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/raid/linear.h>
  20
  21 /*
  22  * find which device holds a particular offset
  23  */
  24 static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
  25 {
  26         dev_info_t *hash;
  27         linear_conf_t *conf = mddev_to_conf(mddev);
  28         sector_t idx = sector >> conf->sector_shift;
  29
  30         /*
  31          * sector_div(a,b) returns the remainer and sets a to a/b
  32          */
  33         (void)sector_div(idx, conf->spacing);
  34         hash = conf->hash_table[idx];
  35
  36         while (sector >= hash->num_sectors + hash->start_sector)
  37                 hash++;
  38         return hash;
  39 }
  40
  41 /**
  42  *      linear_mergeable_bvec -- tell bio layer if two requests can be merged
  43  *      @q: request queue
  44  *      @bvm: properties of new bio
  45  *      @biovec: the request that could be merged to it.
  46  *
  47  *      Return amount of bytes we can take at this offset
  48  */
  49 static int linear_mergeable_bvec(struct request_queue *q,
  50                                  struct bvec_merge_data *bvm,
  51                                  struct bio_vec *biovec)
  52 {
  53         mddev_t *mddev = q->queuedata;
  54         dev_info_t *dev0;
  55         unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
  56         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
  57
  58         dev0 = which_dev(mddev, sector);
  59         maxsectors = dev0->num_sectors - (sector - dev0->start_sector);
  60
  61         if (maxsectors < bio_sectors)
  62                 maxsectors = 0;
  63         else
  64                 maxsectors -= bio_sectors;
  65
  66         if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
  67                 return biovec->bv_len;
  68         /* The bytes available at this offset could be really big,
  69          * so we cap at 2^31 to avoid overflow */
  70         if (maxsectors > (1 << (31-9)))
  71                 return 1<<31;
  72         return maxsectors << 9;
  73 }
  74
  75 static void linear_unplug(struct request_queue *q)
  76 {
  77         mddev_t *mddev = q->queuedata;
  78         linear_conf_t *conf = mddev_to_conf(mddev);
  79         int i;
  80
  81         for (i=0; i < mddev->raid_disks; i++) {
  82                 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
  83                 blk_unplug(r_queue);
  84         }
  85 }
  86
  87 static int linear_congested(void *data, int bits)
  88 {
  89         mddev_t *mddev = data;
  90         linear_conf_t *conf = mddev_to_conf(mddev);
  91         int i, ret = 0;
  92
  93         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
  94                 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
  95                 ret |= bdi_congested(&q->backing_dev_info, bits);
  96         }
  97         return ret;
  98 }
  99
 100 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
 101 {
 102         linear_conf_t *conf;
 103         dev_info_t **table;
 104         mdk_rdev_t *rdev;
 105         int i, nb_zone, cnt;
 106         sector_t min_sectors;
 107         sector_t curr_sector;
 108
 109         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
 110                         GFP_KERNEL);
 111         if (!conf)
 112                 return NULL;
 113
 114         cnt = 0;
 115         conf->array_sectors = 0;
 116
 117         list_for_each_entry(rdev, &mddev->disks, same_set) {
 118                 int j = rdev->raid_disk;
 119                 dev_info_t *disk = conf->disks + j;
 120
 121                 if (j < 0 || j >= raid_disks || disk->rdev) {
 122                         printk("linear: disk numbering problem. Aborting!\n");
 123                         goto out;
 124                 }
 125
 126                 disk->rdev = rdev;
 127
 128                 blk_queue_stack_limits(mddev->queue,
 129                                        rdev->bdev->bd_disk->queue);
 130                 /* as we don't honour merge_bvec_fn, we must never risk
 131                  * violating it, so limit ->max_sector to one PAGE, as
 132                  * a one page request is never in violation.
 133                  */
 134                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
 135                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
 136                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 137
 138                 disk->num_sectors = rdev->size * 2;
 139                 conf->array_sectors += rdev->size * 2;
 140
 141                 cnt++;
 142         }
 143         if (cnt != raid_disks) {
 144                 printk("linear: not enough drives present. Aborting!\n");
 145                 goto out;
 146         }
 147
 148         min_sectors = conf->array_sectors;
 149         sector_div(min_sectors, PAGE_SIZE/sizeof(struct dev_info *));
 150         if (min_sectors == 0)
 151                 min_sectors = 1;
 152
 153         /* min_sectors is the minimum spacing that will fit the hash
 154          * table in one PAGE.  This may be much smaller than needed.
 155          * We find the smallest non-terminal set of consecutive devices
 156          * that is larger than min_sectors and use the size of that as
 157          * the actual spacing
 158          */
 159         conf->spacing = conf->array_sectors;
 160         for (i=0; i < cnt-1 ; i++) {
 161                 sector_t tmp = 0;
 162                 int j;
 163                 for (j = i; j < cnt - 1 && tmp < min_sectors; j++)
 164                         tmp += conf->disks[j].num_sectors;
 165                 if (tmp >= min_sectors && tmp < conf->spacing)
 166                         conf->spacing = tmp;
 167         }
 168
 169         /* spacing may be too large for sector_div to work with,
 170          * so we might need to pre-shift
 171          */
 172         conf->sector_shift = 0;
 173         if (sizeof(sector_t) > sizeof(u32)) {
 174                 sector_t space = conf->spacing;
 175                 while (space > (sector_t)(~(u32)0)) {
 176                         space >>= 1;
 177                         conf->sector_shift++;
 178                 }
 179         }
 180         /*
 181          * This code was restructured to work around a gcc-2.95.3 internal
 182          * compiler error.  Alter it with care.
 183          */
 184         {
 185                 sector_t sz;
 186                 unsigned round;
 187                 unsigned long base;
 188
 189                 sz = conf->array_sectors >> conf->sector_shift;
 190                 sz += 1; /* force round-up */
 191                 base = conf->spacing >> conf->sector_shift;
 192                 round = sector_div(sz, base);
 193                 nb_zone = sz + (round ? 1 : 0);
 194         }
 195         BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
 196
 197         conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
 198                                         GFP_KERNEL);
 199         if (!conf->hash_table)
 200                 goto out;
 201
 202         /*
 203          * Here we generate the linear hash table
 204          * First calculate the device offsets.
 205          */
 206         conf->disks[0].start_sector = 0;
 207         for (i = 1; i < raid_disks; i++)
 208                 conf->disks[i].start_sector =
 209                         conf->disks[i-1].start_sector +
 210                         conf->disks[i-1].num_sectors;
 211
 212         table = conf->hash_table;
 213         i = 0;
 214         for (curr_sector = 0;
 215              curr_sector < conf->array_sectors;
 216              curr_sector += conf->spacing) {
 217
 218                 while (i < raid_disks-1 &&
 219                        curr_sector >= conf->disks[i+1].start_sector)
 220                         i++;
 221
 222                 *table ++ = conf->disks + i;
 223         }
 224
 225         if (conf->sector_shift) {
 226                 conf->spacing >>= conf->sector_shift;
 227                 /* round spacing up so that when we divide by it,
 228                  * we err on the side of "too-low", which is safest.
 229                  */
 230                 conf->spacing++;
 231         }
 232
 233         BUG_ON(table - conf->hash_table > nb_zone);
 234
 235         return conf;
 236
 237 out:
 238         kfree(conf);
 239         return NULL;
 240 }
 241
 242 static int linear_run (mddev_t *mddev)
 243 {
 244         linear_conf_t *conf;
 245
 246         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 247         conf = linear_conf(mddev, mddev->raid_disks);
 248
 249         if (!conf)
 250                 return 1;
 251         mddev->private = conf;
 252         mddev->array_sectors = conf->array_sectors;
 253
 254         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 255         mddev->queue->unplug_fn = linear_unplug;
 256         mddev->queue->backing_dev_info.congested_fn = linear_congested;
 257         mddev->queue->backing_dev_info.congested_data = mddev;
 258         return 0;
 259 }
 260
 261 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
 262 {
 263         /* Adding a drive to a linear array allows the array to grow.
 264          * It is permitted if the new drive has a matching superblock
 265          * already on it, with raid_disk equal to raid_disks.
 266          * It is achieved by creating a new linear_private_data structure
 267          * and swapping it in in-place of the current one.
 268          * The current one is never freed until the array is stopped.
 269          * This avoids races.
 270          */
 271         linear_conf_t *newconf;
 272
 273         if (rdev->saved_raid_disk != mddev->raid_disks)
 274                 return -EINVAL;
 275
 276         rdev->raid_disk = rdev->saved_raid_disk;
 277
 278         newconf = linear_conf(mddev,mddev->raid_disks+1);
 279
 280         if (!newconf)
 281                 return -ENOMEM;
 282
 283         newconf->prev = mddev_to_conf(mddev);
 284         mddev->private = newconf;
 285         mddev->raid_disks++;
 286         mddev->array_sectors = newconf->array_sectors;
 287         set_capacity(mddev->gendisk, mddev->array_sectors);
 288         return 0;
 289 }
 290
 291 static int linear_stop (mddev_t *mddev)
 292 {
 293         linear_conf_t *conf = mddev_to_conf(mddev);
 294
 295         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 296         do {
 297                 linear_conf_t *t = conf->prev;
 298                 kfree(conf->hash_table);
 299                 kfree(conf);
 300                 conf = t;
 301         } while (conf);
 302
 303         return 0;
 304 }
 305
 306 static int linear_make_request (struct request_queue *q, struct bio *bio)
 307 {
 308         const int rw = bio_data_dir(bio);
 309         mddev_t *mddev = q->queuedata;
 310         dev_info_t *tmp_dev;
 311         int cpu;
 312
 313         if (unlikely(bio_barrier(bio))) {
 314                 bio_endio(bio, -EOPNOTSUPP);
 315                 return 0;
 316         }
 317
 318         cpu = part_stat_lock();
 319         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 320         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 321                       bio_sectors(bio));
 322         part_stat_unlock();
 323
 324         tmp_dev = which_dev(mddev, bio->bi_sector);
 325
 326         if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors +
 327                                         tmp_dev->start_sector)
 328                      || (bio->bi_sector <
 329                          tmp_dev->start_sector))) {
 330                 char b[BDEVNAME_SIZE];
 331
 332                 printk("linear_make_request: Sector %llu out of bounds on "
 333                         "dev %s: %llu sectors, offset %llu\n",
 334                         (unsigned long long)bio->bi_sector,
 335                         bdevname(tmp_dev->rdev->bdev, b),
 336                         (unsigned long long)tmp_dev->num_sectors,
 337                         (unsigned long long)tmp_dev->start_sector);
 338                 bio_io_error(bio);
 339                 return 0;
 340         }
 341         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
 342                      tmp_dev->start_sector + tmp_dev->num_sectors)) {
 343                 /* This bio crosses a device boundary, so we have to
 344                  * split it.
 345                  */
 346                 struct bio_pair *bp;
 347
 348                 bp = bio_split(bio,
 349                                tmp_dev->start_sector + tmp_dev->num_sectors
 350                                - bio->bi_sector);
 351
 352                 if (linear_make_request(q, &bp->bio1))
 353                         generic_make_request(&bp->bio1);
 354                 if (linear_make_request(q, &bp->bio2))
 355                         generic_make_request(&bp->bio2);
 356                 bio_pair_release(bp);
 357                 return 0;
 358         }
 359
 360         bio->bi_bdev = tmp_dev->rdev->bdev;
 361         bio->bi_sector = bio->bi_sector - tmp_dev->start_sector
 362                 + tmp_dev->rdev->data_offset;
 363
 364         return 1;
 365 }
 366
 367 static void linear_status (struct seq_file *seq, mddev_t *mddev)
 368 {
 369
 370         seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
 371 }
 372
 373
 374 static struct mdk_personality linear_personality =
 375 {
 376         .name           = "linear",
 377         .level          = LEVEL_LINEAR,
 378         .owner          = THIS_MODULE,
 379         .make_request   = linear_make_request,
 380         .run            = linear_run,
 381         .stop           = linear_stop,
 382         .status         = linear_status,
 383         .hot_add_disk   = linear_add,
 384 };
 385
 386 static int __init linear_init (void)
 387 {
 388         return register_md_personality (&linear_personality);
 389 }
 390
 391 static void linear_exit (void)
 392 {
 393         unregister_md_personality (&linear_personality);
 394 }
 395
 396
 397 module_init(linear_init);
 398 module_exit(linear_exit);
 399 MODULE_LICENSE("GPL");
 400 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 401 MODULE_ALIAS("md-linear");
 402 MODULE_ALIAS("md-level--1");