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