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