| 1 | /* |
| 2 | linear.c : Multiple Devices driver for Linux |
| 3 | Copyright (C) 1994-96 Marc ZYNGIER |
| 4 | <zyngier@ufr-info-p7.ibp.fr> or |
| 5 | <maz@gloups.fdn.fr> |
| 6 | |
| 7 | Linear mode management functions. |
| 8 | |
| 9 | This program is free software; you can redistribute it and/or modify |
| 10 | it under the terms of the GNU General Public License as published by |
| 11 | the Free Software Foundation; either version 2, or (at your option) |
| 12 | any later version. |
| 13 | |
| 14 | You should have received a copy of the GNU General Public License |
| 15 | (for example /usr/src/linux/COPYING); if not, write to the Free |
| 16 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
| 17 | */ |
| 18 | |
| 19 | #include <linux/blkdev.h> |
| 20 | #include <linux/raid/md_u.h> |
| 21 | #include <linux/seq_file.h> |
| 22 | #include <linux/slab.h> |
| 23 | #include "md.h" |
| 24 | #include "linear.h" |
| 25 | |
| 26 | /* |
| 27 | * find which device holds a particular offset |
| 28 | */ |
| 29 | static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector) |
| 30 | { |
| 31 | int lo, mid, hi; |
| 32 | linear_conf_t *conf; |
| 33 | |
| 34 | lo = 0; |
| 35 | hi = mddev->raid_disks - 1; |
| 36 | conf = rcu_dereference(mddev->private); |
| 37 | |
| 38 | /* |
| 39 | * Binary Search |
| 40 | */ |
| 41 | |
| 42 | while (hi > lo) { |
| 43 | |
| 44 | mid = (hi + lo) / 2; |
| 45 | if (sector < conf->disks[mid].end_sector) |
| 46 | hi = mid; |
| 47 | else |
| 48 | lo = mid + 1; |
| 49 | } |
| 50 | |
| 51 | return conf->disks + lo; |
| 52 | } |
| 53 | |
| 54 | /** |
| 55 | * linear_mergeable_bvec -- tell bio layer if two requests can be merged |
| 56 | * @q: request queue |
| 57 | * @bvm: properties of new bio |
| 58 | * @biovec: the request that could be merged to it. |
| 59 | * |
| 60 | * Return amount of bytes we can take at this offset |
| 61 | */ |
| 62 | static int linear_mergeable_bvec(struct request_queue *q, |
| 63 | struct bvec_merge_data *bvm, |
| 64 | struct bio_vec *biovec) |
| 65 | { |
| 66 | mddev_t *mddev = q->queuedata; |
| 67 | dev_info_t *dev0; |
| 68 | unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9; |
| 69 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
| 70 | |
| 71 | rcu_read_lock(); |
| 72 | dev0 = which_dev(mddev, sector); |
| 73 | maxsectors = dev0->end_sector - sector; |
| 74 | rcu_read_unlock(); |
| 75 | |
| 76 | if (maxsectors < bio_sectors) |
| 77 | maxsectors = 0; |
| 78 | else |
| 79 | maxsectors -= bio_sectors; |
| 80 | |
| 81 | if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0) |
| 82 | return biovec->bv_len; |
| 83 | /* The bytes available at this offset could be really big, |
| 84 | * so we cap at 2^31 to avoid overflow */ |
| 85 | if (maxsectors > (1 << (31-9))) |
| 86 | return 1<<31; |
| 87 | return maxsectors << 9; |
| 88 | } |
| 89 | |
| 90 | static int linear_congested(void *data, int bits) |
| 91 | { |
| 92 | mddev_t *mddev = data; |
| 93 | linear_conf_t *conf; |
| 94 | int i, ret = 0; |
| 95 | |
| 96 | if (mddev_congested(mddev, bits)) |
| 97 | return 1; |
| 98 | |
| 99 | rcu_read_lock(); |
| 100 | conf = rcu_dereference(mddev->private); |
| 101 | |
| 102 | for (i = 0; i < mddev->raid_disks && !ret ; i++) { |
| 103 | struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev); |
| 104 | ret |= bdi_congested(&q->backing_dev_info, bits); |
| 105 | } |
| 106 | |
| 107 | rcu_read_unlock(); |
| 108 | return ret; |
| 109 | } |
| 110 | |
| 111 | static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks) |
| 112 | { |
| 113 | linear_conf_t *conf; |
| 114 | sector_t array_sectors; |
| 115 | |
| 116 | rcu_read_lock(); |
| 117 | conf = rcu_dereference(mddev->private); |
| 118 | WARN_ONCE(sectors || raid_disks, |
| 119 | "%s does not support generic reshape\n", __func__); |
| 120 | array_sectors = conf->array_sectors; |
| 121 | rcu_read_unlock(); |
| 122 | |
| 123 | return array_sectors; |
| 124 | } |
| 125 | |
| 126 | static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks) |
| 127 | { |
| 128 | linear_conf_t *conf; |
| 129 | mdk_rdev_t *rdev; |
| 130 | int i, cnt; |
| 131 | |
| 132 | conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t), |
| 133 | GFP_KERNEL); |
| 134 | if (!conf) |
| 135 | return NULL; |
| 136 | |
| 137 | cnt = 0; |
| 138 | conf->array_sectors = 0; |
| 139 | |
| 140 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
| 141 | int j = rdev->raid_disk; |
| 142 | dev_info_t *disk = conf->disks + j; |
| 143 | sector_t sectors; |
| 144 | |
| 145 | if (j < 0 || j >= raid_disks || disk->rdev) { |
| 146 | printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n", |
| 147 | mdname(mddev)); |
| 148 | goto out; |
| 149 | } |
| 150 | |
| 151 | disk->rdev = rdev; |
| 152 | if (mddev->chunk_sectors) { |
| 153 | sectors = rdev->sectors; |
| 154 | sector_div(sectors, mddev->chunk_sectors); |
| 155 | rdev->sectors = sectors * mddev->chunk_sectors; |
| 156 | } |
| 157 | |
| 158 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 159 | rdev->data_offset << 9); |
| 160 | /* as we don't honour merge_bvec_fn, we must never risk |
| 161 | * violating it, so limit max_segments to 1 lying within |
| 162 | * a single page. |
| 163 | */ |
| 164 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
| 165 | blk_queue_max_segments(mddev->queue, 1); |
| 166 | blk_queue_segment_boundary(mddev->queue, |
| 167 | PAGE_CACHE_SIZE - 1); |
| 168 | } |
| 169 | |
| 170 | conf->array_sectors += rdev->sectors; |
| 171 | cnt++; |
| 172 | |
| 173 | } |
| 174 | if (cnt != raid_disks) { |
| 175 | printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n", |
| 176 | mdname(mddev)); |
| 177 | goto out; |
| 178 | } |
| 179 | |
| 180 | /* |
| 181 | * Here we calculate the device offsets. |
| 182 | */ |
| 183 | conf->disks[0].end_sector = conf->disks[0].rdev->sectors; |
| 184 | |
| 185 | for (i = 1; i < raid_disks; i++) |
| 186 | conf->disks[i].end_sector = |
| 187 | conf->disks[i-1].end_sector + |
| 188 | conf->disks[i].rdev->sectors; |
| 189 | |
| 190 | return conf; |
| 191 | |
| 192 | out: |
| 193 | kfree(conf); |
| 194 | return NULL; |
| 195 | } |
| 196 | |
| 197 | static int linear_run (mddev_t *mddev) |
| 198 | { |
| 199 | linear_conf_t *conf; |
| 200 | |
| 201 | if (md_check_no_bitmap(mddev)) |
| 202 | return -EINVAL; |
| 203 | conf = linear_conf(mddev, mddev->raid_disks); |
| 204 | |
| 205 | if (!conf) |
| 206 | return 1; |
| 207 | mddev->private = conf; |
| 208 | md_set_array_sectors(mddev, linear_size(mddev, 0, 0)); |
| 209 | |
| 210 | blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec); |
| 211 | mddev->queue->backing_dev_info.congested_fn = linear_congested; |
| 212 | mddev->queue->backing_dev_info.congested_data = mddev; |
| 213 | return md_integrity_register(mddev); |
| 214 | } |
| 215 | |
| 216 | static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev) |
| 217 | { |
| 218 | /* Adding a drive to a linear array allows the array to grow. |
| 219 | * It is permitted if the new drive has a matching superblock |
| 220 | * already on it, with raid_disk equal to raid_disks. |
| 221 | * It is achieved by creating a new linear_private_data structure |
| 222 | * and swapping it in in-place of the current one. |
| 223 | * The current one is never freed until the array is stopped. |
| 224 | * This avoids races. |
| 225 | */ |
| 226 | linear_conf_t *newconf, *oldconf; |
| 227 | |
| 228 | if (rdev->saved_raid_disk != mddev->raid_disks) |
| 229 | return -EINVAL; |
| 230 | |
| 231 | rdev->raid_disk = rdev->saved_raid_disk; |
| 232 | |
| 233 | newconf = linear_conf(mddev,mddev->raid_disks+1); |
| 234 | |
| 235 | if (!newconf) |
| 236 | return -ENOMEM; |
| 237 | |
| 238 | oldconf = rcu_dereference(mddev->private); |
| 239 | mddev->raid_disks++; |
| 240 | rcu_assign_pointer(mddev->private, newconf); |
| 241 | md_set_array_sectors(mddev, linear_size(mddev, 0, 0)); |
| 242 | set_capacity(mddev->gendisk, mddev->array_sectors); |
| 243 | revalidate_disk(mddev->gendisk); |
| 244 | kfree_rcu(oldconf, rcu); |
| 245 | return 0; |
| 246 | } |
| 247 | |
| 248 | static int linear_stop (mddev_t *mddev) |
| 249 | { |
| 250 | linear_conf_t *conf = mddev->private; |
| 251 | |
| 252 | /* |
| 253 | * We do not require rcu protection here since |
| 254 | * we hold reconfig_mutex for both linear_add and |
| 255 | * linear_stop, so they cannot race. |
| 256 | * We should make sure any old 'conf's are properly |
| 257 | * freed though. |
| 258 | */ |
| 259 | rcu_barrier(); |
| 260 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
| 261 | kfree(conf); |
| 262 | mddev->private = NULL; |
| 263 | |
| 264 | return 0; |
| 265 | } |
| 266 | |
| 267 | static int linear_make_request (mddev_t *mddev, struct bio *bio) |
| 268 | { |
| 269 | dev_info_t *tmp_dev; |
| 270 | sector_t start_sector; |
| 271 | |
| 272 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
| 273 | md_flush_request(mddev, bio); |
| 274 | return 0; |
| 275 | } |
| 276 | |
| 277 | rcu_read_lock(); |
| 278 | tmp_dev = which_dev(mddev, bio->bi_sector); |
| 279 | start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors; |
| 280 | |
| 281 | |
| 282 | if (unlikely(bio->bi_sector >= (tmp_dev->end_sector) |
| 283 | || (bio->bi_sector < start_sector))) { |
| 284 | char b[BDEVNAME_SIZE]; |
| 285 | |
| 286 | printk(KERN_ERR |
| 287 | "md/linear:%s: make_request: Sector %llu out of bounds on " |
| 288 | "dev %s: %llu sectors, offset %llu\n", |
| 289 | mdname(mddev), |
| 290 | (unsigned long long)bio->bi_sector, |
| 291 | bdevname(tmp_dev->rdev->bdev, b), |
| 292 | (unsigned long long)tmp_dev->rdev->sectors, |
| 293 | (unsigned long long)start_sector); |
| 294 | rcu_read_unlock(); |
| 295 | bio_io_error(bio); |
| 296 | return 0; |
| 297 | } |
| 298 | if (unlikely(bio->bi_sector + (bio->bi_size >> 9) > |
| 299 | tmp_dev->end_sector)) { |
| 300 | /* This bio crosses a device boundary, so we have to |
| 301 | * split it. |
| 302 | */ |
| 303 | struct bio_pair *bp; |
| 304 | sector_t end_sector = tmp_dev->end_sector; |
| 305 | |
| 306 | rcu_read_unlock(); |
| 307 | |
| 308 | bp = bio_split(bio, end_sector - bio->bi_sector); |
| 309 | |
| 310 | if (linear_make_request(mddev, &bp->bio1)) |
| 311 | generic_make_request(&bp->bio1); |
| 312 | if (linear_make_request(mddev, &bp->bio2)) |
| 313 | generic_make_request(&bp->bio2); |
| 314 | bio_pair_release(bp); |
| 315 | return 0; |
| 316 | } |
| 317 | |
| 318 | bio->bi_bdev = tmp_dev->rdev->bdev; |
| 319 | bio->bi_sector = bio->bi_sector - start_sector |
| 320 | + tmp_dev->rdev->data_offset; |
| 321 | rcu_read_unlock(); |
| 322 | |
| 323 | return 1; |
| 324 | } |
| 325 | |
| 326 | static void linear_status (struct seq_file *seq, mddev_t *mddev) |
| 327 | { |
| 328 | |
| 329 | seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2); |
| 330 | } |
| 331 | |
| 332 | |
| 333 | static struct mdk_personality linear_personality = |
| 334 | { |
| 335 | .name = "linear", |
| 336 | .level = LEVEL_LINEAR, |
| 337 | .owner = THIS_MODULE, |
| 338 | .make_request = linear_make_request, |
| 339 | .run = linear_run, |
| 340 | .stop = linear_stop, |
| 341 | .status = linear_status, |
| 342 | .hot_add_disk = linear_add, |
| 343 | .size = linear_size, |
| 344 | }; |
| 345 | |
| 346 | static int __init linear_init (void) |
| 347 | { |
| 348 | return register_md_personality (&linear_personality); |
| 349 | } |
| 350 | |
| 351 | static void linear_exit (void) |
| 352 | { |
| 353 | unregister_md_personality (&linear_personality); |
| 354 | } |
| 355 | |
| 356 | |
| 357 | module_init(linear_init); |
| 358 | module_exit(linear_exit); |
| 359 | MODULE_LICENSE("GPL"); |
| 360 | MODULE_DESCRIPTION("Linear device concatenation personality for MD"); |
| 361 | MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/ |
| 362 | MODULE_ALIAS("md-linear"); |
| 363 | MODULE_ALIAS("md-level--1"); |