err.no Git - linux-2.6/blob - 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         md_wakeup_thread(mddev->thread);
 260 }
 261
 262 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 263 {
 264         mdk_rdev_t * rdev;
 265         struct list_head *tmp;
 266
 267         ITERATE_RDEV(mddev,rdev,tmp) {
 268                 if (rdev->desc_nr == nr)
 269                         return rdev;
 270         }
 271         return NULL;
 272 }
 273
 274 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 275 {
 276         struct list_head *tmp;
 277         mdk_rdev_t *rdev;
 278
 279         ITERATE_RDEV(mddev,rdev,tmp) {
 280                 if (rdev->bdev->bd_dev == dev)
 281                         return rdev;
 282         }
 283         return NULL;
 284 }
 285
 286 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 287 {
 288         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 289         return MD_NEW_SIZE_BLOCKS(size);
 290 }
 291
 292 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 293 {
 294         sector_t size;
 295
 296         size = rdev->sb_offset;
 297
 298         if (chunk_size)
 299                 size &= ~((sector_t)chunk_size/1024 - 1);
 300         return size;
 301 }
 302
 303 static int alloc_disk_sb(mdk_rdev_t * rdev)
 304 {
 305         if (rdev->sb_page)
 306                 MD_BUG();
 307
 308         rdev->sb_page = alloc_page(GFP_KERNEL);
 309         if (!rdev->sb_page) {
 310                 printk(KERN_ALERT "md: out of memory.\n");
 311                 return -EINVAL;
 312         }
 313
 314         return 0;
 315 }
 316
 317 static void free_disk_sb(mdk_rdev_t * rdev)
 318 {
 319         if (rdev->sb_page) {
 320                 page_cache_release(rdev->sb_page);
 321                 rdev->sb_loaded = 0;
 322                 rdev->sb_page = NULL;
 323                 rdev->sb_offset = 0;
 324                 rdev->size = 0;
 325         }
 326 }
 327
 328
 329 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 330 {
 331         mdk_rdev_t *rdev = bio->bi_private;
 332         if (bio->bi_size)
 333                 return 1;
 334
 335         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 336                 md_error(rdev->mddev, rdev);
 337
 338         if (atomic_dec_and_test(&rdev->mddev->pending_writes))
 339                 wake_up(&rdev->mddev->sb_wait);
 340         bio_put(bio);
 341         return 0;
 342 }
 343
 344 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 345                    sector_t sector, int size, struct page *page)
 346 {
 347         /* write first size bytes of page to sector of rdev
 348          * Increment mddev->pending_writes before returning
 349          * and decrement it on completion, waking up sb_wait
 350          * if zero is reached.
 351          * If an error occurred, call md_error
 352          */
 353         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 354
 355         bio->bi_bdev = rdev->bdev;
 356         bio->bi_sector = sector;
 357         bio_add_page(bio, page, size, 0);
 358         bio->bi_private = rdev;
 359         bio->bi_end_io = super_written;
 360         atomic_inc(&mddev->pending_writes);
 361         submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
 362 }
 363
 364 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 365 {
 366         if (bio->bi_size)
 367                 return 1;
 368
 369         complete((struct completion*)bio->bi_private);
 370         return 0;
 371 }
 372
 373 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 374                    struct page *page, int rw)
 375 {
 376         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 377         struct completion event;
 378         int ret;
 379
 380         rw |= (1 << BIO_RW_SYNC);
 381
 382         bio->bi_bdev = bdev;
 383         bio->bi_sector = sector;
 384         bio_add_page(bio, page, size, 0);
 385         init_completion(&event);
 386         bio->bi_private = &event;
 387         bio->bi_end_io = bi_complete;
 388         submit_bio(rw, bio);
 389         wait_for_completion(&event);
 390
 391         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 392         bio_put(bio);
 393         return ret;
 394 }
 395
 396 static int read_disk_sb(mdk_rdev_t * rdev)
 397 {
 398         char b[BDEVNAME_SIZE];
 399         if (!rdev->sb_page) {
 400                 MD_BUG();
 401                 return -EINVAL;
 402         }
 403         if (rdev->sb_loaded)
 404                 return 0;
 405
 406
 407         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
 408                 goto fail;
 409         rdev->sb_loaded = 1;
 410         return 0;
 411
 412 fail:
 413         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 414                 bdevname(rdev->bdev,b));
 415         return -EINVAL;
 416 }
 417
 418 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 419 {
 420         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 421                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 422                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 423                 (sb1->set_uuid3 == sb2->set_uuid3))
 424
 425                 return 1;
 426
 427         return 0;
 428 }
 429
 430
 431 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 432 {
 433         int ret;
 434         mdp_super_t *tmp1, *tmp2;
 435
 436         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 437         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 438
 439         if (!tmp1 || !tmp2) {
 440                 ret = 0;
 441                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 442                 goto abort;
 443         }
 444
 445         *tmp1 = *sb1;
 446         *tmp2 = *sb2;
 447
 448         /*
 449          * nr_disks is not constant
 450          */
 451         tmp1->nr_disks = 0;
 452         tmp2->nr_disks = 0;
 453
 454         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 455                 ret = 0;
 456         else
 457                 ret = 1;
 458
 459 abort:
 460         kfree(tmp1);
 461         kfree(tmp2);
 462         return ret;
 463 }
 464
 465 static unsigned int calc_sb_csum(mdp_super_t * sb)
 466 {
 467         unsigned int disk_csum, csum;
 468
 469         disk_csum = sb->sb_csum;
 470         sb->sb_csum = 0;
 471         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 472         sb->sb_csum = disk_csum;
 473         return csum;
 474 }
 475
 476
 477 /*
 478  * Handle superblock details.
 479  * We want to be able to handle multiple superblock formats
 480  * so we have a common interface to them all, and an array of
 481  * different handlers.
 482  * We rely on user-space to write the initial superblock, and support
 483  * reading and updating of superblocks.
 484  * Interface methods are:
 485  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 486  *      loads and validates a superblock on dev.
 487  *      if refdev != NULL, compare superblocks on both devices
 488  *    Return:
 489  *      0 - dev has a superblock that is compatible with refdev
 490  *      1 - dev has a superblock that is compatible and newer than refdev
 491  *          so dev should be used as the refdev in future
 492  *     -EINVAL superblock incompatible or invalid
 493  *     -othererror e.g. -EIO
 494  *
 495  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 496  *      Verify that dev is acceptable into mddev.
 497  *       The first time, mddev->raid_disks will be 0, and data from
 498  *       dev should be merged in.  Subsequent calls check that dev
 499  *       is new enough.  Return 0 or -EINVAL
 500  *
 501  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 502  *     Update the superblock for rdev with data in mddev
 503  *     This does not write to disc.
 504  *
 505  */
 506
 507 struct super_type  {
 508         char            *name;
 509         struct module   *owner;
 510         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 511         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 512         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 513 };
 514
 515 /*
 516  * load_super for 0.90.0
 517  */
 518 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 519 {
 520         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 521         mdp_super_t *sb;
 522         int ret;
 523         sector_t sb_offset;
 524
 525         /*
 526          * Calculate the position of the superblock,
 527          * it's at the end of the disk.
 528          *
 529          * It also happens to be a multiple of 4Kb.
 530          */
 531         sb_offset = calc_dev_sboffset(rdev->bdev);
 532         rdev->sb_offset = sb_offset;
 533
 534         ret = read_disk_sb(rdev);
 535         if (ret) return ret;
 536
 537         ret = -EINVAL;
 538
 539         bdevname(rdev->bdev, b);
 540         sb = (mdp_super_t*)page_address(rdev->sb_page);
 541
 542         if (sb->md_magic != MD_SB_MAGIC) {
 543                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 544                        b);
 545                 goto abort;
 546         }
 547
 548         if (sb->major_version != 0 ||
 549             sb->minor_version != 90) {
 550                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 551                         sb->major_version, sb->minor_version,
 552                         b);
 553                 goto abort;
 554         }
 555
 556         if (sb->raid_disks <= 0)
 557                 goto abort;
 558
 559         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 560                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 561                         b);
 562                 goto abort;
 563         }
 564
 565         rdev->preferred_minor = sb->md_minor;
 566         rdev->data_offset = 0;
 567
 568         if (sb->level == LEVEL_MULTIPATH)
 569                 rdev->desc_nr = -1;
 570         else
 571                 rdev->desc_nr = sb->this_disk.number;
 572
 573         if (refdev == 0)
 574                 ret = 1;
 575         else {
 576                 __u64 ev1, ev2;
 577                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 578                 if (!uuid_equal(refsb, sb)) {
 579                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 580                                 b, bdevname(refdev->bdev,b2));
 581                         goto abort;
 582                 }
 583                 if (!sb_equal(refsb, sb)) {
 584                         printk(KERN_WARNING "md: %s has same UUID"
 585                                " but different superblock to %s\n",
 586                                b, bdevname(refdev->bdev, b2));
 587                         goto abort;
 588                 }
 589                 ev1 = md_event(sb);
 590                 ev2 = md_event(refsb);
 591                 if (ev1 > ev2)
 592                         ret = 1;
 593                 else
 594                         ret = 0;
 595         }
 596         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 597
 598  abort:
 599         return ret;
 600 }
 601
 602 /*
 603  * validate_super for 0.90.0
 604  */
 605 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 606 {
 607         mdp_disk_t *desc;
 608         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 609
 610         rdev->raid_disk = -1;
 611         rdev->in_sync = 0;
 612         if (mddev->raid_disks == 0) {
 613                 mddev->major_version = 0;
 614                 mddev->minor_version = sb->minor_version;
 615                 mddev->patch_version = sb->patch_version;
 616                 mddev->persistent = ! sb->not_persistent;
 617                 mddev->chunk_size = sb->chunk_size;
 618                 mddev->ctime = sb->ctime;
 619                 mddev->utime = sb->utime;
 620                 mddev->level = sb->level;
 621                 mddev->layout = sb->layout;
 622                 mddev->raid_disks = sb->raid_disks;
 623                 mddev->size = sb->size;
 624                 mddev->events = md_event(sb);
 625                 mddev->bitmap_offset = 0;
 626                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 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 = mddev->default_bitmap_offset;
 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                 mddev->default_bitmap_offset = 0;
 944                 if (mddev->minor_version == 0)
 945                         mddev->default_bitmap_offset = -(64*1024)/512;
 946
 947                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 948                 memcpy(mddev->uuid, sb->set_uuid, 16);
 949
 950                 mddev->max_disks =  (4096-256)/2;
 951
 952                 if ((le32_to_cpu(sb->feature_map) & 1) &&
 953                     mddev->bitmap_file == NULL ) {
 954                         if (mddev->level != 1) {
 955                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
 956                                 return -EINVAL;
 957                         }
 958                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
 959                 }
 960         } else if (mddev->pers == NULL) {
 961                 /* Insist of good event counter while assembling */
 962                 __u64 ev1 = le64_to_cpu(sb->events);
 963                 ++ev1;
 964                 if (ev1 < mddev->events)
 965                         return -EINVAL;
 966         } else if (mddev->bitmap) {
 967                 /* If adding to array with a bitmap, then we can accept an
 968                  * older device, but not too old.
 969                  */
 970                 __u64 ev1 = le64_to_cpu(sb->events);
 971                 if (ev1 < mddev->bitmap->events_cleared)
 972                         return 0;
 973         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 974                 return 0;
 975
 976         if (mddev->level != LEVEL_MULTIPATH) {
 977                 int role;
 978                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
 979                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
 980                 switch(role) {
 981                 case 0xffff: /* spare */
 982                         rdev->faulty = 0;
 983                         break;
 984                 case 0xfffe: /* faulty */
 985                         rdev->faulty = 1;
 986                         break;
 987                 default:
 988                         rdev->in_sync = 1;
 989                         rdev->faulty = 0;
 990                         rdev->raid_disk = role;
 991                         break;
 992                 }
 993         } else /* MULTIPATH are always insync */
 994                 rdev->in_sync = 1;
 995
 996         return 0;
 997 }
 998
 999 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1000 {
1001         struct mdp_superblock_1 *sb;
1002         struct list_head *tmp;
1003         mdk_rdev_t *rdev2;
1004         int max_dev, i;
1005         /* make rdev->sb match mddev and rdev data. */
1006
1007         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1008
1009         sb->feature_map = 0;
1010         sb->pad0 = 0;
1011         memset(sb->pad1, 0, sizeof(sb->pad1));
1012         memset(sb->pad2, 0, sizeof(sb->pad2));
1013         memset(sb->pad3, 0, sizeof(sb->pad3));
1014
1015         sb->utime = cpu_to_le64((__u64)mddev->utime);
1016         sb->events = cpu_to_le64(mddev->events);
1017         if (mddev->in_sync)
1018                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1019         else
1020                 sb->resync_offset = cpu_to_le64(0);
1021
1022         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1023                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1024                 sb->feature_map = cpu_to_le32(1);
1025         }
1026
1027         max_dev = 0;
1028         ITERATE_RDEV(mddev,rdev2,tmp)
1029                 if (rdev2->desc_nr+1 > max_dev)
1030                         max_dev = rdev2->desc_nr+1;
1031
1032         sb->max_dev = cpu_to_le32(max_dev);
1033         for (i=0; i<max_dev;i++)
1034                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1035
1036         ITERATE_RDEV(mddev,rdev2,tmp) {
1037                 i = rdev2->desc_nr;
1038                 if (rdev2->faulty)
1039                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1040                 else if (rdev2->in_sync)
1041                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1042                 else
1043                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1044         }
1045
1046         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1047         sb->sb_csum = calc_sb_1_csum(sb);
1048 }
1049
1050
1051 static struct super_type super_types[] = {
1052         [0] = {
1053                 .name   = "0.90.0",
1054                 .owner  = THIS_MODULE,
1055                 .load_super     = super_90_load,
1056                 .validate_super = super_90_validate,
1057                 .sync_super     = super_90_sync,
1058         },
1059         [1] = {
1060                 .name   = "md-1",
1061                 .owner  = THIS_MODULE,
1062                 .load_super     = super_1_load,
1063                 .validate_super = super_1_validate,
1064                 .sync_super     = super_1_sync,
1065         },
1066 };
1067
1068 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1069 {
1070         struct list_head *tmp;
1071         mdk_rdev_t *rdev;
1072
1073         ITERATE_RDEV(mddev,rdev,tmp)
1074                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1075                         return rdev;
1076
1077         return NULL;
1078 }
1079
1080 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1081 {
1082         struct list_head *tmp;
1083         mdk_rdev_t *rdev;
1084
1085         ITERATE_RDEV(mddev1,rdev,tmp)
1086                 if (match_dev_unit(mddev2, rdev))
1087                         return 1;
1088
1089         return 0;
1090 }
1091
1092 static LIST_HEAD(pending_raid_disks);
1093
1094 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1095 {
1096         mdk_rdev_t *same_pdev;
1097         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1098
1099         if (rdev->mddev) {
1100                 MD_BUG();
1101                 return -EINVAL;
1102         }
1103         same_pdev = match_dev_unit(mddev, rdev);
1104         if (same_pdev)
1105                 printk(KERN_WARNING
1106                         "%s: WARNING: %s appears to be on the same physical"
1107                         " disk as %s. True\n     protection against single-disk"
1108                         " failure might be compromised.\n",
1109                         mdname(mddev), bdevname(rdev->bdev,b),
1110                         bdevname(same_pdev->bdev,b2));
1111
1112         /* Verify rdev->desc_nr is unique.
1113          * If it is -1, assign a free number, else
1114          * check number is not in use
1115          */
1116         if (rdev->desc_nr < 0) {
1117                 int choice = 0;
1118                 if (mddev->pers) choice = mddev->raid_disks;
1119                 while (find_rdev_nr(mddev, choice))
1120                         choice++;
1121                 rdev->desc_nr = choice;
1122         } else {
1123                 if (find_rdev_nr(mddev, rdev->desc_nr))
1124                         return -EBUSY;
1125         }
1126
1127         list_add(&rdev->same_set, &mddev->disks);
1128         rdev->mddev = mddev;
1129         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1130         return 0;
1131 }
1132
1133 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1134 {
1135         char b[BDEVNAME_SIZE];
1136         if (!rdev->mddev) {
1137                 MD_BUG();
1138                 return;
1139         }
1140         list_del_init(&rdev->same_set);
1141         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1142         rdev->mddev = NULL;
1143 }
1144
1145 /*
1146  * prevent the device from being mounted, repartitioned or
1147  * otherwise reused by a RAID array (or any other kernel
1148  * subsystem), by bd_claiming the device.
1149  */
1150 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1151 {
1152         int err = 0;
1153         struct block_device *bdev;
1154         char b[BDEVNAME_SIZE];
1155
1156         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1157         if (IS_ERR(bdev)) {
1158                 printk(KERN_ERR "md: could not open %s.\n",
1159                         __bdevname(dev, b));
1160                 return PTR_ERR(bdev);
1161         }
1162         err = bd_claim(bdev, rdev);
1163         if (err) {
1164                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1165                         bdevname(bdev, b));
1166                 blkdev_put(bdev);
1167                 return err;
1168         }
1169         rdev->bdev = bdev;
1170         return err;
1171 }
1172
1173 static void unlock_rdev(mdk_rdev_t *rdev)
1174 {
1175         struct block_device *bdev = rdev->bdev;
1176         rdev->bdev = NULL;
1177         if (!bdev)
1178                 MD_BUG();
1179         bd_release(bdev);
1180         blkdev_put(bdev);
1181 }
1182
1183 void md_autodetect_dev(dev_t dev);
1184
1185 static void export_rdev(mdk_rdev_t * rdev)
1186 {
1187         char b[BDEVNAME_SIZE];
1188         printk(KERN_INFO "md: export_rdev(%s)\n",
1189                 bdevname(rdev->bdev,b));
1190         if (rdev->mddev)
1191                 MD_BUG();
1192         free_disk_sb(rdev);
1193         list_del_init(&rdev->same_set);
1194 #ifndef MODULE
1195         md_autodetect_dev(rdev->bdev->bd_dev);
1196 #endif
1197         unlock_rdev(rdev);
1198         kfree(rdev);
1199 }
1200
1201 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1202 {
1203         unbind_rdev_from_array(rdev);
1204         export_rdev(rdev);
1205 }
1206
1207 static void export_array(mddev_t *mddev)
1208 {
1209         struct list_head *tmp;
1210         mdk_rdev_t *rdev;
1211
1212         ITERATE_RDEV(mddev,rdev,tmp) {
1213                 if (!rdev->mddev) {
1214                         MD_BUG();
1215                         continue;
1216                 }
1217                 kick_rdev_from_array(rdev);
1218         }
1219         if (!list_empty(&mddev->disks))
1220                 MD_BUG();
1221         mddev->raid_disks = 0;
1222         mddev->major_version = 0;
1223 }
1224
1225 static void print_desc(mdp_disk_t *desc)
1226 {
1227         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1228                 desc->major,desc->minor,desc->raid_disk,desc->state);
1229 }
1230
1231 static void print_sb(mdp_super_t *sb)
1232 {
1233         int i;
1234
1235         printk(KERN_INFO
1236                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1237                 sb->major_version, sb->minor_version, sb->patch_version,
1238                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1239                 sb->ctime);
1240         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1241                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1242                 sb->md_minor, sb->layout, sb->chunk_size);
1243         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1244                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1245                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1246                 sb->failed_disks, sb->spare_disks,
1247                 sb->sb_csum, (unsigned long)sb->events_lo);
1248
1249         printk(KERN_INFO);
1250         for (i = 0; i < MD_SB_DISKS; i++) {
1251                 mdp_disk_t *desc;
1252
1253                 desc = sb->disks + i;
1254                 if (desc->number || desc->major || desc->minor ||
1255                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1256                         printk("     D %2d: ", i);
1257                         print_desc(desc);
1258                 }
1259         }
1260         printk(KERN_INFO "md:     THIS: ");
1261         print_desc(&sb->this_disk);
1262
1263 }
1264
1265 static void print_rdev(mdk_rdev_t *rdev)
1266 {
1267         char b[BDEVNAME_SIZE];
1268         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1269                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1270                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1271         if (rdev->sb_loaded) {
1272                 printk(KERN_INFO "md: rdev superblock:\n");
1273                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1274         } else
1275                 printk(KERN_INFO "md: no rdev superblock!\n");
1276 }
1277
1278 void md_print_devices(void)
1279 {
1280         struct list_head *tmp, *tmp2;
1281         mdk_rdev_t *rdev;
1282         mddev_t *mddev;
1283         char b[BDEVNAME_SIZE];
1284
1285         printk("\n");
1286         printk("md:     **********************************\n");
1287         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1288         printk("md:     **********************************\n");
1289         ITERATE_MDDEV(mddev,tmp) {
1290
1291                 if (mddev->bitmap)
1292                         bitmap_print_sb(mddev->bitmap);
1293                 else
1294                         printk("%s: ", mdname(mddev));
1295                 ITERATE_RDEV(mddev,rdev,tmp2)
1296                         printk("<%s>", bdevname(rdev->bdev,b));
1297                 printk("\n");
1298
1299                 ITERATE_RDEV(mddev,rdev,tmp2)
1300                         print_rdev(rdev);
1301         }
1302         printk("md:     **********************************\n");
1303         printk("\n");
1304 }
1305
1306
1307 static void sync_sbs(mddev_t * mddev)
1308 {
1309         mdk_rdev_t *rdev;
1310         struct list_head *tmp;
1311
1312         ITERATE_RDEV(mddev,rdev,tmp) {
1313                 super_types[mddev->major_version].
1314                         sync_super(mddev, rdev);
1315                 rdev->sb_loaded = 1;
1316         }
1317 }
1318
1319 static void md_update_sb(mddev_t * mddev)
1320 {
1321         int err;
1322         struct list_head *tmp;
1323         mdk_rdev_t *rdev;
1324         int sync_req;
1325
1326 repeat:
1327         spin_lock(&mddev->write_lock);
1328         sync_req = mddev->in_sync;
1329         mddev->utime = get_seconds();
1330         mddev->events ++;
1331
1332         if (!mddev->events) {
1333                 /*
1334                  * oops, this 64-bit counter should never wrap.
1335                  * Either we are in around ~1 trillion A.C., assuming
1336                  * 1 reboot per second, or we have a bug:
1337                  */
1338                 MD_BUG();
1339                 mddev->events --;
1340         }
1341         mddev->sb_dirty = 2;
1342         sync_sbs(mddev);
1343
1344         /*
1345          * do not write anything to disk if using
1346          * nonpersistent superblocks
1347          */
1348         if (!mddev->persistent) {
1349                 mddev->sb_dirty = 0;
1350                 spin_unlock(&mddev->write_lock);
1351                 wake_up(&mddev->sb_wait);
1352                 return;
1353         }
1354         spin_unlock(&mddev->write_lock);
1355
1356         dprintk(KERN_INFO
1357                 "md: updating %s RAID superblock on device (in sync %d)\n",
1358                 mdname(mddev),mddev->in_sync);
1359
1360         err = bitmap_update_sb(mddev->bitmap);
1361         ITERATE_RDEV(mddev,rdev,tmp) {
1362                 char b[BDEVNAME_SIZE];
1363                 dprintk(KERN_INFO "md: ");
1364                 if (rdev->faulty)
1365                         dprintk("(skipping faulty ");
1366
1367                 dprintk("%s ", bdevname(rdev->bdev,b));
1368                 if (!rdev->faulty) {
1369                         md_super_write(mddev,rdev,
1370                                        rdev->sb_offset<<1, MD_SB_BYTES,
1371                                        rdev->sb_page);
1372                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1373                                 bdevname(rdev->bdev,b),
1374                                 (unsigned long long)rdev->sb_offset);
1375
1376                 } else
1377                         dprintk(")\n");
1378                 if (mddev->level == LEVEL_MULTIPATH)
1379                         /* only need to write one superblock... */
1380                         break;
1381         }
1382         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1383         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1384
1385         spin_lock(&mddev->write_lock);
1386         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1387                 /* have to write it out again */
1388                 spin_unlock(&mddev->write_lock);
1389                 goto repeat;
1390         }
1391         mddev->sb_dirty = 0;
1392         spin_unlock(&mddev->write_lock);
1393         wake_up(&mddev->sb_wait);
1394
1395 }
1396
1397 /*
1398  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1399  *
1400  * mark the device faulty if:
1401  *
1402  *   - the device is nonexistent (zero size)
1403  *   - the device has no valid superblock
1404  *
1405  * a faulty rdev _never_ has rdev->sb set.
1406  */
1407 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1408 {
1409         char b[BDEVNAME_SIZE];
1410         int err;
1411         mdk_rdev_t *rdev;
1412         sector_t size;
1413
1414         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1415         if (!rdev) {
1416                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1417                 return ERR_PTR(-ENOMEM);
1418         }
1419         memset(rdev, 0, sizeof(*rdev));
1420
1421         if ((err = alloc_disk_sb(rdev)))
1422                 goto abort_free;
1423
1424         err = lock_rdev(rdev, newdev);
1425         if (err)
1426                 goto abort_free;
1427
1428         rdev->desc_nr = -1;
1429         rdev->faulty = 0;
1430         rdev->in_sync = 0;
1431         rdev->data_offset = 0;
1432         atomic_set(&rdev->nr_pending, 0);
1433
1434         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1435         if (!size) {
1436                 printk(KERN_WARNING
1437                         "md: %s has zero or unknown size, marking faulty!\n",
1438                         bdevname(rdev->bdev,b));
1439                 err = -EINVAL;
1440                 goto abort_free;
1441         }
1442
1443         if (super_format >= 0) {
1444                 err = super_types[super_format].
1445                         load_super(rdev, NULL, super_minor);
1446                 if (err == -EINVAL) {
1447                         printk(KERN_WARNING
1448                                 "md: %s has invalid sb, not importing!\n",
1449                                 bdevname(rdev->bdev,b));
1450                         goto abort_free;
1451                 }
1452                 if (err < 0) {
1453                         printk(KERN_WARNING
1454                                 "md: could not read %s's sb, not importing!\n",
1455                                 bdevname(rdev->bdev,b));
1456                         goto abort_free;
1457                 }
1458         }
1459         INIT_LIST_HEAD(&rdev->same_set);
1460
1461         return rdev;
1462
1463 abort_free:
1464         if (rdev->sb_page) {
1465                 if (rdev->bdev)
1466                         unlock_rdev(rdev);
1467                 free_disk_sb(rdev);
1468         }
1469         kfree(rdev);
1470         return ERR_PTR(err);
1471 }
1472
1473 /*
1474  * Check a full RAID array for plausibility
1475  */
1476
1477
1478 static void analyze_sbs(mddev_t * mddev)
1479 {
1480         int i;
1481         struct list_head *tmp;
1482         mdk_rdev_t *rdev, *freshest;
1483         char b[BDEVNAME_SIZE];
1484
1485         freshest = NULL;
1486         ITERATE_RDEV(mddev,rdev,tmp)
1487                 switch (super_types[mddev->major_version].
1488                         load_super(rdev, freshest, mddev->minor_version)) {
1489                 case 1:
1490                         freshest = rdev;
1491                         break;
1492                 case 0:
1493                         break;
1494                 default:
1495                         printk( KERN_ERR \
1496                                 "md: fatal superblock inconsistency in %s"
1497                                 " -- removing from array\n",
1498                                 bdevname(rdev->bdev,b));
1499                         kick_rdev_from_array(rdev);
1500                 }
1501
1502
1503         super_types[mddev->major_version].
1504                 validate_super(mddev, freshest);
1505
1506         i = 0;
1507         ITERATE_RDEV(mddev,rdev,tmp) {
1508                 if (rdev != freshest)
1509                         if (super_types[mddev->major_version].
1510                             validate_super(mddev, rdev)) {
1511                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1512                                         " from array!\n",
1513                                         bdevname(rdev->bdev,b));
1514                                 kick_rdev_from_array(rdev);
1515                                 continue;
1516                         }
1517                 if (mddev->level == LEVEL_MULTIPATH) {
1518                         rdev->desc_nr = i++;
1519                         rdev->raid_disk = rdev->desc_nr;
1520                         rdev->in_sync = 1;
1521                 }
1522         }
1523
1524
1525
1526         if (mddev->recovery_cp != MaxSector &&
1527             mddev->level >= 1)
1528                 printk(KERN_ERR "md: %s: raid array is not clean"
1529                        " -- starting background reconstruction\n",
1530                        mdname(mddev));
1531
1532 }
1533
1534 int mdp_major = 0;
1535
1536 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1537 {
1538         static DECLARE_MUTEX(disks_sem);
1539         mddev_t *mddev = mddev_find(dev);
1540         struct gendisk *disk;
1541         int partitioned = (MAJOR(dev) != MD_MAJOR);
1542         int shift = partitioned ? MdpMinorShift : 0;
1543         int unit = MINOR(dev) >> shift;
1544
1545         if (!mddev)
1546                 return NULL;
1547
1548         down(&disks_sem);
1549         if (mddev->gendisk) {
1550                 up(&disks_sem);
1551                 mddev_put(mddev);
1552                 return NULL;
1553         }
1554         disk = alloc_disk(1 << shift);
1555         if (!disk) {
1556                 up(&disks_sem);
1557                 mddev_put(mddev);
1558                 return NULL;
1559         }
1560         disk->major = MAJOR(dev);
1561         disk->first_minor = unit << shift;
1562         if (partitioned) {
1563                 sprintf(disk->disk_name, "md_d%d", unit);
1564                 sprintf(disk->devfs_name, "md/d%d", unit);
1565         } else {
1566                 sprintf(disk->disk_name, "md%d", unit);
1567                 sprintf(disk->devfs_name, "md/%d", unit);
1568         }
1569         disk->fops = &md_fops;
1570         disk->private_data = mddev;
1571         disk->queue = mddev->queue;
1572         add_disk(disk);
1573         mddev->gendisk = disk;
1574         up(&disks_sem);
1575         return NULL;
1576 }
1577
1578 void md_wakeup_thread(mdk_thread_t *thread);
1579
1580 static void md_safemode_timeout(unsigned long data)
1581 {
1582         mddev_t *mddev = (mddev_t *) data;
1583
1584         mddev->safemode = 1;
1585         md_wakeup_thread(mddev->thread);
1586 }
1587
1588
1589 static int do_md_run(mddev_t * mddev)
1590 {
1591         int pnum, err;
1592         int chunk_size;
1593         struct list_head *tmp;
1594         mdk_rdev_t *rdev;
1595         struct gendisk *disk;
1596         char b[BDEVNAME_SIZE];
1597
1598         if (list_empty(&mddev->disks))
1599                 /* cannot run an array with no devices.. */
1600                 return -EINVAL;
1601
1602         if (mddev->pers)
1603                 return -EBUSY;
1604
1605         /*
1606          * Analyze all RAID superblock(s)
1607          */
1608         if (!mddev->raid_disks)
1609                 analyze_sbs(mddev);
1610
1611         chunk_size = mddev->chunk_size;
1612         pnum = level_to_pers(mddev->level);
1613
1614         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1615                 if (!chunk_size) {
1616                         /*
1617                          * 'default chunksize' in the old md code used to
1618                          * be PAGE_SIZE, baaad.
1619                          * we abort here to be on the safe side. We don't
1620                          * want to continue the bad practice.
1621                          */
1622                         printk(KERN_ERR
1623                                 "no chunksize specified, see 'man raidtab'\n");
1624                         return -EINVAL;
1625                 }
1626                 if (chunk_size > MAX_CHUNK_SIZE) {
1627                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1628                                 chunk_size, MAX_CHUNK_SIZE);
1629                         return -EINVAL;
1630                 }
1631                 /*
1632                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1633                  */
1634                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1635                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1636                         return -EINVAL;
1637                 }
1638                 if (chunk_size < PAGE_SIZE) {
1639                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1640                                 chunk_size, PAGE_SIZE);
1641                         return -EINVAL;
1642                 }
1643
1644                 /* devices must have minimum size of one chunk */
1645                 ITERATE_RDEV(mddev,rdev,tmp) {
1646                         if (rdev->faulty)
1647                                 continue;
1648                         if (rdev->size < chunk_size / 1024) {
1649                                 printk(KERN_WARNING
1650                                         "md: Dev %s smaller than chunk_size:"
1651                                         " %lluk < %dk\n",
1652                                         bdevname(rdev->bdev,b),
1653                                         (unsigned long long)rdev->size,
1654                                         chunk_size / 1024);
1655                                 return -EINVAL;
1656                         }
1657                 }
1658         }
1659
1660 #ifdef CONFIG_KMOD
1661         if (!pers[pnum])
1662         {
1663                 request_module("md-personality-%d", pnum);
1664         }
1665 #endif
1666
1667         /*
1668          * Drop all container device buffers, from now on
1669          * the only valid external interface is through the md
1670          * device.
1671          * Also find largest hardsector size
1672          */
1673         ITERATE_RDEV(mddev,rdev,tmp) {
1674                 if (rdev->faulty)
1675                         continue;
1676                 sync_blockdev(rdev->bdev);
1677                 invalidate_bdev(rdev->bdev, 0);
1678         }
1679
1680         md_probe(mddev->unit, NULL, NULL);
1681         disk = mddev->gendisk;
1682         if (!disk)
1683                 return -ENOMEM;
1684
1685         spin_lock(&pers_lock);
1686         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1687                 spin_unlock(&pers_lock);
1688                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1689                        pnum);
1690                 return -EINVAL;
1691         }
1692
1693         mddev->pers = pers[pnum];
1694         spin_unlock(&pers_lock);
1695
1696         mddev->recovery = 0;
1697         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1698
1699         /* before we start the array running, initialise the bitmap */
1700         err = bitmap_create(mddev);
1701         if (err)
1702                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1703                         mdname(mddev), err);
1704         else
1705                 err = mddev->pers->run(mddev);
1706         if (err) {
1707                 printk(KERN_ERR "md: pers->run() failed ...\n");
1708                 module_put(mddev->pers->owner);
1709                 mddev->pers = NULL;
1710                 bitmap_destroy(mddev);
1711                 return err;
1712         }
1713         atomic_set(&mddev->writes_pending,0);
1714         mddev->safemode = 0;
1715         mddev->safemode_timer.function = md_safemode_timeout;
1716         mddev->safemode_timer.data = (unsigned long) mddev;
1717         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1718         mddev->in_sync = 1;
1719
1720         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1721         md_wakeup_thread(mddev->thread);
1722
1723         if (mddev->sb_dirty)
1724                 md_update_sb(mddev);
1725
1726         set_capacity(disk, mddev->array_size<<1);
1727
1728         /* If we call blk_queue_make_request here, it will
1729          * re-initialise max_sectors etc which may have been
1730          * refined inside -> run.  So just set the bits we need to set.
1731          * Most initialisation happended when we called
1732          * blk_queue_make_request(..., md_fail_request)
1733          * earlier.
1734          */
1735         mddev->queue->queuedata = mddev;
1736         mddev->queue->make_request_fn = mddev->pers->make_request;
1737
1738         mddev->changed = 1;
1739         return 0;
1740 }
1741
1742 static int restart_array(mddev_t *mddev)
1743 {
1744         struct gendisk *disk = mddev->gendisk;
1745         int err;
1746
1747         /*
1748          * Complain if it has no devices
1749          */
1750         err = -ENXIO;
1751         if (list_empty(&mddev->disks))
1752                 goto out;
1753
1754         if (mddev->pers) {
1755                 err = -EBUSY;
1756                 if (!mddev->ro)
1757                         goto out;
1758
1759                 mddev->safemode = 0;
1760                 mddev->ro = 0;
1761                 set_disk_ro(disk, 0);
1762
1763                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1764                         mdname(mddev));
1765                 /*
1766                  * Kick recovery or resync if necessary
1767                  */
1768                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1769                 md_wakeup_thread(mddev->thread);
1770                 err = 0;
1771         } else {
1772                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1773                         mdname(mddev));
1774                 err = -EINVAL;
1775         }
1776
1777 out:
1778         return err;
1779 }
1780
1781 static int do_md_stop(mddev_t * mddev, int ro)
1782 {
1783         int err = 0;
1784         struct gendisk *disk = mddev->gendisk;
1785
1786         if (mddev->pers) {
1787                 if (atomic_read(&mddev->active)>2) {
1788                         printk("md: %s still in use.\n",mdname(mddev));
1789                         return -EBUSY;
1790                 }
1791
1792                 if (mddev->sync_thread) {
1793                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1794                         md_unregister_thread(mddev->sync_thread);
1795                         mddev->sync_thread = NULL;
1796                 }
1797
1798                 del_timer_sync(&mddev->safemode_timer);
1799
1800                 invalidate_partition(disk, 0);
1801
1802                 if (ro) {
1803                         err  = -ENXIO;
1804                         if (mddev->ro)
1805                                 goto out;
1806                         mddev->ro = 1;
1807                 } else {
1808                         bitmap_flush(mddev);
1809                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1810                         if (mddev->ro)
1811                                 set_disk_ro(disk, 0);
1812                         blk_queue_make_request(mddev->queue, md_fail_request);
1813                         mddev->pers->stop(mddev);
1814                         module_put(mddev->pers->owner);
1815                         mddev->pers = NULL;
1816                         if (mddev->ro)
1817                                 mddev->ro = 0;
1818                 }
1819                 if (!mddev->in_sync) {
1820                         /* mark array as shutdown cleanly */
1821                         mddev->in_sync = 1;
1822                         md_update_sb(mddev);
1823                 }
1824                 if (ro)
1825                         set_disk_ro(disk, 1);
1826         }
1827
1828         bitmap_destroy(mddev);
1829         if (mddev->bitmap_file) {
1830                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1831                 fput(mddev->bitmap_file);
1832                 mddev->bitmap_file = NULL;
1833         }
1834         mddev->bitmap_offset = 0;
1835
1836         /*
1837          * Free resources if final stop
1838          */
1839         if (!ro) {
1840                 struct gendisk *disk;
1841                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1842
1843                 export_array(mddev);
1844
1845                 mddev->array_size = 0;
1846                 disk = mddev->gendisk;
1847                 if (disk)
1848                         set_capacity(disk, 0);
1849                 mddev->changed = 1;
1850         } else
1851                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1852                         mdname(mddev));
1853         err = 0;
1854 out:
1855         return err;
1856 }
1857
1858 static void autorun_array(mddev_t *mddev)
1859 {
1860         mdk_rdev_t *rdev;
1861         struct list_head *tmp;
1862         int err;
1863
1864         if (list_empty(&mddev->disks))
1865                 return;
1866
1867         printk(KERN_INFO "md: running: ");
1868
1869         ITERATE_RDEV(mddev,rdev,tmp) {
1870                 char b[BDEVNAME_SIZE];
1871                 printk("<%s>", bdevname(rdev->bdev,b));
1872         }
1873         printk("\n");
1874
1875         err = do_md_run (mddev);
1876         if (err) {
1877                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1878                 do_md_stop (mddev, 0);
1879         }
1880 }
1881
1882 /*
1883  * lets try to run arrays based on all disks that have arrived
1884  * until now. (those are in pending_raid_disks)
1885  *
1886  * the method: pick the first pending disk, collect all disks with
1887  * the same UUID, remove all from the pending list and put them into
1888  * the 'same_array' list. Then order this list based on superblock
1889  * update time (freshest comes first), kick out 'old' disks and
1890  * compare superblocks. If everything's fine then run it.
1891  *
1892  * If "unit" is allocated, then bump its reference count
1893  */
1894 static void autorun_devices(int part)
1895 {
1896         struct list_head candidates;
1897         struct list_head *tmp;
1898         mdk_rdev_t *rdev0, *rdev;
1899         mddev_t *mddev;
1900         char b[BDEVNAME_SIZE];
1901
1902         printk(KERN_INFO "md: autorun ...\n");
1903         while (!list_empty(&pending_raid_disks)) {
1904                 dev_t dev;
1905                 rdev0 = list_entry(pending_raid_disks.next,
1906                                          mdk_rdev_t, same_set);
1907
1908                 printk(KERN_INFO "md: considering %s ...\n",
1909                         bdevname(rdev0->bdev,b));
1910                 INIT_LIST_HEAD(&candidates);
1911                 ITERATE_RDEV_PENDING(rdev,tmp)
1912                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1913                                 printk(KERN_INFO "md:  adding %s ...\n",
1914                                         bdevname(rdev->bdev,b));
1915                                 list_move(&rdev->same_set, &candidates);
1916                         }
1917                 /*
1918                  * now we have a set of devices, with all of them having
1919                  * mostly sane superblocks. It's time to allocate the
1920                  * mddev.
1921                  */
1922                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1923                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1924                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1925                         break;
1926                 }
1927                 if (part)
1928                         dev = MKDEV(mdp_major,
1929                                     rdev0->preferred_minor << MdpMinorShift);
1930                 else
1931                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1932
1933                 md_probe(dev, NULL, NULL);
1934                 mddev = mddev_find(dev);
1935                 if (!mddev) {
1936                         printk(KERN_ERR
1937                                 "md: cannot allocate memory for md drive.\n");
1938                         break;
1939                 }
1940                 if (mddev_lock(mddev))
1941                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1942                                mdname(mddev));
1943                 else if (mddev->raid_disks || mddev->major_version
1944                          || !list_empty(&mddev->disks)) {
1945                         printk(KERN_WARNING
1946                                 "md: %s already running, cannot run %s\n",
1947                                 mdname(mddev), bdevname(rdev0->bdev,b));
1948                         mddev_unlock(mddev);
1949                 } else {
1950                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1951                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1952                                 list_del_init(&rdev->same_set);
1953                                 if (bind_rdev_to_array(rdev, mddev))
1954                                         export_rdev(rdev);
1955                         }
1956                         autorun_array(mddev);
1957                         mddev_unlock(mddev);
1958                 }
1959                 /* on success, candidates will be empty, on error
1960                  * it won't...
1961                  */
1962                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1963                         export_rdev(rdev);
1964                 mddev_put(mddev);
1965         }
1966         printk(KERN_INFO "md: ... autorun DONE.\n");
1967 }
1968
1969 /*
1970  * import RAID devices based on one partition
1971  * if possible, the array gets run as well.
1972  */
1973
1974 static int autostart_array(dev_t startdev)
1975 {
1976         char b[BDEVNAME_SIZE];
1977         int err = -EINVAL, i;
1978         mdp_super_t *sb = NULL;
1979         mdk_rdev_t *start_rdev = NULL, *rdev;
1980
1981         start_rdev = md_import_device(startdev, 0, 0);
1982         if (IS_ERR(start_rdev))
1983                 return err;
1984
1985
1986         /* NOTE: this can only work for 0.90.0 superblocks */
1987         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1988         if (sb->major_version != 0 ||
1989             sb->minor_version != 90 ) {
1990                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1991                 export_rdev(start_rdev);
1992                 return err;
1993         }
1994
1995         if (start_rdev->faulty) {
1996                 printk(KERN_WARNING
1997                         "md: can not autostart based on faulty %s!\n",
1998                         bdevname(start_rdev->bdev,b));
1999                 export_rdev(start_rdev);
2000                 return err;
2001         }
2002         list_add(&start_rdev->same_set, &pending_raid_disks);
2003
2004         for (i = 0; i < MD_SB_DISKS; i++) {
2005                 mdp_disk_t *desc = sb->disks + i;
2006                 dev_t dev = MKDEV(desc->major, desc->minor);
2007
2008                 if (!dev)
2009                         continue;
2010                 if (dev == startdev)
2011                         continue;
2012                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2013                         continue;
2014                 rdev = md_import_device(dev, 0, 0);
2015                 if (IS_ERR(rdev))
2016                         continue;
2017
2018                 list_add(&rdev->same_set, &pending_raid_disks);
2019         }
2020
2021         /*
2022          * possibly return codes
2023          */
2024         autorun_devices(0);
2025         return 0;
2026
2027 }
2028
2029
2030 static int get_version(void __user * arg)
2031 {
2032         mdu_version_t ver;
2033
2034         ver.major = MD_MAJOR_VERSION;
2035         ver.minor = MD_MINOR_VERSION;
2036         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2037
2038         if (copy_to_user(arg, &ver, sizeof(ver)))
2039                 return -EFAULT;
2040
2041         return 0;
2042 }
2043
2044 static int get_array_info(mddev_t * mddev, void __user * arg)
2045 {
2046         mdu_array_info_t info;
2047         int nr,working,active,failed,spare;
2048         mdk_rdev_t *rdev;
2049         struct list_head *tmp;
2050
2051         nr=working=active=failed=spare=0;
2052         ITERATE_RDEV(mddev,rdev,tmp) {
2053                 nr++;
2054                 if (rdev->faulty)
2055                         failed++;
2056                 else {
2057                         working++;
2058                         if (rdev->in_sync)
2059                                 active++;
2060                         else
2061                                 spare++;
2062                 }
2063         }
2064
2065         info.major_version = mddev->major_version;
2066         info.minor_version = mddev->minor_version;
2067         info.patch_version = MD_PATCHLEVEL_VERSION;
2068         info.ctime         = mddev->ctime;
2069         info.level         = mddev->level;
2070         info.size          = mddev->size;
2071         info.nr_disks      = nr;
2072         info.raid_disks    = mddev->raid_disks;
2073         info.md_minor      = mddev->md_minor;
2074         info.not_persistent= !mddev->persistent;
2075
2076         info.utime         = mddev->utime;
2077         info.state         = 0;
2078         if (mddev->in_sync)
2079                 info.state = (1<<MD_SB_CLEAN);
2080         if (mddev->bitmap && mddev->bitmap_offset)
2081                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2082         info.active_disks  = active;
2083         info.working_disks = working;
2084         info.failed_disks  = failed;
2085         info.spare_disks   = spare;
2086
2087         info.layout        = mddev->layout;
2088         info.chunk_size    = mddev->chunk_size;
2089
2090         if (copy_to_user(arg, &info, sizeof(info)))
2091                 return -EFAULT;
2092
2093         return 0;
2094 }
2095
2096 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2097 {
2098         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2099         char *ptr, *buf = NULL;
2100         int err = -ENOMEM;
2101
2102         file = kmalloc(sizeof(*file), GFP_KERNEL);
2103         if (!file)
2104                 goto out;
2105
2106         /* bitmap disabled, zero the first byte and copy out */
2107         if (!mddev->bitmap || !mddev->bitmap->file) {
2108                 file->pathname[0] = '\0';
2109                 goto copy_out;
2110         }
2111
2112         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2113         if (!buf)
2114                 goto out;
2115
2116         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2117         if (!ptr)
2118                 goto out;
2119
2120         strcpy(file->pathname, ptr);
2121
2122 copy_out:
2123         err = 0;
2124         if (copy_to_user(arg, file, sizeof(*file)))
2125                 err = -EFAULT;
2126 out:
2127         kfree(buf);
2128         kfree(file);
2129         return err;
2130 }
2131
2132 static int get_disk_info(mddev_t * mddev, void __user * arg)
2133 {
2134         mdu_disk_info_t info;
2135         unsigned int nr;
2136         mdk_rdev_t *rdev;
2137
2138         if (copy_from_user(&info, arg, sizeof(info)))
2139                 return -EFAULT;
2140
2141         nr = info.number;
2142
2143         rdev = find_rdev_nr(mddev, nr);
2144         if (rdev) {
2145                 info.major = MAJOR(rdev->bdev->bd_dev);
2146                 info.minor = MINOR(rdev->bdev->bd_dev);
2147                 info.raid_disk = rdev->raid_disk;
2148                 info.state = 0;
2149                 if (rdev->faulty)
2150                         info.state |= (1<<MD_DISK_FAULTY);
2151                 else if (rdev->in_sync) {
2152                         info.state |= (1<<MD_DISK_ACTIVE);
2153                         info.state |= (1<<MD_DISK_SYNC);
2154                 }
2155         } else {
2156                 info.major = info.minor = 0;
2157                 info.raid_disk = -1;
2158                 info.state = (1<<MD_DISK_REMOVED);
2159         }
2160
2161         if (copy_to_user(arg, &info, sizeof(info)))
2162                 return -EFAULT;
2163
2164         return 0;
2165 }
2166
2167 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2168 {
2169         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2170         mdk_rdev_t *rdev;
2171         dev_t dev = MKDEV(info->major,info->minor);
2172
2173         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2174                 return -EOVERFLOW;
2175
2176         if (!mddev->raid_disks) {
2177                 int err;
2178                 /* expecting a device which has a superblock */
2179                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2180                 if (IS_ERR(rdev)) {
2181                         printk(KERN_WARNING
2182                                 "md: md_import_device returned %ld\n",
2183                                 PTR_ERR(rdev));
2184                         return PTR_ERR(rdev);
2185                 }
2186                 if (!list_empty(&mddev->disks)) {
2187                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2188                                                         mdk_rdev_t, same_set);
2189                         int err = super_types[mddev->major_version]
2190                                 .load_super(rdev, rdev0, mddev->minor_version);
2191                         if (err < 0) {
2192                                 printk(KERN_WARNING
2193                                         "md: %s has different UUID to %s\n",
2194                                         bdevname(rdev->bdev,b),
2195                                         bdevname(rdev0->bdev,b2));
2196                                 export_rdev(rdev);
2197                                 return -EINVAL;
2198                         }
2199                 }
2200                 err = bind_rdev_to_array(rdev, mddev);
2201                 if (err)
2202                         export_rdev(rdev);
2203                 return err;
2204         }
2205
2206         /*
2207          * add_new_disk can be used once the array is assembled
2208          * to add "hot spares".  They must already have a superblock
2209          * written
2210          */
2211         if (mddev->pers) {
2212                 int err;
2213                 if (!mddev->pers->hot_add_disk) {
2214                         printk(KERN_WARNING
2215                                 "%s: personality does not support diskops!\n",
2216                                mdname(mddev));
2217                         return -EINVAL;
2218                 }
2219                 rdev = md_import_device(dev, mddev->major_version,
2220                                         mddev->minor_version);
2221                 if (IS_ERR(rdev)) {
2222                         printk(KERN_WARNING
2223                                 "md: md_import_device returned %ld\n",
2224                                 PTR_ERR(rdev));
2225                         return PTR_ERR(rdev);
2226                 }
2227                 /* set save_raid_disk if appropriate */
2228                 if (!mddev->persistent) {
2229                         if (info->state & (1<<MD_DISK_SYNC)  &&
2230                             info->raid_disk < mddev->raid_disks)
2231                                 rdev->raid_disk = info->raid_disk;
2232                         else
2233                                 rdev->raid_disk = -1;
2234                 } else
2235                         super_types[mddev->major_version].
2236                                 validate_super(mddev, rdev);
2237                 rdev->saved_raid_disk = rdev->raid_disk;
2238
2239                 rdev->in_sync = 0; /* just to be sure */
2240                 rdev->raid_disk = -1;
2241                 err = bind_rdev_to_array(rdev, mddev);
2242                 if (err)
2243                         export_rdev(rdev);
2244
2245                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2246                 md_wakeup_thread(mddev->thread);
2247                 return err;
2248         }
2249
2250         /* otherwise, add_new_disk is only allowed
2251          * for major_version==0 superblocks
2252          */
2253         if (mddev->major_version != 0) {
2254                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2255                        mdname(mddev));
2256                 return -EINVAL;
2257         }
2258
2259         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2260                 int err;
2261                 rdev = md_import_device (dev, -1, 0);
2262                 if (IS_ERR(rdev)) {
2263                         printk(KERN_WARNING
2264                                 "md: error, md_import_device() returned %ld\n",
2265                                 PTR_ERR(rdev));
2266                         return PTR_ERR(rdev);
2267                 }
2268                 rdev->desc_nr = info->number;
2269                 if (info->raid_disk < mddev->raid_disks)
2270                         rdev->raid_disk = info->raid_disk;
2271                 else
2272                         rdev->raid_disk = -1;
2273
2274                 rdev->faulty = 0;
2275                 if (rdev->raid_disk < mddev->raid_disks)
2276                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2277                 else
2278                         rdev->in_sync = 0;
2279
2280                 err = bind_rdev_to_array(rdev, mddev);
2281                 if (err) {
2282                         export_rdev(rdev);
2283                         return err;
2284                 }
2285
2286                 if (!mddev->persistent) {
2287                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2288                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2289                 } else
2290                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2291                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2292
2293                 if (!mddev->size || (mddev->size > rdev->size))
2294                         mddev->size = rdev->size;
2295         }
2296
2297         return 0;
2298 }
2299
2300 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2301 {
2302         char b[BDEVNAME_SIZE];
2303         mdk_rdev_t *rdev;
2304
2305         if (!mddev->pers)
2306                 return -ENODEV;
2307
2308         rdev = find_rdev(mddev, dev);
2309         if (!rdev)
2310                 return -ENXIO;
2311
2312         if (rdev->raid_disk >= 0)
2313                 goto busy;
2314
2315         kick_rdev_from_array(rdev);
2316         md_update_sb(mddev);
2317
2318         return 0;
2319 busy:
2320         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2321                 bdevname(rdev->bdev,b), mdname(mddev));
2322         return -EBUSY;
2323 }
2324
2325 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2326 {
2327         char b[BDEVNAME_SIZE];
2328         int err;
2329         unsigned int size;
2330         mdk_rdev_t *rdev;
2331
2332         if (!mddev->pers)
2333                 return -ENODEV;
2334
2335         if (mddev->major_version != 0) {
2336                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2337                         " version-0 superblocks.\n",
2338                         mdname(mddev));
2339                 return -EINVAL;
2340         }
2341         if (!mddev->pers->hot_add_disk) {
2342                 printk(KERN_WARNING
2343                         "%s: personality does not support diskops!\n",
2344                         mdname(mddev));
2345                 return -EINVAL;
2346         }
2347
2348         rdev = md_import_device (dev, -1, 0);
2349         if (IS_ERR(rdev)) {
2350                 printk(KERN_WARNING
2351                         "md: error, md_import_device() returned %ld\n",
2352                         PTR_ERR(rdev));
2353                 return -EINVAL;
2354         }
2355
2356         if (mddev->persistent)
2357                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2358         else
2359                 rdev->sb_offset =
2360                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2361
2362         size = calc_dev_size(rdev, mddev->chunk_size);
2363         rdev->size = size;
2364
2365         if (size < mddev->size) {
2366                 printk(KERN_WARNING
2367                         "%s: disk size %llu blocks < array size %llu\n",
2368                         mdname(mddev), (unsigned long long)size,
2369                         (unsigned long long)mddev->size);
2370                 err = -ENOSPC;
2371                 goto abort_export;
2372         }
2373
2374         if (rdev->faulty) {
2375                 printk(KERN_WARNING
2376                         "md: can not hot-add faulty %s disk to %s!\n",
2377                         bdevname(rdev->bdev,b), mdname(mddev));
2378                 err = -EINVAL;
2379                 goto abort_export;
2380         }
2381         rdev->in_sync = 0;
2382         rdev->desc_nr = -1;
2383         bind_rdev_to_array(rdev, mddev);
2384
2385         /*
2386          * The rest should better be atomic, we can have disk failures
2387          * noticed in interrupt contexts ...
2388          */
2389
2390         if (rdev->desc_nr == mddev->max_disks) {
2391                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2392                         mdname(mddev));
2393                 err = -EBUSY;
2394                 goto abort_unbind_export;
2395         }
2396
2397         rdev->raid_disk = -1;
2398
2399         md_update_sb(mddev);
2400
2401         /*
2402          * Kick recovery, maybe this spare has to be added to the
2403          * array immediately.
2404          */
2405         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2406         md_wakeup_thread(mddev->thread);
2407
2408         return 0;
2409
2410 abort_unbind_export:
2411         unbind_rdev_from_array(rdev);
2412
2413 abort_export:
2414         export_rdev(rdev);
2415         return err;
2416 }
2417
2418 /* similar to deny_write_access, but accounts for our holding a reference
2419  * to the file ourselves */
2420 static int deny_bitmap_write_access(struct file * file)
2421 {
2422         struct inode *inode = file->f_mapping->host;
2423
2424         spin_lock(&inode->i_lock);
2425         if (atomic_read(&inode->i_writecount) > 1) {
2426                 spin_unlock(&inode->i_lock);
2427                 return -ETXTBSY;
2428         }
2429         atomic_set(&inode->i_writecount, -1);
2430         spin_unlock(&inode->i_lock);
2431
2432         return 0;
2433 }
2434
2435 static int set_bitmap_file(mddev_t *mddev, int fd)
2436 {
2437         int err;
2438
2439         if (mddev->pers) {
2440                 if (!mddev->pers->quiesce)
2441                         return -EBUSY;
2442                 if (mddev->recovery || mddev->sync_thread)
2443                         return -EBUSY;
2444                 /* we should be able to change the bitmap.. */
2445         }
2446
2447
2448         if (fd >= 0) {
2449                 if (mddev->bitmap)
2450                         return -EEXIST; /* cannot add when bitmap is present */
2451                 mddev->bitmap_file = fget(fd);
2452
2453                 if (mddev->bitmap_file == NULL) {
2454                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2455                                mdname(mddev));
2456                         return -EBADF;
2457                 }
2458
2459                 err = deny_bitmap_write_access(mddev->bitmap_file);
2460                 if (err) {
2461                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2462                                mdname(mddev));
2463                         fput(mddev->bitmap_file);
2464                         mddev->bitmap_file = NULL;
2465                         return err;
2466                 }
2467                 mddev->bitmap_offset = 0; /* file overrides offset */
2468         } else if (mddev->bitmap == NULL)
2469                 return -ENOENT; /* cannot remove what isn't there */
2470         err = 0;
2471         if (mddev->pers) {
2472                 mddev->pers->quiesce(mddev, 1);
2473                 if (fd >= 0)
2474                         err = bitmap_create(mddev);
2475                 if (fd < 0 || err)
2476                         bitmap_destroy(mddev);
2477                 mddev->pers->quiesce(mddev, 0);
2478         } else if (fd < 0) {
2479                 if (mddev->bitmap_file)
2480                         fput(mddev->bitmap_file);
2481                 mddev->bitmap_file = NULL;
2482         }
2483
2484         return err;
2485 }
2486
2487 /*
2488  * set_array_info is used two different ways
2489  * The original usage is when creating a new array.
2490  * In this usage, raid_disks is > 0 and it together with
2491  *  level, size, not_persistent,layout,chunksize determine the
2492  *  shape of the array.
2493  *  This will always create an array with a type-0.90.0 superblock.
2494  * The newer usage is when assembling an array.
2495  *  In this case raid_disks will be 0, and the major_version field is
2496  *  use to determine which style super-blocks are to be found on the devices.
2497  *  The minor and patch _version numbers are also kept incase the
2498  *  super_block handler wishes to interpret them.
2499  */
2500 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2501 {
2502
2503         if (info->raid_disks == 0) {
2504                 /* just setting version number for superblock loading */
2505                 if (info->major_version < 0 ||
2506                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2507                     super_types[info->major_version].name == NULL) {
2508                         /* maybe try to auto-load a module? */
2509                         printk(KERN_INFO
2510                                 "md: superblock version %d not known\n",
2511                                 info->major_version);
2512                         return -EINVAL;
2513                 }
2514                 mddev->major_version = info->major_version;
2515                 mddev->minor_version = info->minor_version;
2516                 mddev->patch_version = info->patch_version;
2517                 return 0;
2518         }
2519         mddev->major_version = MD_MAJOR_VERSION;
2520         mddev->minor_version = MD_MINOR_VERSION;
2521         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2522         mddev->ctime         = get_seconds();
2523
2524         mddev->level         = info->level;
2525         mddev->size          = info->size;
2526         mddev->raid_disks    = info->raid_disks;
2527         /* don't set md_minor, it is determined by which /dev/md* was
2528          * openned
2529          */
2530         if (info->state & (1<<MD_SB_CLEAN))
2531                 mddev->recovery_cp = MaxSector;
2532         else
2533                 mddev->recovery_cp = 0;
2534         mddev->persistent    = ! info->not_persistent;
2535
2536         mddev->layout        = info->layout;
2537         mddev->chunk_size    = info->chunk_size;
2538
2539         mddev->max_disks     = MD_SB_DISKS;
2540
2541         mddev->sb_dirty      = 1;
2542
2543         /*
2544          * Generate a 128 bit UUID
2545          */
2546         get_random_bytes(mddev->uuid, 16);
2547
2548         return 0;
2549 }
2550
2551 /*
2552  * update_array_info is used to change the configuration of an
2553  * on-line array.
2554  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2555  * fields in the info are checked against the array.
2556  * Any differences that cannot be handled will cause an error.
2557  * Normally, only one change can be managed at a time.
2558  */
2559 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2560 {
2561         int rv = 0;
2562         int cnt = 0;
2563         int state = 0;
2564
2565         /* calculate expected state,ignoring low bits */
2566         if (mddev->bitmap && mddev->bitmap_offset)
2567                 state |= (1 << MD_SB_BITMAP_PRESENT);
2568
2569         if (mddev->major_version != info->major_version ||
2570             mddev->minor_version != info->minor_version ||
2571 /*          mddev->patch_version != info->patch_version || */
2572             mddev->ctime         != info->ctime         ||
2573             mddev->level         != info->level         ||
2574 /*          mddev->layout        != info->layout        || */
2575             !mddev->persistent   != info->not_persistent||
2576             mddev->chunk_size    != info->chunk_size    ||
2577             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2578             ((state^info->state) & 0xfffffe00)
2579                 )
2580                 return -EINVAL;
2581         /* Check there is only one change */
2582         if (mddev->size != info->size) cnt++;
2583         if (mddev->raid_disks != info->raid_disks) cnt++;
2584         if (mddev->layout != info->layout) cnt++;
2585         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2586         if (cnt == 0) return 0;
2587         if (cnt > 1) return -EINVAL;
2588
2589         if (mddev->layout != info->layout) {
2590                 /* Change layout
2591                  * we don't need to do anything at the md level, the
2592                  * personality will take care of it all.
2593                  */
2594                 if (mddev->pers->reconfig == NULL)
2595                         return -EINVAL;
2596                 else
2597                         return mddev->pers->reconfig(mddev, info->layout, -1);
2598         }
2599         if (mddev->size != info->size) {
2600                 mdk_rdev_t * rdev;
2601                 struct list_head *tmp;
2602                 if (mddev->pers->resize == NULL)
2603                         return -EINVAL;
2604                 /* The "size" is the amount of each device that is used.
2605                  * This can only make sense for arrays with redundancy.
2606                  * linear and raid0 always use whatever space is available
2607                  * We can only consider changing the size if no resync
2608                  * or reconstruction is happening, and if the new size
2609                  * is acceptable. It must fit before the sb_offset or,
2610                  * if that is <data_offset, it must fit before the
2611                  * size of each device.
2612                  * If size is zero, we find the largest size that fits.
2613                  */
2614                 if (mddev->sync_thread)
2615                         return -EBUSY;
2616                 ITERATE_RDEV(mddev,rdev,tmp) {
2617                         sector_t avail;
2618                         int fit = (info->size == 0);
2619                         if (rdev->sb_offset > rdev->data_offset)
2620                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2621                         else
2622                                 avail = get_capacity(rdev->bdev->bd_disk)
2623                                         - rdev->data_offset;
2624                         if (fit && (info->size == 0 || info->size > avail/2))
2625                                 info->size = avail/2;
2626                         if (avail < ((sector_t)info->size << 1))
2627                                 return -ENOSPC;
2628                 }
2629                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2630                 if (!rv) {
2631                         struct block_device *bdev;
2632
2633                         bdev = bdget_disk(mddev->gendisk, 0);
2634                         if (bdev) {
2635                                 down(&bdev->bd_inode->i_sem);
2636                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2637                                 up(&bdev->bd_inode->i_sem);
2638                                 bdput(bdev);
2639                         }
2640                 }
2641         }
2642         if (mddev->raid_disks    != info->raid_disks) {
2643                 /* change the number of raid disks */
2644                 if (mddev->pers->reshape == NULL)
2645                         return -EINVAL;
2646                 if (info->raid_disks <= 0 ||
2647                     info->raid_disks >= mddev->max_disks)
2648                         return -EINVAL;
2649                 if (mddev->sync_thread)
2650                         return -EBUSY;
2651                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2652                 if (!rv) {
2653                         struct block_device *bdev;
2654
2655                         bdev = bdget_disk(mddev->gendisk, 0);
2656                         if (bdev) {
2657                                 down(&bdev->bd_inode->i_sem);
2658                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2659                                 up(&bdev->bd_inode->i_sem);
2660                                 bdput(bdev);
2661                         }
2662                 }
2663         }
2664         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2665                 if (mddev->pers->quiesce == NULL)
2666                         return -EINVAL;
2667                 if (mddev->recovery || mddev->sync_thread)
2668                         return -EBUSY;
2669                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2670                         /* add the bitmap */
2671                         if (mddev->bitmap)
2672                                 return -EEXIST;
2673                         if (mddev->default_bitmap_offset == 0)
2674                                 return -EINVAL;
2675                         mddev->bitmap_offset = mddev->default_bitmap_offset;
2676                         mddev->pers->quiesce(mddev, 1);
2677                         rv = bitmap_create(mddev);
2678                         if (rv)
2679                                 bitmap_destroy(mddev);
2680                         mddev->pers->quiesce(mddev, 0);
2681                 } else {
2682                         /* remove the bitmap */
2683                         if (!mddev->bitmap)
2684                                 return -ENOENT;
2685                         if (mddev->bitmap->file)
2686                                 return -EINVAL;
2687                         mddev->pers->quiesce(mddev, 1);
2688                         bitmap_destroy(mddev);
2689                         mddev->pers->quiesce(mddev, 0);
2690                         mddev->bitmap_offset = 0;
2691                 }
2692         }
2693         md_update_sb(mddev);
2694         return rv;
2695 }
2696
2697 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2698 {
2699         mdk_rdev_t *rdev;
2700
2701         if (mddev->pers == NULL)
2702                 return -ENODEV;
2703
2704         rdev = find_rdev(mddev, dev);
2705         if (!rdev)
2706                 return -ENODEV;
2707
2708         md_error(mddev, rdev);
2709         return 0;
2710 }
2711
2712 static int md_ioctl(struct inode *inode, struct file *file,
2713                         unsigned int cmd, unsigned long arg)
2714 {
2715         int err = 0;
2716         void __user *argp = (void __user *)arg;
2717         struct hd_geometry __user *loc = argp;
2718         mddev_t *mddev = NULL;
2719
2720         if (!capable(CAP_SYS_ADMIN))
2721                 return -EACCES;
2722
2723         /*
2724          * Commands dealing with the RAID driver but not any
2725          * particular array:
2726          */
2727         switch (cmd)
2728         {
2729                 case RAID_VERSION:
2730                         err = get_version(argp);
2731                         goto done;
2732
2733                 case PRINT_RAID_DEBUG:
2734                         err = 0;
2735                         md_print_devices();
2736                         goto done;
2737
2738 #ifndef MODULE
2739                 case RAID_AUTORUN:
2740                         err = 0;
2741                         autostart_arrays(arg);
2742                         goto done;
2743 #endif
2744                 default:;
2745         }
2746
2747         /*
2748          * Commands creating/starting a new array:
2749          */
2750
2751         mddev = inode->i_bdev->bd_disk->private_data;
2752
2753         if (!mddev) {
2754                 BUG();
2755                 goto abort;
2756         }
2757
2758
2759         if (cmd == START_ARRAY) {
2760                 /* START_ARRAY doesn't need to lock the array as autostart_array
2761                  * does the locking, and it could even be a different array
2762                  */
2763                 static int cnt = 3;
2764                 if (cnt > 0 ) {
2765                         printk(KERN_WARNING
2766                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2767                                "This will not be supported beyond 2.6\n",
2768                                current->comm, current->pid);
2769                         cnt--;
2770                 }
2771                 err = autostart_array(new_decode_dev(arg));
2772                 if (err) {
2773                         printk(KERN_WARNING "md: autostart failed!\n");
2774                         goto abort;
2775                 }
2776                 goto done;
2777         }
2778
2779         err = mddev_lock(mddev);
2780         if (err) {
2781                 printk(KERN_INFO
2782                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2783                         err, cmd);
2784                 goto abort;
2785         }
2786
2787         switch (cmd)
2788         {
2789                 case SET_ARRAY_INFO:
2790                         {
2791                                 mdu_array_info_t info;
2792                                 if (!arg)
2793                                         memset(&info, 0, sizeof(info));
2794                                 else if (copy_from_user(&info, argp, sizeof(info))) {
2795                                         err = -EFAULT;
2796                                         goto abort_unlock;
2797                                 }
2798                                 if (mddev->pers) {
2799                                         err = update_array_info(mddev, &info);
2800                                         if (err) {
2801                                                 printk(KERN_WARNING "md: couldn't update"
2802                                                        " array info. %d\n", err);
2803                                                 goto abort_unlock;
2804                                         }
2805                                         goto done_unlock;
2806                                 }
2807                                 if (!list_empty(&mddev->disks)) {
2808                                         printk(KERN_WARNING
2809                                                "md: array %s already has disks!\n",
2810                                                mdname(mddev));
2811                                         err = -EBUSY;
2812                                         goto abort_unlock;
2813                                 }
2814                                 if (mddev->raid_disks) {
2815                                         printk(KERN_WARNING
2816                                                "md: array %s already initialised!\n",
2817                                                mdname(mddev));
2818                                         err = -EBUSY;
2819                                         goto abort_unlock;
2820                                 }
2821                                 err = set_array_info(mddev, &info);
2822                                 if (err) {
2823                                         printk(KERN_WARNING "md: couldn't set"
2824                                                " array info. %d\n", err);
2825                                         goto abort_unlock;
2826                                 }
2827                         }
2828                         goto done_unlock;
2829
2830                 default:;
2831         }
2832
2833         /*
2834          * Commands querying/configuring an existing array:
2835          */
2836         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2837          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2838         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2839                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2840                 err = -ENODEV;
2841                 goto abort_unlock;
2842         }
2843
2844         /*
2845          * Commands even a read-only array can execute:
2846          */
2847         switch (cmd)
2848         {
2849                 case GET_ARRAY_INFO:
2850                         err = get_array_info(mddev, argp);
2851                         goto done_unlock;
2852
2853                 case GET_BITMAP_FILE:
2854                         err = get_bitmap_file(mddev, argp);
2855                         goto done_unlock;
2856
2857                 case GET_DISK_INFO:
2858                         err = get_disk_info(mddev, argp);
2859                         goto done_unlock;
2860
2861                 case RESTART_ARRAY_RW:
2862                         err = restart_array(mddev);
2863                         goto done_unlock;
2864
2865                 case STOP_ARRAY:
2866                         err = do_md_stop (mddev, 0);
2867                         goto done_unlock;
2868
2869                 case STOP_ARRAY_RO:
2870                         err = do_md_stop (mddev, 1);
2871                         goto done_unlock;
2872
2873         /*
2874          * We have a problem here : there is no easy way to give a CHS
2875          * virtual geometry. We currently pretend that we have a 2 heads
2876          * 4 sectors (with a BIG number of cylinders...). This drives
2877          * dosfs just mad... ;-)
2878          */
2879                 case HDIO_GETGEO:
2880                         if (!loc) {
2881                                 err = -EINVAL;
2882                                 goto abort_unlock;
2883                         }
2884                         err = put_user (2, (char __user *) &loc->heads);
2885                         if (err)
2886                                 goto abort_unlock;
2887                         err = put_user (4, (char __user *) &loc->sectors);
2888                         if (err)
2889                                 goto abort_unlock;
2890                         err = put_user(get_capacity(mddev->gendisk)/8,
2891                                         (short __user *) &loc->cylinders);
2892                         if (err)
2893                                 goto abort_unlock;
2894                         err = put_user (get_start_sect(inode->i_bdev),
2895                                                 (long __user *) &loc->start);
2896                         goto done_unlock;
2897         }
2898
2899         /*
2900          * The remaining ioctls are changing the state of the
2901          * superblock, so we do not allow read-only arrays
2902          * here:
2903          */
2904         if (mddev->ro) {
2905                 err = -EROFS;
2906                 goto abort_unlock;
2907         }
2908
2909         switch (cmd)
2910         {
2911                 case ADD_NEW_DISK:
2912                 {
2913                         mdu_disk_info_t info;
2914                         if (copy_from_user(&info, argp, sizeof(info)))
2915                                 err = -EFAULT;
2916                         else
2917                                 err = add_new_disk(mddev, &info);
2918                         goto done_unlock;
2919                 }
2920
2921                 case HOT_REMOVE_DISK:
2922                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2923                         goto done_unlock;
2924
2925                 case HOT_ADD_DISK:
2926                         err = hot_add_disk(mddev, new_decode_dev(arg));
2927                         goto done_unlock;
2928
2929                 case SET_DISK_FAULTY:
2930                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2931                         goto done_unlock;
2932
2933                 case RUN_ARRAY:
2934                         err = do_md_run (mddev);
2935                         goto done_unlock;
2936
2937                 case SET_BITMAP_FILE:
2938                         err = set_bitmap_file(mddev, (int)arg);
2939                         goto done_unlock;
2940
2941                 default:
2942                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2943                                 printk(KERN_WARNING "md: %s(pid %d) used"
2944                                         " obsolete MD ioctl, upgrade your"
2945                                         " software to use new ictls.\n",
2946                                         current->comm, current->pid);
2947                         err = -EINVAL;
2948                         goto abort_unlock;
2949         }
2950
2951 done_unlock:
2952 abort_unlock:
2953         mddev_unlock(mddev);
2954
2955         return err;
2956 done:
2957         if (err)
2958                 MD_BUG();
2959 abort:
2960         return err;
2961 }
2962
2963 static int md_open(struct inode *inode, struct file *file)
2964 {
2965         /*
2966          * Succeed if we can lock the mddev, which confirms that
2967          * it isn't being stopped right now.
2968          */
2969         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2970         int err;
2971
2972         if ((err = mddev_lock(mddev)))
2973                 goto out;
2974
2975         err = 0;
2976         mddev_get(mddev);
2977         mddev_unlock(mddev);
2978
2979         check_disk_change(inode->i_bdev);
2980  out:
2981         return err;
2982 }
2983
2984 static int md_release(struct inode *inode, struct file * file)
2985 {
2986         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2987
2988         if (!mddev)
2989                 BUG();
2990         mddev_put(mddev);
2991
2992         return 0;
2993 }
2994
2995 static int md_media_changed(struct gendisk *disk)
2996 {
2997         mddev_t *mddev = disk->private_data;
2998
2999         return mddev->changed;
3000 }
3001
3002 static int md_revalidate(struct gendisk *disk)
3003 {
3004         mddev_t *mddev = disk->private_data;
3005
3006         mddev->changed = 0;
3007         return 0;
3008 }
3009 static struct block_device_operations md_fops =
3010 {
3011         .owner          = THIS_MODULE,
3012         .open           = md_open,
3013         .release        = md_release,
3014         .ioctl          = md_ioctl,
3015         .media_changed  = md_media_changed,
3016         .revalidate_disk= md_revalidate,
3017 };
3018
3019 static int md_thread(void * arg)
3020 {
3021         mdk_thread_t *thread = arg;
3022
3023         lock_kernel();
3024
3025         /*
3026          * Detach thread
3027          */
3028
3029         daemonize(thread->name, mdname(thread->mddev));
3030
3031         current->exit_signal = SIGCHLD;
3032         allow_signal(SIGKILL);
3033         thread->tsk = current;
3034
3035         /*
3036          * md_thread is a 'system-thread', it's priority should be very
3037          * high. We avoid resource deadlocks individually in each
3038          * raid personality. (RAID5 does preallocation) We also use RR and
3039          * the very same RT priority as kswapd, thus we will never get
3040          * into a priority inversion deadlock.
3041          *
3042          * we definitely have to have equal or higher priority than
3043          * bdflush, otherwise bdflush will deadlock if there are too
3044          * many dirty RAID5 blocks.
3045          */
3046         unlock_kernel();
3047
3048         complete(thread->event);
3049         while (thread->run) {
3050                 void (*run)(mddev_t *);
3051
3052                 wait_event_interruptible_timeout(thread->wqueue,
3053                                                  test_bit(THREAD_WAKEUP, &thread->flags),
3054                                                  thread->timeout);
3055                 try_to_freeze();
3056
3057                 clear_bit(THREAD_WAKEUP, &thread->flags);
3058
3059                 run = thread->run;
3060                 if (run)
3061                         run(thread->mddev);
3062
3063                 if (signal_pending(current))
3064                         flush_signals(current);
3065         }
3066         complete(thread->event);
3067         return 0;
3068 }
3069
3070 void md_wakeup_thread(mdk_thread_t *thread)
3071 {
3072         if (thread) {
3073                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3074                 set_bit(THREAD_WAKEUP, &thread->flags);
3075                 wake_up(&thread->wqueue);
3076         }
3077 }
3078
3079 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3080                                  const char *name)
3081 {
3082         mdk_thread_t *thread;
3083         int ret;
3084         struct completion event;
3085
3086         thread = (mdk_thread_t *) kmalloc
3087                                 (sizeof(mdk_thread_t), GFP_KERNEL);
3088         if (!thread)
3089                 return NULL;
3090
3091         memset(thread, 0, sizeof(mdk_thread_t));
3092         init_waitqueue_head(&thread->wqueue);
3093
3094         init_completion(&event);
3095         thread->event = &event;
3096         thread->run = run;
3097         thread->mddev = mddev;
3098         thread->name = name;
3099         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3100         ret = kernel_thread(md_thread, thread, 0);
3101         if (ret < 0) {
3102                 kfree(thread);
3103                 return NULL;
3104         }
3105         wait_for_completion(&event);
3106         return thread;
3107 }
3108
3109 void md_unregister_thread(mdk_thread_t *thread)
3110 {
3111         struct completion event;
3112
3113         init_completion(&event);
3114
3115         thread->event = &event;
3116
3117         /* As soon as ->run is set to NULL, the task could disappear,
3118          * so we need to hold tasklist_lock until we have sent the signal
3119          */
3120         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3121         read_lock(&tasklist_lock);
3122         thread->run = NULL;
3123         send_sig(SIGKILL, thread->tsk, 1);
3124         read_unlock(&tasklist_lock);
3125         wait_for_completion(&event);
3126         kfree(thread);
3127 }
3128
3129 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3130 {
3131         if (!mddev) {
3132                 MD_BUG();
3133                 return;
3134         }
3135
3136         if (!rdev || rdev->faulty)
3137                 return;
3138 /*
3139         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3140                 mdname(mddev),
3141                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3142                 __builtin_return_address(0),__builtin_return_address(1),
3143                 __builtin_return_address(2),__builtin_return_address(3));
3144 */
3145         if (!mddev->pers->error_handler)
3146                 return;
3147         mddev->pers->error_handler(mddev,rdev);
3148         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3149         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3150         md_wakeup_thread(mddev->thread);
3151 }
3152
3153 /* seq_file implementation /proc/mdstat */
3154
3155 static void status_unused(struct seq_file *seq)
3156 {
3157         int i = 0;
3158         mdk_rdev_t *rdev;
3159         struct list_head *tmp;
3160
3161         seq_printf(seq, "unused devices: ");
3162
3163         ITERATE_RDEV_PENDING(rdev,tmp) {
3164                 char b[BDEVNAME_SIZE];
3165                 i++;
3166                 seq_printf(seq, "%s ",
3167                               bdevname(rdev->bdev,b));
3168         }
3169         if (!i)
3170                 seq_printf(seq, "<none>");
3171
3172         seq_printf(seq, "\n");
3173 }
3174
3175
3176 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3177 {
3178         unsigned long max_blocks, resync, res, dt, db, rt;
3179
3180         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3181
3182         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3183                 max_blocks = mddev->resync_max_sectors >> 1;
3184         else
3185                 max_blocks = mddev->size;
3186
3187         /*
3188          * Should not happen.
3189          */
3190         if (!max_blocks) {
3191                 MD_BUG();
3192                 return;
3193         }
3194         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3195         {
3196                 int i, x = res/50, y = 20-x;
3197                 seq_printf(seq, "[");
3198                 for (i = 0; i < x; i++)
3199                         seq_printf(seq, "=");
3200                 seq_printf(seq, ">");
3201                 for (i = 0; i < y; i++)
3202                         seq_printf(seq, ".");
3203                 seq_printf(seq, "] ");
3204         }
3205         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3206                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3207                        "resync" : "recovery"),
3208                       res/10, res % 10, resync, max_blocks);
3209
3210         /*
3211          * We do not want to overflow, so the order of operands and
3212          * the * 100 / 100 trick are important. We do a +1 to be
3213          * safe against division by zero. We only estimate anyway.
3214          *
3215          * dt: time from mark until now
3216          * db: blocks written from mark until now
3217          * rt: remaining time
3218          */
3219         dt = ((jiffies - mddev->resync_mark) / HZ);
3220         if (!dt) dt++;
3221         db = resync - (mddev->resync_mark_cnt/2);
3222         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3223
3224         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3225
3226         seq_printf(seq, " speed=%ldK/sec", db/dt);
3227 }
3228
3229 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3230 {
3231         struct list_head *tmp;
3232         loff_t l = *pos;
3233         mddev_t *mddev;
3234
3235         if (l >= 0x10000)
3236                 return NULL;
3237         if (!l--)
3238                 /* header */
3239                 return (void*)1;
3240
3241         spin_lock(&all_mddevs_lock);
3242         list_for_each(tmp,&all_mddevs)
3243                 if (!l--) {
3244                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3245                         mddev_get(mddev);
3246                         spin_unlock(&all_mddevs_lock);
3247                         return mddev;
3248                 }
3249         spin_unlock(&all_mddevs_lock);
3250         if (!l--)
3251                 return (void*)2;/* tail */
3252         return NULL;
3253 }
3254
3255 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3256 {
3257         struct list_head *tmp;
3258         mddev_t *next_mddev, *mddev = v;
3259
3260         ++*pos;
3261         if (v == (void*)2)
3262                 return NULL;
3263
3264         spin_lock(&all_mddevs_lock);
3265         if (v == (void*)1)
3266                 tmp = all_mddevs.next;
3267         else
3268                 tmp = mddev->all_mddevs.next;
3269         if (tmp != &all_mddevs)
3270                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3271         else {
3272                 next_mddev = (void*)2;
3273                 *pos = 0x10000;
3274         }
3275         spin_unlock(&all_mddevs_lock);
3276
3277         if (v != (void*)1)
3278                 mddev_put(mddev);
3279         return next_mddev;
3280
3281 }
3282
3283 static void md_seq_stop(struct seq_file *seq, void *v)
3284 {
3285         mddev_t *mddev = v;
3286
3287         if (mddev && v != (void*)1 && v != (void*)2)
3288                 mddev_put(mddev);
3289 }
3290
3291 static int md_seq_show(struct seq_file *seq, void *v)
3292 {
3293         mddev_t *mddev = v;
3294         sector_t size;
3295         struct list_head *tmp2;
3296         mdk_rdev_t *rdev;
3297         int i;
3298         struct bitmap *bitmap;
3299
3300         if (v == (void*)1) {
3301                 seq_printf(seq, "Personalities : ");
3302                 spin_lock(&pers_lock);
3303                 for (i = 0; i < MAX_PERSONALITY; i++)
3304                         if (pers[i])
3305                                 seq_printf(seq, "[%s] ", pers[i]->name);
3306
3307                 spin_unlock(&pers_lock);
3308                 seq_printf(seq, "\n");
3309                 return 0;
3310         }
3311         if (v == (void*)2) {
3312                 status_unused(seq);
3313                 return 0;
3314         }
3315
3316         if (mddev_lock(mddev)!=0)
3317                 return -EINTR;
3318         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3319                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3320                                                 mddev->pers ? "" : "in");
3321                 if (mddev->pers) {
3322                         if (mddev->ro)
3323                                 seq_printf(seq, " (read-only)");
3324                         seq_printf(seq, " %s", mddev->pers->name);
3325                 }
3326
3327                 size = 0;
3328                 ITERATE_RDEV(mddev,rdev,tmp2) {
3329                         char b[BDEVNAME_SIZE];
3330                         seq_printf(seq, " %s[%d]",
3331                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3332                         if (rdev->faulty) {
3333                                 seq_printf(seq, "(F)");
3334                                 continue;
3335                         }
3336                         size += rdev->size;
3337                 }
3338
3339                 if (!list_empty(&mddev->disks)) {
3340                         if (mddev->pers)
3341                                 seq_printf(seq, "\n      %llu blocks",
3342                                         (unsigned long long)mddev->array_size);
3343                         else
3344                                 seq_printf(seq, "\n      %llu blocks",
3345                                         (unsigned long long)size);
3346                 }
3347
3348                 if (mddev->pers) {
3349                         mddev->pers->status (seq, mddev);
3350                         seq_printf(seq, "\n      ");
3351                         if (mddev->curr_resync > 2) {
3352                                 status_resync (seq, mddev);
3353                                 seq_printf(seq, "\n      ");
3354                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3355                                 seq_printf(seq, "       resync=DELAYED\n      ");
3356                 } else
3357                         seq_printf(seq, "\n       ");
3358
3359                 if ((bitmap = mddev->bitmap)) {
3360                         unsigned long chunk_kb;
3361                         unsigned long flags;
3362                         spin_lock_irqsave(&bitmap->lock, flags);
3363                         chunk_kb = bitmap->chunksize >> 10;
3364                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3365                                 "%lu%s chunk",
3366                                 bitmap->pages - bitmap->missing_pages,
3367                                 bitmap->pages,
3368                                 (bitmap->pages - bitmap->missing_pages)
3369                                         << (PAGE_SHIFT - 10),
3370                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3371                                 chunk_kb ? "KB" : "B");
3372                         if (bitmap->file) {
3373                                 seq_printf(seq, ", file: ");
3374                                 seq_path(seq, bitmap->file->f_vfsmnt,
3375                                          bitmap->file->f_dentry," \t\n");
3376                         }
3377
3378                         seq_printf(seq, "\n");
3379                         spin_unlock_irqrestore(&bitmap->lock, flags);
3380                 }
3381
3382                 seq_printf(seq, "\n");
3383         }
3384         mddev_unlock(mddev);
3385
3386         return 0;
3387 }
3388
3389 static struct seq_operations md_seq_ops = {
3390         .start  = md_seq_start,
3391         .next   = md_seq_next,
3392         .stop   = md_seq_stop,
3393         .show   = md_seq_show,
3394 };
3395
3396 static int md_seq_open(struct inode *inode, struct file *file)
3397 {
3398         int error;
3399
3400         error = seq_open(file, &md_seq_ops);
3401         return error;
3402 }
3403
3404 static struct file_operations md_seq_fops = {
3405         .open           = md_seq_open,
3406         .read           = seq_read,
3407         .llseek         = seq_lseek,
3408         .release        = seq_release,
3409 };
3410
3411 int register_md_personality(int pnum, mdk_personality_t *p)
3412 {
3413         if (pnum >= MAX_PERSONALITY) {
3414                 printk(KERN_ERR
3415                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3416                        p->name, pnum, MAX_PERSONALITY-1);
3417                 return -EINVAL;
3418         }
3419
3420         spin_lock(&pers_lock);
3421         if (pers[pnum]) {
3422                 spin_unlock(&pers_lock);
3423                 return -EBUSY;
3424         }
3425
3426         pers[pnum] = p;
3427         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3428         spin_unlock(&pers_lock);
3429         return 0;
3430 }
3431
3432 int unregister_md_personality(int pnum)
3433 {
3434         if (pnum >= MAX_PERSONALITY)
3435                 return -EINVAL;
3436
3437         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3438         spin_lock(&pers_lock);
3439         pers[pnum] = NULL;
3440         spin_unlock(&pers_lock);
3441         return 0;
3442 }
3443
3444 static int is_mddev_idle(mddev_t *mddev)
3445 {
3446         mdk_rdev_t * rdev;
3447         struct list_head *tmp;
3448         int idle;
3449         unsigned long curr_events;
3450
3451         idle = 1;
3452         ITERATE_RDEV(mddev,rdev,tmp) {
3453                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3454                 curr_events = disk_stat_read(disk, read_sectors) +
3455                                 disk_stat_read(disk, write_sectors) -
3456                                 atomic_read(&disk->sync_io);
3457                 /* Allow some slack between valud of curr_events and last_events,
3458                  * as there are some uninteresting races.
3459                  * Note: the following is an unsigned comparison.
3460                  */
3461                 if ((curr_events - rdev->last_events + 32) > 64) {
3462                         rdev->last_events = curr_events;
3463                         idle = 0;
3464                 }
3465         }
3466         return idle;
3467 }
3468
3469 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3470 {
3471         /* another "blocks" (512byte) blocks have been synced */
3472         atomic_sub(blocks, &mddev->recovery_active);
3473         wake_up(&mddev->recovery_wait);
3474         if (!ok) {
3475                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3476                 md_wakeup_thread(mddev->thread);
3477                 // stop recovery, signal do_sync ....
3478         }
3479 }
3480
3481
3482 /* md_write_start(mddev, bi)
3483  * If we need to update some array metadata (e.g. 'active' flag
3484  * in superblock) before writing, schedule a superblock update
3485  * and wait for it to complete.
3486  */
3487 void md_write_start(mddev_t *mddev, struct bio *bi)
3488 {
3489         DEFINE_WAIT(w);
3490         if (bio_data_dir(bi) != WRITE)
3491                 return;
3492
3493         atomic_inc(&mddev->writes_pending);
3494         if (mddev->in_sync) {
3495                 spin_lock(&mddev->write_lock);
3496                 if (mddev->in_sync) {
3497                         mddev->in_sync = 0;
3498                         mddev->sb_dirty = 1;
3499                         md_wakeup_thread(mddev->thread);
3500                 }
3501                 spin_unlock(&mddev->write_lock);
3502         }
3503         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3504 }
3505
3506 void md_write_end(mddev_t *mddev)
3507 {
3508         if (atomic_dec_and_test(&mddev->writes_pending)) {
3509                 if (mddev->safemode == 2)
3510                         md_wakeup_thread(mddev->thread);
3511                 else
3512                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3513         }
3514 }
3515
3516 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3517
3518 #define SYNC_MARKS      10
3519 #define SYNC_MARK_STEP  (3*HZ)
3520 static void md_do_sync(mddev_t *mddev)
3521 {
3522         mddev_t *mddev2;
3523         unsigned int currspeed = 0,
3524                  window;
3525         sector_t max_sectors,j, io_sectors;
3526         unsigned long mark[SYNC_MARKS];
3527         sector_t mark_cnt[SYNC_MARKS];
3528         int last_mark,m;
3529         struct list_head *tmp;
3530         sector_t last_check;
3531         int skipped = 0;
3532
3533         /* just incase thread restarts... */
3534         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3535                 return;
3536
3537         /* we overload curr_resync somewhat here.
3538          * 0 == not engaged in resync at all
3539          * 2 == checking that there is no conflict with another sync
3540          * 1 == like 2, but have yielded to allow conflicting resync to
3541          *              commense
3542          * other == active in resync - this many blocks
3543          *
3544          * Before starting a resync we must have set curr_resync to
3545          * 2, and then checked that every "conflicting" array has curr_resync
3546          * less than ours.  When we find one that is the same or higher
3547          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3548          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3549          * This will mean we have to start checking from the beginning again.
3550          *
3551          */
3552
3553         do {
3554                 mddev->curr_resync = 2;
3555
3556         try_again:
3557                 if (signal_pending(current)) {
3558                         flush_signals(current);
3559                         goto skip;
3560                 }
3561                 ITERATE_MDDEV(mddev2,tmp) {
3562                         if (mddev2 == mddev)
3563                                 continue;
3564                         if (mddev2->curr_resync &&
3565                             match_mddev_units(mddev,mddev2)) {
3566                                 DEFINE_WAIT(wq);
3567                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3568                                         /* arbitrarily yield */
3569                                         mddev->curr_resync = 1;
3570                                         wake_up(&resync_wait);
3571                                 }
3572                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3573                                         /* no need to wait here, we can wait the next
3574                                          * time 'round when curr_resync == 2
3575                                          */
3576                                         continue;
3577                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3578                                 if (!signal_pending(current)
3579                                     && mddev2->curr_resync >= mddev->curr_resync) {
3580                                         printk(KERN_INFO "md: delaying resync of %s"
3581                                                " until %s has finished resync (they"
3582                                                " share one or more physical units)\n",
3583                                                mdname(mddev), mdname(mddev2));
3584                                         mddev_put(mddev2);
3585                                         schedule();
3586                                         finish_wait(&resync_wait, &wq);
3587                                         goto try_again;
3588                                 }
3589                                 finish_wait(&resync_wait, &wq);
3590                         }
3591                 }
3592         } while (mddev->curr_resync < 2);
3593
3594         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3595                 /* resync follows the size requested by the personality,
3596                  * which defaults to physical size, but can be virtual size
3597                  */
3598                 max_sectors = mddev->resync_max_sectors;
3599         else
3600                 /* recovery follows the physical size of devices */
3601                 max_sectors = mddev->size << 1;
3602
3603         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3604         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3605                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3606         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3607                "(but not more than %d KB/sec) for reconstruction.\n",
3608                sysctl_speed_limit_max);
3609
3610         is_mddev_idle(mddev); /* this also initializes IO event counters */
3611         /* we don't use the checkpoint if there's a bitmap */
3612         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3613                 j = mddev->recovery_cp;
3614         else
3615                 j = 0;
3616         io_sectors = 0;
3617         for (m = 0; m < SYNC_MARKS; m++) {
3618                 mark[m] = jiffies;
3619                 mark_cnt[m] = io_sectors;
3620         }
3621         last_mark = 0;
3622         mddev->resync_mark = mark[last_mark];
3623         mddev->resync_mark_cnt = mark_cnt[last_mark];
3624
3625         /*
3626          * Tune reconstruction:
3627          */
3628         window = 32*(PAGE_SIZE/512);
3629         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3630                 window/2,(unsigned long long) max_sectors/2);
3631
3632         atomic_set(&mddev->recovery_active, 0);
3633         init_waitqueue_head(&mddev->recovery_wait);
3634         last_check = 0;
3635
3636         if (j>2) {
3637                 printk(KERN_INFO
3638                         "md: resuming recovery of %s from checkpoint.\n",
3639                         mdname(mddev));
3640                 mddev->curr_resync = j;
3641         }
3642
3643         while (j < max_sectors) {
3644                 sector_t sectors;
3645
3646                 skipped = 0;
3647                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3648                                             currspeed < sysctl_speed_limit_min);
3649                 if (sectors == 0) {
3650                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3651                         goto out;
3652                 }
3653
3654                 if (!skipped) { /* actual IO requested */
3655                         io_sectors += sectors;
3656                         atomic_add(sectors, &mddev->recovery_active);
3657                 }
3658
3659                 j += sectors;
3660                 if (j>1) mddev->curr_resync = j;
3661
3662
3663                 if (last_check + window > io_sectors || j == max_sectors)
3664                         continue;
3665
3666                 last_check = io_sectors;
3667
3668                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3669                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3670                         break;
3671
3672         repeat:
3673                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3674                         /* step marks */
3675                         int next = (last_mark+1) % SYNC_MARKS;
3676
3677                         mddev->resync_mark = mark[next];
3678                         mddev->resync_mark_cnt = mark_cnt[next];
3679                         mark[next] = jiffies;
3680                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3681                         last_mark = next;
3682                 }
3683
3684
3685                 if (signal_pending(current)) {
3686                         /*
3687                          * got a signal, exit.
3688                          */
3689                         printk(KERN_INFO
3690                                 "md: md_do_sync() got signal ... exiting\n");
3691                         flush_signals(current);
3692                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3693                         goto out;
3694                 }
3695
3696                 /*
3697                  * this loop exits only if either when we are slower than
3698                  * the 'hard' speed limit, or the system was IO-idle for
3699                  * a jiffy.
3700                  * the system might be non-idle CPU-wise, but we only care
3701                  * about not overloading the IO subsystem. (things like an
3702                  * e2fsck being done on the RAID array should execute fast)
3703                  */
3704                 mddev->queue->unplug_fn(mddev->queue);
3705                 cond_resched();
3706
3707                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3708                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
3709
3710                 if (currspeed > sysctl_speed_limit_min) {
3711                         if ((currspeed > sysctl_speed_limit_max) ||
3712                                         !is_mddev_idle(mddev)) {
3713                                 msleep_interruptible(250);
3714                                 goto repeat;
3715                         }
3716                 }
3717         }
3718         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3719         /*
3720          * this also signals 'finished resyncing' to md_stop
3721          */
3722  out:
3723         mddev->queue->unplug_fn(mddev->queue);
3724
3725         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3726
3727         /* tell personality that we are finished */
3728         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3729
3730         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3731             mddev->curr_resync > 2 &&
3732             mddev->curr_resync >= mddev->recovery_cp) {
3733                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3734                         printk(KERN_INFO
3735                                 "md: checkpointing recovery of %s.\n",
3736                                 mdname(mddev));
3737                         mddev->recovery_cp = mddev->curr_resync;
3738                 } else
3739                         mddev->recovery_cp = MaxSector;
3740         }
3741
3742  skip:
3743         mddev->curr_resync = 0;
3744         wake_up(&resync_wait);
3745         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3746         md_wakeup_thread(mddev->thread);
3747 }
3748
3749
3750 /*
3751  * This routine is regularly called by all per-raid-array threads to
3752  * deal with generic issues like resync and super-block update.
3753  * Raid personalities that don't have a thread (linear/raid0) do not
3754  * need this as they never do any recovery or update the superblock.
3755  *
3756  * It does not do any resync itself, but rather "forks" off other threads
3757  * to do that as needed.
3758  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3759  * "->recovery" and create a thread at ->sync_thread.
3760  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3761  * and wakeups up this thread which will reap the thread and finish up.
3762  * This thread also removes any faulty devices (with nr_pending == 0).
3763  *
3764  * The overall approach is:
3765  *  1/ if the superblock needs updating, update it.
3766  *  2/ If a recovery thread is running, don't do anything else.
3767  *  3/ If recovery has finished, clean up, possibly marking spares active.
3768  *  4/ If there are any faulty devices, remove them.
3769  *  5/ If array is degraded, try to add spares devices
3770  *  6/ If array has spares or is not in-sync, start a resync thread.
3771  */
3772 void md_check_recovery(mddev_t *mddev)
3773 {
3774         mdk_rdev_t *rdev;
3775         struct list_head *rtmp;
3776
3777
3778         if (mddev->bitmap)
3779                 bitmap_daemon_work(mddev->bitmap);
3780
3781         if (mddev->ro)
3782                 return;
3783
3784         if (signal_pending(current)) {
3785                 if (mddev->pers->sync_request) {
3786                         printk(KERN_INFO "md: %s in immediate safe mode\n",
3787                                mdname(mddev));
3788                         mddev->safemode = 2;
3789                 }
3790                 flush_signals(current);
3791         }
3792
3793         if ( ! (
3794                 mddev->sb_dirty ||
3795                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3796                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3797                 (mddev->safemode == 1) ||
3798                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3799                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3800                 ))
3801                 return;
3802
3803         if (mddev_trylock(mddev)==0) {
3804                 int spares =0;
3805
3806                 spin_lock(&mddev->write_lock);
3807                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3808                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3809                         mddev->in_sync = 1;
3810                         mddev->sb_dirty = 1;
3811                 }
3812                 if (mddev->safemode == 1)
3813                         mddev->safemode = 0;
3814                 spin_unlock(&mddev->write_lock);
3815
3816                 if (mddev->sb_dirty)
3817                         md_update_sb(mddev);
3818
3819
3820                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3821                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3822                         /* resync/recovery still happening */
3823                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3824                         goto unlock;
3825                 }
3826                 if (mddev->sync_thread) {
3827                         /* resync has finished, collect result */
3828                         md_unregister_thread(mddev->sync_thread);
3829                         mddev->sync_thread = NULL;
3830                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3831                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3832                                 /* success...*/
3833                                 /* activate any spares */
3834                                 mddev->pers->spare_active(mddev);
3835                         }
3836                         md_update_sb(mddev);
3837
3838                         /* if array is no-longer degraded, then any saved_raid_disk
3839                          * information must be scrapped
3840                          */
3841                         if (!mddev->degraded)
3842                                 ITERATE_RDEV(mddev,rdev,rtmp)
3843                                         rdev->saved_raid_disk = -1;
3844
3845                         mddev->recovery = 0;
3846                         /* flag recovery needed just to double check */
3847                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3848                         goto unlock;
3849                 }
3850                 if (mddev->recovery)
3851                         /* probably just the RECOVERY_NEEDED flag */
3852                         mddev->recovery = 0;
3853
3854                 /* no recovery is running.
3855                  * remove any failed drives, then
3856                  * add spares if possible.
3857                  * Spare are also removed and re-added, to allow
3858                  * the personality to fail the re-add.
3859                  */
3860                 ITERATE_RDEV(mddev,rdev,rtmp)
3861                         if (rdev->raid_disk >= 0 &&
3862                             (rdev->faulty || ! rdev->in_sync) &&
3863                             atomic_read(&rdev->nr_pending)==0) {
3864                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3865                                         rdev->raid_disk = -1;
3866                         }
3867
3868                 if (mddev->degraded) {
3869                         ITERATE_RDEV(mddev,rdev,rtmp)
3870                                 if (rdev->raid_disk < 0
3871                                     && !rdev->faulty) {
3872                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3873                                                 spares++;
3874                                         else
3875                                                 break;
3876                                 }
3877                 }
3878
3879                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3880                         /* nothing we can do ... */
3881                         goto unlock;
3882                 }
3883                 if (mddev->pers->sync_request) {
3884                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3885                         if (!spares)
3886                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3887                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3888                                 /* We are adding a device or devices to an array
3889                                  * which has the bitmap stored on all devices.
3890                                  * So make sure all bitmap pages get written
3891                                  */
3892                                 bitmap_write_all(mddev->bitmap);
3893                         }
3894                         mddev->sync_thread = md_register_thread(md_do_sync,
3895                                                                 mddev,
3896                                                                 "%s_resync");
3897                         if (!mddev->sync_thread) {
3898                                 printk(KERN_ERR "%s: could not start resync"
3899                                         " thread...\n",
3900                                         mdname(mddev));
3901                                 /* leave the spares where they are, it shouldn't hurt */
3902                                 mddev->recovery = 0;
3903                         } else {
3904                                 md_wakeup_thread(mddev->sync_thread);
3905                         }
3906                 }
3907         unlock:
3908                 mddev_unlock(mddev);
3909         }
3910 }
3911
3912 static int md_notify_reboot(struct notifier_block *this,
3913                             unsigned long code, void *x)
3914 {
3915         struct list_head *tmp;
3916         mddev_t *mddev;
3917
3918         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3919
3920                 printk(KERN_INFO "md: stopping all md devices.\n");
3921
3922                 ITERATE_MDDEV(mddev,tmp)
3923                         if (mddev_trylock(mddev)==0)
3924                                 do_md_stop (mddev, 1);
3925                 /*
3926                  * certain more exotic SCSI devices are known to be
3927                  * volatile wrt too early system reboots. While the
3928                  * right place to handle this issue is the given
3929                  * driver, we do want to have a safe RAID driver ...
3930                  */
3931                 mdelay(1000*1);
3932         }
3933         return NOTIFY_DONE;
3934 }
3935
3936 static struct notifier_block md_notifier = {
3937         .notifier_call  = md_notify_reboot,
3938         .next           = NULL,
3939         .priority       = INT_MAX, /* before any real devices */
3940 };
3941
3942 static void md_geninit(void)
3943 {
3944         struct proc_dir_entry *p;
3945
3946         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3947
3948         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3949         if (p)
3950                 p->proc_fops = &md_seq_fops;
3951 }
3952
3953 static int __init md_init(void)
3954 {
3955         int minor;
3956
3957         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3958                         " MD_SB_DISKS=%d\n",
3959                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3960                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3961         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3962                         BITMAP_MINOR);
3963
3964         if (register_blkdev(MAJOR_NR, "md"))
3965                 return -1;
3966         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3967                 unregister_blkdev(MAJOR_NR, "md");
3968                 return -1;
3969         }
3970         devfs_mk_dir("md");
3971         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3972                                 md_probe, NULL, NULL);
3973         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3974                             md_probe, NULL, NULL);
3975
3976         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3977                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3978                                 S_IFBLK|S_IRUSR|S_IWUSR,
3979                                 "md/%d", minor);
3980
3981         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3982                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3983                               S_IFBLK|S_IRUSR|S_IWUSR,
3984                               "md/mdp%d", minor);
3985
3986
3987         register_reboot_notifier(&md_notifier);
3988         raid_table_header = register_sysctl_table(raid_root_table, 1);
3989
3990         md_geninit();
3991         return (0);
3992 }
3993
3994
3995 #ifndef MODULE
3996
3997 /*
3998  * Searches all registered partitions for autorun RAID arrays
3999  * at boot time.
4000  */
4001 static dev_t detected_devices[128];
4002 static int dev_cnt;
4003
4004 void md_autodetect_dev(dev_t dev)
4005 {
4006         if (dev_cnt >= 0 && dev_cnt < 127)
4007                 detected_devices[dev_cnt++] = dev;
4008 }
4009
4010
4011 static void autostart_arrays(int part)
4012 {
4013         mdk_rdev_t *rdev;
4014         int i;
4015
4016         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4017
4018         for (i = 0; i < dev_cnt; i++) {
4019                 dev_t dev = detected_devices[i];
4020
4021                 rdev = md_import_device(dev,0, 0);
4022                 if (IS_ERR(rdev))
4023                         continue;
4024
4025                 if (rdev->faulty) {
4026                         MD_BUG();
4027                         continue;
4028                 }
4029                 list_add(&rdev->same_set, &pending_raid_disks);
4030         }
4031         dev_cnt = 0;
4032
4033         autorun_devices(part);
4034 }
4035
4036 #endif
4037
4038 static __exit void md_exit(void)
4039 {
4040         mddev_t *mddev;
4041         struct list_head *tmp;
4042         int i;
4043         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4044         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4045         for (i=0; i < MAX_MD_DEVS; i++)
4046                 devfs_remove("md/%d", i);
4047         for (i=0; i < MAX_MD_DEVS; i++)
4048                 devfs_remove("md/d%d", i);
4049
4050         devfs_remove("md");
4051
4052         unregister_blkdev(MAJOR_NR,"md");
4053         unregister_blkdev(mdp_major, "mdp");
4054         unregister_reboot_notifier(&md_notifier);
4055         unregister_sysctl_table(raid_table_header);
4056         remove_proc_entry("mdstat", NULL);
4057         ITERATE_MDDEV(mddev,tmp) {
4058                 struct gendisk *disk = mddev->gendisk;
4059                 if (!disk)
4060                         continue;
4061                 export_array(mddev);
4062                 del_gendisk(disk);
4063                 put_disk(disk);
4064                 mddev->gendisk = NULL;
4065                 mddev_put(mddev);
4066         }
4067 }
4068
4069 module_init(md_init)
4070 module_exit(md_exit)
4071
4072 EXPORT_SYMBOL(register_md_personality);
4073 EXPORT_SYMBOL(unregister_md_personality);
4074 EXPORT_SYMBOL(md_error);
4075 EXPORT_SYMBOL(md_done_sync);
4076 EXPORT_SYMBOL(md_write_start);
4077 EXPORT_SYMBOL(md_write_end);
4078 EXPORT_SYMBOL(md_register_thread);
4079 EXPORT_SYMBOL(md_unregister_thread);
4080 EXPORT_SYMBOL(md_wakeup_thread);
4081 EXPORT_SYMBOL(md_print_devices);
4082 EXPORT_SYMBOL(md_check_recovery);
4083 MODULE_LICENSE("GPL");
4084 MODULE_ALIAS("md");
4085 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);