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         int lo, mid, hi;
  31         linear_conf_t *conf = mddev->private;
  32
  33         lo = 0;
  34         hi = mddev->raid_disks - 1;
  35
  36         /*
  37          * Binary Search
  38          */
  39
  40         while (hi > lo) {
  41
  42                 mid = (hi + lo) / 2;
  43                 if (sector < conf->disks[mid].end_sector)
  44                         hi = mid;
  45                 else
  46                         lo = mid + 1;
  47         }
  48
  49         return conf->disks + lo;
  50 }
  51
  52 /**
  53  *      linear_mergeable_bvec -- tell bio layer if two requests can be merged
  54  *      @q: request queue
  55  *      @bvm: properties of new bio
  56  *      @biovec: the request that could be merged to it.
  57  *
  58  *      Return amount of bytes we can take at this offset
  59  */
  60 static int linear_mergeable_bvec(struct request_queue *q,
  61                                  struct bvec_merge_data *bvm,
  62                                  struct bio_vec *biovec)
  63 {
  64         mddev_t *mddev = q->queuedata;
  65         dev_info_t *dev0;
  66         unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
  67         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
  68
  69         dev0 = which_dev(mddev, sector);
  70         maxsectors = dev0->end_sector - sector;
  71
  72         if (maxsectors < bio_sectors)
  73                 maxsectors = 0;
  74         else
  75                 maxsectors -= bio_sectors;
  76
  77         if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
  78                 return biovec->bv_len;
  79         /* The bytes available at this offset could be really big,
  80          * so we cap at 2^31 to avoid overflow */
  81         if (maxsectors > (1 << (31-9)))
  82                 return 1<<31;
  83         return maxsectors << 9;
  84 }
  85
  86 static void linear_unplug(struct request_queue *q)
  87 {
  88         mddev_t *mddev = q->queuedata;
  89         linear_conf_t *conf = mddev->private;
  90         int i;
  91
  92         for (i=0; i < mddev->raid_disks; i++) {
  93                 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
  94                 blk_unplug(r_queue);
  95         }
  96 }
  97
  98 static int linear_congested(void *data, int bits)
  99 {
 100         mddev_t *mddev = data;
 101         linear_conf_t *conf = mddev->private;
 102         int i, ret = 0;
 103
 104         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
 105                 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
 106                 ret |= bdi_congested(&q->backing_dev_info, bits);
 107         }
 108         return ret;
 109 }
 110
 111 static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 112 {
 113         linear_conf_t *conf = mddev->private;
 114
 115         WARN_ONCE(sectors || raid_disks,
 116                   "%s does not support generic reshape\n", __func__);
 117
 118         return conf->array_sectors;
 119 }
 120
 121 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
 122 {
 123         linear_conf_t *conf;
 124         mdk_rdev_t *rdev;
 125         int i, cnt;
 126
 127         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
 128                         GFP_KERNEL);
 129         if (!conf)
 130                 return NULL;
 131
 132         cnt = 0;
 133         conf->array_sectors = 0;
 134
 135         list_for_each_entry(rdev, &mddev->disks, same_set) {
 136                 int j = rdev->raid_disk;
 137                 dev_info_t *disk = conf->disks + j;
 138
 139                 if (j < 0 || j >= raid_disks || disk->rdev) {
 140                         printk("linear: disk numbering problem. Aborting!\n");
 141                         goto out;
 142                 }
 143
 144                 disk->rdev = rdev;
 145
 146                 blk_queue_stack_limits(mddev->queue,
 147                                        rdev->bdev->bd_disk->queue);
 148                 /* as we don't honour merge_bvec_fn, we must never risk
 149                  * violating it, so limit ->max_sector to one PAGE, as
 150                  * a one page request is never in violation.
 151                  */
 152                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
 153                     queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
 154                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 155
 156                 conf->array_sectors += rdev->sectors;
 157                 cnt++;
 158
 159         }
 160         if (cnt != raid_disks) {
 161                 printk("linear: not enough drives present. Aborting!\n");
 162                 goto out;
 163         }
 164
 165         /*
 166          * Here we calculate the device offsets.
 167          */
 168         conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
 169
 170         for (i = 1; i < raid_disks; i++)
 171                 conf->disks[i].end_sector =
 172                         conf->disks[i-1].end_sector +
 173                         conf->disks[i].rdev->sectors;
 174
 175         return conf;
 176
 177 out:
 178         kfree(conf);
 179         return NULL;
 180 }
 181
 182 static int linear_run (mddev_t *mddev)
 183 {
 184         linear_conf_t *conf;
 185
 186         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 187         conf = linear_conf(mddev, mddev->raid_disks);
 188
 189         if (!conf)
 190                 return 1;
 191         mddev->private = conf;
 192         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 193
 194         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 195         mddev->queue->unplug_fn = linear_unplug;
 196         mddev->queue->backing_dev_info.congested_fn = linear_congested;
 197         mddev->queue->backing_dev_info.congested_data = mddev;
 198         return 0;
 199 }
 200
 201 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
 202 {
 203         /* Adding a drive to a linear array allows the array to grow.
 204          * It is permitted if the new drive has a matching superblock
 205          * already on it, with raid_disk equal to raid_disks.
 206          * It is achieved by creating a new linear_private_data structure
 207          * and swapping it in in-place of the current one.
 208          * The current one is never freed until the array is stopped.
 209          * This avoids races.
 210          */
 211         linear_conf_t *newconf;
 212
 213         if (rdev->saved_raid_disk != mddev->raid_disks)
 214                 return -EINVAL;
 215
 216         rdev->raid_disk = rdev->saved_raid_disk;
 217
 218         newconf = linear_conf(mddev,mddev->raid_disks+1);
 219
 220         if (!newconf)
 221                 return -ENOMEM;
 222
 223         newconf->prev = mddev->private;
 224         mddev->private = newconf;
 225         mddev->raid_disks++;
 226         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 227         set_capacity(mddev->gendisk, mddev->array_sectors);
 228         return 0;
 229 }
 230
 231 static int linear_stop (mddev_t *mddev)
 232 {
 233         linear_conf_t *conf = mddev->private;
 234
 235         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 236         do {
 237                 linear_conf_t *t = conf->prev;
 238                 kfree(conf);
 239                 conf = t;
 240         } while (conf);
 241
 242         return 0;
 243 }
 244
 245 static int linear_make_request (struct request_queue *q, struct bio *bio)
 246 {
 247         const int rw = bio_data_dir(bio);
 248         mddev_t *mddev = q->queuedata;
 249         dev_info_t *tmp_dev;
 250         sector_t start_sector;
 251         int cpu;
 252
 253         if (unlikely(bio_barrier(bio))) {
 254                 bio_endio(bio, -EOPNOTSUPP);
 255                 return 0;
 256         }
 257
 258         cpu = part_stat_lock();
 259         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 260         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 261                       bio_sectors(bio));
 262         part_stat_unlock();
 263
 264         tmp_dev = which_dev(mddev, bio->bi_sector);
 265         start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
 266
 267         if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
 268                      || (bio->bi_sector < start_sector))) {
 269                 char b[BDEVNAME_SIZE];
 270
 271                 printk("linear_make_request: Sector %llu out of bounds on "
 272                         "dev %s: %llu sectors, offset %llu\n",
 273                         (unsigned long long)bio->bi_sector,
 274                         bdevname(tmp_dev->rdev->bdev, b),
 275                         (unsigned long long)tmp_dev->rdev->sectors,
 276                         (unsigned long long)start_sector);
 277                 bio_io_error(bio);
 278                 return 0;
 279         }
 280         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
 281                      tmp_dev->end_sector)) {
 282                 /* This bio crosses a device boundary, so we have to
 283                  * split it.
 284                  */
 285                 struct bio_pair *bp;
 286
 287                 bp = bio_split(bio,
 288                                tmp_dev->end_sector - bio->bi_sector);
 289
 290                 if (linear_make_request(q, &bp->bio1))
 291                         generic_make_request(&bp->bio1);
 292                 if (linear_make_request(q, &bp->bio2))
 293                         generic_make_request(&bp->bio2);
 294                 bio_pair_release(bp);
 295                 return 0;
 296         }
 297
 298         bio->bi_bdev = tmp_dev->rdev->bdev;
 299         bio->bi_sector = bio->bi_sector - start_sector
 300                 + tmp_dev->rdev->data_offset;
 301
 302         return 1;
 303 }
 304
 305 static void linear_status (struct seq_file *seq, mddev_t *mddev)
 306 {
 307
 308         seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
 309 }
 310
 311
 312 static struct mdk_personality linear_personality =
 313 {
 314         .name           = "linear",
 315         .level          = LEVEL_LINEAR,
 316         .owner          = THIS_MODULE,
 317         .make_request   = linear_make_request,
 318         .run            = linear_run,
 319         .stop           = linear_stop,
 320         .status         = linear_status,
 321         .hot_add_disk   = linear_add,
 322         .size           = linear_size,
 323 };
 324
 325 static int __init linear_init (void)
 326 {
 327         return register_md_personality (&linear_personality);
 328 }
 329
 330 static void linear_exit (void)
 331 {
 332         unregister_md_personality (&linear_personality);
 333 }
 334
 335
 336 module_init(linear_init);
 337 module_exit(linear_exit);
 338 MODULE_LICENSE("GPL");
 339 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 340 MODULE_ALIAS("md-linear");
 341 MODULE_ALIAS("md-level--1");