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 = rcu_dereference(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  *      linear_mergeable_bvec -- tell bio layer if two requests can be merged
  57  *      @q: request queue
  58  *      @bvm: properties of new bio
  59  *      @biovec: the request that could be merged to it.
  60  *
  61  *      Return amount of bytes we can take at this offset
  62  */
  63 static int linear_mergeable_bvec(struct request_queue *q,
  64                                  struct bvec_merge_data *bvm,
  65                                  struct bio_vec *biovec)
  66 {
  67         struct mddev *mddev = q->queuedata;
  68         struct dev_info *dev0;
  69         unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
  70         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
  71         int maxbytes = biovec->bv_len;
  72         struct request_queue *subq;
  73
  74         rcu_read_lock();
  75         dev0 = which_dev(mddev, sector);
  76         maxsectors = dev0->end_sector - sector;
  77         subq = bdev_get_queue(dev0->rdev->bdev);
  78         if (subq->merge_bvec_fn) {
  79                 bvm->bi_bdev = dev0->rdev->bdev;
  80                 bvm->bi_sector -= dev0->end_sector - dev0->rdev->sectors;
  81                 maxbytes = min(maxbytes, subq->merge_bvec_fn(subq, bvm,
  82                                                              biovec));
  83         }
  84         rcu_read_unlock();
  85
  86         if (maxsectors < bio_sectors)
  87                 maxsectors = 0;
  88         else
  89                 maxsectors -= bio_sectors;
  90
  91         if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
  92                 return maxbytes;
  93
  94         if (maxsectors > (maxbytes >> 9))
  95                 return maxbytes;
  96         else
  97                 return maxsectors << 9;
  98 }
  99
 100 static int linear_congested(void *data, int bits)
 101 {
 102         struct mddev *mddev = data;
 103         struct linear_conf *conf;
 104         int i, ret = 0;
 105
 106         if (mddev_congested(mddev, bits))
 107                 return 1;
 108
 109         rcu_read_lock();
 110         conf = rcu_dereference(mddev->private);
 111
 112         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
 113                 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
 114                 ret |= bdi_congested(&q->backing_dev_info, bits);
 115         }
 116
 117         rcu_read_unlock();
 118         return ret;
 119 }
 120
 121 static sector_t linear_size(struct mddev *mddev, sector_t sectors, int raid_disks)
 122 {
 123         struct linear_conf *conf;
 124         sector_t array_sectors;
 125
 126         rcu_read_lock();
 127         conf = rcu_dereference(mddev->private);
 128         WARN_ONCE(sectors || raid_disks,
 129                   "%s does not support generic reshape\n", __func__);
 130         array_sectors = conf->array_sectors;
 131         rcu_read_unlock();
 132
 133         return array_sectors;
 134 }
 135
 136 static struct linear_conf *linear_conf(struct mddev *mddev, int raid_disks)
 137 {
 138         struct linear_conf *conf;
 139         struct md_rdev *rdev;
 140         int i, cnt;
 141         bool discard_supported = false;
 142
 143         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(struct dev_info),
 144                         GFP_KERNEL);
 145         if (!conf)
 146                 return NULL;
 147
 148         cnt = 0;
 149         conf->array_sectors = 0;
 150
 151         rdev_for_each(rdev, mddev) {
 152                 int j = rdev->raid_disk;
 153                 struct dev_info *disk = conf->disks + j;
 154                 sector_t sectors;
 155
 156                 if (j < 0 || j >= raid_disks || disk->rdev) {
 157                         printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n",
 158                                mdname(mddev));
 159                         goto out;
 160                 }
 161
 162                 disk->rdev = rdev;
 163                 if (mddev->chunk_sectors) {
 164                         sectors = rdev->sectors;
 165                         sector_div(sectors, mddev->chunk_sectors);
 166                         rdev->sectors = sectors * mddev->chunk_sectors;
 167                 }
 168
 169                 disk_stack_limits(mddev->gendisk, rdev->bdev,
 170                                   rdev->data_offset << 9);
 171
 172                 conf->array_sectors += rdev->sectors;
 173                 cnt++;
 174
 175                 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
 176                         discard_supported = true;
 177         }
 178         if (cnt != raid_disks) {
 179                 printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n",
 180                        mdname(mddev));
 181                 goto out;
 182         }
 183
 184         if (!discard_supported)
 185                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
 186         else
 187                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
 188
 189         /*
 190          * Here we calculate the device offsets.
 191          */
 192         conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
 193
 194         for (i = 1; i < raid_disks; i++)
 195                 conf->disks[i].end_sector =
 196                         conf->disks[i-1].end_sector +
 197                         conf->disks[i].rdev->sectors;
 198
 199         return conf;
 200
 201 out:
 202         kfree(conf);
 203         return NULL;
 204 }
 205
 206 static int linear_run (struct mddev *mddev)
 207 {
 208         struct linear_conf *conf;
 209         int ret;
 210
 211         if (md_check_no_bitmap(mddev))
 212                 return -EINVAL;
 213         conf = linear_conf(mddev, mddev->raid_disks);
 214
 215         if (!conf)
 216                 return 1;
 217         mddev->private = conf;
 218         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 219
 220         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 221         mddev->queue->backing_dev_info.congested_fn = linear_congested;
 222         mddev->queue->backing_dev_info.congested_data = mddev;
 223
 224         ret =  md_integrity_register(mddev);
 225         if (ret) {
 226                 kfree(conf);
 227                 mddev->private = NULL;
 228         }
 229         return ret;
 230 }
 231
 232 static int linear_add(struct mddev *mddev, struct md_rdev *rdev)
 233 {
 234         /* Adding a drive to a linear array allows the array to grow.
 235          * It is permitted if the new drive has a matching superblock
 236          * already on it, with raid_disk equal to raid_disks.
 237          * It is achieved by creating a new linear_private_data structure
 238          * and swapping it in in-place of the current one.
 239          * The current one is never freed until the array is stopped.
 240          * This avoids races.
 241          */
 242         struct linear_conf *newconf, *oldconf;
 243
 244         if (rdev->saved_raid_disk != mddev->raid_disks)
 245                 return -EINVAL;
 246
 247         rdev->raid_disk = rdev->saved_raid_disk;
 248         rdev->saved_raid_disk = -1;
 249
 250         newconf = linear_conf(mddev,mddev->raid_disks+1);
 251
 252         if (!newconf)
 253                 return -ENOMEM;
 254
 255         oldconf = rcu_dereference_protected(mddev->private,
 256                                             lockdep_is_held(
 257                                                     &mddev->reconfig_mutex));
 258         mddev->raid_disks++;
 259         rcu_assign_pointer(mddev->private, newconf);
 260         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 261         set_capacity(mddev->gendisk, mddev->array_sectors);
 262         revalidate_disk(mddev->gendisk);
 263         kfree_rcu(oldconf, rcu);
 264         return 0;
 265 }
 266
 267 static int linear_stop (struct mddev *mddev)
 268 {
 269         struct linear_conf *conf =
 270                 rcu_dereference_protected(mddev->private,
 271                                           lockdep_is_held(
 272                                                   &mddev->reconfig_mutex));
 273
 274         /*
 275          * We do not require rcu protection here since
 276          * we hold reconfig_mutex for both linear_add and
 277          * linear_stop, so they cannot race.
 278          * We should make sure any old 'conf's are properly
 279          * freed though.
 280          */
 281         rcu_barrier();
 282         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 283         kfree(conf);
 284         mddev->private = NULL;
 285
 286         return 0;
 287 }
 288
 289 static void linear_make_request(struct mddev *mddev, struct bio *bio)
 290 {
 291         char b[BDEVNAME_SIZE];
 292         struct dev_info *tmp_dev;
 293         struct bio *split;
 294         sector_t start_sector, end_sector, data_offset;
 295
 296         if (unlikely(bio->bi_rw & REQ_FLUSH)) {
 297                 md_flush_request(mddev, bio);
 298                 return;
 299         }
 300
 301         do {
 302                 rcu_read_lock();
 303
 304                 tmp_dev = which_dev(mddev, bio->bi_iter.bi_sector);
 305                 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
 306                 end_sector = tmp_dev->end_sector;
 307                 data_offset = tmp_dev->rdev->data_offset;
 308                 bio->bi_bdev = tmp_dev->rdev->bdev;
 309
 310                 rcu_read_unlock();
 311
 312                 if (unlikely(bio->bi_iter.bi_sector >= end_sector ||
 313                              bio->bi_iter.bi_sector < start_sector))
 314                         goto out_of_bounds;
 315
 316                 if (unlikely(bio_end_sector(bio) > end_sector)) {
 317                         /* This bio crosses a device boundary, so we have to
 318                          * split it.
 319                          */
 320                         split = bio_split(bio, end_sector -
 321                                           bio->bi_iter.bi_sector,
 322                                           GFP_NOIO, fs_bio_set);
 323                         bio_chain(split, bio);
 324                 } else {
 325                         split = bio;
 326                 }
 327
 328                 split->bi_iter.bi_sector = split->bi_iter.bi_sector -
 329                         start_sector + data_offset;
 330
 331                 if (unlikely((split->bi_rw & REQ_DISCARD) &&
 332                          !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
 333                         /* Just ignore it */
 334                         bio_endio(split, 0);
 335                 } else
 336                         generic_make_request(split);
 337         } while (split != bio);
 338         return;
 339
 340 out_of_bounds:
 341         printk(KERN_ERR
 342                "md/linear:%s: make_request: Sector %llu out of bounds on "
 343                "dev %s: %llu sectors, offset %llu\n",
 344                mdname(mddev),
 345                (unsigned long long)bio->bi_iter.bi_sector,
 346                bdevname(tmp_dev->rdev->bdev, b),
 347                (unsigned long long)tmp_dev->rdev->sectors,
 348                (unsigned long long)start_sector);
 349         bio_io_error(bio);
 350 }
 351
 352 static void linear_status (struct seq_file *seq, struct mddev *mddev)
 353 {
 354
 355         seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
 356 }
 357
 358
 359 static struct md_personality linear_personality =
 360 {
 361         .name           = "linear",
 362         .level          = LEVEL_LINEAR,
 363         .owner          = THIS_MODULE,
 364         .make_request   = linear_make_request,
 365         .run            = linear_run,
 366         .stop           = linear_stop,
 367         .status         = linear_status,
 368         .hot_add_disk   = linear_add,
 369         .size           = linear_size,
 370 };
 371
 372 static int __init linear_init (void)
 373 {
 374         return register_md_personality (&linear_personality);
 375 }
 376
 377 static void linear_exit (void)
 378 {
 379         unregister_md_personality (&linear_personality);
 380 }
 381
 382
 383 module_init(linear_init);
 384 module_exit(linear_exit);
 385 MODULE_LICENSE("GPL");
 386 MODULE_DESCRIPTION("Linear device concatenation personality for MD");
 387 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 388 MODULE_ALIAS("md-linear");
 389 MODULE_ALIAS("md-level--1");