drivers/md/linear.c

   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         block >>= conf->preshift;
  42         (void)sector_div(block, conf->hash_spacing);
  43         hash = conf->hash_table[block];
  44
  45         while ((sector>>1) >= (hash->size + hash->offset))
  46                 hash++;
  47         return hash;
  48 }
  49
  50 /**
  51  *      linear_mergeable_bvec -- tell bio layer if two requests can be merged
  52  *      @q: request queue
  53  *      @bio: the buffer head that's been built up so far
  54  *      @biovec: the request that could be merged to it.
  55  *
  56  *      Return amount of bytes we can take at this offset
  57  */
  58 static int linear_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec)
  59 {
  60         mddev_t *mddev = q->queuedata;
  61         dev_info_t *dev0;
  62         unsigned long maxsectors, bio_sectors = bio->bi_size >> 9;
  63         sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
  64
  65         dev0 = which_dev(mddev, sector);
  66         maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1));
  67
  68         if (maxsectors < bio_sectors)
  69                 maxsectors = 0;
  70         else
  71                 maxsectors -= bio_sectors;
  72
  73         if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
  74                 return biovec->bv_len;
  75         /* The bytes available at this offset could be really big,
  76          * so we cap at 2^31 to avoid overflow */
  77         if (maxsectors > (1 << (31-9)))
  78                 return 1<<31;
  79         return maxsectors << 9;
  80 }
  81
  82 static void linear_unplug(request_queue_t *q)
  83 {
  84         mddev_t *mddev = q->queuedata;
  85         linear_conf_t *conf = mddev_to_conf(mddev);
  86         int i;
  87
  88         for (i=0; i < mddev->raid_disks; i++) {
  89                 request_queue_t *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
  90                 if (r_queue->unplug_fn)
  91                         r_queue->unplug_fn(r_queue);
  92         }
  93 }
  94
  95 static int linear_issue_flush(request_queue_t *q, struct gendisk *disk,
  96                               sector_t *error_sector)
  97 {
  98         mddev_t *mddev = q->queuedata;
  99         linear_conf_t *conf = mddev_to_conf(mddev);
 100         int i, ret = 0;
 101
 102         for (i=0; i < mddev->raid_disks && ret == 0; i++) {
 103                 struct block_device *bdev = conf->disks[i].rdev->bdev;
 104                 request_queue_t *r_queue = bdev_get_queue(bdev);
 105
 106                 if (!r_queue->issue_flush_fn)
 107                         ret = -EOPNOTSUPP;
 108                 else
 109                         ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector);
 110         }
 111         return ret;
 112 }
 113
 114 static int linear_congested(void *data, int bits)
 115 {
 116         mddev_t *mddev = data;
 117         linear_conf_t *conf = mddev_to_conf(mddev);
 118         int i, ret = 0;
 119
 120         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
 121                 request_queue_t *q = bdev_get_queue(conf->disks[i].rdev->bdev);
 122                 ret |= bdi_congested(&q->backing_dev_info, bits);
 123         }
 124         return ret;
 125 }
 126
 127 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
 128 {
 129         linear_conf_t *conf;
 130         dev_info_t **table;
 131         mdk_rdev_t *rdev;
 132         int i, nb_zone, cnt;
 133         sector_t min_spacing;
 134         sector_t curr_offset;
 135         struct list_head *tmp;
 136
 137         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
 138                         GFP_KERNEL);
 139         if (!conf)
 140                 return NULL;
 141
 142         cnt = 0;
 143         conf->array_size = 0;
 144
 145         ITERATE_RDEV(mddev,rdev,tmp) {
 146                 int j = rdev->raid_disk;
 147                 dev_info_t *disk = conf->disks + j;
 148
 149                 if (j < 0 || j > raid_disks || disk->rdev) {
 150                         printk("linear: disk numbering problem. Aborting!\n");
 151                         goto out;
 152                 }
 153
 154                 disk->rdev = rdev;
 155
 156                 blk_queue_stack_limits(mddev->queue,
 157                                        rdev->bdev->bd_disk->queue);
 158                 /* as we don't honour merge_bvec_fn, we must never risk
 159                  * violating it, so limit ->max_sector to one PAGE, as
 160                  * a one page request is never in violation.
 161                  */
 162                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
 163                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
 164                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 165
 166                 disk->size = rdev->size;
 167                 conf->array_size += rdev->size;
 168
 169                 cnt++;
 170         }
 171         if (cnt != raid_disks) {
 172                 printk("linear: not enough drives present. Aborting!\n");
 173                 goto out;
 174         }
 175
 176         min_spacing = conf->array_size;
 177         sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *));
 178
 179         /* min_spacing is the minimum spacing that will fit the hash
 180          * table in one PAGE.  This may be much smaller than needed.
 181          * We find the smallest non-terminal set of consecutive devices
 182          * that is larger than min_spacing as use the size of that as
 183          * the actual spacing
 184          */
 185         conf->hash_spacing = conf->array_size;
 186         for (i=0; i < cnt-1 ; i++) {
 187                 sector_t sz = 0;
 188                 int j;
 189                 for (j = i; j < cnt - 1 && sz < min_spacing; j++)
 190                         sz += conf->disks[j].size;
 191                 if (sz >= min_spacing && sz < conf->hash_spacing)
 192                         conf->hash_spacing = sz;
 193         }
 194
 195         /* hash_spacing may be too large for sector_div to work with,
 196          * so we might need to pre-shift
 197          */
 198         conf->preshift = 0;
 199         if (sizeof(sector_t) > sizeof(u32)) {
 200                 sector_t space = conf->hash_spacing;
 201                 while (space > (sector_t)(~(u32)0)) {
 202                         space >>= 1;
 203                         conf->preshift++;
 204                 }
 205         }
 206         /*
 207          * This code was restructured to work around a gcc-2.95.3 internal
 208          * compiler error.  Alter it with care.
 209          */
 210         {
 211                 sector_t sz;
 212                 unsigned round;
 213                 unsigned long base;
 214
 215                 sz = conf->array_size >> conf->preshift;
 216                 sz += 1; /* force round-up */
 217                 base = conf->hash_spacing >> conf->preshift;
 218                 round = sector_div(sz, base);
 219                 nb_zone = sz + (round ? 1 : 0);
 220         }
 221         BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
 222
 223         conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
 224                                         GFP_KERNEL);
 225         if (!conf->hash_table)
 226                 goto out;
 227
 228         /*
 229          * Here we generate the linear hash table
 230          * First calculate the device offsets.
 231          */
 232         conf->disks[0].offset = 0;
 233         for (i = 1; i < raid_disks; i++)
 234                 conf->disks[i].offset =
 235                         conf->disks[i-1].offset +
 236                         conf->disks[i-1].size;
 237
 238         table = conf->hash_table;
 239         curr_offset = 0;
 240         i = 0;
 241         for (curr_offset = 0;
 242              curr_offset < conf->array_size;
 243              curr_offset += conf->hash_spacing) {
 244
 245                 while (i < raid_disks-1 &&
 246                        curr_offset >= conf->disks[i+1].offset)
 247                         i++;
 248
 249                 *table ++ = conf->disks + i;
 250         }
 251
 252         if (conf->preshift) {
 253                 conf->hash_spacing >>= conf->preshift;
 254                 /* round hash_spacing up so that when we divide by it,
 255                  * we err on the side of "too-low", which is safest.
 256                  */
 257                 conf->hash_spacing++;
 258         }
 259
 260         BUG_ON(table - conf->hash_table > nb_zone);
 261
 262         return conf;
 263
 264 out:
 265         kfree(conf);
 266         return NULL;
 267 }
 268
 269 static int linear_run (mddev_t *mddev)
 270 {
 271         linear_conf_t *conf;
 272
 273         conf = linear_conf(mddev, mddev->raid_disks);
 274
 275         if (!conf)
 276                 return 1;
 277         mddev->private = conf;
 278         mddev->array_size = conf->array_size;
 279
 280         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 281         mddev->queue->unplug_fn = linear_unplug;
 282         mddev->queue->issue_flush_fn = linear_issue_flush;
 283         mddev->queue->backing_dev_info.congested_fn = linear_congested;
 284         mddev->queue->backing_dev_info.congested_data = mddev;
 285         return 0;
 286 }
 287
 288 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
 289 {
 290         /* Adding a drive to a linear array allows the array to grow.
 291          * It is permitted if the new drive has a matching superblock
 292          * already on it, with raid_disk equal to raid_disks.
 293          * It is achieved by creating a new linear_private_data structure
 294          * and swapping it in in-place of the current one.
 295          * The current one is never freed until the array is stopped.
 296          * This avoids races.
 297          */
 298         linear_conf_t *newconf;
 299
 300         if (rdev->saved_raid_disk != mddev->raid_disks)
 301                 return -EINVAL;
 302
 303         rdev->raid_disk = rdev->saved_raid_disk;
 304
 305         newconf = linear_conf(mddev,mddev->raid_disks+1);
 306
 307         if (!newconf)
 308                 return -ENOMEM;
 309
 310         newconf->prev = mddev_to_conf(mddev);
 311         mddev->private = newconf;
 312         mddev->raid_disks++;
 313         mddev->array_size = newconf->array_size;
 314         set_capacity(mddev->gendisk, mddev->array_size << 1);
 315         return 0;
 316 }
 317
 318 static int linear_stop (mddev_t *mddev)
 319 {
 320         linear_conf_t *conf = mddev_to_conf(mddev);
 321
 322         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 323         do {
 324                 linear_conf_t *t = conf->prev;
 325                 kfree(conf->hash_table);
 326                 kfree(conf);
 327                 conf = t;
 328         } while (conf);
 329
 330         return 0;
 331 }
 332
 333 static int linear_make_request (request_queue_t *q, struct bio *bio)
 334 {
 335         const int rw = bio_data_dir(bio);
 336         mddev_t *mddev = q->queuedata;
 337         dev_info_t *tmp_dev;
 338         sector_t block;
 339
 340         if (unlikely(bio_barrier(bio))) {
 341                 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
 342                 return 0;
 343         }
 344
 345         disk_stat_inc(mddev->gendisk, ios[rw]);
 346         disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
 347
 348         tmp_dev = which_dev(mddev, bio->bi_sector);
 349         block = bio->bi_sector >> 1;
 350
 351         if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
 352                      || block < tmp_dev->offset)) {
 353                 char b[BDEVNAME_SIZE];
 354
 355                 printk("linear_make_request: Block %llu out of bounds on "
 356                         "dev %s size %llu offset %llu\n",
 357                         (unsigned long long)block,
 358                         bdevname(tmp_dev->rdev->bdev, b),
 359                         (unsigned long long)tmp_dev->size,
 360                         (unsigned long long)tmp_dev->offset);
 361                 bio_io_error(bio, bio->bi_size);
 362                 return 0;
 363         }
 364         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
 365                      (tmp_dev->offset + tmp_dev->size)<<1)) {
 366                 /* This bio crosses a device boundary, so we have to
 367                  * split it.
 368                  */
 369                 struct bio_pair *bp;
 370                 bp = bio_split(bio, bio_split_pool,
 371                                ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
 372                 if (linear_make_request(q, &bp->bio1))
 373                         generic_make_request(&bp->bio1);
 374                 if (linear_make_request(q, &bp->bio2))
 375                         generic_make_request(&bp->bio2);
 376                 bio_pair_release(bp);
 377                 return 0;
 378         }
 379
 380         bio->bi_bdev = tmp_dev->rdev->bdev;
 381         bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
 382
 383         return 1;
 384 }
 385
 386 static void linear_status (struct seq_file *seq, mddev_t *mddev)
 387 {
 388
 389 #undef MD_DEBUG
 390 #ifdef MD_DEBUG
 391         int j;
 392         linear_conf_t *conf = mddev_to_conf(mddev);
 393         sector_t s = 0;
 394
 395         seq_printf(seq, "      ");
 396         for (j = 0; j < mddev->raid_disks; j++)
 397         {
 398                 char b[BDEVNAME_SIZE];
 399                 s += conf->smallest_size;
 400                 seq_printf(seq, "[%s",
 401                            bdevname(conf->hash_table[j][0].rdev->bdev,b));
 402
 403                 while (s > conf->hash_table[j][0].offset +
 404                            conf->hash_table[j][0].size)
 405                         seq_printf(seq, "/%s] ",
 406                                    bdevname(conf->hash_table[j][1].rdev->bdev,b));
 407                 else
 408                         seq_printf(seq, "] ");
 409         }
 410         seq_printf(seq, "\n");
 411 #endif
 412         seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
 413 }
 414
 415
 416 static struct mdk_personality linear_personality =
 417 {
 418         .name           = "linear",
 419         .level          = LEVEL_LINEAR,
 420         .owner          = THIS_MODULE,
 421         .make_request   = linear_make_request,
 422         .run            = linear_run,
 423         .stop           = linear_stop,
 424         .status         = linear_status,
 425         .hot_add_disk   = linear_add,
 426 };
 427
 428 static int __init linear_init (void)
 429 {
 430         return register_md_personality (&linear_personality);
 431 }
 432
 433 static void linear_exit (void)
 434 {
 435         unregister_md_personality (&linear_personality);
 436 }
 437
 438
 439 module_init(linear_init);
 440 module_exit(linear_exit);
 441 MODULE_LICENSE("GPL");
 442 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 443 MODULE_ALIAS("md-linear");
 444 MODULE_ALIAS("md-level--1");