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/module.h>
  20
  21 #include <linux/raid/md.h>
  22 #include <linux/slab.h>
  23 #include <linux/raid/linear.h>
  24
  25 #define MAJOR_NR MD_MAJOR
  26 #define MD_DRIVER
  27 #define MD_PERSONALITY
  28
  29 /*
  30  * find which device holds a particular offset
  31  */
  32 static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
  33 {
  34         struct linear_hash *hash;
  35         linear_conf_t *conf = mddev_to_conf(mddev);
  36         sector_t block = sector >> 1;
  37
  38         /*
  39          * sector_div(a,b) returns the remainer and sets a to a/b
  40          */
  41         (void)sector_div(block, conf->smallest->size);
  42         hash = conf->hash_table + block;
  43
  44         if ((sector>>1) >= (hash->dev0->size + hash->dev0->offset))
  45                 return hash->dev1;
  46         else
  47                 return hash->dev0;
  48 }
  49
  50 /**
  51  *      linear_mergeable_bvec -- tell bio layer if a two requests can be merged
  52  *      @q: request queue
  53  *      @bio: the buffer head that's been built up so far
  54  *      @biovec: the request that could be merged to it.
  55  *
  56  *      Return amount of bytes we can take at this offset
  57  */
  58 static int linear_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec)
  59 {
  60         mddev_t *mddev = q->queuedata;
  61         dev_info_t *dev0;
  62         unsigned long maxsectors, bio_sectors = bio->bi_size >> 9;
  63         sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
  64
  65         dev0 = which_dev(mddev, sector);
  66         maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1));
  67
  68         if (maxsectors < bio_sectors)
  69                 maxsectors = 0;
  70         else
  71                 maxsectors -= bio_sectors;
  72
  73         if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
  74                 return biovec->bv_len;
  75         /* The bytes available at this offset could be really big,
  76          * so we cap at 2^31 to avoid overflow */
  77         if (maxsectors > (1 << (31-9)))
  78                 return 1<<31;
  79         return maxsectors << 9;
  80 }
  81
  82 static void linear_unplug(request_queue_t *q)
  83 {
  84         mddev_t *mddev = q->queuedata;
  85         linear_conf_t *conf = mddev_to_conf(mddev);
  86         int i;
  87
  88         for (i=0; i < mddev->raid_disks; i++) {
  89                 request_queue_t *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
  90                 if (r_queue->unplug_fn)
  91                         r_queue->unplug_fn(r_queue);
  92         }
  93 }
  94
  95 static int linear_issue_flush(request_queue_t *q, struct gendisk *disk,
  96                               sector_t *error_sector)
  97 {
  98         mddev_t *mddev = q->queuedata;
  99         linear_conf_t *conf = mddev_to_conf(mddev);
 100         int i, ret = 0;
 101
 102         for (i=0; i < mddev->raid_disks; i++) {
 103                 struct block_device *bdev = conf->disks[i].rdev->bdev;
 104                 request_queue_t *r_queue = bdev_get_queue(bdev);
 105
 106                 if (!r_queue->issue_flush_fn) {
 107                         ret = -EOPNOTSUPP;
 108                         break;
 109                 }
 110                 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector);
 111                 if (ret)
 112                         break;
 113         }
 114         return ret;
 115 }
 116
 117 static int linear_run (mddev_t *mddev)
 118 {
 119         linear_conf_t *conf;
 120         struct linear_hash *table;
 121         mdk_rdev_t *rdev;
 122         int i, nb_zone, cnt;
 123         sector_t size;
 124         unsigned int curr_offset;
 125         struct list_head *tmp;
 126
 127         conf = kmalloc (sizeof (*conf) + mddev->raid_disks*sizeof(dev_info_t),
 128                         GFP_KERNEL);
 129         if (!conf)
 130                 goto out;
 131         memset(conf, 0, sizeof(*conf) + mddev->raid_disks*sizeof(dev_info_t));
 132         mddev->private = conf;
 133
 134         /*
 135          * Find the smallest device.
 136          */
 137
 138         conf->smallest = NULL;
 139         cnt = 0;
 140         mddev->array_size = 0;
 141
 142         ITERATE_RDEV(mddev,rdev,tmp) {
 143                 int j = rdev->raid_disk;
 144                 dev_info_t *disk = conf->disks + j;
 145
 146                 if (j < 0 || j > mddev->raid_disks || disk->rdev) {
 147                         printk("linear: disk numbering problem. Aborting!\n");
 148                         goto out;
 149                 }
 150
 151                 disk->rdev = rdev;
 152
 153                 blk_queue_stack_limits(mddev->queue,
 154                                        rdev->bdev->bd_disk->queue);
 155                 /* as we don't honour merge_bvec_fn, we must never risk
 156                  * violating it, so limit ->max_sector to one PAGE, as
 157                  * a one page request is never in violation.
 158                  */
 159                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
 160                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
 161                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 162
 163                 disk->size = rdev->size;
 164                 mddev->array_size += rdev->size;
 165
 166                 if (!conf->smallest || (disk->size < conf->smallest->size))
 167                         conf->smallest = disk;
 168                 cnt++;
 169         }
 170         if (cnt != mddev->raid_disks) {
 171                 printk("linear: not enough drives present. Aborting!\n");
 172                 goto out;
 173         }
 174
 175         /*
 176          * This code was restructured to work around a gcc-2.95.3 internal
 177          * compiler error.  Alter it with care.
 178          */
 179         {
 180                 sector_t sz;
 181                 unsigned round;
 182                 unsigned long base;
 183
 184                 sz = mddev->array_size;
 185                 base = conf->smallest->size;
 186                 round = sector_div(sz, base);
 187                 nb_zone = conf->nr_zones = sz + (round ? 1 : 0);
 188         }
 189
 190         conf->hash_table = kmalloc (sizeof (struct linear_hash) * nb_zone,
 191                                         GFP_KERNEL);
 192         if (!conf->hash_table)
 193                 goto out;
 194
 195         /*
 196          * Here we generate the linear hash table
 197          */
 198         table = conf->hash_table;
 199         size = 0;
 200         curr_offset = 0;
 201         for (i = 0; i < cnt; i++) {
 202                 dev_info_t *disk = conf->disks + i;
 203
 204                 disk->offset = curr_offset;
 205                 curr_offset += disk->size;
 206
 207                 if (size < 0) {
 208                         table[-1].dev1 = disk;
 209                 }
 210                 size += disk->size;
 211
 212                 while (size>0) {
 213                         table->dev0 = disk;
 214                         table->dev1 = NULL;
 215                         size -= conf->smallest->size;
 216                         table++;
 217                 }
 218         }
 219         if (table-conf->hash_table != nb_zone)
 220                 BUG();
 221
 222         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 223         mddev->queue->unplug_fn = linear_unplug;
 224         mddev->queue->issue_flush_fn = linear_issue_flush;
 225         return 0;
 226
 227 out:
 228         if (conf)
 229                 kfree(conf);
 230         return 1;
 231 }
 232
 233 static int linear_stop (mddev_t *mddev)
 234 {
 235         linear_conf_t *conf = mddev_to_conf(mddev);
 236
 237         kfree(conf->hash_table);
 238         kfree(conf);
 239
 240         return 0;
 241 }
 242
 243 static int linear_make_request (request_queue_t *q, struct bio *bio)
 244 {
 245         mddev_t *mddev = q->queuedata;
 246         dev_info_t *tmp_dev;
 247         sector_t block;
 248
 249         if (bio_data_dir(bio)==WRITE) {
 250                 disk_stat_inc(mddev->gendisk, writes);
 251                 disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio));
 252         } else {
 253                 disk_stat_inc(mddev->gendisk, reads);
 254                 disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio));
 255         }
 256
 257         tmp_dev = which_dev(mddev, bio->bi_sector);
 258         block = bio->bi_sector >> 1;
 259
 260         if (unlikely(!tmp_dev)) {
 261                 printk("linear_make_request: hash->dev1==NULL for block %llu\n",
 262                         (unsigned long long)block);
 263                 bio_io_error(bio, bio->bi_size);
 264                 return 0;
 265         }
 266
 267         if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
 268                      || block < tmp_dev->offset)) {
 269                 char b[BDEVNAME_SIZE];
 270
 271                 printk("linear_make_request: Block %llu out of bounds on "
 272                         "dev %s size %llu offset %llu\n",
 273                         (unsigned long long)block,
 274                         bdevname(tmp_dev->rdev->bdev, b),
 275                         (unsigned long long)tmp_dev->size,
 276                         (unsigned long long)tmp_dev->offset);
 277                 bio_io_error(bio, bio->bi_size);
 278                 return 0;
 279         }
 280         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
 281                      (tmp_dev->offset + tmp_dev->size)<<1)) {
 282                 /* This bio crosses a device boundary, so we have to
 283                  * split it.
 284                  */
 285                 struct bio_pair *bp;
 286                 bp = bio_split(bio, bio_split_pool,
 287                                (bio->bi_sector + (bio->bi_size >> 9) -
 288                                 (tmp_dev->offset + tmp_dev->size))<<1);
 289                 if (linear_make_request(q, &bp->bio1))
 290                         generic_make_request(&bp->bio1);
 291                 if (linear_make_request(q, &bp->bio2))
 292                         generic_make_request(&bp->bio2);
 293                 bio_pair_release(bp);
 294                 return 0;
 295         }
 296
 297         bio->bi_bdev = tmp_dev->rdev->bdev;
 298         bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
 299
 300         return 1;
 301 }
 302
 303 static void linear_status (struct seq_file *seq, mddev_t *mddev)
 304 {
 305
 306 #undef MD_DEBUG
 307 #ifdef MD_DEBUG
 308         int j;
 309         linear_conf_t *conf = mddev_to_conf(mddev);
 310
 311         seq_printf(seq, "      ");
 312         for (j = 0; j < conf->nr_zones; j++)
 313         {
 314                 char b[BDEVNAME_SIZE];
 315                 seq_printf(seq, "[%s",
 316                            bdevname(conf->hash_table[j].dev0->rdev->bdev,b));
 317
 318                 if (conf->hash_table[j].dev1)
 319                         seq_printf(seq, "/%s] ",
 320                                    bdevname(conf->hash_table[j].dev1->rdev->bdev,b));
 321                 else
 322                         seq_printf(seq, "] ");
 323         }
 324         seq_printf(seq, "\n");
 325 #endif
 326         seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
 327 }
 328
 329
 330 static mdk_personality_t linear_personality=
 331 {
 332         .name           = "linear",
 333         .owner          = THIS_MODULE,
 334         .make_request   = linear_make_request,
 335         .run            = linear_run,
 336         .stop           = linear_stop,
 337         .status         = linear_status,
 338 };
 339
 340 static int __init linear_init (void)
 341 {
 342         return register_md_personality (LINEAR, &linear_personality);
 343 }
 344
 345 static void linear_exit (void)
 346 {
 347         unregister_md_personality (LINEAR);
 348 }
 349
 350
 351 module_init(linear_init);
 352 module_exit(linear_exit);
 353 MODULE_LICENSE("GPL");
 354 MODULE_ALIAS("md-personality-1"); /* LINEAR */