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/blkdev.h>
  20 #include <linux/raid/md_u.h>
  21 #include <linux/seq_file.h>
  22 #include "md.h"
  23 #include "linear.h"
  24
  25 /*
  26  * find which device holds a particular offset
  27  */
  28 static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
  29 {
  30         dev_info_t *hash;
  31         linear_conf_t *conf = mddev_to_conf(mddev);
  32         sector_t idx = sector >> conf->sector_shift;
  33
  34         /*
  35          * sector_div(a,b) returns the remainer and sets a to a/b
  36          */
  37         (void)sector_div(idx, conf->spacing);
  38         hash = conf->hash_table[idx];
  39
  40         while (sector >= hash->num_sectors + hash->start_sector)
  41                 hash++;
  42         return hash;
  43 }
  44
  45 /**
  46  *      linear_mergeable_bvec -- tell bio layer if two requests can be merged
  47  *      @q: request queue
  48  *      @bvm: properties of new bio
  49  *      @biovec: the request that could be merged to it.
  50  *
  51  *      Return amount of bytes we can take at this offset
  52  */
  53 static int linear_mergeable_bvec(struct request_queue *q,
  54                                  struct bvec_merge_data *bvm,
  55                                  struct bio_vec *biovec)
  56 {
  57         mddev_t *mddev = q->queuedata;
  58         dev_info_t *dev0;
  59         unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
  60         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
  61
  62         dev0 = which_dev(mddev, sector);
  63         maxsectors = dev0->num_sectors - (sector - dev0->start_sector);
  64
  65         if (maxsectors < bio_sectors)
  66                 maxsectors = 0;
  67         else
  68                 maxsectors -= bio_sectors;
  69
  70         if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
  71                 return biovec->bv_len;
  72         /* The bytes available at this offset could be really big,
  73          * so we cap at 2^31 to avoid overflow */
  74         if (maxsectors > (1 << (31-9)))
  75                 return 1<<31;
  76         return maxsectors << 9;
  77 }
  78
  79 static void linear_unplug(struct request_queue *q)
  80 {
  81         mddev_t *mddev = q->queuedata;
  82         linear_conf_t *conf = mddev_to_conf(mddev);
  83         int i;
  84
  85         for (i=0; i < mddev->raid_disks; i++) {
  86                 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
  87                 blk_unplug(r_queue);
  88         }
  89 }
  90
  91 static int linear_congested(void *data, int bits)
  92 {
  93         mddev_t *mddev = data;
  94         linear_conf_t *conf = mddev_to_conf(mddev);
  95         int i, ret = 0;
  96
  97         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
  98                 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
  99                 ret |= bdi_congested(&q->backing_dev_info, bits);
 100         }
 101         return ret;
 102 }
 103
 104 static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 105 {
 106         linear_conf_t *conf = mddev_to_conf(mddev);
 107
 108         WARN_ONCE(sectors || raid_disks,
 109                   "%s does not support generic reshape\n", __func__);
 110
 111         return conf->array_sectors;
 112 }
 113
 114 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
 115 {
 116         linear_conf_t *conf;
 117         dev_info_t **table;
 118         mdk_rdev_t *rdev;
 119         int i, nb_zone, cnt;
 120         sector_t min_sectors;
 121         sector_t curr_sector;
 122
 123         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
 124                         GFP_KERNEL);
 125         if (!conf)
 126                 return NULL;
 127
 128         cnt = 0;
 129         conf->array_sectors = 0;
 130
 131         list_for_each_entry(rdev, &mddev->disks, same_set) {
 132                 int j = rdev->raid_disk;
 133                 dev_info_t *disk = conf->disks + j;
 134
 135                 if (j < 0 || j >= raid_disks || disk->rdev) {
 136                         printk("linear: disk numbering problem. Aborting!\n");
 137                         goto out;
 138                 }
 139
 140                 disk->rdev = rdev;
 141
 142                 blk_queue_stack_limits(mddev->queue,
 143                                        rdev->bdev->bd_disk->queue);
 144                 /* as we don't honour merge_bvec_fn, we must never risk
 145                  * violating it, so limit ->max_sector to one PAGE, as
 146                  * a one page request is never in violation.
 147                  */
 148                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
 149                     queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
 150                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 151
 152                 disk->num_sectors = rdev->sectors;
 153                 conf->array_sectors += rdev->sectors;
 154
 155                 cnt++;
 156         }
 157         if (cnt != raid_disks) {
 158                 printk("linear: not enough drives present. Aborting!\n");
 159                 goto out;
 160         }
 161
 162         min_sectors = conf->array_sectors;
 163         sector_div(min_sectors, PAGE_SIZE/sizeof(struct dev_info *));
 164         if (min_sectors == 0)
 165                 min_sectors = 1;
 166
 167         /* min_sectors is the minimum spacing that will fit the hash
 168          * table in one PAGE.  This may be much smaller than needed.
 169          * We find the smallest non-terminal set of consecutive devices
 170          * that is larger than min_sectors and use the size of that as
 171          * the actual spacing
 172          */
 173         conf->spacing = conf->array_sectors;
 174         for (i=0; i < cnt-1 ; i++) {
 175                 sector_t tmp = 0;
 176                 int j;
 177                 for (j = i; j < cnt - 1 && tmp < min_sectors; j++)
 178                         tmp += conf->disks[j].num_sectors;
 179                 if (tmp >= min_sectors && tmp < conf->spacing)
 180                         conf->spacing = tmp;
 181         }
 182
 183         /* spacing may be too large for sector_div to work with,
 184          * so we might need to pre-shift
 185          */
 186         conf->sector_shift = 0;
 187         if (sizeof(sector_t) > sizeof(u32)) {
 188                 sector_t space = conf->spacing;
 189                 while (space > (sector_t)(~(u32)0)) {
 190                         space >>= 1;
 191                         conf->sector_shift++;
 192                 }
 193         }
 194         /*
 195          * This code was restructured to work around a gcc-2.95.3 internal
 196          * compiler error.  Alter it with care.
 197          */
 198         {
 199                 sector_t sz;
 200                 unsigned round;
 201                 unsigned long base;
 202
 203                 sz = conf->array_sectors >> conf->sector_shift;
 204                 sz += 1; /* force round-up */
 205                 base = conf->spacing >> conf->sector_shift;
 206                 round = sector_div(sz, base);
 207                 nb_zone = sz + (round ? 1 : 0);
 208         }
 209         BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
 210
 211         conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
 212                                         GFP_KERNEL);
 213         if (!conf->hash_table)
 214                 goto out;
 215
 216         /*
 217          * Here we generate the linear hash table
 218          * First calculate the device offsets.
 219          */
 220         conf->disks[0].start_sector = 0;
 221         for (i = 1; i < raid_disks; i++)
 222                 conf->disks[i].start_sector =
 223                         conf->disks[i-1].start_sector +
 224                         conf->disks[i-1].num_sectors;
 225
 226         table = conf->hash_table;
 227         i = 0;
 228         for (curr_sector = 0;
 229              curr_sector < conf->array_sectors;
 230              curr_sector += conf->spacing) {
 231
 232                 while (i < raid_disks-1 &&
 233                        curr_sector >= conf->disks[i+1].start_sector)
 234                         i++;
 235
 236                 *table ++ = conf->disks + i;
 237         }
 238
 239         if (conf->sector_shift) {
 240                 conf->spacing >>= conf->sector_shift;
 241                 /* round spacing up so that when we divide by it,
 242                  * we err on the side of "too-low", which is safest.
 243                  */
 244                 conf->spacing++;
 245         }
 246
 247         BUG_ON(table - conf->hash_table > nb_zone);
 248
 249         return conf;
 250
 251 out:
 252         kfree(conf);
 253         return NULL;
 254 }
 255
 256 static int linear_run (mddev_t *mddev)
 257 {
 258         linear_conf_t *conf;
 259
 260         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 261         conf = linear_conf(mddev, mddev->raid_disks);
 262
 263         if (!conf)
 264                 return 1;
 265         mddev->private = conf;
 266         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 267
 268         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 269         mddev->queue->unplug_fn = linear_unplug;
 270         mddev->queue->backing_dev_info.congested_fn = linear_congested;
 271         mddev->queue->backing_dev_info.congested_data = mddev;
 272         return 0;
 273 }
 274
 275 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
 276 {
 277         /* Adding a drive to a linear array allows the array to grow.
 278          * It is permitted if the new drive has a matching superblock
 279          * already on it, with raid_disk equal to raid_disks.
 280          * It is achieved by creating a new linear_private_data structure
 281          * and swapping it in in-place of the current one.
 282          * The current one is never freed until the array is stopped.
 283          * This avoids races.
 284          */
 285         linear_conf_t *newconf;
 286
 287         if (rdev->saved_raid_disk != mddev->raid_disks)
 288                 return -EINVAL;
 289
 290         rdev->raid_disk = rdev->saved_raid_disk;
 291
 292         newconf = linear_conf(mddev,mddev->raid_disks+1);
 293
 294         if (!newconf)
 295                 return -ENOMEM;
 296
 297         newconf->prev = mddev_to_conf(mddev);
 298         mddev->private = newconf;
 299         mddev->raid_disks++;
 300         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 301         set_capacity(mddev->gendisk, mddev->array_sectors);
 302         return 0;
 303 }
 304
 305 static int linear_stop (mddev_t *mddev)
 306 {
 307         linear_conf_t *conf = mddev_to_conf(mddev);
 308
 309         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 310         do {
 311                 linear_conf_t *t = conf->prev;
 312                 kfree(conf->hash_table);
 313                 kfree(conf);
 314                 conf = t;
 315         } while (conf);
 316
 317         return 0;
 318 }
 319
 320 static int linear_make_request (struct request_queue *q, struct bio *bio)
 321 {
 322         const int rw = bio_data_dir(bio);
 323         mddev_t *mddev = q->queuedata;
 324         dev_info_t *tmp_dev;
 325         int cpu;
 326
 327         if (unlikely(bio_barrier(bio))) {
 328                 bio_endio(bio, -EOPNOTSUPP);
 329                 return 0;
 330         }
 331
 332         cpu = part_stat_lock();
 333         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 334         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 335                       bio_sectors(bio));
 336         part_stat_unlock();
 337
 338         tmp_dev = which_dev(mddev, bio->bi_sector);
 339
 340         if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors +
 341                                         tmp_dev->start_sector)
 342                      || (bio->bi_sector <
 343                          tmp_dev->start_sector))) {
 344                 char b[BDEVNAME_SIZE];
 345
 346                 printk("linear_make_request: Sector %llu out of bounds on "
 347                         "dev %s: %llu sectors, offset %llu\n",
 348                         (unsigned long long)bio->bi_sector,
 349                         bdevname(tmp_dev->rdev->bdev, b),
 350                         (unsigned long long)tmp_dev->num_sectors,
 351                         (unsigned long long)tmp_dev->start_sector);
 352                 bio_io_error(bio);
 353                 return 0;
 354         }
 355         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
 356                      tmp_dev->start_sector + tmp_dev->num_sectors)) {
 357                 /* This bio crosses a device boundary, so we have to
 358                  * split it.
 359                  */
 360                 struct bio_pair *bp;
 361
 362                 bp = bio_split(bio,
 363                                tmp_dev->start_sector + tmp_dev->num_sectors
 364                                - bio->bi_sector);
 365
 366                 if (linear_make_request(q, &bp->bio1))
 367                         generic_make_request(&bp->bio1);
 368                 if (linear_make_request(q, &bp->bio2))
 369                         generic_make_request(&bp->bio2);
 370                 bio_pair_release(bp);
 371                 return 0;
 372         }
 373
 374         bio->bi_bdev = tmp_dev->rdev->bdev;
 375         bio->bi_sector = bio->bi_sector - tmp_dev->start_sector
 376                 + tmp_dev->rdev->data_offset;
 377
 378         return 1;
 379 }
 380
 381 static void linear_status (struct seq_file *seq, mddev_t *mddev)
 382 {
 383
 384         seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
 385 }
 386
 387
 388 static struct mdk_personality linear_personality =
 389 {
 390         .name           = "linear",
 391         .level          = LEVEL_LINEAR,
 392         .owner          = THIS_MODULE,
 393         .make_request   = linear_make_request,
 394         .run            = linear_run,
 395         .stop           = linear_stop,
 396         .status         = linear_status,
 397         .hot_add_disk   = linear_add,
 398         .size           = linear_size,
 399 };
 400
 401 static int __init linear_init (void)
 402 {
 403         return register_md_personality (&linear_personality);
 404 }
 405
 406 static void linear_exit (void)
 407 {
 408         unregister_md_personality (&linear_personality);
 409 }
 410
 411
 412 module_init(linear_init);
 413 module_exit(linear_exit);
 414 MODULE_LICENSE("GPL");
 415 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 416 MODULE_ALIAS("md-linear");
 417 MODULE_ALIAS("md-level--1");