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 <linux/module.h>
  23 #include <linux/slab.h>
  24 #include "md.h"
  25 #include "linear.h"
  26
  27 /*
  28  * find which device holds a particular offset
  29  */
  30 static inline struct dev_info *which_dev(struct mddev *mddev, sector_t sector)
  31 {
  32         int lo, mid, hi;
  33         struct linear_conf *conf;
  34
  35         lo = 0;
  36         hi = mddev->raid_disks - 1;
  37         conf = mddev->private;
  38
  39         /*
  40          * Binary Search
  41          */
  42
  43         while (hi > lo) {
  44
  45                 mid = (hi + lo) / 2;
  46                 if (sector < conf->disks[mid].end_sector)
  47                         hi = mid;
  48                 else
  49                         lo = mid + 1;
  50         }
  51
  52         return conf->disks + lo;
  53 }
  54
  55 /*
  56  * In linear_congested() conf->raid_disks is used as a copy of
  57  * mddev->raid_disks to iterate conf->disks[], because conf->raid_disks
  58  * and conf->disks[] are created in linear_conf(), they are always
  59  * consitent with each other, but mddev->raid_disks does not.
  60  */
  61 static int linear_congested(struct mddev *mddev, int bits)
  62 {
  63         struct linear_conf *conf;
  64         int i, ret = 0;
  65
  66         rcu_read_lock();
  67         conf = rcu_dereference(mddev->private);
  68
  69         for (i = 0; i < conf->raid_disks && !ret ; i++) {
  70                 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
  71                 ret |= bdi_congested(&q->backing_dev_info, bits);
  72         }
  73
  74         rcu_read_unlock();
  75         return ret;
  76 }
  77
  78 static sector_t linear_size(struct mddev *mddev, sector_t sectors, int raid_disks)
  79 {
  80         struct linear_conf *conf;
  81         sector_t array_sectors;
  82
  83         conf = mddev->private;
  84         WARN_ONCE(sectors || raid_disks,
  85                   "%s does not support generic reshape\n", __func__);
  86         array_sectors = conf->array_sectors;
  87
  88         return array_sectors;
  89 }
  90
  91 static struct linear_conf *linear_conf(struct mddev *mddev, int raid_disks)
  92 {
  93         struct linear_conf *conf;
  94         struct md_rdev *rdev;
  95         int i, cnt;
  96         bool discard_supported = false;
  97
  98         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(struct dev_info),
  99                         GFP_KERNEL);
 100         if (!conf)
 101                 return NULL;
 102
 103         cnt = 0;
 104         conf->array_sectors = 0;
 105
 106         rdev_for_each(rdev, mddev) {
 107                 int j = rdev->raid_disk;
 108                 struct dev_info *disk = conf->disks + j;
 109                 sector_t sectors;
 110
 111                 if (j < 0 || j >= raid_disks || disk->rdev) {
 112                         printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n",
 113                                mdname(mddev));
 114                         goto out;
 115                 }
 116
 117                 disk->rdev = rdev;
 118                 if (mddev->chunk_sectors) {
 119                         sectors = rdev->sectors;
 120                         sector_div(sectors, mddev->chunk_sectors);
 121                         rdev->sectors = sectors * mddev->chunk_sectors;
 122                 }
 123
 124                 disk_stack_limits(mddev->gendisk, rdev->bdev,
 125                                   rdev->data_offset << 9);
 126
 127                 conf->array_sectors += rdev->sectors;
 128                 cnt++;
 129
 130                 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
 131                         discard_supported = true;
 132         }
 133         if (cnt != raid_disks) {
 134                 printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n",
 135                        mdname(mddev));
 136                 goto out;
 137         }
 138
 139         if (!discard_supported)
 140                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
 141         else
 142                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
 143
 144         /*
 145          * Here we calculate the device offsets.
 146          */
 147         conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
 148
 149         for (i = 1; i < raid_disks; i++)
 150                 conf->disks[i].end_sector =
 151                         conf->disks[i-1].end_sector +
 152                         conf->disks[i].rdev->sectors;
 153
 154         /*
 155          * conf->raid_disks is copy of mddev->raid_disks. The reason to
 156          * keep a copy of mddev->raid_disks in struct linear_conf is,
 157          * mddev->raid_disks may not be consistent with pointers number of
 158          * conf->disks[] when it is updated in linear_add() and used to
 159          * iterate old conf->disks[] earray in linear_congested().
 160          * Here conf->raid_disks is always consitent with number of
 161          * pointers in conf->disks[] array, and mddev->private is updated
 162          * with rcu_assign_pointer() in linear_addr(), such race can be
 163          * avoided.
 164          */
 165         conf->raid_disks = raid_disks;
 166
 167         return conf;
 168
 169 out:
 170         kfree(conf);
 171         return NULL;
 172 }
 173
 174 static int linear_run (struct mddev *mddev)
 175 {
 176         struct linear_conf *conf;
 177         int ret;
 178
 179         if (md_check_no_bitmap(mddev))
 180                 return -EINVAL;
 181         conf = linear_conf(mddev, mddev->raid_disks);
 182
 183         if (!conf)
 184                 return 1;
 185         mddev->private = conf;
 186         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 187
 188         ret =  md_integrity_register(mddev);
 189         if (ret) {
 190                 kfree(conf);
 191                 mddev->private = NULL;
 192         }
 193         return ret;
 194 }
 195
 196 static int linear_add(struct mddev *mddev, struct md_rdev *rdev)
 197 {
 198         /* Adding a drive to a linear array allows the array to grow.
 199          * It is permitted if the new drive has a matching superblock
 200          * already on it, with raid_disk equal to raid_disks.
 201          * It is achieved by creating a new linear_private_data structure
 202          * and swapping it in in-place of the current one.
 203          * The current one is never freed until the array is stopped.
 204          * This avoids races.
 205          */
 206         struct linear_conf *newconf, *oldconf;
 207
 208         if (rdev->saved_raid_disk != mddev->raid_disks)
 209                 return -EINVAL;
 210
 211         rdev->raid_disk = rdev->saved_raid_disk;
 212         rdev->saved_raid_disk = -1;
 213
 214         newconf = linear_conf(mddev,mddev->raid_disks+1);
 215
 216         if (!newconf)
 217                 return -ENOMEM;
 218
 219         /* newconf->raid_disks already keeps a copy of * the increased
 220          * value of mddev->raid_disks, WARN_ONCE() is just used to make
 221          * sure of this. It is possible that oldconf is still referenced
 222          * in linear_congested(), therefore kfree_rcu() is used to free
 223          * oldconf until no one uses it anymore.
 224          */
 225         mddev_suspend(mddev);
 226         oldconf = rcu_dereference_protected(mddev->private,
 227                         lockdep_is_held(&mddev->reconfig_mutex));
 228         mddev->raid_disks++;
 229         WARN_ONCE(mddev->raid_disks != newconf->raid_disks,
 230                 "copied raid_disks doesn't match mddev->raid_disks");
 231         rcu_assign_pointer(mddev->private, newconf);
 232         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 233         set_capacity(mddev->gendisk, mddev->array_sectors);
 234         mddev_resume(mddev);
 235         revalidate_disk(mddev->gendisk);
 236         kfree_rcu(oldconf, rcu);
 237         return 0;
 238 }
 239
 240 static void linear_free(struct mddev *mddev, void *priv)
 241 {
 242         struct linear_conf *conf = priv;
 243
 244         kfree(conf);
 245 }
 246
 247 static void linear_make_request(struct mddev *mddev, struct bio *bio)
 248 {
 249         char b[BDEVNAME_SIZE];
 250         struct dev_info *tmp_dev;
 251         struct bio *split;
 252         sector_t start_sector, end_sector, data_offset;
 253
 254         if (unlikely(bio->bi_rw & REQ_FLUSH)) {
 255                 md_flush_request(mddev, bio);
 256                 return;
 257         }
 258
 259         do {
 260                 tmp_dev = which_dev(mddev, bio->bi_iter.bi_sector);
 261                 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
 262                 end_sector = tmp_dev->end_sector;
 263                 data_offset = tmp_dev->rdev->data_offset;
 264                 bio->bi_bdev = tmp_dev->rdev->bdev;
 265
 266                 if (unlikely(bio->bi_iter.bi_sector >= end_sector ||
 267                              bio->bi_iter.bi_sector < start_sector))
 268                         goto out_of_bounds;
 269
 270                 if (unlikely(bio_end_sector(bio) > end_sector)) {
 271                         /* This bio crosses a device boundary, so we have to
 272                          * split it.
 273                          */
 274                         split = bio_split(bio, end_sector -
 275                                           bio->bi_iter.bi_sector,
 276                                           GFP_NOIO, fs_bio_set);
 277                         bio_chain(split, bio);
 278                 } else {
 279                         split = bio;
 280                 }
 281
 282                 split->bi_iter.bi_sector = split->bi_iter.bi_sector -
 283                         start_sector + data_offset;
 284
 285                 if (unlikely((split->bi_rw & REQ_DISCARD) &&
 286                          !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
 287                         /* Just ignore it */
 288                         bio_endio(split);
 289                 } else
 290                         generic_make_request(split);
 291         } while (split != bio);
 292         return;
 293
 294 out_of_bounds:
 295         printk(KERN_ERR
 296                "md/linear:%s: make_request: Sector %llu out of bounds on "
 297                "dev %s: %llu sectors, offset %llu\n",
 298                mdname(mddev),
 299                (unsigned long long)bio->bi_iter.bi_sector,
 300                bdevname(tmp_dev->rdev->bdev, b),
 301                (unsigned long long)tmp_dev->rdev->sectors,
 302                (unsigned long long)start_sector);
 303         bio_io_error(bio);
 304 }
 305
 306 static void linear_status (struct seq_file *seq, struct mddev *mddev)
 307 {
 308
 309         seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
 310 }
 311
 312 static void linear_quiesce(struct mddev *mddev, int state)
 313 {
 314 }
 315
 316 static struct md_personality linear_personality =
 317 {
 318         .name           = "linear",
 319         .level          = LEVEL_LINEAR,
 320         .owner          = THIS_MODULE,
 321         .make_request   = linear_make_request,
 322         .run            = linear_run,
 323         .free           = linear_free,
 324         .status         = linear_status,
 325         .hot_add_disk   = linear_add,
 326         .size           = linear_size,
 327         .quiesce        = linear_quiesce,
 328         .congested      = linear_congested,
 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 module_init(linear_init);
 342 module_exit(linear_exit);
 343 MODULE_LICENSE("GPL");
 344 MODULE_DESCRIPTION("Linear device concatenation personality for MD");
 345 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 346 MODULE_ALIAS("md-linear");
 347 MODULE_ALIAS("md-level--1");