drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/linkage.h>
  38 #include <linux/raid/md.h>
  39 #include <linux/raid/bitmap.h>
  40 #include <linux/sysctl.h>
  41 #include <linux/devfs_fs_kernel.h>
  42 #include <linux/buffer_head.h> /* for invalidate_bdev */
  43 #include <linux/suspend.h>
  44
  45 #include <linux/init.h>
  46
  47 #include <linux/file.h>
  48
  49 #ifdef CONFIG_KMOD
  50 #include <linux/kmod.h>
  51 #endif
  52
  53 #include <asm/unaligned.h>
  54
  55 #define MAJOR_NR MD_MAJOR
  56 #define MD_DRIVER
  57
  58 /* 63 partitions with the alternate major number (mdp) */
  59 #define MdpMinorShift 6
  60
  61 #define DEBUG 0
  62 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  63
  64
  65 #ifndef MODULE
  66 static void autostart_arrays (int part);
  67 #endif
  68
  69 static mdk_personality_t *pers[MAX_PERSONALITY];
  70 static DEFINE_SPINLOCK(pers_lock);
  71
  72 /*
  73  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  74  * is 1000 KB/sec, so the extra system load does not show up that much.
  75  * Increase it if you want to have more _guaranteed_ speed. Note that
  76  * the RAID driver will use the maximum available bandwith if the IO
  77  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  78  * speed limit - in case reconstruction slows down your system despite
  79  * idle IO detection.
  80  *
  81  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  82  */
  83
  84 static int sysctl_speed_limit_min = 1000;
  85 static int sysctl_speed_limit_max = 200000;
  86
  87 static struct ctl_table_header *raid_table_header;
  88
  89 static ctl_table raid_table[] = {
  90         {
  91                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  92                 .procname       = "speed_limit_min",
  93                 .data           = &sysctl_speed_limit_min,
  94                 .maxlen         = sizeof(int),
  95                 .mode           = 0644,
  96                 .proc_handler   = &proc_dointvec,
  97         },
  98         {
  99                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 100                 .procname       = "speed_limit_max",
 101                 .data           = &sysctl_speed_limit_max,
 102                 .maxlen         = sizeof(int),
 103                 .mode           = 0644,
 104                 .proc_handler   = &proc_dointvec,
 105         },
 106         { .ctl_name = 0 }
 107 };
 108
 109 static ctl_table raid_dir_table[] = {
 110         {
 111                 .ctl_name       = DEV_RAID,
 112                 .procname       = "raid",
 113                 .maxlen         = 0,
 114                 .mode           = 0555,
 115                 .child          = raid_table,
 116         },
 117         { .ctl_name = 0 }
 118 };
 119
 120 static ctl_table raid_root_table[] = {
 121         {
 122                 .ctl_name       = CTL_DEV,
 123                 .procname       = "dev",
 124                 .maxlen         = 0,
 125                 .mode           = 0555,
 126                 .child          = raid_dir_table,
 127         },
 128         { .ctl_name = 0 }
 129 };
 130
 131 static struct block_device_operations md_fops;
 132
 133 /*
 134  * Enables to iterate over all existing md arrays
 135  * all_mddevs_lock protects this list.
 136  */
 137 static LIST_HEAD(all_mddevs);
 138 static DEFINE_SPINLOCK(all_mddevs_lock);
 139
 140
 141 /*
 142  * iterates through all used mddevs in the system.
 143  * We take care to grab the all_mddevs_lock whenever navigating
 144  * the list, and to always hold a refcount when unlocked.
 145  * Any code which breaks out of this loop while own
 146  * a reference to the current mddev and must mddev_put it.
 147  */
 148 #define ITERATE_MDDEV(mddev,tmp)                                        \
 149                                                                         \
 150         for (({ spin_lock(&all_mddevs_lock);                            \
 151                 tmp = all_mddevs.next;                                  \
 152                 mddev = NULL;});                                        \
 153              ({ if (tmp != &all_mddevs)                                 \
 154                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 155                 spin_unlock(&all_mddevs_lock);                          \
 156                 if (mddev) mddev_put(mddev);                            \
 157                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 158                 tmp != &all_mddevs;});                                  \
 159              ({ spin_lock(&all_mddevs_lock);                            \
 160                 tmp = tmp->next;})                                      \
 161                 )
 162
 163
 164 static int md_fail_request (request_queue_t *q, struct bio *bio)
 165 {
 166         bio_io_error(bio, bio->bi_size);
 167         return 0;
 168 }
 169
 170 static inline mddev_t *mddev_get(mddev_t *mddev)
 171 {
 172         atomic_inc(&mddev->active);
 173         return mddev;
 174 }
 175
 176 static void mddev_put(mddev_t *mddev)
 177 {
 178         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 179                 return;
 180         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 181                 list_del(&mddev->all_mddevs);
 182                 blk_put_queue(mddev->queue);
 183                 kfree(mddev);
 184         }
 185         spin_unlock(&all_mddevs_lock);
 186 }
 187
 188 static mddev_t * mddev_find(dev_t unit)
 189 {
 190         mddev_t *mddev, *new = NULL;
 191
 192  retry:
 193         spin_lock(&all_mddevs_lock);
 194         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 195                 if (mddev->unit == unit) {
 196                         mddev_get(mddev);
 197                         spin_unlock(&all_mddevs_lock);
 198                         kfree(new);
 199                         return mddev;
 200                 }
 201
 202         if (new) {
 203                 list_add(&new->all_mddevs, &all_mddevs);
 204                 spin_unlock(&all_mddevs_lock);
 205                 return new;
 206         }
 207         spin_unlock(&all_mddevs_lock);
 208
 209         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 210         if (!new)
 211                 return NULL;
 212
 213         memset(new, 0, sizeof(*new));
 214
 215         new->unit = unit;
 216         if (MAJOR(unit) == MD_MAJOR)
 217                 new->md_minor = MINOR(unit);
 218         else
 219                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 220
 221         init_MUTEX(&new->reconfig_sem);
 222         INIT_LIST_HEAD(&new->disks);
 223         INIT_LIST_HEAD(&new->all_mddevs);
 224         init_timer(&new->safemode_timer);
 225         atomic_set(&new->active, 1);
 226         spin_lock_init(&new->write_lock);
 227         init_waitqueue_head(&new->sb_wait);
 228
 229         new->queue = blk_alloc_queue(GFP_KERNEL);
 230         if (!new->queue) {
 231                 kfree(new);
 232                 return NULL;
 233         }
 234
 235         blk_queue_make_request(new->queue, md_fail_request);
 236
 237         goto retry;
 238 }
 239
 240 static inline int mddev_lock(mddev_t * mddev)
 241 {
 242         return down_interruptible(&mddev->reconfig_sem);
 243 }
 244
 245 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 246 {
 247         down(&mddev->reconfig_sem);
 248 }
 249
 250 static inline int mddev_trylock(mddev_t * mddev)
 251 {
 252         return down_trylock(&mddev->reconfig_sem);
 253 }
 254
 255 static inline void mddev_unlock(mddev_t * mddev)
 256 {
 257         up(&mddev->reconfig_sem);
 258
 259         if (mddev->thread)
 260                 md_wakeup_thread(mddev->thread);
 261 }
 262
 263 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 264 {
 265         mdk_rdev_t * rdev;
 266         struct list_head *tmp;
 267
 268         ITERATE_RDEV(mddev,rdev,tmp) {
 269                 if (rdev->desc_nr == nr)
 270                         return rdev;
 271         }
 272         return NULL;
 273 }
 274
 275 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 276 {
 277         struct list_head *tmp;
 278         mdk_rdev_t *rdev;
 279
 280         ITERATE_RDEV(mddev,rdev,tmp) {
 281                 if (rdev->bdev->bd_dev == dev)
 282                         return rdev;
 283         }
 284         return NULL;
 285 }
 286
 287 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 288 {
 289         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 290         return MD_NEW_SIZE_BLOCKS(size);
 291 }
 292
 293 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 294 {
 295         sector_t size;
 296
 297         size = rdev->sb_offset;
 298
 299         if (chunk_size)
 300                 size &= ~((sector_t)chunk_size/1024 - 1);
 301         return size;
 302 }
 303
 304 static int alloc_disk_sb(mdk_rdev_t * rdev)
 305 {
 306         if (rdev->sb_page)
 307                 MD_BUG();
 308
 309         rdev->sb_page = alloc_page(GFP_KERNEL);
 310         if (!rdev->sb_page) {
 311                 printk(KERN_ALERT "md: out of memory.\n");
 312                 return -EINVAL;
 313         }
 314
 315         return 0;
 316 }
 317
 318 static void free_disk_sb(mdk_rdev_t * rdev)
 319 {
 320         if (rdev->sb_page) {
 321                 page_cache_release(rdev->sb_page);
 322                 rdev->sb_loaded = 0;
 323                 rdev->sb_page = NULL;
 324                 rdev->sb_offset = 0;
 325                 rdev->size = 0;
 326         }
 327 }
 328
 329
 330 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 331 {
 332         mdk_rdev_t *rdev = bio->bi_private;
 333         if (bio->bi_size)
 334                 return 1;
 335
 336         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 337                 md_error(rdev->mddev, rdev);
 338
 339         if (atomic_dec_and_test(&rdev->mddev->pending_writes))
 340                 wake_up(&rdev->mddev->sb_wait);
 341         bio_put(bio);
 342         return 0;
 343 }
 344
 345 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 346                    sector_t sector, int size, struct page *page)
 347 {
 348         /* write first size bytes of page to sector of rdev
 349          * Increment mddev->pending_writes before returning
 350          * and decrement it on completion, waking up sb_wait
 351          * if zero is reached.
 352          * If an error occurred, call md_error
 353          */
 354         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 355
 356         bio->bi_bdev = rdev->bdev;
 357         bio->bi_sector = sector;
 358         bio_add_page(bio, page, size, 0);
 359         bio->bi_private = rdev;
 360         bio->bi_end_io = super_written;
 361         atomic_inc(&mddev->pending_writes);
 362         submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
 363 }
 364
 365 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 366 {
 367         if (bio->bi_size)
 368                 return 1;
 369
 370         complete((struct completion*)bio->bi_private);
 371         return 0;
 372 }
 373
 374 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 375                    struct page *page, int rw)
 376 {
 377         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 378         struct completion event;
 379         int ret;
 380
 381         rw |= (1 << BIO_RW_SYNC);
 382
 383         bio->bi_bdev = bdev;
 384         bio->bi_sector = sector;
 385         bio_add_page(bio, page, size, 0);
 386         init_completion(&event);
 387         bio->bi_private = &event;
 388         bio->bi_end_io = bi_complete;
 389         submit_bio(rw, bio);
 390         wait_for_completion(&event);
 391
 392         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 393         bio_put(bio);
 394         return ret;
 395 }
 396
 397 static int read_disk_sb(mdk_rdev_t * rdev)
 398 {
 399         char b[BDEVNAME_SIZE];
 400         if (!rdev->sb_page) {
 401                 MD_BUG();
 402                 return -EINVAL;
 403         }
 404         if (rdev->sb_loaded)
 405                 return 0;
 406
 407
 408         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
 409                 goto fail;
 410         rdev->sb_loaded = 1;
 411         return 0;
 412
 413 fail:
 414         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 415                 bdevname(rdev->bdev,b));
 416         return -EINVAL;
 417 }
 418
 419 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 420 {
 421         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 422                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 423                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 424                 (sb1->set_uuid3 == sb2->set_uuid3))
 425
 426                 return 1;
 427
 428         return 0;
 429 }
 430
 431
 432 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 433 {
 434         int ret;
 435         mdp_super_t *tmp1, *tmp2;
 436
 437         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 438         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 439
 440         if (!tmp1 || !tmp2) {
 441                 ret = 0;
 442                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 443                 goto abort;
 444         }
 445
 446         *tmp1 = *sb1;
 447         *tmp2 = *sb2;
 448
 449         /*
 450          * nr_disks is not constant
 451          */
 452         tmp1->nr_disks = 0;
 453         tmp2->nr_disks = 0;
 454
 455         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 456                 ret = 0;
 457         else
 458                 ret = 1;
 459
 460 abort:
 461         kfree(tmp1);
 462         kfree(tmp2);
 463         return ret;
 464 }
 465
 466 static unsigned int calc_sb_csum(mdp_super_t * sb)
 467 {
 468         unsigned int disk_csum, csum;
 469
 470         disk_csum = sb->sb_csum;
 471         sb->sb_csum = 0;
 472         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 473         sb->sb_csum = disk_csum;
 474         return csum;
 475 }
 476
 477
 478 /*
 479  * Handle superblock details.
 480  * We want to be able to handle multiple superblock formats
 481  * so we have a common interface to them all, and an array of
 482  * different handlers.
 483  * We rely on user-space to write the initial superblock, and support
 484  * reading and updating of superblocks.
 485  * Interface methods are:
 486  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 487  *      loads and validates a superblock on dev.
 488  *      if refdev != NULL, compare superblocks on both devices
 489  *    Return:
 490  *      0 - dev has a superblock that is compatible with refdev
 491  *      1 - dev has a superblock that is compatible and newer than refdev
 492  *          so dev should be used as the refdev in future
 493  *     -EINVAL superblock incompatible or invalid
 494  *     -othererror e.g. -EIO
 495  *
 496  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 497  *      Verify that dev is acceptable into mddev.
 498  *       The first time, mddev->raid_disks will be 0, and data from
 499  *       dev should be merged in.  Subsequent calls check that dev
 500  *       is new enough.  Return 0 or -EINVAL
 501  *
 502  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 503  *     Update the superblock for rdev with data in mddev
 504  *     This does not write to disc.
 505  *
 506  */
 507
 508 struct super_type  {
 509         char            *name;
 510         struct module   *owner;
 511         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 512         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 513         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 514 };
 515
 516 /*
 517  * load_super for 0.90.0
 518  */
 519 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 520 {
 521         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 522         mdp_super_t *sb;
 523         int ret;
 524         sector_t sb_offset;
 525
 526         /*
 527          * Calculate the position of the superblock,
 528          * it's at the end of the disk.
 529          *
 530          * It also happens to be a multiple of 4Kb.
 531          */
 532         sb_offset = calc_dev_sboffset(rdev->bdev);
 533         rdev->sb_offset = sb_offset;
 534
 535         ret = read_disk_sb(rdev);
 536         if (ret) return ret;
 537
 538         ret = -EINVAL;
 539
 540         bdevname(rdev->bdev, b);
 541         sb = (mdp_super_t*)page_address(rdev->sb_page);
 542
 543         if (sb->md_magic != MD_SB_MAGIC) {
 544                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 545                        b);
 546                 goto abort;
 547         }
 548
 549         if (sb->major_version != 0 ||
 550             sb->minor_version != 90) {
 551                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 552                         sb->major_version, sb->minor_version,
 553                         b);
 554                 goto abort;
 555         }
 556
 557         if (sb->raid_disks <= 0)
 558                 goto abort;
 559
 560         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 561                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 562                         b);
 563                 goto abort;
 564         }
 565
 566         rdev->preferred_minor = sb->md_minor;
 567         rdev->data_offset = 0;
 568
 569         if (sb->level == LEVEL_MULTIPATH)
 570                 rdev->desc_nr = -1;
 571         else
 572                 rdev->desc_nr = sb->this_disk.number;
 573
 574         if (refdev == 0)
 575                 ret = 1;
 576         else {
 577                 __u64 ev1, ev2;
 578                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 579                 if (!uuid_equal(refsb, sb)) {
 580                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 581                                 b, bdevname(refdev->bdev,b2));
 582                         goto abort;
 583                 }
 584                 if (!sb_equal(refsb, sb)) {
 585                         printk(KERN_WARNING "md: %s has same UUID"
 586                                " but different superblock to %s\n",
 587                                b, bdevname(refdev->bdev, b2));
 588                         goto abort;
 589                 }
 590                 ev1 = md_event(sb);
 591                 ev2 = md_event(refsb);
 592                 if (ev1 > ev2)
 593                         ret = 1;
 594                 else
 595                         ret = 0;
 596         }
 597         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 598
 599  abort:
 600         return ret;
 601 }
 602
 603 /*
 604  * validate_super for 0.90.0
 605  */
 606 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 607 {
 608         mdp_disk_t *desc;
 609         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 610
 611         rdev->raid_disk = -1;
 612         rdev->in_sync = 0;
 613         if (mddev->raid_disks == 0) {
 614                 mddev->major_version = 0;
 615                 mddev->minor_version = sb->minor_version;
 616                 mddev->patch_version = sb->patch_version;
 617                 mddev->persistent = ! sb->not_persistent;
 618                 mddev->chunk_size = sb->chunk_size;
 619                 mddev->ctime = sb->ctime;
 620                 mddev->utime = sb->utime;
 621                 mddev->level = sb->level;
 622                 mddev->layout = sb->layout;
 623                 mddev->raid_disks = sb->raid_disks;
 624                 mddev->size = sb->size;
 625                 mddev->events = md_event(sb);
 626                 mddev->bitmap_offset = 0;
 627
 628                 if (sb->state & (1<<MD_SB_CLEAN))
 629                         mddev->recovery_cp = MaxSector;
 630                 else {
 631                         if (sb->events_hi == sb->cp_events_hi &&
 632                                 sb->events_lo == sb->cp_events_lo) {
 633                                 mddev->recovery_cp = sb->recovery_cp;
 634                         } else
 635                                 mddev->recovery_cp = 0;
 636                 }
 637
 638                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 639                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 640                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 641                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 642
 643                 mddev->max_disks = MD_SB_DISKS;
 644
 645                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 646                     mddev->bitmap_file == NULL) {
 647                         if (mddev->level != 1) {
 648                                 /* FIXME use a better test */
 649                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
 650                                 return -EINVAL;
 651                         }
 652                         mddev->bitmap_offset = (MD_SB_BYTES >> 9);
 653                 }
 654
 655         } else if (mddev->pers == NULL) {
 656                 /* Insist on good event counter while assembling */
 657                 __u64 ev1 = md_event(sb);
 658                 ++ev1;
 659                 if (ev1 < mddev->events)
 660                         return -EINVAL;
 661         } else if (mddev->bitmap) {
 662                 /* if adding to array with a bitmap, then we can accept an
 663                  * older device ... but not too old.
 664                  */
 665                 __u64 ev1 = md_event(sb);
 666                 if (ev1 < mddev->bitmap->events_cleared)
 667                         return 0;
 668         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 669                 return 0;
 670
 671         if (mddev->level != LEVEL_MULTIPATH) {
 672                 rdev->faulty = 0;
 673                 desc = sb->disks + rdev->desc_nr;
 674
 675                 if (desc->state & (1<<MD_DISK_FAULTY))
 676                         rdev->faulty = 1;
 677                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 678                          desc->raid_disk < mddev->raid_disks) {
 679                         rdev->in_sync = 1;
 680                         rdev->raid_disk = desc->raid_disk;
 681                 }
 682         } else /* MULTIPATH are always insync */
 683                 rdev->in_sync = 1;
 684         return 0;
 685 }
 686
 687 /*
 688  * sync_super for 0.90.0
 689  */
 690 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 691 {
 692         mdp_super_t *sb;
 693         struct list_head *tmp;
 694         mdk_rdev_t *rdev2;
 695         int next_spare = mddev->raid_disks;
 696
 697         /* make rdev->sb match mddev data..
 698          *
 699          * 1/ zero out disks
 700          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 701          * 3/ any empty disks < next_spare become removed
 702          *
 703          * disks[0] gets initialised to REMOVED because
 704          * we cannot be sure from other fields if it has
 705          * been initialised or not.
 706          */
 707         int i;
 708         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 709
 710         sb = (mdp_super_t*)page_address(rdev->sb_page);
 711
 712         memset(sb, 0, sizeof(*sb));
 713
 714         sb->md_magic = MD_SB_MAGIC;
 715         sb->major_version = mddev->major_version;
 716         sb->minor_version = mddev->minor_version;
 717         sb->patch_version = mddev->patch_version;
 718         sb->gvalid_words  = 0; /* ignored */
 719         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 720         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 721         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 722         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 723
 724         sb->ctime = mddev->ctime;
 725         sb->level = mddev->level;
 726         sb->size  = mddev->size;
 727         sb->raid_disks = mddev->raid_disks;
 728         sb->md_minor = mddev->md_minor;
 729         sb->not_persistent = !mddev->persistent;
 730         sb->utime = mddev->utime;
 731         sb->state = 0;
 732         sb->events_hi = (mddev->events>>32);
 733         sb->events_lo = (u32)mddev->events;
 734
 735         if (mddev->in_sync)
 736         {
 737                 sb->recovery_cp = mddev->recovery_cp;
 738                 sb->cp_events_hi = (mddev->events>>32);
 739                 sb->cp_events_lo = (u32)mddev->events;
 740                 if (mddev->recovery_cp == MaxSector)
 741                         sb->state = (1<< MD_SB_CLEAN);
 742         } else
 743                 sb->recovery_cp = 0;
 744
 745         sb->layout = mddev->layout;
 746         sb->chunk_size = mddev->chunk_size;
 747
 748         if (mddev->bitmap && mddev->bitmap_file == NULL)
 749                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 750
 751         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 752         ITERATE_RDEV(mddev,rdev2,tmp) {
 753                 mdp_disk_t *d;
 754                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 755                         rdev2->desc_nr = rdev2->raid_disk;
 756                 else
 757                         rdev2->desc_nr = next_spare++;
 758                 d = &sb->disks[rdev2->desc_nr];
 759                 nr_disks++;
 760                 d->number = rdev2->desc_nr;
 761                 d->major = MAJOR(rdev2->bdev->bd_dev);
 762                 d->minor = MINOR(rdev2->bdev->bd_dev);
 763                 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
 764                         d->raid_disk = rdev2->raid_disk;
 765                 else
 766                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 767                 if (rdev2->faulty) {
 768                         d->state = (1<<MD_DISK_FAULTY);
 769                         failed++;
 770                 } else if (rdev2->in_sync) {
 771                         d->state = (1<<MD_DISK_ACTIVE);
 772                         d->state |= (1<<MD_DISK_SYNC);
 773                         active++;
 774                         working++;
 775                 } else {
 776                         d->state = 0;
 777                         spare++;
 778                         working++;
 779                 }
 780         }
 781
 782         /* now set the "removed" and "faulty" bits on any missing devices */
 783         for (i=0 ; i < mddev->raid_disks ; i++) {
 784                 mdp_disk_t *d = &sb->disks[i];
 785                 if (d->state == 0 && d->number == 0) {
 786                         d->number = i;
 787                         d->raid_disk = i;
 788                         d->state = (1<<MD_DISK_REMOVED);
 789                         d->state |= (1<<MD_DISK_FAULTY);
 790                         failed++;
 791                 }
 792         }
 793         sb->nr_disks = nr_disks;
 794         sb->active_disks = active;
 795         sb->working_disks = working;
 796         sb->failed_disks = failed;
 797         sb->spare_disks = spare;
 798
 799         sb->this_disk = sb->disks[rdev->desc_nr];
 800         sb->sb_csum = calc_sb_csum(sb);
 801 }
 802
 803 /*
 804  * version 1 superblock
 805  */
 806
 807 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 808 {
 809         unsigned int disk_csum, csum;
 810         unsigned long long newcsum;
 811         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 812         unsigned int *isuper = (unsigned int*)sb;
 813         int i;
 814
 815         disk_csum = sb->sb_csum;
 816         sb->sb_csum = 0;
 817         newcsum = 0;
 818         for (i=0; size>=4; size -= 4 )
 819                 newcsum += le32_to_cpu(*isuper++);
 820
 821         if (size == 2)
 822                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 823
 824         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 825         sb->sb_csum = disk_csum;
 826         return cpu_to_le32(csum);
 827 }
 828
 829 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 830 {
 831         struct mdp_superblock_1 *sb;
 832         int ret;
 833         sector_t sb_offset;
 834         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 835
 836         /*
 837          * Calculate the position of the superblock.
 838          * It is always aligned to a 4K boundary and
 839          * depeding on minor_version, it can be:
 840          * 0: At least 8K, but less than 12K, from end of device
 841          * 1: At start of device
 842          * 2: 4K from start of device.
 843          */
 844         switch(minor_version) {
 845         case 0:
 846                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 847                 sb_offset -= 8*2;
 848                 sb_offset &= ~(sector_t)(4*2-1);
 849                 /* convert from sectors to K */
 850                 sb_offset /= 2;
 851                 break;
 852         case 1:
 853                 sb_offset = 0;
 854                 break;
 855         case 2:
 856                 sb_offset = 4;
 857                 break;
 858         default:
 859                 return -EINVAL;
 860         }
 861         rdev->sb_offset = sb_offset;
 862
 863         ret = read_disk_sb(rdev);
 864         if (ret) return ret;
 865
 866
 867         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 868
 869         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 870             sb->major_version != cpu_to_le32(1) ||
 871             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 872             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 873             sb->feature_map != 0)
 874                 return -EINVAL;
 875
 876         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 877                 printk("md: invalid superblock checksum on %s\n",
 878                         bdevname(rdev->bdev,b));
 879                 return -EINVAL;
 880         }
 881         if (le64_to_cpu(sb->data_size) < 10) {
 882                 printk("md: data_size too small on %s\n",
 883                        bdevname(rdev->bdev,b));
 884                 return -EINVAL;
 885         }
 886         rdev->preferred_minor = 0xffff;
 887         rdev->data_offset = le64_to_cpu(sb->data_offset);
 888
 889         if (refdev == 0)
 890                 return 1;
 891         else {
 892                 __u64 ev1, ev2;
 893                 struct mdp_superblock_1 *refsb =
 894                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 895
 896                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 897                     sb->level != refsb->level ||
 898                     sb->layout != refsb->layout ||
 899                     sb->chunksize != refsb->chunksize) {
 900                         printk(KERN_WARNING "md: %s has strangely different"
 901                                 " superblock to %s\n",
 902                                 bdevname(rdev->bdev,b),
 903                                 bdevname(refdev->bdev,b2));
 904                         return -EINVAL;
 905                 }
 906                 ev1 = le64_to_cpu(sb->events);
 907                 ev2 = le64_to_cpu(refsb->events);
 908
 909                 if (ev1 > ev2)
 910                         return 1;
 911         }
 912         if (minor_version)
 913                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 914         else
 915                 rdev->size = rdev->sb_offset;
 916         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 917                 return -EINVAL;
 918         rdev->size = le64_to_cpu(sb->data_size)/2;
 919         if (le32_to_cpu(sb->chunksize))
 920                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 921         return 0;
 922 }
 923
 924 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 925 {
 926         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 927
 928         rdev->raid_disk = -1;
 929         rdev->in_sync = 0;
 930         if (mddev->raid_disks == 0) {
 931                 mddev->major_version = 1;
 932                 mddev->patch_version = 0;
 933                 mddev->persistent = 1;
 934                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 935                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 936                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 937                 mddev->level = le32_to_cpu(sb->level);
 938                 mddev->layout = le32_to_cpu(sb->layout);
 939                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 940                 mddev->size = le64_to_cpu(sb->size)/2;
 941                 mddev->events = le64_to_cpu(sb->events);
 942                 mddev->bitmap_offset = 0;
 943
 944                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 945                 memcpy(mddev->uuid, sb->set_uuid, 16);
 946
 947                 mddev->max_disks =  (4096-256)/2;
 948
 949                 if ((le32_to_cpu(sb->feature_map) & 1) &&
 950                     mddev->bitmap_file == NULL ) {
 951                         if (mddev->level != 1) {
 952                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
 953                                 return -EINVAL;
 954                         }
 955                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
 956                 }
 957         } else if (mddev->pers == NULL) {
 958                 /* Insist of good event counter while assembling */
 959                 __u64 ev1 = le64_to_cpu(sb->events);
 960                 ++ev1;
 961                 if (ev1 < mddev->events)
 962                         return -EINVAL;
 963         } else if (mddev->bitmap) {
 964                 /* If adding to array with a bitmap, then we can accept an
 965                  * older device, but not too old.
 966                  */
 967                 __u64 ev1 = le64_to_cpu(sb->events);
 968                 if (ev1 < mddev->bitmap->events_cleared)
 969                         return 0;
 970         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 971                 return 0;
 972
 973         if (mddev->level != LEVEL_MULTIPATH) {
 974                 int role;
 975                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
 976                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
 977                 switch(role) {
 978                 case 0xffff: /* spare */
 979                         rdev->faulty = 0;
 980                         break;
 981                 case 0xfffe: /* faulty */
 982                         rdev->faulty = 1;
 983                         break;
 984                 default:
 985                         rdev->in_sync = 1;
 986                         rdev->faulty = 0;
 987                         rdev->raid_disk = role;
 988                         break;
 989                 }
 990         } else /* MULTIPATH are always insync */
 991                 rdev->in_sync = 1;
 992
 993         return 0;
 994 }
 995
 996 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 997 {
 998         struct mdp_superblock_1 *sb;
 999         struct list_head *tmp;
1000         mdk_rdev_t *rdev2;
1001         int max_dev, i;
1002         /* make rdev->sb match mddev and rdev data. */
1003
1004         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1005
1006         sb->feature_map = 0;
1007         sb->pad0 = 0;
1008         memset(sb->pad1, 0, sizeof(sb->pad1));
1009         memset(sb->pad2, 0, sizeof(sb->pad2));
1010         memset(sb->pad3, 0, sizeof(sb->pad3));
1011
1012         sb->utime = cpu_to_le64((__u64)mddev->utime);
1013         sb->events = cpu_to_le64(mddev->events);
1014         if (mddev->in_sync)
1015                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1016         else
1017                 sb->resync_offset = cpu_to_le64(0);
1018
1019         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1020                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1021                 sb->feature_map = cpu_to_le32(1);
1022         }
1023
1024         max_dev = 0;
1025         ITERATE_RDEV(mddev,rdev2,tmp)
1026                 if (rdev2->desc_nr+1 > max_dev)
1027                         max_dev = rdev2->desc_nr+1;
1028
1029         sb->max_dev = cpu_to_le32(max_dev);
1030         for (i=0; i<max_dev;i++)
1031                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1032
1033         ITERATE_RDEV(mddev,rdev2,tmp) {
1034                 i = rdev2->desc_nr;
1035                 if (rdev2->faulty)
1036                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1037                 else if (rdev2->in_sync)
1038                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1039                 else
1040                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1041         }
1042
1043         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1044         sb->sb_csum = calc_sb_1_csum(sb);
1045 }
1046
1047
1048 static struct super_type super_types[] = {
1049         [0] = {
1050                 .name   = "0.90.0",
1051                 .owner  = THIS_MODULE,
1052                 .load_super     = super_90_load,
1053                 .validate_super = super_90_validate,
1054                 .sync_super     = super_90_sync,
1055         },
1056         [1] = {
1057                 .name   = "md-1",
1058                 .owner  = THIS_MODULE,
1059                 .load_super     = super_1_load,
1060                 .validate_super = super_1_validate,
1061                 .sync_super     = super_1_sync,
1062         },
1063 };
1064
1065 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1066 {
1067         struct list_head *tmp;
1068         mdk_rdev_t *rdev;
1069
1070         ITERATE_RDEV(mddev,rdev,tmp)
1071                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1072                         return rdev;
1073
1074         return NULL;
1075 }
1076
1077 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1078 {
1079         struct list_head *tmp;
1080         mdk_rdev_t *rdev;
1081
1082         ITERATE_RDEV(mddev1,rdev,tmp)
1083                 if (match_dev_unit(mddev2, rdev))
1084                         return 1;
1085
1086         return 0;
1087 }
1088
1089 static LIST_HEAD(pending_raid_disks);
1090
1091 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1092 {
1093         mdk_rdev_t *same_pdev;
1094         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1095
1096         if (rdev->mddev) {
1097                 MD_BUG();
1098                 return -EINVAL;
1099         }
1100         same_pdev = match_dev_unit(mddev, rdev);
1101         if (same_pdev)
1102                 printk(KERN_WARNING
1103                         "%s: WARNING: %s appears to be on the same physical"
1104                         " disk as %s. True\n     protection against single-disk"
1105                         " failure might be compromised.\n",
1106                         mdname(mddev), bdevname(rdev->bdev,b),
1107                         bdevname(same_pdev->bdev,b2));
1108
1109         /* Verify rdev->desc_nr is unique.
1110          * If it is -1, assign a free number, else
1111          * check number is not in use
1112          */
1113         if (rdev->desc_nr < 0) {
1114                 int choice = 0;
1115                 if (mddev->pers) choice = mddev->raid_disks;
1116                 while (find_rdev_nr(mddev, choice))
1117                         choice++;
1118                 rdev->desc_nr = choice;
1119         } else {
1120                 if (find_rdev_nr(mddev, rdev->desc_nr))
1121                         return -EBUSY;
1122         }
1123
1124         list_add(&rdev->same_set, &mddev->disks);
1125         rdev->mddev = mddev;
1126         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1127         return 0;
1128 }
1129
1130 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1131 {
1132         char b[BDEVNAME_SIZE];
1133         if (!rdev->mddev) {
1134                 MD_BUG();
1135                 return;
1136         }
1137         list_del_init(&rdev->same_set);
1138         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1139         rdev->mddev = NULL;
1140 }
1141
1142 /*
1143  * prevent the device from being mounted, repartitioned or
1144  * otherwise reused by a RAID array (or any other kernel
1145  * subsystem), by bd_claiming the device.
1146  */
1147 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1148 {
1149         int err = 0;
1150         struct block_device *bdev;
1151         char b[BDEVNAME_SIZE];
1152
1153         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1154         if (IS_ERR(bdev)) {
1155                 printk(KERN_ERR "md: could not open %s.\n",
1156                         __bdevname(dev, b));
1157                 return PTR_ERR(bdev);
1158         }
1159         err = bd_claim(bdev, rdev);
1160         if (err) {
1161                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1162                         bdevname(bdev, b));
1163                 blkdev_put(bdev);
1164                 return err;
1165         }
1166         rdev->bdev = bdev;
1167         return err;
1168 }
1169
1170 static void unlock_rdev(mdk_rdev_t *rdev)
1171 {
1172         struct block_device *bdev = rdev->bdev;
1173         rdev->bdev = NULL;
1174         if (!bdev)
1175                 MD_BUG();
1176         bd_release(bdev);
1177         blkdev_put(bdev);
1178 }
1179
1180 void md_autodetect_dev(dev_t dev);
1181
1182 static void export_rdev(mdk_rdev_t * rdev)
1183 {
1184         char b[BDEVNAME_SIZE];
1185         printk(KERN_INFO "md: export_rdev(%s)\n",
1186                 bdevname(rdev->bdev,b));
1187         if (rdev->mddev)
1188                 MD_BUG();
1189         free_disk_sb(rdev);
1190         list_del_init(&rdev->same_set);
1191 #ifndef MODULE
1192         md_autodetect_dev(rdev->bdev->bd_dev);
1193 #endif
1194         unlock_rdev(rdev);
1195         kfree(rdev);
1196 }
1197
1198 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1199 {
1200         unbind_rdev_from_array(rdev);
1201         export_rdev(rdev);
1202 }
1203
1204 static void export_array(mddev_t *mddev)
1205 {
1206         struct list_head *tmp;
1207         mdk_rdev_t *rdev;
1208
1209         ITERATE_RDEV(mddev,rdev,tmp) {
1210                 if (!rdev->mddev) {
1211                         MD_BUG();
1212                         continue;
1213                 }
1214                 kick_rdev_from_array(rdev);
1215         }
1216         if (!list_empty(&mddev->disks))
1217                 MD_BUG();
1218         mddev->raid_disks = 0;
1219         mddev->major_version = 0;
1220 }
1221
1222 static void print_desc(mdp_disk_t *desc)
1223 {
1224         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1225                 desc->major,desc->minor,desc->raid_disk,desc->state);
1226 }
1227
1228 static void print_sb(mdp_super_t *sb)
1229 {
1230         int i;
1231
1232         printk(KERN_INFO
1233                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1234                 sb->major_version, sb->minor_version, sb->patch_version,
1235                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1236                 sb->ctime);
1237         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1238                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1239                 sb->md_minor, sb->layout, sb->chunk_size);
1240         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1241                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1242                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1243                 sb->failed_disks, sb->spare_disks,
1244                 sb->sb_csum, (unsigned long)sb->events_lo);
1245
1246         printk(KERN_INFO);
1247         for (i = 0; i < MD_SB_DISKS; i++) {
1248                 mdp_disk_t *desc;
1249
1250                 desc = sb->disks + i;
1251                 if (desc->number || desc->major || desc->minor ||
1252                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1253                         printk("     D %2d: ", i);
1254                         print_desc(desc);
1255                 }
1256         }
1257         printk(KERN_INFO "md:     THIS: ");
1258         print_desc(&sb->this_disk);
1259
1260 }
1261
1262 static void print_rdev(mdk_rdev_t *rdev)
1263 {
1264         char b[BDEVNAME_SIZE];
1265         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1266                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1267                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1268         if (rdev->sb_loaded) {
1269                 printk(KERN_INFO "md: rdev superblock:\n");
1270                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1271         } else
1272                 printk(KERN_INFO "md: no rdev superblock!\n");
1273 }
1274
1275 void md_print_devices(void)
1276 {
1277         struct list_head *tmp, *tmp2;
1278         mdk_rdev_t *rdev;
1279         mddev_t *mddev;
1280         char b[BDEVNAME_SIZE];
1281
1282         printk("\n");
1283         printk("md:     **********************************\n");
1284         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1285         printk("md:     **********************************\n");
1286         ITERATE_MDDEV(mddev,tmp) {
1287
1288                 if (mddev->bitmap)
1289                         bitmap_print_sb(mddev->bitmap);
1290                 else
1291                         printk("%s: ", mdname(mddev));
1292                 ITERATE_RDEV(mddev,rdev,tmp2)
1293                         printk("<%s>", bdevname(rdev->bdev,b));
1294                 printk("\n");
1295
1296                 ITERATE_RDEV(mddev,rdev,tmp2)
1297                         print_rdev(rdev);
1298         }
1299         printk("md:     **********************************\n");
1300         printk("\n");
1301 }
1302
1303
1304 static void sync_sbs(mddev_t * mddev)
1305 {
1306         mdk_rdev_t *rdev;
1307         struct list_head *tmp;
1308
1309         ITERATE_RDEV(mddev,rdev,tmp) {
1310                 super_types[mddev->major_version].
1311                         sync_super(mddev, rdev);
1312                 rdev->sb_loaded = 1;
1313         }
1314 }
1315
1316 static void md_update_sb(mddev_t * mddev)
1317 {
1318         int err;
1319         struct list_head *tmp;
1320         mdk_rdev_t *rdev;
1321         int sync_req;
1322
1323 repeat:
1324         spin_lock(&mddev->write_lock);
1325         sync_req = mddev->in_sync;
1326         mddev->utime = get_seconds();
1327         mddev->events ++;
1328
1329         if (!mddev->events) {
1330                 /*
1331                  * oops, this 64-bit counter should never wrap.
1332                  * Either we are in around ~1 trillion A.C., assuming
1333                  * 1 reboot per second, or we have a bug:
1334                  */
1335                 MD_BUG();
1336                 mddev->events --;
1337         }
1338         mddev->sb_dirty = 2;
1339         sync_sbs(mddev);
1340
1341         /*
1342          * do not write anything to disk if using
1343          * nonpersistent superblocks
1344          */
1345         if (!mddev->persistent) {
1346                 mddev->sb_dirty = 0;
1347                 spin_unlock(&mddev->write_lock);
1348                 wake_up(&mddev->sb_wait);
1349                 return;
1350         }
1351         spin_unlock(&mddev->write_lock);
1352
1353         dprintk(KERN_INFO
1354                 "md: updating %s RAID superblock on device (in sync %d)\n",
1355                 mdname(mddev),mddev->in_sync);
1356
1357         err = bitmap_update_sb(mddev->bitmap);
1358         ITERATE_RDEV(mddev,rdev,tmp) {
1359                 char b[BDEVNAME_SIZE];
1360                 dprintk(KERN_INFO "md: ");
1361                 if (rdev->faulty)
1362                         dprintk("(skipping faulty ");
1363
1364                 dprintk("%s ", bdevname(rdev->bdev,b));
1365                 if (!rdev->faulty) {
1366                         md_super_write(mddev,rdev,
1367                                        rdev->sb_offset<<1, MD_SB_BYTES,
1368                                        rdev->sb_page);
1369                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1370                                 bdevname(rdev->bdev,b),
1371                                 (unsigned long long)rdev->sb_offset);
1372
1373                 } else
1374                         dprintk(")\n");
1375                 if (mddev->level == LEVEL_MULTIPATH)
1376                         /* only need to write one superblock... */
1377                         break;
1378         }
1379         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1380         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1381
1382         spin_lock(&mddev->write_lock);
1383         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1384                 /* have to write it out again */
1385                 spin_unlock(&mddev->write_lock);
1386                 goto repeat;
1387         }
1388         mddev->sb_dirty = 0;
1389         spin_unlock(&mddev->write_lock);
1390         wake_up(&mddev->sb_wait);
1391
1392 }
1393
1394 /*
1395  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1396  *
1397  * mark the device faulty if:
1398  *
1399  *   - the device is nonexistent (zero size)
1400  *   - the device has no valid superblock
1401  *
1402  * a faulty rdev _never_ has rdev->sb set.
1403  */
1404 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1405 {
1406         char b[BDEVNAME_SIZE];
1407         int err;
1408         mdk_rdev_t *rdev;
1409         sector_t size;
1410
1411         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1412         if (!rdev) {
1413                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1414                 return ERR_PTR(-ENOMEM);
1415         }
1416         memset(rdev, 0, sizeof(*rdev));
1417
1418         if ((err = alloc_disk_sb(rdev)))
1419                 goto abort_free;
1420
1421         err = lock_rdev(rdev, newdev);
1422         if (err)
1423                 goto abort_free;
1424
1425         rdev->desc_nr = -1;
1426         rdev->faulty = 0;
1427         rdev->in_sync = 0;
1428         rdev->data_offset = 0;
1429         atomic_set(&rdev->nr_pending, 0);
1430
1431         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1432         if (!size) {
1433                 printk(KERN_WARNING
1434                         "md: %s has zero or unknown size, marking faulty!\n",
1435                         bdevname(rdev->bdev,b));
1436                 err = -EINVAL;
1437                 goto abort_free;
1438         }
1439
1440         if (super_format >= 0) {
1441                 err = super_types[super_format].
1442                         load_super(rdev, NULL, super_minor);
1443                 if (err == -EINVAL) {
1444                         printk(KERN_WARNING
1445                                 "md: %s has invalid sb, not importing!\n",
1446                                 bdevname(rdev->bdev,b));
1447                         goto abort_free;
1448                 }
1449                 if (err < 0) {
1450                         printk(KERN_WARNING
1451                                 "md: could not read %s's sb, not importing!\n",
1452                                 bdevname(rdev->bdev,b));
1453                         goto abort_free;
1454                 }
1455         }
1456         INIT_LIST_HEAD(&rdev->same_set);
1457
1458         return rdev;
1459
1460 abort_free:
1461         if (rdev->sb_page) {
1462                 if (rdev->bdev)
1463                         unlock_rdev(rdev);
1464                 free_disk_sb(rdev);
1465         }
1466         kfree(rdev);
1467         return ERR_PTR(err);
1468 }
1469
1470 /*
1471  * Check a full RAID array for plausibility
1472  */
1473
1474
1475 static void analyze_sbs(mddev_t * mddev)
1476 {
1477         int i;
1478         struct list_head *tmp;
1479         mdk_rdev_t *rdev, *freshest;
1480         char b[BDEVNAME_SIZE];
1481
1482         freshest = NULL;
1483         ITERATE_RDEV(mddev,rdev,tmp)
1484                 switch (super_types[mddev->major_version].
1485                         load_super(rdev, freshest, mddev->minor_version)) {
1486                 case 1:
1487                         freshest = rdev;
1488                         break;
1489                 case 0:
1490                         break;
1491                 default:
1492                         printk( KERN_ERR \
1493                                 "md: fatal superblock inconsistency in %s"
1494                                 " -- removing from array\n",
1495                                 bdevname(rdev->bdev,b));
1496                         kick_rdev_from_array(rdev);
1497                 }
1498
1499
1500         super_types[mddev->major_version].
1501                 validate_super(mddev, freshest);
1502
1503         i = 0;
1504         ITERATE_RDEV(mddev,rdev,tmp) {
1505                 if (rdev != freshest)
1506                         if (super_types[mddev->major_version].
1507                             validate_super(mddev, rdev)) {
1508                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1509                                         " from array!\n",
1510                                         bdevname(rdev->bdev,b));
1511                                 kick_rdev_from_array(rdev);
1512                                 continue;
1513                         }
1514                 if (mddev->level == LEVEL_MULTIPATH) {
1515                         rdev->desc_nr = i++;
1516                         rdev->raid_disk = rdev->desc_nr;
1517                         rdev->in_sync = 1;
1518                 }
1519         }
1520
1521
1522
1523         if (mddev->recovery_cp != MaxSector &&
1524             mddev->level >= 1)
1525                 printk(KERN_ERR "md: %s: raid array is not clean"
1526                        " -- starting background reconstruction\n",
1527                        mdname(mddev));
1528
1529 }
1530
1531 int mdp_major = 0;
1532
1533 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1534 {
1535         static DECLARE_MUTEX(disks_sem);
1536         mddev_t *mddev = mddev_find(dev);
1537         struct gendisk *disk;
1538         int partitioned = (MAJOR(dev) != MD_MAJOR);
1539         int shift = partitioned ? MdpMinorShift : 0;
1540         int unit = MINOR(dev) >> shift;
1541
1542         if (!mddev)
1543                 return NULL;
1544
1545         down(&disks_sem);
1546         if (mddev->gendisk) {
1547                 up(&disks_sem);
1548                 mddev_put(mddev);
1549                 return NULL;
1550         }
1551         disk = alloc_disk(1 << shift);
1552         if (!disk) {
1553                 up(&disks_sem);
1554                 mddev_put(mddev);
1555                 return NULL;
1556         }
1557         disk->major = MAJOR(dev);
1558         disk->first_minor = unit << shift;
1559         if (partitioned) {
1560                 sprintf(disk->disk_name, "md_d%d", unit);
1561                 sprintf(disk->devfs_name, "md/d%d", unit);
1562         } else {
1563                 sprintf(disk->disk_name, "md%d", unit);
1564                 sprintf(disk->devfs_name, "md/%d", unit);
1565         }
1566         disk->fops = &md_fops;
1567         disk->private_data = mddev;
1568         disk->queue = mddev->queue;
1569         add_disk(disk);
1570         mddev->gendisk = disk;
1571         up(&disks_sem);
1572         return NULL;
1573 }
1574
1575 void md_wakeup_thread(mdk_thread_t *thread);
1576
1577 static void md_safemode_timeout(unsigned long data)
1578 {
1579         mddev_t *mddev = (mddev_t *) data;
1580
1581         mddev->safemode = 1;
1582         md_wakeup_thread(mddev->thread);
1583 }
1584
1585
1586 static int do_md_run(mddev_t * mddev)
1587 {
1588         int pnum, err;
1589         int chunk_size;
1590         struct list_head *tmp;
1591         mdk_rdev_t *rdev;
1592         struct gendisk *disk;
1593         char b[BDEVNAME_SIZE];
1594
1595         if (list_empty(&mddev->disks))
1596                 /* cannot run an array with no devices.. */
1597                 return -EINVAL;
1598
1599         if (mddev->pers)
1600                 return -EBUSY;
1601
1602         /*
1603          * Analyze all RAID superblock(s)
1604          */
1605         if (!mddev->raid_disks)
1606                 analyze_sbs(mddev);
1607
1608         chunk_size = mddev->chunk_size;
1609         pnum = level_to_pers(mddev->level);
1610
1611         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1612                 if (!chunk_size) {
1613                         /*
1614                          * 'default chunksize' in the old md code used to
1615                          * be PAGE_SIZE, baaad.
1616                          * we abort here to be on the safe side. We don't
1617                          * want to continue the bad practice.
1618                          */
1619                         printk(KERN_ERR
1620                                 "no chunksize specified, see 'man raidtab'\n");
1621                         return -EINVAL;
1622                 }
1623                 if (chunk_size > MAX_CHUNK_SIZE) {
1624                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1625                                 chunk_size, MAX_CHUNK_SIZE);
1626                         return -EINVAL;
1627                 }
1628                 /*
1629                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1630                  */
1631                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1632                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1633                         return -EINVAL;
1634                 }
1635                 if (chunk_size < PAGE_SIZE) {
1636                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1637                                 chunk_size, PAGE_SIZE);
1638                         return -EINVAL;
1639                 }
1640
1641                 /* devices must have minimum size of one chunk */
1642                 ITERATE_RDEV(mddev,rdev,tmp) {
1643                         if (rdev->faulty)
1644                                 continue;
1645                         if (rdev->size < chunk_size / 1024) {
1646                                 printk(KERN_WARNING
1647                                         "md: Dev %s smaller than chunk_size:"
1648                                         " %lluk < %dk\n",
1649                                         bdevname(rdev->bdev,b),
1650                                         (unsigned long long)rdev->size,
1651                                         chunk_size / 1024);
1652                                 return -EINVAL;
1653                         }
1654                 }
1655         }
1656
1657 #ifdef CONFIG_KMOD
1658         if (!pers[pnum])
1659         {
1660                 request_module("md-personality-%d", pnum);
1661         }
1662 #endif
1663
1664         /*
1665          * Drop all container device buffers, from now on
1666          * the only valid external interface is through the md
1667          * device.
1668          * Also find largest hardsector size
1669          */
1670         ITERATE_RDEV(mddev,rdev,tmp) {
1671                 if (rdev->faulty)
1672                         continue;
1673                 sync_blockdev(rdev->bdev);
1674                 invalidate_bdev(rdev->bdev, 0);
1675         }
1676
1677         md_probe(mddev->unit, NULL, NULL);
1678         disk = mddev->gendisk;
1679         if (!disk)
1680                 return -ENOMEM;
1681
1682         spin_lock(&pers_lock);
1683         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1684                 spin_unlock(&pers_lock);
1685                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1686                        pnum);
1687                 return -EINVAL;
1688         }
1689
1690         mddev->pers = pers[pnum];
1691         spin_unlock(&pers_lock);
1692
1693         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1694
1695         /* before we start the array running, initialise the bitmap */
1696         err = bitmap_create(mddev);
1697         if (err)
1698                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1699                         mdname(mddev), err);
1700         else
1701                 err = mddev->pers->run(mddev);
1702         if (err) {
1703                 printk(KERN_ERR "md: pers->run() failed ...\n");
1704                 module_put(mddev->pers->owner);
1705                 mddev->pers = NULL;
1706                 bitmap_destroy(mddev);
1707                 return err;
1708         }
1709         atomic_set(&mddev->writes_pending,0);
1710         mddev->safemode = 0;
1711         mddev->safemode_timer.function = md_safemode_timeout;
1712         mddev->safemode_timer.data = (unsigned long) mddev;
1713         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1714         mddev->in_sync = 1;
1715
1716         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1717
1718         if (mddev->sb_dirty)
1719                 md_update_sb(mddev);
1720
1721         set_capacity(disk, mddev->array_size<<1);
1722
1723         /* If we call blk_queue_make_request here, it will
1724          * re-initialise max_sectors etc which may have been
1725          * refined inside -> run.  So just set the bits we need to set.
1726          * Most initialisation happended when we called
1727          * blk_queue_make_request(..., md_fail_request)
1728          * earlier.
1729          */
1730         mddev->queue->queuedata = mddev;
1731         mddev->queue->make_request_fn = mddev->pers->make_request;
1732
1733         mddev->changed = 1;
1734         return 0;
1735 }
1736
1737 static int restart_array(mddev_t *mddev)
1738 {
1739         struct gendisk *disk = mddev->gendisk;
1740         int err;
1741
1742         /*
1743          * Complain if it has no devices
1744          */
1745         err = -ENXIO;
1746         if (list_empty(&mddev->disks))
1747                 goto out;
1748
1749         if (mddev->pers) {
1750                 err = -EBUSY;
1751                 if (!mddev->ro)
1752                         goto out;
1753
1754                 mddev->safemode = 0;
1755                 mddev->ro = 0;
1756                 set_disk_ro(disk, 0);
1757
1758                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1759                         mdname(mddev));
1760                 /*
1761                  * Kick recovery or resync if necessary
1762                  */
1763                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1764                 md_wakeup_thread(mddev->thread);
1765                 err = 0;
1766         } else {
1767                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1768                         mdname(mddev));
1769                 err = -EINVAL;
1770         }
1771
1772 out:
1773         return err;
1774 }
1775
1776 static int do_md_stop(mddev_t * mddev, int ro)
1777 {
1778         int err = 0;
1779         struct gendisk *disk = mddev->gendisk;
1780
1781         if (mddev->pers) {
1782                 if (atomic_read(&mddev->active)>2) {
1783                         printk("md: %s still in use.\n",mdname(mddev));
1784                         return -EBUSY;
1785                 }
1786
1787                 if (mddev->sync_thread) {
1788                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1789                         md_unregister_thread(mddev->sync_thread);
1790                         mddev->sync_thread = NULL;
1791                 }
1792
1793                 del_timer_sync(&mddev->safemode_timer);
1794
1795                 invalidate_partition(disk, 0);
1796
1797                 if (ro) {
1798                         err  = -ENXIO;
1799                         if (mddev->ro)
1800                                 goto out;
1801                         mddev->ro = 1;
1802                 } else {
1803                         bitmap_flush(mddev);
1804                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1805                         if (mddev->ro)
1806                                 set_disk_ro(disk, 0);
1807                         blk_queue_make_request(mddev->queue, md_fail_request);
1808                         mddev->pers->stop(mddev);
1809                         module_put(mddev->pers->owner);
1810                         mddev->pers = NULL;
1811                         if (mddev->ro)
1812                                 mddev->ro = 0;
1813                 }
1814                 if (!mddev->in_sync) {
1815                         /* mark array as shutdown cleanly */
1816                         mddev->in_sync = 1;
1817                         md_update_sb(mddev);
1818                 }
1819                 if (ro)
1820                         set_disk_ro(disk, 1);
1821         }
1822
1823         bitmap_destroy(mddev);
1824         if (mddev->bitmap_file) {
1825                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1826                 fput(mddev->bitmap_file);
1827                 mddev->bitmap_file = NULL;
1828         }
1829         mddev->bitmap_offset = 0;
1830
1831         /*
1832          * Free resources if final stop
1833          */
1834         if (!ro) {
1835                 struct gendisk *disk;
1836                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1837
1838                 export_array(mddev);
1839
1840                 mddev->array_size = 0;
1841                 disk = mddev->gendisk;
1842                 if (disk)
1843                         set_capacity(disk, 0);
1844                 mddev->changed = 1;
1845         } else
1846                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1847                         mdname(mddev));
1848         err = 0;
1849 out:
1850         return err;
1851 }
1852
1853 static void autorun_array(mddev_t *mddev)
1854 {
1855         mdk_rdev_t *rdev;
1856         struct list_head *tmp;
1857         int err;
1858
1859         if (list_empty(&mddev->disks))
1860                 return;
1861
1862         printk(KERN_INFO "md: running: ");
1863
1864         ITERATE_RDEV(mddev,rdev,tmp) {
1865                 char b[BDEVNAME_SIZE];
1866                 printk("<%s>", bdevname(rdev->bdev,b));
1867         }
1868         printk("\n");
1869
1870         err = do_md_run (mddev);
1871         if (err) {
1872                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1873                 do_md_stop (mddev, 0);
1874         }
1875 }
1876
1877 /*
1878  * lets try to run arrays based on all disks that have arrived
1879  * until now. (those are in pending_raid_disks)
1880  *
1881  * the method: pick the first pending disk, collect all disks with
1882  * the same UUID, remove all from the pending list and put them into
1883  * the 'same_array' list. Then order this list based on superblock
1884  * update time (freshest comes first), kick out 'old' disks and
1885  * compare superblocks. If everything's fine then run it.
1886  *
1887  * If "unit" is allocated, then bump its reference count
1888  */
1889 static void autorun_devices(int part)
1890 {
1891         struct list_head candidates;
1892         struct list_head *tmp;
1893         mdk_rdev_t *rdev0, *rdev;
1894         mddev_t *mddev;
1895         char b[BDEVNAME_SIZE];
1896
1897         printk(KERN_INFO "md: autorun ...\n");
1898         while (!list_empty(&pending_raid_disks)) {
1899                 dev_t dev;
1900                 rdev0 = list_entry(pending_raid_disks.next,
1901                                          mdk_rdev_t, same_set);
1902
1903                 printk(KERN_INFO "md: considering %s ...\n",
1904                         bdevname(rdev0->bdev,b));
1905                 INIT_LIST_HEAD(&candidates);
1906                 ITERATE_RDEV_PENDING(rdev,tmp)
1907                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1908                                 printk(KERN_INFO "md:  adding %s ...\n",
1909                                         bdevname(rdev->bdev,b));
1910                                 list_move(&rdev->same_set, &candidates);
1911                         }
1912                 /*
1913                  * now we have a set of devices, with all of them having
1914                  * mostly sane superblocks. It's time to allocate the
1915                  * mddev.
1916                  */
1917                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1918                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1919                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1920                         break;
1921                 }
1922                 if (part)
1923                         dev = MKDEV(mdp_major,
1924                                     rdev0->preferred_minor << MdpMinorShift);
1925                 else
1926                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1927
1928                 md_probe(dev, NULL, NULL);
1929                 mddev = mddev_find(dev);
1930                 if (!mddev) {
1931                         printk(KERN_ERR
1932                                 "md: cannot allocate memory for md drive.\n");
1933                         break;
1934                 }
1935                 if (mddev_lock(mddev))
1936                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1937                                mdname(mddev));
1938                 else if (mddev->raid_disks || mddev->major_version
1939                          || !list_empty(&mddev->disks)) {
1940                         printk(KERN_WARNING
1941                                 "md: %s already running, cannot run %s\n",
1942                                 mdname(mddev), bdevname(rdev0->bdev,b));
1943                         mddev_unlock(mddev);
1944                 } else {
1945                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1946                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1947                                 list_del_init(&rdev->same_set);
1948                                 if (bind_rdev_to_array(rdev, mddev))
1949                                         export_rdev(rdev);
1950                         }
1951                         autorun_array(mddev);
1952                         mddev_unlock(mddev);
1953                 }
1954                 /* on success, candidates will be empty, on error
1955                  * it won't...
1956                  */
1957                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1958                         export_rdev(rdev);
1959                 mddev_put(mddev);
1960         }
1961         printk(KERN_INFO "md: ... autorun DONE.\n");
1962 }
1963
1964 /*
1965  * import RAID devices based on one partition
1966  * if possible, the array gets run as well.
1967  */
1968
1969 static int autostart_array(dev_t startdev)
1970 {
1971         char b[BDEVNAME_SIZE];
1972         int err = -EINVAL, i;
1973         mdp_super_t *sb = NULL;
1974         mdk_rdev_t *start_rdev = NULL, *rdev;
1975
1976         start_rdev = md_import_device(startdev, 0, 0);
1977         if (IS_ERR(start_rdev))
1978                 return err;
1979
1980
1981         /* NOTE: this can only work for 0.90.0 superblocks */
1982         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1983         if (sb->major_version != 0 ||
1984             sb->minor_version != 90 ) {
1985                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1986                 export_rdev(start_rdev);
1987                 return err;
1988         }
1989
1990         if (start_rdev->faulty) {
1991                 printk(KERN_WARNING
1992                         "md: can not autostart based on faulty %s!\n",
1993                         bdevname(start_rdev->bdev,b));
1994                 export_rdev(start_rdev);
1995                 return err;
1996         }
1997         list_add(&start_rdev->same_set, &pending_raid_disks);
1998
1999         for (i = 0; i < MD_SB_DISKS; i++) {
2000                 mdp_disk_t *desc = sb->disks + i;
2001                 dev_t dev = MKDEV(desc->major, desc->minor);
2002
2003                 if (!dev)
2004                         continue;
2005                 if (dev == startdev)
2006                         continue;
2007                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2008                         continue;
2009                 rdev = md_import_device(dev, 0, 0);
2010                 if (IS_ERR(rdev))
2011                         continue;
2012
2013                 list_add(&rdev->same_set, &pending_raid_disks);
2014         }
2015
2016         /*
2017          * possibly return codes
2018          */
2019         autorun_devices(0);
2020         return 0;
2021
2022 }
2023
2024
2025 static int get_version(void __user * arg)
2026 {
2027         mdu_version_t ver;
2028
2029         ver.major = MD_MAJOR_VERSION;
2030         ver.minor = MD_MINOR_VERSION;
2031         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2032
2033         if (copy_to_user(arg, &ver, sizeof(ver)))
2034                 return -EFAULT;
2035
2036         return 0;
2037 }
2038
2039 static int get_array_info(mddev_t * mddev, void __user * arg)
2040 {
2041         mdu_array_info_t info;
2042         int nr,working,active,failed,spare;
2043         mdk_rdev_t *rdev;
2044         struct list_head *tmp;
2045
2046         nr=working=active=failed=spare=0;
2047         ITERATE_RDEV(mddev,rdev,tmp) {
2048                 nr++;
2049                 if (rdev->faulty)
2050                         failed++;
2051                 else {
2052                         working++;
2053                         if (rdev->in_sync)
2054                                 active++;
2055                         else
2056                                 spare++;
2057                 }
2058         }
2059
2060         info.major_version = mddev->major_version;
2061         info.minor_version = mddev->minor_version;
2062         info.patch_version = MD_PATCHLEVEL_VERSION;
2063         info.ctime         = mddev->ctime;
2064         info.level         = mddev->level;
2065         info.size          = mddev->size;
2066         info.nr_disks      = nr;
2067         info.raid_disks    = mddev->raid_disks;
2068         info.md_minor      = mddev->md_minor;
2069         info.not_persistent= !mddev->persistent;
2070
2071         info.utime         = mddev->utime;
2072         info.state         = 0;
2073         if (mddev->in_sync)
2074                 info.state = (1<<MD_SB_CLEAN);
2075         info.active_disks  = active;
2076         info.working_disks = working;
2077         info.failed_disks  = failed;
2078         info.spare_disks   = spare;
2079
2080         info.layout        = mddev->layout;
2081         info.chunk_size    = mddev->chunk_size;
2082
2083         if (copy_to_user(arg, &info, sizeof(info)))
2084                 return -EFAULT;
2085
2086         return 0;
2087 }
2088
2089 static int get_bitmap_file(mddev_t * mddev, void * arg)
2090 {
2091         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2092         char *ptr, *buf = NULL;
2093         int err = -ENOMEM;
2094
2095         file = kmalloc(sizeof(*file), GFP_KERNEL);
2096         if (!file)
2097                 goto out;
2098
2099         /* bitmap disabled, zero the first byte and copy out */
2100         if (!mddev->bitmap || !mddev->bitmap->file) {
2101                 file->pathname[0] = '\0';
2102                 goto copy_out;
2103         }
2104
2105         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2106         if (!buf)
2107                 goto out;
2108
2109         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2110         if (!ptr)
2111                 goto out;
2112
2113         strcpy(file->pathname, ptr);
2114
2115 copy_out:
2116         err = 0;
2117         if (copy_to_user(arg, file, sizeof(*file)))
2118                 err = -EFAULT;
2119 out:
2120         kfree(buf);
2121         kfree(file);
2122         return err;
2123 }
2124
2125 static int get_disk_info(mddev_t * mddev, void __user * arg)
2126 {
2127         mdu_disk_info_t info;
2128         unsigned int nr;
2129         mdk_rdev_t *rdev;
2130
2131         if (copy_from_user(&info, arg, sizeof(info)))
2132                 return -EFAULT;
2133
2134         nr = info.number;
2135
2136         rdev = find_rdev_nr(mddev, nr);
2137         if (rdev) {
2138                 info.major = MAJOR(rdev->bdev->bd_dev);
2139                 info.minor = MINOR(rdev->bdev->bd_dev);
2140                 info.raid_disk = rdev->raid_disk;
2141                 info.state = 0;
2142                 if (rdev->faulty)
2143                         info.state |= (1<<MD_DISK_FAULTY);
2144                 else if (rdev->in_sync) {
2145                         info.state |= (1<<MD_DISK_ACTIVE);
2146                         info.state |= (1<<MD_DISK_SYNC);
2147                 }
2148         } else {
2149                 info.major = info.minor = 0;
2150                 info.raid_disk = -1;
2151                 info.state = (1<<MD_DISK_REMOVED);
2152         }
2153
2154         if (copy_to_user(arg, &info, sizeof(info)))
2155                 return -EFAULT;
2156
2157         return 0;
2158 }
2159
2160 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2161 {
2162         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2163         mdk_rdev_t *rdev;
2164         dev_t dev = MKDEV(info->major,info->minor);
2165
2166         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2167                 return -EOVERFLOW;
2168
2169         if (!mddev->raid_disks) {
2170                 int err;
2171                 /* expecting a device which has a superblock */
2172                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2173                 if (IS_ERR(rdev)) {
2174                         printk(KERN_WARNING
2175                                 "md: md_import_device returned %ld\n",
2176                                 PTR_ERR(rdev));
2177                         return PTR_ERR(rdev);
2178                 }
2179                 if (!list_empty(&mddev->disks)) {
2180                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2181                                                         mdk_rdev_t, same_set);
2182                         int err = super_types[mddev->major_version]
2183                                 .load_super(rdev, rdev0, mddev->minor_version);
2184                         if (err < 0) {
2185                                 printk(KERN_WARNING
2186                                         "md: %s has different UUID to %s\n",
2187                                         bdevname(rdev->bdev,b),
2188                                         bdevname(rdev0->bdev,b2));
2189                                 export_rdev(rdev);
2190                                 return -EINVAL;
2191                         }
2192                 }
2193                 err = bind_rdev_to_array(rdev, mddev);
2194                 if (err)
2195                         export_rdev(rdev);
2196                 return err;
2197         }
2198
2199         /*
2200          * add_new_disk can be used once the array is assembled
2201          * to add "hot spares".  They must already have a superblock
2202          * written
2203          */
2204         if (mddev->pers) {
2205                 int err;
2206                 if (!mddev->pers->hot_add_disk) {
2207                         printk(KERN_WARNING
2208                                 "%s: personality does not support diskops!\n",
2209                                mdname(mddev));
2210                         return -EINVAL;
2211                 }
2212                 rdev = md_import_device(dev, mddev->major_version,
2213                                         mddev->minor_version);
2214                 if (IS_ERR(rdev)) {
2215                         printk(KERN_WARNING
2216                                 "md: md_import_device returned %ld\n",
2217                                 PTR_ERR(rdev));
2218                         return PTR_ERR(rdev);
2219                 }
2220                 /* set save_raid_disk if appropriate */
2221                 if (!mddev->persistent) {
2222                         if (info->state & (1<<MD_DISK_SYNC)  &&
2223                             info->raid_disk < mddev->raid_disks)
2224                                 rdev->raid_disk = info->raid_disk;
2225                         else
2226                                 rdev->raid_disk = -1;
2227                 } else
2228                         super_types[mddev->major_version].
2229                                 validate_super(mddev, rdev);
2230                 rdev->saved_raid_disk = rdev->raid_disk;
2231
2232                 rdev->in_sync = 0; /* just to be sure */
2233                 rdev->raid_disk = -1;
2234                 err = bind_rdev_to_array(rdev, mddev);
2235                 if (err)
2236                         export_rdev(rdev);
2237
2238                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2239                 if (mddev->thread)
2240                         md_wakeup_thread(mddev->thread);
2241                 return err;
2242         }
2243
2244         /* otherwise, add_new_disk is only allowed
2245          * for major_version==0 superblocks
2246          */
2247         if (mddev->major_version != 0) {
2248                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2249                        mdname(mddev));
2250                 return -EINVAL;
2251         }
2252
2253         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2254                 int err;
2255                 rdev = md_import_device (dev, -1, 0);
2256                 if (IS_ERR(rdev)) {
2257                         printk(KERN_WARNING
2258                                 "md: error, md_import_device() returned %ld\n",
2259                                 PTR_ERR(rdev));
2260                         return PTR_ERR(rdev);
2261                 }
2262                 rdev->desc_nr = info->number;
2263                 if (info->raid_disk < mddev->raid_disks)
2264                         rdev->raid_disk = info->raid_disk;
2265                 else
2266                         rdev->raid_disk = -1;
2267
2268                 rdev->faulty = 0;
2269                 if (rdev->raid_disk < mddev->raid_disks)
2270                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2271                 else
2272                         rdev->in_sync = 0;
2273
2274                 err = bind_rdev_to_array(rdev, mddev);
2275                 if (err) {
2276                         export_rdev(rdev);
2277                         return err;
2278                 }
2279
2280                 if (!mddev->persistent) {
2281                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2282                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2283                 } else
2284                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2285                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2286
2287                 if (!mddev->size || (mddev->size > rdev->size))
2288                         mddev->size = rdev->size;
2289         }
2290
2291         return 0;
2292 }
2293
2294 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2295 {
2296         char b[BDEVNAME_SIZE];
2297         mdk_rdev_t *rdev;
2298
2299         if (!mddev->pers)
2300                 return -ENODEV;
2301
2302         rdev = find_rdev(mddev, dev);
2303         if (!rdev)
2304                 return -ENXIO;
2305
2306         if (rdev->raid_disk >= 0)
2307                 goto busy;
2308
2309         kick_rdev_from_array(rdev);
2310         md_update_sb(mddev);
2311
2312         return 0;
2313 busy:
2314         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2315                 bdevname(rdev->bdev,b), mdname(mddev));
2316         return -EBUSY;
2317 }
2318
2319 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2320 {
2321         char b[BDEVNAME_SIZE];
2322         int err;
2323         unsigned int size;
2324         mdk_rdev_t *rdev;
2325
2326         if (!mddev->pers)
2327                 return -ENODEV;
2328
2329         if (mddev->major_version != 0) {
2330                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2331                         " version-0 superblocks.\n",
2332                         mdname(mddev));
2333                 return -EINVAL;
2334         }
2335         if (!mddev->pers->hot_add_disk) {
2336                 printk(KERN_WARNING
2337                         "%s: personality does not support diskops!\n",
2338                         mdname(mddev));
2339                 return -EINVAL;
2340         }
2341
2342         rdev = md_import_device (dev, -1, 0);
2343         if (IS_ERR(rdev)) {
2344                 printk(KERN_WARNING
2345                         "md: error, md_import_device() returned %ld\n",
2346                         PTR_ERR(rdev));
2347                 return -EINVAL;
2348         }
2349
2350         if (mddev->persistent)
2351                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2352         else
2353                 rdev->sb_offset =
2354                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2355
2356         size = calc_dev_size(rdev, mddev->chunk_size);
2357         rdev->size = size;
2358
2359         if (size < mddev->size) {
2360                 printk(KERN_WARNING
2361                         "%s: disk size %llu blocks < array size %llu\n",
2362                         mdname(mddev), (unsigned long long)size,
2363                         (unsigned long long)mddev->size);
2364                 err = -ENOSPC;
2365                 goto abort_export;
2366         }
2367
2368         if (rdev->faulty) {
2369                 printk(KERN_WARNING
2370                         "md: can not hot-add faulty %s disk to %s!\n",
2371                         bdevname(rdev->bdev,b), mdname(mddev));
2372                 err = -EINVAL;
2373                 goto abort_export;
2374         }
2375         rdev->in_sync = 0;
2376         rdev->desc_nr = -1;
2377         bind_rdev_to_array(rdev, mddev);
2378
2379         /*
2380          * The rest should better be atomic, we can have disk failures
2381          * noticed in interrupt contexts ...
2382          */
2383
2384         if (rdev->desc_nr == mddev->max_disks) {
2385                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2386                         mdname(mddev));
2387                 err = -EBUSY;
2388                 goto abort_unbind_export;
2389         }
2390
2391         rdev->raid_disk = -1;
2392
2393         md_update_sb(mddev);
2394
2395         /*
2396          * Kick recovery, maybe this spare has to be added to the
2397          * array immediately.
2398          */
2399         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2400         md_wakeup_thread(mddev->thread);
2401
2402         return 0;
2403
2404 abort_unbind_export:
2405         unbind_rdev_from_array(rdev);
2406
2407 abort_export:
2408         export_rdev(rdev);
2409         return err;
2410 }
2411
2412 /* similar to deny_write_access, but accounts for our holding a reference
2413  * to the file ourselves */
2414 static int deny_bitmap_write_access(struct file * file)
2415 {
2416         struct inode *inode = file->f_mapping->host;
2417
2418         spin_lock(&inode->i_lock);
2419         if (atomic_read(&inode->i_writecount) > 1) {
2420                 spin_unlock(&inode->i_lock);
2421                 return -ETXTBSY;
2422         }
2423         atomic_set(&inode->i_writecount, -1);
2424         spin_unlock(&inode->i_lock);
2425
2426         return 0;
2427 }
2428
2429 static int set_bitmap_file(mddev_t *mddev, int fd)
2430 {
2431         int err;
2432
2433         if (mddev->pers)
2434                 return -EBUSY;
2435
2436         mddev->bitmap_file = fget(fd);
2437
2438         if (mddev->bitmap_file == NULL) {
2439                 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2440                         mdname(mddev));
2441                 return -EBADF;
2442         }
2443
2444         err = deny_bitmap_write_access(mddev->bitmap_file);
2445         if (err) {
2446                 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2447                         mdname(mddev));
2448                 fput(mddev->bitmap_file);
2449                 mddev->bitmap_file = NULL;
2450         } else
2451                 mddev->bitmap_offset = 0; /* file overrides offset */
2452         return err;
2453 }
2454
2455 /*
2456  * set_array_info is used two different ways
2457  * The original usage is when creating a new array.
2458  * In this usage, raid_disks is > 0 and it together with
2459  *  level, size, not_persistent,layout,chunksize determine the
2460  *  shape of the array.
2461  *  This will always create an array with a type-0.90.0 superblock.
2462  * The newer usage is when assembling an array.
2463  *  In this case raid_disks will be 0, and the major_version field is
2464  *  use to determine which style super-blocks are to be found on the devices.
2465  *  The minor and patch _version numbers are also kept incase the
2466  *  super_block handler wishes to interpret them.
2467  */
2468 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2469 {
2470
2471         if (info->raid_disks == 0) {
2472                 /* just setting version number for superblock loading */
2473                 if (info->major_version < 0 ||
2474                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2475                     super_types[info->major_version].name == NULL) {
2476                         /* maybe try to auto-load a module? */
2477                         printk(KERN_INFO
2478                                 "md: superblock version %d not known\n",
2479                                 info->major_version);
2480                         return -EINVAL;
2481                 }
2482                 mddev->major_version = info->major_version;
2483                 mddev->minor_version = info->minor_version;
2484                 mddev->patch_version = info->patch_version;
2485                 return 0;
2486         }
2487         mddev->major_version = MD_MAJOR_VERSION;
2488         mddev->minor_version = MD_MINOR_VERSION;
2489         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2490         mddev->ctime         = get_seconds();
2491
2492         mddev->level         = info->level;
2493         mddev->size          = info->size;
2494         mddev->raid_disks    = info->raid_disks;
2495         /* don't set md_minor, it is determined by which /dev/md* was
2496          * openned
2497          */
2498         if (info->state & (1<<MD_SB_CLEAN))
2499                 mddev->recovery_cp = MaxSector;
2500         else
2501                 mddev->recovery_cp = 0;
2502         mddev->persistent    = ! info->not_persistent;
2503
2504         mddev->layout        = info->layout;
2505         mddev->chunk_size    = info->chunk_size;
2506
2507         mddev->max_disks     = MD_SB_DISKS;
2508
2509         mddev->sb_dirty      = 1;
2510
2511         /*
2512          * Generate a 128 bit UUID
2513          */
2514         get_random_bytes(mddev->uuid, 16);
2515
2516         return 0;
2517 }
2518
2519 /*
2520  * update_array_info is used to change the configuration of an
2521  * on-line array.
2522  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2523  * fields in the info are checked against the array.
2524  * Any differences that cannot be handled will cause an error.
2525  * Normally, only one change can be managed at a time.
2526  */
2527 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2528 {
2529         int rv = 0;
2530         int cnt = 0;
2531
2532         if (mddev->major_version != info->major_version ||
2533             mddev->minor_version != info->minor_version ||
2534 /*          mddev->patch_version != info->patch_version || */
2535             mddev->ctime         != info->ctime         ||
2536             mddev->level         != info->level         ||
2537 /*          mddev->layout        != info->layout        || */
2538             !mddev->persistent   != info->not_persistent||
2539             mddev->chunk_size    != info->chunk_size    )
2540                 return -EINVAL;
2541         /* Check there is only one change */
2542         if (mddev->size != info->size) cnt++;
2543         if (mddev->raid_disks != info->raid_disks) cnt++;
2544         if (mddev->layout != info->layout) cnt++;
2545         if (cnt == 0) return 0;
2546         if (cnt > 1) return -EINVAL;
2547
2548         if (mddev->layout != info->layout) {
2549                 /* Change layout
2550                  * we don't need to do anything at the md level, the
2551                  * personality will take care of it all.
2552                  */
2553                 if (mddev->pers->reconfig == NULL)
2554                         return -EINVAL;
2555                 else
2556                         return mddev->pers->reconfig(mddev, info->layout, -1);
2557         }
2558         if (mddev->size != info->size) {
2559                 mdk_rdev_t * rdev;
2560                 struct list_head *tmp;
2561                 if (mddev->pers->resize == NULL)
2562                         return -EINVAL;
2563                 /* The "size" is the amount of each device that is used.
2564                  * This can only make sense for arrays with redundancy.
2565                  * linear and raid0 always use whatever space is available
2566                  * We can only consider changing the size if no resync
2567                  * or reconstruction is happening, and if the new size
2568                  * is acceptable. It must fit before the sb_offset or,
2569                  * if that is <data_offset, it must fit before the
2570                  * size of each device.
2571                  * If size is zero, we find the largest size that fits.
2572                  */
2573                 if (mddev->sync_thread)
2574                         return -EBUSY;
2575                 ITERATE_RDEV(mddev,rdev,tmp) {
2576                         sector_t avail;
2577                         int fit = (info->size == 0);
2578                         if (rdev->sb_offset > rdev->data_offset)
2579                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2580                         else
2581                                 avail = get_capacity(rdev->bdev->bd_disk)
2582                                         - rdev->data_offset;
2583                         if (fit && (info->size == 0 || info->size > avail/2))
2584                                 info->size = avail/2;
2585                         if (avail < ((sector_t)info->size << 1))
2586                                 return -ENOSPC;
2587                 }
2588                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2589                 if (!rv) {
2590                         struct block_device *bdev;
2591
2592                         bdev = bdget_disk(mddev->gendisk, 0);
2593                         if (bdev) {
2594                                 down(&bdev->bd_inode->i_sem);
2595                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2596                                 up(&bdev->bd_inode->i_sem);
2597                                 bdput(bdev);
2598                         }
2599                 }
2600         }
2601         if (mddev->raid_disks    != info->raid_disks) {
2602                 /* change the number of raid disks */
2603                 if (mddev->pers->reshape == NULL)
2604                         return -EINVAL;
2605                 if (info->raid_disks <= 0 ||
2606                     info->raid_disks >= mddev->max_disks)
2607                         return -EINVAL;
2608                 if (mddev->sync_thread)
2609                         return -EBUSY;
2610                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2611                 if (!rv) {
2612                         struct block_device *bdev;
2613
2614                         bdev = bdget_disk(mddev->gendisk, 0);
2615                         if (bdev) {
2616                                 down(&bdev->bd_inode->i_sem);
2617                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2618                                 up(&bdev->bd_inode->i_sem);
2619                                 bdput(bdev);
2620                         }
2621                 }
2622         }
2623         md_update_sb(mddev);
2624         return rv;
2625 }
2626
2627 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2628 {
2629         mdk_rdev_t *rdev;
2630
2631         if (mddev->pers == NULL)
2632                 return -ENODEV;
2633
2634         rdev = find_rdev(mddev, dev);
2635         if (!rdev)
2636                 return -ENODEV;
2637
2638         md_error(mddev, rdev);
2639         return 0;
2640 }
2641
2642 static int md_ioctl(struct inode *inode, struct file *file,
2643                         unsigned int cmd, unsigned long arg)
2644 {
2645         int err = 0;
2646         void __user *argp = (void __user *)arg;
2647         struct hd_geometry __user *loc = argp;
2648         mddev_t *mddev = NULL;
2649
2650         if (!capable(CAP_SYS_ADMIN))
2651                 return -EACCES;
2652
2653         /*
2654          * Commands dealing with the RAID driver but not any
2655          * particular array:
2656          */
2657         switch (cmd)
2658         {
2659                 case RAID_VERSION:
2660                         err = get_version(argp);
2661                         goto done;
2662
2663                 case PRINT_RAID_DEBUG:
2664                         err = 0;
2665                         md_print_devices();
2666                         goto done;
2667
2668 #ifndef MODULE
2669                 case RAID_AUTORUN:
2670                         err = 0;
2671                         autostart_arrays(arg);
2672                         goto done;
2673 #endif
2674                 default:;
2675         }
2676
2677         /*
2678          * Commands creating/starting a new array:
2679          */
2680
2681         mddev = inode->i_bdev->bd_disk->private_data;
2682
2683         if (!mddev) {
2684                 BUG();
2685                 goto abort;
2686         }
2687
2688
2689         if (cmd == START_ARRAY) {
2690                 /* START_ARRAY doesn't need to lock the array as autostart_array
2691                  * does the locking, and it could even be a different array
2692                  */
2693                 static int cnt = 3;
2694                 if (cnt > 0 ) {
2695                         printk(KERN_WARNING
2696                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2697                                "This will not be supported beyond 2.6\n",
2698                                current->comm, current->pid);
2699                         cnt--;
2700                 }
2701                 err = autostart_array(new_decode_dev(arg));
2702                 if (err) {
2703                         printk(KERN_WARNING "md: autostart failed!\n");
2704                         goto abort;
2705                 }
2706                 goto done;
2707         }
2708
2709         err = mddev_lock(mddev);
2710         if (err) {
2711                 printk(KERN_INFO
2712                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2713                         err, cmd);
2714                 goto abort;
2715         }
2716
2717         switch (cmd)
2718         {
2719                 case SET_ARRAY_INFO:
2720                         {
2721                                 mdu_array_info_t info;
2722                                 if (!arg)
2723                                         memset(&info, 0, sizeof(info));
2724                                 else if (copy_from_user(&info, argp, sizeof(info))) {
2725                                         err = -EFAULT;
2726                                         goto abort_unlock;
2727                                 }
2728                                 if (mddev->pers) {
2729                                         err = update_array_info(mddev, &info);
2730                                         if (err) {
2731                                                 printk(KERN_WARNING "md: couldn't update"
2732                                                        " array info. %d\n", err);
2733                                                 goto abort_unlock;
2734                                         }
2735                                         goto done_unlock;
2736                                 }
2737                                 if (!list_empty(&mddev->disks)) {
2738                                         printk(KERN_WARNING
2739                                                "md: array %s already has disks!\n",
2740                                                mdname(mddev));
2741                                         err = -EBUSY;
2742                                         goto abort_unlock;
2743                                 }
2744                                 if (mddev->raid_disks) {
2745                                         printk(KERN_WARNING
2746                                                "md: array %s already initialised!\n",
2747                                                mdname(mddev));
2748                                         err = -EBUSY;
2749                                         goto abort_unlock;
2750                                 }
2751                                 err = set_array_info(mddev, &info);
2752                                 if (err) {
2753                                         printk(KERN_WARNING "md: couldn't set"
2754                                                " array info. %d\n", err);
2755                                         goto abort_unlock;
2756                                 }
2757                         }
2758                         goto done_unlock;
2759
2760                 default:;
2761         }
2762
2763         /*
2764          * Commands querying/configuring an existing array:
2765          */
2766         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2767          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2768         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2769                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2770                 err = -ENODEV;
2771                 goto abort_unlock;
2772         }
2773
2774         /*
2775          * Commands even a read-only array can execute:
2776          */
2777         switch (cmd)
2778         {
2779                 case GET_ARRAY_INFO:
2780                         err = get_array_info(mddev, argp);
2781                         goto done_unlock;
2782
2783                 case GET_BITMAP_FILE:
2784                         err = get_bitmap_file(mddev, (void *)arg);
2785                         goto done_unlock;
2786
2787                 case GET_DISK_INFO:
2788                         err = get_disk_info(mddev, argp);
2789                         goto done_unlock;
2790
2791                 case RESTART_ARRAY_RW:
2792                         err = restart_array(mddev);
2793                         goto done_unlock;
2794
2795                 case STOP_ARRAY:
2796                         err = do_md_stop (mddev, 0);
2797                         goto done_unlock;
2798
2799                 case STOP_ARRAY_RO:
2800                         err = do_md_stop (mddev, 1);
2801                         goto done_unlock;
2802
2803         /*
2804          * We have a problem here : there is no easy way to give a CHS
2805          * virtual geometry. We currently pretend that we have a 2 heads
2806          * 4 sectors (with a BIG number of cylinders...). This drives
2807          * dosfs just mad... ;-)
2808          */
2809                 case HDIO_GETGEO:
2810                         if (!loc) {
2811                                 err = -EINVAL;
2812                                 goto abort_unlock;
2813                         }
2814                         err = put_user (2, (char __user *) &loc->heads);
2815                         if (err)
2816                                 goto abort_unlock;
2817                         err = put_user (4, (char __user *) &loc->sectors);
2818                         if (err)
2819                                 goto abort_unlock;
2820                         err = put_user(get_capacity(mddev->gendisk)/8,
2821                                         (short __user *) &loc->cylinders);
2822                         if (err)
2823                                 goto abort_unlock;
2824                         err = put_user (get_start_sect(inode->i_bdev),
2825                                                 (long __user *) &loc->start);
2826                         goto done_unlock;
2827         }
2828
2829         /*
2830          * The remaining ioctls are changing the state of the
2831          * superblock, so we do not allow read-only arrays
2832          * here:
2833          */
2834         if (mddev->ro) {
2835                 err = -EROFS;
2836                 goto abort_unlock;
2837         }
2838
2839         switch (cmd)
2840         {
2841                 case ADD_NEW_DISK:
2842                 {
2843                         mdu_disk_info_t info;
2844                         if (copy_from_user(&info, argp, sizeof(info)))
2845                                 err = -EFAULT;
2846                         else
2847                                 err = add_new_disk(mddev, &info);
2848                         goto done_unlock;
2849                 }
2850
2851                 case HOT_REMOVE_DISK:
2852                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2853                         goto done_unlock;
2854
2855                 case HOT_ADD_DISK:
2856                         err = hot_add_disk(mddev, new_decode_dev(arg));
2857                         goto done_unlock;
2858
2859                 case SET_DISK_FAULTY:
2860                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2861                         goto done_unlock;
2862
2863                 case RUN_ARRAY:
2864                         err = do_md_run (mddev);
2865                         goto done_unlock;
2866
2867                 case SET_BITMAP_FILE:
2868                         err = set_bitmap_file(mddev, (int)arg);
2869                         goto done_unlock;
2870
2871                 default:
2872                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2873                                 printk(KERN_WARNING "md: %s(pid %d) used"
2874                                         " obsolete MD ioctl, upgrade your"
2875                                         " software to use new ictls.\n",
2876                                         current->comm, current->pid);
2877                         err = -EINVAL;
2878                         goto abort_unlock;
2879         }
2880
2881 done_unlock:
2882 abort_unlock:
2883         mddev_unlock(mddev);
2884
2885         return err;
2886 done:
2887         if (err)
2888                 MD_BUG();
2889 abort:
2890         return err;
2891 }
2892
2893 static int md_open(struct inode *inode, struct file *file)
2894 {
2895         /*
2896          * Succeed if we can lock the mddev, which confirms that
2897          * it isn't being stopped right now.
2898          */
2899         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2900         int err;
2901
2902         if ((err = mddev_lock(mddev)))
2903                 goto out;
2904
2905         err = 0;
2906         mddev_get(mddev);
2907         mddev_unlock(mddev);
2908
2909         check_disk_change(inode->i_bdev);
2910  out:
2911         return err;
2912 }
2913
2914 static int md_release(struct inode *inode, struct file * file)
2915 {
2916         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2917
2918         if (!mddev)
2919                 BUG();
2920         mddev_put(mddev);
2921
2922         return 0;
2923 }
2924
2925 static int md_media_changed(struct gendisk *disk)
2926 {
2927         mddev_t *mddev = disk->private_data;
2928
2929         return mddev->changed;
2930 }
2931
2932 static int md_revalidate(struct gendisk *disk)
2933 {
2934         mddev_t *mddev = disk->private_data;
2935
2936         mddev->changed = 0;
2937         return 0;
2938 }
2939 static struct block_device_operations md_fops =
2940 {
2941         .owner          = THIS_MODULE,
2942         .open           = md_open,
2943         .release        = md_release,
2944         .ioctl          = md_ioctl,
2945         .media_changed  = md_media_changed,
2946         .revalidate_disk= md_revalidate,
2947 };
2948
2949 static int md_thread(void * arg)
2950 {
2951         mdk_thread_t *thread = arg;
2952
2953         lock_kernel();
2954
2955         /*
2956          * Detach thread
2957          */
2958
2959         daemonize(thread->name, mdname(thread->mddev));
2960
2961         current->exit_signal = SIGCHLD;
2962         allow_signal(SIGKILL);
2963         thread->tsk = current;
2964
2965         /*
2966          * md_thread is a 'system-thread', it's priority should be very
2967          * high. We avoid resource deadlocks individually in each
2968          * raid personality. (RAID5 does preallocation) We also use RR and
2969          * the very same RT priority as kswapd, thus we will never get
2970          * into a priority inversion deadlock.
2971          *
2972          * we definitely have to have equal or higher priority than
2973          * bdflush, otherwise bdflush will deadlock if there are too
2974          * many dirty RAID5 blocks.
2975          */
2976         unlock_kernel();
2977
2978         complete(thread->event);
2979         while (thread->run) {
2980                 void (*run)(mddev_t *);
2981
2982                 wait_event_interruptible_timeout(thread->wqueue,
2983                                                  test_bit(THREAD_WAKEUP, &thread->flags),
2984                                                  thread->timeout);
2985                 try_to_freeze();
2986
2987                 clear_bit(THREAD_WAKEUP, &thread->flags);
2988
2989                 run = thread->run;
2990                 if (run)
2991                         run(thread->mddev);
2992
2993                 if (signal_pending(current))
2994                         flush_signals(current);
2995         }
2996         complete(thread->event);
2997         return 0;
2998 }
2999
3000 void md_wakeup_thread(mdk_thread_t *thread)
3001 {
3002         if (thread) {
3003                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3004                 set_bit(THREAD_WAKEUP, &thread->flags);
3005                 wake_up(&thread->wqueue);
3006         }
3007 }
3008
3009 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3010                                  const char *name)
3011 {
3012         mdk_thread_t *thread;
3013         int ret;
3014         struct completion event;
3015
3016         thread = (mdk_thread_t *) kmalloc
3017                                 (sizeof(mdk_thread_t), GFP_KERNEL);
3018         if (!thread)
3019                 return NULL;
3020
3021         memset(thread, 0, sizeof(mdk_thread_t));
3022         init_waitqueue_head(&thread->wqueue);
3023
3024         init_completion(&event);
3025         thread->event = &event;
3026         thread->run = run;
3027         thread->mddev = mddev;
3028         thread->name = name;
3029         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3030         ret = kernel_thread(md_thread, thread, 0);
3031         if (ret < 0) {
3032                 kfree(thread);
3033                 return NULL;
3034         }
3035         wait_for_completion(&event);
3036         return thread;
3037 }
3038
3039 void md_unregister_thread(mdk_thread_t *thread)
3040 {
3041         struct completion event;
3042
3043         init_completion(&event);
3044
3045         thread->event = &event;
3046
3047         /* As soon as ->run is set to NULL, the task could disappear,
3048          * so we need to hold tasklist_lock until we have sent the signal
3049          */
3050         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3051         read_lock(&tasklist_lock);
3052         thread->run = NULL;
3053         send_sig(SIGKILL, thread->tsk, 1);
3054         read_unlock(&tasklist_lock);
3055         wait_for_completion(&event);
3056         kfree(thread);
3057 }
3058
3059 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3060 {
3061         if (!mddev) {
3062                 MD_BUG();
3063                 return;
3064         }
3065
3066         if (!rdev || rdev->faulty)
3067                 return;
3068 /*
3069         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3070                 mdname(mddev),
3071                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3072                 __builtin_return_address(0),__builtin_return_address(1),
3073                 __builtin_return_address(2),__builtin_return_address(3));
3074 */
3075         if (!mddev->pers->error_handler)
3076                 return;
3077         mddev->pers->error_handler(mddev,rdev);
3078         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3079         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3080         md_wakeup_thread(mddev->thread);
3081 }
3082
3083 /* seq_file implementation /proc/mdstat */
3084
3085 static void status_unused(struct seq_file *seq)
3086 {
3087         int i = 0;
3088         mdk_rdev_t *rdev;
3089         struct list_head *tmp;
3090
3091         seq_printf(seq, "unused devices: ");
3092
3093         ITERATE_RDEV_PENDING(rdev,tmp) {
3094                 char b[BDEVNAME_SIZE];
3095                 i++;
3096                 seq_printf(seq, "%s ",
3097                               bdevname(rdev->bdev,b));
3098         }
3099         if (!i)
3100                 seq_printf(seq, "<none>");
3101
3102         seq_printf(seq, "\n");
3103 }
3104
3105
3106 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3107 {
3108         unsigned long max_blocks, resync, res, dt, db, rt;
3109
3110         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3111
3112         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3113                 max_blocks = mddev->resync_max_sectors >> 1;
3114         else
3115                 max_blocks = mddev->size;
3116
3117         /*
3118          * Should not happen.
3119          */
3120         if (!max_blocks) {
3121                 MD_BUG();
3122                 return;
3123         }
3124         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3125         {
3126                 int i, x = res/50, y = 20-x;
3127                 seq_printf(seq, "[");
3128                 for (i = 0; i < x; i++)
3129                         seq_printf(seq, "=");
3130                 seq_printf(seq, ">");
3131                 for (i = 0; i < y; i++)
3132                         seq_printf(seq, ".");
3133                 seq_printf(seq, "] ");
3134         }
3135         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3136                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3137                        "resync" : "recovery"),
3138                       res/10, res % 10, resync, max_blocks);
3139
3140         /*
3141          * We do not want to overflow, so the order of operands and
3142          * the * 100 / 100 trick are important. We do a +1 to be
3143          * safe against division by zero. We only estimate anyway.
3144          *
3145          * dt: time from mark until now
3146          * db: blocks written from mark until now
3147          * rt: remaining time
3148          */
3149         dt = ((jiffies - mddev->resync_mark) / HZ);
3150         if (!dt) dt++;
3151         db = resync - (mddev->resync_mark_cnt/2);
3152         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3153
3154         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3155
3156         seq_printf(seq, " speed=%ldK/sec", db/dt);
3157 }
3158
3159 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3160 {
3161         struct list_head *tmp;
3162         loff_t l = *pos;
3163         mddev_t *mddev;
3164
3165         if (l >= 0x10000)
3166                 return NULL;
3167         if (!l--)
3168                 /* header */
3169                 return (void*)1;
3170
3171         spin_lock(&all_mddevs_lock);
3172         list_for_each(tmp,&all_mddevs)
3173                 if (!l--) {
3174                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3175                         mddev_get(mddev);
3176                         spin_unlock(&all_mddevs_lock);
3177                         return mddev;
3178                 }
3179         spin_unlock(&all_mddevs_lock);
3180         if (!l--)
3181                 return (void*)2;/* tail */
3182         return NULL;
3183 }
3184
3185 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3186 {
3187         struct list_head *tmp;
3188         mddev_t *next_mddev, *mddev = v;
3189
3190         ++*pos;
3191         if (v == (void*)2)
3192                 return NULL;
3193
3194         spin_lock(&all_mddevs_lock);
3195         if (v == (void*)1)
3196                 tmp = all_mddevs.next;
3197         else
3198                 tmp = mddev->all_mddevs.next;
3199         if (tmp != &all_mddevs)
3200                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3201         else {
3202                 next_mddev = (void*)2;
3203                 *pos = 0x10000;
3204         }
3205         spin_unlock(&all_mddevs_lock);
3206
3207         if (v != (void*)1)
3208                 mddev_put(mddev);
3209         return next_mddev;
3210
3211 }
3212
3213 static void md_seq_stop(struct seq_file *seq, void *v)
3214 {
3215         mddev_t *mddev = v;
3216
3217         if (mddev && v != (void*)1 && v != (void*)2)
3218                 mddev_put(mddev);
3219 }
3220
3221 static int md_seq_show(struct seq_file *seq, void *v)
3222 {
3223         mddev_t *mddev = v;
3224         sector_t size;
3225         struct list_head *tmp2;
3226         mdk_rdev_t *rdev;
3227         int i;
3228         struct bitmap *bitmap;
3229
3230         if (v == (void*)1) {
3231                 seq_printf(seq, "Personalities : ");
3232                 spin_lock(&pers_lock);
3233                 for (i = 0; i < MAX_PERSONALITY; i++)
3234                         if (pers[i])
3235                                 seq_printf(seq, "[%s] ", pers[i]->name);
3236
3237                 spin_unlock(&pers_lock);
3238                 seq_printf(seq, "\n");
3239                 return 0;
3240         }
3241         if (v == (void*)2) {
3242                 status_unused(seq);
3243                 return 0;
3244         }
3245
3246         if (mddev_lock(mddev)!=0)
3247                 return -EINTR;
3248         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3249                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3250                                                 mddev->pers ? "" : "in");
3251                 if (mddev->pers) {
3252                         if (mddev->ro)
3253                                 seq_printf(seq, " (read-only)");
3254                         seq_printf(seq, " %s", mddev->pers->name);
3255                 }
3256
3257                 size = 0;
3258                 ITERATE_RDEV(mddev,rdev,tmp2) {
3259                         char b[BDEVNAME_SIZE];
3260                         seq_printf(seq, " %s[%d]",
3261                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3262                         if (rdev->faulty) {
3263                                 seq_printf(seq, "(F)");
3264                                 continue;
3265                         }
3266                         size += rdev->size;
3267                 }
3268
3269                 if (!list_empty(&mddev->disks)) {
3270                         if (mddev->pers)
3271                                 seq_printf(seq, "\n      %llu blocks",
3272                                         (unsigned long long)mddev->array_size);
3273                         else
3274                                 seq_printf(seq, "\n      %llu blocks",
3275                                         (unsigned long long)size);
3276                 }
3277
3278                 if (mddev->pers) {
3279                         mddev->pers->status (seq, mddev);
3280                         seq_printf(seq, "\n      ");
3281                         if (mddev->curr_resync > 2) {
3282                                 status_resync (seq, mddev);
3283                                 seq_printf(seq, "\n      ");
3284                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3285                                 seq_printf(seq, "       resync=DELAYED\n      ");
3286                 } else
3287                         seq_printf(seq, "\n       ");
3288
3289                 if ((bitmap = mddev->bitmap)) {
3290                         unsigned long chunk_kb;
3291                         unsigned long flags;
3292                         spin_lock_irqsave(&bitmap->lock, flags);
3293                         chunk_kb = bitmap->chunksize >> 10;
3294                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3295                                 "%lu%s chunk",
3296                                 bitmap->pages - bitmap->missing_pages,
3297                                 bitmap->pages,
3298                                 (bitmap->pages - bitmap->missing_pages)
3299                                         << (PAGE_SHIFT - 10),
3300                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3301                                 chunk_kb ? "KB" : "B");
3302                         if (bitmap->file) {
3303                                 seq_printf(seq, ", file: ");
3304                                 seq_path(seq, bitmap->file->f_vfsmnt,
3305                                          bitmap->file->f_dentry," \t\n");
3306                         }
3307
3308                         seq_printf(seq, "\n");
3309                         spin_unlock_irqrestore(&bitmap->lock, flags);
3310                 }
3311
3312                 seq_printf(seq, "\n");
3313         }
3314         mddev_unlock(mddev);
3315
3316         return 0;
3317 }
3318
3319 static struct seq_operations md_seq_ops = {
3320         .start  = md_seq_start,
3321         .next   = md_seq_next,
3322         .stop   = md_seq_stop,
3323         .show   = md_seq_show,
3324 };
3325
3326 static int md_seq_open(struct inode *inode, struct file *file)
3327 {
3328         int error;
3329
3330         error = seq_open(file, &md_seq_ops);
3331         return error;
3332 }
3333
3334 static struct file_operations md_seq_fops = {
3335         .open           = md_seq_open,
3336         .read           = seq_read,
3337         .llseek         = seq_lseek,
3338         .release        = seq_release,
3339 };
3340
3341 int register_md_personality(int pnum, mdk_personality_t *p)
3342 {
3343         if (pnum >= MAX_PERSONALITY) {
3344                 printk(KERN_ERR
3345                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3346                        p->name, pnum, MAX_PERSONALITY-1);
3347                 return -EINVAL;
3348         }
3349
3350         spin_lock(&pers_lock);
3351         if (pers[pnum]) {
3352                 spin_unlock(&pers_lock);
3353                 return -EBUSY;
3354         }
3355
3356         pers[pnum] = p;
3357         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3358         spin_unlock(&pers_lock);
3359         return 0;
3360 }
3361
3362 int unregister_md_personality(int pnum)
3363 {
3364         if (pnum >= MAX_PERSONALITY)
3365                 return -EINVAL;
3366
3367         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3368         spin_lock(&pers_lock);
3369         pers[pnum] = NULL;
3370         spin_unlock(&pers_lock);
3371         return 0;
3372 }
3373
3374 static int is_mddev_idle(mddev_t *mddev)
3375 {
3376         mdk_rdev_t * rdev;
3377         struct list_head *tmp;
3378         int idle;
3379         unsigned long curr_events;
3380
3381         idle = 1;
3382         ITERATE_RDEV(mddev,rdev,tmp) {
3383                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3384                 curr_events = disk_stat_read(disk, read_sectors) +
3385                                 disk_stat_read(disk, write_sectors) -
3386                                 atomic_read(&disk->sync_io);
3387                 /* Allow some slack between valud of curr_events and last_events,
3388                  * as there are some uninteresting races.
3389                  * Note: the following is an unsigned comparison.
3390                  */
3391                 if ((curr_events - rdev->last_events + 32) > 64) {
3392                         rdev->last_events = curr_events;
3393                         idle = 0;
3394                 }
3395         }
3396         return idle;
3397 }
3398
3399 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3400 {
3401         /* another "blocks" (512byte) blocks have been synced */
3402         atomic_sub(blocks, &mddev->recovery_active);
3403         wake_up(&mddev->recovery_wait);
3404         if (!ok) {
3405                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3406                 md_wakeup_thread(mddev->thread);
3407                 // stop recovery, signal do_sync ....
3408         }
3409 }
3410
3411
3412 /* md_write_start(mddev, bi)
3413  * If we need to update some array metadata (e.g. 'active' flag
3414  * in superblock) before writing, schedule a superblock update
3415  * and wait for it to complete.
3416  */
3417 void md_write_start(mddev_t *mddev, struct bio *bi)
3418 {
3419         DEFINE_WAIT(w);
3420         if (bio_data_dir(bi) != WRITE)
3421                 return;
3422
3423         atomic_inc(&mddev->writes_pending);
3424         if (mddev->in_sync) {
3425                 spin_lock(&mddev->write_lock);
3426                 if (mddev->in_sync) {
3427                         mddev->in_sync = 0;
3428                         mddev->sb_dirty = 1;
3429                         md_wakeup_thread(mddev->thread);
3430                 }
3431                 spin_unlock(&mddev->write_lock);
3432         }
3433         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3434 }
3435
3436 void md_write_end(mddev_t *mddev)
3437 {
3438         if (atomic_dec_and_test(&mddev->writes_pending)) {
3439                 if (mddev->safemode == 2)
3440                         md_wakeup_thread(mddev->thread);
3441                 else
3442                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3443         }
3444 }
3445
3446 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3447
3448 #define SYNC_MARKS      10
3449 #define SYNC_MARK_STEP  (3*HZ)
3450 static void md_do_sync(mddev_t *mddev)
3451 {
3452         mddev_t *mddev2;
3453         unsigned int currspeed = 0,
3454                  window;
3455         sector_t max_sectors,j, io_sectors;
3456         unsigned long mark[SYNC_MARKS];
3457         sector_t mark_cnt[SYNC_MARKS];
3458         int last_mark,m;
3459         struct list_head *tmp;
3460         sector_t last_check;
3461         int skipped = 0;
3462
3463         /* just incase thread restarts... */
3464         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3465                 return;
3466
3467         /* we overload curr_resync somewhat here.
3468          * 0 == not engaged in resync at all
3469          * 2 == checking that there is no conflict with another sync
3470          * 1 == like 2, but have yielded to allow conflicting resync to
3471          *              commense
3472          * other == active in resync - this many blocks
3473          *
3474          * Before starting a resync we must have set curr_resync to
3475          * 2, and then checked that every "conflicting" array has curr_resync
3476          * less than ours.  When we find one that is the same or higher
3477          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3478          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3479          * This will mean we have to start checking from the beginning again.
3480          *
3481          */
3482
3483         do {
3484                 mddev->curr_resync = 2;
3485
3486         try_again:
3487                 if (signal_pending(current)) {
3488                         flush_signals(current);
3489                         goto skip;
3490                 }
3491                 ITERATE_MDDEV(mddev2,tmp) {
3492                         if (mddev2 == mddev)
3493                                 continue;
3494                         if (mddev2->curr_resync &&
3495                             match_mddev_units(mddev,mddev2)) {
3496                                 DEFINE_WAIT(wq);
3497                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3498                                         /* arbitrarily yield */
3499                                         mddev->curr_resync = 1;
3500                                         wake_up(&resync_wait);
3501                                 }
3502                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3503                                         /* no need to wait here, we can wait the next
3504                                          * time 'round when curr_resync == 2
3505                                          */
3506                                         continue;
3507                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3508                                 if (!signal_pending(current)
3509                                     && mddev2->curr_resync >= mddev->curr_resync) {
3510                                         printk(KERN_INFO "md: delaying resync of %s"
3511                                                " until %s has finished resync (they"
3512                                                " share one or more physical units)\n",
3513                                                mdname(mddev), mdname(mddev2));
3514                                         mddev_put(mddev2);
3515                                         schedule();
3516                                         finish_wait(&resync_wait, &wq);
3517                                         goto try_again;
3518                                 }
3519                                 finish_wait(&resync_wait, &wq);
3520                         }
3521                 }
3522         } while (mddev->curr_resync < 2);
3523
3524         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3525                 /* resync follows the size requested by the personality,
3526                  * which defaults to physical size, but can be virtual size
3527                  */
3528                 max_sectors = mddev->resync_max_sectors;
3529         else
3530                 /* recovery follows the physical size of devices */
3531                 max_sectors = mddev->size << 1;
3532
3533         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3534         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3535                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3536         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3537                "(but not more than %d KB/sec) for reconstruction.\n",
3538                sysctl_speed_limit_max);
3539
3540         is_mddev_idle(mddev); /* this also initializes IO event counters */
3541         /* we don't use the checkpoint if there's a bitmap */
3542         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3543                 j = mddev->recovery_cp;
3544         else
3545                 j = 0;
3546         io_sectors = 0;
3547         for (m = 0; m < SYNC_MARKS; m++) {
3548                 mark[m] = jiffies;
3549                 mark_cnt[m] = io_sectors;
3550         }
3551         last_mark = 0;
3552         mddev->resync_mark = mark[last_mark];
3553         mddev->resync_mark_cnt = mark_cnt[last_mark];
3554
3555         /*
3556          * Tune reconstruction:
3557          */
3558         window = 32*(PAGE_SIZE/512);
3559         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3560                 window/2,(unsigned long long) max_sectors/2);
3561
3562         atomic_set(&mddev->recovery_active, 0);
3563         init_waitqueue_head(&mddev->recovery_wait);
3564         last_check = 0;
3565
3566         if (j>2) {
3567                 printk(KERN_INFO
3568                         "md: resuming recovery of %s from checkpoint.\n",
3569                         mdname(mddev));
3570                 mddev->curr_resync = j;
3571         }
3572
3573         while (j < max_sectors) {
3574                 sector_t sectors;
3575
3576                 skipped = 0;
3577                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3578                                             currspeed < sysctl_speed_limit_min);
3579                 if (sectors == 0) {
3580                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3581                         goto out;
3582                 }
3583
3584                 if (!skipped) { /* actual IO requested */
3585                         io_sectors += sectors;
3586                         atomic_add(sectors, &mddev->recovery_active);
3587                 }
3588
3589                 j += sectors;
3590                 if (j>1) mddev->curr_resync = j;
3591
3592
3593                 if (last_check + window > io_sectors || j == max_sectors)
3594                         continue;
3595
3596                 last_check = io_sectors;
3597
3598                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3599                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3600                         break;
3601
3602         repeat:
3603                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3604                         /* step marks */
3605                         int next = (last_mark+1) % SYNC_MARKS;
3606
3607                         mddev->resync_mark = mark[next];
3608                         mddev->resync_mark_cnt = mark_cnt[next];
3609                         mark[next] = jiffies;
3610                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3611                         last_mark = next;
3612                 }
3613
3614
3615                 if (signal_pending(current)) {
3616                         /*
3617                          * got a signal, exit.
3618                          */
3619                         printk(KERN_INFO
3620                                 "md: md_do_sync() got signal ... exiting\n");
3621                         flush_signals(current);
3622                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3623                         goto out;
3624                 }
3625
3626                 /*
3627                  * this loop exits only if either when we are slower than
3628                  * the 'hard' speed limit, or the system was IO-idle for
3629                  * a jiffy.
3630                  * the system might be non-idle CPU-wise, but we only care
3631                  * about not overloading the IO subsystem. (things like an
3632                  * e2fsck being done on the RAID array should execute fast)
3633                  */
3634                 mddev->queue->unplug_fn(mddev->queue);
3635                 cond_resched();
3636
3637                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3638                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
3639
3640                 if (currspeed > sysctl_speed_limit_min) {
3641                         if ((currspeed > sysctl_speed_limit_max) ||
3642                                         !is_mddev_idle(mddev)) {
3643                                 msleep_interruptible(250);
3644                                 goto repeat;
3645                         }
3646                 }
3647         }
3648         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3649         /*
3650          * this also signals 'finished resyncing' to md_stop
3651          */
3652  out:
3653         mddev->queue->unplug_fn(mddev->queue);
3654
3655         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3656
3657         /* tell personality that we are finished */
3658         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3659
3660         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3661             mddev->curr_resync > 2 &&
3662             mddev->curr_resync >= mddev->recovery_cp) {
3663                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3664                         printk(KERN_INFO
3665                                 "md: checkpointing recovery of %s.\n",
3666                                 mdname(mddev));
3667                         mddev->recovery_cp = mddev->curr_resync;
3668                 } else
3669                         mddev->recovery_cp = MaxSector;
3670         }
3671
3672  skip:
3673         mddev->curr_resync = 0;
3674         wake_up(&resync_wait);
3675         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3676         md_wakeup_thread(mddev->thread);
3677 }
3678
3679
3680 /*
3681  * This routine is regularly called by all per-raid-array threads to
3682  * deal with generic issues like resync and super-block update.
3683  * Raid personalities that don't have a thread (linear/raid0) do not
3684  * need this as they never do any recovery or update the superblock.
3685  *
3686  * It does not do any resync itself, but rather "forks" off other threads
3687  * to do that as needed.
3688  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3689  * "->recovery" and create a thread at ->sync_thread.
3690  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3691  * and wakeups up this thread which will reap the thread and finish up.
3692  * This thread also removes any faulty devices (with nr_pending == 0).
3693  *
3694  * The overall approach is:
3695  *  1/ if the superblock needs updating, update it.
3696  *  2/ If a recovery thread is running, don't do anything else.
3697  *  3/ If recovery has finished, clean up, possibly marking spares active.
3698  *  4/ If there are any faulty devices, remove them.
3699  *  5/ If array is degraded, try to add spares devices
3700  *  6/ If array has spares or is not in-sync, start a resync thread.
3701  */
3702 void md_check_recovery(mddev_t *mddev)
3703 {
3704         mdk_rdev_t *rdev;
3705         struct list_head *rtmp;
3706
3707
3708         if (mddev->bitmap)
3709                 bitmap_daemon_work(mddev->bitmap);
3710
3711         if (mddev->ro)
3712                 return;
3713
3714         if (signal_pending(current)) {
3715                 if (mddev->pers->sync_request) {
3716                         printk(KERN_INFO "md: %s in immediate safe mode\n",
3717                                mdname(mddev));
3718                         mddev->safemode = 2;
3719                 }
3720                 flush_signals(current);
3721         }
3722
3723         if ( ! (
3724                 mddev->sb_dirty ||
3725                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3726                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3727                 (mddev->safemode == 1) ||
3728                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3729                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3730                 ))
3731                 return;
3732
3733         if (mddev_trylock(mddev)==0) {
3734                 int spares =0;
3735
3736                 spin_lock(&mddev->write_lock);
3737                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3738                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3739                         mddev->in_sync = 1;
3740                         mddev->sb_dirty = 1;
3741                 }
3742                 if (mddev->safemode == 1)
3743                         mddev->safemode = 0;
3744                 spin_unlock(&mddev->write_lock);
3745
3746                 if (mddev->sb_dirty)
3747                         md_update_sb(mddev);
3748
3749
3750                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3751                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3752                         /* resync/recovery still happening */
3753                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3754                         goto unlock;
3755                 }
3756                 if (mddev->sync_thread) {
3757                         /* resync has finished, collect result */
3758                         md_unregister_thread(mddev->sync_thread);
3759                         mddev->sync_thread = NULL;
3760                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3761                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3762                                 /* success...*/
3763                                 /* activate any spares */
3764                                 mddev->pers->spare_active(mddev);
3765                         }
3766                         md_update_sb(mddev);
3767
3768                         /* if array is no-longer degraded, then any saved_raid_disk
3769                          * information must be scrapped
3770                          */
3771                         if (!mddev->degraded)
3772                                 ITERATE_RDEV(mddev,rdev,rtmp)
3773                                         rdev->saved_raid_disk = -1;
3774
3775                         mddev->recovery = 0;
3776                         /* flag recovery needed just to double check */
3777                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3778                         goto unlock;
3779                 }
3780                 if (mddev->recovery)
3781                         /* probably just the RECOVERY_NEEDED flag */
3782                         mddev->recovery = 0;
3783
3784                 /* no recovery is running.
3785                  * remove any failed drives, then
3786                  * add spares if possible.
3787                  * Spare are also removed and re-added, to allow
3788                  * the personality to fail the re-add.
3789                  */
3790                 ITERATE_RDEV(mddev,rdev,rtmp)
3791                         if (rdev->raid_disk >= 0 &&
3792                             (rdev->faulty || ! rdev->in_sync) &&
3793                             atomic_read(&rdev->nr_pending)==0) {
3794                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3795                                         rdev->raid_disk = -1;
3796                         }
3797
3798                 if (mddev->degraded) {
3799                         ITERATE_RDEV(mddev,rdev,rtmp)
3800                                 if (rdev->raid_disk < 0
3801                                     && !rdev->faulty) {
3802                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3803                                                 spares++;
3804                                         else
3805                                                 break;
3806                                 }
3807                 }
3808
3809                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3810                         /* nothing we can do ... */
3811                         goto unlock;
3812                 }
3813                 if (mddev->pers->sync_request) {
3814                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3815                         if (!spares)
3816                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3817                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3818                                 /* We are adding a device or devices to an array
3819                                  * which has the bitmap stored on all devices.
3820                                  * So make sure all bitmap pages get written
3821                                  */
3822                                 bitmap_write_all(mddev->bitmap);
3823                         }
3824                         mddev->sync_thread = md_register_thread(md_do_sync,
3825                                                                 mddev,
3826                                                                 "%s_resync");
3827                         if (!mddev->sync_thread) {
3828                                 printk(KERN_ERR "%s: could not start resync"
3829                                         " thread...\n",
3830                                         mdname(mddev));
3831                                 /* leave the spares where they are, it shouldn't hurt */
3832                                 mddev->recovery = 0;
3833                         } else {
3834                                 md_wakeup_thread(mddev->sync_thread);
3835                         }
3836                 }
3837         unlock:
3838                 mddev_unlock(mddev);
3839         }
3840 }
3841
3842 static int md_notify_reboot(struct notifier_block *this,
3843                             unsigned long code, void *x)
3844 {
3845         struct list_head *tmp;
3846         mddev_t *mddev;
3847
3848         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3849
3850                 printk(KERN_INFO "md: stopping all md devices.\n");
3851
3852                 ITERATE_MDDEV(mddev,tmp)
3853                         if (mddev_trylock(mddev)==0)
3854                                 do_md_stop (mddev, 1);
3855                 /*
3856                  * certain more exotic SCSI devices are known to be
3857                  * volatile wrt too early system reboots. While the
3858                  * right place to handle this issue is the given
3859                  * driver, we do want to have a safe RAID driver ...
3860                  */
3861                 mdelay(1000*1);
3862         }
3863         return NOTIFY_DONE;
3864 }
3865
3866 static struct notifier_block md_notifier = {
3867         .notifier_call  = md_notify_reboot,
3868         .next           = NULL,
3869         .priority       = INT_MAX, /* before any real devices */
3870 };
3871
3872 static void md_geninit(void)
3873 {
3874         struct proc_dir_entry *p;
3875
3876         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3877
3878         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3879         if (p)
3880                 p->proc_fops = &md_seq_fops;
3881 }
3882
3883 static int __init md_init(void)
3884 {
3885         int minor;
3886
3887         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3888                         " MD_SB_DISKS=%d\n",
3889                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3890                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3891         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3892                         BITMAP_MINOR);
3893
3894         if (register_blkdev(MAJOR_NR, "md"))
3895                 return -1;
3896         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3897                 unregister_blkdev(MAJOR_NR, "md");
3898                 return -1;
3899         }
3900         devfs_mk_dir("md");
3901         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3902                                 md_probe, NULL, NULL);
3903         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3904                             md_probe, NULL, NULL);
3905
3906         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3907                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3908                                 S_IFBLK|S_IRUSR|S_IWUSR,
3909                                 "md/%d", minor);
3910
3911         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3912                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3913                               S_IFBLK|S_IRUSR|S_IWUSR,
3914                               "md/mdp%d", minor);
3915
3916
3917         register_reboot_notifier(&md_notifier);
3918         raid_table_header = register_sysctl_table(raid_root_table, 1);
3919
3920         md_geninit();
3921         return (0);
3922 }
3923
3924
3925 #ifndef MODULE
3926
3927 /*
3928  * Searches all registered partitions for autorun RAID arrays
3929  * at boot time.
3930  */
3931 static dev_t detected_devices[128];
3932 static int dev_cnt;
3933
3934 void md_autodetect_dev(dev_t dev)
3935 {
3936         if (dev_cnt >= 0 && dev_cnt < 127)
3937                 detected_devices[dev_cnt++] = dev;
3938 }
3939
3940
3941 static void autostart_arrays(int part)
3942 {
3943         mdk_rdev_t *rdev;
3944         int i;
3945
3946         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3947
3948         for (i = 0; i < dev_cnt; i++) {
3949                 dev_t dev = detected_devices[i];
3950
3951                 rdev = md_import_device(dev,0, 0);
3952                 if (IS_ERR(rdev))
3953                         continue;
3954
3955                 if (rdev->faulty) {
3956                         MD_BUG();
3957                         continue;
3958                 }
3959                 list_add(&rdev->same_set, &pending_raid_disks);
3960         }
3961         dev_cnt = 0;
3962
3963         autorun_devices(part);
3964 }
3965
3966 #endif
3967
3968 static __exit void md_exit(void)
3969 {
3970         mddev_t *mddev;
3971         struct list_head *tmp;
3972         int i;
3973         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3974         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3975         for (i=0; i < MAX_MD_DEVS; i++)
3976                 devfs_remove("md/%d", i);
3977         for (i=0; i < MAX_MD_DEVS; i++)
3978                 devfs_remove("md/d%d", i);
3979
3980         devfs_remove("md");
3981
3982         unregister_blkdev(MAJOR_NR,"md");
3983         unregister_blkdev(mdp_major, "mdp");
3984         unregister_reboot_notifier(&md_notifier);
3985         unregister_sysctl_table(raid_table_header);
3986         remove_proc_entry("mdstat", NULL);
3987         ITERATE_MDDEV(mddev,tmp) {
3988                 struct gendisk *disk = mddev->gendisk;
3989                 if (!disk)
3990                         continue;
3991                 export_array(mddev);
3992                 del_gendisk(disk);
3993                 put_disk(disk);
3994                 mddev->gendisk = NULL;
3995                 mddev_put(mddev);
3996         }
3997 }
3998
3999 module_init(md_init)
4000 module_exit(md_exit)
4001
4002 EXPORT_SYMBOL(register_md_personality);
4003 EXPORT_SYMBOL(unregister_md_personality);
4004 EXPORT_SYMBOL(md_error);
4005 EXPORT_SYMBOL(md_done_sync);
4006 EXPORT_SYMBOL(md_write_start);
4007 EXPORT_SYMBOL(md_write_end);
4008 EXPORT_SYMBOL(md_register_thread);
4009 EXPORT_SYMBOL(md_unregister_thread);
4010 EXPORT_SYMBOL(md_wakeup_thread);
4011 EXPORT_SYMBOL(md_print_devices);
4012 EXPORT_SYMBOL(md_check_recovery);
4013 MODULE_LICENSE("GPL");
4014 MODULE_ALIAS("md");