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                 sector_t sectors;
 139
 140                 if (j < 0 || j >= raid_disks || disk->rdev) {
 141                         printk("linear: disk numbering problem. Aborting!\n");
 142                         goto out;
 143                 }
 144
 145                 disk->rdev = rdev;
 146                 if (mddev->chunk_sectors) {
 147                         sectors = rdev->sectors;
 148                         sector_div(sectors, mddev->chunk_sectors);
 149                         rdev->sectors = sectors * mddev->chunk_sectors;
 150                 }
 151
 152                 blk_queue_stack_limits(mddev->queue,
 153                                        rdev->bdev->bd_disk->queue);
 154                 /* as we don't honour merge_bvec_fn, we must never risk
 155                  * violating it, so limit ->max_sector to one PAGE, as
 156                  * a one page request is never in violation.
 157                  */
 158                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
 159                     queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
 160                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 161
 162                 conf->array_sectors += rdev->sectors;
 163                 cnt++;
 164
 165         }
 166         if (cnt != raid_disks) {
 167                 printk("linear: not enough drives present. Aborting!\n");
 168                 goto out;
 169         }
 170
 171         /*
 172          * Here we calculate the device offsets.
 173          */
 174         conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
 175
 176         for (i = 1; i < raid_disks; i++)
 177                 conf->disks[i].end_sector =
 178                         conf->disks[i-1].end_sector +
 179                         conf->disks[i].rdev->sectors;
 180
 181         return conf;
 182
 183 out:
 184         kfree(conf);
 185         return NULL;
 186 }
 187
 188 static int linear_run (mddev_t *mddev)
 189 {
 190         linear_conf_t *conf;
 191
 192         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 193         conf = linear_conf(mddev, mddev->raid_disks);
 194
 195         if (!conf)
 196                 return 1;
 197         mddev->private = conf;
 198         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 199
 200         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 201         mddev->queue->unplug_fn = linear_unplug;
 202         mddev->queue->backing_dev_info.congested_fn = linear_congested;
 203         mddev->queue->backing_dev_info.congested_data = mddev;
 204         return 0;
 205 }
 206
 207 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
 208 {
 209         /* Adding a drive to a linear array allows the array to grow.
 210          * It is permitted if the new drive has a matching superblock
 211          * already on it, with raid_disk equal to raid_disks.
 212          * It is achieved by creating a new linear_private_data structure
 213          * and swapping it in in-place of the current one.
 214          * The current one is never freed until the array is stopped.
 215          * This avoids races.
 216          */
 217         linear_conf_t *newconf;
 218
 219         if (rdev->saved_raid_disk != mddev->raid_disks)
 220                 return -EINVAL;
 221
 222         rdev->raid_disk = rdev->saved_raid_disk;
 223
 224         newconf = linear_conf(mddev,mddev->raid_disks+1);
 225
 226         if (!newconf)
 227                 return -ENOMEM;
 228
 229         newconf->prev = mddev->private;
 230         mddev->private = newconf;
 231         mddev->raid_disks++;
 232         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 233         set_capacity(mddev->gendisk, mddev->array_sectors);
 234         return 0;
 235 }
 236
 237 static int linear_stop (mddev_t *mddev)
 238 {
 239         linear_conf_t *conf = mddev->private;
 240
 241         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 242         do {
 243                 linear_conf_t *t = conf->prev;
 244                 kfree(conf);
 245                 conf = t;
 246         } while (conf);
 247
 248         return 0;
 249 }
 250
 251 static int linear_make_request (struct request_queue *q, struct bio *bio)
 252 {
 253         const int rw = bio_data_dir(bio);
 254         mddev_t *mddev = q->queuedata;
 255         dev_info_t *tmp_dev;
 256         sector_t start_sector;
 257         int cpu;
 258
 259         if (unlikely(bio_barrier(bio))) {
 260                 bio_endio(bio, -EOPNOTSUPP);
 261                 return 0;
 262         }
 263
 264         cpu = part_stat_lock();
 265         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 266         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 267                       bio_sectors(bio));
 268         part_stat_unlock();
 269
 270         tmp_dev = which_dev(mddev, bio->bi_sector);
 271         start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
 272
 273         if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
 274                      || (bio->bi_sector < start_sector))) {
 275                 char b[BDEVNAME_SIZE];
 276
 277                 printk("linear_make_request: Sector %llu out of bounds on "
 278                         "dev %s: %llu sectors, offset %llu\n",
 279                         (unsigned long long)bio->bi_sector,
 280                         bdevname(tmp_dev->rdev->bdev, b),
 281                         (unsigned long long)tmp_dev->rdev->sectors,
 282                         (unsigned long long)start_sector);
 283                 bio_io_error(bio);
 284                 return 0;
 285         }
 286         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
 287                      tmp_dev->end_sector)) {
 288                 /* This bio crosses a device boundary, so we have to
 289                  * split it.
 290                  */
 291                 struct bio_pair *bp;
 292
 293                 bp = bio_split(bio,
 294                                tmp_dev->end_sector - bio->bi_sector);
 295
 296                 if (linear_make_request(q, &bp->bio1))
 297                         generic_make_request(&bp->bio1);
 298                 if (linear_make_request(q, &bp->bio2))
 299                         generic_make_request(&bp->bio2);
 300                 bio_pair_release(bp);
 301                 return 0;
 302         }
 303
 304         bio->bi_bdev = tmp_dev->rdev->bdev;
 305         bio->bi_sector = bio->bi_sector - start_sector
 306                 + tmp_dev->rdev->data_offset;
 307
 308         return 1;
 309 }
 310
 311 static void linear_status (struct seq_file *seq, mddev_t *mddev)
 312 {
 313
 314         seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
 315 }
 316
 317
 318 static struct mdk_personality linear_personality =
 319 {
 320         .name           = "linear",
 321         .level          = LEVEL_LINEAR,
 322         .owner          = THIS_MODULE,
 323         .make_request   = linear_make_request,
 324         .run            = linear_run,
 325         .stop           = linear_stop,
 326         .status         = linear_status,
 327         .hot_add_disk   = linear_add,
 328         .size           = linear_size,
 329 };
 330
 331 static int __init linear_init (void)
 332 {
 333         return register_md_personality (&linear_personality);
 334 }
 335
 336 static void linear_exit (void)
 337 {
 338         unregister_md_personality (&linear_personality);
 339 }
 340
 341
 342 module_init(linear_init);
 343 module_exit(linear_exit);
 344 MODULE_LICENSE("GPL");
 345 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 346 MODULE_ALIAS("md-linear");
 347 MODULE_ALIAS("md-level--1");