Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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/module.h> |
| 20 | |
| 21 | #include <linux/raid/md.h> |
| 22 | #include <linux/slab.h> |
| 23 | #include <linux/raid/linear.h> |
| 24 | |
| 25 | #define MAJOR_NR MD_MAJOR |
| 26 | #define MD_DRIVER |
| 27 | #define MD_PERSONALITY |
| 28 | |
| 29 | /* |
| 30 | * find which device holds a particular offset |
| 31 | */ |
| 32 | static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector) |
| 33 | { |
| 34 | dev_info_t *hash; |
| 35 | linear_conf_t *conf = mddev_to_conf(mddev); |
| 36 | sector_t block = sector >> 1; |
| 37 | |
| 38 | /* |
| 39 | * sector_div(a,b) returns the remainer and sets a to a/b |
| 40 | */ |
| 41 | (void)sector_div(block, conf->smallest->size); |
| 42 | hash = conf->hash_table[block]; |
| 43 | |
| 44 | while ((sector>>1) >= (hash->size + hash->offset)) |
| 45 | hash++; |
| 46 | return hash; |
| 47 | } |
| 48 | |
| 49 | /** |
| 50 | * linear_mergeable_bvec -- tell bio layer if a two requests can be merged |
| 51 | * @q: request queue |
| 52 | * @bio: the buffer head that's been built up so far |
| 53 | * @biovec: the request that could be merged to it. |
| 54 | * |
| 55 | * Return amount of bytes we can take at this offset |
| 56 | */ |
| 57 | static int linear_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec) |
| 58 | { |
| 59 | mddev_t *mddev = q->queuedata; |
| 60 | dev_info_t *dev0; |
| 61 | unsigned long maxsectors, bio_sectors = bio->bi_size >> 9; |
| 62 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); |
| 63 | |
| 64 | dev0 = which_dev(mddev, sector); |
| 65 | maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1)); |
| 66 | |
| 67 | if (maxsectors < bio_sectors) |
| 68 | maxsectors = 0; |
| 69 | else |
| 70 | maxsectors -= bio_sectors; |
| 71 | |
| 72 | if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0) |
| 73 | return biovec->bv_len; |
| 74 | /* The bytes available at this offset could be really big, |
| 75 | * so we cap at 2^31 to avoid overflow */ |
| 76 | if (maxsectors > (1 << (31-9))) |
| 77 | return 1<<31; |
| 78 | return maxsectors << 9; |
| 79 | } |
| 80 | |
| 81 | static void linear_unplug(request_queue_t *q) |
| 82 | { |
| 83 | mddev_t *mddev = q->queuedata; |
| 84 | linear_conf_t *conf = mddev_to_conf(mddev); |
| 85 | int i; |
| 86 | |
| 87 | for (i=0; i < mddev->raid_disks; i++) { |
| 88 | request_queue_t *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev); |
| 89 | if (r_queue->unplug_fn) |
| 90 | r_queue->unplug_fn(r_queue); |
| 91 | } |
| 92 | } |
| 93 | |
| 94 | static int linear_issue_flush(request_queue_t *q, struct gendisk *disk, |
| 95 | sector_t *error_sector) |
| 96 | { |
| 97 | mddev_t *mddev = q->queuedata; |
| 98 | linear_conf_t *conf = mddev_to_conf(mddev); |
| 99 | int i, ret = 0; |
| 100 | |
| 101 | for (i=0; i < mddev->raid_disks && ret == 0; i++) { |
| 102 | struct block_device *bdev = conf->disks[i].rdev->bdev; |
| 103 | request_queue_t *r_queue = bdev_get_queue(bdev); |
| 104 | |
| 105 | if (!r_queue->issue_flush_fn) |
| 106 | ret = -EOPNOTSUPP; |
| 107 | else |
| 108 | ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector); |
| 109 | } |
| 110 | return ret; |
| 111 | } |
| 112 | |
| 113 | static int linear_run (mddev_t *mddev) |
| 114 | { |
| 115 | linear_conf_t *conf; |
| 116 | dev_info_t **table; |
| 117 | mdk_rdev_t *rdev; |
| 118 | int i, nb_zone, cnt; |
| 119 | sector_t start; |
| 120 | sector_t curr_offset; |
| 121 | struct list_head *tmp; |
| 122 | |
| 123 | conf = kmalloc (sizeof (*conf) + mddev->raid_disks*sizeof(dev_info_t), |
| 124 | GFP_KERNEL); |
| 125 | if (!conf) |
| 126 | goto out; |
| 127 | memset(conf, 0, sizeof(*conf) + mddev->raid_disks*sizeof(dev_info_t)); |
| 128 | mddev->private = conf; |
| 129 | |
| 130 | /* |
| 131 | * Find the smallest device. |
| 132 | */ |
| 133 | |
| 134 | conf->smallest = NULL; |
| 135 | cnt = 0; |
| 136 | mddev->array_size = 0; |
| 137 | |
| 138 | ITERATE_RDEV(mddev,rdev,tmp) { |
| 139 | int j = rdev->raid_disk; |
| 140 | dev_info_t *disk = conf->disks + j; |
| 141 | |
| 142 | if (j < 0 || j > mddev->raid_disks || disk->rdev) { |
| 143 | printk("linear: disk numbering problem. Aborting!\n"); |
| 144 | goto out; |
| 145 | } |
| 146 | |
| 147 | disk->rdev = rdev; |
| 148 | |
| 149 | blk_queue_stack_limits(mddev->queue, |
| 150 | rdev->bdev->bd_disk->queue); |
| 151 | /* as we don't honour merge_bvec_fn, we must never risk |
| 152 | * violating it, so limit ->max_sector to one PAGE, as |
| 153 | * a one page request is never in violation. |
| 154 | */ |
| 155 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && |
| 156 | mddev->queue->max_sectors > (PAGE_SIZE>>9)) |
| 157 | blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); |
| 158 | |
| 159 | disk->size = rdev->size; |
| 160 | mddev->array_size += rdev->size; |
| 161 | |
| 162 | if (!conf->smallest || (disk->size < conf->smallest->size)) |
| 163 | conf->smallest = disk; |
| 164 | cnt++; |
| 165 | } |
| 166 | if (cnt != mddev->raid_disks) { |
| 167 | printk("linear: not enough drives present. Aborting!\n"); |
| 168 | goto out; |
| 169 | } |
| 170 | |
| 171 | /* |
| 172 | * This code was restructured to work around a gcc-2.95.3 internal |
| 173 | * compiler error. Alter it with care. |
| 174 | */ |
| 175 | { |
| 176 | sector_t sz; |
| 177 | unsigned round; |
| 178 | unsigned long base; |
| 179 | |
| 180 | sz = mddev->array_size; |
| 181 | base = conf->smallest->size; |
| 182 | round = sector_div(sz, base); |
| 183 | nb_zone = conf->nr_zones = sz + (round ? 1 : 0); |
| 184 | } |
| 185 | |
| 186 | conf->hash_table = kmalloc (sizeof (dev_info_t*) * nb_zone, |
| 187 | GFP_KERNEL); |
| 188 | if (!conf->hash_table) |
| 189 | goto out; |
| 190 | |
| 191 | /* |
| 192 | * Here we generate the linear hash table |
| 193 | */ |
| 194 | table = conf->hash_table; |
| 195 | start = 0; |
| 196 | curr_offset = 0; |
| 197 | for (i = 0; i < cnt; i++) { |
| 198 | dev_info_t *disk = conf->disks + i; |
| 199 | |
| 200 | disk->offset = curr_offset; |
| 201 | curr_offset += disk->size; |
| 202 | |
| 203 | /* 'curr_offset' is the end of this disk |
| 204 | * 'start' is the start of table |
| 205 | */ |
| 206 | while (start < curr_offset) { |
| 207 | *table++ = disk; |
| 208 | start += conf->smallest->size; |
| 209 | } |
| 210 | } |
| 211 | if (table-conf->hash_table != nb_zone) |
| 212 | BUG(); |
| 213 | |
| 214 | blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec); |
| 215 | mddev->queue->unplug_fn = linear_unplug; |
| 216 | mddev->queue->issue_flush_fn = linear_issue_flush; |
| 217 | return 0; |
| 218 | |
| 219 | out: |
| 220 | if (conf) |
| 221 | kfree(conf); |
| 222 | return 1; |
| 223 | } |
| 224 | |
| 225 | static int linear_stop (mddev_t *mddev) |
| 226 | { |
| 227 | linear_conf_t *conf = mddev_to_conf(mddev); |
| 228 | |
| 229 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
| 230 | kfree(conf->hash_table); |
| 231 | kfree(conf); |
| 232 | |
| 233 | return 0; |
| 234 | } |
| 235 | |
| 236 | static int linear_make_request (request_queue_t *q, struct bio *bio) |
| 237 | { |
| 238 | mddev_t *mddev = q->queuedata; |
| 239 | dev_info_t *tmp_dev; |
| 240 | sector_t block; |
| 241 | |
| 242 | if (bio_data_dir(bio)==WRITE) { |
| 243 | disk_stat_inc(mddev->gendisk, writes); |
| 244 | disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio)); |
| 245 | } else { |
| 246 | disk_stat_inc(mddev->gendisk, reads); |
| 247 | disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio)); |
| 248 | } |
| 249 | |
| 250 | tmp_dev = which_dev(mddev, bio->bi_sector); |
| 251 | block = bio->bi_sector >> 1; |
| 252 | |
| 253 | if (unlikely(block >= (tmp_dev->size + tmp_dev->offset) |
| 254 | || block < tmp_dev->offset)) { |
| 255 | char b[BDEVNAME_SIZE]; |
| 256 | |
| 257 | printk("linear_make_request: Block %llu out of bounds on " |
| 258 | "dev %s size %llu offset %llu\n", |
| 259 | (unsigned long long)block, |
| 260 | bdevname(tmp_dev->rdev->bdev, b), |
| 261 | (unsigned long long)tmp_dev->size, |
| 262 | (unsigned long long)tmp_dev->offset); |
| 263 | bio_io_error(bio, bio->bi_size); |
| 264 | return 0; |
| 265 | } |
| 266 | if (unlikely(bio->bi_sector + (bio->bi_size >> 9) > |
| 267 | (tmp_dev->offset + tmp_dev->size)<<1)) { |
| 268 | /* This bio crosses a device boundary, so we have to |
| 269 | * split it. |
| 270 | */ |
| 271 | struct bio_pair *bp; |
NeilBrown | 29ac8e0 | 2005-05-16 21:53:15 -0700 | [diff] [blame] | 272 | bp = bio_split(bio, bio_split_pool, |
| 273 | ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | if (linear_make_request(q, &bp->bio1)) |
| 275 | generic_make_request(&bp->bio1); |
| 276 | if (linear_make_request(q, &bp->bio2)) |
| 277 | generic_make_request(&bp->bio2); |
| 278 | bio_pair_release(bp); |
| 279 | return 0; |
| 280 | } |
| 281 | |
| 282 | bio->bi_bdev = tmp_dev->rdev->bdev; |
| 283 | bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset; |
| 284 | |
| 285 | return 1; |
| 286 | } |
| 287 | |
| 288 | static void linear_status (struct seq_file *seq, mddev_t *mddev) |
| 289 | { |
| 290 | |
| 291 | #undef MD_DEBUG |
| 292 | #ifdef MD_DEBUG |
| 293 | int j; |
| 294 | linear_conf_t *conf = mddev_to_conf(mddev); |
| 295 | sector_t s = 0; |
| 296 | |
| 297 | seq_printf(seq, " "); |
| 298 | for (j = 0; j < conf->nr_zones; j++) |
| 299 | { |
| 300 | char b[BDEVNAME_SIZE]; |
| 301 | s += conf->smallest_size; |
| 302 | seq_printf(seq, "[%s", |
| 303 | bdevname(conf->hash_table[j][0].rdev->bdev,b)); |
| 304 | |
| 305 | while (s > conf->hash_table[j][0].offset + |
| 306 | conf->hash_table[j][0].size) |
| 307 | seq_printf(seq, "/%s] ", |
| 308 | bdevname(conf->hash_table[j][1].rdev->bdev,b)); |
| 309 | else |
| 310 | seq_printf(seq, "] "); |
| 311 | } |
| 312 | seq_printf(seq, "\n"); |
| 313 | #endif |
| 314 | seq_printf(seq, " %dk rounding", mddev->chunk_size/1024); |
| 315 | } |
| 316 | |
| 317 | |
| 318 | static mdk_personality_t linear_personality= |
| 319 | { |
| 320 | .name = "linear", |
| 321 | .owner = THIS_MODULE, |
| 322 | .make_request = linear_make_request, |
| 323 | .run = linear_run, |
| 324 | .stop = linear_stop, |
| 325 | .status = linear_status, |
| 326 | }; |
| 327 | |
| 328 | static int __init linear_init (void) |
| 329 | { |
| 330 | return register_md_personality (LINEAR, &linear_personality); |
| 331 | } |
| 332 | |
| 333 | static void linear_exit (void) |
| 334 | { |
| 335 | unregister_md_personality (LINEAR); |
| 336 | } |
| 337 | |
| 338 | |
| 339 | module_init(linear_init); |
| 340 | module_exit(linear_exit); |
| 341 | MODULE_LICENSE("GPL"); |
| 342 | MODULE_ALIAS("md-personality-1"); /* LINEAR */ |