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