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 <[email protected]>
  11    - boot support for linear and striped mode by Harald Hoyer <[email protected]>
  12    - kerneld support by Boris Tobotras <[email protected]>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <[email protected]>
  15    - Devfs support by Richard Gooch <[email protected]>
  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 <[email protected]>.
  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 DEFINE_SPINLOCK(pers_lock);
  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 DEFINE_SPINLOCK(all_mddevs_lock);
 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 == 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+1 > max_dev)
 939                         max_dev = rdev2->desc_nr+1;
 940
 941         sb->max_dev = cpu_to_le32(max_dev);
 942         for (i=0; i<max_dev;i++)
 943                 sb->dev_roles[i] = 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         sb->sb_csum = calc_sb_1_csum(sb);
 957 }
 958
 959
 960 struct super_type super_types[] = {
 961         [0] = {
 962                 .name   = "0.90.0",
 963                 .owner  = THIS_MODULE,
 964                 .load_super     = super_90_load,
 965                 .validate_super = super_90_validate,
 966                 .sync_super     = super_90_sync,
 967         },
 968         [1] = {
 969                 .name   = "md-1",
 970                 .owner  = THIS_MODULE,
 971                 .load_super     = super_1_load,
 972                 .validate_super = super_1_validate,
 973                 .sync_super     = super_1_sync,
 974         },
 975 };
 976
 977 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
 978 {
 979         struct list_head *tmp;
 980         mdk_rdev_t *rdev;
 981
 982         ITERATE_RDEV(mddev,rdev,tmp)
 983                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
 984                         return rdev;
 985
 986         return NULL;
 987 }
 988
 989 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
 990 {
 991         struct list_head *tmp;
 992         mdk_rdev_t *rdev;
 993
 994         ITERATE_RDEV(mddev1,rdev,tmp)
 995                 if (match_dev_unit(mddev2, rdev))
 996                         return 1;
 997
 998         return 0;
 999 }
1000
1001 static LIST_HEAD(pending_raid_disks);
1002
1003 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1004 {
1005         mdk_rdev_t *same_pdev;
1006         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1007
1008         if (rdev->mddev) {
1009                 MD_BUG();
1010                 return -EINVAL;
1011         }
1012         same_pdev = match_dev_unit(mddev, rdev);
1013         if (same_pdev)
1014                 printk(KERN_WARNING
1015                         "%s: WARNING: %s appears to be on the same physical"
1016                         " disk as %s. True\n     protection against single-disk"
1017                         " failure might be compromised.\n",
1018                         mdname(mddev), bdevname(rdev->bdev,b),
1019                         bdevname(same_pdev->bdev,b2));
1020
1021         /* Verify rdev->desc_nr is unique.
1022          * If it is -1, assign a free number, else
1023          * check number is not in use
1024          */
1025         if (rdev->desc_nr < 0) {
1026                 int choice = 0;
1027                 if (mddev->pers) choice = mddev->raid_disks;
1028                 while (find_rdev_nr(mddev, choice))
1029                         choice++;
1030                 rdev->desc_nr = choice;
1031         } else {
1032                 if (find_rdev_nr(mddev, rdev->desc_nr))
1033                         return -EBUSY;
1034         }
1035
1036         list_add(&rdev->same_set, &mddev->disks);
1037         rdev->mddev = mddev;
1038         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1039         return 0;
1040 }
1041
1042 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1043 {
1044         char b[BDEVNAME_SIZE];
1045         if (!rdev->mddev) {
1046                 MD_BUG();
1047                 return;
1048         }
1049         list_del_init(&rdev->same_set);
1050         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1051         rdev->mddev = NULL;
1052 }
1053
1054 /*
1055  * prevent the device from being mounted, repartitioned or
1056  * otherwise reused by a RAID array (or any other kernel
1057  * subsystem), by bd_claiming the device.
1058  */
1059 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1060 {
1061         int err = 0;
1062         struct block_device *bdev;
1063         char b[BDEVNAME_SIZE];
1064
1065         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1066         if (IS_ERR(bdev)) {
1067                 printk(KERN_ERR "md: could not open %s.\n",
1068                         __bdevname(dev, b));
1069                 return PTR_ERR(bdev);
1070         }
1071         err = bd_claim(bdev, rdev);
1072         if (err) {
1073                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1074                         bdevname(bdev, b));
1075                 blkdev_put(bdev);
1076                 return err;
1077         }
1078         rdev->bdev = bdev;
1079         return err;
1080 }
1081
1082 static void unlock_rdev(mdk_rdev_t *rdev)
1083 {
1084         struct block_device *bdev = rdev->bdev;
1085         rdev->bdev = NULL;
1086         if (!bdev)
1087                 MD_BUG();
1088         bd_release(bdev);
1089         blkdev_put(bdev);
1090 }
1091
1092 void md_autodetect_dev(dev_t dev);
1093
1094 static void export_rdev(mdk_rdev_t * rdev)
1095 {
1096         char b[BDEVNAME_SIZE];
1097         printk(KERN_INFO "md: export_rdev(%s)\n",
1098                 bdevname(rdev->bdev,b));
1099         if (rdev->mddev)
1100                 MD_BUG();
1101         free_disk_sb(rdev);
1102         list_del_init(&rdev->same_set);
1103 #ifndef MODULE
1104         md_autodetect_dev(rdev->bdev->bd_dev);
1105 #endif
1106         unlock_rdev(rdev);
1107         kfree(rdev);
1108 }
1109
1110 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1111 {
1112         unbind_rdev_from_array(rdev);
1113         export_rdev(rdev);
1114 }
1115
1116 static void export_array(mddev_t *mddev)
1117 {
1118         struct list_head *tmp;
1119         mdk_rdev_t *rdev;
1120
1121         ITERATE_RDEV(mddev,rdev,tmp) {
1122                 if (!rdev->mddev) {
1123                         MD_BUG();
1124                         continue;
1125                 }
1126                 kick_rdev_from_array(rdev);
1127         }
1128         if (!list_empty(&mddev->disks))
1129                 MD_BUG();
1130         mddev->raid_disks = 0;
1131         mddev->major_version = 0;
1132 }
1133
1134 static void print_desc(mdp_disk_t *desc)
1135 {
1136         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1137                 desc->major,desc->minor,desc->raid_disk,desc->state);
1138 }
1139
1140 static void print_sb(mdp_super_t *sb)
1141 {
1142         int i;
1143
1144         printk(KERN_INFO
1145                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1146                 sb->major_version, sb->minor_version, sb->patch_version,
1147                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1148                 sb->ctime);
1149         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1150                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1151                 sb->md_minor, sb->layout, sb->chunk_size);
1152         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1153                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1154                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1155                 sb->failed_disks, sb->spare_disks,
1156                 sb->sb_csum, (unsigned long)sb->events_lo);
1157
1158         printk(KERN_INFO);
1159         for (i = 0; i < MD_SB_DISKS; i++) {
1160                 mdp_disk_t *desc;
1161
1162                 desc = sb->disks + i;
1163                 if (desc->number || desc->major || desc->minor ||
1164                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1165                         printk("     D %2d: ", i);
1166                         print_desc(desc);
1167                 }
1168         }
1169         printk(KERN_INFO "md:     THIS: ");
1170         print_desc(&sb->this_disk);
1171
1172 }
1173
1174 static void print_rdev(mdk_rdev_t *rdev)
1175 {
1176         char b[BDEVNAME_SIZE];
1177         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1178                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1179                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1180         if (rdev->sb_loaded) {
1181                 printk(KERN_INFO "md: rdev superblock:\n");
1182                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1183         } else
1184                 printk(KERN_INFO "md: no rdev superblock!\n");
1185 }
1186
1187 void md_print_devices(void)
1188 {
1189         struct list_head *tmp, *tmp2;
1190         mdk_rdev_t *rdev;
1191         mddev_t *mddev;
1192         char b[BDEVNAME_SIZE];
1193
1194         printk("\n");
1195         printk("md:     **********************************\n");
1196         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1197         printk("md:     **********************************\n");
1198         ITERATE_MDDEV(mddev,tmp) {
1199                 printk("%s: ", mdname(mddev));
1200
1201                 ITERATE_RDEV(mddev,rdev,tmp2)
1202                         printk("<%s>", bdevname(rdev->bdev,b));
1203                 printk("\n");
1204
1205                 ITERATE_RDEV(mddev,rdev,tmp2)
1206                         print_rdev(rdev);
1207         }
1208         printk("md:     **********************************\n");
1209         printk("\n");
1210 }
1211
1212
1213 static int write_disk_sb(mdk_rdev_t * rdev)
1214 {
1215         char b[BDEVNAME_SIZE];
1216         if (!rdev->sb_loaded) {
1217                 MD_BUG();
1218                 return 1;
1219         }
1220         if (rdev->faulty) {
1221                 MD_BUG();
1222                 return 1;
1223         }
1224
1225         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1226                 bdevname(rdev->bdev,b),
1227                (unsigned long long)rdev->sb_offset);
1228
1229         if (sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, WRITE))
1230                 return 0;
1231
1232         printk("md: write_disk_sb failed for device %s\n",
1233                 bdevname(rdev->bdev,b));
1234         return 1;
1235 }
1236
1237 static void sync_sbs(mddev_t * mddev)
1238 {
1239         mdk_rdev_t *rdev;
1240         struct list_head *tmp;
1241
1242         ITERATE_RDEV(mddev,rdev,tmp) {
1243                 super_types[mddev->major_version].
1244                         sync_super(mddev, rdev);
1245                 rdev->sb_loaded = 1;
1246         }
1247 }
1248
1249 static void md_update_sb(mddev_t * mddev)
1250 {
1251         int err, count = 100;
1252         struct list_head *tmp;
1253         mdk_rdev_t *rdev;
1254
1255         mddev->sb_dirty = 0;
1256 repeat:
1257         mddev->utime = get_seconds();
1258         mddev->events ++;
1259
1260         if (!mddev->events) {
1261                 /*
1262                  * oops, this 64-bit counter should never wrap.
1263                  * Either we are in around ~1 trillion A.C., assuming
1264                  * 1 reboot per second, or we have a bug:
1265                  */
1266                 MD_BUG();
1267                 mddev->events --;
1268         }
1269         sync_sbs(mddev);
1270
1271         /*
1272          * do not write anything to disk if using
1273          * nonpersistent superblocks
1274          */
1275         if (!mddev->persistent)
1276                 return;
1277
1278         dprintk(KERN_INFO
1279                 "md: updating %s RAID superblock on device (in sync %d)\n",
1280                 mdname(mddev),mddev->in_sync);
1281
1282         err = 0;
1283         ITERATE_RDEV(mddev,rdev,tmp) {
1284                 char b[BDEVNAME_SIZE];
1285                 dprintk(KERN_INFO "md: ");
1286                 if (rdev->faulty)
1287                         dprintk("(skipping faulty ");
1288
1289                 dprintk("%s ", bdevname(rdev->bdev,b));
1290                 if (!rdev->faulty) {
1291                         err += write_disk_sb(rdev);
1292                 } else
1293                         dprintk(")\n");
1294                 if (!err && mddev->level == LEVEL_MULTIPATH)
1295                         /* only need to write one superblock... */
1296                         break;
1297         }
1298         if (err) {
1299                 if (--count) {
1300                         printk(KERN_ERR "md: errors occurred during superblock"
1301                                 " update, repeating\n");
1302                         goto repeat;
1303                 }
1304                 printk(KERN_ERR \
1305                         "md: excessive errors occurred during superblock update, exiting\n");
1306         }
1307 }
1308
1309 /*
1310  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1311  *
1312  * mark the device faulty if:
1313  *
1314  *   - the device is nonexistent (zero size)
1315  *   - the device has no valid superblock
1316  *
1317  * a faulty rdev _never_ has rdev->sb set.
1318  */
1319 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1320 {
1321         char b[BDEVNAME_SIZE];
1322         int err;
1323         mdk_rdev_t *rdev;
1324         sector_t size;
1325
1326         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1327         if (!rdev) {
1328                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1329                 return ERR_PTR(-ENOMEM);
1330         }
1331         memset(rdev, 0, sizeof(*rdev));
1332
1333         if ((err = alloc_disk_sb(rdev)))
1334                 goto abort_free;
1335
1336         err = lock_rdev(rdev, newdev);
1337         if (err)
1338                 goto abort_free;
1339
1340         rdev->desc_nr = -1;
1341         rdev->faulty = 0;
1342         rdev->in_sync = 0;
1343         rdev->data_offset = 0;
1344         atomic_set(&rdev->nr_pending, 0);
1345
1346         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1347         if (!size) {
1348                 printk(KERN_WARNING
1349                         "md: %s has zero or unknown size, marking faulty!\n",
1350                         bdevname(rdev->bdev,b));
1351                 err = -EINVAL;
1352                 goto abort_free;
1353         }
1354
1355         if (super_format >= 0) {
1356                 err = super_types[super_format].
1357                         load_super(rdev, NULL, super_minor);
1358                 if (err == -EINVAL) {
1359                         printk(KERN_WARNING
1360                                 "md: %s has invalid sb, not importing!\n",
1361                                 bdevname(rdev->bdev,b));
1362                         goto abort_free;
1363                 }
1364                 if (err < 0) {
1365                         printk(KERN_WARNING
1366                                 "md: could not read %s's sb, not importing!\n",
1367                                 bdevname(rdev->bdev,b));
1368                         goto abort_free;
1369                 }
1370         }
1371         INIT_LIST_HEAD(&rdev->same_set);
1372
1373         return rdev;
1374
1375 abort_free:
1376         if (rdev->sb_page) {
1377                 if (rdev->bdev)
1378                         unlock_rdev(rdev);
1379                 free_disk_sb(rdev);
1380         }
1381         kfree(rdev);
1382         return ERR_PTR(err);
1383 }
1384
1385 /*
1386  * Check a full RAID array for plausibility
1387  */
1388
1389
1390 static int analyze_sbs(mddev_t * mddev)
1391 {
1392         int i;
1393         struct list_head *tmp;
1394         mdk_rdev_t *rdev, *freshest;
1395         char b[BDEVNAME_SIZE];
1396
1397         freshest = NULL;
1398         ITERATE_RDEV(mddev,rdev,tmp)
1399                 switch (super_types[mddev->major_version].
1400                         load_super(rdev, freshest, mddev->minor_version)) {
1401                 case 1:
1402                         freshest = rdev;
1403                         break;
1404                 case 0:
1405                         break;
1406                 default:
1407                         printk( KERN_ERR \
1408                                 "md: fatal superblock inconsistency in %s"
1409                                 " -- removing from array\n",
1410                                 bdevname(rdev->bdev,b));
1411                         kick_rdev_from_array(rdev);
1412                 }
1413
1414
1415         super_types[mddev->major_version].
1416                 validate_super(mddev, freshest);
1417
1418         i = 0;
1419         ITERATE_RDEV(mddev,rdev,tmp) {
1420                 if (rdev != freshest)
1421                         if (super_types[mddev->major_version].
1422                             validate_super(mddev, rdev)) {
1423                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1424                                         " from array!\n",
1425                                         bdevname(rdev->bdev,b));
1426                                 kick_rdev_from_array(rdev);
1427                                 continue;
1428                         }
1429                 if (mddev->level == LEVEL_MULTIPATH) {
1430                         rdev->desc_nr = i++;
1431                         rdev->raid_disk = rdev->desc_nr;
1432                         rdev->in_sync = 1;
1433                 }
1434         }
1435
1436
1437
1438         if (mddev->recovery_cp != MaxSector &&
1439             mddev->level >= 1)
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                 sprintf(disk->devfs_name, "md/d%d", unit);
1478         } else {
1479                 sprintf(disk->disk_name, "md%d", unit);
1480                 sprintf(disk->devfs_name, "md/%d", unit);
1481         }
1482         disk->fops = &md_fops;
1483         disk->private_data = mddev;
1484         disk->queue = mddev->queue;
1485         add_disk(disk);
1486         mddev->gendisk = disk;
1487         up(&disks_sem);
1488         return NULL;
1489 }
1490
1491 void md_wakeup_thread(mdk_thread_t *thread);
1492
1493 static void md_safemode_timeout(unsigned long data)
1494 {
1495         mddev_t *mddev = (mddev_t *) data;
1496
1497         mddev->safemode = 1;
1498         md_wakeup_thread(mddev->thread);
1499 }
1500
1501
1502 static int do_md_run(mddev_t * mddev)
1503 {
1504         int pnum, err;
1505         int chunk_size;
1506         struct list_head *tmp;
1507         mdk_rdev_t *rdev;
1508         struct gendisk *disk;
1509         char b[BDEVNAME_SIZE];
1510
1511         if (list_empty(&mddev->disks)) {
1512                 MD_BUG();
1513                 return -EINVAL;
1514         }
1515
1516         if (mddev->pers)
1517                 return -EBUSY;
1518
1519         /*
1520          * Analyze all RAID superblock(s)
1521          */
1522         if (!mddev->raid_disks && analyze_sbs(mddev)) {
1523                 MD_BUG();
1524                 return -EINVAL;
1525         }
1526
1527         chunk_size = mddev->chunk_size;
1528         pnum = level_to_pers(mddev->level);
1529
1530         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1531                 if (!chunk_size) {
1532                         /*
1533                          * 'default chunksize' in the old md code used to
1534                          * be PAGE_SIZE, baaad.
1535                          * we abort here to be on the safe side. We don't
1536                          * want to continue the bad practice.
1537                          */
1538                         printk(KERN_ERR
1539                                 "no chunksize specified, see 'man raidtab'\n");
1540                         return -EINVAL;
1541                 }
1542                 if (chunk_size > MAX_CHUNK_SIZE) {
1543                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1544                                 chunk_size, MAX_CHUNK_SIZE);
1545                         return -EINVAL;
1546                 }
1547                 /*
1548                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1549                  */
1550                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1551                         MD_BUG();
1552                         return -EINVAL;
1553                 }
1554                 if (chunk_size < PAGE_SIZE) {
1555                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1556                                 chunk_size, PAGE_SIZE);
1557                         return -EINVAL;
1558                 }
1559
1560                 /* devices must have minimum size of one chunk */
1561                 ITERATE_RDEV(mddev,rdev,tmp) {
1562                         if (rdev->faulty)
1563                                 continue;
1564                         if (rdev->size < chunk_size / 1024) {
1565                                 printk(KERN_WARNING
1566                                         "md: Dev %s smaller than chunk_size:"
1567                                         " %lluk < %dk\n",
1568                                         bdevname(rdev->bdev,b),
1569                                         (unsigned long long)rdev->size,
1570                                         chunk_size / 1024);
1571                                 return -EINVAL;
1572                         }
1573                 }
1574         }
1575
1576         if (pnum >= MAX_PERSONALITY) {
1577                 MD_BUG();
1578                 return -EINVAL;
1579         }
1580
1581 #ifdef CONFIG_KMOD
1582         if (!pers[pnum])
1583         {
1584                 request_module("md-personality-%d", pnum);
1585         }
1586 #endif
1587
1588         /*
1589          * Drop all container device buffers, from now on
1590          * the only valid external interface is through the md
1591          * device.
1592          * Also find largest hardsector size
1593          */
1594         ITERATE_RDEV(mddev,rdev,tmp) {
1595                 if (rdev->faulty)
1596                         continue;
1597                 sync_blockdev(rdev->bdev);
1598                 invalidate_bdev(rdev->bdev, 0);
1599         }
1600
1601         md_probe(mddev->unit, NULL, NULL);
1602         disk = mddev->gendisk;
1603         if (!disk)
1604                 return -ENOMEM;
1605
1606         spin_lock(&pers_lock);
1607         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1608                 spin_unlock(&pers_lock);
1609                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1610                        pnum);
1611                 return -EINVAL;
1612         }
1613
1614         mddev->pers = pers[pnum];
1615         spin_unlock(&pers_lock);
1616
1617         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1618
1619         err = mddev->pers->run(mddev);
1620         if (err) {
1621                 printk(KERN_ERR "md: pers->run() failed ...\n");
1622                 module_put(mddev->pers->owner);
1623                 mddev->pers = NULL;
1624                 return -EINVAL;
1625         }
1626         atomic_set(&mddev->writes_pending,0);
1627         mddev->safemode = 0;
1628         mddev->safemode_timer.function = md_safemode_timeout;
1629         mddev->safemode_timer.data = (unsigned long) mddev;
1630         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1631         mddev->in_sync = 1;
1632
1633         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1634
1635         if (mddev->sb_dirty)
1636                 md_update_sb(mddev);
1637
1638         set_capacity(disk, mddev->array_size<<1);
1639
1640         /* If we call blk_queue_make_request here, it will
1641          * re-initialise max_sectors etc which may have been
1642          * refined inside -> run.  So just set the bits we need to set.
1643          * Most initialisation happended when we called
1644          * blk_queue_make_request(..., md_fail_request)
1645          * earlier.
1646          */
1647         mddev->queue->queuedata = mddev;
1648         mddev->queue->make_request_fn = mddev->pers->make_request;
1649
1650         mddev->changed = 1;
1651         return 0;
1652 }
1653
1654 static int restart_array(mddev_t *mddev)
1655 {
1656         struct gendisk *disk = mddev->gendisk;
1657         int err;
1658
1659         /*
1660          * Complain if it has no devices
1661          */
1662         err = -ENXIO;
1663         if (list_empty(&mddev->disks))
1664                 goto out;
1665
1666         if (mddev->pers) {
1667                 err = -EBUSY;
1668                 if (!mddev->ro)
1669                         goto out;
1670
1671                 mddev->safemode = 0;
1672                 mddev->ro = 0;
1673                 set_disk_ro(disk, 0);
1674
1675                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1676                         mdname(mddev));
1677                 /*
1678                  * Kick recovery or resync if necessary
1679                  */
1680                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1681                 md_wakeup_thread(mddev->thread);
1682                 err = 0;
1683         } else {
1684                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1685                         mdname(mddev));
1686                 err = -EINVAL;
1687         }
1688
1689 out:
1690         return err;
1691 }
1692
1693 static int do_md_stop(mddev_t * mddev, int ro)
1694 {
1695         int err = 0;
1696         struct gendisk *disk = mddev->gendisk;
1697
1698         if (mddev->pers) {
1699                 if (atomic_read(&mddev->active)>2) {
1700                         printk("md: %s still in use.\n",mdname(mddev));
1701                         return -EBUSY;
1702                 }
1703
1704                 if (mddev->sync_thread) {
1705                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1706                         md_unregister_thread(mddev->sync_thread);
1707                         mddev->sync_thread = NULL;
1708                 }
1709
1710                 del_timer_sync(&mddev->safemode_timer);
1711
1712                 invalidate_partition(disk, 0);
1713
1714                 if (ro) {
1715                         err  = -ENXIO;
1716                         if (mddev->ro)
1717                                 goto out;
1718                         mddev->ro = 1;
1719                 } else {
1720                         if (mddev->ro)
1721                                 set_disk_ro(disk, 0);
1722                         blk_queue_make_request(mddev->queue, md_fail_request);
1723                         mddev->pers->stop(mddev);
1724                         module_put(mddev->pers->owner);
1725                         mddev->pers = NULL;
1726                         if (mddev->ro)
1727                                 mddev->ro = 0;
1728                 }
1729                 if (!mddev->in_sync) {
1730                         /* mark array as shutdown cleanly */
1731                         mddev->in_sync = 1;
1732                         md_update_sb(mddev);
1733                 }
1734                 if (ro)
1735                         set_disk_ro(disk, 1);
1736         }
1737         /*
1738          * Free resources if final stop
1739          */
1740         if (!ro) {
1741                 struct gendisk *disk;
1742                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1743
1744                 export_array(mddev);
1745
1746                 mddev->array_size = 0;
1747                 disk = mddev->gendisk;
1748                 if (disk)
1749                         set_capacity(disk, 0);
1750                 mddev->changed = 1;
1751         } else
1752                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1753                         mdname(mddev));
1754         err = 0;
1755 out:
1756         return err;
1757 }
1758
1759 static void autorun_array(mddev_t *mddev)
1760 {
1761         mdk_rdev_t *rdev;
1762         struct list_head *tmp;
1763         int err;
1764
1765         if (list_empty(&mddev->disks)) {
1766                 MD_BUG();
1767                 return;
1768         }
1769
1770         printk(KERN_INFO "md: running: ");
1771
1772         ITERATE_RDEV(mddev,rdev,tmp) {
1773                 char b[BDEVNAME_SIZE];
1774                 printk("<%s>", bdevname(rdev->bdev,b));
1775         }
1776         printk("\n");
1777
1778         err = do_md_run (mddev);
1779         if (err) {
1780                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1781                 do_md_stop (mddev, 0);
1782         }
1783 }
1784
1785 /*
1786  * lets try to run arrays based on all disks that have arrived
1787  * until now. (those are in pending_raid_disks)
1788  *
1789  * the method: pick the first pending disk, collect all disks with
1790  * the same UUID, remove all from the pending list and put them into
1791  * the 'same_array' list. Then order this list based on superblock
1792  * update time (freshest comes first), kick out 'old' disks and
1793  * compare superblocks. If everything's fine then run it.
1794  *
1795  * If "unit" is allocated, then bump its reference count
1796  */
1797 static void autorun_devices(int part)
1798 {
1799         struct list_head candidates;
1800         struct list_head *tmp;
1801         mdk_rdev_t *rdev0, *rdev;
1802         mddev_t *mddev;
1803         char b[BDEVNAME_SIZE];
1804
1805         printk(KERN_INFO "md: autorun ...\n");
1806         while (!list_empty(&pending_raid_disks)) {
1807                 dev_t dev;
1808                 rdev0 = list_entry(pending_raid_disks.next,
1809                                          mdk_rdev_t, same_set);
1810
1811                 printk(KERN_INFO "md: considering %s ...\n",
1812                         bdevname(rdev0->bdev,b));
1813                 INIT_LIST_HEAD(&candidates);
1814                 ITERATE_RDEV_PENDING(rdev,tmp)
1815                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1816                                 printk(KERN_INFO "md:  adding %s ...\n",
1817                                         bdevname(rdev->bdev,b));
1818                                 list_move(&rdev->same_set, &candidates);
1819                         }
1820                 /*
1821                  * now we have a set of devices, with all of them having
1822                  * mostly sane superblocks. It's time to allocate the
1823                  * mddev.
1824                  */
1825                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1826                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1827                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1828                         break;
1829                 }
1830                 if (part)
1831                         dev = MKDEV(mdp_major,
1832                                     rdev0->preferred_minor << MdpMinorShift);
1833                 else
1834                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1835
1836                 md_probe(dev, NULL, NULL);
1837                 mddev = mddev_find(dev);
1838                 if (!mddev) {
1839                         printk(KERN_ERR
1840                                 "md: cannot allocate memory for md drive.\n");
1841                         break;
1842                 }
1843                 if (mddev_lock(mddev))
1844                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1845                                mdname(mddev));
1846                 else if (mddev->raid_disks || mddev->major_version
1847                          || !list_empty(&mddev->disks)) {
1848                         printk(KERN_WARNING
1849                                 "md: %s already running, cannot run %s\n",
1850                                 mdname(mddev), bdevname(rdev0->bdev,b));
1851                         mddev_unlock(mddev);
1852                 } else {
1853                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1854                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1855                                 list_del_init(&rdev->same_set);
1856                                 if (bind_rdev_to_array(rdev, mddev))
1857                                         export_rdev(rdev);
1858                         }
1859                         autorun_array(mddev);
1860                         mddev_unlock(mddev);
1861                 }
1862                 /* on success, candidates will be empty, on error
1863                  * it won't...
1864                  */
1865                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1866                         export_rdev(rdev);
1867                 mddev_put(mddev);
1868         }
1869         printk(KERN_INFO "md: ... autorun DONE.\n");
1870 }
1871
1872 /*
1873  * import RAID devices based on one partition
1874  * if possible, the array gets run as well.
1875  */
1876
1877 static int autostart_array(dev_t startdev)
1878 {
1879         char b[BDEVNAME_SIZE];
1880         int err = -EINVAL, i;
1881         mdp_super_t *sb = NULL;
1882         mdk_rdev_t *start_rdev = NULL, *rdev;
1883
1884         start_rdev = md_import_device(startdev, 0, 0);
1885         if (IS_ERR(start_rdev))
1886                 return err;
1887
1888
1889         /* NOTE: this can only work for 0.90.0 superblocks */
1890         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1891         if (sb->major_version != 0 ||
1892             sb->minor_version != 90 ) {
1893                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1894                 export_rdev(start_rdev);
1895                 return err;
1896         }
1897
1898         if (start_rdev->faulty) {
1899                 printk(KERN_WARNING
1900                         "md: can not autostart based on faulty %s!\n",
1901                         bdevname(start_rdev->bdev,b));
1902                 export_rdev(start_rdev);
1903                 return err;
1904         }
1905         list_add(&start_rdev->same_set, &pending_raid_disks);
1906
1907         for (i = 0; i < MD_SB_DISKS; i++) {
1908                 mdp_disk_t *desc = sb->disks + i;
1909                 dev_t dev = MKDEV(desc->major, desc->minor);
1910
1911                 if (!dev)
1912                         continue;
1913                 if (dev == startdev)
1914                         continue;
1915                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
1916                         continue;
1917                 rdev = md_import_device(dev, 0, 0);
1918                 if (IS_ERR(rdev))
1919                         continue;
1920
1921                 list_add(&rdev->same_set, &pending_raid_disks);
1922         }
1923
1924         /*
1925          * possibly return codes
1926          */
1927         autorun_devices(0);
1928         return 0;
1929
1930 }
1931
1932
1933 static int get_version(void __user * arg)
1934 {
1935         mdu_version_t ver;
1936
1937         ver.major = MD_MAJOR_VERSION;
1938         ver.minor = MD_MINOR_VERSION;
1939         ver.patchlevel = MD_PATCHLEVEL_VERSION;
1940
1941         if (copy_to_user(arg, &ver, sizeof(ver)))
1942                 return -EFAULT;
1943
1944         return 0;
1945 }
1946
1947 static int get_array_info(mddev_t * mddev, void __user * arg)
1948 {
1949         mdu_array_info_t info;
1950         int nr,working,active,failed,spare;
1951         mdk_rdev_t *rdev;
1952         struct list_head *tmp;
1953
1954         nr=working=active=failed=spare=0;
1955         ITERATE_RDEV(mddev,rdev,tmp) {
1956                 nr++;
1957                 if (rdev->faulty)
1958                         failed++;
1959                 else {
1960                         working++;
1961                         if (rdev->in_sync)
1962                                 active++;
1963                         else
1964                                 spare++;
1965                 }
1966         }
1967
1968         info.major_version = mddev->major_version;
1969         info.minor_version = mddev->minor_version;
1970         info.patch_version = MD_PATCHLEVEL_VERSION;
1971         info.ctime         = mddev->ctime;
1972         info.level         = mddev->level;
1973         info.size          = mddev->size;
1974         info.nr_disks      = nr;
1975         info.raid_disks    = mddev->raid_disks;
1976         info.md_minor      = mddev->md_minor;
1977         info.not_persistent= !mddev->persistent;
1978
1979         info.utime         = mddev->utime;
1980         info.state         = 0;
1981         if (mddev->in_sync)
1982                 info.state = (1<<MD_SB_CLEAN);
1983         info.active_disks  = active;
1984         info.working_disks = working;
1985         info.failed_disks  = failed;
1986         info.spare_disks   = spare;
1987
1988         info.layout        = mddev->layout;
1989         info.chunk_size    = mddev->chunk_size;
1990
1991         if (copy_to_user(arg, &info, sizeof(info)))
1992                 return -EFAULT;
1993
1994         return 0;
1995 }
1996
1997 static int get_disk_info(mddev_t * mddev, void __user * arg)
1998 {
1999         mdu_disk_info_t info;
2000         unsigned int nr;
2001         mdk_rdev_t *rdev;
2002
2003         if (copy_from_user(&info, arg, sizeof(info)))
2004                 return -EFAULT;
2005
2006         nr = info.number;
2007
2008         rdev = find_rdev_nr(mddev, nr);
2009         if (rdev) {
2010                 info.major = MAJOR(rdev->bdev->bd_dev);
2011                 info.minor = MINOR(rdev->bdev->bd_dev);
2012                 info.raid_disk = rdev->raid_disk;
2013                 info.state = 0;
2014                 if (rdev->faulty)
2015                         info.state |= (1<<MD_DISK_FAULTY);
2016                 else if (rdev->in_sync) {
2017                         info.state |= (1<<MD_DISK_ACTIVE);
2018                         info.state |= (1<<MD_DISK_SYNC);
2019                 }
2020         } else {
2021                 info.major = info.minor = 0;
2022                 info.raid_disk = -1;
2023                 info.state = (1<<MD_DISK_REMOVED);
2024         }
2025
2026         if (copy_to_user(arg, &info, sizeof(info)))
2027                 return -EFAULT;
2028
2029         return 0;
2030 }
2031
2032 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2033 {
2034         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2035         mdk_rdev_t *rdev;
2036         dev_t dev = MKDEV(info->major,info->minor);
2037
2038         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2039                 return -EOVERFLOW;
2040
2041         if (!mddev->raid_disks) {
2042                 int err;
2043                 /* expecting a device which has a superblock */
2044                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2045                 if (IS_ERR(rdev)) {
2046                         printk(KERN_WARNING
2047                                 "md: md_import_device returned %ld\n",
2048                                 PTR_ERR(rdev));
2049                         return PTR_ERR(rdev);
2050                 }
2051                 if (!list_empty(&mddev->disks)) {
2052                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2053                                                         mdk_rdev_t, same_set);
2054                         int err = super_types[mddev->major_version]
2055                                 .load_super(rdev, rdev0, mddev->minor_version);
2056                         if (err < 0) {
2057                                 printk(KERN_WARNING
2058                                         "md: %s has different UUID to %s\n",
2059                                         bdevname(rdev->bdev,b),
2060                                         bdevname(rdev0->bdev,b2));
2061                                 export_rdev(rdev);
2062                                 return -EINVAL;
2063                         }
2064                 }
2065                 err = bind_rdev_to_array(rdev, mddev);
2066                 if (err)
2067                         export_rdev(rdev);
2068                 return err;
2069         }
2070
2071         /*
2072          * add_new_disk can be used once the array is assembled
2073          * to add "hot spares".  They must already have a superblock
2074          * written
2075          */
2076         if (mddev->pers) {
2077                 int err;
2078                 if (!mddev->pers->hot_add_disk) {
2079                         printk(KERN_WARNING
2080                                 "%s: personality does not support diskops!\n",
2081                                mdname(mddev));
2082                         return -EINVAL;
2083                 }
2084                 rdev = md_import_device(dev, mddev->major_version,
2085                                         mddev->minor_version);
2086                 if (IS_ERR(rdev)) {
2087                         printk(KERN_WARNING
2088                                 "md: md_import_device returned %ld\n",
2089                                 PTR_ERR(rdev));
2090                         return PTR_ERR(rdev);
2091                 }
2092                 rdev->in_sync = 0; /* just to be sure */
2093                 rdev->raid_disk = -1;
2094                 err = bind_rdev_to_array(rdev, mddev);
2095                 if (err)
2096                         export_rdev(rdev);
2097                 if (mddev->thread)
2098                         md_wakeup_thread(mddev->thread);
2099                 return err;
2100         }
2101
2102         /* otherwise, add_new_disk is only allowed
2103          * for major_version==0 superblocks
2104          */
2105         if (mddev->major_version != 0) {
2106                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2107                        mdname(mddev));
2108                 return -EINVAL;
2109         }
2110
2111         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2112                 int err;
2113                 rdev = md_import_device (dev, -1, 0);
2114                 if (IS_ERR(rdev)) {
2115                         printk(KERN_WARNING
2116                                 "md: error, md_import_device() returned %ld\n",
2117                                 PTR_ERR(rdev));
2118                         return PTR_ERR(rdev);
2119                 }
2120                 rdev->desc_nr = info->number;
2121                 if (info->raid_disk < mddev->raid_disks)
2122                         rdev->raid_disk = info->raid_disk;
2123                 else
2124                         rdev->raid_disk = -1;
2125
2126                 rdev->faulty = 0;
2127                 if (rdev->raid_disk < mddev->raid_disks)
2128                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2129                 else
2130                         rdev->in_sync = 0;
2131
2132                 err = bind_rdev_to_array(rdev, mddev);
2133                 if (err) {
2134                         export_rdev(rdev);
2135                         return err;
2136                 }
2137
2138                 if (!mddev->persistent) {
2139                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2140                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2141                 } else
2142                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2143                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2144
2145                 if (!mddev->size || (mddev->size > rdev->size))
2146                         mddev->size = rdev->size;
2147         }
2148
2149         return 0;
2150 }
2151
2152 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2153 {
2154         char b[BDEVNAME_SIZE];
2155         mdk_rdev_t *rdev;
2156
2157         if (!mddev->pers)
2158                 return -ENODEV;
2159
2160         rdev = find_rdev(mddev, dev);
2161         if (!rdev)
2162                 return -ENXIO;
2163
2164         if (rdev->raid_disk >= 0)
2165                 goto busy;
2166
2167         kick_rdev_from_array(rdev);
2168         md_update_sb(mddev);
2169
2170         return 0;
2171 busy:
2172         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2173                 bdevname(rdev->bdev,b), mdname(mddev));
2174         return -EBUSY;
2175 }
2176
2177 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2178 {
2179         char b[BDEVNAME_SIZE];
2180         int err;
2181         unsigned int size;
2182         mdk_rdev_t *rdev;
2183
2184         if (!mddev->pers)
2185                 return -ENODEV;
2186
2187         if (mddev->major_version != 0) {
2188                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2189                         " version-0 superblocks.\n",
2190                         mdname(mddev));
2191                 return -EINVAL;
2192         }
2193         if (!mddev->pers->hot_add_disk) {
2194                 printk(KERN_WARNING
2195                         "%s: personality does not support diskops!\n",
2196                         mdname(mddev));
2197                 return -EINVAL;
2198         }
2199
2200         rdev = md_import_device (dev, -1, 0);
2201         if (IS_ERR(rdev)) {
2202                 printk(KERN_WARNING
2203                         "md: error, md_import_device() returned %ld\n",
2204                         PTR_ERR(rdev));
2205                 return -EINVAL;
2206         }
2207
2208         if (mddev->persistent)
2209                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2210         else
2211                 rdev->sb_offset =
2212                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2213
2214         size = calc_dev_size(rdev, mddev->chunk_size);
2215         rdev->size = size;
2216
2217         if (size < mddev->size) {
2218                 printk(KERN_WARNING
2219                         "%s: disk size %llu blocks < array size %llu\n",
2220                         mdname(mddev), (unsigned long long)size,
2221                         (unsigned long long)mddev->size);
2222                 err = -ENOSPC;
2223                 goto abort_export;
2224         }
2225
2226         if (rdev->faulty) {
2227                 printk(KERN_WARNING
2228                         "md: can not hot-add faulty %s disk to %s!\n",
2229                         bdevname(rdev->bdev,b), mdname(mddev));
2230                 err = -EINVAL;
2231                 goto abort_export;
2232         }
2233         rdev->in_sync = 0;
2234         rdev->desc_nr = -1;
2235         bind_rdev_to_array(rdev, mddev);
2236
2237         /*
2238          * The rest should better be atomic, we can have disk failures
2239          * noticed in interrupt contexts ...
2240          */
2241
2242         if (rdev->desc_nr == mddev->max_disks) {
2243                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2244                         mdname(mddev));
2245                 err = -EBUSY;
2246                 goto abort_unbind_export;
2247         }
2248
2249         rdev->raid_disk = -1;
2250
2251         md_update_sb(mddev);
2252
2253         /*
2254          * Kick recovery, maybe this spare has to be added to the
2255          * array immediately.
2256          */
2257         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2258         md_wakeup_thread(mddev->thread);
2259
2260         return 0;
2261
2262 abort_unbind_export:
2263         unbind_rdev_from_array(rdev);
2264
2265 abort_export:
2266         export_rdev(rdev);
2267         return err;
2268 }
2269
2270 /*
2271  * set_array_info is used two different ways
2272  * The original usage is when creating a new array.
2273  * In this usage, raid_disks is > 0 and it together with
2274  *  level, size, not_persistent,layout,chunksize determine the
2275  *  shape of the array.
2276  *  This will always create an array with a type-0.90.0 superblock.
2277  * The newer usage is when assembling an array.
2278  *  In this case raid_disks will be 0, and the major_version field is
2279  *  use to determine which style super-blocks are to be found on the devices.
2280  *  The minor and patch _version numbers are also kept incase the
2281  *  super_block handler wishes to interpret them.
2282  */
2283 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2284 {
2285
2286         if (info->raid_disks == 0) {
2287                 /* just setting version number for superblock loading */
2288                 if (info->major_version < 0 ||
2289                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2290                     super_types[info->major_version].name == NULL) {
2291                         /* maybe try to auto-load a module? */
2292                         printk(KERN_INFO
2293                                 "md: superblock version %d not known\n",
2294                                 info->major_version);
2295                         return -EINVAL;
2296                 }
2297                 mddev->major_version = info->major_version;
2298                 mddev->minor_version = info->minor_version;
2299                 mddev->patch_version = info->patch_version;
2300                 return 0;
2301         }
2302         mddev->major_version = MD_MAJOR_VERSION;
2303         mddev->minor_version = MD_MINOR_VERSION;
2304         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2305         mddev->ctime         = get_seconds();
2306
2307         mddev->level         = info->level;
2308         mddev->size          = info->size;
2309         mddev->raid_disks    = info->raid_disks;
2310         /* don't set md_minor, it is determined by which /dev/md* was
2311          * openned
2312          */
2313         if (info->state & (1<<MD_SB_CLEAN))
2314                 mddev->recovery_cp = MaxSector;
2315         else
2316                 mddev->recovery_cp = 0;
2317         mddev->persistent    = ! info->not_persistent;
2318
2319         mddev->layout        = info->layout;
2320         mddev->chunk_size    = info->chunk_size;
2321
2322         mddev->max_disks     = MD_SB_DISKS;
2323
2324         mddev->sb_dirty      = 1;
2325
2326         /*
2327          * Generate a 128 bit UUID
2328          */
2329         get_random_bytes(mddev->uuid, 16);
2330
2331         return 0;
2332 }
2333
2334 /*
2335  * update_array_info is used to change the configuration of an
2336  * on-line array.
2337  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2338  * fields in the info are checked against the array.
2339  * Any differences that cannot be handled will cause an error.
2340  * Normally, only one change can be managed at a time.
2341  */
2342 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2343 {
2344         int rv = 0;
2345         int cnt = 0;
2346
2347         if (mddev->major_version != info->major_version ||
2348             mddev->minor_version != info->minor_version ||
2349 /*          mddev->patch_version != info->patch_version || */
2350             mddev->ctime         != info->ctime         ||
2351             mddev->level         != info->level         ||
2352 /*          mddev->layout        != info->layout        || */
2353             !mddev->persistent   != info->not_persistent||
2354             mddev->chunk_size    != info->chunk_size    )
2355                 return -EINVAL;
2356         /* Check there is only one change */
2357         if (mddev->size != info->size) cnt++;
2358         if (mddev->raid_disks != info->raid_disks) cnt++;
2359         if (mddev->layout != info->layout) cnt++;
2360         if (cnt == 0) return 0;
2361         if (cnt > 1) return -EINVAL;
2362
2363         if (mddev->layout != info->layout) {
2364                 /* Change layout
2365                  * we don't need to do anything at the md level, the
2366                  * personality will take care of it all.
2367                  */
2368                 if (mddev->pers->reconfig == NULL)
2369                         return -EINVAL;
2370                 else
2371                         return mddev->pers->reconfig(mddev, info->layout, -1);
2372         }
2373         if (mddev->size != info->size) {
2374                 mdk_rdev_t * rdev;
2375                 struct list_head *tmp;
2376                 if (mddev->pers->resize == NULL)
2377                         return -EINVAL;
2378                 /* The "size" is the amount of each device that is used.
2379                  * This can only make sense for arrays with redundancy.
2380                  * linear and raid0 always use whatever space is available
2381                  * We can only consider changing the size if no resync
2382                  * or reconstruction is happening, and if the new size
2383                  * is acceptable. It must fit before the sb_offset or,
2384                  * if that is <data_offset, it must fit before the
2385                  * size of each device.
2386                  * If size is zero, we find the largest size that fits.
2387                  */
2388                 if (mddev->sync_thread)
2389                         return -EBUSY;
2390                 ITERATE_RDEV(mddev,rdev,tmp) {
2391                         sector_t avail;
2392                         int fit = (info->size == 0);
2393                         if (rdev->sb_offset > rdev->data_offset)
2394                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2395                         else
2396                                 avail = get_capacity(rdev->bdev->bd_disk)
2397                                         - rdev->data_offset;
2398                         if (fit && (info->size == 0 || info->size > avail/2))
2399                                 info->size = avail/2;
2400                         if (avail < ((sector_t)info->size << 1))
2401                                 return -ENOSPC;
2402                 }
2403                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2404                 if (!rv) {
2405                         struct block_device *bdev;
2406
2407                         bdev = bdget_disk(mddev->gendisk, 0);
2408                         if (bdev) {
2409                                 down(&bdev->bd_inode->i_sem);
2410                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2411                                 up(&bdev->bd_inode->i_sem);
2412                                 bdput(bdev);
2413                         }
2414                 }
2415         }
2416         if (mddev->raid_disks    != info->raid_disks) {
2417                 /* change the number of raid disks */
2418                 if (mddev->pers->reshape == NULL)
2419                         return -EINVAL;
2420                 if (info->raid_disks <= 0 ||
2421                     info->raid_disks >= mddev->max_disks)
2422                         return -EINVAL;
2423                 if (mddev->sync_thread)
2424                         return -EBUSY;
2425                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2426                 if (!rv) {
2427                         struct block_device *bdev;
2428
2429                         bdev = bdget_disk(mddev->gendisk, 0);
2430                         if (bdev) {
2431                                 down(&bdev->bd_inode->i_sem);
2432                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2433                                 up(&bdev->bd_inode->i_sem);
2434                                 bdput(bdev);
2435                         }
2436                 }
2437         }
2438         md_update_sb(mddev);
2439         return rv;
2440 }
2441
2442 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2443 {
2444         mdk_rdev_t *rdev;
2445
2446         if (mddev->pers == NULL)
2447                 return -ENODEV;
2448
2449         rdev = find_rdev(mddev, dev);
2450         if (!rdev)
2451                 return -ENODEV;
2452
2453         md_error(mddev, rdev);
2454         return 0;
2455 }
2456
2457 static int md_ioctl(struct inode *inode, struct file *file,
2458                         unsigned int cmd, unsigned long arg)
2459 {
2460         int err = 0;
2461         void __user *argp = (void __user *)arg;
2462         struct hd_geometry __user *loc = argp;
2463         mddev_t *mddev = NULL;
2464
2465         if (!capable(CAP_SYS_ADMIN))
2466                 return -EACCES;
2467
2468         /*
2469          * Commands dealing with the RAID driver but not any
2470          * particular array:
2471          */
2472         switch (cmd)
2473         {
2474                 case RAID_VERSION:
2475                         err = get_version(argp);
2476                         goto done;
2477
2478                 case PRINT_RAID_DEBUG:
2479                         err = 0;
2480                         md_print_devices();
2481                         goto done;
2482
2483 #ifndef MODULE
2484                 case RAID_AUTORUN:
2485                         err = 0;
2486                         autostart_arrays(arg);
2487                         goto done;
2488 #endif
2489                 default:;
2490         }
2491
2492         /*
2493          * Commands creating/starting a new array:
2494          */
2495
2496         mddev = inode->i_bdev->bd_disk->private_data;
2497
2498         if (!mddev) {
2499                 BUG();
2500                 goto abort;
2501         }
2502
2503
2504         if (cmd == START_ARRAY) {
2505                 /* START_ARRAY doesn't need to lock the array as autostart_array
2506                  * does the locking, and it could even be a different array
2507                  */
2508                 static int cnt = 3;
2509                 if (cnt > 0 ) {
2510                         printk(KERN_WARNING
2511                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2512                                "This will not be supported beyond 2.6\n",
2513                                current->comm, current->pid);
2514                         cnt--;
2515                 }
2516                 err = autostart_array(new_decode_dev(arg));
2517                 if (err) {
2518                         printk(KERN_WARNING "md: autostart failed!\n");
2519                         goto abort;
2520                 }
2521                 goto done;
2522         }
2523
2524         err = mddev_lock(mddev);
2525         if (err) {
2526                 printk(KERN_INFO
2527                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2528                         err, cmd);
2529                 goto abort;
2530         }
2531
2532         switch (cmd)
2533         {
2534                 case SET_ARRAY_INFO:
2535                         {
2536                                 mdu_array_info_t info;
2537                                 if (!arg)
2538                                         memset(&info, 0, sizeof(info));
2539                                 else if (copy_from_user(&info, argp, sizeof(info))) {
2540                                         err = -EFAULT;
2541                                         goto abort_unlock;
2542                                 }
2543                                 if (mddev->pers) {
2544                                         err = update_array_info(mddev, &info);
2545                                         if (err) {
2546                                                 printk(KERN_WARNING "md: couldn't update"
2547                                                        " array info. %d\n", err);
2548                                                 goto abort_unlock;
2549                                         }
2550                                         goto done_unlock;
2551                                 }
2552                                 if (!list_empty(&mddev->disks)) {
2553                                         printk(KERN_WARNING
2554                                                "md: array %s already has disks!\n",
2555                                                mdname(mddev));
2556                                         err = -EBUSY;
2557                                         goto abort_unlock;
2558                                 }
2559                                 if (mddev->raid_disks) {
2560                                         printk(KERN_WARNING
2561                                                "md: array %s already initialised!\n",
2562                                                mdname(mddev));
2563                                         err = -EBUSY;
2564                                         goto abort_unlock;
2565                                 }
2566                                 err = set_array_info(mddev, &info);
2567                                 if (err) {
2568                                         printk(KERN_WARNING "md: couldn't set"
2569                                                " array info. %d\n", err);
2570                                         goto abort_unlock;
2571                                 }
2572                         }
2573                         goto done_unlock;
2574
2575                 default:;
2576         }
2577
2578         /*
2579          * Commands querying/configuring an existing array:
2580          */
2581         /* if we are initialised yet, only ADD_NEW_DISK or STOP_ARRAY is allowed */
2582         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY && cmd != RUN_ARRAY) {
2583                 err = -ENODEV;
2584                 goto abort_unlock;
2585         }
2586
2587         /*
2588          * Commands even a read-only array can execute:
2589          */
2590         switch (cmd)
2591         {
2592                 case GET_ARRAY_INFO:
2593                         err = get_array_info(mddev, argp);
2594                         goto done_unlock;
2595
2596                 case GET_DISK_INFO:
2597                         err = get_disk_info(mddev, argp);
2598                         goto done_unlock;
2599
2600                 case RESTART_ARRAY_RW:
2601                         err = restart_array(mddev);
2602                         goto done_unlock;
2603
2604                 case STOP_ARRAY:
2605                         err = do_md_stop (mddev, 0);
2606                         goto done_unlock;
2607
2608                 case STOP_ARRAY_RO:
2609                         err = do_md_stop (mddev, 1);
2610                         goto done_unlock;
2611
2612         /*
2613          * We have a problem here : there is no easy way to give a CHS
2614          * virtual geometry. We currently pretend that we have a 2 heads
2615          * 4 sectors (with a BIG number of cylinders...). This drives
2616          * dosfs just mad... ;-)
2617          */
2618                 case HDIO_GETGEO:
2619                         if (!loc) {
2620                                 err = -EINVAL;
2621                                 goto abort_unlock;
2622                         }
2623                         err = put_user (2, (char __user *) &loc->heads);
2624                         if (err)
2625                                 goto abort_unlock;
2626                         err = put_user (4, (char __user *) &loc->sectors);
2627                         if (err)
2628                                 goto abort_unlock;
2629                         err = put_user(get_capacity(mddev->gendisk)/8,
2630                                         (short __user *) &loc->cylinders);
2631                         if (err)
2632                                 goto abort_unlock;
2633                         err = put_user (get_start_sect(inode->i_bdev),
2634                                                 (long __user *) &loc->start);
2635                         goto done_unlock;
2636         }
2637
2638         /*
2639          * The remaining ioctls are changing the state of the
2640          * superblock, so we do not allow read-only arrays
2641          * here:
2642          */
2643         if (mddev->ro) {
2644                 err = -EROFS;
2645                 goto abort_unlock;
2646         }
2647
2648         switch (cmd)
2649         {
2650                 case ADD_NEW_DISK:
2651                 {
2652                         mdu_disk_info_t info;
2653                         if (copy_from_user(&info, argp, sizeof(info)))
2654                                 err = -EFAULT;
2655                         else
2656                                 err = add_new_disk(mddev, &info);
2657                         goto done_unlock;
2658                 }
2659
2660                 case HOT_REMOVE_DISK:
2661                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2662                         goto done_unlock;
2663
2664                 case HOT_ADD_DISK:
2665                         err = hot_add_disk(mddev, new_decode_dev(arg));
2666                         goto done_unlock;
2667
2668                 case SET_DISK_FAULTY:
2669                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2670                         goto done_unlock;
2671
2672                 case RUN_ARRAY:
2673                         err = do_md_run (mddev);
2674                         goto done_unlock;
2675
2676                 default:
2677                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2678                                 printk(KERN_WARNING "md: %s(pid %d) used"
2679                                         " obsolete MD ioctl, upgrade your"
2680                                         " software to use new ictls.\n",
2681                                         current->comm, current->pid);
2682                         err = -EINVAL;
2683                         goto abort_unlock;
2684         }
2685
2686 done_unlock:
2687 abort_unlock:
2688         mddev_unlock(mddev);
2689
2690         return err;
2691 done:
2692         if (err)
2693                 MD_BUG();
2694 abort:
2695         return err;
2696 }
2697
2698 static int md_open(struct inode *inode, struct file *file)
2699 {
2700         /*
2701          * Succeed if we can lock the mddev, which confirms that
2702          * it isn't being stopped right now.
2703          */
2704         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2705         int err;
2706
2707         if ((err = mddev_lock(mddev)))
2708                 goto out;
2709
2710         err = 0;
2711         mddev_get(mddev);
2712         mddev_unlock(mddev);
2713
2714         check_disk_change(inode->i_bdev);
2715  out:
2716         return err;
2717 }
2718
2719 static int md_release(struct inode *inode, struct file * file)
2720 {
2721         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2722
2723         if (!mddev)
2724                 BUG();
2725         mddev_put(mddev);
2726
2727         return 0;
2728 }
2729
2730 static int md_media_changed(struct gendisk *disk)
2731 {
2732         mddev_t *mddev = disk->private_data;
2733
2734         return mddev->changed;
2735 }
2736
2737 static int md_revalidate(struct gendisk *disk)
2738 {
2739         mddev_t *mddev = disk->private_data;
2740
2741         mddev->changed = 0;
2742         return 0;
2743 }
2744 static struct block_device_operations md_fops =
2745 {
2746         .owner          = THIS_MODULE,
2747         .open           = md_open,
2748         .release        = md_release,
2749         .ioctl          = md_ioctl,
2750         .media_changed  = md_media_changed,
2751         .revalidate_disk= md_revalidate,
2752 };
2753
2754 int md_thread(void * arg)
2755 {
2756         mdk_thread_t *thread = arg;
2757
2758         lock_kernel();
2759
2760         /*
2761          * Detach thread
2762          */
2763
2764         daemonize(thread->name, mdname(thread->mddev));
2765
2766         current->exit_signal = SIGCHLD;
2767         allow_signal(SIGKILL);
2768         thread->tsk = current;
2769
2770         /*
2771          * md_thread is a 'system-thread', it's priority should be very
2772          * high. We avoid resource deadlocks individually in each
2773          * raid personality. (RAID5 does preallocation) We also use RR and
2774          * the very same RT priority as kswapd, thus we will never get
2775          * into a priority inversion deadlock.
2776          *
2777          * we definitely have to have equal or higher priority than
2778          * bdflush, otherwise bdflush will deadlock if there are too
2779          * many dirty RAID5 blocks.
2780          */
2781         unlock_kernel();
2782
2783         complete(thread->event);
2784         while (thread->run) {
2785                 void (*run)(mddev_t *);
2786
2787                 wait_event_interruptible(thread->wqueue,
2788                                          test_bit(THREAD_WAKEUP, &thread->flags));
2789                 if (current->flags & PF_FREEZE)
2790                         refrigerator(PF_FREEZE);
2791
2792                 clear_bit(THREAD_WAKEUP, &thread->flags);
2793
2794                 run = thread->run;
2795                 if (run)
2796                         run(thread->mddev);
2797
2798                 if (signal_pending(current))
2799                         flush_signals(current);
2800         }
2801         complete(thread->event);
2802         return 0;
2803 }
2804
2805 void md_wakeup_thread(mdk_thread_t *thread)
2806 {
2807         if (thread) {
2808                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
2809                 set_bit(THREAD_WAKEUP, &thread->flags);
2810                 wake_up(&thread->wqueue);
2811         }
2812 }
2813
2814 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
2815                                  const char *name)
2816 {
2817         mdk_thread_t *thread;
2818         int ret;
2819         struct completion event;
2820
2821         thread = (mdk_thread_t *) kmalloc
2822                                 (sizeof(mdk_thread_t), GFP_KERNEL);
2823         if (!thread)
2824                 return NULL;
2825
2826         memset(thread, 0, sizeof(mdk_thread_t));
2827         init_waitqueue_head(&thread->wqueue);
2828
2829         init_completion(&event);
2830         thread->event = &event;
2831         thread->run = run;
2832         thread->mddev = mddev;
2833         thread->name = name;
2834         ret = kernel_thread(md_thread, thread, 0);
2835         if (ret < 0) {
2836                 kfree(thread);
2837                 return NULL;
2838         }
2839         wait_for_completion(&event);
2840         return thread;
2841 }
2842
2843 static void md_interrupt_thread(mdk_thread_t *thread)
2844 {
2845         if (!thread->tsk) {
2846                 MD_BUG();
2847                 return;
2848         }
2849         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
2850         send_sig(SIGKILL, thread->tsk, 1);
2851 }
2852
2853 void md_unregister_thread(mdk_thread_t *thread)
2854 {
2855         struct completion event;
2856
2857         init_completion(&event);
2858
2859         thread->event = &event;
2860         thread->run = NULL;
2861         thread->name = NULL;
2862         md_interrupt_thread(thread);
2863         wait_for_completion(&event);
2864         kfree(thread);
2865 }
2866
2867 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
2868 {
2869         if (!mddev) {
2870                 MD_BUG();
2871                 return;
2872         }
2873
2874         if (!rdev || rdev->faulty)
2875                 return;
2876
2877         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
2878                 mdname(mddev),
2879                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
2880                 __builtin_return_address(0),__builtin_return_address(1),
2881                 __builtin_return_address(2),__builtin_return_address(3));
2882
2883         if (!mddev->pers->error_handler)
2884                 return;
2885         mddev->pers->error_handler(mddev,rdev);
2886         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2887         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2888         md_wakeup_thread(mddev->thread);
2889 }
2890
2891 /* seq_file implementation /proc/mdstat */
2892
2893 static void status_unused(struct seq_file *seq)
2894 {
2895         int i = 0;
2896         mdk_rdev_t *rdev;
2897         struct list_head *tmp;
2898
2899         seq_printf(seq, "unused devices: ");
2900
2901         ITERATE_RDEV_PENDING(rdev,tmp) {
2902                 char b[BDEVNAME_SIZE];
2903                 i++;
2904                 seq_printf(seq, "%s ",
2905                               bdevname(rdev->bdev,b));
2906         }
2907         if (!i)
2908                 seq_printf(seq, "<none>");
2909
2910         seq_printf(seq, "\n");
2911 }
2912
2913
2914 static void status_resync(struct seq_file *seq, mddev_t * mddev)
2915 {
2916         unsigned long max_blocks, resync, res, dt, db, rt;
2917
2918         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
2919
2920         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2921                 max_blocks = mddev->resync_max_sectors >> 1;
2922         else
2923                 max_blocks = mddev->size;
2924
2925         /*
2926          * Should not happen.
2927          */
2928         if (!max_blocks) {
2929                 MD_BUG();
2930                 return;
2931         }
2932         res = (resync/1024)*1000/(max_blocks/1024 + 1);
2933         {
2934                 int i, x = res/50, y = 20-x;
2935                 seq_printf(seq, "[");
2936                 for (i = 0; i < x; i++)
2937                         seq_printf(seq, "=");
2938                 seq_printf(seq, ">");
2939                 for (i = 0; i < y; i++)
2940                         seq_printf(seq, ".");
2941                 seq_printf(seq, "] ");
2942         }
2943         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
2944                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
2945                        "resync" : "recovery"),
2946                       res/10, res % 10, resync, max_blocks);
2947
2948         /*
2949          * We do not want to overflow, so the order of operands and
2950          * the * 100 / 100 trick are important. We do a +1 to be
2951          * safe against division by zero. We only estimate anyway.
2952          *
2953          * dt: time from mark until now
2954          * db: blocks written from mark until now
2955          * rt: remaining time
2956          */
2957         dt = ((jiffies - mddev->resync_mark) / HZ);
2958         if (!dt) dt++;
2959         db = resync - (mddev->resync_mark_cnt/2);
2960         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
2961
2962         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
2963
2964         seq_printf(seq, " speed=%ldK/sec", db/dt);
2965 }
2966
2967 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
2968 {
2969         struct list_head *tmp;
2970         loff_t l = *pos;
2971         mddev_t *mddev;
2972
2973         if (l >= 0x10000)
2974                 return NULL;
2975         if (!l--)
2976                 /* header */
2977                 return (void*)1;
2978
2979         spin_lock(&all_mddevs_lock);
2980         list_for_each(tmp,&all_mddevs)
2981                 if (!l--) {
2982                         mddev = list_entry(tmp, mddev_t, all_mddevs);
2983                         mddev_get(mddev);
2984                         spin_unlock(&all_mddevs_lock);
2985                         return mddev;
2986                 }
2987         spin_unlock(&all_mddevs_lock);
2988         if (!l--)
2989                 return (void*)2;/* tail */
2990         return NULL;
2991 }
2992
2993 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2994 {
2995         struct list_head *tmp;
2996         mddev_t *next_mddev, *mddev = v;
2997
2998         ++*pos;
2999         if (v == (void*)2)
3000                 return NULL;
3001
3002         spin_lock(&all_mddevs_lock);
3003         if (v == (void*)1)
3004                 tmp = all_mddevs.next;
3005         else
3006                 tmp = mddev->all_mddevs.next;
3007         if (tmp != &all_mddevs)
3008                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3009         else {
3010                 next_mddev = (void*)2;
3011                 *pos = 0x10000;
3012         }
3013         spin_unlock(&all_mddevs_lock);
3014
3015         if (v != (void*)1)
3016                 mddev_put(mddev);
3017         return next_mddev;
3018
3019 }
3020
3021 static void md_seq_stop(struct seq_file *seq, void *v)
3022 {
3023         mddev_t *mddev = v;
3024
3025         if (mddev && v != (void*)1 && v != (void*)2)
3026                 mddev_put(mddev);
3027 }
3028
3029 static int md_seq_show(struct seq_file *seq, void *v)
3030 {
3031         mddev_t *mddev = v;
3032         sector_t size;
3033         struct list_head *tmp2;
3034         mdk_rdev_t *rdev;
3035         int i;
3036
3037         if (v == (void*)1) {
3038                 seq_printf(seq, "Personalities : ");
3039                 spin_lock(&pers_lock);
3040                 for (i = 0; i < MAX_PERSONALITY; i++)
3041                         if (pers[i])
3042                                 seq_printf(seq, "[%s] ", pers[i]->name);
3043
3044                 spin_unlock(&pers_lock);
3045                 seq_printf(seq, "\n");
3046                 return 0;
3047         }
3048         if (v == (void*)2) {
3049                 status_unused(seq);
3050                 return 0;
3051         }
3052
3053         if (mddev_lock(mddev)!=0)
3054                 return -EINTR;
3055         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3056                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3057                                                 mddev->pers ? "" : "in");
3058                 if (mddev->pers) {
3059                         if (mddev->ro)
3060                                 seq_printf(seq, " (read-only)");
3061                         seq_printf(seq, " %s", mddev->pers->name);
3062                 }
3063
3064                 size = 0;
3065                 ITERATE_RDEV(mddev,rdev,tmp2) {
3066                         char b[BDEVNAME_SIZE];
3067                         seq_printf(seq, " %s[%d]",
3068                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3069                         if (rdev->faulty) {
3070                                 seq_printf(seq, "(F)");
3071                                 continue;
3072                         }
3073                         size += rdev->size;
3074                 }
3075
3076                 if (!list_empty(&mddev->disks)) {
3077                         if (mddev->pers)
3078                                 seq_printf(seq, "\n      %llu blocks",
3079                                         (unsigned long long)mddev->array_size);
3080                         else
3081                                 seq_printf(seq, "\n      %llu blocks",
3082                                         (unsigned long long)size);
3083                 }
3084
3085                 if (mddev->pers) {
3086                         mddev->pers->status (seq, mddev);
3087                         seq_printf(seq, "\n      ");
3088                         if (mddev->curr_resync > 2)
3089                                 status_resync (seq, mddev);
3090                         else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3091                                 seq_printf(seq, "       resync=DELAYED");
3092                 }
3093
3094                 seq_printf(seq, "\n");
3095         }
3096         mddev_unlock(mddev);
3097
3098         return 0;
3099 }
3100
3101 static struct seq_operations md_seq_ops = {
3102         .start  = md_seq_start,
3103         .next   = md_seq_next,
3104         .stop   = md_seq_stop,
3105         .show   = md_seq_show,
3106 };
3107
3108 static int md_seq_open(struct inode *inode, struct file *file)
3109 {
3110         int error;
3111
3112         error = seq_open(file, &md_seq_ops);
3113         return error;
3114 }
3115
3116 static struct file_operations md_seq_fops = {
3117         .open           = md_seq_open,
3118         .read           = seq_read,
3119         .llseek         = seq_lseek,
3120         .release        = seq_release,
3121 };
3122
3123 int register_md_personality(int pnum, mdk_personality_t *p)
3124 {
3125         if (pnum >= MAX_PERSONALITY) {
3126                 printk(KERN_ERR
3127                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3128                        p->name, pnum, MAX_PERSONALITY-1);
3129                 return -EINVAL;
3130         }
3131
3132         spin_lock(&pers_lock);
3133         if (pers[pnum]) {
3134                 spin_unlock(&pers_lock);
3135                 MD_BUG();
3136                 return -EBUSY;
3137         }
3138
3139         pers[pnum] = p;
3140         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3141         spin_unlock(&pers_lock);
3142         return 0;
3143 }
3144
3145 int unregister_md_personality(int pnum)
3146 {
3147         if (pnum >= MAX_PERSONALITY) {
3148                 MD_BUG();
3149                 return -EINVAL;
3150         }
3151
3152         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3153         spin_lock(&pers_lock);
3154         pers[pnum] = NULL;
3155         spin_unlock(&pers_lock);
3156         return 0;
3157 }
3158
3159 static int is_mddev_idle(mddev_t *mddev)
3160 {
3161         mdk_rdev_t * rdev;
3162         struct list_head *tmp;
3163         int idle;
3164         unsigned long curr_events;
3165
3166         idle = 1;
3167         ITERATE_RDEV(mddev,rdev,tmp) {
3168                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3169                 curr_events = disk_stat_read(disk, read_sectors) +
3170                                 disk_stat_read(disk, write_sectors) -
3171                                 atomic_read(&disk->sync_io);
3172                 /* Allow some slack between valud of curr_events and last_events,
3173                  * as there are some uninteresting races.
3174                  * Note: the following is an unsigned comparison.
3175                  */
3176                 if ((curr_events - rdev->last_events + 32) > 64) {
3177                         rdev->last_events = curr_events;
3178                         idle = 0;
3179                 }
3180         }
3181         return idle;
3182 }
3183
3184 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3185 {
3186         /* another "blocks" (512byte) blocks have been synced */
3187         atomic_sub(blocks, &mddev->recovery_active);
3188         wake_up(&mddev->recovery_wait);
3189         if (!ok) {
3190                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3191                 md_wakeup_thread(mddev->thread);
3192                 // stop recovery, signal do_sync ....
3193         }
3194 }
3195
3196
3197 void md_write_start(mddev_t *mddev)
3198 {
3199         if (!atomic_read(&mddev->writes_pending)) {
3200                 mddev_lock_uninterruptible(mddev);
3201                 if (mddev->in_sync) {
3202                         mddev->in_sync = 0;
3203                         del_timer(&mddev->safemode_timer);
3204                         md_update_sb(mddev);
3205                 }
3206                 atomic_inc(&mddev->writes_pending);
3207                 mddev_unlock(mddev);
3208         } else
3209                 atomic_inc(&mddev->writes_pending);
3210 }
3211
3212 void md_write_end(mddev_t *mddev)
3213 {
3214         if (atomic_dec_and_test(&mddev->writes_pending)) {
3215                 if (mddev->safemode == 2)
3216                         md_wakeup_thread(mddev->thread);
3217                 else
3218                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3219         }
3220 }
3221
3222 static inline void md_enter_safemode(mddev_t *mddev)
3223 {
3224         if (!mddev->safemode) return;
3225         if (mddev->safemode == 2 &&
3226             (atomic_read(&mddev->writes_pending) || mddev->in_sync ||
3227                     mddev->recovery_cp != MaxSector))
3228                 return; /* avoid the lock */
3229         mddev_lock_uninterruptible(mddev);
3230         if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3231             !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3232                 mddev->in_sync = 1;
3233                 md_update_sb(mddev);
3234         }
3235         mddev_unlock(mddev);
3236
3237         if (mddev->safemode == 1)
3238                 mddev->safemode = 0;
3239 }
3240
3241 void md_handle_safemode(mddev_t *mddev)
3242 {
3243         if (signal_pending(current)) {
3244                 printk(KERN_INFO "md: %s in immediate safe mode\n",
3245                         mdname(mddev));
3246                 mddev->safemode = 2;
3247                 flush_signals(current);
3248         }
3249         md_enter_safemode(mddev);
3250 }
3251
3252
3253 DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3254
3255 #define SYNC_MARKS      10
3256 #define SYNC_MARK_STEP  (3*HZ)
3257 static void md_do_sync(mddev_t *mddev)
3258 {
3259         mddev_t *mddev2;
3260         unsigned int currspeed = 0,
3261                  window;
3262         sector_t max_sectors,j;
3263         unsigned long mark[SYNC_MARKS];
3264         sector_t mark_cnt[SYNC_MARKS];
3265         int last_mark,m;
3266         struct list_head *tmp;
3267         sector_t last_check;
3268
3269         /* just incase thread restarts... */
3270         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3271                 return;
3272
3273         /* we overload curr_resync somewhat here.
3274          * 0 == not engaged in resync at all
3275          * 2 == checking that there is no conflict with another sync
3276          * 1 == like 2, but have yielded to allow conflicting resync to
3277          *              commense
3278          * other == active in resync - this many blocks
3279          *
3280          * Before starting a resync we must have set curr_resync to
3281          * 2, and then checked that every "conflicting" array has curr_resync
3282          * less than ours.  When we find one that is the same or higher
3283          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3284          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3285          * This will mean we have to start checking from the beginning again.
3286          *
3287          */
3288
3289         do {
3290                 mddev->curr_resync = 2;
3291
3292         try_again:
3293                 if (signal_pending(current)) {
3294                         flush_signals(current);
3295                         goto skip;
3296                 }
3297                 ITERATE_MDDEV(mddev2,tmp) {
3298                         printk(".");
3299                         if (mddev2 == mddev)
3300                                 continue;
3301                         if (mddev2->curr_resync &&
3302                             match_mddev_units(mddev,mddev2)) {
3303                                 DEFINE_WAIT(wq);
3304                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3305                                         /* arbitrarily yield */
3306                                         mddev->curr_resync = 1;
3307                                         wake_up(&resync_wait);
3308                                 }
3309                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3310                                         /* no need to wait here, we can wait the next
3311                                          * time 'round when curr_resync == 2
3312                                          */
3313                                         continue;
3314                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3315                                 if (!signal_pending(current)
3316                                     && mddev2->curr_resync >= mddev->curr_resync) {
3317                                         printk(KERN_INFO "md: delaying resync of %s"
3318                                                " until %s has finished resync (they"
3319                                                " share one or more physical units)\n",
3320                                                mdname(mddev), mdname(mddev2));
3321                                         mddev_put(mddev2);
3322                                         schedule();
3323                                         finish_wait(&resync_wait, &wq);
3324                                         goto try_again;
3325                                 }
3326                                 finish_wait(&resync_wait, &wq);
3327                         }
3328                 }
3329         } while (mddev->curr_resync < 2);
3330
3331         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3332                 /* resync follows the size requested by the personality,
3333                  * which default to physical size, but can be virtual size
3334                  */
3335                 max_sectors = mddev->resync_max_sectors;
3336         else
3337                 /* recovery follows the physical size of devices */
3338                 max_sectors = mddev->size << 1;
3339
3340         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3341         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3342                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3343         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3344                "(but not more than %d KB/sec) for reconstruction.\n",
3345                sysctl_speed_limit_max);
3346
3347         is_mddev_idle(mddev); /* this also initializes IO event counters */
3348         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3349                 j = mddev->recovery_cp;
3350         else
3351                 j = 0;
3352         for (m = 0; m < SYNC_MARKS; m++) {
3353                 mark[m] = jiffies;
3354                 mark_cnt[m] = j;
3355         }
3356         last_mark = 0;
3357         mddev->resync_mark = mark[last_mark];
3358         mddev->resync_mark_cnt = mark_cnt[last_mark];
3359
3360         /*
3361          * Tune reconstruction:
3362          */
3363         window = 32*(PAGE_SIZE/512);
3364         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3365                 window/2,(unsigned long long) max_sectors/2);
3366
3367         atomic_set(&mddev->recovery_active, 0);
3368         init_waitqueue_head(&mddev->recovery_wait);
3369         last_check = 0;
3370
3371         if (j>2) {
3372                 printk(KERN_INFO
3373                         "md: resuming recovery of %s from checkpoint.\n",
3374                         mdname(mddev));
3375                 mddev->curr_resync = j;
3376         }
3377
3378         while (j < max_sectors) {
3379                 int sectors;
3380
3381                 sectors = mddev->pers->sync_request(mddev, j, currspeed < sysctl_speed_limit_min);
3382                 if (sectors < 0) {
3383                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3384                         goto out;
3385                 }
3386                 atomic_add(sectors, &mddev->recovery_active);
3387                 j += sectors;
3388                 if (j>1) mddev->curr_resync = j;
3389
3390                 if (last_check + window > j || j == max_sectors)
3391                         continue;
3392
3393                 last_check = j;
3394
3395                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3396                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3397                         break;
3398
3399         repeat:
3400                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3401                         /* step marks */
3402                         int next = (last_mark+1) % SYNC_MARKS;
3403
3404                         mddev->resync_mark = mark[next];
3405                         mddev->resync_mark_cnt = mark_cnt[next];
3406                         mark[next] = jiffies;
3407                         mark_cnt[next] = j - atomic_read(&mddev->recovery_active);
3408                         last_mark = next;
3409                 }
3410
3411
3412                 if (signal_pending(current)) {
3413                         /*
3414                          * got a signal, exit.
3415                          */
3416                         printk(KERN_INFO
3417                                 "md: md_do_sync() got signal ... exiting\n");
3418                         flush_signals(current);
3419                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3420                         goto out;
3421                 }
3422
3423                 /*
3424                  * this loop exits only if either when we are slower than
3425                  * the 'hard' speed limit, or the system was IO-idle for
3426                  * a jiffy.
3427                  * the system might be non-idle CPU-wise, but we only care
3428                  * about not overloading the IO subsystem. (things like an
3429                  * e2fsck being done on the RAID array should execute fast)
3430                  */
3431                 mddev->queue->unplug_fn(mddev->queue);
3432                 cond_resched();
3433
3434                 currspeed = ((unsigned long)(j-mddev->resync_mark_cnt))/2/((jiffies-mddev->resync_mark)/HZ +1) +1;
3435
3436                 if (currspeed > sysctl_speed_limit_min) {
3437                         if ((currspeed > sysctl_speed_limit_max) ||
3438                                         !is_mddev_idle(mddev)) {
3439                                 msleep_interruptible(250);
3440                                 goto repeat;
3441                         }
3442                 }
3443         }
3444         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3445         /*
3446          * this also signals 'finished resyncing' to md_stop
3447          */
3448  out:
3449         mddev->queue->unplug_fn(mddev->queue);
3450
3451         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3452
3453         /* tell personality that we are finished */
3454         mddev->pers->sync_request(mddev, max_sectors, 1);
3455
3456         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3457             mddev->curr_resync > 2 &&
3458             mddev->curr_resync >= mddev->recovery_cp) {
3459                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3460                         printk(KERN_INFO
3461                                 "md: checkpointing recovery of %s.\n",
3462                                 mdname(mddev));
3463                         mddev->recovery_cp = mddev->curr_resync;
3464                 } else
3465                         mddev->recovery_cp = MaxSector;
3466         }
3467
3468         md_enter_safemode(mddev);
3469  skip:
3470         mddev->curr_resync = 0;
3471         wake_up(&resync_wait);
3472         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3473         md_wakeup_thread(mddev->thread);
3474 }
3475
3476
3477 /*
3478  * This routine is regularly called by all per-raid-array threads to
3479  * deal with generic issues like resync and super-block update.
3480  * Raid personalities that don't have a thread (linear/raid0) do not
3481  * need this as they never do any recovery or update the superblock.
3482  *
3483  * It does not do any resync itself, but rather "forks" off other threads
3484  * to do that as needed.
3485  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3486  * "->recovery" and create a thread at ->sync_thread.
3487  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3488  * and wakeups up this thread which will reap the thread and finish up.
3489  * This thread also removes any faulty devices (with nr_pending == 0).
3490  *
3491  * The overall approach is:
3492  *  1/ if the superblock needs updating, update it.
3493  *  2/ If a recovery thread is running, don't do anything else.
3494  *  3/ If recovery has finished, clean up, possibly marking spares active.
3495  *  4/ If there are any faulty devices, remove them.
3496  *  5/ If array is degraded, try to add spares devices
3497  *  6/ If array has spares or is not in-sync, start a resync thread.
3498  */
3499 void md_check_recovery(mddev_t *mddev)
3500 {
3501         mdk_rdev_t *rdev;
3502         struct list_head *rtmp;
3503
3504
3505         dprintk(KERN_INFO "md: recovery thread got woken up ...\n");
3506
3507         if (mddev->ro)
3508                 return;
3509         if ( ! (
3510                 mddev->sb_dirty ||
3511                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3512                 test_bit(MD_RECOVERY_DONE, &mddev->recovery)
3513                 ))
3514                 return;
3515         if (mddev_trylock(mddev)==0) {
3516                 int spares =0;
3517                 if (mddev->sb_dirty)
3518                         md_update_sb(mddev);
3519                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3520                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3521                         /* resync/recovery still happening */
3522                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3523                         goto unlock;
3524                 }
3525                 if (mddev->sync_thread) {
3526                         /* resync has finished, collect result */
3527                         md_unregister_thread(mddev->sync_thread);
3528                         mddev->sync_thread = NULL;
3529                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3530                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3531                                 /* success...*/
3532                                 /* activate any spares */
3533                                 mddev->pers->spare_active(mddev);
3534                         }
3535                         md_update_sb(mddev);
3536                         mddev->recovery = 0;
3537                         /* flag recovery needed just to double check */
3538                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3539                         goto unlock;
3540                 }
3541                 if (mddev->recovery)
3542                         /* probably just the RECOVERY_NEEDED flag */
3543                         mddev->recovery = 0;
3544
3545                 /* no recovery is running.
3546                  * remove any failed drives, then
3547                  * add spares if possible.
3548                  * Spare are also removed and re-added, to allow
3549                  * the personality to fail the re-add.
3550                  */
3551                 ITERATE_RDEV(mddev,rdev,rtmp)
3552                         if (rdev->raid_disk >= 0 &&
3553                             (rdev->faulty || ! rdev->in_sync) &&
3554                             atomic_read(&rdev->nr_pending)==0) {
3555                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3556                                         rdev->raid_disk = -1;
3557                         }
3558
3559                 if (mddev->degraded) {
3560                         ITERATE_RDEV(mddev,rdev,rtmp)
3561                                 if (rdev->raid_disk < 0
3562                                     && !rdev->faulty) {
3563                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3564                                                 spares++;
3565                                         else
3566                                                 break;
3567                                 }
3568                 }
3569
3570                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3571                         /* nothing we can do ... */
3572                         goto unlock;
3573                 }
3574                 if (mddev->pers->sync_request) {
3575                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3576                         if (!spares)
3577                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3578                         mddev->sync_thread = md_register_thread(md_do_sync,
3579                                                                 mddev,
3580                                                                 "%s_resync");
3581                         if (!mddev->sync_thread) {
3582                                 printk(KERN_ERR "%s: could not start resync"
3583                                         " thread...\n",
3584                                         mdname(mddev));
3585                                 /* leave the spares where they are, it shouldn't hurt */
3586                                 mddev->recovery = 0;
3587                         } else {
3588                                 md_wakeup_thread(mddev->sync_thread);
3589                         }
3590                 }
3591         unlock:
3592                 mddev_unlock(mddev);
3593         }
3594 }
3595
3596 int md_notify_reboot(struct notifier_block *this,
3597                                         unsigned long code, void *x)
3598 {
3599         struct list_head *tmp;
3600         mddev_t *mddev;
3601
3602         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3603
3604                 printk(KERN_INFO "md: stopping all md devices.\n");
3605
3606                 ITERATE_MDDEV(mddev,tmp)
3607                         if (mddev_trylock(mddev)==0)
3608                                 do_md_stop (mddev, 1);
3609                 /*
3610                  * certain more exotic SCSI devices are known to be
3611                  * volatile wrt too early system reboots. While the
3612                  * right place to handle this issue is the given
3613                  * driver, we do want to have a safe RAID driver ...
3614                  */
3615                 mdelay(1000*1);
3616         }
3617         return NOTIFY_DONE;
3618 }
3619
3620 struct notifier_block md_notifier = {
3621         .notifier_call  = md_notify_reboot,
3622         .next           = NULL,
3623         .priority       = INT_MAX, /* before any real devices */
3624 };
3625
3626 static void md_geninit(void)
3627 {
3628         struct proc_dir_entry *p;
3629
3630         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3631
3632         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3633         if (p)
3634                 p->proc_fops = &md_seq_fops;
3635 }
3636
3637 int __init md_init(void)
3638 {
3639         int minor;
3640
3641         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3642                         " MD_SB_DISKS=%d\n",
3643                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3644                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3645
3646         if (register_blkdev(MAJOR_NR, "md"))
3647                 return -1;
3648         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3649                 unregister_blkdev(MAJOR_NR, "md");
3650                 return -1;
3651         }
3652         devfs_mk_dir("md");
3653         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3654                                 md_probe, NULL, NULL);
3655         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3656                             md_probe, NULL, NULL);
3657
3658         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3659                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3660                                 S_IFBLK|S_IRUSR|S_IWUSR,
3661                                 "md/%d", minor);
3662
3663         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3664                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3665                               S_IFBLK|S_IRUSR|S_IWUSR,
3666                               "md/mdp%d", minor);
3667
3668
3669         register_reboot_notifier(&md_notifier);
3670         raid_table_header = register_sysctl_table(raid_root_table, 1);
3671
3672         md_geninit();
3673         return (0);
3674 }
3675
3676
3677 #ifndef MODULE
3678
3679 /*
3680  * Searches all registered partitions for autorun RAID arrays
3681  * at boot time.
3682  */
3683 static dev_t detected_devices[128];
3684 static int dev_cnt;
3685
3686 void md_autodetect_dev(dev_t dev)
3687 {
3688         if (dev_cnt >= 0 && dev_cnt < 127)
3689                 detected_devices[dev_cnt++] = dev;
3690 }
3691
3692
3693 static void autostart_arrays(int part)
3694 {
3695         mdk_rdev_t *rdev;
3696         int i;
3697
3698         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3699
3700         for (i = 0; i < dev_cnt; i++) {
3701                 dev_t dev = detected_devices[i];
3702
3703                 rdev = md_import_device(dev,0, 0);
3704                 if (IS_ERR(rdev))
3705                         continue;
3706
3707                 if (rdev->faulty) {
3708                         MD_BUG();
3709                         continue;
3710                 }
3711                 list_add(&rdev->same_set, &pending_raid_disks);
3712         }
3713         dev_cnt = 0;
3714
3715         autorun_devices(part);
3716 }
3717
3718 #endif
3719
3720 static __exit void md_exit(void)
3721 {
3722         mddev_t *mddev;
3723         struct list_head *tmp;
3724         int i;
3725         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3726         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3727         for (i=0; i < MAX_MD_DEVS; i++)
3728                 devfs_remove("md/%d", i);
3729         for (i=0; i < MAX_MD_DEVS; i++)
3730                 devfs_remove("md/d%d", i);
3731
3732         devfs_remove("md");
3733
3734         unregister_blkdev(MAJOR_NR,"md");
3735         unregister_blkdev(mdp_major, "mdp");
3736         unregister_reboot_notifier(&md_notifier);
3737         unregister_sysctl_table(raid_table_header);
3738         remove_proc_entry("mdstat", NULL);
3739         ITERATE_MDDEV(mddev,tmp) {
3740                 struct gendisk *disk = mddev->gendisk;
3741                 if (!disk)
3742                         continue;
3743                 export_array(mddev);
3744                 del_gendisk(disk);
3745                 put_disk(disk);
3746                 mddev->gendisk = NULL;
3747                 mddev_put(mddev);
3748         }
3749 }
3750
3751 module_init(md_init)
3752 module_exit(md_exit)
3753
3754 EXPORT_SYMBOL(register_md_personality);
3755 EXPORT_SYMBOL(unregister_md_personality);
3756 EXPORT_SYMBOL(md_error);
3757 EXPORT_SYMBOL(md_done_sync);
3758 EXPORT_SYMBOL(md_write_start);
3759 EXPORT_SYMBOL(md_write_end);
3760 EXPORT_SYMBOL(md_handle_safemode);
3761 EXPORT_SYMBOL(md_register_thread);
3762 EXPORT_SYMBOL(md_unregister_thread);
3763 EXPORT_SYMBOL(md_wakeup_thread);
3764 EXPORT_SYMBOL(md_print_devices);
3765 EXPORT_SYMBOL(md_check_recovery);
3766 MODULE_LICENSE("GPL");