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