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 | * |
| 12 | * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> |
| 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 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 34 | #include "dm-bio-list.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include <linux/raid/raid1.h> |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 36 | #include <linux/raid/bitmap.h> |
| 37 | |
| 38 | #define DEBUG 0 |
| 39 | #if DEBUG |
| 40 | #define PRINTK(x...) printk(x) |
| 41 | #else |
| 42 | #define PRINTK(x...) |
| 43 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | |
| 45 | /* |
| 46 | * Number of guaranteed r1bios in case of extreme VM load: |
| 47 | */ |
| 48 | #define NR_RAID1_BIOS 256 |
| 49 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | |
| 51 | static void unplug_slaves(mddev_t *mddev); |
| 52 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 53 | static void allow_barrier(conf_t *conf); |
| 54 | static void lower_barrier(conf_t *conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 56 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | { |
| 58 | struct pool_info *pi = data; |
| 59 | r1bio_t *r1_bio; |
| 60 | int size = offsetof(r1bio_t, bios[pi->raid_disks]); |
| 61 | |
| 62 | /* allocate a r1bio with room for raid_disks entries in the bios array */ |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 63 | r1_bio = kzalloc(size, gfp_flags); |
| 64 | if (!r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | unplug_slaves(pi->mddev); |
| 66 | |
| 67 | return r1_bio; |
| 68 | } |
| 69 | |
| 70 | static void r1bio_pool_free(void *r1_bio, void *data) |
| 71 | { |
| 72 | kfree(r1_bio); |
| 73 | } |
| 74 | |
| 75 | #define RESYNC_BLOCK_SIZE (64*1024) |
| 76 | //#define RESYNC_BLOCK_SIZE PAGE_SIZE |
| 77 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
| 78 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
| 79 | #define RESYNC_WINDOW (2048*1024) |
| 80 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 81 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | { |
| 83 | struct pool_info *pi = data; |
| 84 | struct page *page; |
| 85 | r1bio_t *r1_bio; |
| 86 | struct bio *bio; |
| 87 | int i, j; |
| 88 | |
| 89 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); |
| 90 | if (!r1_bio) { |
| 91 | unplug_slaves(pi->mddev); |
| 92 | return NULL; |
| 93 | } |
| 94 | |
| 95 | /* |
| 96 | * Allocate bios : 1 for reading, n-1 for writing |
| 97 | */ |
| 98 | for (j = pi->raid_disks ; j-- ; ) { |
| 99 | bio = bio_alloc(gfp_flags, RESYNC_PAGES); |
| 100 | if (!bio) |
| 101 | goto out_free_bio; |
| 102 | r1_bio->bios[j] = bio; |
| 103 | } |
| 104 | /* |
| 105 | * Allocate RESYNC_PAGES data pages and attach them to |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 106 | * the first bio. |
| 107 | * If this is a user-requested check/repair, allocate |
| 108 | * RESYNC_PAGES for each bio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | */ |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 110 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
| 111 | j = pi->raid_disks; |
| 112 | else |
| 113 | j = 1; |
| 114 | while(j--) { |
| 115 | bio = r1_bio->bios[j]; |
| 116 | for (i = 0; i < RESYNC_PAGES; i++) { |
| 117 | page = alloc_page(gfp_flags); |
| 118 | if (unlikely(!page)) |
| 119 | goto out_free_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 121 | bio->bi_io_vec[i].bv_page = page; |
| 122 | } |
| 123 | } |
| 124 | /* If not user-requests, copy the page pointers to all bios */ |
| 125 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { |
| 126 | for (i=0; i<RESYNC_PAGES ; i++) |
| 127 | for (j=1; j<pi->raid_disks; j++) |
| 128 | r1_bio->bios[j]->bi_io_vec[i].bv_page = |
| 129 | r1_bio->bios[0]->bi_io_vec[i].bv_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | } |
| 131 | |
| 132 | r1_bio->master_bio = NULL; |
| 133 | |
| 134 | return r1_bio; |
| 135 | |
| 136 | out_free_pages: |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 137 | for (i=0; i < RESYNC_PAGES ; i++) |
| 138 | for (j=0 ; j < pi->raid_disks; j++) |
NeilBrown | 1345b1d | 2006-01-06 00:20:40 -0800 | [diff] [blame] | 139 | safe_put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 140 | j = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | out_free_bio: |
| 142 | while ( ++j < pi->raid_disks ) |
| 143 | bio_put(r1_bio->bios[j]); |
| 144 | r1bio_pool_free(r1_bio, data); |
| 145 | return NULL; |
| 146 | } |
| 147 | |
| 148 | static void r1buf_pool_free(void *__r1_bio, void *data) |
| 149 | { |
| 150 | struct pool_info *pi = data; |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 151 | int i,j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | r1bio_t *r1bio = __r1_bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 154 | for (i = 0; i < RESYNC_PAGES; i++) |
| 155 | for (j = pi->raid_disks; j-- ;) { |
| 156 | if (j == 0 || |
| 157 | r1bio->bios[j]->bi_io_vec[i].bv_page != |
| 158 | r1bio->bios[0]->bi_io_vec[i].bv_page) |
NeilBrown | 1345b1d | 2006-01-06 00:20:40 -0800 | [diff] [blame] | 159 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 160 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | for (i=0 ; i < pi->raid_disks; i++) |
| 162 | bio_put(r1bio->bios[i]); |
| 163 | |
| 164 | r1bio_pool_free(r1bio, data); |
| 165 | } |
| 166 | |
| 167 | static void put_all_bios(conf_t *conf, r1bio_t *r1_bio) |
| 168 | { |
| 169 | int i; |
| 170 | |
| 171 | for (i = 0; i < conf->raid_disks; i++) { |
| 172 | struct bio **bio = r1_bio->bios + i; |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 173 | if (*bio && *bio != IO_BLOCKED) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | bio_put(*bio); |
| 175 | *bio = NULL; |
| 176 | } |
| 177 | } |
| 178 | |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 179 | static void free_r1bio(r1bio_t *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | conf_t *conf = mddev_to_conf(r1_bio->mddev); |
| 182 | |
| 183 | /* |
| 184 | * Wake up any possible resync thread that waits for the device |
| 185 | * to go idle. |
| 186 | */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 187 | allow_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | |
| 189 | put_all_bios(conf, r1_bio); |
| 190 | mempool_free(r1_bio, conf->r1bio_pool); |
| 191 | } |
| 192 | |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 193 | static void put_buf(r1bio_t *r1_bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | { |
| 195 | conf_t *conf = mddev_to_conf(r1_bio->mddev); |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 196 | int i; |
| 197 | |
| 198 | for (i=0; i<conf->raid_disks; i++) { |
| 199 | struct bio *bio = r1_bio->bios[i]; |
| 200 | if (bio->bi_end_io) |
| 201 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); |
| 202 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | |
| 204 | mempool_free(r1_bio, conf->r1buf_pool); |
| 205 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 206 | lower_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | } |
| 208 | |
| 209 | static void reschedule_retry(r1bio_t *r1_bio) |
| 210 | { |
| 211 | unsigned long flags; |
| 212 | mddev_t *mddev = r1_bio->mddev; |
| 213 | conf_t *conf = mddev_to_conf(mddev); |
| 214 | |
| 215 | spin_lock_irqsave(&conf->device_lock, flags); |
| 216 | list_add(&r1_bio->retry_list, &conf->retry_list); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 217 | conf->nr_queued ++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 218 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 219 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 220 | wake_up(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | md_wakeup_thread(mddev->thread); |
| 222 | } |
| 223 | |
| 224 | /* |
| 225 | * raid_end_bio_io() is called when we have finished servicing a mirrored |
| 226 | * operation and are ready to return a success/failure code to the buffer |
| 227 | * cache layer. |
| 228 | */ |
| 229 | static void raid_end_bio_io(r1bio_t *r1_bio) |
| 230 | { |
| 231 | struct bio *bio = r1_bio->master_bio; |
| 232 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 233 | /* if nobody has done the final endio yet, do it now */ |
| 234 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { |
| 235 | PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n", |
| 236 | (bio_data_dir(bio) == WRITE) ? "write" : "read", |
| 237 | (unsigned long long) bio->bi_sector, |
| 238 | (unsigned long long) bio->bi_sector + |
| 239 | (bio->bi_size >> 9) - 1); |
| 240 | |
| 241 | bio_endio(bio, bio->bi_size, |
| 242 | test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO); |
| 243 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | free_r1bio(r1_bio); |
| 245 | } |
| 246 | |
| 247 | /* |
| 248 | * Update disk head position estimator based on IRQ completion info. |
| 249 | */ |
| 250 | static inline void update_head_pos(int disk, r1bio_t *r1_bio) |
| 251 | { |
| 252 | conf_t *conf = mddev_to_conf(r1_bio->mddev); |
| 253 | |
| 254 | conf->mirrors[disk].head_position = |
| 255 | r1_bio->sector + (r1_bio->sectors); |
| 256 | } |
| 257 | |
| 258 | static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error) |
| 259 | { |
| 260 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
| 261 | r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); |
| 262 | int mirror; |
| 263 | conf_t *conf = mddev_to_conf(r1_bio->mddev); |
| 264 | |
| 265 | if (bio->bi_size) |
| 266 | return 1; |
| 267 | |
| 268 | mirror = r1_bio->read_disk; |
| 269 | /* |
| 270 | * this branch is our 'one mirror IO has finished' event handler: |
| 271 | */ |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 272 | update_head_pos(mirror, r1_bio); |
| 273 | |
| 274 | if (uptodate || conf->working_disks <= 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | /* |
| 276 | * Set R1BIO_Uptodate in our master bio, so that |
| 277 | * we will return a good error code for to the higher |
| 278 | * levels even if IO on some other mirrored buffer fails. |
| 279 | * |
| 280 | * The 'master' represents the composite IO operation to |
| 281 | * user-side. So if something waits for IO, then it will |
| 282 | * wait for the 'master' bio. |
| 283 | */ |
NeilBrown | 220946c | 2006-01-06 00:20:27 -0800 | [diff] [blame] | 284 | if (uptodate) |
| 285 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | raid_end_bio_io(r1_bio); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 288 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | /* |
| 290 | * oops, read error: |
| 291 | */ |
| 292 | char b[BDEVNAME_SIZE]; |
| 293 | if (printk_ratelimit()) |
| 294 | printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n", |
| 295 | bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector); |
| 296 | reschedule_retry(r1_bio); |
| 297 | } |
| 298 | |
| 299 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
| 300 | return 0; |
| 301 | } |
| 302 | |
| 303 | static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error) |
| 304 | { |
| 305 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
| 306 | r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 307 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | conf_t *conf = mddev_to_conf(r1_bio->mddev); |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 309 | struct bio *to_put = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | |
| 311 | if (bio->bi_size) |
| 312 | return 1; |
| 313 | |
| 314 | for (mirror = 0; mirror < conf->raid_disks; mirror++) |
| 315 | if (r1_bio->bios[mirror] == bio) |
| 316 | break; |
| 317 | |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 318 | if (error == -ENOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) { |
| 319 | set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags); |
| 320 | set_bit(R1BIO_BarrierRetry, &r1_bio->state); |
| 321 | r1_bio->mddev->barriers_work = 0; |
| 322 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | /* |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 324 | * this branch is our 'one mirror IO has finished' event handler: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 325 | */ |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 326 | r1_bio->bios[mirror] = NULL; |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 327 | to_put = bio; |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 328 | if (!uptodate) { |
| 329 | md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); |
| 330 | /* an I/O failed, we can't clear the bitmap */ |
| 331 | set_bit(R1BIO_Degraded, &r1_bio->state); |
| 332 | } else |
| 333 | /* |
| 334 | * Set R1BIO_Uptodate in our master bio, so that |
| 335 | * we will return a good error code for to the higher |
| 336 | * levels even if IO on some other mirrored buffer fails. |
| 337 | * |
| 338 | * The 'master' represents the composite IO operation to |
| 339 | * user-side. So if something waits for IO, then it will |
| 340 | * wait for the 'master' bio. |
| 341 | */ |
| 342 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 344 | update_head_pos(mirror, r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 346 | if (behind) { |
| 347 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) |
| 348 | atomic_dec(&r1_bio->behind_remaining); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 349 | |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 350 | /* In behind mode, we ACK the master bio once the I/O has safely |
| 351 | * reached all non-writemostly disks. Setting the Returned bit |
| 352 | * ensures that this gets done only once -- we don't ever want to |
| 353 | * return -EIO here, instead we'll wait */ |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 354 | |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 355 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && |
| 356 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
| 357 | /* Maybe we can return now */ |
| 358 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { |
| 359 | struct bio *mbio = r1_bio->master_bio; |
| 360 | PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n", |
| 361 | (unsigned long long) mbio->bi_sector, |
| 362 | (unsigned long long) mbio->bi_sector + |
| 363 | (mbio->bi_size >> 9) - 1); |
| 364 | bio_endio(mbio, mbio->bi_size, 0); |
| 365 | } |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 366 | } |
| 367 | } |
| 368 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | /* |
| 370 | * |
| 371 | * Let's see if all mirrored write operations have finished |
| 372 | * already. |
| 373 | */ |
| 374 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 375 | if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) { |
| 376 | reschedule_retry(r1_bio); |
| 377 | /* Don't dec_pending yet, we want to hold |
| 378 | * the reference over the retry |
| 379 | */ |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 380 | goto out; |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 381 | } |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 382 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
| 383 | /* free extra copy of the data pages */ |
| 384 | int i = bio->bi_vcnt; |
| 385 | while (i--) |
NeilBrown | 1345b1d | 2006-01-06 00:20:40 -0800 | [diff] [blame] | 386 | safe_put_page(bio->bi_io_vec[i].bv_page); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 387 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 388 | /* clear the bitmap if all writes complete successfully */ |
| 389 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
| 390 | r1_bio->sectors, |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 391 | !test_bit(R1BIO_Degraded, &r1_bio->state), |
| 392 | behind); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 393 | md_write_end(r1_bio->mddev); |
| 394 | raid_end_bio_io(r1_bio); |
| 395 | } |
| 396 | |
| 397 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 398 | out: |
| 399 | if (to_put) |
| 400 | bio_put(to_put); |
| 401 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | return 0; |
| 403 | } |
| 404 | |
| 405 | |
| 406 | /* |
| 407 | * This routine returns the disk from which the requested read should |
| 408 | * be done. There is a per-array 'next expected sequential IO' sector |
| 409 | * number - if this matches on the next IO then we use the last disk. |
| 410 | * There is also a per-disk 'last know head position' sector that is |
| 411 | * maintained from IRQ contexts, both the normal and the resync IO |
| 412 | * completion handlers update this position correctly. If there is no |
| 413 | * perfect sequential match then we pick the disk whose head is closest. |
| 414 | * |
| 415 | * If there are 2 mirrors in the same 2 devices, performance degrades |
| 416 | * because position is mirror, not device based. |
| 417 | * |
| 418 | * The rdev for the device selected will have nr_pending incremented. |
| 419 | */ |
| 420 | static int read_balance(conf_t *conf, r1bio_t *r1_bio) |
| 421 | { |
| 422 | const unsigned long this_sector = r1_bio->sector; |
| 423 | int new_disk = conf->last_used, disk = new_disk; |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 424 | int wonly_disk = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 425 | const int sectors = r1_bio->sectors; |
| 426 | sector_t new_distance, current_distance; |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 427 | mdk_rdev_t *rdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | |
| 429 | rcu_read_lock(); |
| 430 | /* |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 431 | * Check if we can balance. We can balance on the whole |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | * device if no resync is going on, or below the resync window. |
| 433 | * We take the first readable disk when above the resync window. |
| 434 | */ |
| 435 | retry: |
| 436 | if (conf->mddev->recovery_cp < MaxSector && |
| 437 | (this_sector + sectors >= conf->next_resync)) { |
| 438 | /* Choose the first operation device, for consistancy */ |
| 439 | new_disk = 0; |
| 440 | |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 441 | for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev); |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 442 | r1_bio->bios[new_disk] == IO_BLOCKED || |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 443 | !rdev || !test_bit(In_sync, &rdev->flags) |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 444 | || test_bit(WriteMostly, &rdev->flags); |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 445 | rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) { |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 446 | |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 447 | if (rdev && test_bit(In_sync, &rdev->flags) && |
| 448 | r1_bio->bios[new_disk] != IO_BLOCKED) |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 449 | wonly_disk = new_disk; |
| 450 | |
| 451 | if (new_disk == conf->raid_disks - 1) { |
| 452 | new_disk = wonly_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | break; |
| 454 | } |
| 455 | } |
| 456 | goto rb_out; |
| 457 | } |
| 458 | |
| 459 | |
| 460 | /* make sure the disk is operational */ |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 461 | for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev); |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 462 | r1_bio->bios[new_disk] == IO_BLOCKED || |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 463 | !rdev || !test_bit(In_sync, &rdev->flags) || |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 464 | test_bit(WriteMostly, &rdev->flags); |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 465 | rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) { |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 466 | |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 467 | if (rdev && test_bit(In_sync, &rdev->flags) && |
| 468 | r1_bio->bios[new_disk] != IO_BLOCKED) |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 469 | wonly_disk = new_disk; |
| 470 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 471 | if (new_disk <= 0) |
| 472 | new_disk = conf->raid_disks; |
| 473 | new_disk--; |
| 474 | if (new_disk == disk) { |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 475 | new_disk = wonly_disk; |
| 476 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 477 | } |
| 478 | } |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 479 | |
| 480 | if (new_disk < 0) |
| 481 | goto rb_out; |
| 482 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | disk = new_disk; |
| 484 | /* now disk == new_disk == starting point for search */ |
| 485 | |
| 486 | /* |
| 487 | * Don't change to another disk for sequential reads: |
| 488 | */ |
| 489 | if (conf->next_seq_sect == this_sector) |
| 490 | goto rb_out; |
| 491 | if (this_sector == conf->mirrors[new_disk].head_position) |
| 492 | goto rb_out; |
| 493 | |
| 494 | current_distance = abs(this_sector - conf->mirrors[disk].head_position); |
| 495 | |
| 496 | /* Find the disk whose head is closest */ |
| 497 | |
| 498 | do { |
| 499 | if (disk <= 0) |
| 500 | disk = conf->raid_disks; |
| 501 | disk--; |
| 502 | |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 503 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 504 | |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 505 | if (!rdev || r1_bio->bios[disk] == IO_BLOCKED || |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 506 | !test_bit(In_sync, &rdev->flags) || |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 507 | test_bit(WriteMostly, &rdev->flags)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 508 | continue; |
| 509 | |
| 510 | if (!atomic_read(&rdev->nr_pending)) { |
| 511 | new_disk = disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | break; |
| 513 | } |
| 514 | new_distance = abs(this_sector - conf->mirrors[disk].head_position); |
| 515 | if (new_distance < current_distance) { |
| 516 | current_distance = new_distance; |
| 517 | new_disk = disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | } |
| 519 | } while (disk != conf->last_used); |
| 520 | |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 521 | rb_out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 522 | |
| 523 | |
| 524 | if (new_disk >= 0) { |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 525 | rdev = rcu_dereference(conf->mirrors[new_disk].rdev); |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 526 | if (!rdev) |
| 527 | goto retry; |
| 528 | atomic_inc(&rdev->nr_pending); |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 529 | if (!test_bit(In_sync, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | /* cannot risk returning a device that failed |
| 531 | * before we inc'ed nr_pending |
| 532 | */ |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 533 | rdev_dec_pending(rdev, conf->mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | goto retry; |
| 535 | } |
NeilBrown | 8ddf9ef | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 536 | conf->next_seq_sect = this_sector + sectors; |
| 537 | conf->last_used = new_disk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | } |
| 539 | rcu_read_unlock(); |
| 540 | |
| 541 | return new_disk; |
| 542 | } |
| 543 | |
| 544 | static void unplug_slaves(mddev_t *mddev) |
| 545 | { |
| 546 | conf_t *conf = mddev_to_conf(mddev); |
| 547 | int i; |
| 548 | |
| 549 | rcu_read_lock(); |
| 550 | for (i=0; i<mddev->raid_disks; i++) { |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 551 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 552 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | request_queue_t *r_queue = bdev_get_queue(rdev->bdev); |
| 554 | |
| 555 | atomic_inc(&rdev->nr_pending); |
| 556 | rcu_read_unlock(); |
| 557 | |
| 558 | if (r_queue->unplug_fn) |
| 559 | r_queue->unplug_fn(r_queue); |
| 560 | |
| 561 | rdev_dec_pending(rdev, mddev); |
| 562 | rcu_read_lock(); |
| 563 | } |
| 564 | } |
| 565 | rcu_read_unlock(); |
| 566 | } |
| 567 | |
| 568 | static void raid1_unplug(request_queue_t *q) |
| 569 | { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 570 | mddev_t *mddev = q->queuedata; |
| 571 | |
| 572 | unplug_slaves(mddev); |
| 573 | md_wakeup_thread(mddev->thread); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | } |
| 575 | |
| 576 | static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk, |
| 577 | sector_t *error_sector) |
| 578 | { |
| 579 | mddev_t *mddev = q->queuedata; |
| 580 | conf_t *conf = mddev_to_conf(mddev); |
| 581 | int i, ret = 0; |
| 582 | |
| 583 | rcu_read_lock(); |
| 584 | for (i=0; i<mddev->raid_disks && ret == 0; i++) { |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 585 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 586 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | struct block_device *bdev = rdev->bdev; |
| 588 | request_queue_t *r_queue = bdev_get_queue(bdev); |
| 589 | |
| 590 | if (!r_queue->issue_flush_fn) |
| 591 | ret = -EOPNOTSUPP; |
| 592 | else { |
| 593 | atomic_inc(&rdev->nr_pending); |
| 594 | rcu_read_unlock(); |
| 595 | ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, |
| 596 | error_sector); |
| 597 | rdev_dec_pending(rdev, mddev); |
| 598 | rcu_read_lock(); |
| 599 | } |
| 600 | } |
| 601 | } |
| 602 | rcu_read_unlock(); |
| 603 | return ret; |
| 604 | } |
| 605 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 606 | /* Barriers.... |
| 607 | * Sometimes we need to suspend IO while we do something else, |
| 608 | * either some resync/recovery, or reconfigure the array. |
| 609 | * To do this we raise a 'barrier'. |
| 610 | * The 'barrier' is a counter that can be raised multiple times |
| 611 | * to count how many activities are happening which preclude |
| 612 | * normal IO. |
| 613 | * We can only raise the barrier if there is no pending IO. |
| 614 | * i.e. if nr_pending == 0. |
| 615 | * We choose only to raise the barrier if no-one is waiting for the |
| 616 | * barrier to go down. This means that as soon as an IO request |
| 617 | * is ready, no other operations which require a barrier will start |
| 618 | * until the IO request has had a chance. |
| 619 | * |
| 620 | * So: regular IO calls 'wait_barrier'. When that returns there |
| 621 | * is no backgroup IO happening, It must arrange to call |
| 622 | * allow_barrier when it has finished its IO. |
| 623 | * backgroup IO calls must call raise_barrier. Once that returns |
| 624 | * there is no normal IO happeing. It must arrange to call |
| 625 | * lower_barrier when the particular background IO completes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | */ |
| 627 | #define RESYNC_DEPTH 32 |
| 628 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 629 | static void raise_barrier(conf_t *conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | { |
| 631 | spin_lock_irq(&conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 632 | |
| 633 | /* Wait until no block IO is waiting */ |
| 634 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, |
| 635 | conf->resync_lock, |
| 636 | raid1_unplug(conf->mddev->queue)); |
| 637 | |
| 638 | /* block any new IO from starting */ |
| 639 | conf->barrier++; |
| 640 | |
| 641 | /* No wait for all pending IO to complete */ |
| 642 | wait_event_lock_irq(conf->wait_barrier, |
| 643 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, |
| 644 | conf->resync_lock, |
| 645 | raid1_unplug(conf->mddev->queue)); |
| 646 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | spin_unlock_irq(&conf->resync_lock); |
| 648 | } |
| 649 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 650 | static void lower_barrier(conf_t *conf) |
| 651 | { |
| 652 | unsigned long flags; |
| 653 | spin_lock_irqsave(&conf->resync_lock, flags); |
| 654 | conf->barrier--; |
| 655 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
| 656 | wake_up(&conf->wait_barrier); |
| 657 | } |
| 658 | |
| 659 | static void wait_barrier(conf_t *conf) |
| 660 | { |
| 661 | spin_lock_irq(&conf->resync_lock); |
| 662 | if (conf->barrier) { |
| 663 | conf->nr_waiting++; |
| 664 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, |
| 665 | conf->resync_lock, |
| 666 | raid1_unplug(conf->mddev->queue)); |
| 667 | conf->nr_waiting--; |
| 668 | } |
| 669 | conf->nr_pending++; |
| 670 | spin_unlock_irq(&conf->resync_lock); |
| 671 | } |
| 672 | |
| 673 | static void allow_barrier(conf_t *conf) |
| 674 | { |
| 675 | unsigned long flags; |
| 676 | spin_lock_irqsave(&conf->resync_lock, flags); |
| 677 | conf->nr_pending--; |
| 678 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
| 679 | wake_up(&conf->wait_barrier); |
| 680 | } |
| 681 | |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 682 | static void freeze_array(conf_t *conf) |
| 683 | { |
| 684 | /* stop syncio and normal IO and wait for everything to |
| 685 | * go quite. |
| 686 | * We increment barrier and nr_waiting, and then |
| 687 | * wait until barrier+nr_pending match nr_queued+2 |
| 688 | */ |
| 689 | spin_lock_irq(&conf->resync_lock); |
| 690 | conf->barrier++; |
| 691 | conf->nr_waiting++; |
| 692 | wait_event_lock_irq(conf->wait_barrier, |
| 693 | conf->barrier+conf->nr_pending == conf->nr_queued+2, |
| 694 | conf->resync_lock, |
| 695 | raid1_unplug(conf->mddev->queue)); |
| 696 | spin_unlock_irq(&conf->resync_lock); |
| 697 | } |
| 698 | static void unfreeze_array(conf_t *conf) |
| 699 | { |
| 700 | /* reverse the effect of the freeze */ |
| 701 | spin_lock_irq(&conf->resync_lock); |
| 702 | conf->barrier--; |
| 703 | conf->nr_waiting--; |
| 704 | wake_up(&conf->wait_barrier); |
| 705 | spin_unlock_irq(&conf->resync_lock); |
| 706 | } |
| 707 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 708 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 709 | /* duplicate the data pages for behind I/O */ |
| 710 | static struct page **alloc_behind_pages(struct bio *bio) |
| 711 | { |
| 712 | int i; |
| 713 | struct bio_vec *bvec; |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 714 | struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *), |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 715 | GFP_NOIO); |
| 716 | if (unlikely(!pages)) |
| 717 | goto do_sync_io; |
| 718 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 719 | bio_for_each_segment(bvec, bio, i) { |
| 720 | pages[i] = alloc_page(GFP_NOIO); |
| 721 | if (unlikely(!pages[i])) |
| 722 | goto do_sync_io; |
| 723 | memcpy(kmap(pages[i]) + bvec->bv_offset, |
| 724 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); |
| 725 | kunmap(pages[i]); |
| 726 | kunmap(bvec->bv_page); |
| 727 | } |
| 728 | |
| 729 | return pages; |
| 730 | |
| 731 | do_sync_io: |
| 732 | if (pages) |
| 733 | for (i = 0; i < bio->bi_vcnt && pages[i]; i++) |
NeilBrown | 2d1f3b5 | 2006-01-06 00:20:31 -0800 | [diff] [blame] | 734 | put_page(pages[i]); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 735 | kfree(pages); |
| 736 | PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); |
| 737 | return NULL; |
| 738 | } |
| 739 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | static int make_request(request_queue_t *q, struct bio * bio) |
| 741 | { |
| 742 | mddev_t *mddev = q->queuedata; |
| 743 | conf_t *conf = mddev_to_conf(mddev); |
| 744 | mirror_info_t *mirror; |
| 745 | r1bio_t *r1_bio; |
| 746 | struct bio *read_bio; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 747 | int i, targets = 0, disks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 748 | mdk_rdev_t *rdev; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 749 | struct bitmap *bitmap = mddev->bitmap; |
| 750 | unsigned long flags; |
| 751 | struct bio_list bl; |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 752 | struct page **behind_pages = NULL; |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 753 | const int rw = bio_data_dir(bio); |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 754 | int do_barriers; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 755 | |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 756 | if (unlikely(!mddev->barriers_work && bio_barrier(bio))) { |
NeilBrown | e5dcdd8 | 2005-09-09 16:23:41 -0700 | [diff] [blame] | 757 | bio_endio(bio, bio->bi_size, -EOPNOTSUPP); |
| 758 | return 0; |
| 759 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | |
| 761 | /* |
| 762 | * Register the new request and wait if the reconstruction |
| 763 | * thread has put up a bar for new requests. |
| 764 | * Continue immediately if no resync is active currently. |
| 765 | */ |
NeilBrown | 3d310eb | 2005-06-21 17:17:26 -0700 | [diff] [blame] | 766 | md_write_start(mddev, bio); /* wait on superblock update early */ |
| 767 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 768 | wait_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 769 | |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 770 | disk_stat_inc(mddev->gendisk, ios[rw]); |
| 771 | disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | |
| 773 | /* |
| 774 | * make_request() can abort the operation when READA is being |
| 775 | * used and no empty request is available. |
| 776 | * |
| 777 | */ |
| 778 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); |
| 779 | |
| 780 | r1_bio->master_bio = bio; |
| 781 | r1_bio->sectors = bio->bi_size >> 9; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 782 | r1_bio->state = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | r1_bio->mddev = mddev; |
| 784 | r1_bio->sector = bio->bi_sector; |
| 785 | |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 786 | if (rw == READ) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 787 | /* |
| 788 | * read balancing logic: |
| 789 | */ |
| 790 | int rdisk = read_balance(conf, r1_bio); |
| 791 | |
| 792 | if (rdisk < 0) { |
| 793 | /* couldn't find anywhere to read from */ |
| 794 | raid_end_bio_io(r1_bio); |
| 795 | return 0; |
| 796 | } |
| 797 | mirror = conf->mirrors + rdisk; |
| 798 | |
| 799 | r1_bio->read_disk = rdisk; |
| 800 | |
| 801 | read_bio = bio_clone(bio, GFP_NOIO); |
| 802 | |
| 803 | r1_bio->bios[rdisk] = read_bio; |
| 804 | |
| 805 | read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; |
| 806 | read_bio->bi_bdev = mirror->rdev->bdev; |
| 807 | read_bio->bi_end_io = raid1_end_read_request; |
| 808 | read_bio->bi_rw = READ; |
| 809 | read_bio->bi_private = r1_bio; |
| 810 | |
| 811 | generic_make_request(read_bio); |
| 812 | return 0; |
| 813 | } |
| 814 | |
| 815 | /* |
| 816 | * WRITE: |
| 817 | */ |
| 818 | /* first select target devices under spinlock and |
| 819 | * inc refcount on their rdev. Record them by setting |
| 820 | * bios[x] to bio |
| 821 | */ |
| 822 | disks = conf->raid_disks; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 823 | #if 0 |
| 824 | { static int first=1; |
| 825 | if (first) printk("First Write sector %llu disks %d\n", |
| 826 | (unsigned long long)r1_bio->sector, disks); |
| 827 | first = 0; |
| 828 | } |
| 829 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | rcu_read_lock(); |
| 831 | for (i = 0; i < disks; i++) { |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 832 | if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL && |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 833 | !test_bit(Faulty, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | atomic_inc(&rdev->nr_pending); |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 835 | if (test_bit(Faulty, &rdev->flags)) { |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 836 | rdev_dec_pending(rdev, mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 837 | r1_bio->bios[i] = NULL; |
| 838 | } else |
| 839 | r1_bio->bios[i] = bio; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 840 | targets++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 841 | } else |
| 842 | r1_bio->bios[i] = NULL; |
| 843 | } |
| 844 | rcu_read_unlock(); |
| 845 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 846 | BUG_ON(targets == 0); /* we never fail the last device */ |
| 847 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 848 | if (targets < conf->raid_disks) { |
| 849 | /* array is degraded, we will not clear the bitmap |
| 850 | * on I/O completion (see raid1_end_write_request) */ |
| 851 | set_bit(R1BIO_Degraded, &r1_bio->state); |
| 852 | } |
NeilBrown | 06d91a5 | 2005-06-21 17:17:12 -0700 | [diff] [blame] | 853 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 854 | /* do behind I/O ? */ |
| 855 | if (bitmap && |
| 856 | atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind && |
| 857 | (behind_pages = alloc_behind_pages(bio)) != NULL) |
| 858 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
| 859 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 860 | atomic_set(&r1_bio->remaining, 0); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 861 | atomic_set(&r1_bio->behind_remaining, 0); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 862 | |
NeilBrown | 04b857f | 2006-03-09 17:33:46 -0800 | [diff] [blame] | 863 | do_barriers = bio_barrier(bio); |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 864 | if (do_barriers) |
| 865 | set_bit(R1BIO_Barrier, &r1_bio->state); |
| 866 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 867 | bio_list_init(&bl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 868 | for (i = 0; i < disks; i++) { |
| 869 | struct bio *mbio; |
| 870 | if (!r1_bio->bios[i]) |
| 871 | continue; |
| 872 | |
| 873 | mbio = bio_clone(bio, GFP_NOIO); |
| 874 | r1_bio->bios[i] = mbio; |
| 875 | |
| 876 | mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset; |
| 877 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
| 878 | mbio->bi_end_io = raid1_end_write_request; |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 879 | mbio->bi_rw = WRITE | do_barriers; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 880 | mbio->bi_private = r1_bio; |
| 881 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 882 | if (behind_pages) { |
| 883 | struct bio_vec *bvec; |
| 884 | int j; |
| 885 | |
| 886 | /* Yes, I really want the '__' version so that |
| 887 | * we clear any unused pointer in the io_vec, rather |
| 888 | * than leave them unchanged. This is important |
| 889 | * because when we come to free the pages, we won't |
| 890 | * know the originial bi_idx, so we just free |
| 891 | * them all |
| 892 | */ |
| 893 | __bio_for_each_segment(bvec, mbio, j, 0) |
| 894 | bvec->bv_page = behind_pages[j]; |
| 895 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
| 896 | atomic_inc(&r1_bio->behind_remaining); |
| 897 | } |
| 898 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 899 | atomic_inc(&r1_bio->remaining); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 900 | |
| 901 | bio_list_add(&bl, mbio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 902 | } |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 903 | kfree(behind_pages); /* the behind pages are attached to the bios now */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 904 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 905 | bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors, |
| 906 | test_bit(R1BIO_BehindIO, &r1_bio->state)); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 907 | spin_lock_irqsave(&conf->device_lock, flags); |
| 908 | bio_list_merge(&conf->pending_bio_list, &bl); |
| 909 | bio_list_init(&bl); |
| 910 | |
| 911 | blk_plug_device(mddev->queue); |
| 912 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 913 | |
| 914 | #if 0 |
| 915 | while ((bio = bio_list_pop(&bl)) != NULL) |
| 916 | generic_make_request(bio); |
| 917 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 918 | |
| 919 | return 0; |
| 920 | } |
| 921 | |
| 922 | static void status(struct seq_file *seq, mddev_t *mddev) |
| 923 | { |
| 924 | conf_t *conf = mddev_to_conf(mddev); |
| 925 | int i; |
| 926 | |
| 927 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
| 928 | conf->working_disks); |
| 929 | for (i = 0; i < conf->raid_disks; i++) |
| 930 | seq_printf(seq, "%s", |
| 931 | conf->mirrors[i].rdev && |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 932 | test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 933 | seq_printf(seq, "]"); |
| 934 | } |
| 935 | |
| 936 | |
| 937 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) |
| 938 | { |
| 939 | char b[BDEVNAME_SIZE]; |
| 940 | conf_t *conf = mddev_to_conf(mddev); |
| 941 | |
| 942 | /* |
| 943 | * If it is not operational, then we have already marked it as dead |
| 944 | * else if it is the last working disks, ignore the error, let the |
| 945 | * next level up know. |
| 946 | * else mark the drive as failed |
| 947 | */ |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 948 | if (test_bit(In_sync, &rdev->flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 949 | && conf->working_disks == 1) |
| 950 | /* |
| 951 | * Don't fail the drive, act as though we were just a |
| 952 | * normal single drive |
| 953 | */ |
| 954 | return; |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 955 | if (test_bit(In_sync, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | mddev->degraded++; |
| 957 | conf->working_disks--; |
| 958 | /* |
| 959 | * if recovery is running, make sure it aborts. |
| 960 | */ |
| 961 | set_bit(MD_RECOVERY_ERR, &mddev->recovery); |
| 962 | } |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 963 | clear_bit(In_sync, &rdev->flags); |
| 964 | set_bit(Faulty, &rdev->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 965 | mddev->sb_dirty = 1; |
| 966 | printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n" |
| 967 | " Operation continuing on %d devices\n", |
| 968 | bdevname(rdev->bdev,b), conf->working_disks); |
| 969 | } |
| 970 | |
| 971 | static void print_conf(conf_t *conf) |
| 972 | { |
| 973 | int i; |
| 974 | mirror_info_t *tmp; |
| 975 | |
| 976 | printk("RAID1 conf printout:\n"); |
| 977 | if (!conf) { |
| 978 | printk("(!conf)\n"); |
| 979 | return; |
| 980 | } |
| 981 | printk(" --- wd:%d rd:%d\n", conf->working_disks, |
| 982 | conf->raid_disks); |
| 983 | |
| 984 | for (i = 0; i < conf->raid_disks; i++) { |
| 985 | char b[BDEVNAME_SIZE]; |
| 986 | tmp = conf->mirrors + i; |
| 987 | if (tmp->rdev) |
| 988 | printk(" disk %d, wo:%d, o:%d, dev:%s\n", |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 989 | i, !test_bit(In_sync, &tmp->rdev->flags), !test_bit(Faulty, &tmp->rdev->flags), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 990 | bdevname(tmp->rdev->bdev,b)); |
| 991 | } |
| 992 | } |
| 993 | |
| 994 | static void close_sync(conf_t *conf) |
| 995 | { |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 996 | wait_barrier(conf); |
| 997 | allow_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | |
| 999 | mempool_destroy(conf->r1buf_pool); |
| 1000 | conf->r1buf_pool = NULL; |
| 1001 | } |
| 1002 | |
| 1003 | static int raid1_spare_active(mddev_t *mddev) |
| 1004 | { |
| 1005 | int i; |
| 1006 | conf_t *conf = mddev->private; |
| 1007 | mirror_info_t *tmp; |
| 1008 | |
| 1009 | /* |
| 1010 | * Find all failed disks within the RAID1 configuration |
| 1011 | * and mark them readable |
| 1012 | */ |
| 1013 | for (i = 0; i < conf->raid_disks; i++) { |
| 1014 | tmp = conf->mirrors + i; |
| 1015 | if (tmp->rdev |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1016 | && !test_bit(Faulty, &tmp->rdev->flags) |
| 1017 | && !test_bit(In_sync, &tmp->rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1018 | conf->working_disks++; |
| 1019 | mddev->degraded--; |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1020 | set_bit(In_sync, &tmp->rdev->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1021 | } |
| 1022 | } |
| 1023 | |
| 1024 | print_conf(conf); |
| 1025 | return 0; |
| 1026 | } |
| 1027 | |
| 1028 | |
| 1029 | static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) |
| 1030 | { |
| 1031 | conf_t *conf = mddev->private; |
| 1032 | int found = 0; |
NeilBrown | 41158c7 | 2005-06-21 17:17:25 -0700 | [diff] [blame] | 1033 | int mirror = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | mirror_info_t *p; |
| 1035 | |
| 1036 | for (mirror=0; mirror < mddev->raid_disks; mirror++) |
| 1037 | if ( !(p=conf->mirrors+mirror)->rdev) { |
| 1038 | |
| 1039 | blk_queue_stack_limits(mddev->queue, |
| 1040 | rdev->bdev->bd_disk->queue); |
| 1041 | /* as we don't honour merge_bvec_fn, we must never risk |
| 1042 | * violating it, so limit ->max_sector to one PAGE, as |
| 1043 | * a one page request is never in violation. |
| 1044 | */ |
| 1045 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && |
| 1046 | mddev->queue->max_sectors > (PAGE_SIZE>>9)) |
| 1047 | blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); |
| 1048 | |
| 1049 | p->head_position = 0; |
| 1050 | rdev->raid_disk = mirror; |
| 1051 | found = 1; |
NeilBrown | 6aea114a | 2005-11-28 13:44:13 -0800 | [diff] [blame] | 1052 | /* As all devices are equivalent, we don't need a full recovery |
| 1053 | * if this was recently any drive of the array |
| 1054 | */ |
| 1055 | if (rdev->saved_raid_disk < 0) |
NeilBrown | 41158c7 | 2005-06-21 17:17:25 -0700 | [diff] [blame] | 1056 | conf->fullsync = 1; |
Suzanne Wood | d6065f7 | 2005-11-08 21:39:27 -0800 | [diff] [blame] | 1057 | rcu_assign_pointer(p->rdev, rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | break; |
| 1059 | } |
| 1060 | |
| 1061 | print_conf(conf); |
| 1062 | return found; |
| 1063 | } |
| 1064 | |
| 1065 | static int raid1_remove_disk(mddev_t *mddev, int number) |
| 1066 | { |
| 1067 | conf_t *conf = mddev->private; |
| 1068 | int err = 0; |
| 1069 | mdk_rdev_t *rdev; |
| 1070 | mirror_info_t *p = conf->mirrors+ number; |
| 1071 | |
| 1072 | print_conf(conf); |
| 1073 | rdev = p->rdev; |
| 1074 | if (rdev) { |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1075 | if (test_bit(In_sync, &rdev->flags) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1076 | atomic_read(&rdev->nr_pending)) { |
| 1077 | err = -EBUSY; |
| 1078 | goto abort; |
| 1079 | } |
| 1080 | p->rdev = NULL; |
Paul E. McKenney | fbd568a3e | 2005-05-01 08:59:04 -0700 | [diff] [blame] | 1081 | synchronize_rcu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1082 | if (atomic_read(&rdev->nr_pending)) { |
| 1083 | /* lost the race, try later */ |
| 1084 | err = -EBUSY; |
| 1085 | p->rdev = rdev; |
| 1086 | } |
| 1087 | } |
| 1088 | abort: |
| 1089 | |
| 1090 | print_conf(conf); |
| 1091 | return err; |
| 1092 | } |
| 1093 | |
| 1094 | |
| 1095 | static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error) |
| 1096 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1098 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1099 | |
| 1100 | if (bio->bi_size) |
| 1101 | return 1; |
| 1102 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1103 | for (i=r1_bio->mddev->raid_disks; i--; ) |
| 1104 | if (r1_bio->bios[i] == bio) |
| 1105 | break; |
| 1106 | BUG_ON(i < 0); |
| 1107 | update_head_pos(i, r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1108 | /* |
| 1109 | * we have read a block, now it needs to be re-written, |
| 1110 | * or re-read if the read failed. |
| 1111 | * We don't do much here, just schedule handling by raid1d |
| 1112 | */ |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1113 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1114 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1115 | |
| 1116 | if (atomic_dec_and_test(&r1_bio->remaining)) |
| 1117 | reschedule_retry(r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | return 0; |
| 1119 | } |
| 1120 | |
| 1121 | static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error) |
| 1122 | { |
| 1123 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
| 1124 | r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); |
| 1125 | mddev_t *mddev = r1_bio->mddev; |
| 1126 | conf_t *conf = mddev_to_conf(mddev); |
| 1127 | int i; |
| 1128 | int mirror=0; |
| 1129 | |
| 1130 | if (bio->bi_size) |
| 1131 | return 1; |
| 1132 | |
| 1133 | for (i = 0; i < conf->raid_disks; i++) |
| 1134 | if (r1_bio->bios[i] == bio) { |
| 1135 | mirror = i; |
| 1136 | break; |
| 1137 | } |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1138 | if (!uptodate) { |
| 1139 | int sync_blocks = 0; |
| 1140 | sector_t s = r1_bio->sector; |
| 1141 | long sectors_to_go = r1_bio->sectors; |
| 1142 | /* make sure these bits doesn't get cleared. */ |
| 1143 | do { |
| 1144 | bitmap_end_sync(mddev->bitmap, r1_bio->sector, |
| 1145 | &sync_blocks, 1); |
| 1146 | s += sync_blocks; |
| 1147 | sectors_to_go -= sync_blocks; |
| 1148 | } while (sectors_to_go > 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | md_error(mddev, conf->mirrors[mirror].rdev); |
NeilBrown | 6b1117d | 2006-03-31 02:31:57 -0800 | [diff] [blame] | 1150 | } |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1151 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1152 | update_head_pos(mirror, r1_bio); |
| 1153 | |
| 1154 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
| 1155 | md_done_sync(mddev, r1_bio->sectors, uptodate); |
| 1156 | put_buf(r1_bio); |
| 1157 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1158 | return 0; |
| 1159 | } |
| 1160 | |
| 1161 | static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) |
| 1162 | { |
| 1163 | conf_t *conf = mddev_to_conf(mddev); |
| 1164 | int i; |
| 1165 | int disks = conf->raid_disks; |
| 1166 | struct bio *bio, *wbio; |
| 1167 | |
| 1168 | bio = r1_bio->bios[r1_bio->read_disk]; |
| 1169 | |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1170 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1171 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
| 1172 | /* We have read all readable devices. If we haven't |
| 1173 | * got the block, then there is no hope left. |
| 1174 | * If we have, then we want to do a comparison |
| 1175 | * and skip the write if everything is the same. |
| 1176 | * If any blocks failed to read, then we need to |
| 1177 | * attempt an over-write |
| 1178 | */ |
| 1179 | int primary; |
| 1180 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
| 1181 | for (i=0; i<mddev->raid_disks; i++) |
| 1182 | if (r1_bio->bios[i]->bi_end_io == end_sync_read) |
| 1183 | md_error(mddev, conf->mirrors[i].rdev); |
| 1184 | |
| 1185 | md_done_sync(mddev, r1_bio->sectors, 1); |
| 1186 | put_buf(r1_bio); |
| 1187 | return; |
| 1188 | } |
| 1189 | for (primary=0; primary<mddev->raid_disks; primary++) |
| 1190 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
| 1191 | test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { |
| 1192 | r1_bio->bios[primary]->bi_end_io = NULL; |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 1193 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1194 | break; |
| 1195 | } |
| 1196 | r1_bio->read_disk = primary; |
| 1197 | for (i=0; i<mddev->raid_disks; i++) |
| 1198 | if (r1_bio->bios[i]->bi_end_io == end_sync_read && |
| 1199 | test_bit(BIO_UPTODATE, &r1_bio->bios[i]->bi_flags)) { |
| 1200 | int j; |
| 1201 | int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9); |
| 1202 | struct bio *pbio = r1_bio->bios[primary]; |
| 1203 | struct bio *sbio = r1_bio->bios[i]; |
| 1204 | for (j = vcnt; j-- ; ) |
| 1205 | if (memcmp(page_address(pbio->bi_io_vec[j].bv_page), |
| 1206 | page_address(sbio->bi_io_vec[j].bv_page), |
| 1207 | PAGE_SIZE)) |
| 1208 | break; |
| 1209 | if (j >= 0) |
| 1210 | mddev->resync_mismatches += r1_bio->sectors; |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 1211 | if (j < 0 || test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1212 | sbio->bi_end_io = NULL; |
NeilBrown | 03c902e | 2006-01-06 00:20:46 -0800 | [diff] [blame] | 1213 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); |
| 1214 | } else { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1215 | /* fixup the bio for reuse */ |
| 1216 | sbio->bi_vcnt = vcnt; |
| 1217 | sbio->bi_size = r1_bio->sectors << 9; |
| 1218 | sbio->bi_idx = 0; |
| 1219 | sbio->bi_phys_segments = 0; |
| 1220 | sbio->bi_hw_segments = 0; |
| 1221 | sbio->bi_hw_front_size = 0; |
| 1222 | sbio->bi_hw_back_size = 0; |
| 1223 | sbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
| 1224 | sbio->bi_flags |= 1 << BIO_UPTODATE; |
| 1225 | sbio->bi_next = NULL; |
| 1226 | sbio->bi_sector = r1_bio->sector + |
| 1227 | conf->mirrors[i].rdev->data_offset; |
| 1228 | sbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
| 1229 | } |
| 1230 | } |
| 1231 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1232 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1233 | /* ouch - failed to read all of that. |
| 1234 | * Try some synchronous reads of other devices to get |
| 1235 | * good data, much like with normal read errors. Only |
| 1236 | * read into the pages we already have so they we don't |
| 1237 | * need to re-issue the read request. |
| 1238 | * We don't need to freeze the array, because being in an |
| 1239 | * active sync request, there is no normal IO, and |
| 1240 | * no overlapping syncs. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | */ |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1242 | sector_t sect = r1_bio->sector; |
| 1243 | int sectors = r1_bio->sectors; |
| 1244 | int idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1246 | while(sectors) { |
| 1247 | int s = sectors; |
| 1248 | int d = r1_bio->read_disk; |
| 1249 | int success = 0; |
| 1250 | mdk_rdev_t *rdev; |
| 1251 | |
| 1252 | if (s > (PAGE_SIZE>>9)) |
| 1253 | s = PAGE_SIZE >> 9; |
| 1254 | do { |
| 1255 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { |
| 1256 | rdev = conf->mirrors[d].rdev; |
| 1257 | if (sync_page_io(rdev->bdev, |
| 1258 | sect + rdev->data_offset, |
| 1259 | s<<9, |
| 1260 | bio->bi_io_vec[idx].bv_page, |
| 1261 | READ)) { |
| 1262 | success = 1; |
| 1263 | break; |
| 1264 | } |
| 1265 | } |
| 1266 | d++; |
| 1267 | if (d == conf->raid_disks) |
| 1268 | d = 0; |
| 1269 | } while (!success && d != r1_bio->read_disk); |
| 1270 | |
| 1271 | if (success) { |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1272 | int start = d; |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1273 | /* write it back and re-read */ |
| 1274 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
| 1275 | while (d != r1_bio->read_disk) { |
| 1276 | if (d == 0) |
| 1277 | d = conf->raid_disks; |
| 1278 | d--; |
| 1279 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) |
| 1280 | continue; |
| 1281 | rdev = conf->mirrors[d].rdev; |
NeilBrown | 4dbcdc7 | 2006-01-06 00:20:52 -0800 | [diff] [blame] | 1282 | atomic_add(s, &rdev->corrected_errors); |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1283 | if (sync_page_io(rdev->bdev, |
| 1284 | sect + rdev->data_offset, |
| 1285 | s<<9, |
| 1286 | bio->bi_io_vec[idx].bv_page, |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1287 | WRITE) == 0) |
| 1288 | md_error(mddev, rdev); |
| 1289 | } |
| 1290 | d = start; |
| 1291 | while (d != r1_bio->read_disk) { |
| 1292 | if (d == 0) |
| 1293 | d = conf->raid_disks; |
| 1294 | d--; |
| 1295 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) |
| 1296 | continue; |
| 1297 | rdev = conf->mirrors[d].rdev; |
| 1298 | if (sync_page_io(rdev->bdev, |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1299 | sect + rdev->data_offset, |
| 1300 | s<<9, |
| 1301 | bio->bi_io_vec[idx].bv_page, |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1302 | READ) == 0) |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1303 | md_error(mddev, rdev); |
NeilBrown | 69382e8 | 2006-01-06 00:20:22 -0800 | [diff] [blame] | 1304 | } |
| 1305 | } else { |
| 1306 | char b[BDEVNAME_SIZE]; |
| 1307 | /* Cannot read from anywhere, array is toast */ |
| 1308 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); |
| 1309 | printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error" |
| 1310 | " for block %llu\n", |
| 1311 | bdevname(bio->bi_bdev,b), |
| 1312 | (unsigned long long)r1_bio->sector); |
| 1313 | md_done_sync(mddev, r1_bio->sectors, 0); |
| 1314 | put_buf(r1_bio); |
| 1315 | return; |
| 1316 | } |
| 1317 | sectors -= s; |
| 1318 | sect += s; |
| 1319 | idx ++; |
| 1320 | } |
| 1321 | } |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1322 | |
| 1323 | /* |
| 1324 | * schedule writes |
| 1325 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1326 | atomic_set(&r1_bio->remaining, 1); |
| 1327 | for (i = 0; i < disks ; i++) { |
| 1328 | wbio = r1_bio->bios[i]; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1329 | if (wbio->bi_end_io == NULL || |
| 1330 | (wbio->bi_end_io == end_sync_read && |
| 1331 | (i == r1_bio->read_disk || |
| 1332 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | continue; |
| 1334 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1335 | wbio->bi_rw = WRITE; |
| 1336 | wbio->bi_end_io = end_sync_write; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | atomic_inc(&r1_bio->remaining); |
| 1338 | md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1339 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1340 | generic_make_request(wbio); |
| 1341 | } |
| 1342 | |
| 1343 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1344 | /* if we're here, all write(s) have completed, so clean up */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1345 | md_done_sync(mddev, r1_bio->sectors, 1); |
| 1346 | put_buf(r1_bio); |
| 1347 | } |
| 1348 | } |
| 1349 | |
| 1350 | /* |
| 1351 | * This is a kernel thread which: |
| 1352 | * |
| 1353 | * 1. Retries failed read operations on working mirrors. |
| 1354 | * 2. Updates the raid superblock when problems encounter. |
| 1355 | * 3. Performs writes following reads for array syncronising. |
| 1356 | */ |
| 1357 | |
| 1358 | static void raid1d(mddev_t *mddev) |
| 1359 | { |
| 1360 | r1bio_t *r1_bio; |
| 1361 | struct bio *bio; |
| 1362 | unsigned long flags; |
| 1363 | conf_t *conf = mddev_to_conf(mddev); |
| 1364 | struct list_head *head = &conf->retry_list; |
| 1365 | int unplug=0; |
| 1366 | mdk_rdev_t *rdev; |
| 1367 | |
| 1368 | md_check_recovery(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1369 | |
| 1370 | for (;;) { |
| 1371 | char b[BDEVNAME_SIZE]; |
| 1372 | spin_lock_irqsave(&conf->device_lock, flags); |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1373 | |
| 1374 | if (conf->pending_bio_list.head) { |
| 1375 | bio = bio_list_get(&conf->pending_bio_list); |
| 1376 | blk_remove_plug(mddev->queue); |
| 1377 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 1378 | /* flush any pending bitmap writes to disk before proceeding w/ I/O */ |
| 1379 | if (bitmap_unplug(mddev->bitmap) != 0) |
| 1380 | printk("%s: bitmap file write failed!\n", mdname(mddev)); |
| 1381 | |
| 1382 | while (bio) { /* submit pending writes */ |
| 1383 | struct bio *next = bio->bi_next; |
| 1384 | bio->bi_next = NULL; |
| 1385 | generic_make_request(bio); |
| 1386 | bio = next; |
| 1387 | } |
| 1388 | unplug = 1; |
| 1389 | |
| 1390 | continue; |
| 1391 | } |
| 1392 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1393 | if (list_empty(head)) |
| 1394 | break; |
| 1395 | r1_bio = list_entry(head->prev, r1bio_t, retry_list); |
| 1396 | list_del(head->prev); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1397 | conf->nr_queued--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1398 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 1399 | |
| 1400 | mddev = r1_bio->mddev; |
| 1401 | conf = mddev_to_conf(mddev); |
| 1402 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
| 1403 | sync_request_write(mddev, r1_bio); |
| 1404 | unplug = 1; |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 1405 | } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) { |
| 1406 | /* some requests in the r1bio were BIO_RW_BARRIER |
| 1407 | * requests which failed with -ENOTSUPP. Hohumm.. |
| 1408 | * Better resubmit without the barrier. |
| 1409 | * We know which devices to resubmit for, because |
| 1410 | * all others have had their bios[] entry cleared. |
| 1411 | */ |
| 1412 | int i; |
| 1413 | clear_bit(R1BIO_BarrierRetry, &r1_bio->state); |
| 1414 | clear_bit(R1BIO_Barrier, &r1_bio->state); |
| 1415 | for (i=0; i < conf->raid_disks; i++) |
NeilBrown | 2f88912 | 2006-03-27 01:18:19 -0800 | [diff] [blame] | 1416 | if (r1_bio->bios[i]) |
| 1417 | atomic_inc(&r1_bio->remaining); |
| 1418 | for (i=0; i < conf->raid_disks; i++) |
NeilBrown | a9701a3 | 2005-11-08 21:39:34 -0800 | [diff] [blame] | 1419 | if (r1_bio->bios[i]) { |
| 1420 | struct bio_vec *bvec; |
| 1421 | int j; |
| 1422 | |
| 1423 | bio = bio_clone(r1_bio->master_bio, GFP_NOIO); |
| 1424 | /* copy pages from the failed bio, as |
| 1425 | * this might be a write-behind device */ |
| 1426 | __bio_for_each_segment(bvec, bio, j, 0) |
| 1427 | bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page; |
| 1428 | bio_put(r1_bio->bios[i]); |
| 1429 | bio->bi_sector = r1_bio->sector + |
| 1430 | conf->mirrors[i].rdev->data_offset; |
| 1431 | bio->bi_bdev = conf->mirrors[i].rdev->bdev; |
| 1432 | bio->bi_end_io = raid1_end_write_request; |
| 1433 | bio->bi_rw = WRITE; |
| 1434 | bio->bi_private = r1_bio; |
| 1435 | r1_bio->bios[i] = bio; |
| 1436 | generic_make_request(bio); |
| 1437 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1438 | } else { |
| 1439 | int disk; |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1440 | |
| 1441 | /* we got a read error. Maybe the drive is bad. Maybe just |
| 1442 | * the block and we can fix it. |
| 1443 | * We freeze all other IO, and try reading the block from |
| 1444 | * other devices. When we find one, we re-write |
| 1445 | * and check it that fixes the read error. |
| 1446 | * This is all done synchronously while the array is |
| 1447 | * frozen |
| 1448 | */ |
| 1449 | sector_t sect = r1_bio->sector; |
| 1450 | int sectors = r1_bio->sectors; |
| 1451 | freeze_array(conf); |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 1452 | if (mddev->ro == 0) while(sectors) { |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1453 | int s = sectors; |
| 1454 | int d = r1_bio->read_disk; |
| 1455 | int success = 0; |
| 1456 | |
| 1457 | if (s > (PAGE_SIZE>>9)) |
| 1458 | s = PAGE_SIZE >> 9; |
| 1459 | |
| 1460 | do { |
| 1461 | rdev = conf->mirrors[d].rdev; |
| 1462 | if (rdev && |
| 1463 | test_bit(In_sync, &rdev->flags) && |
| 1464 | sync_page_io(rdev->bdev, |
| 1465 | sect + rdev->data_offset, |
| 1466 | s<<9, |
| 1467 | conf->tmppage, READ)) |
| 1468 | success = 1; |
| 1469 | else { |
| 1470 | d++; |
| 1471 | if (d == conf->raid_disks) |
| 1472 | d = 0; |
| 1473 | } |
| 1474 | } while (!success && d != r1_bio->read_disk); |
| 1475 | |
| 1476 | if (success) { |
| 1477 | /* write it back and re-read */ |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1478 | int start = d; |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1479 | while (d != r1_bio->read_disk) { |
| 1480 | if (d==0) |
| 1481 | d = conf->raid_disks; |
| 1482 | d--; |
| 1483 | rdev = conf->mirrors[d].rdev; |
NeilBrown | 4dbcdc7 | 2006-01-06 00:20:52 -0800 | [diff] [blame] | 1484 | atomic_add(s, &rdev->corrected_errors); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1485 | if (rdev && |
| 1486 | test_bit(In_sync, &rdev->flags)) { |
| 1487 | if (sync_page_io(rdev->bdev, |
| 1488 | sect + rdev->data_offset, |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1489 | s<<9, conf->tmppage, WRITE) == 0) |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1490 | /* Well, this device is dead */ |
| 1491 | md_error(mddev, rdev); |
NeilBrown | 097426f | 2006-01-06 00:20:37 -0800 | [diff] [blame] | 1492 | } |
| 1493 | } |
| 1494 | d = start; |
| 1495 | while (d != r1_bio->read_disk) { |
| 1496 | if (d==0) |
| 1497 | d = conf->raid_disks; |
| 1498 | d--; |
| 1499 | rdev = conf->mirrors[d].rdev; |
| 1500 | if (rdev && |
| 1501 | test_bit(In_sync, &rdev->flags)) { |
| 1502 | if (sync_page_io(rdev->bdev, |
| 1503 | sect + rdev->data_offset, |
| 1504 | s<<9, conf->tmppage, READ) == 0) |
| 1505 | /* Well, this device is dead */ |
| 1506 | md_error(mddev, rdev); |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1507 | } |
| 1508 | } |
| 1509 | } else { |
| 1510 | /* Cannot read from anywhere -- bye bye array */ |
| 1511 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); |
| 1512 | break; |
| 1513 | } |
| 1514 | sectors -= s; |
| 1515 | sect += s; |
| 1516 | } |
| 1517 | |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1518 | unfreeze_array(conf); |
| 1519 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1520 | bio = r1_bio->bios[r1_bio->read_disk]; |
| 1521 | if ((disk=read_balance(conf, r1_bio)) == -1) { |
| 1522 | printk(KERN_ALERT "raid1: %s: unrecoverable I/O" |
| 1523 | " read error for block %llu\n", |
| 1524 | bdevname(bio->bi_bdev,b), |
| 1525 | (unsigned long long)r1_bio->sector); |
| 1526 | raid_end_bio_io(r1_bio); |
| 1527 | } else { |
NeilBrown | cf30a47 | 2006-01-06 00:20:23 -0800 | [diff] [blame] | 1528 | r1_bio->bios[r1_bio->read_disk] = |
| 1529 | mddev->ro ? IO_BLOCKED : NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1530 | r1_bio->read_disk = disk; |
| 1531 | bio_put(bio); |
| 1532 | bio = bio_clone(r1_bio->master_bio, GFP_NOIO); |
| 1533 | r1_bio->bios[r1_bio->read_disk] = bio; |
| 1534 | rdev = conf->mirrors[disk].rdev; |
| 1535 | if (printk_ratelimit()) |
| 1536 | printk(KERN_ERR "raid1: %s: redirecting sector %llu to" |
| 1537 | " another mirror\n", |
| 1538 | bdevname(rdev->bdev,b), |
| 1539 | (unsigned long long)r1_bio->sector); |
| 1540 | bio->bi_sector = r1_bio->sector + rdev->data_offset; |
| 1541 | bio->bi_bdev = rdev->bdev; |
| 1542 | bio->bi_end_io = raid1_end_read_request; |
| 1543 | bio->bi_rw = READ; |
| 1544 | bio->bi_private = r1_bio; |
| 1545 | unplug = 1; |
| 1546 | generic_make_request(bio); |
| 1547 | } |
| 1548 | } |
| 1549 | } |
| 1550 | spin_unlock_irqrestore(&conf->device_lock, flags); |
| 1551 | if (unplug) |
| 1552 | unplug_slaves(mddev); |
| 1553 | } |
| 1554 | |
| 1555 | |
| 1556 | static int init_resync(conf_t *conf) |
| 1557 | { |
| 1558 | int buffs; |
| 1559 | |
| 1560 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; |
| 1561 | if (conf->r1buf_pool) |
| 1562 | BUG(); |
| 1563 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
| 1564 | conf->poolinfo); |
| 1565 | if (!conf->r1buf_pool) |
| 1566 | return -ENOMEM; |
| 1567 | conf->next_resync = 0; |
| 1568 | return 0; |
| 1569 | } |
| 1570 | |
| 1571 | /* |
| 1572 | * perform a "sync" on one "block" |
| 1573 | * |
| 1574 | * We need to make sure that no normal I/O request - particularly write |
| 1575 | * requests - conflict with active sync requests. |
| 1576 | * |
| 1577 | * This is achieved by tracking pending requests and a 'barrier' concept |
| 1578 | * that can be installed to exclude normal IO requests. |
| 1579 | */ |
| 1580 | |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 1581 | 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] | 1582 | { |
| 1583 | conf_t *conf = mddev_to_conf(mddev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1584 | r1bio_t *r1_bio; |
| 1585 | struct bio *bio; |
| 1586 | sector_t max_sector, nr_sectors; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1587 | int disk = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1588 | int i; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1589 | int wonly = -1; |
| 1590 | int write_targets = 0, read_targets = 0; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1591 | int sync_blocks; |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1592 | int still_degraded = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1593 | |
| 1594 | if (!conf->r1buf_pool) |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1595 | { |
| 1596 | /* |
| 1597 | printk("sync start - bitmap %p\n", mddev->bitmap); |
| 1598 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1599 | if (init_resync(conf)) |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 1600 | return 0; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1601 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1602 | |
| 1603 | max_sector = mddev->size << 1; |
| 1604 | if (sector_nr >= max_sector) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1605 | /* If we aborted, we need to abort the |
| 1606 | * sync on the 'current' bitmap chunk (there will |
| 1607 | * only be one in raid1 resync. |
| 1608 | * We can find the current addess in mddev->curr_resync |
| 1609 | */ |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1610 | if (mddev->curr_resync < max_sector) /* aborted */ |
| 1611 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1612 | &sync_blocks, 1); |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1613 | else /* completed sync */ |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1614 | conf->fullsync = 0; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1615 | |
| 1616 | bitmap_close_sync(mddev->bitmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1617 | close_sync(conf); |
| 1618 | return 0; |
| 1619 | } |
| 1620 | |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1621 | /* before building a request, check if we can skip these blocks.. |
| 1622 | * This call the bitmap_start_sync doesn't actually record anything |
| 1623 | */ |
| 1624 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
NeilBrown | e5de485f | 2005-11-08 21:39:38 -0800 | [diff] [blame] | 1625 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1626 | /* We can skip this block, and probably several more */ |
| 1627 | *skipped = 1; |
| 1628 | return sync_blocks; |
| 1629 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | /* |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1631 | * If there is non-resync activity waiting for a turn, |
| 1632 | * and resync is going fast enough, |
| 1633 | * then let it though before starting on this new sync request. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1634 | */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1635 | if (!go_faster && conf->nr_waiting) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1636 | msleep_interruptible(1000); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1637 | |
| 1638 | raise_barrier(conf); |
| 1639 | |
| 1640 | conf->next_resync = sector_nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1641 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1643 | rcu_read_lock(); |
| 1644 | /* |
| 1645 | * If we get a correctably read error during resync or recovery, |
| 1646 | * we might want to read from a different device. So we |
| 1647 | * flag all drives that could conceivably be read from for READ, |
| 1648 | * and any others (which will be non-In_sync devices) for WRITE. |
| 1649 | * If a read fails, we try reading from something else for which READ |
| 1650 | * is OK. |
| 1651 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1652 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1653 | r1_bio->mddev = mddev; |
| 1654 | r1_bio->sector = sector_nr; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1655 | r1_bio->state = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1656 | set_bit(R1BIO_IsSync, &r1_bio->state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | |
| 1658 | for (i=0; i < conf->raid_disks; i++) { |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1659 | mdk_rdev_t *rdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1660 | bio = r1_bio->bios[i]; |
| 1661 | |
| 1662 | /* take from bio_init */ |
| 1663 | bio->bi_next = NULL; |
| 1664 | bio->bi_flags |= 1 << BIO_UPTODATE; |
| 1665 | bio->bi_rw = 0; |
| 1666 | bio->bi_vcnt = 0; |
| 1667 | bio->bi_idx = 0; |
| 1668 | bio->bi_phys_segments = 0; |
| 1669 | bio->bi_hw_segments = 0; |
| 1670 | bio->bi_size = 0; |
| 1671 | bio->bi_end_io = NULL; |
| 1672 | bio->bi_private = NULL; |
| 1673 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1674 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
| 1675 | if (rdev == NULL || |
| 1676 | test_bit(Faulty, &rdev->flags)) { |
NeilBrown | e3b9703 | 2005-08-04 12:53:34 -0700 | [diff] [blame] | 1677 | still_degraded = 1; |
| 1678 | continue; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1679 | } else if (!test_bit(In_sync, &rdev->flags)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1680 | bio->bi_rw = WRITE; |
| 1681 | bio->bi_end_io = end_sync_write; |
| 1682 | write_targets ++; |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1683 | } else { |
| 1684 | /* may need to read from here */ |
| 1685 | bio->bi_rw = READ; |
| 1686 | bio->bi_end_io = end_sync_read; |
| 1687 | if (test_bit(WriteMostly, &rdev->flags)) { |
| 1688 | if (wonly < 0) |
| 1689 | wonly = i; |
| 1690 | } else { |
| 1691 | if (disk < 0) |
| 1692 | disk = i; |
| 1693 | } |
| 1694 | read_targets++; |
| 1695 | } |
| 1696 | atomic_inc(&rdev->nr_pending); |
| 1697 | bio->bi_sector = sector_nr + rdev->data_offset; |
| 1698 | bio->bi_bdev = rdev->bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | bio->bi_private = r1_bio; |
| 1700 | } |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1701 | rcu_read_unlock(); |
| 1702 | if (disk < 0) |
| 1703 | disk = wonly; |
| 1704 | r1_bio->read_disk = disk; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1705 | |
NeilBrown | 3e198f7 | 2006-01-06 00:20:21 -0800 | [diff] [blame] | 1706 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
| 1707 | /* extra read targets are also write targets */ |
| 1708 | write_targets += read_targets-1; |
| 1709 | |
| 1710 | if (write_targets == 0 || read_targets == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1711 | /* There is nowhere to write, so all non-sync |
| 1712 | * drives must be failed - so we are finished |
| 1713 | */ |
NeilBrown | 57afd89 | 2005-06-21 17:17:13 -0700 | [diff] [blame] | 1714 | sector_t rv = max_sector - sector_nr; |
| 1715 | *skipped = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1716 | put_buf(r1_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1717 | return rv; |
| 1718 | } |
| 1719 | |
| 1720 | nr_sectors = 0; |
NeilBrown | 289e99e | 2005-06-21 17:17:24 -0700 | [diff] [blame] | 1721 | sync_blocks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1722 | do { |
| 1723 | struct page *page; |
| 1724 | int len = PAGE_SIZE; |
| 1725 | if (sector_nr + (len>>9) > max_sector) |
| 1726 | len = (max_sector - sector_nr) << 9; |
| 1727 | if (len == 0) |
| 1728 | break; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1729 | if (sync_blocks == 0) { |
| 1730 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
NeilBrown | e5de485f | 2005-11-08 21:39:38 -0800 | [diff] [blame] | 1731 | &sync_blocks, still_degraded) && |
| 1732 | !conf->fullsync && |
| 1733 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1734 | break; |
| 1735 | if (sync_blocks < (PAGE_SIZE>>9)) |
| 1736 | BUG(); |
| 1737 | if (len > (sync_blocks<<9)) |
| 1738 | len = sync_blocks<<9; |
NeilBrown | ab7a30c | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1739 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1740 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1741 | for (i=0 ; i < conf->raid_disks; i++) { |
| 1742 | bio = r1_bio->bios[i]; |
| 1743 | if (bio->bi_end_io) { |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1744 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1745 | if (bio_add_page(bio, page, len, 0) == 0) { |
| 1746 | /* stop here */ |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1747 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1748 | while (i > 0) { |
| 1749 | i--; |
| 1750 | bio = r1_bio->bios[i]; |
NeilBrown | 6a806c5 | 2005-07-15 03:56:35 -0700 | [diff] [blame] | 1751 | if (bio->bi_end_io==NULL) |
| 1752 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | /* remove last page from this bio */ |
| 1754 | bio->bi_vcnt--; |
| 1755 | bio->bi_size -= len; |
| 1756 | bio->bi_flags &= ~(1<< BIO_SEG_VALID); |
| 1757 | } |
| 1758 | goto bio_full; |
| 1759 | } |
| 1760 | } |
| 1761 | } |
| 1762 | nr_sectors += len>>9; |
| 1763 | sector_nr += len>>9; |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1764 | sync_blocks -= (len>>9); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1765 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
| 1766 | bio_full: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1767 | r1_bio->sectors = nr_sectors; |
| 1768 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1769 | /* For a user-requested sync, we read all readable devices and do a |
| 1770 | * compare |
| 1771 | */ |
| 1772 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
| 1773 | atomic_set(&r1_bio->remaining, read_targets); |
| 1774 | for (i=0; i<conf->raid_disks; i++) { |
| 1775 | bio = r1_bio->bios[i]; |
| 1776 | if (bio->bi_end_io == end_sync_read) { |
| 1777 | md_sync_acct(conf->mirrors[i].rdev->bdev, nr_sectors); |
| 1778 | generic_make_request(bio); |
| 1779 | } |
| 1780 | } |
| 1781 | } else { |
| 1782 | atomic_set(&r1_bio->remaining, 1); |
| 1783 | bio = r1_bio->bios[r1_bio->read_disk]; |
| 1784 | md_sync_acct(conf->mirrors[r1_bio->read_disk].rdev->bdev, |
| 1785 | nr_sectors); |
| 1786 | generic_make_request(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1787 | |
NeilBrown | d11c171 | 2006-01-06 00:20:26 -0800 | [diff] [blame] | 1788 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1789 | |
| 1790 | return nr_sectors; |
| 1791 | } |
| 1792 | |
| 1793 | static int run(mddev_t *mddev) |
| 1794 | { |
| 1795 | conf_t *conf; |
| 1796 | int i, j, disk_idx; |
| 1797 | mirror_info_t *disk; |
| 1798 | mdk_rdev_t *rdev; |
| 1799 | struct list_head *tmp; |
| 1800 | |
| 1801 | if (mddev->level != 1) { |
| 1802 | printk("raid1: %s: raid level not set to mirroring (%d)\n", |
| 1803 | mdname(mddev), mddev->level); |
| 1804 | goto out; |
| 1805 | } |
NeilBrown | f670557 | 2006-03-27 01:18:11 -0800 | [diff] [blame] | 1806 | if (mddev->reshape_position != MaxSector) { |
| 1807 | printk("raid1: %s: reshape_position set but not supported\n", |
| 1808 | mdname(mddev)); |
| 1809 | goto out; |
| 1810 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1811 | /* |
| 1812 | * copy the already verified devices into our private RAID1 |
| 1813 | * bookkeeping area. [whatever we allocate in run(), |
| 1814 | * should be freed in stop()] |
| 1815 | */ |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 1816 | conf = kzalloc(sizeof(conf_t), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1817 | mddev->private = conf; |
| 1818 | if (!conf) |
| 1819 | goto out_no_mem; |
| 1820 | |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 1821 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1822 | GFP_KERNEL); |
| 1823 | if (!conf->mirrors) |
| 1824 | goto out_no_mem; |
| 1825 | |
NeilBrown | ddaf22a | 2006-01-06 00:20:19 -0800 | [diff] [blame] | 1826 | conf->tmppage = alloc_page(GFP_KERNEL); |
| 1827 | if (!conf->tmppage) |
| 1828 | goto out_no_mem; |
| 1829 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1830 | conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
| 1831 | if (!conf->poolinfo) |
| 1832 | goto out_no_mem; |
| 1833 | conf->poolinfo->mddev = mddev; |
| 1834 | conf->poolinfo->raid_disks = mddev->raid_disks; |
| 1835 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
| 1836 | r1bio_pool_free, |
| 1837 | conf->poolinfo); |
| 1838 | if (!conf->r1bio_pool) |
| 1839 | goto out_no_mem; |
| 1840 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1841 | ITERATE_RDEV(mddev, rdev, tmp) { |
| 1842 | disk_idx = rdev->raid_disk; |
| 1843 | if (disk_idx >= mddev->raid_disks |
| 1844 | || disk_idx < 0) |
| 1845 | continue; |
| 1846 | disk = conf->mirrors + disk_idx; |
| 1847 | |
| 1848 | disk->rdev = rdev; |
| 1849 | |
| 1850 | blk_queue_stack_limits(mddev->queue, |
| 1851 | rdev->bdev->bd_disk->queue); |
| 1852 | /* as we don't honour merge_bvec_fn, we must never risk |
| 1853 | * violating it, so limit ->max_sector to one PAGE, as |
| 1854 | * a one page request is never in violation. |
| 1855 | */ |
| 1856 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && |
| 1857 | mddev->queue->max_sectors > (PAGE_SIZE>>9)) |
| 1858 | blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); |
| 1859 | |
| 1860 | disk->head_position = 0; |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1861 | if (!test_bit(Faulty, &rdev->flags) && test_bit(In_sync, &rdev->flags)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1862 | conf->working_disks++; |
| 1863 | } |
| 1864 | conf->raid_disks = mddev->raid_disks; |
| 1865 | conf->mddev = mddev; |
| 1866 | spin_lock_init(&conf->device_lock); |
| 1867 | INIT_LIST_HEAD(&conf->retry_list); |
| 1868 | if (conf->working_disks == 1) |
| 1869 | mddev->recovery_cp = MaxSector; |
| 1870 | |
| 1871 | spin_lock_init(&conf->resync_lock); |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 1872 | init_waitqueue_head(&conf->wait_barrier); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1873 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1874 | bio_list_init(&conf->pending_bio_list); |
| 1875 | bio_list_init(&conf->flushing_bio_list); |
| 1876 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1877 | if (!conf->working_disks) { |
| 1878 | printk(KERN_ERR "raid1: no operational mirrors for %s\n", |
| 1879 | mdname(mddev)); |
| 1880 | goto out_free_conf; |
| 1881 | } |
| 1882 | |
| 1883 | mddev->degraded = 0; |
| 1884 | for (i = 0; i < conf->raid_disks; i++) { |
| 1885 | |
| 1886 | disk = conf->mirrors + i; |
| 1887 | |
| 1888 | if (!disk->rdev) { |
| 1889 | disk->head_position = 0; |
| 1890 | mddev->degraded++; |
| 1891 | } |
| 1892 | } |
| 1893 | |
| 1894 | /* |
| 1895 | * find the first working one and use it as a starting point |
| 1896 | * to read balancing. |
| 1897 | */ |
| 1898 | for (j = 0; j < conf->raid_disks && |
| 1899 | (!conf->mirrors[j].rdev || |
NeilBrown | b2d444d | 2005-11-08 21:39:31 -0800 | [diff] [blame] | 1900 | !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1901 | /* nothing */; |
| 1902 | conf->last_used = j; |
| 1903 | |
| 1904 | |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1905 | mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1"); |
| 1906 | if (!mddev->thread) { |
| 1907 | printk(KERN_ERR |
| 1908 | "raid1: couldn't allocate thread for %s\n", |
| 1909 | mdname(mddev)); |
| 1910 | goto out_free_conf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1911 | } |
NeilBrown | 191ea9b | 2005-06-21 17:17:23 -0700 | [diff] [blame] | 1912 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1913 | printk(KERN_INFO |
| 1914 | "raid1: raid set %s active with %d out of %d mirrors\n", |
| 1915 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
| 1916 | mddev->raid_disks); |
| 1917 | /* |
| 1918 | * Ok, everything is just fine now |
| 1919 | */ |
| 1920 | mddev->array_size = mddev->size; |
| 1921 | |
NeilBrown | 7a5febe | 2005-05-16 21:53:16 -0700 | [diff] [blame] | 1922 | mddev->queue->unplug_fn = raid1_unplug; |
| 1923 | mddev->queue->issue_flush_fn = raid1_issue_flush; |
| 1924 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1925 | return 0; |
| 1926 | |
| 1927 | out_no_mem: |
| 1928 | printk(KERN_ERR "raid1: couldn't allocate memory for %s\n", |
| 1929 | mdname(mddev)); |
| 1930 | |
| 1931 | out_free_conf: |
| 1932 | if (conf) { |
| 1933 | if (conf->r1bio_pool) |
| 1934 | mempool_destroy(conf->r1bio_pool); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 1935 | kfree(conf->mirrors); |
NeilBrown | 1345b1d | 2006-01-06 00:20:40 -0800 | [diff] [blame] | 1936 | safe_put_page(conf->tmppage); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 1937 | kfree(conf->poolinfo); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1938 | kfree(conf); |
| 1939 | mddev->private = NULL; |
| 1940 | } |
| 1941 | out: |
| 1942 | return -EIO; |
| 1943 | } |
| 1944 | |
| 1945 | static int stop(mddev_t *mddev) |
| 1946 | { |
| 1947 | conf_t *conf = mddev_to_conf(mddev); |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 1948 | struct bitmap *bitmap = mddev->bitmap; |
| 1949 | int behind_wait = 0; |
| 1950 | |
| 1951 | /* wait for behind writes to complete */ |
| 1952 | while (bitmap && atomic_read(&bitmap->behind_writes) > 0) { |
| 1953 | behind_wait++; |
| 1954 | printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait); |
| 1955 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 1956 | schedule_timeout(HZ); /* wait a second */ |
| 1957 | /* need to kick something here to make sure I/O goes? */ |
| 1958 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1959 | |
| 1960 | md_unregister_thread(mddev->thread); |
| 1961 | mddev->thread = NULL; |
| 1962 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
| 1963 | if (conf->r1bio_pool) |
| 1964 | mempool_destroy(conf->r1bio_pool); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 1965 | kfree(conf->mirrors); |
| 1966 | kfree(conf->poolinfo); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1967 | kfree(conf); |
| 1968 | mddev->private = NULL; |
| 1969 | return 0; |
| 1970 | } |
| 1971 | |
| 1972 | static int raid1_resize(mddev_t *mddev, sector_t sectors) |
| 1973 | { |
| 1974 | /* no resync is happening, and there is enough space |
| 1975 | * on all devices, so we can resize. |
| 1976 | * We need to make sure resync covers any new space. |
| 1977 | * If the array is shrinking we should possibly wait until |
| 1978 | * any io in the removed space completes, but it hardly seems |
| 1979 | * worth it. |
| 1980 | */ |
| 1981 | mddev->array_size = sectors>>1; |
| 1982 | set_capacity(mddev->gendisk, mddev->array_size << 1); |
| 1983 | mddev->changed = 1; |
| 1984 | if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) { |
| 1985 | mddev->recovery_cp = mddev->size << 1; |
| 1986 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
| 1987 | } |
| 1988 | mddev->size = mddev->array_size; |
NeilBrown | 4b5c7ae | 2005-07-27 11:43:28 -0700 | [diff] [blame] | 1989 | mddev->resync_max_sectors = sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1990 | return 0; |
| 1991 | } |
| 1992 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 1993 | static int raid1_reshape(mddev_t *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1994 | { |
| 1995 | /* We need to: |
| 1996 | * 1/ resize the r1bio_pool |
| 1997 | * 2/ resize conf->mirrors |
| 1998 | * |
| 1999 | * We allocate a new r1bio_pool if we can. |
| 2000 | * Then raise a device barrier and wait until all IO stops. |
| 2001 | * Then resize conf->mirrors and swap in the new r1bio pool. |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2002 | * |
| 2003 | * At the same time, we "pack" the devices so that all the missing |
| 2004 | * devices have the higher raid_disk numbers. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2005 | */ |
| 2006 | mempool_t *newpool, *oldpool; |
| 2007 | struct pool_info *newpoolinfo; |
| 2008 | mirror_info_t *newmirrors; |
| 2009 | conf_t *conf = mddev_to_conf(mddev); |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2010 | int cnt, raid_disks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2011 | |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2012 | int d, d2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2013 | |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2014 | /* Cannot change chunk_size, layout, or level */ |
| 2015 | if (mddev->chunk_size != mddev->new_chunk || |
| 2016 | mddev->layout != mddev->new_layout || |
| 2017 | mddev->level != mddev->new_level) { |
| 2018 | mddev->new_chunk = mddev->chunk_size; |
| 2019 | mddev->new_layout = mddev->layout; |
| 2020 | mddev->new_level = mddev->level; |
| 2021 | return -EINVAL; |
| 2022 | } |
| 2023 | |
| 2024 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
| 2025 | |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2026 | if (raid_disks < conf->raid_disks) { |
| 2027 | cnt=0; |
| 2028 | for (d= 0; d < conf->raid_disks; d++) |
| 2029 | if (conf->mirrors[d].rdev) |
| 2030 | cnt++; |
| 2031 | if (cnt > raid_disks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2032 | return -EBUSY; |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2033 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2034 | |
| 2035 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); |
| 2036 | if (!newpoolinfo) |
| 2037 | return -ENOMEM; |
| 2038 | newpoolinfo->mddev = mddev; |
| 2039 | newpoolinfo->raid_disks = raid_disks; |
| 2040 | |
| 2041 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
| 2042 | r1bio_pool_free, newpoolinfo); |
| 2043 | if (!newpool) { |
| 2044 | kfree(newpoolinfo); |
| 2045 | return -ENOMEM; |
| 2046 | } |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 2047 | newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2048 | if (!newmirrors) { |
| 2049 | kfree(newpoolinfo); |
| 2050 | mempool_destroy(newpool); |
| 2051 | return -ENOMEM; |
| 2052 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2053 | |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2054 | raise_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2055 | |
| 2056 | /* ok, everything is stopped */ |
| 2057 | oldpool = conf->r1bio_pool; |
| 2058 | conf->r1bio_pool = newpool; |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2059 | |
| 2060 | for (d=d2=0; d < conf->raid_disks; d++) |
| 2061 | if (conf->mirrors[d].rdev) { |
| 2062 | conf->mirrors[d].rdev->raid_disk = d2; |
| 2063 | newmirrors[d2++].rdev = conf->mirrors[d].rdev; |
| 2064 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2065 | kfree(conf->mirrors); |
| 2066 | conf->mirrors = newmirrors; |
| 2067 | kfree(conf->poolinfo); |
| 2068 | conf->poolinfo = newpoolinfo; |
| 2069 | |
| 2070 | mddev->degraded += (raid_disks - conf->raid_disks); |
| 2071 | conf->raid_disks = mddev->raid_disks = raid_disks; |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2072 | mddev->delta_disks = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2073 | |
NeilBrown | 6ea9c07 | 2005-06-21 17:17:09 -0700 | [diff] [blame] | 2074 | conf->last_used = 0; /* just make sure it is in-range */ |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2075 | lower_barrier(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2076 | |
| 2077 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
| 2078 | md_wakeup_thread(mddev->thread); |
| 2079 | |
| 2080 | mempool_destroy(oldpool); |
| 2081 | return 0; |
| 2082 | } |
| 2083 | |
NeilBrown | 500af87 | 2005-09-09 16:23:58 -0700 | [diff] [blame] | 2084 | static void raid1_quiesce(mddev_t *mddev, int state) |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2085 | { |
| 2086 | conf_t *conf = mddev_to_conf(mddev); |
| 2087 | |
| 2088 | switch(state) { |
NeilBrown | 9e6603d | 2005-09-09 16:23:48 -0700 | [diff] [blame] | 2089 | case 1: |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2090 | raise_barrier(conf); |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2091 | break; |
NeilBrown | 9e6603d | 2005-09-09 16:23:48 -0700 | [diff] [blame] | 2092 | case 0: |
NeilBrown | 17999be | 2006-01-06 00:20:12 -0800 | [diff] [blame] | 2093 | lower_barrier(conf); |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2094 | break; |
| 2095 | } |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2096 | } |
| 2097 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2098 | |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2099 | static struct mdk_personality raid1_personality = |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2100 | { |
| 2101 | .name = "raid1", |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2102 | .level = 1, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2103 | .owner = THIS_MODULE, |
| 2104 | .make_request = make_request, |
| 2105 | .run = run, |
| 2106 | .stop = stop, |
| 2107 | .status = status, |
| 2108 | .error_handler = error, |
| 2109 | .hot_add_disk = raid1_add_disk, |
| 2110 | .hot_remove_disk= raid1_remove_disk, |
| 2111 | .spare_active = raid1_spare_active, |
| 2112 | .sync_request = sync_request, |
| 2113 | .resize = raid1_resize, |
NeilBrown | 63c70c4 | 2006-03-27 01:18:13 -0800 | [diff] [blame] | 2114 | .check_reshape = raid1_reshape, |
NeilBrown | 36fa306 | 2005-09-09 16:23:45 -0700 | [diff] [blame] | 2115 | .quiesce = raid1_quiesce, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2116 | }; |
| 2117 | |
| 2118 | static int __init raid_init(void) |
| 2119 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2120 | return register_md_personality(&raid1_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2121 | } |
| 2122 | |
| 2123 | static void raid_exit(void) |
| 2124 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2125 | unregister_md_personality(&raid1_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2126 | } |
| 2127 | |
| 2128 | module_init(raid_init); |
| 2129 | module_exit(raid_exit); |
| 2130 | MODULE_LICENSE("GPL"); |
| 2131 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
NeilBrown | d9d166c | 2006-01-06 00:20:51 -0800 | [diff] [blame] | 2132 | MODULE_ALIAS("md-raid1"); |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 2133 | MODULE_ALIAS("md-level-1"); |