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                 rdev->flags = 0;
 674                 desc = sb->disks + rdev->desc_nr;
 675
 676                 if (desc->state & (1<<MD_DISK_FAULTY))
 677                         rdev->faulty = 1;
 678                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 679                          desc->raid_disk < mddev->raid_disks) {
 680                         rdev->in_sync = 1;
 681                         rdev->raid_disk = desc->raid_disk;
 682                 }
 683                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 684                         set_bit(WriteMostly, &rdev->flags);
 685         } else /* MULTIPATH are always insync */
 686                 rdev->in_sync = 1;
 687         return 0;
 688 }
 689
 690 /*
 691  * sync_super for 0.90.0
 692  */
 693 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 694 {
 695         mdp_super_t *sb;
 696         struct list_head *tmp;
 697         mdk_rdev_t *rdev2;
 698         int next_spare = mddev->raid_disks;
 699
 700         /* make rdev->sb match mddev data..
 701          *
 702          * 1/ zero out disks
 703          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 704          * 3/ any empty disks < next_spare become removed
 705          *
 706          * disks[0] gets initialised to REMOVED because
 707          * we cannot be sure from other fields if it has
 708          * been initialised or not.
 709          */
 710         int i;
 711         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 712
 713         sb = (mdp_super_t*)page_address(rdev->sb_page);
 714
 715         memset(sb, 0, sizeof(*sb));
 716
 717         sb->md_magic = MD_SB_MAGIC;
 718         sb->major_version = mddev->major_version;
 719         sb->minor_version = mddev->minor_version;
 720         sb->patch_version = mddev->patch_version;
 721         sb->gvalid_words  = 0; /* ignored */
 722         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 723         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 724         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 725         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 726
 727         sb->ctime = mddev->ctime;
 728         sb->level = mddev->level;
 729         sb->size  = mddev->size;
 730         sb->raid_disks = mddev->raid_disks;
 731         sb->md_minor = mddev->md_minor;
 732         sb->not_persistent = !mddev->persistent;
 733         sb->utime = mddev->utime;
 734         sb->state = 0;
 735         sb->events_hi = (mddev->events>>32);
 736         sb->events_lo = (u32)mddev->events;
 737
 738         if (mddev->in_sync)
 739         {
 740                 sb->recovery_cp = mddev->recovery_cp;
 741                 sb->cp_events_hi = (mddev->events>>32);
 742                 sb->cp_events_lo = (u32)mddev->events;
 743                 if (mddev->recovery_cp == MaxSector)
 744                         sb->state = (1<< MD_SB_CLEAN);
 745         } else
 746                 sb->recovery_cp = 0;
 747
 748         sb->layout = mddev->layout;
 749         sb->chunk_size = mddev->chunk_size;
 750
 751         if (mddev->bitmap && mddev->bitmap_file == NULL)
 752                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 753
 754         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 755         ITERATE_RDEV(mddev,rdev2,tmp) {
 756                 mdp_disk_t *d;
 757                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 758                         rdev2->desc_nr = rdev2->raid_disk;
 759                 else
 760                         rdev2->desc_nr = next_spare++;
 761                 d = &sb->disks[rdev2->desc_nr];
 762                 nr_disks++;
 763                 d->number = rdev2->desc_nr;
 764                 d->major = MAJOR(rdev2->bdev->bd_dev);
 765                 d->minor = MINOR(rdev2->bdev->bd_dev);
 766                 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
 767                         d->raid_disk = rdev2->raid_disk;
 768                 else
 769                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 770                 if (rdev2->faulty) {
 771                         d->state = (1<<MD_DISK_FAULTY);
 772                         failed++;
 773                 } else if (rdev2->in_sync) {
 774                         d->state = (1<<MD_DISK_ACTIVE);
 775                         d->state |= (1<<MD_DISK_SYNC);
 776                         active++;
 777                         working++;
 778                 } else {
 779                         d->state = 0;
 780                         spare++;
 781                         working++;
 782                 }
 783                 if (test_bit(WriteMostly, &rdev2->flags))
 784                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 785         }
 786
 787         /* now set the "removed" and "faulty" bits on any missing devices */
 788         for (i=0 ; i < mddev->raid_disks ; i++) {
 789                 mdp_disk_t *d = &sb->disks[i];
 790                 if (d->state == 0 && d->number == 0) {
 791                         d->number = i;
 792                         d->raid_disk = i;
 793                         d->state = (1<<MD_DISK_REMOVED);
 794                         d->state |= (1<<MD_DISK_FAULTY);
 795                         failed++;
 796                 }
 797         }
 798         sb->nr_disks = nr_disks;
 799         sb->active_disks = active;
 800         sb->working_disks = working;
 801         sb->failed_disks = failed;
 802         sb->spare_disks = spare;
 803
 804         sb->this_disk = sb->disks[rdev->desc_nr];
 805         sb->sb_csum = calc_sb_csum(sb);
 806 }
 807
 808 /*
 809  * version 1 superblock
 810  */
 811
 812 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 813 {
 814         unsigned int disk_csum, csum;
 815         unsigned long long newcsum;
 816         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 817         unsigned int *isuper = (unsigned int*)sb;
 818         int i;
 819
 820         disk_csum = sb->sb_csum;
 821         sb->sb_csum = 0;
 822         newcsum = 0;
 823         for (i=0; size>=4; size -= 4 )
 824                 newcsum += le32_to_cpu(*isuper++);
 825
 826         if (size == 2)
 827                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 828
 829         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 830         sb->sb_csum = disk_csum;
 831         return cpu_to_le32(csum);
 832 }
 833
 834 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 835 {
 836         struct mdp_superblock_1 *sb;
 837         int ret;
 838         sector_t sb_offset;
 839         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 840
 841         /*
 842          * Calculate the position of the superblock.
 843          * It is always aligned to a 4K boundary and
 844          * depeding on minor_version, it can be:
 845          * 0: At least 8K, but less than 12K, from end of device
 846          * 1: At start of device
 847          * 2: 4K from start of device.
 848          */
 849         switch(minor_version) {
 850         case 0:
 851                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 852                 sb_offset -= 8*2;
 853                 sb_offset &= ~(sector_t)(4*2-1);
 854                 /* convert from sectors to K */
 855                 sb_offset /= 2;
 856                 break;
 857         case 1:
 858                 sb_offset = 0;
 859                 break;
 860         case 2:
 861                 sb_offset = 4;
 862                 break;
 863         default:
 864                 return -EINVAL;
 865         }
 866         rdev->sb_offset = sb_offset;
 867
 868         ret = read_disk_sb(rdev);
 869         if (ret) return ret;
 870
 871
 872         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 873
 874         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 875             sb->major_version != cpu_to_le32(1) ||
 876             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 877             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 878             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 879                 return -EINVAL;
 880
 881         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 882                 printk("md: invalid superblock checksum on %s\n",
 883                         bdevname(rdev->bdev,b));
 884                 return -EINVAL;
 885         }
 886         if (le64_to_cpu(sb->data_size) < 10) {
 887                 printk("md: data_size too small on %s\n",
 888                        bdevname(rdev->bdev,b));
 889                 return -EINVAL;
 890         }
 891         rdev->preferred_minor = 0xffff;
 892         rdev->data_offset = le64_to_cpu(sb->data_offset);
 893
 894         if (refdev == 0)
 895                 return 1;
 896         else {
 897                 __u64 ev1, ev2;
 898                 struct mdp_superblock_1 *refsb =
 899                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 900
 901                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 902                     sb->level != refsb->level ||
 903                     sb->layout != refsb->layout ||
 904                     sb->chunksize != refsb->chunksize) {
 905                         printk(KERN_WARNING "md: %s has strangely different"
 906                                 " superblock to %s\n",
 907                                 bdevname(rdev->bdev,b),
 908                                 bdevname(refdev->bdev,b2));
 909                         return -EINVAL;
 910                 }
 911                 ev1 = le64_to_cpu(sb->events);
 912                 ev2 = le64_to_cpu(refsb->events);
 913
 914                 if (ev1 > ev2)
 915                         return 1;
 916         }
 917         if (minor_version)
 918                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 919         else
 920                 rdev->size = rdev->sb_offset;
 921         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 922                 return -EINVAL;
 923         rdev->size = le64_to_cpu(sb->data_size)/2;
 924         if (le32_to_cpu(sb->chunksize))
 925                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 926         return 0;
 927 }
 928
 929 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 930 {
 931         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 932
 933         rdev->raid_disk = -1;
 934         rdev->in_sync = 0;
 935         if (mddev->raid_disks == 0) {
 936                 mddev->major_version = 1;
 937                 mddev->patch_version = 0;
 938                 mddev->persistent = 1;
 939                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 940                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 941                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 942                 mddev->level = le32_to_cpu(sb->level);
 943                 mddev->layout = le32_to_cpu(sb->layout);
 944                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 945                 mddev->size = le64_to_cpu(sb->size)/2;
 946                 mddev->events = le64_to_cpu(sb->events);
 947                 mddev->bitmap_offset = 0;
 948                 mddev->default_bitmap_offset = 0;
 949                 if (mddev->minor_version == 0)
 950                         mddev->default_bitmap_offset = -(64*1024)/512;
 951
 952                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 953                 memcpy(mddev->uuid, sb->set_uuid, 16);
 954
 955                 mddev->max_disks =  (4096-256)/2;
 956
 957                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
 958                     mddev->bitmap_file == NULL ) {
 959                         if (mddev->level != 1) {
 960                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
 961                                 return -EINVAL;
 962                         }
 963                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
 964                 }
 965         } else if (mddev->pers == NULL) {
 966                 /* Insist of good event counter while assembling */
 967                 __u64 ev1 = le64_to_cpu(sb->events);
 968                 ++ev1;
 969                 if (ev1 < mddev->events)
 970                         return -EINVAL;
 971         } else if (mddev->bitmap) {
 972                 /* If adding to array with a bitmap, then we can accept an
 973                  * older device, but not too old.
 974                  */
 975                 __u64 ev1 = le64_to_cpu(sb->events);
 976                 if (ev1 < mddev->bitmap->events_cleared)
 977                         return 0;
 978         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 979                 return 0;
 980
 981         if (mddev->level != LEVEL_MULTIPATH) {
 982                 int role;
 983                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
 984                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
 985                 switch(role) {
 986                 case 0xffff: /* spare */
 987                         rdev->faulty = 0;
 988                         break;
 989                 case 0xfffe: /* faulty */
 990                         rdev->faulty = 1;
 991                         break;
 992                 default:
 993                         rdev->in_sync = 1;
 994                         rdev->faulty = 0;
 995                         rdev->raid_disk = role;
 996                         break;
 997                 }
 998                 rdev->flags = 0;
 999                 if (sb->devflags & WriteMostly1)
1000                         set_bit(WriteMostly, &rdev->flags);
1001         } else /* MULTIPATH are always insync */
1002                 rdev->in_sync = 1;
1003
1004         return 0;
1005 }
1006
1007 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1008 {
1009         struct mdp_superblock_1 *sb;
1010         struct list_head *tmp;
1011         mdk_rdev_t *rdev2;
1012         int max_dev, i;
1013         /* make rdev->sb match mddev and rdev data. */
1014
1015         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1016
1017         sb->feature_map = 0;
1018         sb->pad0 = 0;
1019         memset(sb->pad1, 0, sizeof(sb->pad1));
1020         memset(sb->pad2, 0, sizeof(sb->pad2));
1021         memset(sb->pad3, 0, sizeof(sb->pad3));
1022
1023         sb->utime = cpu_to_le64((__u64)mddev->utime);
1024         sb->events = cpu_to_le64(mddev->events);
1025         if (mddev->in_sync)
1026                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1027         else
1028                 sb->resync_offset = cpu_to_le64(0);
1029
1030         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1031                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1032                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1033         }
1034
1035         max_dev = 0;
1036         ITERATE_RDEV(mddev,rdev2,tmp)
1037                 if (rdev2->desc_nr+1 > max_dev)
1038                         max_dev = rdev2->desc_nr+1;
1039
1040         sb->max_dev = cpu_to_le32(max_dev);
1041         for (i=0; i<max_dev;i++)
1042                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1043
1044         ITERATE_RDEV(mddev,rdev2,tmp) {
1045                 i = rdev2->desc_nr;
1046                 if (rdev2->faulty)
1047                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1048                 else if (rdev2->in_sync)
1049                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1050                 else
1051                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1052         }
1053
1054         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1055         sb->sb_csum = calc_sb_1_csum(sb);
1056 }
1057
1058
1059 static struct super_type super_types[] = {
1060         [0] = {
1061                 .name   = "0.90.0",
1062                 .owner  = THIS_MODULE,
1063                 .load_super     = super_90_load,
1064                 .validate_super = super_90_validate,
1065                 .sync_super     = super_90_sync,
1066         },
1067         [1] = {
1068                 .name   = "md-1",
1069                 .owner  = THIS_MODULE,
1070                 .load_super     = super_1_load,
1071                 .validate_super = super_1_validate,
1072                 .sync_super     = super_1_sync,
1073         },
1074 };
1075
1076 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1077 {
1078         struct list_head *tmp;
1079         mdk_rdev_t *rdev;
1080
1081         ITERATE_RDEV(mddev,rdev,tmp)
1082                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1083                         return rdev;
1084
1085         return NULL;
1086 }
1087
1088 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1089 {
1090         struct list_head *tmp;
1091         mdk_rdev_t *rdev;
1092
1093         ITERATE_RDEV(mddev1,rdev,tmp)
1094                 if (match_dev_unit(mddev2, rdev))
1095                         return 1;
1096
1097         return 0;
1098 }
1099
1100 static LIST_HEAD(pending_raid_disks);
1101
1102 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1103 {
1104         mdk_rdev_t *same_pdev;
1105         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1106
1107         if (rdev->mddev) {
1108                 MD_BUG();
1109                 return -EINVAL;
1110         }
1111         same_pdev = match_dev_unit(mddev, rdev);
1112         if (same_pdev)
1113                 printk(KERN_WARNING
1114                         "%s: WARNING: %s appears to be on the same physical"
1115                         " disk as %s. True\n     protection against single-disk"
1116                         " failure might be compromised.\n",
1117                         mdname(mddev), bdevname(rdev->bdev,b),
1118                         bdevname(same_pdev->bdev,b2));
1119
1120         /* Verify rdev->desc_nr is unique.
1121          * If it is -1, assign a free number, else
1122          * check number is not in use
1123          */
1124         if (rdev->desc_nr < 0) {
1125                 int choice = 0;
1126                 if (mddev->pers) choice = mddev->raid_disks;
1127                 while (find_rdev_nr(mddev, choice))
1128                         choice++;
1129                 rdev->desc_nr = choice;
1130         } else {
1131                 if (find_rdev_nr(mddev, rdev->desc_nr))
1132                         return -EBUSY;
1133         }
1134
1135         list_add(&rdev->same_set, &mddev->disks);
1136         rdev->mddev = mddev;
1137         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1138         return 0;
1139 }
1140
1141 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1142 {
1143         char b[BDEVNAME_SIZE];
1144         if (!rdev->mddev) {
1145                 MD_BUG();
1146                 return;
1147         }
1148         list_del_init(&rdev->same_set);
1149         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1150         rdev->mddev = NULL;
1151 }
1152
1153 /*
1154  * prevent the device from being mounted, repartitioned or
1155  * otherwise reused by a RAID array (or any other kernel
1156  * subsystem), by bd_claiming the device.
1157  */
1158 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1159 {
1160         int err = 0;
1161         struct block_device *bdev;
1162         char b[BDEVNAME_SIZE];
1163
1164         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1165         if (IS_ERR(bdev)) {
1166                 printk(KERN_ERR "md: could not open %s.\n",
1167                         __bdevname(dev, b));
1168                 return PTR_ERR(bdev);
1169         }
1170         err = bd_claim(bdev, rdev);
1171         if (err) {
1172                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1173                         bdevname(bdev, b));
1174                 blkdev_put(bdev);
1175                 return err;
1176         }
1177         rdev->bdev = bdev;
1178         return err;
1179 }
1180
1181 static void unlock_rdev(mdk_rdev_t *rdev)
1182 {
1183         struct block_device *bdev = rdev->bdev;
1184         rdev->bdev = NULL;
1185         if (!bdev)
1186                 MD_BUG();
1187         bd_release(bdev);
1188         blkdev_put(bdev);
1189 }
1190
1191 void md_autodetect_dev(dev_t dev);
1192
1193 static void export_rdev(mdk_rdev_t * rdev)
1194 {
1195         char b[BDEVNAME_SIZE];
1196         printk(KERN_INFO "md: export_rdev(%s)\n",
1197                 bdevname(rdev->bdev,b));
1198         if (rdev->mddev)
1199                 MD_BUG();
1200         free_disk_sb(rdev);
1201         list_del_init(&rdev->same_set);
1202 #ifndef MODULE
1203         md_autodetect_dev(rdev->bdev->bd_dev);
1204 #endif
1205         unlock_rdev(rdev);
1206         kfree(rdev);
1207 }
1208
1209 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1210 {
1211         unbind_rdev_from_array(rdev);
1212         export_rdev(rdev);
1213 }
1214
1215 static void export_array(mddev_t *mddev)
1216 {
1217         struct list_head *tmp;
1218         mdk_rdev_t *rdev;
1219
1220         ITERATE_RDEV(mddev,rdev,tmp) {
1221                 if (!rdev->mddev) {
1222                         MD_BUG();
1223                         continue;
1224                 }
1225                 kick_rdev_from_array(rdev);
1226         }
1227         if (!list_empty(&mddev->disks))
1228                 MD_BUG();
1229         mddev->raid_disks = 0;
1230         mddev->major_version = 0;
1231 }
1232
1233 static void print_desc(mdp_disk_t *desc)
1234 {
1235         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1236                 desc->major,desc->minor,desc->raid_disk,desc->state);
1237 }
1238
1239 static void print_sb(mdp_super_t *sb)
1240 {
1241         int i;
1242
1243         printk(KERN_INFO
1244                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1245                 sb->major_version, sb->minor_version, sb->patch_version,
1246                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1247                 sb->ctime);
1248         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1249                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1250                 sb->md_minor, sb->layout, sb->chunk_size);
1251         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1252                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1253                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1254                 sb->failed_disks, sb->spare_disks,
1255                 sb->sb_csum, (unsigned long)sb->events_lo);
1256
1257         printk(KERN_INFO);
1258         for (i = 0; i < MD_SB_DISKS; i++) {
1259                 mdp_disk_t *desc;
1260
1261                 desc = sb->disks + i;
1262                 if (desc->number || desc->major || desc->minor ||
1263                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1264                         printk("     D %2d: ", i);
1265                         print_desc(desc);
1266                 }
1267         }
1268         printk(KERN_INFO "md:     THIS: ");
1269         print_desc(&sb->this_disk);
1270
1271 }
1272
1273 static void print_rdev(mdk_rdev_t *rdev)
1274 {
1275         char b[BDEVNAME_SIZE];
1276         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1277                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1278                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1279         if (rdev->sb_loaded) {
1280                 printk(KERN_INFO "md: rdev superblock:\n");
1281                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1282         } else
1283                 printk(KERN_INFO "md: no rdev superblock!\n");
1284 }
1285
1286 void md_print_devices(void)
1287 {
1288         struct list_head *tmp, *tmp2;
1289         mdk_rdev_t *rdev;
1290         mddev_t *mddev;
1291         char b[BDEVNAME_SIZE];
1292
1293         printk("\n");
1294         printk("md:     **********************************\n");
1295         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1296         printk("md:     **********************************\n");
1297         ITERATE_MDDEV(mddev,tmp) {
1298
1299                 if (mddev->bitmap)
1300                         bitmap_print_sb(mddev->bitmap);
1301                 else
1302                         printk("%s: ", mdname(mddev));
1303                 ITERATE_RDEV(mddev,rdev,tmp2)
1304                         printk("<%s>", bdevname(rdev->bdev,b));
1305                 printk("\n");
1306
1307                 ITERATE_RDEV(mddev,rdev,tmp2)
1308                         print_rdev(rdev);
1309         }
1310         printk("md:     **********************************\n");
1311         printk("\n");
1312 }
1313
1314
1315 static void sync_sbs(mddev_t * mddev)
1316 {
1317         mdk_rdev_t *rdev;
1318         struct list_head *tmp;
1319
1320         ITERATE_RDEV(mddev,rdev,tmp) {
1321                 super_types[mddev->major_version].
1322                         sync_super(mddev, rdev);
1323                 rdev->sb_loaded = 1;
1324         }
1325 }
1326
1327 static void md_update_sb(mddev_t * mddev)
1328 {
1329         int err;
1330         struct list_head *tmp;
1331         mdk_rdev_t *rdev;
1332         int sync_req;
1333
1334 repeat:
1335         spin_lock(&mddev->write_lock);
1336         sync_req = mddev->in_sync;
1337         mddev->utime = get_seconds();
1338         mddev->events ++;
1339
1340         if (!mddev->events) {
1341                 /*
1342                  * oops, this 64-bit counter should never wrap.
1343                  * Either we are in around ~1 trillion A.C., assuming
1344                  * 1 reboot per second, or we have a bug:
1345                  */
1346                 MD_BUG();
1347                 mddev->events --;
1348         }
1349         mddev->sb_dirty = 2;
1350         sync_sbs(mddev);
1351
1352         /*
1353          * do not write anything to disk if using
1354          * nonpersistent superblocks
1355          */
1356         if (!mddev->persistent) {
1357                 mddev->sb_dirty = 0;
1358                 spin_unlock(&mddev->write_lock);
1359                 wake_up(&mddev->sb_wait);
1360                 return;
1361         }
1362         spin_unlock(&mddev->write_lock);
1363
1364         dprintk(KERN_INFO
1365                 "md: updating %s RAID superblock on device (in sync %d)\n",
1366                 mdname(mddev),mddev->in_sync);
1367
1368         err = bitmap_update_sb(mddev->bitmap);
1369         ITERATE_RDEV(mddev,rdev,tmp) {
1370                 char b[BDEVNAME_SIZE];
1371                 dprintk(KERN_INFO "md: ");
1372                 if (rdev->faulty)
1373                         dprintk("(skipping faulty ");
1374
1375                 dprintk("%s ", bdevname(rdev->bdev,b));
1376                 if (!rdev->faulty) {
1377                         md_super_write(mddev,rdev,
1378                                        rdev->sb_offset<<1, MD_SB_BYTES,
1379                                        rdev->sb_page);
1380                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1381                                 bdevname(rdev->bdev,b),
1382                                 (unsigned long long)rdev->sb_offset);
1383
1384                 } else
1385                         dprintk(")\n");
1386                 if (mddev->level == LEVEL_MULTIPATH)
1387                         /* only need to write one superblock... */
1388                         break;
1389         }
1390         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1391         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1392
1393         spin_lock(&mddev->write_lock);
1394         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1395                 /* have to write it out again */
1396                 spin_unlock(&mddev->write_lock);
1397                 goto repeat;
1398         }
1399         mddev->sb_dirty = 0;
1400         spin_unlock(&mddev->write_lock);
1401         wake_up(&mddev->sb_wait);
1402
1403 }
1404
1405 /*
1406  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1407  *
1408  * mark the device faulty if:
1409  *
1410  *   - the device is nonexistent (zero size)
1411  *   - the device has no valid superblock
1412  *
1413  * a faulty rdev _never_ has rdev->sb set.
1414  */
1415 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1416 {
1417         char b[BDEVNAME_SIZE];
1418         int err;
1419         mdk_rdev_t *rdev;
1420         sector_t size;
1421
1422         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1423         if (!rdev) {
1424                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1425                 return ERR_PTR(-ENOMEM);
1426         }
1427         memset(rdev, 0, sizeof(*rdev));
1428
1429         if ((err = alloc_disk_sb(rdev)))
1430                 goto abort_free;
1431
1432         err = lock_rdev(rdev, newdev);
1433         if (err)
1434                 goto abort_free;
1435
1436         rdev->desc_nr = -1;
1437         rdev->faulty = 0;
1438         rdev->in_sync = 0;
1439         rdev->data_offset = 0;
1440         atomic_set(&rdev->nr_pending, 0);
1441
1442         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1443         if (!size) {
1444                 printk(KERN_WARNING
1445                         "md: %s has zero or unknown size, marking faulty!\n",
1446                         bdevname(rdev->bdev,b));
1447                 err = -EINVAL;
1448                 goto abort_free;
1449         }
1450
1451         if (super_format >= 0) {
1452                 err = super_types[super_format].
1453                         load_super(rdev, NULL, super_minor);
1454                 if (err == -EINVAL) {
1455                         printk(KERN_WARNING
1456                                 "md: %s has invalid sb, not importing!\n",
1457                                 bdevname(rdev->bdev,b));
1458                         goto abort_free;
1459                 }
1460                 if (err < 0) {
1461                         printk(KERN_WARNING
1462                                 "md: could not read %s's sb, not importing!\n",
1463                                 bdevname(rdev->bdev,b));
1464                         goto abort_free;
1465                 }
1466         }
1467         INIT_LIST_HEAD(&rdev->same_set);
1468
1469         return rdev;
1470
1471 abort_free:
1472         if (rdev->sb_page) {
1473                 if (rdev->bdev)
1474                         unlock_rdev(rdev);
1475                 free_disk_sb(rdev);
1476         }
1477         kfree(rdev);
1478         return ERR_PTR(err);
1479 }
1480
1481 /*
1482  * Check a full RAID array for plausibility
1483  */
1484
1485
1486 static void analyze_sbs(mddev_t * mddev)
1487 {
1488         int i;
1489         struct list_head *tmp;
1490         mdk_rdev_t *rdev, *freshest;
1491         char b[BDEVNAME_SIZE];
1492
1493         freshest = NULL;
1494         ITERATE_RDEV(mddev,rdev,tmp)
1495                 switch (super_types[mddev->major_version].
1496                         load_super(rdev, freshest, mddev->minor_version)) {
1497                 case 1:
1498                         freshest = rdev;
1499                         break;
1500                 case 0:
1501                         break;
1502                 default:
1503                         printk( KERN_ERR \
1504                                 "md: fatal superblock inconsistency in %s"
1505                                 " -- removing from array\n",
1506                                 bdevname(rdev->bdev,b));
1507                         kick_rdev_from_array(rdev);
1508                 }
1509
1510
1511         super_types[mddev->major_version].
1512                 validate_super(mddev, freshest);
1513
1514         i = 0;
1515         ITERATE_RDEV(mddev,rdev,tmp) {
1516                 if (rdev != freshest)
1517                         if (super_types[mddev->major_version].
1518                             validate_super(mddev, rdev)) {
1519                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1520                                         " from array!\n",
1521                                         bdevname(rdev->bdev,b));
1522                                 kick_rdev_from_array(rdev);
1523                                 continue;
1524                         }
1525                 if (mddev->level == LEVEL_MULTIPATH) {
1526                         rdev->desc_nr = i++;
1527                         rdev->raid_disk = rdev->desc_nr;
1528                         rdev->in_sync = 1;
1529                 }
1530         }
1531
1532
1533
1534         if (mddev->recovery_cp != MaxSector &&
1535             mddev->level >= 1)
1536                 printk(KERN_ERR "md: %s: raid array is not clean"
1537                        " -- starting background reconstruction\n",
1538                        mdname(mddev));
1539
1540 }
1541
1542 int mdp_major = 0;
1543
1544 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1545 {
1546         static DECLARE_MUTEX(disks_sem);
1547         mddev_t *mddev = mddev_find(dev);
1548         struct gendisk *disk;
1549         int partitioned = (MAJOR(dev) != MD_MAJOR);
1550         int shift = partitioned ? MdpMinorShift : 0;
1551         int unit = MINOR(dev) >> shift;
1552
1553         if (!mddev)
1554                 return NULL;
1555
1556         down(&disks_sem);
1557         if (mddev->gendisk) {
1558                 up(&disks_sem);
1559                 mddev_put(mddev);
1560                 return NULL;
1561         }
1562         disk = alloc_disk(1 << shift);
1563         if (!disk) {
1564                 up(&disks_sem);
1565                 mddev_put(mddev);
1566                 return NULL;
1567         }
1568         disk->major = MAJOR(dev);
1569         disk->first_minor = unit << shift;
1570         if (partitioned) {
1571                 sprintf(disk->disk_name, "md_d%d", unit);
1572                 sprintf(disk->devfs_name, "md/d%d", unit);
1573         } else {
1574                 sprintf(disk->disk_name, "md%d", unit);
1575                 sprintf(disk->devfs_name, "md/%d", unit);
1576         }
1577         disk->fops = &md_fops;
1578         disk->private_data = mddev;
1579         disk->queue = mddev->queue;
1580         add_disk(disk);
1581         mddev->gendisk = disk;
1582         up(&disks_sem);
1583         return NULL;
1584 }
1585
1586 void md_wakeup_thread(mdk_thread_t *thread);
1587
1588 static void md_safemode_timeout(unsigned long data)
1589 {
1590         mddev_t *mddev = (mddev_t *) data;
1591
1592         mddev->safemode = 1;
1593         md_wakeup_thread(mddev->thread);
1594 }
1595
1596
1597 static int do_md_run(mddev_t * mddev)
1598 {
1599         int pnum, err;
1600         int chunk_size;
1601         struct list_head *tmp;
1602         mdk_rdev_t *rdev;
1603         struct gendisk *disk;
1604         char b[BDEVNAME_SIZE];
1605
1606         if (list_empty(&mddev->disks))
1607                 /* cannot run an array with no devices.. */
1608                 return -EINVAL;
1609
1610         if (mddev->pers)
1611                 return -EBUSY;
1612
1613         /*
1614          * Analyze all RAID superblock(s)
1615          */
1616         if (!mddev->raid_disks)
1617                 analyze_sbs(mddev);
1618
1619         chunk_size = mddev->chunk_size;
1620         pnum = level_to_pers(mddev->level);
1621
1622         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1623                 if (!chunk_size) {
1624                         /*
1625                          * 'default chunksize' in the old md code used to
1626                          * be PAGE_SIZE, baaad.
1627                          * we abort here to be on the safe side. We don't
1628                          * want to continue the bad practice.
1629                          */
1630                         printk(KERN_ERR
1631                                 "no chunksize specified, see 'man raidtab'\n");
1632                         return -EINVAL;
1633                 }
1634                 if (chunk_size > MAX_CHUNK_SIZE) {
1635                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1636                                 chunk_size, MAX_CHUNK_SIZE);
1637                         return -EINVAL;
1638                 }
1639                 /*
1640                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1641                  */
1642                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1643                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1644                         return -EINVAL;
1645                 }
1646                 if (chunk_size < PAGE_SIZE) {
1647                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1648                                 chunk_size, PAGE_SIZE);
1649                         return -EINVAL;
1650                 }
1651
1652                 /* devices must have minimum size of one chunk */
1653                 ITERATE_RDEV(mddev,rdev,tmp) {
1654                         if (rdev->faulty)
1655                                 continue;
1656                         if (rdev->size < chunk_size / 1024) {
1657                                 printk(KERN_WARNING
1658                                         "md: Dev %s smaller than chunk_size:"
1659                                         " %lluk < %dk\n",
1660                                         bdevname(rdev->bdev,b),
1661                                         (unsigned long long)rdev->size,
1662                                         chunk_size / 1024);
1663                                 return -EINVAL;
1664                         }
1665                 }
1666         }
1667
1668 #ifdef CONFIG_KMOD
1669         if (!pers[pnum])
1670         {
1671                 request_module("md-personality-%d", pnum);
1672         }
1673 #endif
1674
1675         /*
1676          * Drop all container device buffers, from now on
1677          * the only valid external interface is through the md
1678          * device.
1679          * Also find largest hardsector size
1680          */
1681         ITERATE_RDEV(mddev,rdev,tmp) {
1682                 if (rdev->faulty)
1683                         continue;
1684                 sync_blockdev(rdev->bdev);
1685                 invalidate_bdev(rdev->bdev, 0);
1686         }
1687
1688         md_probe(mddev->unit, NULL, NULL);
1689         disk = mddev->gendisk;
1690         if (!disk)
1691                 return -ENOMEM;
1692
1693         spin_lock(&pers_lock);
1694         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1695                 spin_unlock(&pers_lock);
1696                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1697                        pnum);
1698                 return -EINVAL;
1699         }
1700
1701         mddev->pers = pers[pnum];
1702         spin_unlock(&pers_lock);
1703
1704         mddev->recovery = 0;
1705         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1706
1707         /* before we start the array running, initialise the bitmap */
1708         err = bitmap_create(mddev);
1709         if (err)
1710                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1711                         mdname(mddev), err);
1712         else
1713                 err = mddev->pers->run(mddev);
1714         if (err) {
1715                 printk(KERN_ERR "md: pers->run() failed ...\n");
1716                 module_put(mddev->pers->owner);
1717                 mddev->pers = NULL;
1718                 bitmap_destroy(mddev);
1719                 return err;
1720         }
1721         atomic_set(&mddev->writes_pending,0);
1722         mddev->safemode = 0;
1723         mddev->safemode_timer.function = md_safemode_timeout;
1724         mddev->safemode_timer.data = (unsigned long) mddev;
1725         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1726         mddev->in_sync = 1;
1727
1728         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1729         md_wakeup_thread(mddev->thread);
1730
1731         if (mddev->sb_dirty)
1732                 md_update_sb(mddev);
1733
1734         set_capacity(disk, mddev->array_size<<1);
1735
1736         /* If we call blk_queue_make_request here, it will
1737          * re-initialise max_sectors etc which may have been
1738          * refined inside -> run.  So just set the bits we need to set.
1739          * Most initialisation happended when we called
1740          * blk_queue_make_request(..., md_fail_request)
1741          * earlier.
1742          */
1743         mddev->queue->queuedata = mddev;
1744         mddev->queue->make_request_fn = mddev->pers->make_request;
1745
1746         mddev->changed = 1;
1747         return 0;
1748 }
1749
1750 static int restart_array(mddev_t *mddev)
1751 {
1752         struct gendisk *disk = mddev->gendisk;
1753         int err;
1754
1755         /*
1756          * Complain if it has no devices
1757          */
1758         err = -ENXIO;
1759         if (list_empty(&mddev->disks))
1760                 goto out;
1761
1762         if (mddev->pers) {
1763                 err = -EBUSY;
1764                 if (!mddev->ro)
1765                         goto out;
1766
1767                 mddev->safemode = 0;
1768                 mddev->ro = 0;
1769                 set_disk_ro(disk, 0);
1770
1771                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1772                         mdname(mddev));
1773                 /*
1774                  * Kick recovery or resync if necessary
1775                  */
1776                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1777                 md_wakeup_thread(mddev->thread);
1778                 err = 0;
1779         } else {
1780                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1781                         mdname(mddev));
1782                 err = -EINVAL;
1783         }
1784
1785 out:
1786         return err;
1787 }
1788
1789 static int do_md_stop(mddev_t * mddev, int ro)
1790 {
1791         int err = 0;
1792         struct gendisk *disk = mddev->gendisk;
1793
1794         if (mddev->pers) {
1795                 if (atomic_read(&mddev->active)>2) {
1796                         printk("md: %s still in use.\n",mdname(mddev));
1797                         return -EBUSY;
1798                 }
1799
1800                 if (mddev->sync_thread) {
1801                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1802                         md_unregister_thread(mddev->sync_thread);
1803                         mddev->sync_thread = NULL;
1804                 }
1805
1806                 del_timer_sync(&mddev->safemode_timer);
1807
1808                 invalidate_partition(disk, 0);
1809
1810                 if (ro) {
1811                         err  = -ENXIO;
1812                         if (mddev->ro)
1813                                 goto out;
1814                         mddev->ro = 1;
1815                 } else {
1816                         bitmap_flush(mddev);
1817                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1818                         if (mddev->ro)
1819                                 set_disk_ro(disk, 0);
1820                         blk_queue_make_request(mddev->queue, md_fail_request);
1821                         mddev->pers->stop(mddev);
1822                         module_put(mddev->pers->owner);
1823                         mddev->pers = NULL;
1824                         if (mddev->ro)
1825                                 mddev->ro = 0;
1826                 }
1827                 if (!mddev->in_sync) {
1828                         /* mark array as shutdown cleanly */
1829                         mddev->in_sync = 1;
1830                         md_update_sb(mddev);
1831                 }
1832                 if (ro)
1833                         set_disk_ro(disk, 1);
1834         }
1835
1836         bitmap_destroy(mddev);
1837         if (mddev->bitmap_file) {
1838                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1839                 fput(mddev->bitmap_file);
1840                 mddev->bitmap_file = NULL;
1841         }
1842         mddev->bitmap_offset = 0;
1843
1844         /*
1845          * Free resources if final stop
1846          */
1847         if (!ro) {
1848                 struct gendisk *disk;
1849                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1850
1851                 export_array(mddev);
1852
1853                 mddev->array_size = 0;
1854                 disk = mddev->gendisk;
1855                 if (disk)
1856                         set_capacity(disk, 0);
1857                 mddev->changed = 1;
1858         } else
1859                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1860                         mdname(mddev));
1861         err = 0;
1862 out:
1863         return err;
1864 }
1865
1866 static void autorun_array(mddev_t *mddev)
1867 {
1868         mdk_rdev_t *rdev;
1869         struct list_head *tmp;
1870         int err;
1871
1872         if (list_empty(&mddev->disks))
1873                 return;
1874
1875         printk(KERN_INFO "md: running: ");
1876
1877         ITERATE_RDEV(mddev,rdev,tmp) {
1878                 char b[BDEVNAME_SIZE];
1879                 printk("<%s>", bdevname(rdev->bdev,b));
1880         }
1881         printk("\n");
1882
1883         err = do_md_run (mddev);
1884         if (err) {
1885                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1886                 do_md_stop (mddev, 0);
1887         }
1888 }
1889
1890 /*
1891  * lets try to run arrays based on all disks that have arrived
1892  * until now. (those are in pending_raid_disks)
1893  *
1894  * the method: pick the first pending disk, collect all disks with
1895  * the same UUID, remove all from the pending list and put them into
1896  * the 'same_array' list. Then order this list based on superblock
1897  * update time (freshest comes first), kick out 'old' disks and
1898  * compare superblocks. If everything's fine then run it.
1899  *
1900  * If "unit" is allocated, then bump its reference count
1901  */
1902 static void autorun_devices(int part)
1903 {
1904         struct list_head candidates;
1905         struct list_head *tmp;
1906         mdk_rdev_t *rdev0, *rdev;
1907         mddev_t *mddev;
1908         char b[BDEVNAME_SIZE];
1909
1910         printk(KERN_INFO "md: autorun ...\n");
1911         while (!list_empty(&pending_raid_disks)) {
1912                 dev_t dev;
1913                 rdev0 = list_entry(pending_raid_disks.next,
1914                                          mdk_rdev_t, same_set);
1915
1916                 printk(KERN_INFO "md: considering %s ...\n",
1917                         bdevname(rdev0->bdev,b));
1918                 INIT_LIST_HEAD(&candidates);
1919                 ITERATE_RDEV_PENDING(rdev,tmp)
1920                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1921                                 printk(KERN_INFO "md:  adding %s ...\n",
1922                                         bdevname(rdev->bdev,b));
1923                                 list_move(&rdev->same_set, &candidates);
1924                         }
1925                 /*
1926                  * now we have a set of devices, with all of them having
1927                  * mostly sane superblocks. It's time to allocate the
1928                  * mddev.
1929                  */
1930                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1931                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1932                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1933                         break;
1934                 }
1935                 if (part)
1936                         dev = MKDEV(mdp_major,
1937                                     rdev0->preferred_minor << MdpMinorShift);
1938                 else
1939                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1940
1941                 md_probe(dev, NULL, NULL);
1942                 mddev = mddev_find(dev);
1943                 if (!mddev) {
1944                         printk(KERN_ERR
1945                                 "md: cannot allocate memory for md drive.\n");
1946                         break;
1947                 }
1948                 if (mddev_lock(mddev))
1949                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1950                                mdname(mddev));
1951                 else if (mddev->raid_disks || mddev->major_version
1952                          || !list_empty(&mddev->disks)) {
1953                         printk(KERN_WARNING
1954                                 "md: %s already running, cannot run %s\n",
1955                                 mdname(mddev), bdevname(rdev0->bdev,b));
1956                         mddev_unlock(mddev);
1957                 } else {
1958                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1959                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1960                                 list_del_init(&rdev->same_set);
1961                                 if (bind_rdev_to_array(rdev, mddev))
1962                                         export_rdev(rdev);
1963                         }
1964                         autorun_array(mddev);
1965                         mddev_unlock(mddev);
1966                 }
1967                 /* on success, candidates will be empty, on error
1968                  * it won't...
1969                  */
1970                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1971                         export_rdev(rdev);
1972                 mddev_put(mddev);
1973         }
1974         printk(KERN_INFO "md: ... autorun DONE.\n");
1975 }
1976
1977 /*
1978  * import RAID devices based on one partition
1979  * if possible, the array gets run as well.
1980  */
1981
1982 static int autostart_array(dev_t startdev)
1983 {
1984         char b[BDEVNAME_SIZE];
1985         int err = -EINVAL, i;
1986         mdp_super_t *sb = NULL;
1987         mdk_rdev_t *start_rdev = NULL, *rdev;
1988
1989         start_rdev = md_import_device(startdev, 0, 0);
1990         if (IS_ERR(start_rdev))
1991                 return err;
1992
1993
1994         /* NOTE: this can only work for 0.90.0 superblocks */
1995         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1996         if (sb->major_version != 0 ||
1997             sb->minor_version != 90 ) {
1998                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1999                 export_rdev(start_rdev);
2000                 return err;
2001         }
2002
2003         if (start_rdev->faulty) {
2004                 printk(KERN_WARNING
2005                         "md: can not autostart based on faulty %s!\n",
2006                         bdevname(start_rdev->bdev,b));
2007                 export_rdev(start_rdev);
2008                 return err;
2009         }
2010         list_add(&start_rdev->same_set, &pending_raid_disks);
2011
2012         for (i = 0; i < MD_SB_DISKS; i++) {
2013                 mdp_disk_t *desc = sb->disks + i;
2014                 dev_t dev = MKDEV(desc->major, desc->minor);
2015
2016                 if (!dev)
2017                         continue;
2018                 if (dev == startdev)
2019                         continue;
2020                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2021                         continue;
2022                 rdev = md_import_device(dev, 0, 0);
2023                 if (IS_ERR(rdev))
2024                         continue;
2025
2026                 list_add(&rdev->same_set, &pending_raid_disks);
2027         }
2028
2029         /*
2030          * possibly return codes
2031          */
2032         autorun_devices(0);
2033         return 0;
2034
2035 }
2036
2037
2038 static int get_version(void __user * arg)
2039 {
2040         mdu_version_t ver;
2041
2042         ver.major = MD_MAJOR_VERSION;
2043         ver.minor = MD_MINOR_VERSION;
2044         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2045
2046         if (copy_to_user(arg, &ver, sizeof(ver)))
2047                 return -EFAULT;
2048
2049         return 0;
2050 }
2051
2052 static int get_array_info(mddev_t * mddev, void __user * arg)
2053 {
2054         mdu_array_info_t info;
2055         int nr,working,active,failed,spare;
2056         mdk_rdev_t *rdev;
2057         struct list_head *tmp;
2058
2059         nr=working=active=failed=spare=0;
2060         ITERATE_RDEV(mddev,rdev,tmp) {
2061                 nr++;
2062                 if (rdev->faulty)
2063                         failed++;
2064                 else {
2065                         working++;
2066                         if (rdev->in_sync)
2067                                 active++;
2068                         else
2069                                 spare++;
2070                 }
2071         }
2072
2073         info.major_version = mddev->major_version;
2074         info.minor_version = mddev->minor_version;
2075         info.patch_version = MD_PATCHLEVEL_VERSION;
2076         info.ctime         = mddev->ctime;
2077         info.level         = mddev->level;
2078         info.size          = mddev->size;
2079         info.nr_disks      = nr;
2080         info.raid_disks    = mddev->raid_disks;
2081         info.md_minor      = mddev->md_minor;
2082         info.not_persistent= !mddev->persistent;
2083
2084         info.utime         = mddev->utime;
2085         info.state         = 0;
2086         if (mddev->in_sync)
2087                 info.state = (1<<MD_SB_CLEAN);
2088         if (mddev->bitmap && mddev->bitmap_offset)
2089                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2090         info.active_disks  = active;
2091         info.working_disks = working;
2092         info.failed_disks  = failed;
2093         info.spare_disks   = spare;
2094
2095         info.layout        = mddev->layout;
2096         info.chunk_size    = mddev->chunk_size;
2097
2098         if (copy_to_user(arg, &info, sizeof(info)))
2099                 return -EFAULT;
2100
2101         return 0;
2102 }
2103
2104 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2105 {
2106         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2107         char *ptr, *buf = NULL;
2108         int err = -ENOMEM;
2109
2110         file = kmalloc(sizeof(*file), GFP_KERNEL);
2111         if (!file)
2112                 goto out;
2113
2114         /* bitmap disabled, zero the first byte and copy out */
2115         if (!mddev->bitmap || !mddev->bitmap->file) {
2116                 file->pathname[0] = '\0';
2117                 goto copy_out;
2118         }
2119
2120         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2121         if (!buf)
2122                 goto out;
2123
2124         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2125         if (!ptr)
2126                 goto out;
2127
2128         strcpy(file->pathname, ptr);
2129
2130 copy_out:
2131         err = 0;
2132         if (copy_to_user(arg, file, sizeof(*file)))
2133                 err = -EFAULT;
2134 out:
2135         kfree(buf);
2136         kfree(file);
2137         return err;
2138 }
2139
2140 static int get_disk_info(mddev_t * mddev, void __user * arg)
2141 {
2142         mdu_disk_info_t info;
2143         unsigned int nr;
2144         mdk_rdev_t *rdev;
2145
2146         if (copy_from_user(&info, arg, sizeof(info)))
2147                 return -EFAULT;
2148
2149         nr = info.number;
2150
2151         rdev = find_rdev_nr(mddev, nr);
2152         if (rdev) {
2153                 info.major = MAJOR(rdev->bdev->bd_dev);
2154                 info.minor = MINOR(rdev->bdev->bd_dev);
2155                 info.raid_disk = rdev->raid_disk;
2156                 info.state = 0;
2157                 if (rdev->faulty)
2158                         info.state |= (1<<MD_DISK_FAULTY);
2159                 else if (rdev->in_sync) {
2160                         info.state |= (1<<MD_DISK_ACTIVE);
2161                         info.state |= (1<<MD_DISK_SYNC);
2162                 }
2163                 if (test_bit(WriteMostly, &rdev->flags))
2164                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2165         } else {
2166                 info.major = info.minor = 0;
2167                 info.raid_disk = -1;
2168                 info.state = (1<<MD_DISK_REMOVED);
2169         }
2170
2171         if (copy_to_user(arg, &info, sizeof(info)))
2172                 return -EFAULT;
2173
2174         return 0;
2175 }
2176
2177 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2178 {
2179         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2180         mdk_rdev_t *rdev;
2181         dev_t dev = MKDEV(info->major,info->minor);
2182
2183         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2184                 return -EOVERFLOW;
2185
2186         if (!mddev->raid_disks) {
2187                 int err;
2188                 /* expecting a device which has a superblock */
2189                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2190                 if (IS_ERR(rdev)) {
2191                         printk(KERN_WARNING
2192                                 "md: md_import_device returned %ld\n",
2193                                 PTR_ERR(rdev));
2194                         return PTR_ERR(rdev);
2195                 }
2196                 if (!list_empty(&mddev->disks)) {
2197                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2198                                                         mdk_rdev_t, same_set);
2199                         int err = super_types[mddev->major_version]
2200                                 .load_super(rdev, rdev0, mddev->minor_version);
2201                         if (err < 0) {
2202                                 printk(KERN_WARNING
2203                                         "md: %s has different UUID to %s\n",
2204                                         bdevname(rdev->bdev,b),
2205                                         bdevname(rdev0->bdev,b2));
2206                                 export_rdev(rdev);
2207                                 return -EINVAL;
2208                         }
2209                 }
2210                 err = bind_rdev_to_array(rdev, mddev);
2211                 if (err)
2212                         export_rdev(rdev);
2213                 return err;
2214         }
2215
2216         /*
2217          * add_new_disk can be used once the array is assembled
2218          * to add "hot spares".  They must already have a superblock
2219          * written
2220          */
2221         if (mddev->pers) {
2222                 int err;
2223                 if (!mddev->pers->hot_add_disk) {
2224                         printk(KERN_WARNING
2225                                 "%s: personality does not support diskops!\n",
2226                                mdname(mddev));
2227                         return -EINVAL;
2228                 }
2229                 if (mddev->persistent)
2230                         rdev = md_import_device(dev, mddev->major_version,
2231                                                 mddev->minor_version);
2232                 else
2233                         rdev = md_import_device(dev, -1, -1);
2234                 if (IS_ERR(rdev)) {
2235                         printk(KERN_WARNING
2236                                 "md: md_import_device returned %ld\n",
2237                                 PTR_ERR(rdev));
2238                         return PTR_ERR(rdev);
2239                 }
2240                 /* set save_raid_disk if appropriate */
2241                 if (!mddev->persistent) {
2242                         if (info->state & (1<<MD_DISK_SYNC)  &&
2243                             info->raid_disk < mddev->raid_disks)
2244                                 rdev->raid_disk = info->raid_disk;
2245                         else
2246                                 rdev->raid_disk = -1;
2247                 } else
2248                         super_types[mddev->major_version].
2249                                 validate_super(mddev, rdev);
2250                 rdev->saved_raid_disk = rdev->raid_disk;
2251
2252                 rdev->in_sync = 0; /* just to be sure */
2253                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2254                         set_bit(WriteMostly, &rdev->flags);
2255
2256                 rdev->raid_disk = -1;
2257                 err = bind_rdev_to_array(rdev, mddev);
2258                 if (err)
2259                         export_rdev(rdev);
2260
2261                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2262                 md_wakeup_thread(mddev->thread);
2263                 return err;
2264         }
2265
2266         /* otherwise, add_new_disk is only allowed
2267          * for major_version==0 superblocks
2268          */
2269         if (mddev->major_version != 0) {
2270                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2271                        mdname(mddev));
2272                 return -EINVAL;
2273         }
2274
2275         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2276                 int err;
2277                 rdev = md_import_device (dev, -1, 0);
2278                 if (IS_ERR(rdev)) {
2279                         printk(KERN_WARNING
2280                                 "md: error, md_import_device() returned %ld\n",
2281                                 PTR_ERR(rdev));
2282                         return PTR_ERR(rdev);
2283                 }
2284                 rdev->desc_nr = info->number;
2285                 if (info->raid_disk < mddev->raid_disks)
2286                         rdev->raid_disk = info->raid_disk;
2287                 else
2288                         rdev->raid_disk = -1;
2289
2290                 rdev->faulty = 0;
2291                 if (rdev->raid_disk < mddev->raid_disks)
2292                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2293                 else
2294                         rdev->in_sync = 0;
2295
2296                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2297                         set_bit(WriteMostly, &rdev->flags);
2298
2299                 err = bind_rdev_to_array(rdev, mddev);
2300                 if (err) {
2301                         export_rdev(rdev);
2302                         return err;
2303                 }
2304
2305                 if (!mddev->persistent) {
2306                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2307                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2308                 } else
2309                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2310                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2311
2312                 if (!mddev->size || (mddev->size > rdev->size))
2313                         mddev->size = rdev->size;
2314         }
2315
2316         return 0;
2317 }
2318
2319 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2320 {
2321         char b[BDEVNAME_SIZE];
2322         mdk_rdev_t *rdev;
2323
2324         if (!mddev->pers)
2325                 return -ENODEV;
2326
2327         rdev = find_rdev(mddev, dev);
2328         if (!rdev)
2329                 return -ENXIO;
2330
2331         if (rdev->raid_disk >= 0)
2332                 goto busy;
2333
2334         kick_rdev_from_array(rdev);
2335         md_update_sb(mddev);
2336
2337         return 0;
2338 busy:
2339         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2340                 bdevname(rdev->bdev,b), mdname(mddev));
2341         return -EBUSY;
2342 }
2343
2344 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2345 {
2346         char b[BDEVNAME_SIZE];
2347         int err;
2348         unsigned int size;
2349         mdk_rdev_t *rdev;
2350
2351         if (!mddev->pers)
2352                 return -ENODEV;
2353
2354         if (mddev->major_version != 0) {
2355                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2356                         " version-0 superblocks.\n",
2357                         mdname(mddev));
2358                 return -EINVAL;
2359         }
2360         if (!mddev->pers->hot_add_disk) {
2361                 printk(KERN_WARNING
2362                         "%s: personality does not support diskops!\n",
2363                         mdname(mddev));
2364                 return -EINVAL;
2365         }
2366
2367         rdev = md_import_device (dev, -1, 0);
2368         if (IS_ERR(rdev)) {
2369                 printk(KERN_WARNING
2370                         "md: error, md_import_device() returned %ld\n",
2371                         PTR_ERR(rdev));
2372                 return -EINVAL;
2373         }
2374
2375         if (mddev->persistent)
2376                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2377         else
2378                 rdev->sb_offset =
2379                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2380
2381         size = calc_dev_size(rdev, mddev->chunk_size);
2382         rdev->size = size;
2383
2384         if (size < mddev->size) {
2385                 printk(KERN_WARNING
2386                         "%s: disk size %llu blocks < array size %llu\n",
2387                         mdname(mddev), (unsigned long long)size,
2388                         (unsigned long long)mddev->size);
2389                 err = -ENOSPC;
2390                 goto abort_export;
2391         }
2392
2393         if (rdev->faulty) {
2394                 printk(KERN_WARNING
2395                         "md: can not hot-add faulty %s disk to %s!\n",
2396                         bdevname(rdev->bdev,b), mdname(mddev));
2397                 err = -EINVAL;
2398                 goto abort_export;
2399         }
2400         rdev->in_sync = 0;
2401         rdev->desc_nr = -1;
2402         bind_rdev_to_array(rdev, mddev);
2403
2404         /*
2405          * The rest should better be atomic, we can have disk failures
2406          * noticed in interrupt contexts ...
2407          */
2408
2409         if (rdev->desc_nr == mddev->max_disks) {
2410                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2411                         mdname(mddev));
2412                 err = -EBUSY;
2413                 goto abort_unbind_export;
2414         }
2415
2416         rdev->raid_disk = -1;
2417
2418         md_update_sb(mddev);
2419
2420         /*
2421          * Kick recovery, maybe this spare has to be added to the
2422          * array immediately.
2423          */
2424         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2425         md_wakeup_thread(mddev->thread);
2426
2427         return 0;
2428
2429 abort_unbind_export:
2430         unbind_rdev_from_array(rdev);
2431
2432 abort_export:
2433         export_rdev(rdev);
2434         return err;
2435 }
2436
2437 /* similar to deny_write_access, but accounts for our holding a reference
2438  * to the file ourselves */
2439 static int deny_bitmap_write_access(struct file * file)
2440 {
2441         struct inode *inode = file->f_mapping->host;
2442
2443         spin_lock(&inode->i_lock);
2444         if (atomic_read(&inode->i_writecount) > 1) {
2445                 spin_unlock(&inode->i_lock);
2446                 return -ETXTBSY;
2447         }
2448         atomic_set(&inode->i_writecount, -1);
2449         spin_unlock(&inode->i_lock);
2450
2451         return 0;
2452 }
2453
2454 static int set_bitmap_file(mddev_t *mddev, int fd)
2455 {
2456         int err;
2457
2458         if (mddev->pers) {
2459                 if (!mddev->pers->quiesce)
2460                         return -EBUSY;
2461                 if (mddev->recovery || mddev->sync_thread)
2462                         return -EBUSY;
2463                 /* we should be able to change the bitmap.. */
2464         }
2465
2466
2467         if (fd >= 0) {
2468                 if (mddev->bitmap)
2469                         return -EEXIST; /* cannot add when bitmap is present */
2470                 mddev->bitmap_file = fget(fd);
2471
2472                 if (mddev->bitmap_file == NULL) {
2473                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2474                                mdname(mddev));
2475                         return -EBADF;
2476                 }
2477
2478                 err = deny_bitmap_write_access(mddev->bitmap_file);
2479                 if (err) {
2480                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2481                                mdname(mddev));
2482                         fput(mddev->bitmap_file);
2483                         mddev->bitmap_file = NULL;
2484                         return err;
2485                 }
2486                 mddev->bitmap_offset = 0; /* file overrides offset */
2487         } else if (mddev->bitmap == NULL)
2488                 return -ENOENT; /* cannot remove what isn't there */
2489         err = 0;
2490         if (mddev->pers) {
2491                 mddev->pers->quiesce(mddev, 1);
2492                 if (fd >= 0)
2493                         err = bitmap_create(mddev);
2494                 if (fd < 0 || err)
2495                         bitmap_destroy(mddev);
2496                 mddev->pers->quiesce(mddev, 0);
2497         } else if (fd < 0) {
2498                 if (mddev->bitmap_file)
2499                         fput(mddev->bitmap_file);
2500                 mddev->bitmap_file = NULL;
2501         }
2502
2503         return err;
2504 }
2505
2506 /*
2507  * set_array_info is used two different ways
2508  * The original usage is when creating a new array.
2509  * In this usage, raid_disks is > 0 and it together with
2510  *  level, size, not_persistent,layout,chunksize determine the
2511  *  shape of the array.
2512  *  This will always create an array with a type-0.90.0 superblock.
2513  * The newer usage is when assembling an array.
2514  *  In this case raid_disks will be 0, and the major_version field is
2515  *  use to determine which style super-blocks are to be found on the devices.
2516  *  The minor and patch _version numbers are also kept incase the
2517  *  super_block handler wishes to interpret them.
2518  */
2519 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2520 {
2521
2522         if (info->raid_disks == 0) {
2523                 /* just setting version number for superblock loading */
2524                 if (info->major_version < 0 ||
2525                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2526                     super_types[info->major_version].name == NULL) {
2527                         /* maybe try to auto-load a module? */
2528                         printk(KERN_INFO
2529                                 "md: superblock version %d not known\n",
2530                                 info->major_version);
2531                         return -EINVAL;
2532                 }
2533                 mddev->major_version = info->major_version;
2534                 mddev->minor_version = info->minor_version;
2535                 mddev->patch_version = info->patch_version;
2536                 return 0;
2537         }
2538         mddev->major_version = MD_MAJOR_VERSION;
2539         mddev->minor_version = MD_MINOR_VERSION;
2540         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2541         mddev->ctime         = get_seconds();
2542
2543         mddev->level         = info->level;
2544         mddev->size          = info->size;
2545         mddev->raid_disks    = info->raid_disks;
2546         /* don't set md_minor, it is determined by which /dev/md* was
2547          * openned
2548          */
2549         if (info->state & (1<<MD_SB_CLEAN))
2550                 mddev->recovery_cp = MaxSector;
2551         else
2552                 mddev->recovery_cp = 0;
2553         mddev->persistent    = ! info->not_persistent;
2554
2555         mddev->layout        = info->layout;
2556         mddev->chunk_size    = info->chunk_size;
2557
2558         mddev->max_disks     = MD_SB_DISKS;
2559
2560         mddev->sb_dirty      = 1;
2561
2562         /*
2563          * Generate a 128 bit UUID
2564          */
2565         get_random_bytes(mddev->uuid, 16);
2566
2567         return 0;
2568 }
2569
2570 /*
2571  * update_array_info is used to change the configuration of an
2572  * on-line array.
2573  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2574  * fields in the info are checked against the array.
2575  * Any differences that cannot be handled will cause an error.
2576  * Normally, only one change can be managed at a time.
2577  */
2578 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2579 {
2580         int rv = 0;
2581         int cnt = 0;
2582         int state = 0;
2583
2584         /* calculate expected state,ignoring low bits */
2585         if (mddev->bitmap && mddev->bitmap_offset)
2586                 state |= (1 << MD_SB_BITMAP_PRESENT);
2587
2588         if (mddev->major_version != info->major_version ||
2589             mddev->minor_version != info->minor_version ||
2590 /*          mddev->patch_version != info->patch_version || */
2591             mddev->ctime         != info->ctime         ||
2592             mddev->level         != info->level         ||
2593 /*          mddev->layout        != info->layout        || */
2594             !mddev->persistent   != info->not_persistent||
2595             mddev->chunk_size    != info->chunk_size    ||
2596             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2597             ((state^info->state) & 0xfffffe00)
2598                 )
2599                 return -EINVAL;
2600         /* Check there is only one change */
2601         if (mddev->size != info->size) cnt++;
2602         if (mddev->raid_disks != info->raid_disks) cnt++;
2603         if (mddev->layout != info->layout) cnt++;
2604         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2605         if (cnt == 0) return 0;
2606         if (cnt > 1) return -EINVAL;
2607
2608         if (mddev->layout != info->layout) {
2609                 /* Change layout
2610                  * we don't need to do anything at the md level, the
2611                  * personality will take care of it all.
2612                  */
2613                 if (mddev->pers->reconfig == NULL)
2614                         return -EINVAL;
2615                 else
2616                         return mddev->pers->reconfig(mddev, info->layout, -1);
2617         }
2618         if (mddev->size != info->size) {
2619                 mdk_rdev_t * rdev;
2620                 struct list_head *tmp;
2621                 if (mddev->pers->resize == NULL)
2622                         return -EINVAL;
2623                 /* The "size" is the amount of each device that is used.
2624                  * This can only make sense for arrays with redundancy.
2625                  * linear and raid0 always use whatever space is available
2626                  * We can only consider changing the size if no resync
2627                  * or reconstruction is happening, and if the new size
2628                  * is acceptable. It must fit before the sb_offset or,
2629                  * if that is <data_offset, it must fit before the
2630                  * size of each device.
2631                  * If size is zero, we find the largest size that fits.
2632                  */
2633                 if (mddev->sync_thread)
2634                         return -EBUSY;
2635                 ITERATE_RDEV(mddev,rdev,tmp) {
2636                         sector_t avail;
2637                         int fit = (info->size == 0);
2638                         if (rdev->sb_offset > rdev->data_offset)
2639                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2640                         else
2641                                 avail = get_capacity(rdev->bdev->bd_disk)
2642                                         - rdev->data_offset;
2643                         if (fit && (info->size == 0 || info->size > avail/2))
2644                                 info->size = avail/2;
2645                         if (avail < ((sector_t)info->size << 1))
2646                                 return -ENOSPC;
2647                 }
2648                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2649                 if (!rv) {
2650                         struct block_device *bdev;
2651
2652                         bdev = bdget_disk(mddev->gendisk, 0);
2653                         if (bdev) {
2654                                 down(&bdev->bd_inode->i_sem);
2655                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2656                                 up(&bdev->bd_inode->i_sem);
2657                                 bdput(bdev);
2658                         }
2659                 }
2660         }
2661         if (mddev->raid_disks    != info->raid_disks) {
2662                 /* change the number of raid disks */
2663                 if (mddev->pers->reshape == NULL)
2664                         return -EINVAL;
2665                 if (info->raid_disks <= 0 ||
2666                     info->raid_disks >= mddev->max_disks)
2667                         return -EINVAL;
2668                 if (mddev->sync_thread)
2669                         return -EBUSY;
2670                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2671                 if (!rv) {
2672                         struct block_device *bdev;
2673
2674                         bdev = bdget_disk(mddev->gendisk, 0);
2675                         if (bdev) {
2676                                 down(&bdev->bd_inode->i_sem);
2677                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2678                                 up(&bdev->bd_inode->i_sem);
2679                                 bdput(bdev);
2680                         }
2681                 }
2682         }
2683         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2684                 if (mddev->pers->quiesce == NULL)
2685                         return -EINVAL;
2686                 if (mddev->recovery || mddev->sync_thread)
2687                         return -EBUSY;
2688                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2689                         /* add the bitmap */
2690                         if (mddev->bitmap)
2691                                 return -EEXIST;
2692                         if (mddev->default_bitmap_offset == 0)
2693                                 return -EINVAL;
2694                         mddev->bitmap_offset = mddev->default_bitmap_offset;
2695                         mddev->pers->quiesce(mddev, 1);
2696                         rv = bitmap_create(mddev);
2697                         if (rv)
2698                                 bitmap_destroy(mddev);
2699                         mddev->pers->quiesce(mddev, 0);
2700                 } else {
2701                         /* remove the bitmap */
2702                         if (!mddev->bitmap)
2703                                 return -ENOENT;
2704                         if (mddev->bitmap->file)
2705                                 return -EINVAL;
2706                         mddev->pers->quiesce(mddev, 1);
2707                         bitmap_destroy(mddev);
2708                         mddev->pers->quiesce(mddev, 0);
2709                         mddev->bitmap_offset = 0;
2710                 }
2711         }
2712         md_update_sb(mddev);
2713         return rv;
2714 }
2715
2716 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2717 {
2718         mdk_rdev_t *rdev;
2719
2720         if (mddev->pers == NULL)
2721                 return -ENODEV;
2722
2723         rdev = find_rdev(mddev, dev);
2724         if (!rdev)
2725                 return -ENODEV;
2726
2727         md_error(mddev, rdev);
2728         return 0;
2729 }
2730
2731 static int md_ioctl(struct inode *inode, struct file *file,
2732                         unsigned int cmd, unsigned long arg)
2733 {
2734         int err = 0;
2735         void __user *argp = (void __user *)arg;
2736         struct hd_geometry __user *loc = argp;
2737         mddev_t *mddev = NULL;
2738
2739         if (!capable(CAP_SYS_ADMIN))
2740                 return -EACCES;
2741
2742         /*
2743          * Commands dealing with the RAID driver but not any
2744          * particular array:
2745          */
2746         switch (cmd)
2747         {
2748                 case RAID_VERSION:
2749                         err = get_version(argp);
2750                         goto done;
2751
2752                 case PRINT_RAID_DEBUG:
2753                         err = 0;
2754                         md_print_devices();
2755                         goto done;
2756
2757 #ifndef MODULE
2758                 case RAID_AUTORUN:
2759                         err = 0;
2760                         autostart_arrays(arg);
2761                         goto done;
2762 #endif
2763                 default:;
2764         }
2765
2766         /*
2767          * Commands creating/starting a new array:
2768          */
2769
2770         mddev = inode->i_bdev->bd_disk->private_data;
2771
2772         if (!mddev) {
2773                 BUG();
2774                 goto abort;
2775         }
2776
2777
2778         if (cmd == START_ARRAY) {
2779                 /* START_ARRAY doesn't need to lock the array as autostart_array
2780                  * does the locking, and it could even be a different array
2781                  */
2782                 static int cnt = 3;
2783                 if (cnt > 0 ) {
2784                         printk(KERN_WARNING
2785                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2786                                "This will not be supported beyond 2.6\n",
2787                                current->comm, current->pid);
2788                         cnt--;
2789                 }
2790                 err = autostart_array(new_decode_dev(arg));
2791                 if (err) {
2792                         printk(KERN_WARNING "md: autostart failed!\n");
2793                         goto abort;
2794                 }
2795                 goto done;
2796         }
2797
2798         err = mddev_lock(mddev);
2799         if (err) {
2800                 printk(KERN_INFO
2801                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2802                         err, cmd);
2803                 goto abort;
2804         }
2805
2806         switch (cmd)
2807         {
2808                 case SET_ARRAY_INFO:
2809                         {
2810                                 mdu_array_info_t info;
2811                                 if (!arg)
2812                                         memset(&info, 0, sizeof(info));
2813                                 else if (copy_from_user(&info, argp, sizeof(info))) {
2814                                         err = -EFAULT;
2815                                         goto abort_unlock;
2816                                 }
2817                                 if (mddev->pers) {
2818                                         err = update_array_info(mddev, &info);
2819                                         if (err) {
2820                                                 printk(KERN_WARNING "md: couldn't update"
2821                                                        " array info. %d\n", err);
2822                                                 goto abort_unlock;
2823                                         }
2824                                         goto done_unlock;
2825                                 }
2826                                 if (!list_empty(&mddev->disks)) {
2827                                         printk(KERN_WARNING
2828                                                "md: array %s already has disks!\n",
2829                                                mdname(mddev));
2830                                         err = -EBUSY;
2831                                         goto abort_unlock;
2832                                 }
2833                                 if (mddev->raid_disks) {
2834                                         printk(KERN_WARNING
2835                                                "md: array %s already initialised!\n",
2836                                                mdname(mddev));
2837                                         err = -EBUSY;
2838                                         goto abort_unlock;
2839                                 }
2840                                 err = set_array_info(mddev, &info);
2841                                 if (err) {
2842                                         printk(KERN_WARNING "md: couldn't set"
2843                                                " array info. %d\n", err);
2844                                         goto abort_unlock;
2845                                 }
2846                         }
2847                         goto done_unlock;
2848
2849                 default:;
2850         }
2851
2852         /*
2853          * Commands querying/configuring an existing array:
2854          */
2855         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2856          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2857         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2858                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2859                 err = -ENODEV;
2860                 goto abort_unlock;
2861         }
2862
2863         /*
2864          * Commands even a read-only array can execute:
2865          */
2866         switch (cmd)
2867         {
2868                 case GET_ARRAY_INFO:
2869                         err = get_array_info(mddev, argp);
2870                         goto done_unlock;
2871
2872                 case GET_BITMAP_FILE:
2873                         err = get_bitmap_file(mddev, argp);
2874                         goto done_unlock;
2875
2876                 case GET_DISK_INFO:
2877                         err = get_disk_info(mddev, argp);
2878                         goto done_unlock;
2879
2880                 case RESTART_ARRAY_RW:
2881                         err = restart_array(mddev);
2882                         goto done_unlock;
2883
2884                 case STOP_ARRAY:
2885                         err = do_md_stop (mddev, 0);
2886                         goto done_unlock;
2887
2888                 case STOP_ARRAY_RO:
2889                         err = do_md_stop (mddev, 1);
2890                         goto done_unlock;
2891
2892         /*
2893          * We have a problem here : there is no easy way to give a CHS
2894          * virtual geometry. We currently pretend that we have a 2 heads
2895          * 4 sectors (with a BIG number of cylinders...). This drives
2896          * dosfs just mad... ;-)
2897          */
2898                 case HDIO_GETGEO:
2899                         if (!loc) {
2900                                 err = -EINVAL;
2901                                 goto abort_unlock;
2902                         }
2903                         err = put_user (2, (char __user *) &loc->heads);
2904                         if (err)
2905                                 goto abort_unlock;
2906                         err = put_user (4, (char __user *) &loc->sectors);
2907                         if (err)
2908                                 goto abort_unlock;
2909                         err = put_user(get_capacity(mddev->gendisk)/8,
2910                                         (short __user *) &loc->cylinders);
2911                         if (err)
2912                                 goto abort_unlock;
2913                         err = put_user (get_start_sect(inode->i_bdev),
2914                                                 (long __user *) &loc->start);
2915                         goto done_unlock;
2916         }
2917
2918         /*
2919          * The remaining ioctls are changing the state of the
2920          * superblock, so we do not allow read-only arrays
2921          * here:
2922          */
2923         if (mddev->ro) {
2924                 err = -EROFS;
2925                 goto abort_unlock;
2926         }
2927
2928         switch (cmd)
2929         {
2930                 case ADD_NEW_DISK:
2931                 {
2932                         mdu_disk_info_t info;
2933                         if (copy_from_user(&info, argp, sizeof(info)))
2934                                 err = -EFAULT;
2935                         else
2936                                 err = add_new_disk(mddev, &info);
2937                         goto done_unlock;
2938                 }
2939
2940                 case HOT_REMOVE_DISK:
2941                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2942                         goto done_unlock;
2943
2944                 case HOT_ADD_DISK:
2945                         err = hot_add_disk(mddev, new_decode_dev(arg));
2946                         goto done_unlock;
2947
2948                 case SET_DISK_FAULTY:
2949                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2950                         goto done_unlock;
2951
2952                 case RUN_ARRAY:
2953                         err = do_md_run (mddev);
2954                         goto done_unlock;
2955
2956                 case SET_BITMAP_FILE:
2957                         err = set_bitmap_file(mddev, (int)arg);
2958                         goto done_unlock;
2959
2960                 default:
2961                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2962                                 printk(KERN_WARNING "md: %s(pid %d) used"
2963                                         " obsolete MD ioctl, upgrade your"
2964                                         " software to use new ictls.\n",
2965                                         current->comm, current->pid);
2966                         err = -EINVAL;
2967                         goto abort_unlock;
2968         }
2969
2970 done_unlock:
2971 abort_unlock:
2972         mddev_unlock(mddev);
2973
2974         return err;
2975 done:
2976         if (err)
2977                 MD_BUG();
2978 abort:
2979         return err;
2980 }
2981
2982 static int md_open(struct inode *inode, struct file *file)
2983 {
2984         /*
2985          * Succeed if we can lock the mddev, which confirms that
2986          * it isn't being stopped right now.
2987          */
2988         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2989         int err;
2990
2991         if ((err = mddev_lock(mddev)))
2992                 goto out;
2993
2994         err = 0;
2995         mddev_get(mddev);
2996         mddev_unlock(mddev);
2997
2998         check_disk_change(inode->i_bdev);
2999  out:
3000         return err;
3001 }
3002
3003 static int md_release(struct inode *inode, struct file * file)
3004 {
3005         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3006
3007         if (!mddev)
3008                 BUG();
3009         mddev_put(mddev);
3010
3011         return 0;
3012 }
3013
3014 static int md_media_changed(struct gendisk *disk)
3015 {
3016         mddev_t *mddev = disk->private_data;
3017
3018         return mddev->changed;
3019 }
3020
3021 static int md_revalidate(struct gendisk *disk)
3022 {
3023         mddev_t *mddev = disk->private_data;
3024
3025         mddev->changed = 0;
3026         return 0;
3027 }
3028 static struct block_device_operations md_fops =
3029 {
3030         .owner          = THIS_MODULE,
3031         .open           = md_open,
3032         .release        = md_release,
3033         .ioctl          = md_ioctl,
3034         .media_changed  = md_media_changed,
3035         .revalidate_disk= md_revalidate,
3036 };
3037
3038 static int md_thread(void * arg)
3039 {
3040         mdk_thread_t *thread = arg;
3041
3042         lock_kernel();
3043
3044         /*
3045          * Detach thread
3046          */
3047
3048         daemonize(thread->name, mdname(thread->mddev));
3049
3050         current->exit_signal = SIGCHLD;
3051         allow_signal(SIGKILL);
3052         thread->tsk = current;
3053
3054         /*
3055          * md_thread is a 'system-thread', it's priority should be very
3056          * high. We avoid resource deadlocks individually in each
3057          * raid personality. (RAID5 does preallocation) We also use RR and
3058          * the very same RT priority as kswapd, thus we will never get
3059          * into a priority inversion deadlock.
3060          *
3061          * we definitely have to have equal or higher priority than
3062          * bdflush, otherwise bdflush will deadlock if there are too
3063          * many dirty RAID5 blocks.
3064          */
3065         unlock_kernel();
3066
3067         complete(thread->event);
3068         while (thread->run) {
3069                 void (*run)(mddev_t *);
3070
3071                 wait_event_interruptible_timeout(thread->wqueue,
3072                                                  test_bit(THREAD_WAKEUP, &thread->flags),
3073                                                  thread->timeout);
3074                 try_to_freeze();
3075
3076                 clear_bit(THREAD_WAKEUP, &thread->flags);
3077
3078                 run = thread->run;
3079                 if (run)
3080                         run(thread->mddev);
3081
3082                 if (signal_pending(current))
3083                         flush_signals(current);
3084         }
3085         complete(thread->event);
3086         return 0;
3087 }
3088
3089 void md_wakeup_thread(mdk_thread_t *thread)
3090 {
3091         if (thread) {
3092                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3093                 set_bit(THREAD_WAKEUP, &thread->flags);
3094                 wake_up(&thread->wqueue);
3095         }
3096 }
3097
3098 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3099                                  const char *name)
3100 {
3101         mdk_thread_t *thread;
3102         int ret;
3103         struct completion event;
3104
3105         thread = (mdk_thread_t *) kmalloc
3106                                 (sizeof(mdk_thread_t), GFP_KERNEL);
3107         if (!thread)
3108                 return NULL;
3109
3110         memset(thread, 0, sizeof(mdk_thread_t));
3111         init_waitqueue_head(&thread->wqueue);
3112
3113         init_completion(&event);
3114         thread->event = &event;
3115         thread->run = run;
3116         thread->mddev = mddev;
3117         thread->name = name;
3118         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3119         ret = kernel_thread(md_thread, thread, 0);
3120         if (ret < 0) {
3121                 kfree(thread);
3122                 return NULL;
3123         }
3124         wait_for_completion(&event);
3125         return thread;
3126 }
3127
3128 void md_unregister_thread(mdk_thread_t *thread)
3129 {
3130         struct completion event;
3131
3132         init_completion(&event);
3133
3134         thread->event = &event;
3135
3136         /* As soon as ->run is set to NULL, the task could disappear,
3137          * so we need to hold tasklist_lock until we have sent the signal
3138          */
3139         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3140         read_lock(&tasklist_lock);
3141         thread->run = NULL;
3142         send_sig(SIGKILL, thread->tsk, 1);
3143         read_unlock(&tasklist_lock);
3144         wait_for_completion(&event);
3145         kfree(thread);
3146 }
3147
3148 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3149 {
3150         if (!mddev) {
3151                 MD_BUG();
3152                 return;
3153         }
3154
3155         if (!rdev || rdev->faulty)
3156                 return;
3157 /*
3158         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3159                 mdname(mddev),
3160                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3161                 __builtin_return_address(0),__builtin_return_address(1),
3162                 __builtin_return_address(2),__builtin_return_address(3));
3163 */
3164         if (!mddev->pers->error_handler)
3165                 return;
3166         mddev->pers->error_handler(mddev,rdev);
3167         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3168         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3169         md_wakeup_thread(mddev->thread);
3170 }
3171
3172 /* seq_file implementation /proc/mdstat */
3173
3174 static void status_unused(struct seq_file *seq)
3175 {
3176         int i = 0;
3177         mdk_rdev_t *rdev;
3178         struct list_head *tmp;
3179
3180         seq_printf(seq, "unused devices: ");
3181
3182         ITERATE_RDEV_PENDING(rdev,tmp) {
3183                 char b[BDEVNAME_SIZE];
3184                 i++;
3185                 seq_printf(seq, "%s ",
3186                               bdevname(rdev->bdev,b));
3187         }
3188         if (!i)
3189                 seq_printf(seq, "<none>");
3190
3191         seq_printf(seq, "\n");
3192 }
3193
3194
3195 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3196 {
3197         unsigned long max_blocks, resync, res, dt, db, rt;
3198
3199         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3200
3201         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3202                 max_blocks = mddev->resync_max_sectors >> 1;
3203         else
3204                 max_blocks = mddev->size;
3205
3206         /*
3207          * Should not happen.
3208          */
3209         if (!max_blocks) {
3210                 MD_BUG();
3211                 return;
3212         }
3213         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3214         {
3215                 int i, x = res/50, y = 20-x;
3216                 seq_printf(seq, "[");
3217                 for (i = 0; i < x; i++)
3218                         seq_printf(seq, "=");
3219                 seq_printf(seq, ">");
3220                 for (i = 0; i < y; i++)
3221                         seq_printf(seq, ".");
3222                 seq_printf(seq, "] ");
3223         }
3224         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3225                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3226                        "resync" : "recovery"),
3227                       res/10, res % 10, resync, max_blocks);
3228
3229         /*
3230          * We do not want to overflow, so the order of operands and
3231          * the * 100 / 100 trick are important. We do a +1 to be
3232          * safe against division by zero. We only estimate anyway.
3233          *
3234          * dt: time from mark until now
3235          * db: blocks written from mark until now
3236          * rt: remaining time
3237          */
3238         dt = ((jiffies - mddev->resync_mark) / HZ);
3239         if (!dt) dt++;
3240         db = resync - (mddev->resync_mark_cnt/2);
3241         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3242
3243         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3244
3245         seq_printf(seq, " speed=%ldK/sec", db/dt);
3246 }
3247
3248 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3249 {
3250         struct list_head *tmp;
3251         loff_t l = *pos;
3252         mddev_t *mddev;
3253
3254         if (l >= 0x10000)
3255                 return NULL;
3256         if (!l--)
3257                 /* header */
3258                 return (void*)1;
3259
3260         spin_lock(&all_mddevs_lock);
3261         list_for_each(tmp,&all_mddevs)
3262                 if (!l--) {
3263                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3264                         mddev_get(mddev);
3265                         spin_unlock(&all_mddevs_lock);
3266                         return mddev;
3267                 }
3268         spin_unlock(&all_mddevs_lock);
3269         if (!l--)
3270                 return (void*)2;/* tail */
3271         return NULL;
3272 }
3273
3274 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3275 {
3276         struct list_head *tmp;
3277         mddev_t *next_mddev, *mddev = v;
3278
3279         ++*pos;
3280         if (v == (void*)2)
3281                 return NULL;
3282
3283         spin_lock(&all_mddevs_lock);
3284         if (v == (void*)1)
3285                 tmp = all_mddevs.next;
3286         else
3287                 tmp = mddev->all_mddevs.next;
3288         if (tmp != &all_mddevs)
3289                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3290         else {
3291                 next_mddev = (void*)2;
3292                 *pos = 0x10000;
3293         }
3294         spin_unlock(&all_mddevs_lock);
3295
3296         if (v != (void*)1)
3297                 mddev_put(mddev);
3298         return next_mddev;
3299
3300 }
3301
3302 static void md_seq_stop(struct seq_file *seq, void *v)
3303 {
3304         mddev_t *mddev = v;
3305
3306         if (mddev && v != (void*)1 && v != (void*)2)
3307                 mddev_put(mddev);
3308 }
3309
3310 static int md_seq_show(struct seq_file *seq, void *v)
3311 {
3312         mddev_t *mddev = v;
3313         sector_t size;
3314         struct list_head *tmp2;
3315         mdk_rdev_t *rdev;
3316         int i;
3317         struct bitmap *bitmap;
3318
3319         if (v == (void*)1) {
3320                 seq_printf(seq, "Personalities : ");
3321                 spin_lock(&pers_lock);
3322                 for (i = 0; i < MAX_PERSONALITY; i++)
3323                         if (pers[i])
3324                                 seq_printf(seq, "[%s] ", pers[i]->name);
3325
3326                 spin_unlock(&pers_lock);
3327                 seq_printf(seq, "\n");
3328                 return 0;
3329         }
3330         if (v == (void*)2) {
3331                 status_unused(seq);
3332                 return 0;
3333         }
3334
3335         if (mddev_lock(mddev)!=0)
3336                 return -EINTR;
3337         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3338                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3339                                                 mddev->pers ? "" : "in");
3340                 if (mddev->pers) {
3341                         if (mddev->ro)
3342                                 seq_printf(seq, " (read-only)");
3343                         seq_printf(seq, " %s", mddev->pers->name);
3344                 }
3345
3346                 size = 0;
3347                 ITERATE_RDEV(mddev,rdev,tmp2) {
3348                         char b[BDEVNAME_SIZE];
3349                         seq_printf(seq, " %s[%d]",
3350                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3351                         if (test_bit(WriteMostly, &rdev->flags))
3352                                 seq_printf(seq, "(W)");
3353                         if (rdev->faulty) {
3354                                 seq_printf(seq, "(F)");
3355                                 continue;
3356                         }
3357                         size += rdev->size;
3358                 }
3359
3360                 if (!list_empty(&mddev->disks)) {
3361                         if (mddev->pers)
3362                                 seq_printf(seq, "\n      %llu blocks",
3363                                         (unsigned long long)mddev->array_size);
3364                         else
3365                                 seq_printf(seq, "\n      %llu blocks",
3366                                         (unsigned long long)size);
3367                 }
3368
3369                 if (mddev->pers) {
3370                         mddev->pers->status (seq, mddev);
3371                         seq_printf(seq, "\n      ");
3372                         if (mddev->curr_resync > 2) {
3373                                 status_resync (seq, mddev);
3374                                 seq_printf(seq, "\n      ");
3375                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3376                                 seq_printf(seq, "       resync=DELAYED\n      ");
3377                 } else
3378                         seq_printf(seq, "\n       ");
3379
3380                 if ((bitmap = mddev->bitmap)) {
3381                         unsigned long chunk_kb;
3382                         unsigned long flags;
3383                         spin_lock_irqsave(&bitmap->lock, flags);
3384                         chunk_kb = bitmap->chunksize >> 10;
3385                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3386                                 "%lu%s chunk",
3387                                 bitmap->pages - bitmap->missing_pages,
3388                                 bitmap->pages,
3389                                 (bitmap->pages - bitmap->missing_pages)
3390                                         << (PAGE_SHIFT - 10),
3391                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3392                                 chunk_kb ? "KB" : "B");
3393                         if (bitmap->file) {
3394                                 seq_printf(seq, ", file: ");
3395                                 seq_path(seq, bitmap->file->f_vfsmnt,
3396                                          bitmap->file->f_dentry," \t\n");
3397                         }
3398
3399                         seq_printf(seq, "\n");
3400                         spin_unlock_irqrestore(&bitmap->lock, flags);
3401                 }
3402
3403                 seq_printf(seq, "\n");
3404         }
3405         mddev_unlock(mddev);
3406
3407         return 0;
3408 }
3409
3410 static struct seq_operations md_seq_ops = {
3411         .start  = md_seq_start,
3412         .next   = md_seq_next,
3413         .stop   = md_seq_stop,
3414         .show   = md_seq_show,
3415 };
3416
3417 static int md_seq_open(struct inode *inode, struct file *file)
3418 {
3419         int error;
3420
3421         error = seq_open(file, &md_seq_ops);
3422         return error;
3423 }
3424
3425 static struct file_operations md_seq_fops = {
3426         .open           = md_seq_open,
3427         .read           = seq_read,
3428         .llseek         = seq_lseek,
3429         .release        = seq_release,
3430 };
3431
3432 int register_md_personality(int pnum, mdk_personality_t *p)
3433 {
3434         if (pnum >= MAX_PERSONALITY) {
3435                 printk(KERN_ERR
3436                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3437                        p->name, pnum, MAX_PERSONALITY-1);
3438                 return -EINVAL;
3439         }
3440
3441         spin_lock(&pers_lock);
3442         if (pers[pnum]) {
3443                 spin_unlock(&pers_lock);
3444                 return -EBUSY;
3445         }
3446
3447         pers[pnum] = p;
3448         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3449         spin_unlock(&pers_lock);
3450         return 0;
3451 }
3452
3453 int unregister_md_personality(int pnum)
3454 {
3455         if (pnum >= MAX_PERSONALITY)
3456                 return -EINVAL;
3457
3458         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3459         spin_lock(&pers_lock);
3460         pers[pnum] = NULL;
3461         spin_unlock(&pers_lock);
3462         return 0;
3463 }
3464
3465 static int is_mddev_idle(mddev_t *mddev)
3466 {
3467         mdk_rdev_t * rdev;
3468         struct list_head *tmp;
3469         int idle;
3470         unsigned long curr_events;
3471
3472         idle = 1;
3473         ITERATE_RDEV(mddev,rdev,tmp) {
3474                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3475                 curr_events = disk_stat_read(disk, read_sectors) +
3476                                 disk_stat_read(disk, write_sectors) -
3477                                 atomic_read(&disk->sync_io);
3478                 /* Allow some slack between valud of curr_events and last_events,
3479                  * as there are some uninteresting races.
3480                  * Note: the following is an unsigned comparison.
3481                  */
3482                 if ((curr_events - rdev->last_events + 32) > 64) {
3483                         rdev->last_events = curr_events;
3484                         idle = 0;
3485                 }
3486         }
3487         return idle;
3488 }
3489
3490 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3491 {
3492         /* another "blocks" (512byte) blocks have been synced */
3493         atomic_sub(blocks, &mddev->recovery_active);
3494         wake_up(&mddev->recovery_wait);
3495         if (!ok) {
3496                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3497                 md_wakeup_thread(mddev->thread);
3498                 // stop recovery, signal do_sync ....
3499         }
3500 }
3501
3502
3503 /* md_write_start(mddev, bi)
3504  * If we need to update some array metadata (e.g. 'active' flag
3505  * in superblock) before writing, schedule a superblock update
3506  * and wait for it to complete.
3507  */
3508 void md_write_start(mddev_t *mddev, struct bio *bi)
3509 {
3510         DEFINE_WAIT(w);
3511         if (bio_data_dir(bi) != WRITE)
3512                 return;
3513
3514         atomic_inc(&mddev->writes_pending);
3515         if (mddev->in_sync) {
3516                 spin_lock(&mddev->write_lock);
3517                 if (mddev->in_sync) {
3518                         mddev->in_sync = 0;
3519                         mddev->sb_dirty = 1;
3520                         md_wakeup_thread(mddev->thread);
3521                 }
3522                 spin_unlock(&mddev->write_lock);
3523         }
3524         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3525 }
3526
3527 void md_write_end(mddev_t *mddev)
3528 {
3529         if (atomic_dec_and_test(&mddev->writes_pending)) {
3530                 if (mddev->safemode == 2)
3531                         md_wakeup_thread(mddev->thread);
3532                 else
3533                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3534         }
3535 }
3536
3537 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3538
3539 #define SYNC_MARKS      10
3540 #define SYNC_MARK_STEP  (3*HZ)
3541 static void md_do_sync(mddev_t *mddev)
3542 {
3543         mddev_t *mddev2;
3544         unsigned int currspeed = 0,
3545                  window;
3546         sector_t max_sectors,j, io_sectors;
3547         unsigned long mark[SYNC_MARKS];
3548         sector_t mark_cnt[SYNC_MARKS];
3549         int last_mark,m;
3550         struct list_head *tmp;
3551         sector_t last_check;
3552         int skipped = 0;
3553
3554         /* just incase thread restarts... */
3555         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3556                 return;
3557
3558         /* we overload curr_resync somewhat here.
3559          * 0 == not engaged in resync at all
3560          * 2 == checking that there is no conflict with another sync
3561          * 1 == like 2, but have yielded to allow conflicting resync to
3562          *              commense
3563          * other == active in resync - this many blocks
3564          *
3565          * Before starting a resync we must have set curr_resync to
3566          * 2, and then checked that every "conflicting" array has curr_resync
3567          * less than ours.  When we find one that is the same or higher
3568          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3569          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3570          * This will mean we have to start checking from the beginning again.
3571          *
3572          */
3573
3574         do {
3575                 mddev->curr_resync = 2;
3576
3577         try_again:
3578                 if (signal_pending(current)) {
3579                         flush_signals(current);
3580                         goto skip;
3581                 }
3582                 ITERATE_MDDEV(mddev2,tmp) {
3583                         if (mddev2 == mddev)
3584                                 continue;
3585                         if (mddev2->curr_resync &&
3586                             match_mddev_units(mddev,mddev2)) {
3587                                 DEFINE_WAIT(wq);
3588                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3589                                         /* arbitrarily yield */
3590                                         mddev->curr_resync = 1;
3591                                         wake_up(&resync_wait);
3592                                 }
3593                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3594                                         /* no need to wait here, we can wait the next
3595                                          * time 'round when curr_resync == 2
3596                                          */
3597                                         continue;
3598                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3599                                 if (!signal_pending(current)
3600                                     && mddev2->curr_resync >= mddev->curr_resync) {
3601                                         printk(KERN_INFO "md: delaying resync of %s"
3602                                                " until %s has finished resync (they"
3603                                                " share one or more physical units)\n",
3604                                                mdname(mddev), mdname(mddev2));
3605                                         mddev_put(mddev2);
3606                                         schedule();
3607                                         finish_wait(&resync_wait, &wq);
3608                                         goto try_again;
3609                                 }
3610                                 finish_wait(&resync_wait, &wq);
3611                         }
3612                 }
3613         } while (mddev->curr_resync < 2);
3614
3615         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3616                 /* resync follows the size requested by the personality,
3617                  * which defaults to physical size, but can be virtual size
3618                  */
3619                 max_sectors = mddev->resync_max_sectors;
3620         else
3621                 /* recovery follows the physical size of devices */
3622                 max_sectors = mddev->size << 1;
3623
3624         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3625         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3626                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3627         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3628                "(but not more than %d KB/sec) for reconstruction.\n",
3629                sysctl_speed_limit_max);
3630
3631         is_mddev_idle(mddev); /* this also initializes IO event counters */
3632         /* we don't use the checkpoint if there's a bitmap */
3633         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3634                 j = mddev->recovery_cp;
3635         else
3636                 j = 0;
3637         io_sectors = 0;
3638         for (m = 0; m < SYNC_MARKS; m++) {
3639                 mark[m] = jiffies;
3640                 mark_cnt[m] = io_sectors;
3641         }
3642         last_mark = 0;
3643         mddev->resync_mark = mark[last_mark];
3644         mddev->resync_mark_cnt = mark_cnt[last_mark];
3645
3646         /*
3647          * Tune reconstruction:
3648          */
3649         window = 32*(PAGE_SIZE/512);
3650         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3651                 window/2,(unsigned long long) max_sectors/2);
3652
3653         atomic_set(&mddev->recovery_active, 0);
3654         init_waitqueue_head(&mddev->recovery_wait);
3655         last_check = 0;
3656
3657         if (j>2) {
3658                 printk(KERN_INFO
3659                         "md: resuming recovery of %s from checkpoint.\n",
3660                         mdname(mddev));
3661                 mddev->curr_resync = j;
3662         }
3663
3664         while (j < max_sectors) {
3665                 sector_t sectors;
3666
3667                 skipped = 0;
3668                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3669                                             currspeed < sysctl_speed_limit_min);
3670                 if (sectors == 0) {
3671                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3672                         goto out;
3673                 }
3674
3675                 if (!skipped) { /* actual IO requested */
3676                         io_sectors += sectors;
3677                         atomic_add(sectors, &mddev->recovery_active);
3678                 }
3679
3680                 j += sectors;
3681                 if (j>1) mddev->curr_resync = j;
3682
3683
3684                 if (last_check + window > io_sectors || j == max_sectors)
3685                         continue;
3686
3687                 last_check = io_sectors;
3688
3689                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3690                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3691                         break;
3692
3693         repeat:
3694                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3695                         /* step marks */
3696                         int next = (last_mark+1) % SYNC_MARKS;
3697
3698                         mddev->resync_mark = mark[next];
3699                         mddev->resync_mark_cnt = mark_cnt[next];
3700                         mark[next] = jiffies;
3701                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3702                         last_mark = next;
3703                 }
3704
3705
3706                 if (signal_pending(current)) {
3707                         /*
3708                          * got a signal, exit.
3709                          */
3710                         printk(KERN_INFO
3711                                 "md: md_do_sync() got signal ... exiting\n");
3712                         flush_signals(current);
3713                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3714                         goto out;
3715                 }
3716
3717                 /*
3718                  * this loop exits only if either when we are slower than
3719                  * the 'hard' speed limit, or the system was IO-idle for
3720                  * a jiffy.
3721                  * the system might be non-idle CPU-wise, but we only care
3722                  * about not overloading the IO subsystem. (things like an
3723                  * e2fsck being done on the RAID array should execute fast)
3724                  */
3725                 mddev->queue->unplug_fn(mddev->queue);
3726                 cond_resched();
3727
3728                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3729                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
3730
3731                 if (currspeed > sysctl_speed_limit_min) {
3732                         if ((currspeed > sysctl_speed_limit_max) ||
3733                                         !is_mddev_idle(mddev)) {
3734                                 msleep_interruptible(250);
3735                                 goto repeat;
3736                         }
3737                 }
3738         }
3739         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3740         /*
3741          * this also signals 'finished resyncing' to md_stop
3742          */
3743  out:
3744         mddev->queue->unplug_fn(mddev->queue);
3745
3746         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3747
3748         /* tell personality that we are finished */
3749         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3750
3751         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3752             mddev->curr_resync > 2 &&
3753             mddev->curr_resync >= mddev->recovery_cp) {
3754                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3755                         printk(KERN_INFO
3756                                 "md: checkpointing recovery of %s.\n",
3757                                 mdname(mddev));
3758                         mddev->recovery_cp = mddev->curr_resync;
3759                 } else
3760                         mddev->recovery_cp = MaxSector;
3761         }
3762
3763  skip:
3764         mddev->curr_resync = 0;
3765         wake_up(&resync_wait);
3766         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3767         md_wakeup_thread(mddev->thread);
3768 }
3769
3770
3771 /*
3772  * This routine is regularly called by all per-raid-array threads to
3773  * deal with generic issues like resync and super-block update.
3774  * Raid personalities that don't have a thread (linear/raid0) do not
3775  * need this as they never do any recovery or update the superblock.
3776  *
3777  * It does not do any resync itself, but rather "forks" off other threads
3778  * to do that as needed.
3779  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3780  * "->recovery" and create a thread at ->sync_thread.
3781  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3782  * and wakeups up this thread which will reap the thread and finish up.
3783  * This thread also removes any faulty devices (with nr_pending == 0).
3784  *
3785  * The overall approach is:
3786  *  1/ if the superblock needs updating, update it.
3787  *  2/ If a recovery thread is running, don't do anything else.
3788  *  3/ If recovery has finished, clean up, possibly marking spares active.
3789  *  4/ If there are any faulty devices, remove them.
3790  *  5/ If array is degraded, try to add spares devices
3791  *  6/ If array has spares or is not in-sync, start a resync thread.
3792  */
3793 void md_check_recovery(mddev_t *mddev)
3794 {
3795         mdk_rdev_t *rdev;
3796         struct list_head *rtmp;
3797
3798
3799         if (mddev->bitmap)
3800                 bitmap_daemon_work(mddev->bitmap);
3801
3802         if (mddev->ro)
3803                 return;
3804
3805         if (signal_pending(current)) {
3806                 if (mddev->pers->sync_request) {
3807                         printk(KERN_INFO "md: %s in immediate safe mode\n",
3808                                mdname(mddev));
3809                         mddev->safemode = 2;
3810                 }
3811                 flush_signals(current);
3812         }
3813
3814         if ( ! (
3815                 mddev->sb_dirty ||
3816                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3817                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3818                 (mddev->safemode == 1) ||
3819                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3820                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3821                 ))
3822                 return;
3823
3824         if (mddev_trylock(mddev)==0) {
3825                 int spares =0;
3826
3827                 spin_lock(&mddev->write_lock);
3828                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3829                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3830                         mddev->in_sync = 1;
3831                         mddev->sb_dirty = 1;
3832                 }
3833                 if (mddev->safemode == 1)
3834                         mddev->safemode = 0;
3835                 spin_unlock(&mddev->write_lock);
3836
3837                 if (mddev->sb_dirty)
3838                         md_update_sb(mddev);
3839
3840
3841                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3842                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3843                         /* resync/recovery still happening */
3844                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3845                         goto unlock;
3846                 }
3847                 if (mddev->sync_thread) {
3848                         /* resync has finished, collect result */
3849                         md_unregister_thread(mddev->sync_thread);
3850                         mddev->sync_thread = NULL;
3851                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3852                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3853                                 /* success...*/
3854                                 /* activate any spares */
3855                                 mddev->pers->spare_active(mddev);
3856                         }
3857                         md_update_sb(mddev);
3858
3859                         /* if array is no-longer degraded, then any saved_raid_disk
3860                          * information must be scrapped
3861                          */
3862                         if (!mddev->degraded)
3863                                 ITERATE_RDEV(mddev,rdev,rtmp)
3864                                         rdev->saved_raid_disk = -1;
3865
3866                         mddev->recovery = 0;
3867                         /* flag recovery needed just to double check */
3868                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3869                         goto unlock;
3870                 }
3871                 if (mddev->recovery)
3872                         /* probably just the RECOVERY_NEEDED flag */
3873                         mddev->recovery = 0;
3874
3875                 /* no recovery is running.
3876                  * remove any failed drives, then
3877                  * add spares if possible.
3878                  * Spare are also removed and re-added, to allow
3879                  * the personality to fail the re-add.
3880                  */
3881                 ITERATE_RDEV(mddev,rdev,rtmp)
3882                         if (rdev->raid_disk >= 0 &&
3883                             (rdev->faulty || ! rdev->in_sync) &&
3884                             atomic_read(&rdev->nr_pending)==0) {
3885                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3886                                         rdev->raid_disk = -1;
3887                         }
3888
3889                 if (mddev->degraded) {
3890                         ITERATE_RDEV(mddev,rdev,rtmp)
3891                                 if (rdev->raid_disk < 0
3892                                     && !rdev->faulty) {
3893                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3894                                                 spares++;
3895                                         else
3896                                                 break;
3897                                 }
3898                 }
3899
3900                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3901                         /* nothing we can do ... */
3902                         goto unlock;
3903                 }
3904                 if (mddev->pers->sync_request) {
3905                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3906                         if (!spares)
3907                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3908                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3909                                 /* We are adding a device or devices to an array
3910                                  * which has the bitmap stored on all devices.
3911                                  * So make sure all bitmap pages get written
3912                                  */
3913                                 bitmap_write_all(mddev->bitmap);
3914                         }
3915                         mddev->sync_thread = md_register_thread(md_do_sync,
3916                                                                 mddev,
3917                                                                 "%s_resync");
3918                         if (!mddev->sync_thread) {
3919                                 printk(KERN_ERR "%s: could not start resync"
3920                                         " thread...\n",
3921                                         mdname(mddev));
3922                                 /* leave the spares where they are, it shouldn't hurt */
3923                                 mddev->recovery = 0;
3924                         } else {
3925                                 md_wakeup_thread(mddev->sync_thread);
3926                         }
3927                 }
3928         unlock:
3929                 mddev_unlock(mddev);
3930         }
3931 }
3932
3933 static int md_notify_reboot(struct notifier_block *this,
3934                             unsigned long code, void *x)
3935 {
3936         struct list_head *tmp;
3937         mddev_t *mddev;
3938
3939         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3940
3941                 printk(KERN_INFO "md: stopping all md devices.\n");
3942
3943                 ITERATE_MDDEV(mddev,tmp)
3944                         if (mddev_trylock(mddev)==0)
3945                                 do_md_stop (mddev, 1);
3946                 /*
3947                  * certain more exotic SCSI devices are known to be
3948                  * volatile wrt too early system reboots. While the
3949                  * right place to handle this issue is the given
3950                  * driver, we do want to have a safe RAID driver ...
3951                  */
3952                 mdelay(1000*1);
3953         }
3954         return NOTIFY_DONE;
3955 }
3956
3957 static struct notifier_block md_notifier = {
3958         .notifier_call  = md_notify_reboot,
3959         .next           = NULL,
3960         .priority       = INT_MAX, /* before any real devices */
3961 };
3962
3963 static void md_geninit(void)
3964 {
3965         struct proc_dir_entry *p;
3966
3967         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3968
3969         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3970         if (p)
3971                 p->proc_fops = &md_seq_fops;
3972 }
3973
3974 static int __init md_init(void)
3975 {
3976         int minor;
3977
3978         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3979                         " MD_SB_DISKS=%d\n",
3980                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3981                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3982         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3983                         BITMAP_MINOR);
3984
3985         if (register_blkdev(MAJOR_NR, "md"))
3986                 return -1;
3987         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3988                 unregister_blkdev(MAJOR_NR, "md");
3989                 return -1;
3990         }
3991         devfs_mk_dir("md");
3992         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3993                                 md_probe, NULL, NULL);
3994         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3995                             md_probe, NULL, NULL);
3996
3997         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3998                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3999                                 S_IFBLK|S_IRUSR|S_IWUSR,
4000                                 "md/%d", minor);
4001
4002         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4003                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4004                               S_IFBLK|S_IRUSR|S_IWUSR,
4005                               "md/mdp%d", minor);
4006
4007
4008         register_reboot_notifier(&md_notifier);
4009         raid_table_header = register_sysctl_table(raid_root_table, 1);
4010
4011         md_geninit();
4012         return (0);
4013 }
4014
4015
4016 #ifndef MODULE
4017
4018 /*
4019  * Searches all registered partitions for autorun RAID arrays
4020  * at boot time.
4021  */
4022 static dev_t detected_devices[128];
4023 static int dev_cnt;
4024
4025 void md_autodetect_dev(dev_t dev)
4026 {
4027         if (dev_cnt >= 0 && dev_cnt < 127)
4028                 detected_devices[dev_cnt++] = dev;
4029 }
4030
4031
4032 static void autostart_arrays(int part)
4033 {
4034         mdk_rdev_t *rdev;
4035         int i;
4036
4037         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4038
4039         for (i = 0; i < dev_cnt; i++) {
4040                 dev_t dev = detected_devices[i];
4041
4042                 rdev = md_import_device(dev,0, 0);
4043                 if (IS_ERR(rdev))
4044                         continue;
4045
4046                 if (rdev->faulty) {
4047                         MD_BUG();
4048                         continue;
4049                 }
4050                 list_add(&rdev->same_set, &pending_raid_disks);
4051         }
4052         dev_cnt = 0;
4053
4054         autorun_devices(part);
4055 }
4056
4057 #endif
4058
4059 static __exit void md_exit(void)
4060 {
4061         mddev_t *mddev;
4062         struct list_head *tmp;
4063         int i;
4064         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4065         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4066         for (i=0; i < MAX_MD_DEVS; i++)
4067                 devfs_remove("md/%d", i);
4068         for (i=0; i < MAX_MD_DEVS; i++)
4069                 devfs_remove("md/d%d", i);
4070
4071         devfs_remove("md");
4072
4073         unregister_blkdev(MAJOR_NR,"md");
4074         unregister_blkdev(mdp_major, "mdp");
4075         unregister_reboot_notifier(&md_notifier);
4076         unregister_sysctl_table(raid_table_header);
4077         remove_proc_entry("mdstat", NULL);
4078         ITERATE_MDDEV(mddev,tmp) {
4079                 struct gendisk *disk = mddev->gendisk;
4080                 if (!disk)
4081                         continue;
4082                 export_array(mddev);
4083                 del_gendisk(disk);
4084                 put_disk(disk);
4085                 mddev->gendisk = NULL;
4086                 mddev_put(mddev);
4087         }
4088 }
4089
4090 module_init(md_init)
4091 module_exit(md_exit)
4092
4093 EXPORT_SYMBOL(register_md_personality);
4094 EXPORT_SYMBOL(unregister_md_personality);
4095 EXPORT_SYMBOL(md_error);
4096 EXPORT_SYMBOL(md_done_sync);
4097 EXPORT_SYMBOL(md_write_start);
4098 EXPORT_SYMBOL(md_write_end);
4099 EXPORT_SYMBOL(md_register_thread);
4100 EXPORT_SYMBOL(md_unregister_thread);
4101 EXPORT_SYMBOL(md_wakeup_thread);
4102 EXPORT_SYMBOL(md_print_devices);
4103 EXPORT_SYMBOL(md_check_recovery);
4104 MODULE_LICENSE("GPL");
4105 MODULE_ALIAS("md");
4106 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);