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