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/slab.h>
  23 #include "md.h"
  24 #include "linear.h"
  25
  26 /*
  27  * find which device holds a particular offset
  28  */
  29 static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
  30 {
  31         int lo, mid, hi;
  32         linear_conf_t *conf;
  33
  34         lo = 0;
  35         hi = mddev->raid_disks - 1;
  36         conf = rcu_dereference(mddev->private);
  37
  38         /*
  39          * Binary Search
  40          */
  41
  42         while (hi > lo) {
  43
  44                 mid = (hi + lo) / 2;
  45                 if (sector < conf->disks[mid].end_sector)
  46                         hi = mid;
  47                 else
  48                         lo = mid + 1;
  49         }
  50
  51         return conf->disks + lo;
  52 }
  53
  54 /**
  55  *      linear_mergeable_bvec -- tell bio layer if two requests can be merged
  56  *      @q: request queue
  57  *      @bvm: properties of new bio
  58  *      @biovec: the request that could be merged to it.
  59  *
  60  *      Return amount of bytes we can take at this offset
  61  */
  62 static int linear_mergeable_bvec(struct request_queue *q,
  63                                  struct bvec_merge_data *bvm,
  64                                  struct bio_vec *biovec)
  65 {
  66         mddev_t *mddev = q->queuedata;
  67         dev_info_t *dev0;
  68         unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
  69         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
  70
  71         rcu_read_lock();
  72         dev0 = which_dev(mddev, sector);
  73         maxsectors = dev0->end_sector - sector;
  74         rcu_read_unlock();
  75
  76         if (maxsectors < bio_sectors)
  77                 maxsectors = 0;
  78         else
  79                 maxsectors -= bio_sectors;
  80
  81         if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
  82                 return biovec->bv_len;
  83         /* The bytes available at this offset could be really big,
  84          * so we cap at 2^31 to avoid overflow */
  85         if (maxsectors > (1 << (31-9)))
  86                 return 1<<31;
  87         return maxsectors << 9;
  88 }
  89
  90 static void linear_unplug(struct request_queue *q)
  91 {
  92         mddev_t *mddev = q->queuedata;
  93         linear_conf_t *conf;
  94         int i;
  95
  96         rcu_read_lock();
  97         conf = rcu_dereference(mddev->private);
  98
  99         for (i=0; i < mddev->raid_disks; i++) {
 100                 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
 101                 blk_unplug(r_queue);
 102         }
 103         rcu_read_unlock();
 104 }
 105
 106 static int linear_congested(void *data, int bits)
 107 {
 108         mddev_t *mddev = data;
 109         linear_conf_t *conf;
 110         int i, ret = 0;
 111
 112         if (mddev_congested(mddev, bits))
 113                 return 1;
 114
 115         rcu_read_lock();
 116         conf = rcu_dereference(mddev->private);
 117
 118         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
 119                 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
 120                 ret |= bdi_congested(&q->backing_dev_info, bits);
 121         }
 122
 123         rcu_read_unlock();
 124         return ret;
 125 }
 126
 127 static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 128 {
 129         linear_conf_t *conf;
 130         sector_t array_sectors;
 131
 132         rcu_read_lock();
 133         conf = rcu_dereference(mddev->private);
 134         WARN_ONCE(sectors || raid_disks,
 135                   "%s does not support generic reshape\n", __func__);
 136         array_sectors = conf->array_sectors;
 137         rcu_read_unlock();
 138
 139         return array_sectors;
 140 }
 141
 142 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
 143 {
 144         linear_conf_t *conf;
 145         mdk_rdev_t *rdev;
 146         int i, cnt;
 147
 148         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
 149                         GFP_KERNEL);
 150         if (!conf)
 151                 return NULL;
 152
 153         cnt = 0;
 154         conf->array_sectors = 0;
 155
 156         list_for_each_entry(rdev, &mddev->disks, same_set) {
 157                 int j = rdev->raid_disk;
 158                 dev_info_t *disk = conf->disks + j;
 159                 sector_t sectors;
 160
 161                 if (j < 0 || j >= raid_disks || disk->rdev) {
 162                         printk("linear: disk numbering problem. Aborting!\n");
 163                         goto out;
 164                 }
 165
 166                 disk->rdev = rdev;
 167                 if (mddev->chunk_sectors) {
 168                         sectors = rdev->sectors;
 169                         sector_div(sectors, mddev->chunk_sectors);
 170                         rdev->sectors = sectors * mddev->chunk_sectors;
 171                 }
 172
 173                 disk_stack_limits(mddev->gendisk, rdev->bdev,
 174                                   rdev->data_offset << 9);
 175                 /* as we don't honour merge_bvec_fn, we must never risk
 176                  * violating it, so limit max_segments to 1 lying within
 177                  * a single page.
 178                  */
 179                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
 180                         blk_queue_max_segments(mddev->queue, 1);
 181                         blk_queue_segment_boundary(mddev->queue,
 182                                                    PAGE_CACHE_SIZE - 1);
 183                 }
 184
 185                 conf->array_sectors += rdev->sectors;
 186                 cnt++;
 187
 188         }
 189         if (cnt != raid_disks) {
 190                 printk("linear: not enough drives present. Aborting!\n");
 191                 goto out;
 192         }
 193
 194         /*
 195          * Here we calculate the device offsets.
 196          */
 197         conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
 198
 199         for (i = 1; i < raid_disks; i++)
 200                 conf->disks[i].end_sector =
 201                         conf->disks[i-1].end_sector +
 202                         conf->disks[i].rdev->sectors;
 203
 204         return conf;
 205
 206 out:
 207         kfree(conf);
 208         return NULL;
 209 }
 210
 211 static int linear_run (mddev_t *mddev)
 212 {
 213         linear_conf_t *conf;
 214
 215         if (md_check_no_bitmap(mddev))
 216                 return -EINVAL;
 217         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 218         conf = linear_conf(mddev, mddev->raid_disks);
 219
 220         if (!conf)
 221                 return 1;
 222         mddev->private = conf;
 223         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 224
 225         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
 226         mddev->queue->unplug_fn = linear_unplug;
 227         mddev->queue->backing_dev_info.congested_fn = linear_congested;
 228         mddev->queue->backing_dev_info.congested_data = mddev;
 229         md_integrity_register(mddev);
 230         return 0;
 231 }
 232
 233 static void free_conf(struct rcu_head *head)
 234 {
 235         linear_conf_t *conf = container_of(head, linear_conf_t, rcu);
 236         kfree(conf);
 237 }
 238
 239 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
 240 {
 241         /* Adding a drive to a linear array allows the array to grow.
 242          * It is permitted if the new drive has a matching superblock
 243          * already on it, with raid_disk equal to raid_disks.
 244          * It is achieved by creating a new linear_private_data structure
 245          * and swapping it in in-place of the current one.
 246          * The current one is never freed until the array is stopped.
 247          * This avoids races.
 248          */
 249         linear_conf_t *newconf, *oldconf;
 250
 251         if (rdev->saved_raid_disk != mddev->raid_disks)
 252                 return -EINVAL;
 253
 254         rdev->raid_disk = rdev->saved_raid_disk;
 255
 256         newconf = linear_conf(mddev,mddev->raid_disks+1);
 257
 258         if (!newconf)
 259                 return -ENOMEM;
 260
 261         oldconf = rcu_dereference(mddev->private);
 262         mddev->raid_disks++;
 263         rcu_assign_pointer(mddev->private, newconf);
 264         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
 265         set_capacity(mddev->gendisk, mddev->array_sectors);
 266         revalidate_disk(mddev->gendisk);
 267         call_rcu(&oldconf->rcu, free_conf);
 268         return 0;
 269 }
 270
 271 static int linear_stop (mddev_t *mddev)
 272 {
 273         linear_conf_t *conf = mddev->private;
 274
 275         /*
 276          * We do not require rcu protection here since
 277          * we hold reconfig_mutex for both linear_add and
 278          * linear_stop, so they cannot race.
 279          * We should make sure any old 'conf's are properly
 280          * freed though.
 281          */
 282         rcu_barrier();
 283         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 284         kfree(conf);
 285
 286         return 0;
 287 }
 288
 289 static int linear_make_request (struct request_queue *q, struct bio *bio)
 290 {
 291         const int rw = bio_data_dir(bio);
 292         mddev_t *mddev = q->queuedata;
 293         dev_info_t *tmp_dev;
 294         sector_t start_sector;
 295         int cpu;
 296
 297         if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
 298                 md_barrier_request(mddev, bio);
 299                 return 0;
 300         }
 301
 302         cpu = part_stat_lock();
 303         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 304         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 305                       bio_sectors(bio));
 306         part_stat_unlock();
 307
 308         rcu_read_lock();
 309         tmp_dev = which_dev(mddev, bio->bi_sector);
 310         start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
 311
 312
 313         if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
 314                      || (bio->bi_sector < start_sector))) {
 315                 char b[BDEVNAME_SIZE];
 316
 317                 printk("linear_make_request: Sector %llu out of bounds on "
 318                         "dev %s: %llu sectors, offset %llu\n",
 319                         (unsigned long long)bio->bi_sector,
 320                         bdevname(tmp_dev->rdev->bdev, b),
 321                         (unsigned long long)tmp_dev->rdev->sectors,
 322                         (unsigned long long)start_sector);
 323                 rcu_read_unlock();
 324                 bio_io_error(bio);
 325                 return 0;
 326         }
 327         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
 328                      tmp_dev->end_sector)) {
 329                 /* This bio crosses a device boundary, so we have to
 330                  * split it.
 331                  */
 332                 struct bio_pair *bp;
 333                 sector_t end_sector = tmp_dev->end_sector;
 334
 335                 rcu_read_unlock();
 336
 337                 bp = bio_split(bio, end_sector - bio->bi_sector);
 338
 339                 if (linear_make_request(q, &bp->bio1))
 340                         generic_make_request(&bp->bio1);
 341                 if (linear_make_request(q, &bp->bio2))
 342                         generic_make_request(&bp->bio2);
 343                 bio_pair_release(bp);
 344                 return 0;
 345         }
 346
 347         bio->bi_bdev = tmp_dev->rdev->bdev;
 348         bio->bi_sector = bio->bi_sector - start_sector
 349                 + tmp_dev->rdev->data_offset;
 350         rcu_read_unlock();
 351
 352         return 1;
 353 }
 354
 355 static void linear_status (struct seq_file *seq, mddev_t *mddev)
 356 {
 357
 358         seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
 359 }
 360
 361
 362 static struct mdk_personality linear_personality =
 363 {
 364         .name           = "linear",
 365         .level          = LEVEL_LINEAR,
 366         .owner          = THIS_MODULE,
 367         .make_request   = linear_make_request,
 368         .run            = linear_run,
 369         .stop           = linear_stop,
 370         .status         = linear_status,
 371         .hot_add_disk   = linear_add,
 372         .size           = linear_size,
 373 };
 374
 375 static int __init linear_init (void)
 376 {
 377         return register_md_personality (&linear_personality);
 378 }
 379
 380 static void linear_exit (void)
 381 {
 382         unregister_md_personality (&linear_personality);
 383 }
 384
 385
 386 module_init(linear_init);
 387 module_exit(linear_exit);
 388 MODULE_LICENSE("GPL");
 389 MODULE_DESCRIPTION("Linear device concatenation personality for MD");
 390 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
 391 MODULE_ALIAS("md-linear");
 392 MODULE_ALIAS("md-level--1");