 2009-10-16 15:55:32 +1100 | [diff] [blame] | 1643 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1644 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1645 | if (unplug) |
| 1646 | unplug_slaves(mddev); |
| 1647 | } |
| 1648 | |
| 1649 | |
| 1650 | static int init_resync(conf_t *conf) |
| 1651 | { |
| 1652 | int buffs; |
| 1653 | |
| 1654 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; |
Eric Sesterhenn | 9e77c48 | 2006-04-01 01:08:49 +0200 | [diff] [blame] | 1655 | BUG_ON(conf->r1buf_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1656 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
| 1657 | conf->poolinfo); |
| 1658 | if (!conf->r1buf_pool) |
| 1659 | return -ENOMEM; |
| 1660 | conf->next_resync = 0; |
| 1661 | return 0; |
| 1662 | } |
| 1663 | |
| 1664 | /* |
| 1665 | * perform a "sync" on one "block" |
| 1666 | * |
| 1667 | * We need to make sure that no normal I/O request - particularly write |
| 1668 | * requests - conflict with active sync requests. |
| 1669 | * |
| 1670 | * This is achieved by tracking pending requests and a 'barrier' concept |
| 1671 | * that can be installed to exclude normal IO requests. |
| 1672 | */ |
| 1673 | |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 1674 | static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1675 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 1676 | conf_t *conf = mddev->private; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1677 | r1bio_t *r1_bio; |
| 1678 | struct bio *bio; |
| 1679 | sector_t max_sector, nr_sectors; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1680 | int disk = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1681 | int i; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1682 | int wonly = -1; |
| 1683 | int write_targets = 0, read_targets = 0; |
NeilBrown | 57dab0b | 2010-10-19 10:03:39 +1100 | [diff] [blame] | 1684 | sector_t sync_blocks; |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1685 | int still_degraded = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1686 | |
| 1687 | if (!conf->r1buf_pool) |
| 1688 | if (init_resync(conf)) |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 1689 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1690 | |
Andre Noll | 58c0fed | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 1691 | max_sector = mddev->dev_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | if (sector_nr >= max_sector) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1693 | /* If we aborted, we need to abort the |
| 1694 | * sync on the 'current' bitmap chunk (there will |
| 1695 | * only be one in raid1 resync. |
| 1696 | * We can find the current addess in mddev->curr_resync |
| 1697 | */ |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1698 | if (mddev->curr_resync < max_sector) /* aborted */ |
| 1699 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1700 | &sync_blocks, 1); |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1701 | else /* completed sync */ |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1702 | conf->fullsync = 0; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1703 | |
| 1704 | bitmap_close_sync(mddev->bitmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1705 | close_sync(conf); |
| 1706 | return 0; |
| 1707 | } |
| 1708 | |
NeilBrown | 07d84d10 | 2006-06-26 00:27:56 -0700 | [diff] [blame] | 1709 | if (mddev->bitmap == NULL && |
| 1710 | mddev->recovery_cp == MaxSector && |
NeilBrown | 6394cca | 2006-08-27 01:23:50 -0700 | [diff] [blame] | 1711 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
NeilBrown | 07d84d10 | 2006-06-26 00:27:56 -0700 | [diff] [blame] | 1712 | conf->fullsync == 0) { |
| 1713 | *skipped = 1; |
| 1714 | return max_sector - sector_nr; |
| 1715 | } |
NeilBrown | 6394cca | 2006-08-27 01:23:50 -0700 | [diff] [blame] | 1716 | /* before building a request, check if we can skip these blocks.. |
| 1717 | * This call the bitmap_start_sync doesn't actually record anything |
| 1718 | */ |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1719 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
NeilBrown | e5de485f | 2005-11-08 21:39:38 -0800 | [diff] [blame] | 1720 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1721 | /* We can skip this block, and probably several more */ |
| 1722 | *skipped = 1; |
| 1723 | return sync_blocks; |
| 1724 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1725 | /* |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1726 | * If there is non-resync activity waiting for a turn, |
| 1727 | * and resync is going fast enough, |
| 1728 | * then let it though before starting on this new sync request. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1729 | */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1730 | if (!go_faster && conf->nr_waiting) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1731 | msleep_interruptible(1000); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1732 | |
NeilBrown | b47490c | 2008-02-06 01:39:50 -0800 | [diff] [blame] | 1733 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
NeilBrown | 1c4588e | 2010-10-26 17:41:22 +1100 | [diff] [blame] | 1734 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1735 | raise_barrier(conf); |
| 1736 | |
| 1737 | conf->next_resync = sector_nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1739 | rcu_read_lock(); |
| 1740 | /* |
| 1741 | * If we get a correctably read error during resync or recovery, |
| 1742 | * we might want to read from a different device. So we |
| 1743 | * flag all drives that could conceivably be read from for READ, |
| 1744 | * and any others (which will be non-In_sync devices) for WRITE. |
| 1745 | * If a read fails, we try reading from something else for which READ |
| 1746 | * is OK. |
| 1747 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1748 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1749 | r1_bio->mddev = mddev; |
| 1750 | r1_bio->sector = sector_nr; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1751 | r1_bio->state = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1752 | set_bit(R1BIO_IsSync, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | |
| 1754 | for (i=0; i < conf->raid_disks; i++) { |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1755 | mdk_rdev_t *rdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1756 | bio = r1_bio->bios[i]; |
| 1757 | |
| 1758 | /* take from bio_init */ |
| 1759 | bio->bi_next = NULL; |
NeilBrown | db8d9d3 | 2010-10-07 12:00:50 +1100 | [diff] [blame] | 1760 | bio->bi_flags &= ~(BIO_POOL_MASK-1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1761 | bio->bi_flags |= 1 << BIO_UPTODATE; |
NeilBrown | db8d9d3 | 2010-10-07 12:00:50 +1100 | [diff] [blame] | 1762 | bio->bi_comp_cpu = -1; |
NeilBrown | 802ba06 | 2006-12-13 00:34:13 -0800 | [diff] [blame] | 1763 | bio->bi_rw = READ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1764 | bio->bi_vcnt = 0; |
| 1765 | bio->bi_idx = 0; |
| 1766 | bio->bi_phys_segments = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1767 | bio->bi_size = 0; |
| 1768 | bio->bi_end_io = NULL; |
| 1769 | bio->bi_private = NULL; |
| 1770 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1771 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
| 1772 | if (rdev == NULL || |
| 1773 | test_bit(Faulty, &rdev->flags)) { |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1774 | still_degraded = 1; |
| 1775 | continue; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1776 | } else if (!test_bit(In_sync, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1777 | bio->bi_rw = WRITE; |
| 1778 | bio->bi_end_io = end_sync_write; |
| 1779 | write_targets ++; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1780 | } else { |
| 1781 | /* may need to read from here */ |
| 1782 | bio->bi_rw = READ; |
| 1783 | bio->bi_end_io = end_sync_read; |
| 1784 | if (test_bit(WriteMostly, &rdev->flags)) { |
| 1785 | if (wonly < 0) |
| 1786 | wonly = i; |
| 1787 | } else { |
| 1788 | if (disk < 0) |
| 1789 | disk = i; |
| 1790 | } |
| 1791 | read_targets++; |
| 1792 | } |
| 1793 | atomic_inc(&rdev->nr_pending); |
| 1794 | bio->bi_sector = sector_nr + rdev->data_offset; |
| 1795 | bio->bi_bdev = rdev->bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1796 | bio->bi_private = r1_bio; |
| 1797 | } |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1798 | rcu_read_unlock(); |
| 1799 | if (disk < 0) |
| 1800 | disk = wonly; |
| 1801 | r1_bio->read_disk = disk; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1802 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1803 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
| 1804 | /* extra read targets are also write targets */ |
| 1805 | write_targets += read_targets-1; |
| 1806 | |
| 1807 | if (write_targets == 0 || read_targets == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1808 | /* There is nowhere to write, so all non-sync |
| 1809 | * drives must be failed - so we are finished |
| 1810 | */ |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 1811 | sector_t rv = max_sector - sector_nr; |
| 1812 | *skipped = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1813 | put_buf(r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1814 | return rv; |
| 1815 | } |
| 1816 | |
NeilBrown | c620727 | 2008-02-06 01:39:52 -0800 | [diff] [blame] | 1817 | if (max_sector > mddev->resync_max) |
| 1818 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1819 | nr_sectors = 0; |
NeilBrown | 289e99e | 2005-06-21 17:17:24 -0700 | [diff] [blame] | 1820 | sync_blocks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1821 | do { |
| 1822 | struct page *page; |
| 1823 | int len = PAGE_SIZE; |
| 1824 | if (sector_nr + (len>>9) > max_sector) |
| 1825 | len = (max_sector - sector_nr) << 9; |
| 1826 | if (len == 0) |
| 1827 | break; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1828 | if (sync_blocks == 0) { |
| 1829 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
NeilBrown | e5de485f | 2005-11-08 21:39:38 -0800 | [diff] [blame] | 1830 | &sync_blocks, still_degraded) && |
| 1831 | !conf->fullsync && |
| 1832 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1833 | break; |
Eric Sesterhenn | 9e77c48 | 2006-04-01 01:08:49 +0200 | [diff] [blame] | 1834 | BUG_ON(sync_blocks < (PAGE_SIZE>>9)); |
NeilBrown | 7571ae8 | 2010-10-07 11:54:46 +1100 | [diff] [blame] | 1835 | if ((len >> 9) > sync_blocks) |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1836 | len = sync_blocks<<9; |
NeilBrown | ab7a30c | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1837 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1838 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1839 | for (i=0 ; i < conf->raid_disks; i++) { |
| 1840 | bio = r1_bio->bios[i]; |
| 1841 | if (bio->bi_end_io) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1842 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1843 | if (bio_add_page(bio, page, len, 0) == 0) { |
| 1844 | /* stop here */ |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1845 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1846 | while (i > 0) { |
| 1847 | i--; |
| 1848 | bio = r1_bio->bios[i]; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1849 | if (bio->bi_end_io==NULL) |
| 1850 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1851 | /* remove last page from this bio */ |
| 1852 | bio->bi_vcnt--; |
| 1853 | bio->bi_size -= len; |
| 1854 | bio->bi_flags &= ~(1<< BIO_SEG_VALID); |
| 1855 | } |
| 1856 | goto bio_full; |
| 1857 | } |
| 1858 | } |
| 1859 | } |
| 1860 | nr_sectors += len>>9; |
| 1861 | sector_nr += len>>9; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1862 | sync_blocks -= (len>>9); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1863 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
| 1864 | bio_full: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1865 | r1_bio->sectors = nr_sectors; |
| 1866 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1867 | /* For a user-requested sync, we read all readable devices and do a |
| 1868 | * compare |
| 1869 | */ |
| 1870 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
| 1871 | atomic_set(&r1_bio->remaining, read_targets); |
| 1872 | for (i=0; i<conf->raid_disks; i++) { |
| 1873 | bio = r1_bio->bios[i]; |
| 1874 | if (bio->bi_end_io == end_sync_read) { |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1875 | md_sync_acct(bio->bi_bdev, nr_sectors); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1876 | generic_make_request(bio); |
| 1877 | } |
| 1878 | } |
| 1879 | } else { |
| 1880 | atomic_set(&r1_bio->remaining, 1); |
| 1881 | bio = r1_bio->bios[r1_bio->read_disk]; |
NeilBrown | ddac7c7 | 2006-08-31 21:27:36 -0700 | [diff] [blame] | 1882 | md_sync_acct(bio->bi_bdev, nr_sectors); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1883 | generic_make_request(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1884 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1885 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1886 | return nr_sectors; |
| 1887 | } |
| 1888 | |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 1889 | static sector_t raid1_size(mddev_t *mddev, sector_t sectors, int raid_disks) |
| 1890 | { |
| 1891 | if (sectors) |
| 1892 | return sectors; |
| 1893 | |
| 1894 | return mddev->dev_sectors; |
| 1895 | } |
| 1896 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1897 | static conf_t *setup_conf(mddev_t *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1898 | { |
| 1899 | conf_t *conf; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1900 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1901 | mirror_info_t *disk; |
| 1902 | mdk_rdev_t *rdev; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1903 | int err = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1904 | |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 1905 | conf = kzalloc(sizeof(conf_t), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1906 | if (!conf) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1907 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1908 | |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 1909 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1910 | GFP_KERNEL); |
| 1911 | if (!conf->mirrors) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1912 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1913 | |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1914 | conf->tmppage = alloc_page(GFP_KERNEL); |
| 1915 | if (!conf->tmppage) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1916 | goto abort; |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1917 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1918 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1919 | if (!conf->poolinfo) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1920 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1921 | conf->poolinfo->raid_disks = mddev->raid_disks; |
| 1922 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
| 1923 | r1bio_pool_free, |
| 1924 | conf->poolinfo); |
| 1925 | if (!conf->r1bio_pool) |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1926 | goto abort; |
| 1927 | |
NeilBrown | ed9bfdf | 2009-10-16 15:55:44 +1100 | [diff] [blame] | 1928 | conf->poolinfo->mddev = mddev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1929 | |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 1930 | spin_lock_init(&conf->device_lock); |
Cheng Renquan | 159ec1f | 2009-01-09 08:31:08 +1100 | [diff] [blame] | 1931 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1932 | int disk_idx = rdev->raid_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1933 | if (disk_idx >= mddev->raid_disks |
| 1934 | || disk_idx < 0) |
| 1935 | continue; |
| 1936 | disk = conf->mirrors + disk_idx; |
| 1937 | |
| 1938 | disk->rdev = rdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1939 | |
| 1940 | disk->head_position = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1941 | } |
| 1942 | conf->raid_disks = mddev->raid_disks; |
| 1943 | conf->mddev = mddev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1944 | INIT_LIST_HEAD(&conf->retry_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1945 | |
| 1946 | spin_lock_init(&conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1947 | init_waitqueue_head(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1948 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1949 | bio_list_init(&conf->pending_bio_list); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1950 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1951 | conf->last_used = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1952 | for (i = 0; i < conf->raid_disks; i++) { |
| 1953 | |
| 1954 | disk = conf->mirrors + i; |
| 1955 | |
NeilBrown | 5fd6c1d | 2006-06-26 00:27:40 -0700 | [diff] [blame] | 1956 | if (!disk->rdev || |
| 1957 | !test_bit(In_sync, &disk->rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1958 | disk->head_position = 0; |
NeilBrown | 918f023 | 2007-08-22 14:01:52 -0700 | [diff] [blame] | 1959 | if (disk->rdev) |
| 1960 | conf->fullsync = 1; |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1961 | } else if (conf->last_used < 0) |
| 1962 | /* |
| 1963 | * The first working device is used as a |
| 1964 | * starting point to read balancing. |
| 1965 | */ |
| 1966 | conf->last_used = i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1967 | } |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1968 | |
| 1969 | err = -EIO; |
| 1970 | if (conf->last_used < 0) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1971 | printk(KERN_ERR "md/raid1:%s: no operational mirrors\n", |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1972 | mdname(mddev)); |
| 1973 | goto abort; |
NeilBrown | 11ce99e | 2006-10-03 01:15:52 -0700 | [diff] [blame] | 1974 | } |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1975 | err = -ENOMEM; |
| 1976 | conf->thread = md_register_thread(raid1d, mddev, NULL); |
| 1977 | if (!conf->thread) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1978 | printk(KERN_ERR |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 1979 | "md/raid1:%s: couldn't allocate thread\n", |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1980 | mdname(mddev)); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1981 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1982 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1983 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 1984 | return conf; |
| 1985 | |
| 1986 | abort: |
| 1987 | if (conf) { |
| 1988 | if (conf->r1bio_pool) |
| 1989 | mempool_destroy(conf->r1bio_pool); |
| 1990 | kfree(conf->mirrors); |
| 1991 | safe_put_page(conf->tmppage); |
| 1992 | kfree(conf->poolinfo); |
| 1993 | kfree(conf); |
| 1994 | } |
| 1995 | return ERR_PTR(err); |
| 1996 | } |
| 1997 | |
| 1998 | static int run(mddev_t *mddev) |
| 1999 | { |
| 2000 | conf_t *conf; |
| 2001 | int i; |
| 2002 | mdk_rdev_t *rdev; |
| 2003 | |
| 2004 | if (mddev->level != 1) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2005 | 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] | 2006 | mdname(mddev), mddev->level); |
| 2007 | return -EIO; |
| 2008 | } |
| 2009 | if (mddev->reshape_position != MaxSector) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2010 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2011 | mdname(mddev)); |
| 2012 | return -EIO; |
| 2013 | } |
| 2014 | /* |
| 2015 | * copy the already verified devices into our private RAID1 |
| 2016 | * bookkeeping area. [whatever we allocate in run(), |
| 2017 | * should be freed in stop()] |
| 2018 | */ |
| 2019 | if (mddev->private == NULL) |
| 2020 | conf = setup_conf(mddev); |
| 2021 | else |
| 2022 | conf = mddev->private; |
| 2023 | |
| 2024 | if (IS_ERR(conf)) |
| 2025 | return PTR_ERR(conf); |
| 2026 | |
| 2027 | mddev->queue->queue_lock = &conf->device_lock; |
| 2028 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
| 2029 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 2030 | rdev->data_offset << 9); |
| 2031 | /* as we don't honour merge_bvec_fn, we must never risk |
NeilBrown | 627a2d3 | 2010-03-08 16:44:38 +1100 | [diff] [blame] | 2032 | * violating it, so limit ->max_segments to 1 lying within |
| 2033 | * a single page, as a one page request is never in violation. |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2034 | */ |
NeilBrown | 627a2d3 | 2010-03-08 16:44:38 +1100 | [diff] [blame] | 2035 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
| 2036 | blk_queue_max_segments(mddev->queue, 1); |
| 2037 | blk_queue_segment_boundary(mddev->queue, |
| 2038 | PAGE_CACHE_SIZE - 1); |
| 2039 | } |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2040 | } |
| 2041 | |
| 2042 | mddev->degraded = 0; |
| 2043 | for (i=0; i < conf->raid_disks; i++) |
| 2044 | if (conf->mirrors[i].rdev == NULL || |
| 2045 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || |
| 2046 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) |
| 2047 | mddev->degraded++; |
| 2048 | |
| 2049 | if (conf->raid_disks - mddev->degraded == 1) |
| 2050 | mddev->recovery_cp = MaxSector; |
| 2051 | |
Andre Noll | 8c6ac86 | 2009-06-18 08:48:06 +1000 | [diff] [blame] | 2052 | if (mddev->recovery_cp != MaxSector) |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2053 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
Andre Noll | 8c6ac86 | 2009-06-18 08:48:06 +1000 | [diff] [blame] | 2054 | " -- starting background reconstruction\n", |
| 2055 | mdname(mddev)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2056 | printk(KERN_INFO |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2057 | "md/raid1:%s: active with %d out of %d mirrors\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2058 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
| 2059 | mddev->raid_disks); |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2060 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2061 | /* |
| 2062 | * Ok, everything is just fine now |
| 2063 | */ |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2064 | mddev->thread = conf->thread; |
| 2065 | conf->thread = NULL; |
| 2066 | mddev->private = conf; |
| 2067 | |
Dan Williams | 1f40362 | 2009-03-31 14:59:03 +1100 | [diff] [blame] | 2068 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2069 | |
NeilBrown | 7a5febe | 2005-05-16 21:53:16 -0700 | [diff] [blame] | 2070 | mddev->queue->unplug_fn = raid1_unplug; |
NeilBrown | 0d12922 | 2006-10-03 01:15:54 -0700 | [diff] [blame] | 2071 | mddev->queue->backing_dev_info.congested_fn = raid1_congested; |
| 2072 | mddev->queue->backing_dev_info.congested_data = mddev; |
Andre Noll | ac5e711 | 2009-08-03 10:59:47 +1000 | [diff] [blame] | 2073 | md_integrity_register(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2074 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2075 | } |
| 2076 | |
| 2077 | static int stop(mddev_t *mddev) |
| 2078 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 2079 | conf_t *conf = mddev->private; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 2080 | struct bitmap *bitmap = mddev->bitmap; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 2081 | |
| 2082 | /* wait for behind writes to complete */ |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 2083 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2084 | printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n", |
| 2085 | mdname(mddev)); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 2086 | /* need to kick something here to make sure I/O goes? */ |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 2087 | wait_event(bitmap->behind_wait, |
| 2088 | atomic_read(&bitmap->behind_writes) == 0); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 2089 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2090 | |
NeilBrown | 409c57f | 2009-03-31 14:39:39 +1100 | [diff] [blame] | 2091 | raise_barrier(conf); |
| 2092 | lower_barrier(conf); |
| 2093 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2094 | md_unregister_thread(mddev->thread); |
| 2095 | mddev->thread = NULL; |
| 2096 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
| 2097 | if (conf->r1bio_pool) |
| 2098 | mempool_destroy(conf->r1bio_pool); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 2099 | kfree(conf->mirrors); |
| 2100 | kfree(conf->poolinfo); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2101 | kfree(conf); |
| 2102 | mddev->private = NULL; |
| 2103 | return 0; |
| 2104 | } |
| 2105 | |
| 2106 | static int raid1_resize(mddev_t *mddev, sector_t sectors) |
| 2107 | { |
| 2108 | /* no resync is happening, and there is enough space |
| 2109 | * on all devices, so we can resize. |
| 2110 | * We need to make sure resync covers any new space. |
| 2111 | * If the array is shrinking we should possibly wait until |
| 2112 | * any io in the removed space completes, but it hardly seems |
| 2113 | * worth it. |
| 2114 | */ |
Dan Williams | 1f40362 | 2009-03-31 14:59:03 +1100 | [diff] [blame] | 2115 | md_set_array_sectors(mddev, raid1_size(mddev, sectors, 0)); |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 2116 | if (mddev->array_sectors > raid1_size(mddev, sectors, 0)) |
| 2117 | return -EINVAL; |
Andre Noll | f233ea5 | 2008-07-21 17:05:22 +1000 | [diff] [blame] | 2118 | set_capacity(mddev->gendisk, mddev->array_sectors); |
NeilBrown | 449aad3 | 2009-08-03 10:59:58 +1000 | [diff] [blame] | 2119 | revalidate_disk(mddev->gendisk); |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 2120 | if (sectors > mddev->dev_sectors && |
Andre Noll | f233ea5 | 2008-07-21 17:05:22 +1000 | [diff] [blame] | 2121 | mddev->recovery_cp == MaxSector) { |
Andre Noll | 58c0fed | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 2122 | mddev->recovery_cp = mddev->dev_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2123 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
| 2124 | } |
Dan Williams | b522adc | 2009-03-31 15:00:31 +1100 | [diff] [blame] | 2125 | mddev->dev_sectors = sectors; |
NeilBrown | 4b5c7ae | 2005-07-27 11:43:28 -0700 | [diff] [blame] | 2126 | mddev->resync_max_sectors = sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2127 | return 0; |
| 2128 | } |
| 2129 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2130 | static int raid1_reshape(mddev_t *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2131 | { |
| 2132 | /* We need to: |
| 2133 | * 1/ resize the r1bio_pool |
| 2134 | * 2/ resize conf->mirrors |
| 2135 | * |
| 2136 | * We allocate a new r1bio_pool if we can. |
| 2137 | * Then raise a device barrier and wait until all IO stops. |
| 2138 | * Then resize conf->mirrors and swap in the new r1bio pool. |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2139 | * |
| 2140 | * At the same time, we "pack" the devices so that all the missing |
| 2141 | * devices have the higher raid_disk numbers. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2142 | */ |
| 2143 | mempool_t *newpool, *oldpool; |
| 2144 | struct pool_info *newpoolinfo; |
| 2145 | mirror_info_t *newmirrors; |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 2146 | conf_t *conf = mddev->private; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2147 | int cnt, raid_disks; |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 2148 | unsigned long flags; |
Dan Williams | b5470dc | 2008-06-27 21:44:04 -0700 | [diff] [blame] | 2149 | int d, d2, err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2150 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2151 | /* Cannot change chunk_size, layout, or level */ |
Andre Noll | 664e7c4 | 2009-06-18 08:45:27 +1000 | [diff] [blame] | 2152 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2153 | mddev->layout != mddev->new_layout || |
| 2154 | mddev->level != mddev->new_level) { |
Andre Noll | 664e7c4 | 2009-06-18 08:45:27 +1000 | [diff] [blame] | 2155 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2156 | mddev->new_layout = mddev->layout; |
| 2157 | mddev->new_level = mddev->level; |
| 2158 | return -EINVAL; |
| 2159 | } |
| 2160 | |
Dan Williams | b5470dc | 2008-06-27 21:44:04 -0700 | [diff] [blame] | 2161 | err = md_allow_write(mddev); |
| 2162 | if (err) |
| 2163 | return err; |
NeilBrown | 2a2275d | 2007-01-26 00:57:11 -0800 | [diff] [blame] | 2164 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2165 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
| 2166 | |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2167 | if (raid_disks < conf->raid_disks) { |
| 2168 | cnt=0; |
| 2169 | for (d= 0; d < conf->raid_disks; d++) |
| 2170 | if (conf->mirrors[d].rdev) |
| 2171 | cnt++; |
| 2172 | if (cnt > raid_disks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2173 | return -EBUSY; |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2174 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2175 | |
| 2176 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); |
| 2177 | if (!newpoolinfo) |
| 2178 | return -ENOMEM; |
| 2179 | newpoolinfo->mddev = mddev; |
| 2180 | newpoolinfo->raid_disks = raid_disks; |
| 2181 | |
| 2182 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
| 2183 | r1bio_pool_free, newpoolinfo); |
| 2184 | if (!newpool) { |
| 2185 | kfree(newpoolinfo); |
| 2186 | return -ENOMEM; |
| 2187 | } |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 2188 | newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2189 | if (!newmirrors) { |
| 2190 | kfree(newpoolinfo); |
| 2191 | mempool_destroy(newpool); |
| 2192 | return -ENOMEM; |
| 2193 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2195 | raise_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2196 | |
| 2197 | /* ok, everything is stopped */ |
| 2198 | oldpool = conf->r1bio_pool; |
| 2199 | conf->r1bio_pool = newpool; |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2200 | |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 2201 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
| 2202 | mdk_rdev_t *rdev = conf->mirrors[d].rdev; |
| 2203 | if (rdev && rdev->raid_disk != d2) { |
| 2204 | char nm[20]; |
| 2205 | sprintf(nm, "rd%d", rdev->raid_disk); |
| 2206 | sysfs_remove_link(&mddev->kobj, nm); |
| 2207 | rdev->raid_disk = d2; |
| 2208 | sprintf(nm, "rd%d", rdev->raid_disk); |
| 2209 | sysfs_remove_link(&mddev->kobj, nm); |
| 2210 | if (sysfs_create_link(&mddev->kobj, |
| 2211 | &rdev->kobj, nm)) |
| 2212 | printk(KERN_WARNING |
NeilBrown | 9dd1e2f | 2010-05-03 14:30:35 +1000 | [diff] [blame] | 2213 | "md/raid1:%s: cannot register " |
| 2214 | "%s\n", |
| 2215 | mdname(mddev), nm); |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2216 | } |
NeilBrown | a88aa78 | 2007-08-22 14:01:53 -0700 | [diff] [blame] | 2217 | if (rdev) |
| 2218 | newmirrors[d2++].rdev = rdev; |
| 2219 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2220 | kfree(conf->mirrors); |
| 2221 | conf->mirrors = newmirrors; |
| 2222 | kfree(conf->poolinfo); |
| 2223 | conf->poolinfo = newpoolinfo; |
| 2224 | |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 2225 | spin_lock_irqsave(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2226 | mddev->degraded += (raid_disks - conf->raid_disks); |
NeilBrown | c04be0a | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 2227 | spin_unlock_irqrestore(&conf->device_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2228 | conf->raid_disks = mddev->raid_disks = raid_disks; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2229 | mddev->delta_disks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2230 | |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2231 | conf->last_used = 0; /* just make sure it is in-range */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2232 | lower_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2233 | |
| 2234 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
| 2235 | md_wakeup_thread(mddev->thread); |
| 2236 | |
| 2237 | mempool_destroy(oldpool); |
| 2238 | return 0; |
| 2239 | } |
| 2240 | |
NeilBrown | 500af87 | 2005-09-09 16:23:58 -0700 | [diff] [blame] | 2241 | static void raid1_quiesce(mddev_t *mddev, int state) |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2242 | { |
NeilBrown | 070ec55 | 2009-06-16 16:54:21 +1000 | [diff] [blame] | 2243 | conf_t *conf = mddev->private; |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2244 | |
| 2245 | switch(state) { |
NeilBrown | 6eef4b2 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2246 | case 2: /* wake for suspend */ |
| 2247 | wake_up(&conf->wait_barrier); |
| 2248 | break; |
NeilBrown | 9e6603d | 2005-09-09 16:23:48 -0700 | [diff] [blame] | 2249 | case 1: |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2250 | raise_barrier(conf); |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2251 | break; |
NeilBrown | 9e6603d | 2005-09-09 16:23:48 -0700 | [diff] [blame] | 2252 | case 0: |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2253 | lower_barrier(conf); |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2254 | break; |
| 2255 | } |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2256 | } |
| 2257 | |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2258 | static void *raid1_takeover(mddev_t *mddev) |
| 2259 | { |
| 2260 | /* raid1 can take over: |
| 2261 | * raid5 with 2 devices, any layout or chunk size |
| 2262 | */ |
| 2263 | if (mddev->level == 5 && mddev->raid_disks == 2) { |
| 2264 | conf_t *conf; |
| 2265 | mddev->new_level = 1; |
| 2266 | mddev->new_layout = 0; |
| 2267 | mddev->new_chunk_sectors = 0; |
| 2268 | conf = setup_conf(mddev); |
| 2269 | if (!IS_ERR(conf)) |
| 2270 | conf->barrier = 1; |
| 2271 | return conf; |
| 2272 | } |
| 2273 | return ERR_PTR(-EINVAL); |
| 2274 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2275 | |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2276 | static struct mdk_personality raid1_personality = |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2277 | { |
| 2278 | .name = "raid1", |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2279 | .level = 1, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2280 | .owner = THIS_MODULE, |
| 2281 | .make_request = make_request, |
| 2282 | .run = run, |
| 2283 | .stop = stop, |
| 2284 | .status = status, |
| 2285 | .error_handler = error, |
| 2286 | .hot_add_disk = raid1_add_disk, |
| 2287 | .hot_remove_disk= raid1_remove_disk, |
| 2288 | .spare_active = raid1_spare_active, |
| 2289 | .sync_request = sync_request, |
| 2290 | .resize = raid1_resize, |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 2291 | .size = raid1_size, |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2292 | .check_reshape = raid1_reshape, |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2293 | .quiesce = raid1_quiesce, |
NeilBrown | 709ae48 | 2009-12-14 12:49:51 +1100 | [diff] [blame] | 2294 | .takeover = raid1_takeover, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2295 | }; |
| 2296 | |
| 2297 | static int __init raid_init(void) |
| 2298 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2299 | return register_md_personality(&raid1_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2300 | } |
| 2301 | |
| 2302 | static void raid_exit(void) |
| 2303 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2304 | unregister_md_personality(&raid1_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2305 | } |
| 2306 | |
| 2307 | module_init(raid_init); |
| 2308 | module_exit(raid_exit); |
| 2309 | MODULE_LICENSE("GPL"); |
NeilBrown | 0efb9e6 | 2009-12-14 12:49:58 +1100 | [diff] [blame] | 2310 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2311 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
NeilBrown | d9d166c | 2006-01-06 00:20:51 -0800 | [diff] [blame] | 2312 | MODULE_ALIAS("md-raid1"); |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2313 | MODULE_ALIAS("md-level-1"); |