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