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