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