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 dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
  31 {
  32         int lo, mid, hi;
  33         linear_conf_t *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         mddev_t *mddev = q->queuedata;
  68         dev_info_t *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
  72         rcu_read_lock();
  73         dev0 = which_dev(mddev, sector);
  74         maxsectors = dev0->end_sector - sector;
  75         rcu_read_unlock();
  76
  77         if (maxsectors < bio_sectors)
  78                 maxsectors = 0;
  79         else
  80                 maxsectors -= bio_sectors;
  81
  82         if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
  83                 return biovec->bv_len;
  84         /* The bytes available at this offset could be really big,
  85          * so we cap at 2^31 to avoid overflow */
  86         if (maxsectors > (1 << (31-9)))
  87                 return 1<<31;
  88         return maxsectors << 9;
  89 }
  90
  91 static int linear_congested(void *data, int bits)
  92 {
  93         mddev_t *mddev = data;
  94         linear_conf_t *conf;
  95         int i, ret = 0;
  96
  97         if (mddev_congested(mddev, bits))
  98                 return 1;
  99
 100         rcu_read_lock();
 101         conf = rcu_dereference(mddev->private);
 102
 103         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
 104                 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
 105                 ret |= bdi_congested(&q->backing_dev_info, bits);
 106         }
 107
 108         rcu_read_unlock();
 109         return ret;
 110 }
 111
 112 static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 113 {
 114         linear_conf_t *conf;
 115         sector_t array_sectors;
 116
 117         rcu_read_lock();
 118         conf = rcu_dereference(mddev->private);
 119         WARN_ONCE(sectors || raid_disks,
 120                   "%s does not support generic reshape\n", __func__);
 121         array_sectors = conf->array_sectors;
 122         rcu_read_unlock();
 123
 124         return array_sectors;
 125 }
 126
 127 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
 128 {
 129         linear_conf_t *conf;
 130         mdk_rdev_t *rdev;
 131         int i, cnt;
 132
 133         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
 134                         GFP_KERNEL);
 135         if (!conf)
 136                 return NULL;
 137
 138         cnt = 0;
 139         conf->array_sectors = 0;
 140
 141         list_for_each_entry(rdev, &mddev->disks, same_set) {
 142                 int j = rdev->raid_disk;
 143                 dev_info_t *disk = conf->disks + j;
 144                 sector_t sectors;
 145
 146                 if (j < 0 || j >= raid_disks || disk->rdev) {
 147                         printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n",
 148                                mdname(mddev));
 149                         goto out;
 150                 }
 151
 152                 disk->rdev = rdev;
 153                 if (mddev->chunk_sectors) {
 154                         sectors = rdev->sectors;
 155                         sector_div(sectors, mddev->chunk_sectors);
 156                         rdev->sectors = sectors * mddev->chunk_sectors;
 157                 }
 158
 159                 disk_stack_limits(mddev->gendisk, rdev->bdev,
 160                                   rdev->data_offset << 9);
 161                 /* as we don't honour merge_bvec_fn, we must never risk
 162                  * violating it, so limit max_segments to 1 lying within
 163                  * a single page.
 164                  */
 165                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
 166                         blk_queue_max_segments(mddev->queue, 1);
 167                         blk_queue_segment_boundary(mddev->queue,
 168                                                    PAGE_CACHE_SIZE - 1);
 169                 }
 170
 171                 conf->array_sectors += rdev->sectors;
 172                 cnt++;
 173
 174         }
 175         if (cnt != raid_disks) {
 176                 printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n",
 177                        mdname(mddev));
 178                 goto out;
 179         }
 180
 181         /*
 182          * Here we calculate the device offsets.
 183          */
 184         conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
 185
 186         for (i = 1; i < raid_disks; i++)
 187                 conf->disks[i].end_sector =
 188                         conf->disks[i-1].end_sector +
 189                         conf->disks[i].rdev->sectors;
 190
 191         return conf;
 192
 193 out:
 194         kfree(conf);
 195         return NULL;
 196 }
 197
 198 static int linear_run (mddev_t *mddev)
 199 {
 200         linear_conf_t *conf;
 201
 202         if (md_check_no_bitmap(mddev))
 203                 return -EINVAL;
 204         conf = linear_conf(mddev, mddev->raid_disks);
 205
 206         if (!conf)
 207                 return 1;
 208         mddev->private = conf;
 209         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 210
 211         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 212         mddev->queue->backing_dev_info.congested_fn = linear_congested;
 213         mddev->queue->backing_dev_info.congested_data = mddev;
 214         return md_integrity_register(mddev);
 215 }
 216
 217 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
 218 {
 219         /* Adding a drive to a linear array allows the array to grow.
 220          * It is permitted if the new drive has a matching superblock
 221          * already on it, with raid_disk equal to raid_disks.
 222          * It is achieved by creating a new linear_private_data structure
 223          * and swapping it in in-place of the current one.
 224          * The current one is never freed until the array is stopped.
 225          * This avoids races.
 226          */
 227         linear_conf_t *newconf, *oldconf;
 228
 229         if (rdev->saved_raid_disk != mddev->raid_disks)
 230                 return -EINVAL;
 231
 232         rdev->raid_disk = rdev->saved_raid_disk;
 233
 234         newconf = linear_conf(mddev,mddev->raid_disks+1);
 235
 236         if (!newconf)
 237                 return -ENOMEM;
 238
 239         oldconf = rcu_dereference(mddev->private);
 240         mddev->raid_disks++;
 241         rcu_assign_pointer(mddev->private, newconf);
 242         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 243         set_capacity(mddev->gendisk, mddev->array_sectors);
 244         revalidate_disk(mddev->gendisk);
 245         kfree_rcu(oldconf, rcu);
 246         return 0;
 247 }
 248
 249 static int linear_stop (mddev_t *mddev)
 250 {
 251         linear_conf_t *conf = mddev->private;
 252
 253         /*
 254          * We do not require rcu protection here since
 255          * we hold reconfig_mutex for both linear_add and
 256          * linear_stop, so they cannot race.
 257          * We should make sure any old 'conf's are properly
 258          * freed though.
 259          */
 260         rcu_barrier();
 261         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 262         kfree(conf);
 263         mddev->private = NULL;
 264
 265         return 0;
 266 }
 267
 268 static int linear_make_request (mddev_t *mddev, struct bio *bio)
 269 {
 270         dev_info_t *tmp_dev;
 271         sector_t start_sector;
 272
 273         if (unlikely(bio->bi_rw & REQ_FLUSH)) {
 274                 md_flush_request(mddev, bio);
 275                 return 0;
 276         }
 277
 278         rcu_read_lock();
 279         tmp_dev = which_dev(mddev, bio->bi_sector);
 280         start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
 281
 282
 283         if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
 284                      || (bio->bi_sector < start_sector))) {
 285                 char b[BDEVNAME_SIZE];
 286
 287                 printk(KERN_ERR
 288                        "md/linear:%s: make_request: Sector %llu out of bounds on "
 289                        "dev %s: %llu sectors, offset %llu\n",
 290                        mdname(mddev),
 291                        (unsigned long long)bio->bi_sector,
 292                        bdevname(tmp_dev->rdev->bdev, b),
 293                        (unsigned long long)tmp_dev->rdev->sectors,
 294                        (unsigned long long)start_sector);
 295                 rcu_read_unlock();
 296                 bio_io_error(bio);
 297                 return 0;
 298         }
 299         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
 300                      tmp_dev->end_sector)) {
 301                 /* This bio crosses a device boundary, so we have to
 302                  * split it.
 303                  */
 304                 struct bio_pair *bp;
 305                 sector_t end_sector = tmp_dev->end_sector;
 306
 307                 rcu_read_unlock();
 308
 309                 bp = bio_split(bio, end_sector - bio->bi_sector);
 310
 311                 if (linear_make_request(mddev, &bp->bio1))
 312                         generic_make_request(&bp->bio1);
 313                 if (linear_make_request(mddev, &bp->bio2))
 314                         generic_make_request(&bp->bio2);
 315                 bio_pair_release(bp);
 316                 return 0;
 317         }
 318
 319         bio->bi_bdev = tmp_dev->rdev->bdev;
 320         bio->bi_sector = bio->bi_sector - start_sector
 321                 + tmp_dev->rdev->data_offset;
 322         rcu_read_unlock();
 323
 324         return 1;
 325 }
 326
 327 static void linear_status (struct seq_file *seq, mddev_t *mddev)
 328 {
 329
 330         seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
 331 }
 332
 333
 334 static struct mdk_personality linear_personality =
 335 {
 336         .name           = "linear",
 337         .level          = LEVEL_LINEAR,
 338         .owner          = THIS_MODULE,
 339         .make_request   = linear_make_request,
 340         .run            = linear_run,
 341         .stop           = linear_stop,
 342         .status         = linear_status,
 343         .hot_add_disk   = linear_add,
 344         .size           = linear_size,
 345 };
 346
 347 static int __init linear_init (void)
 348 {
 349         return register_md_personality (&linear_personality);
 350 }
 351
 352 static void linear_exit (void)
 353 {
 354         unregister_md_personality (&linear_personality);
 355 }
 356
 357
 358 module_init(linear_init);
 359 module_exit(linear_exit);
 360 MODULE_LICENSE("GPL");
 361 MODULE_DESCRIPTION("Linear device concatenation personality for MD");
 362 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 363 MODULE_ALIAS("md-linear");
 364 MODULE_ALIAS("md-level--1");