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