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