SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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;
  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->sectors = smallest->sectors * cnt;
 157         zone->zone_end = zone->sectors;
 158
 159         curr_zone_end = zone->sectors;
 160
 161         /* now do the other zones */
 162         for (i = 1; i < conf->nr_strip_zones; i++)
 163         {
 164                 zone = conf->strip_zone + i;
 165                 zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
 166
 167                 printk(KERN_INFO "raid0: zone %d\n", i);
 168                 zone->dev_start = smallest->sectors;
 169                 smallest = NULL;
 170                 c = 0;
 171
 172                 for (j=0; j<cnt; j++) {
 173                         char b[BDEVNAME_SIZE];
 174                         rdev = conf->strip_zone[0].dev[j];
 175                         printk(KERN_INFO "raid0: checking %s ...",
 176                                 bdevname(rdev->bdev, b));
 177                         if (rdev->sectors <= zone->dev_start) {
 178                                 printk(KERN_INFO " nope.\n");
 179                                 continue;
 180                         }
 181                         printk(KERN_INFO " contained as device %d\n", c);
 182                         zone->dev[c] = rdev;
 183                         c++;
 184                         if (!smallest || rdev->sectors < smallest->sectors) {
 185                                 smallest = rdev;
 186                                 printk(KERN_INFO "  (%llu) is smallest!.\n",
 187                                         (unsigned long long)rdev->sectors);
 188                         }
 189                 }
 190
 191                 zone->nb_dev = c;
 192                 zone->sectors = (smallest->sectors - zone->dev_start) * c;
 193                 printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
 194                         zone->nb_dev, (unsigned long long)zone->sectors);
 195
 196                 curr_zone_end += zone->sectors;
 197                 zone->zone_end = curr_zone_end;
 198
 199                 printk(KERN_INFO "raid0: current zone start: %llu\n",
 200                         (unsigned long long)smallest->sectors);
 201         }
 202         mddev->queue->unplug_fn = raid0_unplug;
 203         mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 204         mddev->queue->backing_dev_info.congested_data = mddev;
 205
 206         printk(KERN_INFO "raid0: done.\n");
 207         mddev->private = conf;
 208         return 0;
 209 abort:
 210         kfree(conf->strip_zone);
 211         kfree(conf->devlist);
 212         kfree(conf);
 213         mddev->private = NULL;
 214         return err;
 215 }
 216
 217 /**
 218  *      raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 219  *      @q: request queue
 220  *      @bvm: properties of new bio
 221  *      @biovec: the request that could be merged to it.
 222  *
 223  *      Return amount of bytes we can accept at this offset
 224  */
 225 static int raid0_mergeable_bvec(struct request_queue *q,
 226                                 struct bvec_merge_data *bvm,
 227                                 struct bio_vec *biovec)
 228 {
 229         mddev_t *mddev = q->queuedata;
 230         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 231         int max;
 232         unsigned int chunk_sectors = mddev->chunk_size >> 9;
 233         unsigned int bio_sectors = bvm->bi_size >> 9;
 234
 235         max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
 236         if (max < 0) max = 0; /* bio_add cannot handle a negative return */
 237         if (max <= biovec->bv_len && bio_sectors == 0)
 238                 return biovec->bv_len;
 239         else
 240                 return max;
 241 }
 242
 243 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 244 {
 245         sector_t array_sectors = 0;
 246         mdk_rdev_t *rdev;
 247
 248         WARN_ONCE(sectors || raid_disks,
 249                   "%s does not support generic reshape\n", __func__);
 250
 251         list_for_each_entry(rdev, &mddev->disks, same_set)
 252                 array_sectors += rdev->sectors;
 253
 254         return array_sectors;
 255 }
 256
 257 static int raid0_run(mddev_t *mddev)
 258 {
 259         int ret;
 260
 261         if (mddev->chunk_size == 0) {
 262                 printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
 263                 return -EINVAL;
 264         }
 265         printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
 266                mdname(mddev),
 267                mddev->chunk_size >> 9,
 268                (mddev->chunk_size>>1)-1);
 269         blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
 270         blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
 271         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 272
 273         ret = create_strip_zones(mddev);
 274         if (ret < 0)
 275                 return ret;
 276
 277         /* calculate array device size */
 278         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 279
 280         printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
 281                 (unsigned long long)mddev->array_sectors);
 282         /* calculate the max read-ahead size.
 283          * For read-ahead of large files to be effective, we need to
 284          * readahead at least twice a whole stripe. i.e. number of devices
 285          * multiplied by chunk size times 2.
 286          * If an individual device has an ra_pages greater than the
 287          * chunk size, then we will not drive that device as hard as it
 288          * wants.  We consider this a configuration error: a larger
 289          * chunksize should be used in that case.
 290          */
 291         {
 292                 int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
 293                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 294                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 295         }
 296
 297         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 298         return 0;
 299 }
 300
 301 static int raid0_stop (mddev_t *mddev)
 302 {
 303         raid0_conf_t *conf = mddev_to_conf(mddev);
 304
 305         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 306         kfree(conf->strip_zone);
 307         conf->strip_zone = NULL;
 308         kfree(conf);
 309         mddev->private = NULL;
 310
 311         return 0;
 312 }
 313
 314 /* Find the zone which holds a particular offset */
 315 static struct strip_zone *find_zone(struct raid0_private_data *conf,
 316                 sector_t sector)
 317 {
 318         int i;
 319         struct strip_zone *z = conf->strip_zone;
 320
 321         for (i = 0; i < conf->nr_strip_zones; i++)
 322                 if (sector < z[i].zone_end)
 323                         return z + i;
 324         BUG();
 325 }
 326
 327 static int raid0_make_request (struct request_queue *q, struct bio *bio)
 328 {
 329         mddev_t *mddev = q->queuedata;
 330         unsigned int sect_in_chunk, chunksect_bits, chunk_sects;
 331         raid0_conf_t *conf = mddev_to_conf(mddev);
 332         struct strip_zone *zone;
 333         mdk_rdev_t *tmp_dev;
 334         sector_t chunk;
 335         sector_t sector, rsect;
 336         const int rw = bio_data_dir(bio);
 337         int cpu;
 338
 339         if (unlikely(bio_barrier(bio))) {
 340                 bio_endio(bio, -EOPNOTSUPP);
 341                 return 0;
 342         }
 343
 344         cpu = part_stat_lock();
 345         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 346         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 347                       bio_sectors(bio));
 348         part_stat_unlock();
 349
 350         chunk_sects = mddev->chunk_size >> 9;
 351         chunksect_bits = ffz(~chunk_sects);
 352         sector = bio->bi_sector;
 353
 354         if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
 355                 struct bio_pair *bp;
 356                 /* Sanity check -- queue functions should prevent this happening */
 357                 if (bio->bi_vcnt != 1 ||
 358                     bio->bi_idx != 0)
 359                         goto bad_map;
 360                 /* This is a one page bio that upper layers
 361                  * refuse to split for us, so we need to split it.
 362                  */
 363                 bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
 364                 if (raid0_make_request(q, &bp->bio1))
 365                         generic_make_request(&bp->bio1);
 366                 if (raid0_make_request(q, &bp->bio2))
 367                         generic_make_request(&bp->bio2);
 368
 369                 bio_pair_release(bp);
 370                 return 0;
 371         }
 372         zone = find_zone(conf, sector);
 373         sect_in_chunk = bio->bi_sector & (chunk_sects - 1);
 374         {
 375                 sector_t x = (zone->sectors + sector - zone->zone_end)
 376                                 >> chunksect_bits;
 377
 378                 sector_div(x, zone->nb_dev);
 379                 chunk = x;
 380
 381                 x = sector >> chunksect_bits;
 382                 tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
 383         }
 384         rsect = (chunk << chunksect_bits) + zone->dev_start + sect_in_chunk;
 385
 386         bio->bi_bdev = tmp_dev->bdev;
 387         bio->bi_sector = rsect + tmp_dev->data_offset;
 388
 389         /*
 390          * Let the main block layer submit the IO and resolve recursion:
 391          */
 392         return 1;
 393
 394 bad_map:
 395         printk("raid0_make_request bug: can't convert block across chunks"
 396                 " or bigger than %dk %llu %d\n", chunk_sects / 2,
 397                 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 398
 399         bio_io_error(bio);
 400         return 0;
 401 }
 402
 403 static void raid0_status (struct seq_file *seq, mddev_t *mddev)
 404 {
 405 #undef MD_DEBUG
 406 #ifdef MD_DEBUG
 407         int j, k, h;
 408         char b[BDEVNAME_SIZE];
 409         raid0_conf_t *conf = mddev_to_conf(mddev);
 410
 411         h = 0;
 412         for (j = 0; j < conf->nr_strip_zones; j++) {
 413                 seq_printf(seq, "      z%d", j);
 414                 seq_printf(seq, "=[");
 415                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 416                         seq_printf(seq, "%s/", bdevname(
 417                                 conf->strip_zone[j].dev[k]->bdev,b));
 418
 419                 seq_printf(seq, "] ze=%d ds=%d s=%d\n",
 420                                 conf->strip_zone[j].zone_end,
 421                                 conf->strip_zone[j].dev_start,
 422                                 conf->strip_zone[j].sectors);
 423         }
 424 #endif
 425         seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
 426         return;
 427 }
 428
 429 static struct mdk_personality raid0_personality=
 430 {
 431         .name           = "raid0",
 432         .level          = 0,
 433         .owner          = THIS_MODULE,
 434         .make_request   = raid0_make_request,
 435         .run            = raid0_run,
 436         .stop           = raid0_stop,
 437         .status         = raid0_status,
 438         .size           = raid0_size,
 439 };
 440
 441 static int __init raid0_init (void)
 442 {
 443         return register_md_personality (&raid0_personality);
 444 }
 445
 446 static void raid0_exit (void)
 447 {
 448         unregister_md_personality (&raid0_personality);
 449 }
 450
 451 module_init(raid0_init);
 452 module_exit(raid0_exit);
 453 MODULE_LICENSE("GPL");
 454 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 455 MODULE_ALIAS("md-raid0");
 456 MODULE_ALIAS("md-level-0");