drivers/md/raid0.c

   1 /*
   2    raid0.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              Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
   7
   8
   9    RAID-0 management functions.
  10
  11    This program is free software; you can redistribute it and/or modify
  12    it under the terms of the GNU General Public License as published by
  13    the Free Software Foundation; either version 2, or (at your option)
  14    any later version.
  15
  16    You should have received a copy of the GNU General Public License
  17    (for example /usr/src/linux/COPYING); if not, write to the Free
  18    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  19 */
  20
  21 #include <linux/blkdev.h>
  22 #include <linux/seq_file.h>
  23 #include "md.h"
  24 #include "raid0.h"
  25
  26 static void raid0_unplug(struct request_queue *q)
  27 {
  28         mddev_t *mddev = q->queuedata;
  29         raid0_conf_t *conf = mddev_to_conf(mddev);
  30         mdk_rdev_t **devlist = conf->strip_zone[0].dev;
  31         int i;
  32
  33         for (i=0; i<mddev->raid_disks; i++) {
  34                 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
  35
  36                 blk_unplug(r_queue);
  37         }
  38 }
  39
  40 static int raid0_congested(void *data, int bits)
  41 {
  42         mddev_t *mddev = data;
  43         raid0_conf_t *conf = mddev_to_conf(mddev);
  44         mdk_rdev_t **devlist = conf->strip_zone[0].dev;
  45         int i, ret = 0;
  46
  47         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
  48                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  49
  50                 ret |= bdi_congested(&q->backing_dev_info, bits);
  51         }
  52         return ret;
  53 }
  54
  55 static int create_strip_zones (mddev_t *mddev)
  56 {
  57         int i, c, j;
  58         sector_t curr_zone_end;
  59         sector_t min_spacing;
  60         raid0_conf_t *conf = mddev_to_conf(mddev);
  61         mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
  62         struct strip_zone *zone;
  63         int cnt;
  64         char b[BDEVNAME_SIZE];
  65
  66         /*
  67          * The number of 'same size groups'
  68          */
  69         conf->nr_strip_zones = 0;
  70
  71         list_for_each_entry(rdev1, &mddev->disks, same_set) {
  72                 printk(KERN_INFO "raid0: looking at %s\n",
  73                         bdevname(rdev1->bdev,b));
  74                 c = 0;
  75                 list_for_each_entry(rdev2, &mddev->disks, same_set) {
  76                         printk(KERN_INFO "raid0:   comparing %s(%llu)",
  77                                bdevname(rdev1->bdev,b),
  78                                (unsigned long long)rdev1->sectors);
  79                         printk(KERN_INFO " with %s(%llu)\n",
  80                                bdevname(rdev2->bdev,b),
  81                                (unsigned long long)rdev2->sectors);
  82                         if (rdev2 == rdev1) {
  83                                 printk(KERN_INFO "raid0:   END\n");
  84                                 break;
  85                         }
  86                         if (rdev2->sectors == rdev1->sectors) {
  87                                 /*
  88                                  * Not unique, don't count it as a new
  89                                  * group
  90                                  */
  91                                 printk(KERN_INFO "raid0:   EQUAL\n");
  92                                 c = 1;
  93                                 break;
  94                         }
  95                         printk(KERN_INFO "raid0:   NOT EQUAL\n");
  96                 }
  97                 if (!c) {
  98                         printk(KERN_INFO "raid0:   ==> UNIQUE\n");
  99                         conf->nr_strip_zones++;
 100                         printk(KERN_INFO "raid0: %d zones\n",
 101                                 conf->nr_strip_zones);
 102                 }
 103         }
 104         printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
 105
 106         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 107                                 conf->nr_strip_zones, GFP_KERNEL);
 108         if (!conf->strip_zone)
 109                 return 1;
 110         conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
 111                                 conf->nr_strip_zones*mddev->raid_disks,
 112                                 GFP_KERNEL);
 113         if (!conf->devlist)
 114                 return 1;
 115
 116         /* The first zone must contain all devices, so here we check that
 117          * there is a proper alignment of slots to devices and find them all
 118          */
 119         zone = &conf->strip_zone[0];
 120         cnt = 0;
 121         smallest = NULL;
 122         zone->dev = conf->devlist;
 123         list_for_each_entry(rdev1, &mddev->disks, same_set) {
 124                 int j = rdev1->raid_disk;
 125
 126                 if (j < 0 || j >= mddev->raid_disks) {
 127                         printk(KERN_ERR "raid0: bad disk number %d - "
 128                                 "aborting!\n", j);
 129                         goto abort;
 130                 }
 131                 if (zone->dev[j]) {
 132                         printk(KERN_ERR "raid0: multiple devices for %d - "
 133                                 "aborting!\n", j);
 134                         goto abort;
 135                 }
 136                 zone->dev[j] = rdev1;
 137
 138                 blk_queue_stack_limits(mddev->queue,
 139                                        rdev1->bdev->bd_disk->queue);
 140                 /* as we don't honour merge_bvec_fn, we must never risk
 141                  * violating it, so limit ->max_sector to one PAGE, as
 142                  * a one page request is never in violation.
 143                  */
 144
 145                 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
 146                     queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
 147                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 148
 149                 if (!smallest || (rdev1->sectors < smallest->sectors))
 150                         smallest = rdev1;
 151                 cnt++;
 152         }
 153         if (cnt != mddev->raid_disks) {
 154                 printk(KERN_ERR "raid0: too few disks (%d of %d) - "
 155                         "aborting!\n", cnt, mddev->raid_disks);
 156                 goto abort;
 157         }
 158         zone->nb_dev = cnt;
 159         zone->sectors = smallest->sectors * cnt;
 160         zone->zone_end = zone->sectors;
 161
 162         curr_zone_end = zone->sectors;
 163
 164         /* now do the other zones */
 165         for (i = 1; i < conf->nr_strip_zones; i++)
 166         {
 167                 zone = conf->strip_zone + i;
 168                 zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
 169
 170                 printk(KERN_INFO "raid0: zone %d\n", i);
 171                 zone->dev_start = smallest->sectors;
 172                 smallest = NULL;
 173                 c = 0;
 174
 175                 for (j=0; j<cnt; j++) {
 176                         char b[BDEVNAME_SIZE];
 177                         rdev = conf->strip_zone[0].dev[j];
 178                         printk(KERN_INFO "raid0: checking %s ...",
 179                                 bdevname(rdev->bdev, b));
 180                         if (rdev->sectors <= zone->dev_start) {
 181                                 printk(KERN_INFO " nope.\n");
 182                                 continue;
 183                         }
 184                         printk(KERN_INFO " contained as device %d\n", c);
 185                         zone->dev[c] = rdev;
 186                         c++;
 187                         if (!smallest || rdev->sectors < smallest->sectors) {
 188                                 smallest = rdev;
 189                                 printk(KERN_INFO "  (%llu) is smallest!.\n",
 190                                         (unsigned long long)rdev->sectors);
 191                         }
 192                 }
 193
 194                 zone->nb_dev = c;
 195                 zone->sectors = (smallest->sectors - zone->dev_start) * c;
 196                 printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
 197                         zone->nb_dev, (unsigned long long)zone->sectors);
 198
 199                 curr_zone_end += zone->sectors;
 200                 zone->zone_end = curr_zone_end;
 201
 202                 printk(KERN_INFO "raid0: current zone start: %llu\n",
 203                         (unsigned long long)smallest->sectors);
 204         }
 205         /* Now find appropriate hash spacing.
 206          * We want a number which causes most hash entries to cover
 207          * at most two strips, but the hash table must be at most
 208          * 1 PAGE.  We choose the smallest strip, or contiguous collection
 209          * of strips, that has big enough size.  We never consider the last
 210          * strip though as it's size has no bearing on the efficacy of the hash
 211          * table.
 212          */
 213         conf->spacing = curr_zone_end;
 214         min_spacing = curr_zone_end;
 215         sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
 216         for (i=0; i < conf->nr_strip_zones-1; i++) {
 217                 sector_t s = 0;
 218                 for (j = i; j < conf->nr_strip_zones - 1 &&
 219                                 s < min_spacing; j++)
 220                         s += conf->strip_zone[j].sectors;
 221                 if (s >= min_spacing && s < conf->spacing)
 222                         conf->spacing = s;
 223         }
 224
 225         mddev->queue->unplug_fn = raid0_unplug;
 226
 227         mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 228         mddev->queue->backing_dev_info.congested_data = mddev;
 229
 230         printk(KERN_INFO "raid0: done.\n");
 231         return 0;
 232  abort:
 233         return 1;
 234 }
 235
 236 /**
 237  *      raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 238  *      @q: request queue
 239  *      @bvm: properties of new bio
 240  *      @biovec: the request that could be merged to it.
 241  *
 242  *      Return amount of bytes we can accept at this offset
 243  */
 244 static int raid0_mergeable_bvec(struct request_queue *q,
 245                                 struct bvec_merge_data *bvm,
 246                                 struct bio_vec *biovec)
 247 {
 248         mddev_t *mddev = q->queuedata;
 249         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 250         int max;
 251         unsigned int chunk_sectors = mddev->chunk_size >> 9;
 252         unsigned int bio_sectors = bvm->bi_size >> 9;
 253
 254         max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
 255         if (max < 0) max = 0; /* bio_add cannot handle a negative return */
 256         if (max <= biovec->bv_len && bio_sectors == 0)
 257                 return biovec->bv_len;
 258         else
 259                 return max;
 260 }
 261
 262 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 263 {
 264         sector_t array_sectors = 0;
 265         mdk_rdev_t *rdev;
 266
 267         WARN_ONCE(sectors || raid_disks,
 268                   "%s does not support generic reshape\n", __func__);
 269
 270         list_for_each_entry(rdev, &mddev->disks, same_set)
 271                 array_sectors += rdev->sectors;
 272
 273         return array_sectors;
 274 }
 275
 276 static int raid0_run (mddev_t *mddev)
 277 {
 278         unsigned  cur=0, i=0, nb_zone;
 279         s64 sectors;
 280         raid0_conf_t *conf;
 281
 282         if (mddev->chunk_size == 0) {
 283                 printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
 284                 return -EINVAL;
 285         }
 286         printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
 287                mdname(mddev),
 288                mddev->chunk_size >> 9,
 289                (mddev->chunk_size>>1)-1);
 290         blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
 291         blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
 292         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 293
 294         conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
 295         if (!conf)
 296                 goto out;
 297         mddev->private = (void *)conf;
 298
 299         conf->strip_zone = NULL;
 300         conf->devlist = NULL;
 301         if (create_strip_zones (mddev))
 302                 goto out_free_conf;
 303
 304         /* calculate array device size */
 305         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 306
 307         printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
 308                 (unsigned long long)mddev->array_sectors);
 309         printk(KERN_INFO "raid0 : conf->spacing is %llu sectors.\n",
 310                 (unsigned long long)conf->spacing);
 311         {
 312                 sector_t s = raid0_size(mddev, 0, 0);
 313                 sector_t space = conf->spacing;
 314                 int round;
 315                 conf->sector_shift = 0;
 316                 if (sizeof(sector_t) > sizeof(u32)) {
 317                         /*shift down space and s so that sector_div will work */
 318                         while (space > (sector_t) (~(u32)0)) {
 319                                 s >>= 1;
 320                                 space >>= 1;
 321                                 s += 1; /* force round-up */
 322                                 conf->sector_shift++;
 323                         }
 324                 }
 325                 round = sector_div(s, (u32)space) ? 1 : 0;
 326                 nb_zone = s + round;
 327         }
 328         printk(KERN_INFO "raid0 : nb_zone is %d.\n", nb_zone);
 329         sectors = conf->strip_zone[cur].sectors;
 330
 331         for (i=1; i< nb_zone; i++) {
 332                 while (sectors <= conf->spacing) {
 333                         cur++;
 334                         sectors += conf->strip_zone[cur].sectors;
 335                 }
 336                 sectors -= conf->spacing;
 337         }
 338         if (conf->sector_shift) {
 339                 conf->spacing >>= conf->sector_shift;
 340                 /* round spacing up so when we divide by it, we
 341                  * err on the side of too-low, which is safest
 342                  */
 343                 conf->spacing++;
 344         }
 345
 346         /* calculate the max read-ahead size.
 347          * For read-ahead of large files to be effective, we need to
 348          * readahead at least twice a whole stripe. i.e. number of devices
 349          * multiplied by chunk size times 2.
 350          * If an individual device has an ra_pages greater than the
 351          * chunk size, then we will not drive that device as hard as it
 352          * wants.  We consider this a configuration error: a larger
 353          * chunksize should be used in that case.
 354          */
 355         {
 356                 int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
 357                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 358                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 359         }
 360
 361
 362         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 363         return 0;
 364
 365 out_free_conf:
 366         kfree(conf->strip_zone);
 367         kfree(conf->devlist);
 368         kfree(conf);
 369         mddev->private = NULL;
 370 out:
 371         return -ENOMEM;
 372 }
 373
 374 static int raid0_stop (mddev_t *mddev)
 375 {
 376         raid0_conf_t *conf = mddev_to_conf(mddev);
 377
 378         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 379         kfree(conf->strip_zone);
 380         conf->strip_zone = NULL;
 381         kfree(conf);
 382         mddev->private = NULL;
 383
 384         return 0;
 385 }
 386
 387 /* Find the zone which holds a particular offset */
 388 static struct strip_zone *find_zone(struct raid0_private_data *conf,
 389                 sector_t sector)
 390 {
 391         int i;
 392         struct strip_zone *z = conf->strip_zone;
 393
 394         for (i = 0; i < conf->nr_strip_zones; i++)
 395                 if (sector < z[i].zone_end)
 396                         return z + i;
 397         BUG();
 398 }
 399
 400 static int raid0_make_request (struct request_queue *q, struct bio *bio)
 401 {
 402         mddev_t *mddev = q->queuedata;
 403         unsigned int sect_in_chunk, chunksect_bits, chunk_sects;
 404         raid0_conf_t *conf = mddev_to_conf(mddev);
 405         struct strip_zone *zone;
 406         mdk_rdev_t *tmp_dev;
 407         sector_t chunk;
 408         sector_t sector, rsect;
 409         const int rw = bio_data_dir(bio);
 410         int cpu;
 411
 412         if (unlikely(bio_barrier(bio))) {
 413                 bio_endio(bio, -EOPNOTSUPP);
 414                 return 0;
 415         }
 416
 417         cpu = part_stat_lock();
 418         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 419         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 420                       bio_sectors(bio));
 421         part_stat_unlock();
 422
 423         chunk_sects = mddev->chunk_size >> 9;
 424         chunksect_bits = ffz(~chunk_sects);
 425         sector = bio->bi_sector;
 426
 427         if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
 428                 struct bio_pair *bp;
 429                 /* Sanity check -- queue functions should prevent this happening */
 430                 if (bio->bi_vcnt != 1 ||
 431                     bio->bi_idx != 0)
 432                         goto bad_map;
 433                 /* This is a one page bio that upper layers
 434                  * refuse to split for us, so we need to split it.
 435                  */
 436                 bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
 437                 if (raid0_make_request(q, &bp->bio1))
 438                         generic_make_request(&bp->bio1);
 439                 if (raid0_make_request(q, &bp->bio2))
 440                         generic_make_request(&bp->bio2);
 441
 442                 bio_pair_release(bp);
 443                 return 0;
 444         }
 445         zone = find_zone(conf, sector);
 446         sect_in_chunk = bio->bi_sector & (chunk_sects - 1);
 447         {
 448                 sector_t x = (zone->sectors + sector - zone->zone_end)
 449                                 >> chunksect_bits;
 450
 451                 sector_div(x, zone->nb_dev);
 452                 chunk = x;
 453
 454                 x = sector >> chunksect_bits;
 455                 tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
 456         }
 457         rsect = (chunk << chunksect_bits) + zone->dev_start + sect_in_chunk;
 458
 459         bio->bi_bdev = tmp_dev->bdev;
 460         bio->bi_sector = rsect + tmp_dev->data_offset;
 461
 462         /*
 463          * Let the main block layer submit the IO and resolve recursion:
 464          */
 465         return 1;
 466
 467 bad_map:
 468         printk("raid0_make_request bug: can't convert block across chunks"
 469                 " or bigger than %dk %llu %d\n", chunk_sects / 2,
 470                 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 471
 472         bio_io_error(bio);
 473         return 0;
 474 }
 475
 476 static void raid0_status (struct seq_file *seq, mddev_t *mddev)
 477 {
 478 #undef MD_DEBUG
 479 #ifdef MD_DEBUG
 480         int j, k, h;
 481         char b[BDEVNAME_SIZE];
 482         raid0_conf_t *conf = mddev_to_conf(mddev);
 483
 484         h = 0;
 485         for (j = 0; j < conf->nr_strip_zones; j++) {
 486                 seq_printf(seq, "      z%d", j);
 487                 seq_printf(seq, "=[");
 488                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 489                         seq_printf(seq, "%s/", bdevname(
 490                                 conf->strip_zone[j].dev[k]->bdev,b));
 491
 492                 seq_printf(seq, "] ze=%d ds=%d s=%d\n",
 493                                 conf->strip_zone[j].zone_end,
 494                                 conf->strip_zone[j].dev_start,
 495                                 conf->strip_zone[j].sectors);
 496         }
 497 #endif
 498         seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
 499         return;
 500 }
 501
 502 static struct mdk_personality raid0_personality=
 503 {
 504         .name           = "raid0",
 505         .level          = 0,
 506         .owner          = THIS_MODULE,
 507         .make_request   = raid0_make_request,
 508         .run            = raid0_run,
 509         .stop           = raid0_stop,
 510         .status         = raid0_status,
 511         .size           = raid0_size,
 512 };
 513
 514 static int __init raid0_init (void)
 515 {
 516         return register_md_personality (&raid0_personality);
 517 }
 518
 519 static void raid0_exit (void)
 520 {
 521         unregister_md_personality (&raid0_personality);
 522 }
 523
 524 module_init(raid0_init);
 525 module_exit(raid0_exit);
 526 MODULE_LICENSE("GPL");
 527 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 528 MODULE_ALIAS("md-raid0");
 529 MODULE_ALIAS("md-level-0");