drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    This program is free software; you can redistribute it and/or modify
  23    it under the terms of the GNU General Public License as published by
  24    the Free Software Foundation; either version 2, or (at your option)
  25    any later version.
  26
  27    You should have received a copy of the GNU General Public License
  28    (for example /usr/src/linux/COPYING); if not, write to the Free
  29    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  30 */
  31
  32 #include <linux/module.h>
  33 #include <linux/config.h>
  34 #include <linux/linkage.h>
  35 #include <linux/raid/md.h>
  36 #include <linux/sysctl.h>
  37 #include <linux/devfs_fs_kernel.h>
  38 #include <linux/buffer_head.h> /* for invalidate_bdev */
  39 #include <linux/suspend.h>
  40
  41 #include <linux/init.h>
  42
  43 #ifdef CONFIG_KMOD
  44 #include <linux/kmod.h>
  45 #endif
  46
  47 #include <asm/unaligned.h>
  48
  49 #define MAJOR_NR MD_MAJOR
  50 #define MD_DRIVER
  51
  52 /* 63 partitions with the alternate major number (mdp) */
  53 #define MdpMinorShift 6
  54
  55 #define DEBUG 0
  56 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  57
  58
  59 #ifndef MODULE
  60 static void autostart_arrays (int part);
  61 #endif
  62
  63 static mdk_personality_t *pers[MAX_PERSONALITY];
  64 static spinlock_t pers_lock = SPIN_LOCK_UNLOCKED;
  65
  66 /*
  67  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  68  * is 1000 KB/sec, so the extra system load does not show up that much.
  69  * Increase it if you want to have more _guaranteed_ speed. Note that
  70  * the RAID driver will use the maximum available bandwith if the IO
  71  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  72  * speed limit - in case reconstruction slows down your system despite
  73  * idle IO detection.
  74  *
  75  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  76  */
  77
  78 static int sysctl_speed_limit_min = 1000;
  79 static int sysctl_speed_limit_max = 200000;
  80
  81 static struct ctl_table_header *raid_table_header;
  82
  83 static ctl_table raid_table[] = {
  84         {
  85                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  86                 .procname       = "speed_limit_min",
  87                 .data           = &sysctl_speed_limit_min,
  88                 .maxlen         = sizeof(int),
  89                 .mode           = 0644,
  90                 .proc_handler   = &proc_dointvec,
  91         },
  92         {
  93                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
  94                 .procname       = "speed_limit_max",
  95                 .data           = &sysctl_speed_limit_max,
  96                 .maxlen         = sizeof(int),
  97                 .mode           = 0644,
  98                 .proc_handler   = &proc_dointvec,
  99         },
 100         { .ctl_name = 0 }
 101 };
 102
 103 static ctl_table raid_dir_table[] = {
 104         {
 105                 .ctl_name       = DEV_RAID,
 106                 .procname       = "raid",
 107                 .maxlen         = 0,
 108                 .mode           = 0555,
 109                 .child          = raid_table,
 110         },
 111         { .ctl_name = 0 }
 112 };
 113
 114 static ctl_table raid_root_table[] = {
 115         {
 116                 .ctl_name       = CTL_DEV,
 117                 .procname       = "dev",
 118                 .maxlen         = 0,
 119                 .mode           = 0555,
 120                 .child          = raid_dir_table,
 121         },
 122         { .ctl_name = 0 }
 123 };
 124
 125 static struct block_device_operations md_fops;
 126
 127 /*
 128  * Enables to iterate over all existing md arrays
 129  * all_mddevs_lock protects this list.
 130  */
 131 static LIST_HEAD(all_mddevs);
 132 static spinlock_t all_mddevs_lock = SPIN_LOCK_UNLOCKED;
 133
 134
 135 /*
 136  * iterates through all used mddevs in the system.
 137  * We take care to grab the all_mddevs_lock whenever navigating
 138  * the list, and to always hold a refcount when unlocked.
 139  * Any code which breaks out of this loop while own
 140  * a reference to the current mddev and must mddev_put it.
 141  */
 142 #define ITERATE_MDDEV(mddev,tmp)                                        \
 143                                                                         \
 144         for (({ spin_lock(&all_mddevs_lock);                            \
 145                 tmp = all_mddevs.next;                                  \
 146                 mddev = NULL;});                                        \
 147              ({ if (tmp != &all_mddevs)                                 \
 148                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 149                 spin_unlock(&all_mddevs_lock);                          \
 150                 if (mddev) mddev_put(mddev);                            \
 151                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 152                 tmp != &all_mddevs;});                                  \
 153              ({ spin_lock(&all_mddevs_lock);                            \
 154                 tmp = tmp->next;})                                      \
 155                 )
 156
 157 int md_flush_mddev(mddev_t *mddev, sector_t *error_sector)
 158 {
 159         struct list_head *tmp;
 160         mdk_rdev_t *rdev;
 161         int ret = 0;
 162
 163         /*
 164          * this list iteration is done without any locking in md?!
 165          */
 166         ITERATE_RDEV(mddev, rdev, tmp) {
 167                 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
 168                 int err;
 169
 170                 if (!r_queue->issue_flush_fn)
 171                         err = -EOPNOTSUPP;
 172                 else
 173                         err = r_queue->issue_flush_fn(r_queue, rdev->bdev->bd_disk, error_sector);
 174
 175                 if (!ret)
 176                         ret = err;
 177         }
 178
 179         return ret;
 180 }
 181
 182 static int md_flush_all(request_queue_t *q, struct gendisk *disk,
 183                          sector_t *error_sector)
 184 {
 185         mddev_t *mddev = q->queuedata;
 186
 187         return md_flush_mddev(mddev, error_sector);
 188 }
 189
 190 static int md_fail_request (request_queue_t *q, struct bio *bio)
 191 {
 192         bio_io_error(bio, bio->bi_size);
 193         return 0;
 194 }
 195
 196 static inline mddev_t *mddev_get(mddev_t *mddev)
 197 {
 198         atomic_inc(&mddev->active);
 199         return mddev;
 200 }
 201
 202 static void mddev_put(mddev_t *mddev)
 203 {
 204         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 205                 return;
 206         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 207                 list_del(&mddev->all_mddevs);
 208                 blk_put_queue(mddev->queue);
 209                 kfree(mddev);
 210         }
 211         spin_unlock(&all_mddevs_lock);
 212 }
 213
 214 static mddev_t * mddev_find(dev_t unit)
 215 {
 216         mddev_t *mddev, *new = NULL;
 217
 218  retry:
 219         spin_lock(&all_mddevs_lock);
 220         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 221                 if (mddev->unit == unit) {
 222                         mddev_get(mddev);
 223                         spin_unlock(&all_mddevs_lock);
 224                         if (new)
 225                                 kfree(new);
 226                         return mddev;
 227                 }
 228
 229         if (new) {
 230                 list_add(&new->all_mddevs, &all_mddevs);
 231                 spin_unlock(&all_mddevs_lock);
 232                 return new;
 233         }
 234         spin_unlock(&all_mddevs_lock);
 235
 236         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 237         if (!new)
 238                 return NULL;
 239
 240         memset(new, 0, sizeof(*new));
 241
 242         new->unit = unit;
 243         if (MAJOR(unit) == MD_MAJOR)
 244                 new->md_minor = MINOR(unit);
 245         else
 246                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 247
 248         init_MUTEX(&new->reconfig_sem);
 249         INIT_LIST_HEAD(&new->disks);
 250         INIT_LIST_HEAD(&new->all_mddevs);
 251         init_timer(&new->safemode_timer);
 252         atomic_set(&new->active, 1);
 253
 254         new->queue = blk_alloc_queue(GFP_KERNEL);
 255         if (!new->queue) {
 256                 kfree(new);
 257                 return NULL;
 258         }
 259
 260         blk_queue_make_request(new->queue, md_fail_request);
 261
 262         goto retry;
 263 }
 264
 265 static inline int mddev_lock(mddev_t * mddev)
 266 {
 267         return down_interruptible(&mddev->reconfig_sem);
 268 }
 269
 270 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 271 {
 272         down(&mddev->reconfig_sem);
 273 }
 274
 275 static inline int mddev_trylock(mddev_t * mddev)
 276 {
 277         return down_trylock(&mddev->reconfig_sem);
 278 }
 279
 280 static inline void mddev_unlock(mddev_t * mddev)
 281 {
 282         up(&mddev->reconfig_sem);
 283
 284         if (mddev->thread)
 285                 md_wakeup_thread(mddev->thread);
 286 }
 287
 288 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 289 {
 290         mdk_rdev_t * rdev;
 291         struct list_head *tmp;
 292
 293         ITERATE_RDEV(mddev,rdev,tmp) {
 294                 if (rdev->desc_nr == nr)
 295                         return rdev;
 296         }
 297         return NULL;
 298 }
 299
 300 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 301 {
 302         struct list_head *tmp;
 303         mdk_rdev_t *rdev;
 304
 305         ITERATE_RDEV(mddev,rdev,tmp) {
 306                 if (rdev->bdev->bd_dev == dev)
 307                         return rdev;
 308         }
 309         return NULL;
 310 }
 311
 312 inline static sector_t calc_dev_sboffset(struct block_device *bdev)
 313 {
 314         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 315         return MD_NEW_SIZE_BLOCKS(size);
 316 }
 317
 318 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 319 {
 320         sector_t size;
 321
 322         size = rdev->sb_offset;
 323
 324         if (chunk_size)
 325                 size &= ~((sector_t)chunk_size/1024 - 1);
 326         return size;
 327 }
 328
 329 static int alloc_disk_sb(mdk_rdev_t * rdev)
 330 {
 331         if (rdev->sb_page)
 332                 MD_BUG();
 333
 334         rdev->sb_page = alloc_page(GFP_KERNEL);
 335         if (!rdev->sb_page) {
 336                 printk(KERN_ALERT "md: out of memory.\n");
 337                 return -EINVAL;
 338         }
 339
 340         return 0;
 341 }
 342
 343 static void free_disk_sb(mdk_rdev_t * rdev)
 344 {
 345         if (rdev->sb_page) {
 346                 page_cache_release(rdev->sb_page);
 347                 rdev->sb_loaded = 0;
 348                 rdev->sb_page = NULL;
 349                 rdev->sb_offset = 0;
 350                 rdev->size = 0;
 351         }
 352 }
 353
 354
 355 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 356 {
 357         if (bio->bi_size)
 358                 return 1;
 359
 360         complete((struct completion*)bio->bi_private);
 361         return 0;
 362 }
 363
 364 static int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 365                    struct page *page, int rw)
 366 {
 367         struct bio bio;
 368         struct bio_vec vec;
 369         struct completion event;
 370
 371         rw |= (1 << BIO_RW_SYNC);
 372
 373         bio_init(&bio);
 374         bio.bi_io_vec = &vec;
 375         vec.bv_page = page;
 376         vec.bv_len = size;
 377         vec.bv_offset = 0;
 378         bio.bi_vcnt = 1;
 379         bio.bi_idx = 0;
 380         bio.bi_size = size;
 381         bio.bi_bdev = bdev;
 382         bio.bi_sector = sector;
 383         init_completion(&event);
 384         bio.bi_private = &event;
 385         bio.bi_end_io = bi_complete;
 386         submit_bio(rw, &bio);
 387         wait_for_completion(&event);
 388
 389         return test_bit(BIO_UPTODATE, &bio.bi_flags);
 390 }
 391
 392 static int read_disk_sb(mdk_rdev_t * rdev)
 393 {
 394         char b[BDEVNAME_SIZE];
 395         if (!rdev->sb_page) {
 396                 MD_BUG();
 397                 return -EINVAL;
 398         }
 399         if (rdev->sb_loaded)
 400                 return 0;
 401
 402
 403         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
 404                 goto fail;
 405         rdev->sb_loaded = 1;
 406         return 0;
 407
 408 fail:
 409         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 410                 bdevname(rdev->bdev,b));
 411         return -EINVAL;
 412 }
 413
 414 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 415 {
 416         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 417                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 418                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 419                 (sb1->set_uuid3 == sb2->set_uuid3))
 420
 421                 return 1;
 422
 423         return 0;
 424 }
 425
 426
 427 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 428 {
 429         int ret;
 430         mdp_super_t *tmp1, *tmp2;
 431
 432         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 433         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 434
 435         if (!tmp1 || !tmp2) {
 436                 ret = 0;
 437                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 438                 goto abort;
 439         }
 440
 441         *tmp1 = *sb1;
 442         *tmp2 = *sb2;
 443
 444         /*
 445          * nr_disks is not constant
 446          */
 447         tmp1->nr_disks = 0;
 448         tmp2->nr_disks = 0;
 449
 450         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 451                 ret = 0;
 452         else
 453                 ret = 1;
 454
 455 abort:
 456         if (tmp1)
 457                 kfree(tmp1);
 458         if (tmp2)
 459                 kfree(tmp2);
 460
 461         return ret;
 462 }
 463
 464 static unsigned int calc_sb_csum(mdp_super_t * sb)
 465 {
 466         unsigned int disk_csum, csum;
 467
 468         disk_csum = sb->sb_csum;
 469         sb->sb_csum = 0;
 470         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 471         sb->sb_csum = disk_csum;
 472         return csum;
 473 }
 474
 475 /* csum_partial is not consistent between different architectures.
 476  * Some (i386) do a 32bit csum.  Some (alpha) do 16 bit.
 477  * This makes it hard for user-space to know what to do.
 478  * So we use calc_sb_csum to set the checksum to allow working
 479  * with older kernels, but allow calc_sb_csum_common to
 480  * be used when checking if a checksum is correct, to
 481  * make life easier for user-space tools that might write
 482  * a superblock.
 483  */
 484 static unsigned int calc_sb_csum_common(mdp_super_t *super)
 485 {
 486         unsigned int  disk_csum = super->sb_csum;
 487         unsigned long long newcsum = 0;
 488         unsigned int csum;
 489         int i;
 490         unsigned int *superc = (int*) super;
 491         super->sb_csum = 0;
 492
 493         for (i=0; i<MD_SB_BYTES/4; i++)
 494                 newcsum+= superc[i];
 495         csum = (newcsum& 0xffffffff) + (newcsum>>32);
 496         super->sb_csum = disk_csum;
 497         return csum;
 498 }
 499
 500 /*
 501  * Handle superblock details.
 502  * We want to be able to handle multiple superblock formats
 503  * so we have a common interface to them all, and an array of
 504  * different handlers.
 505  * We rely on user-space to write the initial superblock, and support
 506  * reading and updating of superblocks.
 507  * Interface methods are:
 508  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 509  *      loads and validates a superblock on dev.
 510  *      if refdev != NULL, compare superblocks on both devices
 511  *    Return:
 512  *      0 - dev has a superblock that is compatible with refdev
 513  *      1 - dev has a superblock that is compatible and newer than refdev
 514  *          so dev should be used as the refdev in future
 515  *     -EINVAL superblock incompatible or invalid
 516  *     -othererror e.g. -EIO
 517  *
 518  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 519  *      Verify that dev is acceptable into mddev.
 520  *       The first time, mddev->raid_disks will be 0, and data from
 521  *       dev should be merged in.  Subsequent calls check that dev
 522  *       is new enough.  Return 0 or -EINVAL
 523  *
 524  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 525  *     Update the superblock for rdev with data in mddev
 526  *     This does not write to disc.
 527  *
 528  */
 529
 530 struct super_type  {
 531         char            *name;
 532         struct module   *owner;
 533         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 534         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 535         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 536 };
 537
 538 /*
 539  * load_super for 0.90.0
 540  */
 541 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 542 {
 543         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 544         mdp_super_t *sb;
 545         int ret;
 546         sector_t sb_offset;
 547
 548         /*
 549          * Calculate the position of the superblock,
 550          * it's at the end of the disk.
 551          *
 552          * It also happens to be a multiple of 4Kb.
 553          */
 554         sb_offset = calc_dev_sboffset(rdev->bdev);
 555         rdev->sb_offset = sb_offset;
 556
 557         ret = read_disk_sb(rdev);
 558         if (ret) return ret;
 559
 560         ret = -EINVAL;
 561
 562         bdevname(rdev->bdev, b);
 563         sb = (mdp_super_t*)page_address(rdev->sb_page);
 564
 565         if (sb->md_magic != MD_SB_MAGIC) {
 566                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 567                        b);
 568                 goto abort;
 569         }
 570
 571         if (sb->major_version != 0 ||
 572             sb->minor_version != 90) {
 573                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 574                         sb->major_version, sb->minor_version,
 575                         b);
 576                 goto abort;
 577         }
 578
 579         if (sb->raid_disks <= 0)
 580                 goto abort;
 581
 582         if (calc_sb_csum(sb) != sb->sb_csum &&
 583                 calc_sb_csum_common(sb) != sb->sb_csum) {
 584                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 585                         b);
 586                 goto abort;
 587         }
 588
 589         rdev->preferred_minor = sb->md_minor;
 590         rdev->data_offset = 0;
 591
 592         if (sb->level == MULTIPATH)
 593                 rdev->desc_nr = -1;
 594         else
 595                 rdev->desc_nr = sb->this_disk.number;
 596
 597         if (refdev == 0)
 598                 ret = 1;
 599         else {
 600                 __u64 ev1, ev2;
 601                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 602                 if (!uuid_equal(refsb, sb)) {
 603                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 604                                 b, bdevname(refdev->bdev,b2));
 605                         goto abort;
 606                 }
 607                 if (!sb_equal(refsb, sb)) {
 608                         printk(KERN_WARNING "md: %s has same UUID"
 609                                " but different superblock to %s\n",
 610                                b, bdevname(refdev->bdev, b2));
 611                         goto abort;
 612                 }
 613                 ev1 = md_event(sb);
 614                 ev2 = md_event(refsb);
 615                 if (ev1 > ev2)
 616                         ret = 1;
 617                 else
 618                         ret = 0;
 619         }
 620         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 621
 622  abort:
 623         return ret;
 624 }
 625
 626 /*
 627  * validate_super for 0.90.0
 628  */
 629 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 630 {
 631         mdp_disk_t *desc;
 632         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 633
 634         if (mddev->raid_disks == 0) {
 635                 mddev->major_version = 0;
 636                 mddev->minor_version = sb->minor_version;
 637                 mddev->patch_version = sb->patch_version;
 638                 mddev->persistent = ! sb->not_persistent;
 639                 mddev->chunk_size = sb->chunk_size;
 640                 mddev->ctime = sb->ctime;
 641                 mddev->utime = sb->utime;
 642                 mddev->level = sb->level;
 643                 mddev->layout = sb->layout;
 644                 mddev->raid_disks = sb->raid_disks;
 645                 mddev->size = sb->size;
 646                 mddev->events = md_event(sb);
 647
 648                 if (sb->state & (1<<MD_SB_CLEAN))
 649                         mddev->recovery_cp = MaxSector;
 650                 else {
 651                         if (sb->events_hi == sb->cp_events_hi &&
 652                                 sb->events_lo == sb->cp_events_lo) {
 653                                 mddev->recovery_cp = sb->recovery_cp;
 654                         } else
 655                                 mddev->recovery_cp = 0;
 656                 }
 657
 658                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 659                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 660                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 661                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 662
 663                 mddev->max_disks = MD_SB_DISKS;
 664         } else {
 665                 __u64 ev1;
 666                 ev1 = md_event(sb);
 667                 ++ev1;
 668                 if (ev1 < mddev->events)
 669                         return -EINVAL;
 670         }
 671         if (mddev->level != LEVEL_MULTIPATH) {
 672                 rdev->raid_disk = -1;
 673                 rdev->in_sync = rdev->faulty = 0;
 674                 desc = sb->disks + rdev->desc_nr;
 675
 676                 if (desc->state & (1<<MD_DISK_FAULTY))
 677                         rdev->faulty = 1;
 678                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 679                          desc->raid_disk < mddev->raid_disks) {
 680                         rdev->in_sync = 1;
 681                         rdev->raid_disk = desc->raid_disk;
 682                 }
 683         }
 684         return 0;
 685 }
 686
 687 /*
 688  * sync_super for 0.90.0
 689  */
 690 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 691 {
 692         mdp_super_t *sb;
 693         struct list_head *tmp;
 694         mdk_rdev_t *rdev2;
 695         int next_spare = mddev->raid_disks;
 696
 697         /* make rdev->sb match mddev data..
 698          *
 699          * 1/ zero out disks
 700          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 701          * 3/ any empty disks < next_spare become removed
 702          *
 703          * disks[0] gets initialised to REMOVED because
 704          * we cannot be sure from other fields if it has
 705          * been initialised or not.
 706          */
 707         int i;
 708         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 709
 710         sb = (mdp_super_t*)page_address(rdev->sb_page);
 711
 712         memset(sb, 0, sizeof(*sb));
 713
 714         sb->md_magic = MD_SB_MAGIC;
 715         sb->major_version = mddev->major_version;
 716         sb->minor_version = mddev->minor_version;
 717         sb->patch_version = mddev->patch_version;
 718         sb->gvalid_words  = 0; /* ignored */
 719         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 720         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 721         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 722         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 723
 724         sb->ctime = mddev->ctime;
 725         sb->level = mddev->level;
 726         sb->size  = mddev->size;
 727         sb->raid_disks = mddev->raid_disks;
 728         sb->md_minor = mddev->md_minor;
 729         sb->not_persistent = !mddev->persistent;
 730         sb->utime = mddev->utime;
 731         sb->state = 0;
 732         sb->events_hi = (mddev->events>>32);
 733         sb->events_lo = (u32)mddev->events;
 734
 735         if (mddev->in_sync)
 736         {
 737                 sb->recovery_cp = mddev->recovery_cp;
 738                 sb->cp_events_hi = (mddev->events>>32);
 739                 sb->cp_events_lo = (u32)mddev->events;
 740                 if (mddev->recovery_cp == MaxSector)
 741                         sb->state = (1<< MD_SB_CLEAN);
 742         } else
 743                 sb->recovery_cp = 0;
 744
 745         sb->layout = mddev->layout;
 746         sb->chunk_size = mddev->chunk_size;
 747
 748         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 749         ITERATE_RDEV(mddev,rdev2,tmp) {
 750                 mdp_disk_t *d;
 751                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 752                         rdev2->desc_nr = rdev2->raid_disk;
 753                 else
 754                         rdev2->desc_nr = next_spare++;
 755                 d = &sb->disks[rdev2->desc_nr];
 756                 nr_disks++;
 757                 d->number = rdev2->desc_nr;
 758                 d->major = MAJOR(rdev2->bdev->bd_dev);
 759                 d->minor = MINOR(rdev2->bdev->bd_dev);
 760                 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
 761                         d->raid_disk = rdev2->raid_disk;
 762                 else
 763                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 764                 if (rdev2->faulty) {
 765                         d->state = (1<<MD_DISK_FAULTY);
 766                         failed++;
 767                 } else if (rdev2->in_sync) {
 768                         d->state = (1<<MD_DISK_ACTIVE);
 769                         d->state |= (1<<MD_DISK_SYNC);
 770                         active++;
 771                         working++;
 772                 } else {
 773                         d->state = 0;
 774                         spare++;
 775                         working++;
 776                 }
 777         }
 778
 779         /* now set the "removed" and "faulty" bits on any missing devices */
 780         for (i=0 ; i < mddev->raid_disks ; i++) {
 781                 mdp_disk_t *d = &sb->disks[i];
 782                 if (d->state == 0 && d->number == 0) {
 783                         d->number = i;
 784                         d->raid_disk = i;
 785                         d->state = (1<<MD_DISK_REMOVED);
 786                         d->state |= (1<<MD_DISK_FAULTY);
 787                         failed++;
 788                 }
 789         }
 790         sb->nr_disks = nr_disks;
 791         sb->active_disks = active;
 792         sb->working_disks = working;
 793         sb->failed_disks = failed;
 794         sb->spare_disks = spare;
 795
 796         sb->this_disk = sb->disks[rdev->desc_nr];
 797         sb->sb_csum = calc_sb_csum(sb);
 798 }
 799
 800 /*
 801  * version 1 superblock
 802  */
 803
 804 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 805 {
 806         unsigned int disk_csum, csum;
 807         unsigned long long newcsum;
 808         int size = 256 + sb->max_dev*2;
 809         unsigned int *isuper = (unsigned int*)sb;
 810         int i;
 811
 812         disk_csum = sb->sb_csum;
 813         sb->sb_csum = 0;
 814         newcsum = 0;
 815         for (i=0; size>=4; size -= 4 )
 816                 newcsum += le32_to_cpu(*isuper++);
 817
 818         if (size == 2)
 819                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 820
 821         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 822         sb->sb_csum = disk_csum;
 823         return csum;
 824 }
 825
 826 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 827 {
 828         struct mdp_superblock_1 *sb;
 829         int ret;
 830         sector_t sb_offset;
 831         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 832
 833         /*
 834          * Calculate the position of the superblock.
 835          * It is always aligned to a 4K boundary and
 836          * depeding on minor_version, it can be:
 837          * 0: At least 8K, but less than 12K, from end of device
 838          * 1: At start of device
 839          * 2: 4K from start of device.
 840          */
 841         switch(minor_version) {
 842         case 0:
 843                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 844                 sb_offset -= 8*2;
 845                 sb_offset &= ~(4*2);
 846                 /* convert from sectors to K */
 847                 sb_offset /= 2;
 848                 break;
 849         case 1:
 850                 sb_offset = 0;
 851                 break;
 852         case 2:
 853                 sb_offset = 4;
 854                 break;
 855         default:
 856                 return -EINVAL;
 857         }
 858         rdev->sb_offset = sb_offset;
 859
 860         ret = read_disk_sb(rdev);
 861         if (ret) return ret;
 862
 863
 864         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 865
 866         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 867             sb->major_version != cpu_to_le32(1) ||
 868             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 869             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 870             sb->feature_map != 0)
 871                 return -EINVAL;
 872
 873         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 874                 printk("md: invalid superblock checksum on %s\n",
 875                         bdevname(rdev->bdev,b));
 876                 return -EINVAL;
 877         }
 878         rdev->preferred_minor = 0xffff;
 879         rdev->data_offset = le64_to_cpu(sb->data_offset);
 880
 881         if (refdev == 0)
 882                 return 1;
 883         else {
 884                 __u64 ev1, ev2;
 885                 struct mdp_superblock_1 *refsb =
 886                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 887
 888                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 889                     sb->level != refsb->level ||
 890                     sb->layout != refsb->layout ||
 891                     sb->chunksize != refsb->chunksize) {
 892                         printk(KERN_WARNING "md: %s has strangely different"
 893                                 " superblock to %s\n",
 894                                 bdevname(rdev->bdev,b),
 895                                 bdevname(refdev->bdev,b2));
 896                         return -EINVAL;
 897                 }
 898                 ev1 = le64_to_cpu(sb->events);
 899                 ev2 = le64_to_cpu(refsb->events);
 900
 901                 if (ev1 > ev2)
 902                         return 1;
 903         }
 904         if (minor_version)
 905                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 906         else
 907                 rdev->size = rdev->sb_offset;
 908         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 909                 return -EINVAL;
 910         rdev->size = le64_to_cpu(sb->data_size)/2;
 911         if (le32_to_cpu(sb->chunksize))
 912                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 913         return 0;
 914 }
 915
 916 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 917 {
 918         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 919
 920         if (mddev->raid_disks == 0) {
 921                 mddev->major_version = 1;
 922                 mddev->minor_version = 0;
 923                 mddev->patch_version = 0;
 924                 mddev->persistent = 1;
 925                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 926                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 927                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 928                 mddev->level = le32_to_cpu(sb->level);
 929                 mddev->layout = le32_to_cpu(sb->layout);
 930                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 931                 mddev->size = (u32)le64_to_cpu(sb->size);
 932                 mddev->events = le64_to_cpu(sb->events);
 933
 934                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 935                 memcpy(mddev->uuid, sb->set_uuid, 16);
 936
 937                 mddev->max_disks =  (4096-256)/2;
 938         } else {
 939                 __u64 ev1;
 940                 ev1 = le64_to_cpu(sb->events);
 941                 ++ev1;
 942                 if (ev1 < mddev->events)
 943                         return -EINVAL;
 944         }
 945
 946         if (mddev->level != LEVEL_MULTIPATH) {
 947                 int role;
 948                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
 949                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
 950                 switch(role) {
 951                 case 0xffff: /* spare */
 952                         rdev->in_sync = 0;
 953                         rdev->faulty = 0;
 954                         rdev->raid_disk = -1;
 955                         break;
 956                 case 0xfffe: /* faulty */
 957                         rdev->in_sync = 0;
 958                         rdev->faulty = 1;
 959                         rdev->raid_disk = -1;
 960                         break;
 961                 default:
 962                         rdev->in_sync = 1;
 963                         rdev->faulty = 0;
 964                         rdev->raid_disk = role;
 965                         break;
 966                 }
 967         }
 968         return 0;
 969 }
 970
 971 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 972 {
 973         struct mdp_superblock_1 *sb;
 974         struct list_head *tmp;
 975         mdk_rdev_t *rdev2;
 976         int max_dev, i;
 977         /* make rdev->sb match mddev and rdev data. */
 978
 979         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 980
 981         sb->feature_map = 0;
 982         sb->pad0 = 0;
 983         memset(sb->pad1, 0, sizeof(sb->pad1));
 984         memset(sb->pad2, 0, sizeof(sb->pad2));
 985         memset(sb->pad3, 0, sizeof(sb->pad3));
 986
 987         sb->utime = cpu_to_le64((__u64)mddev->utime);
 988         sb->events = cpu_to_le64(mddev->events);
 989         if (mddev->in_sync)
 990                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
 991         else
 992                 sb->resync_offset = cpu_to_le64(0);
 993
 994         max_dev = 0;
 995         ITERATE_RDEV(mddev,rdev2,tmp)
 996                 if (rdev2->desc_nr > max_dev)
 997                         max_dev = rdev2->desc_nr;
 998
 999         sb->max_dev = max_dev;
1000         for (i=0; i<max_dev;i++)
1001                 sb->dev_roles[max_dev] = cpu_to_le16(0xfffe);
1002
1003         ITERATE_RDEV(mddev,rdev2,tmp) {
1004                 i = rdev2->desc_nr;
1005                 if (rdev2->faulty)
1006                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1007                 else if (rdev2->in_sync)
1008                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1009                 else
1010                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1011         }
1012
1013         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1014 }
1015
1016
1017 struct super_type super_types[] = {
1018         [0] = {
1019                 .name   = "0.90.0",
1020                 .owner  = THIS_MODULE,
1021                 .load_super     = super_90_load,
1022                 .validate_super = super_90_validate,
1023                 .sync_super     = super_90_sync,
1024         },
1025         [1] = {
1026                 .name   = "md-1",
1027                 .owner  = THIS_MODULE,
1028                 .load_super     = super_1_load,
1029                 .validate_super = super_1_validate,
1030                 .sync_super     = super_1_sync,
1031         },
1032 };
1033
1034 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1035 {
1036         struct list_head *tmp;
1037         mdk_rdev_t *rdev;
1038
1039         ITERATE_RDEV(mddev,rdev,tmp)
1040                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1041                         return rdev;
1042
1043         return NULL;
1044 }
1045
1046 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1047 {
1048         struct list_head *tmp;
1049         mdk_rdev_t *rdev;
1050
1051         ITERATE_RDEV(mddev1,rdev,tmp)
1052                 if (match_dev_unit(mddev2, rdev))
1053                         return 1;
1054
1055         return 0;
1056 }
1057
1058 static LIST_HEAD(pending_raid_disks);
1059
1060 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1061 {
1062         mdk_rdev_t *same_pdev;
1063         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1064
1065         if (rdev->mddev) {
1066                 MD_BUG();
1067                 return -EINVAL;
1068         }
1069         same_pdev = match_dev_unit(mddev, rdev);
1070         if (same_pdev)
1071                 printk(KERN_WARNING
1072                         "%s: WARNING: %s appears to be on the same physical"
1073                         " disk as %s. True\n     protection against single-disk"
1074                         " failure might be compromised.\n",
1075                         mdname(mddev), bdevname(rdev->bdev,b),
1076                         bdevname(same_pdev->bdev,b2));
1077
1078         /* Verify rdev->desc_nr is unique.
1079          * If it is -1, assign a free number, else
1080          * check number is not in use
1081          */
1082         if (rdev->desc_nr < 0) {
1083                 int choice = 0;
1084                 if (mddev->pers) choice = mddev->raid_disks;
1085                 while (find_rdev_nr(mddev, choice))
1086                         choice++;
1087                 rdev->desc_nr = choice;
1088         } else {
1089                 if (find_rdev_nr(mddev, rdev->desc_nr))
1090                         return -EBUSY;
1091         }
1092
1093         list_add(&rdev->same_set, &mddev->disks);
1094         rdev->mddev = mddev;
1095         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1096         return 0;
1097 }
1098
1099 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1100 {
1101         char b[BDEVNAME_SIZE];
1102         if (!rdev->mddev) {
1103                 MD_BUG();
1104                 return;
1105         }
1106         list_del_init(&rdev->same_set);
1107         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1108         rdev->mddev = NULL;
1109 }
1110
1111 /*
1112  * prevent the device from being mounted, repartitioned or
1113  * otherwise reused by a RAID array (or any other kernel
1114  * subsystem), by bd_claiming the device.
1115  */
1116 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1117 {
1118         int err = 0;
1119         struct block_device *bdev;
1120         char b[BDEVNAME_SIZE];
1121
1122         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1123         if (IS_ERR(bdev)) {
1124                 printk(KERN_ERR "md: could not open %s.\n",
1125                         __bdevname(dev, b));
1126                 return PTR_ERR(bdev);
1127         }
1128         err = bd_claim(bdev, rdev);
1129         if (err) {
1130                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1131                         bdevname(bdev, b));
1132                 blkdev_put(bdev);
1133                 return err;
1134         }
1135         rdev->bdev = bdev;
1136         return err;
1137 }
1138
1139 static void unlock_rdev(mdk_rdev_t *rdev)
1140 {
1141         struct block_device *bdev = rdev->bdev;
1142         rdev->bdev = NULL;
1143         if (!bdev)
1144                 MD_BUG();
1145         bd_release(bdev);
1146         blkdev_put(bdev);
1147 }
1148
1149 void md_autodetect_dev(dev_t dev);
1150
1151 static void export_rdev(mdk_rdev_t * rdev)
1152 {
1153         char b[BDEVNAME_SIZE];
1154         printk(KERN_INFO "md: export_rdev(%s)\n",
1155                 bdevname(rdev->bdev,b));
1156         if (rdev->mddev)
1157                 MD_BUG();
1158         free_disk_sb(rdev);
1159         list_del_init(&rdev->same_set);
1160 #ifndef MODULE
1161         md_autodetect_dev(rdev->bdev->bd_dev);
1162 #endif
1163         unlock_rdev(rdev);
1164         kfree(rdev);
1165 }
1166
1167 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1168 {
1169         unbind_rdev_from_array(rdev);
1170         export_rdev(rdev);
1171 }
1172
1173 static void export_array(mddev_t *mddev)
1174 {
1175         struct list_head *tmp;
1176         mdk_rdev_t *rdev;
1177
1178         ITERATE_RDEV(mddev,rdev,tmp) {
1179                 if (!rdev->mddev) {
1180                         MD_BUG();
1181                         continue;
1182                 }
1183                 kick_rdev_from_array(rdev);
1184         }
1185         if (!list_empty(&mddev->disks))
1186                 MD_BUG();
1187         mddev->raid_disks = 0;
1188         mddev->major_version = 0;
1189 }
1190
1191 static void print_desc(mdp_disk_t *desc)
1192 {
1193         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1194                 desc->major,desc->minor,desc->raid_disk,desc->state);
1195 }
1196
1197 static void print_sb(mdp_super_t *sb)
1198 {
1199         int i;
1200
1201         printk(KERN_INFO
1202                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1203                 sb->major_version, sb->minor_version, sb->patch_version,
1204                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1205                 sb->ctime);
1206         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1207                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1208                 sb->md_minor, sb->layout, sb->chunk_size);
1209         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1210                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1211                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1212                 sb->failed_disks, sb->spare_disks,
1213                 sb->sb_csum, (unsigned long)sb->events_lo);
1214
1215         printk(KERN_INFO);
1216         for (i = 0; i < MD_SB_DISKS; i++) {
1217                 mdp_disk_t *desc;
1218
1219                 desc = sb->disks + i;
1220                 if (desc->number || desc->major || desc->minor ||
1221                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1222                         printk("     D %2d: ", i);
1223                         print_desc(desc);
1224                 }
1225         }
1226         printk(KERN_INFO "md:     THIS: ");
1227         print_desc(&sb->this_disk);
1228
1229 }
1230
1231 static void print_rdev(mdk_rdev_t *rdev)
1232 {
1233         char b[BDEVNAME_SIZE];
1234         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1235                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1236                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1237         if (rdev->sb_loaded) {
1238                 printk(KERN_INFO "md: rdev superblock:\n");
1239                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1240         } else
1241                 printk(KERN_INFO "md: no rdev superblock!\n");
1242 }
1243
1244 void md_print_devices(void)
1245 {
1246         struct list_head *tmp, *tmp2;
1247         mdk_rdev_t *rdev;
1248         mddev_t *mddev;
1249         char b[BDEVNAME_SIZE];
1250
1251         printk("\n");
1252         printk("md:     **********************************\n");
1253         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1254         printk("md:     **********************************\n");
1255         ITERATE_MDDEV(mddev,tmp) {
1256                 printk("%s: ", mdname(mddev));
1257
1258                 ITERATE_RDEV(mddev,rdev,tmp2)
1259                         printk("<%s>", bdevname(rdev->bdev,b));
1260                 printk("\n");
1261
1262                 ITERATE_RDEV(mddev,rdev,tmp2)
1263                         print_rdev(rdev);
1264         }
1265         printk("md:     **********************************\n");
1266         printk("\n");
1267 }
1268
1269
1270 static int write_disk_sb(mdk_rdev_t * rdev)
1271 {
1272         char b[BDEVNAME_SIZE];
1273         if (!rdev->sb_loaded) {
1274                 MD_BUG();
1275                 return 1;
1276         }
1277         if (rdev->faulty) {
1278                 MD_BUG();
1279                 return 1;
1280         }
1281
1282         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1283                 bdevname(rdev->bdev,b),
1284                (unsigned long long)rdev->sb_offset);
1285
1286         if (sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, WRITE))
1287                 return 0;
1288
1289         printk("md: write_disk_sb failed for device %s\n",
1290                 bdevname(rdev->bdev,b));
1291         return 1;
1292 }
1293
1294 static void sync_sbs(mddev_t * mddev)
1295 {
1296         mdk_rdev_t *rdev;
1297         struct list_head *tmp;
1298
1299         ITERATE_RDEV(mddev,rdev,tmp) {
1300                 super_types[mddev->major_version].
1301                         sync_super(mddev, rdev);
1302                 rdev->sb_loaded = 1;
1303         }
1304 }
1305
1306 static void md_update_sb(mddev_t * mddev)
1307 {
1308         int err, count = 100;
1309         struct list_head *tmp;
1310         mdk_rdev_t *rdev;
1311
1312         mddev->sb_dirty = 0;
1313 repeat:
1314         mddev->utime = get_seconds();
1315         mddev->events ++;
1316
1317         if (!mddev->events) {
1318                 /*
1319                  * oops, this 64-bit counter should never wrap.
1320                  * Either we are in around ~1 trillion A.C., assuming
1321                  * 1 reboot per second, or we have a bug:
1322                  */
1323                 MD_BUG();
1324                 mddev->events --;
1325         }
1326         sync_sbs(mddev);
1327
1328         /*
1329          * do not write anything to disk if using
1330          * nonpersistent superblocks
1331          */
1332         if (!mddev->persistent)
1333                 return;
1334
1335         dprintk(KERN_INFO
1336                 "md: updating %s RAID superblock on device (in sync %d)\n",
1337                 mdname(mddev),mddev->in_sync);
1338
1339         err = 0;
1340         ITERATE_RDEV(mddev,rdev,tmp) {
1341                 char b[BDEVNAME_SIZE];
1342                 dprintk(KERN_INFO "md: ");
1343                 if (rdev->faulty)
1344                         dprintk("(skipping faulty ");
1345
1346                 dprintk("%s ", bdevname(rdev->bdev,b));
1347                 if (!rdev->faulty) {
1348                         err += write_disk_sb(rdev);
1349                 } else
1350                         dprintk(")\n");
1351                 if (!err && mddev->level == LEVEL_MULTIPATH)
1352                         /* only need to write one superblock... */
1353                         break;
1354         }
1355         if (err) {
1356                 if (--count) {
1357                         printk(KERN_ERR "md: errors occurred during superblock"
1358                                 " update, repeating\n");
1359                         goto repeat;
1360                 }
1361                 printk(KERN_ERR \
1362                         "md: excessive errors occurred during superblock update, exiting\n");
1363         }
1364 }
1365
1366 /*
1367  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1368  *
1369  * mark the device faulty if:
1370  *
1371  *   - the device is nonexistent (zero size)
1372  *   - the device has no valid superblock
1373  *
1374  * a faulty rdev _never_ has rdev->sb set.
1375  */
1376 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1377 {
1378         char b[BDEVNAME_SIZE];
1379         int err;
1380         mdk_rdev_t *rdev;
1381         sector_t size;
1382
1383         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1384         if (!rdev) {
1385                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1386                 return ERR_PTR(-ENOMEM);
1387         }
1388         memset(rdev, 0, sizeof(*rdev));
1389
1390         if ((err = alloc_disk_sb(rdev)))
1391                 goto abort_free;
1392
1393         err = lock_rdev(rdev, newdev);
1394         if (err)
1395                 goto abort_free;
1396
1397         rdev->desc_nr = -1;
1398         rdev->faulty = 0;
1399         rdev->in_sync = 0;
1400         rdev->data_offset = 0;
1401         atomic_set(&rdev->nr_pending, 0);
1402
1403         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1404         if (!size) {
1405                 printk(KERN_WARNING
1406                         "md: %s has zero or unknown size, marking faulty!\n",
1407                         bdevname(rdev->bdev,b));
1408                 err = -EINVAL;
1409                 goto abort_free;
1410         }
1411
1412         if (super_format >= 0) {
1413                 err = super_types[super_format].
1414                         load_super(rdev, NULL, super_minor);
1415                 if (err == -EINVAL) {
1416                         printk(KERN_WARNING
1417                                 "md: %s has invalid sb, not importing!\n",
1418                                 bdevname(rdev->bdev,b));
1419                         goto abort_free;
1420                 }
1421                 if (err < 0) {
1422                         printk(KERN_WARNING
1423                                 "md: could not read %s's sb, not importing!\n",
1424                                 bdevname(rdev->bdev,b));
1425                         goto abort_free;
1426                 }
1427         }
1428         INIT_LIST_HEAD(&rdev->same_set);
1429
1430         return rdev;
1431
1432 abort_free:
1433         if (rdev->sb_page) {
1434                 if (rdev->bdev)
1435                         unlock_rdev(rdev);
1436                 free_disk_sb(rdev);
1437         }
1438         kfree(rdev);
1439         return ERR_PTR(err);
1440 }
1441
1442 /*
1443  * Check a full RAID array for plausibility
1444  */
1445
1446
1447 static int analyze_sbs(mddev_t * mddev)
1448 {
1449         int i;
1450         struct list_head *tmp;
1451         mdk_rdev_t *rdev, *freshest;
1452         char b[BDEVNAME_SIZE];
1453
1454         freshest = NULL;
1455         ITERATE_RDEV(mddev,rdev,tmp)
1456                 switch (super_types[mddev->major_version].
1457                         load_super(rdev, freshest, mddev->minor_version)) {
1458                 case 1:
1459                         freshest = rdev;
1460                         break;
1461                 case 0:
1462                         break;
1463                 default:
1464                         printk( KERN_ERR \
1465                                 "md: fatal superblock inconsistency in %s"
1466                                 " -- removing from array\n",
1467                                 bdevname(rdev->bdev,b));
1468                         kick_rdev_from_array(rdev);
1469                 }
1470
1471
1472         super_types[mddev->major_version].
1473                 validate_super(mddev, freshest);
1474
1475         i = 0;
1476         ITERATE_RDEV(mddev,rdev,tmp) {
1477                 if (rdev != freshest)
1478                         if (super_types[mddev->major_version].
1479                             validate_super(mddev, rdev)) {
1480                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1481                                         " from array!\n",
1482                                         bdevname(rdev->bdev,b));
1483                                 kick_rdev_from_array(rdev);
1484                                 continue;
1485                         }
1486                 if (mddev->level == LEVEL_MULTIPATH) {
1487                         rdev->desc_nr = i++;
1488                         rdev->raid_disk = rdev->desc_nr;
1489                         rdev->in_sync = 1;
1490                 }
1491         }
1492
1493
1494         /*
1495          * Check if we can support this RAID array
1496          */
1497         if (mddev->major_version != MD_MAJOR_VERSION ||
1498                         mddev->minor_version > MD_MINOR_VERSION) {
1499                 printk(KERN_ALERT
1500                         "md: %s: unsupported raid array version %d.%d.%d\n",
1501                         mdname(mddev), mddev->major_version,
1502                         mddev->minor_version, mddev->patch_version);
1503                 goto abort;
1504         }
1505
1506         if ((mddev->recovery_cp != MaxSector) &&
1507             ((mddev->level == 1) ||
1508              ((mddev->level >= 4) && (mddev->level <= 6))))
1509                 printk(KERN_ERR "md: %s: raid array is not clean"
1510                        " -- starting background reconstruction\n",
1511                        mdname(mddev));
1512
1513         return 0;
1514 abort:
1515         return 1;
1516 }
1517
1518 int mdp_major = 0;
1519
1520 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1521 {
1522         static DECLARE_MUTEX(disks_sem);
1523         mddev_t *mddev = mddev_find(dev);
1524         struct gendisk *disk;
1525         int partitioned = (MAJOR(dev) != MD_MAJOR);
1526         int shift = partitioned ? MdpMinorShift : 0;
1527         int unit = MINOR(dev) >> shift;
1528
1529         if (!mddev)
1530                 return NULL;
1531
1532         down(&disks_sem);
1533         if (mddev->gendisk) {
1534                 up(&disks_sem);
1535                 mddev_put(mddev);
1536                 return NULL;
1537         }
1538         disk = alloc_disk(1 << shift);
1539         if (!disk) {
1540                 up(&disks_sem);
1541                 mddev_put(mddev);
1542                 return NULL;
1543         }
1544         disk->major = MAJOR(dev);
1545         disk->first_minor = unit << shift;
1546         if (partitioned)
1547                 sprintf(disk->disk_name, "md_d%d", unit);
1548         else
1549                 sprintf(disk->disk_name, "md%d", unit);
1550         disk->fops = &md_fops;
1551         disk->private_data = mddev;
1552         disk->queue = mddev->queue;
1553         add_disk(disk);
1554         mddev->gendisk = disk;
1555         up(&disks_sem);
1556         return NULL;
1557 }
1558
1559 void md_wakeup_thread(mdk_thread_t *thread);
1560
1561 static void md_safemode_timeout(unsigned long data)
1562 {
1563         mddev_t *mddev = (mddev_t *) data;
1564
1565         mddev->safemode = 1;
1566         md_wakeup_thread(mddev->thread);
1567 }
1568
1569
1570 static int do_md_run(mddev_t * mddev)
1571 {
1572         int pnum, err;
1573         int chunk_size;
1574         struct list_head *tmp;
1575         mdk_rdev_t *rdev;
1576         struct gendisk *disk;
1577         char b[BDEVNAME_SIZE];
1578
1579         if (list_empty(&mddev->disks)) {
1580                 MD_BUG();
1581                 return -EINVAL;
1582         }
1583
1584         if (mddev->pers)
1585                 return -EBUSY;
1586
1587         /*
1588          * Analyze all RAID superblock(s)
1589          */
1590         if (!mddev->raid_disks && analyze_sbs(mddev)) {
1591                 MD_BUG();
1592                 return -EINVAL;
1593         }
1594
1595         chunk_size = mddev->chunk_size;
1596         pnum = level_to_pers(mddev->level);
1597
1598         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1599                 if (!chunk_size) {
1600                         /*
1601                          * 'default chunksize' in the old md code used to
1602                          * be PAGE_SIZE, baaad.
1603                          * we abort here to be on the safe side. We don't
1604                          * want to continue the bad practice.
1605                          */
1606                         printk(KERN_ERR
1607                                 "no chunksize specified, see 'man raidtab'\n");
1608                         return -EINVAL;
1609                 }
1610                 if (chunk_size > MAX_CHUNK_SIZE) {
1611                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1612                                 chunk_size, MAX_CHUNK_SIZE);
1613                         return -EINVAL;
1614                 }
1615                 /*
1616                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1617                  */
1618                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1619                         MD_BUG();
1620                         return -EINVAL;
1621                 }
1622                 if (chunk_size < PAGE_SIZE) {
1623                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1624                                 chunk_size, PAGE_SIZE);
1625                         return -EINVAL;
1626                 }
1627
1628                 /* devices must have minimum size of one chunk */
1629                 ITERATE_RDEV(mddev,rdev,tmp) {
1630                         if (rdev->faulty)
1631                                 continue;
1632                         if (rdev->size < chunk_size / 1024) {
1633                                 printk(KERN_WARNING
1634                                         "md: Dev %s smaller than chunk_size:"
1635                                         " %lluk < %dk\n",
1636                                         bdevname(rdev->bdev,b),
1637                                         (unsigned long long)rdev->size,
1638                                         chunk_size / 1024);
1639                                 return -EINVAL;
1640                         }
1641                 }
1642         }
1643
1644         if (pnum >= MAX_PERSONALITY) {
1645                 MD_BUG();
1646                 return -EINVAL;
1647         }
1648
1649 #ifdef CONFIG_KMOD
1650         if (!pers[pnum])
1651         {
1652                 request_module("md-personality-%d", pnum);
1653         }
1654 #endif
1655
1656         /*
1657          * Drop all container device buffers, from now on
1658          * the only valid external interface is through the md
1659          * device.
1660          * Also find largest hardsector size
1661          */
1662         ITERATE_RDEV(mddev,rdev,tmp) {
1663                 if (rdev->faulty)
1664                         continue;
1665                 sync_blockdev(rdev->bdev);
1666                 invalidate_bdev(rdev->bdev, 0);
1667         }
1668
1669         md_probe(mddev->unit, NULL, NULL);
1670         disk = mddev->gendisk;
1671         if (!disk)
1672                 return -ENOMEM;
1673
1674         spin_lock(&pers_lock);
1675         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1676                 spin_unlock(&pers_lock);
1677                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1678                        pnum);
1679                 return -EINVAL;
1680         }
1681
1682         mddev->pers = pers[pnum];
1683         spin_unlock(&pers_lock);
1684
1685         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1686
1687         err = mddev->pers->run(mddev);
1688         if (err) {
1689                 printk(KERN_ERR "md: pers->run() failed ...\n");
1690                 module_put(mddev->pers->owner);
1691                 mddev->pers = NULL;
1692                 return -EINVAL;
1693         }
1694         atomic_set(&mddev->writes_pending,0);
1695         mddev->safemode = 0;
1696         mddev->safemode_timer.function = md_safemode_timeout;
1697         mddev->safemode_timer.data = (unsigned long) mddev;
1698         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1699         mddev->in_sync = 1;
1700
1701         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1702
1703         if (mddev->sb_dirty)
1704                 md_update_sb(mddev);
1705
1706         set_capacity(disk, mddev->array_size<<1);
1707
1708         /* If we call blk_queue_make_request here, it will
1709          * re-initialise max_sectors etc which may have been
1710          * refined inside -> run.  So just set the bits we need to set.
1711          * Most initialisation happended when we called
1712          * blk_queue_make_request(..., md_fail_request)
1713          * earlier.
1714          */
1715         mddev->queue->queuedata = mddev;
1716         mddev->queue->make_request_fn = mddev->pers->make_request;
1717         mddev->queue->issue_flush_fn = md_flush_all;
1718
1719         mddev->changed = 1;
1720         return 0;
1721 }
1722
1723 static int restart_array(mddev_t *mddev)
1724 {
1725         struct gendisk *disk = mddev->gendisk;
1726         int err;
1727
1728         /*
1729          * Complain if it has no devices
1730          */
1731         err = -ENXIO;
1732         if (list_empty(&mddev->disks))
1733                 goto out;
1734
1735         if (mddev->pers) {
1736                 err = -EBUSY;
1737                 if (!mddev->ro)
1738                         goto out;
1739
1740                 mddev->safemode = 0;
1741                 mddev->ro = 0;
1742                 set_disk_ro(disk, 0);
1743
1744                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1745                         mdname(mddev));
1746                 /*
1747                  * Kick recovery or resync if necessary
1748                  */
1749                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1750                 md_wakeup_thread(mddev->thread);
1751                 err = 0;
1752         } else {
1753                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1754                         mdname(mddev));
1755                 err = -EINVAL;
1756         }
1757
1758 out:
1759         return err;
1760 }
1761
1762 static int do_md_stop(mddev_t * mddev, int ro)
1763 {
1764         int err = 0;
1765         struct gendisk *disk = mddev->gendisk;
1766
1767         if (mddev->pers) {
1768                 if (atomic_read(&mddev->active)>2) {
1769                         printk("md: %s still in use.\n",mdname(mddev));
1770                         return -EBUSY;
1771                 }
1772
1773                 if (mddev->sync_thread) {
1774                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1775                         md_unregister_thread(mddev->sync_thread);
1776                         mddev->sync_thread = NULL;
1777                 }
1778
1779                 del_timer_sync(&mddev->safemode_timer);
1780
1781                 invalidate_partition(disk, 0);
1782
1783                 if (ro) {
1784                         err  = -ENXIO;
1785                         if (mddev->ro)
1786                                 goto out;
1787                         mddev->ro = 1;
1788                 } else {
1789                         if (mddev->ro)
1790                                 set_disk_ro(disk, 0);
1791                         blk_queue_make_request(mddev->queue, md_fail_request);
1792                         mddev->pers->stop(mddev);
1793                         module_put(mddev->pers->owner);
1794                         mddev->pers = NULL;
1795                         if (mddev->ro)
1796                                 mddev->ro = 0;
1797                 }
1798                 if (!mddev->in_sync) {
1799                         /* mark array as shutdown cleanly */
1800                         mddev->in_sync = 1;
1801                         md_update_sb(mddev);
1802                 }
1803                 if (ro)
1804                         set_disk_ro(disk, 1);
1805         }
1806         /*
1807          * Free resources if final stop
1808          */
1809         if (!ro) {
1810                 struct gendisk *disk;
1811                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1812
1813                 export_array(mddev);
1814
1815                 mddev->array_size = 0;
1816                 disk = mddev->gendisk;
1817                 if (disk)
1818                         set_capacity(disk, 0);
1819                 mddev->changed = 1;
1820         } else
1821                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1822                         mdname(mddev));
1823         err = 0;
1824 out:
1825         return err;
1826 }
1827
1828 static void autorun_array(mddev_t *mddev)
1829 {
1830         mdk_rdev_t *rdev;
1831         struct list_head *tmp;
1832         int err;
1833
1834         if (list_empty(&mddev->disks)) {
1835                 MD_BUG();
1836                 return;
1837         }
1838
1839         printk(KERN_INFO "md: running: ");
1840
1841         ITERATE_RDEV(mddev,rdev,tmp) {
1842                 char b[BDEVNAME_SIZE];
1843                 printk("<%s>", bdevname(rdev->bdev,b));
1844         }
1845         printk("\n");
1846
1847         err = do_md_run (mddev);
1848         if (err) {
1849                 printk(KERN_WARNING "md :do_md_run() returned %d\n", err);
1850                 do_md_stop (mddev, 0);
1851         }
1852 }
1853
1854 /*
1855  * lets try to run arrays based on all disks that have arrived
1856  * until now. (those are in pending_raid_disks)
1857  *
1858  * the method: pick the first pending disk, collect all disks with
1859  * the same UUID, remove all from the pending list and put them into
1860  * the 'same_array' list. Then order this list based on superblock
1861  * update time (freshest comes first), kick out 'old' disks and
1862  * compare superblocks. If everything's fine then run it.
1863  *
1864  * If "unit" is allocated, then bump its reference count
1865  */
1866 static void autorun_devices(int part)
1867 {
1868         struct list_head candidates;
1869         struct list_head *tmp;
1870         mdk_rdev_t *rdev0, *rdev;
1871         mddev_t *mddev;
1872         char b[BDEVNAME_SIZE];
1873
1874         printk(KERN_INFO "md: autorun ...\n");
1875         while (!list_empty(&pending_raid_disks)) {
1876                 dev_t dev;
1877                 rdev0 = list_entry(pending_raid_disks.next,
1878                                          mdk_rdev_t, same_set);
1879
1880                 printk(KERN_INFO "md: considering %s ...\n",
1881                         bdevname(rdev0->bdev,b));
1882                 INIT_LIST_HEAD(&candidates);
1883                 ITERATE_RDEV_PENDING(rdev,tmp)
1884                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1885                                 printk(KERN_INFO "md:  adding %s ...\n",
1886                                         bdevname(rdev->bdev,b));
1887                                 list_move(&rdev->same_set, &candidates);
1888                         }
1889                 /*
1890                  * now we have a set of devices, with all of them having
1891                  * mostly sane superblocks. It's time to allocate the
1892                  * mddev.
1893                  */
1894                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1895                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1896                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1897                         break;
1898                 }
1899                 if (part)
1900                         dev = MKDEV(mdp_major,
1901                                     rdev0->preferred_minor << MdpMinorShift);
1902                 else
1903                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1904
1905                 md_probe(dev, NULL, NULL);
1906                 mddev = mddev_find(dev);
1907                 if (!mddev) {
1908                         printk(KERN_ERR
1909                                 "md: cannot allocate memory for md drive.\n");
1910                         break;
1911                 }
1912                 if (mddev_lock(mddev))
1913                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1914                                mdname(mddev));
1915                 else if (mddev->raid_disks || mddev->major_version
1916                          || !list_empty(&mddev->disks)) {
1917                         printk(KERN_WARNING
1918                                 "md: %s already running, cannot run %s\n",
1919                                 mdname(mddev), bdevname(rdev0->bdev,b));
1920                         mddev_unlock(mddev);
1921                 } else {
1922                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1923                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1924                                 list_del_init(&rdev->same_set);
1925                                 if (bind_rdev_to_array(rdev, mddev))
1926                                         export_rdev(rdev);
1927                         }
1928                         autorun_array(mddev);
1929                         mddev_unlock(mddev);
1930                 }
1931                 /* on success, candidates will be empty, on error
1932                  * it won't...
1933                  */
1934                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1935                         export_rdev(rdev);
1936                 mddev_put(mddev);
1937         }
1938         printk(KERN_INFO "md: ... autorun DONE.\n");
1939 }
1940
1941 /*
1942  * import RAID devices based on one partition
1943  * if possible, the array gets run as well.
1944  */
1945
1946 static int autostart_array(dev_t startdev)
1947 {
1948         char b[BDEVNAME_SIZE];
1949         int err = -EINVAL, i;
1950         mdp_super_t *sb = NULL;
1951         mdk_rdev_t *start_rdev = NULL, *rdev;
1952
1953         start_rdev = md_import_device(startdev, 0, 0);
1954         if (IS_ERR(start_rdev))
1955                 return err;
1956
1957
1958         /* NOTE: this can only work for 0.90.0 superblocks */
1959         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1960         if (sb->major_version != 0 ||
1961             sb->minor_version != 90 ) {
1962                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1963                 export_rdev(start_rdev);
1964                 return err;
1965         }
1966
1967         if (start_rdev->faulty) {
1968                 printk(KERN_WARNING
1969                         "md: can not autostart based on faulty %s!\n",
1970                         bdevname(start_rdev->bdev,b));
1971                 export_rdev(start_rdev);
1972                 return err;
1973         }
1974         list_add(&start_rdev->same_set, &pending_raid_disks);
1975
1976         for (i = 0; i < MD_SB_DISKS; i++) {
1977                 mdp_disk_t *desc = sb->disks + i;
1978                 dev_t dev = MKDEV(desc->major, desc->minor);
1979
1980                 if (!dev)
1981                         continue;
1982                 if (dev == startdev)
1983                         continue;
1984                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
1985                         continue;
1986                 rdev = md_import_device(dev, 0, 0);
1987                 if (IS_ERR(rdev))
1988                         continue;
1989
1990                 list_add(&rdev->same_set, &pending_raid_disks);
1991         }
1992
1993         /*
1994          * possibly return codes
1995          */
1996         autorun_devices(0);
1997         return 0;
1998
1999 }
2000
2001
2002 static int get_version(void __user * arg)
2003 {
2004         mdu_version_t ver;
2005
2006         ver.major = MD_MAJOR_VERSION;
2007         ver.minor = MD_MINOR_VERSION;
2008         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2009
2010         if (copy_to_user(arg, &ver, sizeof(ver)))
2011                 return -EFAULT;
2012
2013         return 0;
2014 }
2015
2016 static int get_array_info(mddev_t * mddev, void __user * arg)
2017 {
2018         mdu_array_info_t info;
2019         int nr,working,active,failed,spare;
2020         mdk_rdev_t *rdev;
2021         struct list_head *tmp;
2022
2023         nr=working=active=failed=spare=0;
2024         ITERATE_RDEV(mddev,rdev,tmp) {
2025                 nr++;
2026                 if (rdev->faulty)
2027                         failed++;
2028                 else {
2029                         working++;
2030                         if (rdev->in_sync)
2031                                 active++;
2032                         else
2033                                 spare++;
2034                 }
2035         }
2036
2037         info.major_version = mddev->major_version;
2038         info.minor_version = mddev->minor_version;
2039         info.patch_version = 1;
2040         info.ctime         = mddev->ctime;
2041         info.level         = mddev->level;
2042         info.size          = mddev->size;
2043         info.nr_disks      = nr;
2044         info.raid_disks    = mddev->raid_disks;
2045         info.md_minor      = mddev->md_minor;
2046         info.not_persistent= !mddev->persistent;
2047
2048         info.utime         = mddev->utime;
2049         info.state         = 0;
2050         if (mddev->in_sync)
2051                 info.state = (1<<MD_SB_CLEAN);
2052         info.active_disks  = active;
2053         info.working_disks = working;
2054         info.failed_disks  = failed;
2055         info.spare_disks   = spare;
2056
2057         info.layout        = mddev->layout;
2058         info.chunk_size    = mddev->chunk_size;
2059
2060         if (copy_to_user(arg, &info, sizeof(info)))
2061                 return -EFAULT;
2062
2063         return 0;
2064 }
2065
2066 static int get_disk_info(mddev_t * mddev, void __user * arg)
2067 {
2068         mdu_disk_info_t info;
2069         unsigned int nr;
2070         mdk_rdev_t *rdev;
2071
2072         if (copy_from_user(&info, arg, sizeof(info)))
2073                 return -EFAULT;
2074
2075         nr = info.number;
2076
2077         rdev = find_rdev_nr(mddev, nr);
2078         if (rdev) {
2079                 info.major = MAJOR(rdev->bdev->bd_dev);
2080                 info.minor = MINOR(rdev->bdev->bd_dev);
2081                 info.raid_disk = rdev->raid_disk;
2082                 info.state = 0;
2083                 if (rdev->faulty)
2084                         info.state |= (1<<MD_DISK_FAULTY);
2085                 else if (rdev->in_sync) {
2086                         info.state |= (1<<MD_DISK_ACTIVE);
2087                         info.state |= (1<<MD_DISK_SYNC);
2088                 }
2089         } else {
2090                 info.major = info.minor = 0;
2091                 info.raid_disk = -1;
2092                 info.state = (1<<MD_DISK_REMOVED);
2093         }
2094
2095         if (copy_to_user(arg, &info, sizeof(info)))
2096                 return -EFAULT;
2097
2098         return 0;
2099 }
2100
2101 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2102 {
2103         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2104         mdk_rdev_t *rdev;
2105         dev_t dev = MKDEV(info->major,info->minor);
2106
2107         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2108                 return -EOVERFLOW;
2109
2110         if (!mddev->raid_disks) {
2111                 int err;
2112                 /* expecting a device which has a superblock */
2113                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2114                 if (IS_ERR(rdev)) {
2115                         printk(KERN_WARNING
2116                                 "md: md_import_device returned %ld\n",
2117                                 PTR_ERR(rdev));
2118                         return PTR_ERR(rdev);
2119                 }
2120                 if (!list_empty(&mddev->disks)) {
2121                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2122                                                         mdk_rdev_t, same_set);
2123                         int err = super_types[mddev->major_version]
2124                                 .load_super(rdev, rdev0, mddev->minor_version);
2125                         if (err < 0) {
2126                                 printk(KERN_WARNING
2127                                         "md: %s has different UUID to %s\n",
2128                                         bdevname(rdev->bdev,b),
2129                                         bdevname(rdev0->bdev,b2));
2130                                 export_rdev(rdev);
2131                                 return -EINVAL;
2132                         }
2133                 }
2134                 err = bind_rdev_to_array(rdev, mddev);
2135                 if (err)
2136                         export_rdev(rdev);
2137                 return err;
2138         }
2139
2140         /*
2141          * add_new_disk can be used once the array is assembled
2142          * to add "hot spares".  They must already have a superblock
2143          * written
2144          */
2145         if (mddev->pers) {
2146                 int err;
2147                 if (!mddev->pers->hot_add_disk) {
2148                         printk(KERN_WARNING
2149                                 "%s: personality does not support diskops!\n",
2150                                mdname(mddev));
2151                         return -EINVAL;
2152                 }
2153                 rdev = md_import_device(dev, mddev->major_version,
2154                                         mddev->minor_version);
2155                 if (IS_ERR(rdev)) {
2156                         printk(KERN_WARNING
2157                                 "md: md_import_device returned %ld\n",
2158                                 PTR_ERR(rdev));
2159                         return PTR_ERR(rdev);
2160                 }
2161                 rdev->in_sync = 0; /* just to be sure */
2162                 rdev->raid_disk = -1;
2163                 err = bind_rdev_to_array(rdev, mddev);
2164                 if (err)
2165                         export_rdev(rdev);
2166                 if (mddev->thread)
2167                         md_wakeup_thread(mddev->thread);
2168                 return err;
2169         }
2170
2171         /* otherwise, add_new_disk is only allowed
2172          * for major_version==0 superblocks
2173          */
2174         if (mddev->major_version != 0) {
2175                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2176                        mdname(mddev));
2177                 return -EINVAL;
2178         }
2179
2180         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2181                 int err;
2182                 rdev = md_import_device (dev, -1, 0);
2183                 if (IS_ERR(rdev)) {
2184                         printk(KERN_WARNING
2185                                 "md: error, md_import_device() returned %ld\n",
2186                                 PTR_ERR(rdev));
2187                         return PTR_ERR(rdev);
2188                 }
2189                 rdev->desc_nr = info->number;
2190                 if (info->raid_disk < mddev->raid_disks)
2191                         rdev->raid_disk = info->raid_disk;
2192                 else
2193                         rdev->raid_disk = -1;
2194
2195                 rdev->faulty = 0;
2196                 if (rdev->raid_disk < mddev->raid_disks)
2197                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2198                 else
2199                         rdev->in_sync = 0;
2200
2201                 err = bind_rdev_to_array(rdev, mddev);
2202                 if (err) {
2203                         export_rdev(rdev);
2204                         return err;
2205                 }
2206
2207                 if (!mddev->persistent) {
2208                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2209                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2210                 } else
2211                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2212                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2213
2214                 if (!mddev->size || (mddev->size > rdev->size))
2215                         mddev->size = rdev->size;
2216         }
2217
2218         return 0;
2219 }
2220
2221 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2222 {
2223         char b[BDEVNAME_SIZE];
2224         mdk_rdev_t *rdev;
2225
2226         if (!mddev->pers)
2227                 return -ENODEV;
2228
2229         rdev = find_rdev(mddev, dev);
2230         if (!rdev)
2231                 return -ENXIO;
2232
2233         if (rdev->raid_disk >= 0)
2234                 goto busy;
2235
2236         kick_rdev_from_array(rdev);
2237         md_update_sb(mddev);
2238
2239         return 0;
2240 busy:
2241         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2242                 bdevname(rdev->bdev,b), mdname(mddev));
2243         return -EBUSY;
2244 }
2245
2246 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2247 {
2248         char b[BDEVNAME_SIZE];
2249         int err;
2250         unsigned int size;
2251         mdk_rdev_t *rdev;
2252
2253         if (!mddev->pers)
2254                 return -ENODEV;
2255
2256         if (mddev->major_version != 0) {
2257                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2258                         " version-0 superblocks.\n",
2259                         mdname(mddev));
2260                 return -EINVAL;
2261         }
2262         if (!mddev->pers->hot_add_disk) {
2263                 printk(KERN_WARNING
2264                         "%s: personality does not support diskops!\n",
2265                         mdname(mddev));
2266                 return -EINVAL;
2267         }
2268
2269         rdev = md_import_device (dev, -1, 0);
2270         if (IS_ERR(rdev)) {
2271                 printk(KERN_WARNING
2272                         "md: error, md_import_device() returned %ld\n",
2273                         PTR_ERR(rdev));
2274                 return -EINVAL;
2275         }
2276
2277         if (mddev->persistent)
2278                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2279         else
2280                 rdev->sb_offset =
2281                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2282
2283         size = calc_dev_size(rdev, mddev->chunk_size);
2284         rdev->size = size;
2285
2286         if (size < mddev->size) {
2287                 printk(KERN_WARNING
2288                         "%s: disk size %llu blocks < array size %llu\n",
2289                         mdname(mddev), (unsigned long long)size,
2290                         (unsigned long long)mddev->size);
2291                 err = -ENOSPC;
2292                 goto abort_export;
2293         }
2294
2295         if (rdev->faulty) {
2296                 printk(KERN_WARNING
2297                         "md: can not hot-add faulty %s disk to %s!\n",
2298                         bdevname(rdev->bdev,b), mdname(mddev));
2299                 err = -EINVAL;
2300                 goto abort_export;
2301         }
2302         rdev->in_sync = 0;
2303         rdev->desc_nr = -1;
2304         bind_rdev_to_array(rdev, mddev);
2305
2306         /*
2307          * The rest should better be atomic, we can have disk failures
2308          * noticed in interrupt contexts ...
2309          */
2310
2311         if (rdev->desc_nr == mddev->max_disks) {
2312                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2313                         mdname(mddev));
2314                 err = -EBUSY;
2315                 goto abort_unbind_export;
2316         }
2317
2318         rdev->raid_disk = -1;
2319
2320         md_update_sb(mddev);
2321
2322         /*
2323          * Kick recovery, maybe this spare has to be added to the
2324          * array immediately.
2325          */
2326         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2327         md_wakeup_thread(mddev->thread);
2328
2329         return 0;
2330
2331 abort_unbind_export:
2332         unbind_rdev_from_array(rdev);
2333
2334 abort_export:
2335         export_rdev(rdev);
2336         return err;
2337 }
2338
2339 /*
2340  * set_array_info is used two different ways
2341  * The original usage is when creating a new array.
2342  * In this usage, raid_disks is > 0 and it together with
2343  *  level, size, not_persistent,layout,chunksize determine the
2344  *  shape of the array.
2345  *  This will always create an array with a type-0.90.0 superblock.
2346  * The newer usage is when assembling an array.
2347  *  In this case raid_disks will be 0, and the major_version field is
2348  *  use to determine which style super-blocks are to be found on the devices.
2349  *  The minor and patch _version numbers are also kept incase the
2350  *  super_block handler wishes to interpret them.
2351  */
2352 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2353 {
2354
2355         if (info->raid_disks == 0) {
2356                 /* just setting version number for superblock loading */
2357                 if (info->major_version < 0 ||
2358                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2359                     super_types[info->major_version].name == NULL) {
2360                         /* maybe try to auto-load a module? */
2361                         printk(KERN_INFO
2362                                 "md: superblock version %d not known\n",
2363                                 info->major_version);
2364                         return -EINVAL;
2365                 }
2366                 mddev->major_version = info->major_version;
2367                 mddev->minor_version = info->minor_version;
2368                 mddev->patch_version = info->patch_version;
2369                 return 0;
2370         }
2371         mddev->major_version = MD_MAJOR_VERSION;
2372         mddev->minor_version = MD_MINOR_VERSION;
2373         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2374         mddev->ctime         = get_seconds();
2375
2376         mddev->level         = info->level;
2377         mddev->size          = info->size;
2378         mddev->raid_disks    = info->raid_disks;
2379         /* don't set md_minor, it is determined by which /dev/md* was
2380          * openned
2381          */
2382         if (info->state & (1<<MD_SB_CLEAN))
2383                 mddev->recovery_cp = MaxSector;
2384         else
2385                 mddev->recovery_cp = 0;
2386         mddev->persistent    = ! info->not_persistent;
2387
2388         mddev->layout        = info->layout;
2389         mddev->chunk_size    = info->chunk_size;
2390
2391         mddev->max_disks     = MD_SB_DISKS;
2392
2393         mddev->sb_dirty      = 1;
2394
2395         /*
2396          * Generate a 128 bit UUID
2397          */
2398         get_random_bytes(mddev->uuid, 16);
2399
2400         return 0;
2401 }
2402
2403 /*
2404  * update_array_info is used to change the configuration of an
2405  * on-line array.
2406  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2407  * fields in the info are checked against the array.
2408  * Any differences that cannot be handled will cause an error.
2409  * Normally, only one change can be managed at a time.
2410  */
2411 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2412 {
2413         int rv = 0;
2414         int cnt = 0;
2415
2416         if (mddev->major_version != info->major_version ||
2417             mddev->minor_version != info->minor_version ||
2418 /*          mddev->patch_version != info->patch_version || */
2419             mddev->ctime         != info->ctime         ||
2420             mddev->level         != info->level         ||
2421             mddev->layout        != info->layout        ||
2422             !mddev->persistent   != info->not_persistent||
2423             mddev->chunk_size    != info->chunk_size    )
2424                 return -EINVAL;
2425         /* Check there is only one change */
2426         if (mddev->size != info->size) cnt++;
2427         if (mddev->raid_disks != info->raid_disks) cnt++;
2428         if (cnt == 0) return 0;
2429         if (cnt > 1) return -EINVAL;
2430
2431         if (mddev->size != info->size) {
2432                 mdk_rdev_t * rdev;
2433                 struct list_head *tmp;
2434                 if (mddev->pers->resize == NULL)
2435                         return -EINVAL;
2436                 /* The "size" is the amount of each device that is used.
2437                  * This can only make sense for arrays with redundancy.
2438                  * linear and raid0 always use whatever space is available
2439                  * We can only consider changing the size of no resync
2440                  * or reconstruction is happening, and if the new size
2441                  * is acceptable. It must fit before the sb_offset or,
2442                  * if that is <data_offset, it must fit before the
2443                  * size of each device.
2444                  * If size is zero, we find the largest size that fits.
2445                  */
2446                 if (mddev->sync_thread)
2447                         return -EBUSY;
2448                 ITERATE_RDEV(mddev,rdev,tmp) {
2449                         sector_t avail;
2450                         int fit = (info->size == 0);
2451                         if (rdev->sb_offset > rdev->data_offset)
2452                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2453                         else
2454                                 avail = get_capacity(rdev->bdev->bd_disk)
2455                                         - rdev->data_offset;
2456                         if (fit && (info->size == 0 || info->size > avail/2))
2457                                 info->size = avail/2;
2458                         if (avail < ((sector_t)info->size << 1))
2459                                 return -ENOSPC;
2460                 }
2461                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2462                 if (!rv) {
2463                         struct block_device *bdev;
2464
2465                         bdev = bdget_disk(mddev->gendisk, 0);
2466                         if (bdev) {
2467                                 down(&bdev->bd_inode->i_sem);
2468                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2469                                 up(&bdev->bd_inode->i_sem);
2470                                 bdput(bdev);
2471                         }
2472                 }
2473         }
2474         if (mddev->raid_disks    != info->raid_disks) {
2475                 /* change the number of raid disks */
2476                 if (mddev->pers->reshape == NULL)
2477                         return -EINVAL;
2478                 if (info->raid_disks <= 0 ||
2479                     info->raid_disks >= mddev->max_disks)
2480                         return -EINVAL;
2481                 if (mddev->sync_thread)
2482                         return -EBUSY;
2483                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2484                 if (!rv) {
2485                         struct block_device *bdev;
2486
2487                         bdev = bdget_disk(mddev->gendisk, 0);
2488                         if (bdev) {
2489                                 down(&bdev->bd_inode->i_sem);
2490                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2491                                 up(&bdev->bd_inode->i_sem);
2492                                 bdput(bdev);
2493                         }
2494                 }
2495         }
2496         md_update_sb(mddev);
2497         return rv;
2498 }
2499
2500 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2501 {
2502         mdk_rdev_t *rdev;
2503
2504         rdev = find_rdev(mddev, dev);
2505         if (!rdev)
2506                 return -ENODEV;
2507
2508         md_error(mddev, rdev);
2509         return 0;
2510 }
2511
2512 static int md_ioctl(struct inode *inode, struct file *file,
2513                         unsigned int cmd, unsigned long arg)
2514 {
2515         int err = 0;
2516         void __user *argp = (void __user *)arg;
2517         struct hd_geometry __user *loc = argp;
2518         mddev_t *mddev = NULL;
2519
2520         if (!capable(CAP_SYS_ADMIN))
2521                 return -EACCES;
2522
2523         /*
2524          * Commands dealing with the RAID driver but not any
2525          * particular array:
2526          */
2527         switch (cmd)
2528         {
2529                 case RAID_VERSION:
2530                         err = get_version(argp);
2531                         goto done;
2532
2533                 case PRINT_RAID_DEBUG:
2534                         err = 0;
2535                         md_print_devices();
2536                         goto done;
2537
2538 #ifndef MODULE
2539                 case RAID_AUTORUN:
2540                         err = 0;
2541                         autostart_arrays(arg);
2542                         goto done;
2543 #endif
2544                 default:;
2545         }
2546
2547         /*
2548          * Commands creating/starting a new array:
2549          */
2550
2551         mddev = inode->i_bdev->bd_disk->private_data;
2552
2553         if (!mddev) {
2554                 BUG();
2555                 goto abort;
2556         }
2557
2558
2559         if (cmd == START_ARRAY) {
2560                 /* START_ARRAY doesn't need to lock the array as autostart_array
2561                  * does the locking, and it could even be a different array
2562                  */
2563                 static int cnt = 3;
2564                 if (cnt > 0 ) {
2565                         printk(KERN_WARNING
2566                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2567                                "This will not be supported beyond 2.6\n",
2568                                current->comm, current->pid);
2569                         cnt--;
2570                 }
2571                 err = autostart_array(new_decode_dev(arg));
2572                 if (err) {
2573                         printk(KERN_WARNING "md: autostart failed!\n");
2574                         goto abort;
2575                 }
2576                 goto done;
2577         }
2578
2579         err = mddev_lock(mddev);
2580         if (err) {
2581                 printk(KERN_INFO
2582                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2583                         err, cmd);
2584                 goto abort;
2585         }
2586
2587         switch (cmd)
2588         {
2589                 case SET_ARRAY_INFO:
2590                         {
2591                                 mdu_array_info_t info;
2592                                 if (!arg)
2593                                         memset(&info, 0, sizeof(info));
2594                                 else if (copy_from_user(&info, argp, sizeof(info))) {
2595                                         err = -EFAULT;
2596                                         goto abort_unlock;
2597                                 }
2598                                 if (mddev->pers) {
2599                                         err = update_array_info(mddev, &info);
2600                                         if (err) {
2601                                                 printk(KERN_WARNING "md: couldn't update"
2602                                                        " array info. %d\n", err);
2603                                                 goto abort_unlock;
2604                                         }
2605                                         goto done_unlock;
2606                                 }
2607                                 if (!list_empty(&mddev->disks)) {
2608                                         printk(KERN_WARNING
2609                                                "md: array %s already has disks!\n",
2610                                                mdname(mddev));
2611                                         err = -EBUSY;
2612                                         goto abort_unlock;
2613                                 }
2614                                 if (mddev->raid_disks) {
2615                                         printk(KERN_WARNING
2616                                                "md: array %s already initialised!\n",
2617                                                mdname(mddev));
2618                                         err = -EBUSY;
2619                                         goto abort_unlock;
2620                                 }
2621                                 err = set_array_info(mddev, &info);
2622                                 if (err) {
2623                                         printk(KERN_WARNING "md: couldn't set"
2624                                                " array info. %d\n", err);
2625                                         goto abort_unlock;
2626                                 }
2627                         }
2628                         goto done_unlock;
2629
2630                 default:;
2631         }
2632
2633         /*
2634          * Commands querying/configuring an existing array:
2635          */
2636         /* if we are initialised yet, only ADD_NEW_DISK or STOP_ARRAY is allowed */
2637         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY && cmd != RUN_ARRAY) {
2638                 err = -ENODEV;
2639                 goto abort_unlock;
2640         }
2641
2642         /*
2643          * Commands even a read-only array can execute:
2644          */
2645         switch (cmd)
2646         {
2647                 case GET_ARRAY_INFO:
2648                         err = get_array_info(mddev, argp);
2649                         goto done_unlock;
2650
2651                 case GET_DISK_INFO:
2652                         err = get_disk_info(mddev, argp);
2653                         goto done_unlock;
2654
2655                 case RESTART_ARRAY_RW:
2656                         err = restart_array(mddev);
2657                         goto done_unlock;
2658
2659                 case STOP_ARRAY:
2660                         err = do_md_stop (mddev, 0);
2661                         goto done_unlock;
2662
2663                 case STOP_ARRAY_RO:
2664                         err = do_md_stop (mddev, 1);
2665                         goto done_unlock;
2666
2667         /*
2668          * We have a problem here : there is no easy way to give a CHS
2669          * virtual geometry. We currently pretend that we have a 2 heads
2670          * 4 sectors (with a BIG number of cylinders...). This drives
2671          * dosfs just mad... ;-)
2672          */
2673                 case HDIO_GETGEO:
2674                         if (!loc) {
2675                                 err = -EINVAL;
2676                                 goto abort_unlock;
2677                         }
2678                         err = put_user (2, (char __user *) &loc->heads);
2679                         if (err)
2680                                 goto abort_unlock;
2681                         err = put_user (4, (char __user *) &loc->sectors);
2682                         if (err)
2683                                 goto abort_unlock;
2684                         err = put_user(get_capacity(mddev->gendisk)/8,
2685                                         (short __user *) &loc->cylinders);
2686                         if (err)
2687                                 goto abort_unlock;
2688                         err = put_user (get_start_sect(inode->i_bdev),
2689                                                 (long __user *) &loc->start);
2690                         goto done_unlock;
2691         }
2692
2693         /*
2694          * The remaining ioctls are changing the state of the
2695          * superblock, so we do not allow read-only arrays
2696          * here:
2697          */
2698         if (mddev->ro) {
2699                 err = -EROFS;
2700                 goto abort_unlock;
2701         }
2702
2703         switch (cmd)
2704         {
2705                 case ADD_NEW_DISK:
2706                 {
2707                         mdu_disk_info_t info;
2708                         if (copy_from_user(&info, argp, sizeof(info)))
2709                                 err = -EFAULT;
2710                         else
2711                                 err = add_new_disk(mddev, &info);
2712                         goto done_unlock;
2713                 }
2714
2715                 case HOT_REMOVE_DISK:
2716                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2717                         goto done_unlock;
2718
2719                 case HOT_ADD_DISK:
2720                         err = hot_add_disk(mddev, new_decode_dev(arg));
2721                         goto done_unlock;
2722
2723                 case SET_DISK_FAULTY:
2724                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2725                         goto done_unlock;
2726
2727                 case RUN_ARRAY:
2728                         err = do_md_run (mddev);
2729                         goto done_unlock;
2730
2731                 default:
2732                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2733                                 printk(KERN_WARNING "md: %s(pid %d) used"
2734                                         " obsolete MD ioctl, upgrade your"
2735                                         " software to use new ictls.\n",
2736                                         current->comm, current->pid);
2737                         err = -EINVAL;
2738                         goto abort_unlock;
2739         }
2740
2741 done_unlock:
2742 abort_unlock:
2743         mddev_unlock(mddev);
2744
2745         return err;
2746 done:
2747         if (err)
2748                 MD_BUG();
2749 abort:
2750         return err;
2751 }
2752
2753 static int md_open(struct inode *inode, struct file *file)
2754 {
2755         /*
2756          * Succeed if we can lock the mddev, which confirms that
2757          * it isn't being stopped right now.
2758          */
2759         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2760         int err;
2761
2762         if ((err = mddev_lock(mddev)))
2763                 goto out;
2764
2765         err = 0;
2766         mddev_get(mddev);
2767         mddev_unlock(mddev);
2768
2769         check_disk_change(inode->i_bdev);
2770  out:
2771         return err;
2772 }
2773
2774 static int md_release(struct inode *inode, struct file * file)
2775 {
2776         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2777
2778         if (!mddev)
2779                 BUG();
2780         mddev_put(mddev);
2781
2782         return 0;
2783 }
2784
2785 static int md_media_changed(struct gendisk *disk)
2786 {
2787         mddev_t *mddev = disk->private_data;
2788
2789         return mddev->changed;
2790 }
2791
2792 static int md_revalidate(struct gendisk *disk)
2793 {
2794         mddev_t *mddev = disk->private_data;
2795
2796         mddev->changed = 0;
2797         return 0;
2798 }
2799 static struct block_device_operations md_fops =
2800 {
2801         .owner          = THIS_MODULE,
2802         .open           = md_open,
2803         .release        = md_release,
2804         .ioctl          = md_ioctl,
2805         .media_changed  = md_media_changed,
2806         .revalidate_disk= md_revalidate,
2807 };
2808
2809 int md_thread(void * arg)
2810 {
2811         mdk_thread_t *thread = arg;
2812
2813         lock_kernel();
2814
2815         /*
2816          * Detach thread
2817          */
2818
2819         daemonize(thread->name, mdname(thread->mddev));
2820
2821         current->exit_signal = SIGCHLD;
2822         allow_signal(SIGKILL);
2823         thread->tsk = current;
2824
2825         /*
2826          * md_thread is a 'system-thread', it's priority should be very
2827          * high. We avoid resource deadlocks individually in each
2828          * raid personality. (RAID5 does preallocation) We also use RR and
2829          * the very same RT priority as kswapd, thus we will never get
2830          * into a priority inversion deadlock.
2831          *
2832          * we definitely have to have equal or higher priority than
2833          * bdflush, otherwise bdflush will deadlock if there are too
2834          * many dirty RAID5 blocks.
2835          */
2836         unlock_kernel();
2837
2838         complete(thread->event);
2839         while (thread->run) {
2840                 void (*run)(mddev_t *);
2841
2842                 wait_event_interruptible(thread->wqueue,
2843                                          test_bit(THREAD_WAKEUP, &thread->flags));
2844                 if (current->flags & PF_FREEZE)
2845                         refrigerator(PF_FREEZE);
2846
2847                 clear_bit(THREAD_WAKEUP, &thread->flags);
2848
2849                 run = thread->run;
2850                 if (run)
2851                         run(thread->mddev);
2852
2853                 if (signal_pending(current))
2854                         flush_signals(current);
2855         }
2856         complete(thread->event);
2857         return 0;
2858 }
2859
2860 void md_wakeup_thread(mdk_thread_t *thread)
2861 {
2862         if (thread) {
2863                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
2864                 set_bit(THREAD_WAKEUP, &thread->flags);
2865                 wake_up(&thread->wqueue);
2866         }
2867 }
2868
2869 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
2870                                  const char *name)
2871 {
2872         mdk_thread_t *thread;
2873         int ret;
2874         struct completion event;
2875
2876         thread = (mdk_thread_t *) kmalloc
2877                                 (sizeof(mdk_thread_t), GFP_KERNEL);
2878         if (!thread)
2879                 return NULL;
2880
2881         memset(thread, 0, sizeof(mdk_thread_t));
2882         init_waitqueue_head(&thread->wqueue);
2883
2884         init_completion(&event);
2885         thread->event = &event;
2886         thread->run = run;
2887         thread->mddev = mddev;
2888         thread->name = name;
2889         ret = kernel_thread(md_thread, thread, 0);
2890         if (ret < 0) {
2891                 kfree(thread);
2892                 return NULL;
2893         }
2894         wait_for_completion(&event);
2895         return thread;
2896 }
2897
2898 static void md_interrupt_thread(mdk_thread_t *thread)
2899 {
2900         if (!thread->tsk) {
2901                 MD_BUG();
2902                 return;
2903         }
2904         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
2905         send_sig(SIGKILL, thread->tsk, 1);
2906 }
2907
2908 void md_unregister_thread(mdk_thread_t *thread)
2909 {
2910         struct completion event;
2911
2912         init_completion(&event);
2913
2914         thread->event = &event;
2915         thread->run = NULL;
2916         thread->name = NULL;
2917         md_interrupt_thread(thread);
2918         wait_for_completion(&event);
2919         kfree(thread);
2920 }
2921
2922 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
2923 {
2924         if (!mddev) {
2925                 MD_BUG();
2926                 return;
2927         }
2928
2929         if (!rdev || rdev->faulty)
2930                 return;
2931
2932         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
2933                 mdname(mddev),
2934                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
2935                 __builtin_return_address(0),__builtin_return_address(1),
2936                 __builtin_return_address(2),__builtin_return_address(3));
2937
2938         if (!mddev->pers->error_handler)
2939                 return;
2940         mddev->pers->error_handler(mddev,rdev);
2941         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2942         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2943         md_wakeup_thread(mddev->thread);
2944 }
2945
2946 /* seq_file implementation /proc/mdstat */
2947
2948 static void status_unused(struct seq_file *seq)
2949 {
2950         int i = 0;
2951         mdk_rdev_t *rdev;
2952         struct list_head *tmp;
2953
2954         seq_printf(seq, "unused devices: ");
2955
2956         ITERATE_RDEV_PENDING(rdev,tmp) {
2957                 char b[BDEVNAME_SIZE];
2958                 i++;
2959                 seq_printf(seq, "%s ",
2960                               bdevname(rdev->bdev,b));
2961         }
2962         if (!i)
2963                 seq_printf(seq, "<none>");
2964
2965         seq_printf(seq, "\n");
2966 }
2967
2968
2969 static void status_resync(struct seq_file *seq, mddev_t * mddev)
2970 {
2971         unsigned long max_blocks, resync, res, dt, db, rt;
2972
2973         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
2974
2975         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2976                 max_blocks = mddev->resync_max_sectors >> 1;
2977         else
2978                 max_blocks = mddev->size;
2979
2980         /*
2981          * Should not happen.
2982          */
2983         if (!max_blocks) {
2984                 MD_BUG();
2985                 return;
2986         }
2987         res = (resync/1024)*1000/(max_blocks/1024 + 1);
2988         {
2989                 int i, x = res/50, y = 20-x;
2990                 seq_printf(seq, "[");
2991                 for (i = 0; i < x; i++)
2992                         seq_printf(seq, "=");
2993                 seq_printf(seq, ">");
2994                 for (i = 0; i < y; i++)
2995                         seq_printf(seq, ".");
2996                 seq_printf(seq, "] ");
2997         }
2998         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
2999                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3000                        "resync" : "recovery"),
3001                       res/10, res % 10, resync, max_blocks);
3002
3003         /*
3004          * We do not want to overflow, so the order of operands and
3005          * the * 100 / 100 trick are important. We do a +1 to be
3006          * safe against division by zero. We only estimate anyway.
3007          *
3008          * dt: time from mark until now
3009          * db: blocks written from mark until now
3010          * rt: remaining time
3011          */
3012         dt = ((jiffies - mddev->resync_mark) / HZ);
3013         if (!dt) dt++;
3014         db = resync - (mddev->resync_mark_cnt/2);
3015         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3016
3017         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3018
3019         seq_printf(seq, " speed=%ldK/sec", db/dt);
3020 }
3021
3022 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3023 {
3024         struct list_head *tmp;
3025         loff_t l = *pos;
3026         mddev_t *mddev;
3027
3028         if (l >= 0x10000)
3029                 return NULL;
3030         if (!l--)
3031                 /* header */
3032                 return (void*)1;
3033
3034         spin_lock(&all_mddevs_lock);
3035         list_for_each(tmp,&all_mddevs)
3036                 if (!l--) {
3037                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3038                         mddev_get(mddev);
3039                         spin_unlock(&all_mddevs_lock);
3040                         return mddev;
3041                 }
3042         spin_unlock(&all_mddevs_lock);
3043         if (!l--)
3044                 return (void*)2;/* tail */
3045         return NULL;
3046 }
3047
3048 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3049 {
3050         struct list_head *tmp;
3051         mddev_t *next_mddev, *mddev = v;
3052
3053         ++*pos;
3054         if (v == (void*)2)
3055                 return NULL;
3056
3057         spin_lock(&all_mddevs_lock);
3058         if (v == (void*)1)
3059                 tmp = all_mddevs.next;
3060         else
3061                 tmp = mddev->all_mddevs.next;
3062         if (tmp != &all_mddevs)
3063                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3064         else {
3065                 next_mddev = (void*)2;
3066                 *pos = 0x10000;
3067         }
3068         spin_unlock(&all_mddevs_lock);
3069
3070         if (v != (void*)1)
3071                 mddev_put(mddev);
3072         return next_mddev;
3073
3074 }
3075
3076 static void md_seq_stop(struct seq_file *seq, void *v)
3077 {
3078         mddev_t *mddev = v;
3079
3080         if (mddev && v != (void*)1 && v != (void*)2)
3081                 mddev_put(mddev);
3082 }
3083
3084 static int md_seq_show(struct seq_file *seq, void *v)
3085 {
3086         mddev_t *mddev = v;
3087         sector_t size;
3088         struct list_head *tmp2;
3089         mdk_rdev_t *rdev;
3090         int i;
3091
3092         if (v == (void*)1) {
3093                 seq_printf(seq, "Personalities : ");
3094                 spin_lock(&pers_lock);
3095                 for (i = 0; i < MAX_PERSONALITY; i++)
3096                         if (pers[i])
3097                                 seq_printf(seq, "[%s] ", pers[i]->name);
3098
3099                 spin_unlock(&pers_lock);
3100                 seq_printf(seq, "\n");
3101                 return 0;
3102         }
3103         if (v == (void*)2) {
3104                 status_unused(seq);
3105                 return 0;
3106         }
3107
3108         if (mddev_lock(mddev)!=0)
3109                 return -EINTR;
3110         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3111                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3112                                                 mddev->pers ? "" : "in");
3113                 if (mddev->pers) {
3114                         if (mddev->ro)
3115                                 seq_printf(seq, " (read-only)");
3116                         seq_printf(seq, " %s", mddev->pers->name);
3117                 }
3118
3119                 size = 0;
3120                 ITERATE_RDEV(mddev,rdev,tmp2) {
3121                         char b[BDEVNAME_SIZE];
3122                         seq_printf(seq, " %s[%d]",
3123                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3124                         if (rdev->faulty) {
3125                                 seq_printf(seq, "(F)");
3126                                 continue;
3127                         }
3128                         size += rdev->size;
3129                 }
3130
3131                 if (!list_empty(&mddev->disks)) {
3132                         if (mddev->pers)
3133                                 seq_printf(seq, "\n      %llu blocks",
3134                                         (unsigned long long)mddev->array_size);
3135                         else
3136                                 seq_printf(seq, "\n      %llu blocks",
3137                                         (unsigned long long)size);
3138                 }
3139
3140                 if (mddev->pers) {
3141                         mddev->pers->status (seq, mddev);
3142                         seq_printf(seq, "\n      ");
3143                         if (mddev->curr_resync > 2)
3144                                 status_resync (seq, mddev);
3145                         else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3146                                 seq_printf(seq, "       resync=DELAYED");
3147                 }
3148
3149                 seq_printf(seq, "\n");
3150         }
3151         mddev_unlock(mddev);
3152
3153         return 0;
3154 }
3155
3156 static struct seq_operations md_seq_ops = {
3157         .start  = md_seq_start,
3158         .next   = md_seq_next,
3159         .stop   = md_seq_stop,
3160         .show   = md_seq_show,
3161 };
3162
3163 static int md_seq_open(struct inode *inode, struct file *file)
3164 {
3165         int error;
3166
3167         error = seq_open(file, &md_seq_ops);
3168         return error;
3169 }
3170
3171 static struct file_operations md_seq_fops = {
3172         .open           = md_seq_open,
3173         .read           = seq_read,
3174         .llseek         = seq_lseek,
3175         .release        = seq_release,
3176 };
3177
3178 int register_md_personality(int pnum, mdk_personality_t *p)
3179 {
3180         if (pnum >= MAX_PERSONALITY) {
3181                 printk(KERN_ERR
3182                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3183                        p->name, pnum, MAX_PERSONALITY-1);
3184                 return -EINVAL;
3185         }
3186
3187         spin_lock(&pers_lock);
3188         if (pers[pnum]) {
3189                 spin_unlock(&pers_lock);
3190                 MD_BUG();
3191                 return -EBUSY;
3192         }
3193
3194         pers[pnum] = p;
3195         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3196         spin_unlock(&pers_lock);
3197         return 0;
3198 }
3199
3200 int unregister_md_personality(int pnum)
3201 {
3202         if (pnum >= MAX_PERSONALITY) {
3203                 MD_BUG();
3204                 return -EINVAL;
3205         }
3206
3207         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3208         spin_lock(&pers_lock);
3209         pers[pnum] = NULL;
3210         spin_unlock(&pers_lock);
3211         return 0;
3212 }
3213
3214 static int is_mddev_idle(mddev_t *mddev)
3215 {
3216         mdk_rdev_t * rdev;
3217         struct list_head *tmp;
3218         int idle;
3219         unsigned long curr_events;
3220
3221         idle = 1;
3222         ITERATE_RDEV(mddev,rdev,tmp) {
3223                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3224                 curr_events = disk_stat_read(disk, read_sectors) +
3225                                 disk_stat_read(disk, write_sectors) -
3226                                 atomic_read(&disk->sync_io);
3227                 /* Allow some slack between valud of curr_events and last_events,
3228                  * as there are some uninteresting races.
3229                  * Note: the following is an unsigned comparison.
3230                  */
3231                 if ((curr_events - rdev->last_events + 32) > 64) {
3232                         rdev->last_events = curr_events;
3233                         idle = 0;
3234                 }
3235         }
3236         return idle;
3237 }
3238
3239 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3240 {
3241         /* another "blocks" (512byte) blocks have been synced */
3242         atomic_sub(blocks, &mddev->recovery_active);
3243         wake_up(&mddev->recovery_wait);
3244         if (!ok) {
3245                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3246                 md_wakeup_thread(mddev->thread);
3247                 // stop recovery, signal do_sync ....
3248         }
3249 }
3250
3251
3252 void md_write_start(mddev_t *mddev)
3253 {
3254         if (!atomic_read(&mddev->writes_pending)) {
3255                 mddev_lock_uninterruptible(mddev);
3256                 if (mddev->in_sync) {
3257                         mddev->in_sync = 0;
3258                         del_timer(&mddev->safemode_timer);
3259                         md_update_sb(mddev);
3260                 }
3261                 atomic_inc(&mddev->writes_pending);
3262                 mddev_unlock(mddev);
3263         } else
3264                 atomic_inc(&mddev->writes_pending);
3265 }
3266
3267 void md_write_end(mddev_t *mddev)
3268 {
3269         if (atomic_dec_and_test(&mddev->writes_pending)) {
3270                 if (mddev->safemode == 2)
3271                         md_wakeup_thread(mddev->thread);
3272                 else
3273                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3274         }
3275 }
3276
3277 static inline void md_enter_safemode(mddev_t *mddev)
3278 {
3279         if (!mddev->safemode) return;
3280         if (mddev->safemode == 2 &&
3281             (atomic_read(&mddev->writes_pending) || mddev->in_sync ||
3282                     mddev->recovery_cp != MaxSector))
3283                 return; /* avoid the lock */
3284         mddev_lock_uninterruptible(mddev);
3285         if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3286             !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3287                 mddev->in_sync = 1;
3288                 md_update_sb(mddev);
3289         }
3290         mddev_unlock(mddev);
3291
3292         if (mddev->safemode == 1)
3293                 mddev->safemode = 0;
3294 }
3295
3296 void md_handle_safemode(mddev_t *mddev)
3297 {
3298         if (signal_pending(current)) {
3299                 printk(KERN_INFO "md: %s in immediate safe mode\n",
3300                         mdname(mddev));
3301                 mddev->safemode = 2;
3302                 flush_signals(current);
3303         }
3304         md_enter_safemode(mddev);
3305 }
3306
3307
3308 DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3309
3310 #define SYNC_MARKS      10
3311 #define SYNC_MARK_STEP  (3*HZ)
3312 static void md_do_sync(mddev_t *mddev)
3313 {
3314         mddev_t *mddev2;
3315         unsigned int currspeed = 0,
3316                  window;
3317         sector_t max_sectors,j;
3318         unsigned long mark[SYNC_MARKS];
3319         sector_t mark_cnt[SYNC_MARKS];
3320         int last_mark,m;
3321         struct list_head *tmp;
3322         sector_t last_check;
3323
3324         /* just incase thread restarts... */
3325         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3326                 return;
3327
3328         /* we overload curr_resync somewhat here.
3329          * 0 == not engaged in resync at all
3330          * 2 == checking that there is no conflict with another sync
3331          * 1 == like 2, but have yielded to allow conflicting resync to
3332          *              commense
3333          * other == active in resync - this many blocks
3334          */
3335         do {
3336                 mddev->curr_resync = 2;
3337
3338                 ITERATE_MDDEV(mddev2,tmp) {
3339                         if (mddev2 == mddev)
3340                                 continue;
3341                         if (mddev2->curr_resync &&
3342                             match_mddev_units(mddev,mddev2)) {
3343                                 printk(KERN_INFO "md: delaying resync of %s"
3344                                         " until %s has finished resync (they"
3345                                         " share one or more physical units)\n",
3346                                        mdname(mddev), mdname(mddev2));
3347                                 if (mddev < mddev2) {/* arbitrarily yield */
3348                                         mddev->curr_resync = 1;
3349                                         wake_up(&resync_wait);
3350                                 }
3351                                 if (wait_event_interruptible(resync_wait,
3352                                                              mddev2->curr_resync < mddev->curr_resync)) {
3353                                         flush_signals(current);
3354                                         mddev_put(mddev2);
3355                                         goto skip;
3356                                 }
3357                         }
3358                         if (mddev->curr_resync == 1) {
3359                                 mddev_put(mddev2);
3360                                 break;
3361                         }
3362                 }
3363         } while (mddev->curr_resync < 2);
3364
3365         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3366                 /* resync follows the size requested by the personality,
3367                  * which default to physical size, but can be virtual size
3368                  */
3369                 max_sectors = mddev->resync_max_sectors;
3370         else
3371                 /* recovery follows the physical size of devices */
3372                 max_sectors = mddev->size << 1;
3373
3374         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3375         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3376                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3377         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3378                "(but not more than %d KB/sec) for reconstruction.\n",
3379                sysctl_speed_limit_max);
3380
3381         is_mddev_idle(mddev); /* this also initializes IO event counters */
3382         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3383                 j = mddev->recovery_cp;
3384         else
3385                 j = 0;
3386         for (m = 0; m < SYNC_MARKS; m++) {
3387                 mark[m] = jiffies;
3388                 mark_cnt[m] = j;
3389         }
3390         last_mark = 0;
3391         mddev->resync_mark = mark[last_mark];
3392         mddev->resync_mark_cnt = mark_cnt[last_mark];
3393
3394         /*
3395          * Tune reconstruction:
3396          */
3397         window = 32*(PAGE_SIZE/512);
3398         printk(KERN_INFO "md: using %dk window, over a total of %Lu blocks.\n",
3399                 window/2,(unsigned long long) max_sectors/2);
3400
3401         atomic_set(&mddev->recovery_active, 0);
3402         init_waitqueue_head(&mddev->recovery_wait);
3403         last_check = 0;
3404
3405         if (j)
3406                 printk(KERN_INFO
3407                         "md: resuming recovery of %s from checkpoint.\n",
3408                         mdname(mddev));
3409
3410         while (j < max_sectors) {
3411                 int sectors;
3412
3413                 sectors = mddev->pers->sync_request(mddev, j, currspeed < sysctl_speed_limit_min);
3414                 if (sectors < 0) {
3415                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3416                         goto out;
3417                 }
3418                 atomic_add(sectors, &mddev->recovery_active);
3419                 j += sectors;
3420                 if (j>1) mddev->curr_resync = j;
3421
3422                 if (last_check + window > j || j == max_sectors)
3423                         continue;
3424
3425                 last_check = j;
3426
3427                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3428                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3429                         break;
3430
3431         repeat:
3432                 if (jiffies >= mark[last_mark] + SYNC_MARK_STEP ) {
3433                         /* step marks */
3434                         int next = (last_mark+1) % SYNC_MARKS;
3435
3436                         mddev->resync_mark = mark[next];
3437                         mddev->resync_mark_cnt = mark_cnt[next];
3438                         mark[next] = jiffies;
3439                         mark_cnt[next] = j - atomic_read(&mddev->recovery_active);
3440                         last_mark = next;
3441                 }
3442
3443
3444                 if (signal_pending(current)) {
3445                         /*
3446                          * got a signal, exit.
3447                          */
3448                         printk(KERN_INFO
3449                                 "md: md_do_sync() got signal ... exiting\n");
3450                         flush_signals(current);
3451                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3452                         goto out;
3453                 }
3454
3455                 /*
3456                  * this loop exits only if either when we are slower than
3457                  * the 'hard' speed limit, or the system was IO-idle for
3458                  * a jiffy.
3459                  * the system might be non-idle CPU-wise, but we only care
3460                  * about not overloading the IO subsystem. (things like an
3461                  * e2fsck being done on the RAID array should execute fast)
3462                  */
3463                 mddev->queue->unplug_fn(mddev->queue);
3464                 cond_resched();
3465
3466                 currspeed = ((unsigned long)(j-mddev->resync_mark_cnt))/2/((jiffies-mddev->resync_mark)/HZ +1) +1;
3467
3468                 if (currspeed > sysctl_speed_limit_min) {
3469                         if ((currspeed > sysctl_speed_limit_max) ||
3470                                         !is_mddev_idle(mddev)) {
3471                                 current->state = TASK_INTERRUPTIBLE;
3472                                 schedule_timeout(HZ/4);
3473                                 goto repeat;
3474                         }
3475                 }
3476         }
3477         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3478         /*
3479          * this also signals 'finished resyncing' to md_stop
3480          */
3481  out:
3482         mddev->queue->unplug_fn(mddev->queue);
3483
3484         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3485
3486         /* tell personality that we are finished */
3487         mddev->pers->sync_request(mddev, max_sectors, 1);
3488
3489         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3490             mddev->curr_resync > 2 &&
3491             mddev->curr_resync > mddev->recovery_cp) {
3492                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3493                         printk(KERN_INFO
3494                                 "md: checkpointing recovery of %s.\n",
3495                                 mdname(mddev));
3496                         mddev->recovery_cp = mddev->curr_resync;
3497                 } else
3498                         mddev->recovery_cp = MaxSector;
3499         }
3500
3501         md_enter_safemode(mddev);
3502  skip:
3503         mddev->curr_resync = 0;
3504         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3505         md_wakeup_thread(mddev->thread);
3506 }
3507
3508
3509 /*
3510  * This routine is regularly called by all per-raid-array threads to
3511  * deal with generic issues like resync and super-block update.
3512  * Raid personalities that don't have a thread (linear/raid0) do not
3513  * need this as they never do any recovery or update the superblock.
3514  *
3515  * It does not do any resync itself, but rather "forks" off other threads
3516  * to do that as needed.
3517  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3518  * "->recovery" and create a thread at ->sync_thread.
3519  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3520  * and wakeups up this thread which will reap the thread and finish up.
3521  * This thread also removes any faulty devices (with nr_pending == 0).
3522  *
3523  * The overall approach is:
3524  *  1/ if the superblock needs updating, update it.
3525  *  2/ If a recovery thread is running, don't do anything else.
3526  *  3/ If recovery has finished, clean up, possibly marking spares active.
3527  *  4/ If there are any faulty devices, remove them.
3528  *  5/ If array is degraded, try to add spares devices
3529  *  6/ If array has spares or is not in-sync, start a resync thread.
3530  */
3531 void md_check_recovery(mddev_t *mddev)
3532 {
3533         mdk_rdev_t *rdev;
3534         struct list_head *rtmp;
3535
3536
3537         dprintk(KERN_INFO "md: recovery thread got woken up ...\n");
3538
3539         if (mddev->ro)
3540                 return;
3541         if ( ! (
3542                 mddev->sb_dirty ||
3543                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3544                 test_bit(MD_RECOVERY_DONE, &mddev->recovery)
3545                 ))
3546                 return;
3547         if (mddev_trylock(mddev)==0) {
3548                 int spares =0;
3549                 if (mddev->sb_dirty)
3550                         md_update_sb(mddev);
3551                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3552                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3553                         /* resync/recovery still happening */
3554                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3555                         goto unlock;
3556                 }
3557                 if (mddev->sync_thread) {
3558                         /* resync has finished, collect result */
3559                         md_unregister_thread(mddev->sync_thread);
3560                         mddev->sync_thread = NULL;
3561                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3562                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3563                                 /* success...*/
3564                                 /* activate any spares */
3565                                 mddev->pers->spare_active(mddev);
3566                         }
3567                         md_update_sb(mddev);
3568                         mddev->recovery = 0;
3569                         /* flag recovery needed just to double check */
3570                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3571                         wake_up(&resync_wait);
3572                         goto unlock;
3573                 }
3574                 if (mddev->recovery) {
3575                         /* probably just the RECOVERY_NEEDED flag */
3576                         mddev->recovery = 0;
3577                         wake_up(&resync_wait);
3578                 }
3579
3580                 /* no recovery is running.
3581                  * remove any failed drives, then
3582                  * add spares if possible
3583                  */
3584                 ITERATE_RDEV(mddev,rdev,rtmp) {
3585                         if (rdev->raid_disk >= 0 &&
3586                             rdev->faulty &&
3587                             atomic_read(&rdev->nr_pending)==0) {
3588                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3589                                         rdev->raid_disk = -1;
3590                         }
3591                         if (!rdev->faulty && rdev->raid_disk >= 0 && !rdev->in_sync)
3592                                 spares++;
3593                 }
3594                 if (mddev->degraded) {
3595                         ITERATE_RDEV(mddev,rdev,rtmp)
3596                                 if (rdev->raid_disk < 0
3597                                     && !rdev->faulty) {
3598                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3599                                                 spares++;
3600                                         else
3601                                                 break;
3602                                 }
3603                 }
3604
3605                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3606                         /* nothing we can do ... */
3607                         goto unlock;
3608                 }
3609                 if (mddev->pers->sync_request) {
3610                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3611                         if (!spares)
3612                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3613                         mddev->sync_thread = md_register_thread(md_do_sync,
3614                                                                 mddev,
3615                                                                 "%s_resync");
3616                         if (!mddev->sync_thread) {
3617                                 printk(KERN_ERR "%s: could not start resync"
3618                                         " thread...\n",
3619                                         mdname(mddev));
3620                                 /* leave the spares where they are, it shouldn't hurt */
3621                                 mddev->recovery = 0;
3622                         } else {
3623                                 md_wakeup_thread(mddev->sync_thread);
3624                         }
3625                 }
3626         unlock:
3627                 mddev_unlock(mddev);
3628         }
3629 }
3630
3631 int md_notify_reboot(struct notifier_block *this,
3632                                         unsigned long code, void *x)
3633 {
3634         struct list_head *tmp;
3635         mddev_t *mddev;
3636
3637         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3638
3639                 printk(KERN_INFO "md: stopping all md devices.\n");
3640
3641                 ITERATE_MDDEV(mddev,tmp)
3642                         if (mddev_trylock(mddev)==0)
3643                                 do_md_stop (mddev, 1);
3644                 /*
3645                  * certain more exotic SCSI devices are known to be
3646                  * volatile wrt too early system reboots. While the
3647                  * right place to handle this issue is the given
3648                  * driver, we do want to have a safe RAID driver ...
3649                  */
3650                 mdelay(1000*1);
3651         }
3652         return NOTIFY_DONE;
3653 }
3654
3655 struct notifier_block md_notifier = {
3656         .notifier_call  = md_notify_reboot,
3657         .next           = NULL,
3658         .priority       = INT_MAX, /* before any real devices */
3659 };
3660
3661 static void md_geninit(void)
3662 {
3663         struct proc_dir_entry *p;
3664
3665         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3666
3667         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3668         if (p)
3669                 p->proc_fops = &md_seq_fops;
3670 }
3671
3672 int __init md_init(void)
3673 {
3674         int minor;
3675
3676         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3677                         " MD_SB_DISKS=%d\n",
3678                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3679                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3680
3681         if (register_blkdev(MAJOR_NR, "md"))
3682                 return -1;
3683         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3684                 unregister_blkdev(MAJOR_NR, "md");
3685                 return -1;
3686         }
3687         devfs_mk_dir("md");
3688         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3689                                 md_probe, NULL, NULL);
3690         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3691                             md_probe, NULL, NULL);
3692
3693         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3694                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3695                                 S_IFBLK|S_IRUSR|S_IWUSR,
3696                                 "md/%d", minor);
3697
3698         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3699                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3700                               S_IFBLK|S_IRUSR|S_IWUSR,
3701                               "md/d%d", minor);
3702
3703
3704         register_reboot_notifier(&md_notifier);
3705         raid_table_header = register_sysctl_table(raid_root_table, 1);
3706
3707         md_geninit();
3708         return (0);
3709 }
3710
3711
3712 #ifndef MODULE
3713
3714 /*
3715  * Searches all registered partitions for autorun RAID arrays
3716  * at boot time.
3717  */
3718 static dev_t detected_devices[128];
3719 static int dev_cnt;
3720
3721 void md_autodetect_dev(dev_t dev)
3722 {
3723         if (dev_cnt >= 0 && dev_cnt < 127)
3724                 detected_devices[dev_cnt++] = dev;
3725 }
3726
3727
3728 static void autostart_arrays(int part)
3729 {
3730         mdk_rdev_t *rdev;
3731         int i;
3732
3733         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3734
3735         for (i = 0; i < dev_cnt; i++) {
3736                 dev_t dev = detected_devices[i];
3737
3738                 rdev = md_import_device(dev,0, 0);
3739                 if (IS_ERR(rdev))
3740                         continue;
3741
3742                 if (rdev->faulty) {
3743                         MD_BUG();
3744                         continue;
3745                 }
3746                 list_add(&rdev->same_set, &pending_raid_disks);
3747         }
3748         dev_cnt = 0;
3749
3750         autorun_devices(part);
3751 }
3752
3753 #endif
3754
3755 static __exit void md_exit(void)
3756 {
3757         mddev_t *mddev;
3758         struct list_head *tmp;
3759         int i;
3760         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3761         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3762         for (i=0; i < MAX_MD_DEVS; i++)
3763                 devfs_remove("md/%d", i);
3764         for (i=0; i < MAX_MD_DEVS; i++)
3765                 devfs_remove("md/d%d", i);
3766
3767         devfs_remove("md");
3768
3769         unregister_blkdev(MAJOR_NR,"md");
3770         unregister_blkdev(mdp_major, "mdp");
3771         unregister_reboot_notifier(&md_notifier);
3772         unregister_sysctl_table(raid_table_header);
3773         remove_proc_entry("mdstat", NULL);
3774         ITERATE_MDDEV(mddev,tmp) {
3775                 struct gendisk *disk = mddev->gendisk;
3776                 if (!disk)
3777                         continue;
3778                 export_array(mddev);
3779                 del_gendisk(disk);
3780                 put_disk(disk);
3781                 mddev->gendisk = NULL;
3782                 mddev_put(mddev);
3783         }
3784 }
3785
3786 module_init(md_init)
3787 module_exit(md_exit)
3788
3789 EXPORT_SYMBOL(register_md_personality);
3790 EXPORT_SYMBOL(unregister_md_personality);
3791 EXPORT_SYMBOL(md_error);
3792 EXPORT_SYMBOL(md_done_sync);
3793 EXPORT_SYMBOL(md_write_start);
3794 EXPORT_SYMBOL(md_write_end);
3795 EXPORT_SYMBOL(md_handle_safemode);
3796 EXPORT_SYMBOL(md_register_thread);
3797 EXPORT_SYMBOL(md_unregister_thread);
3798 EXPORT_SYMBOL(md_wakeup_thread);
3799 EXPORT_SYMBOL(md_print_devices);
3800 EXPORT_SYMBOL(md_check_recovery);
3801 MODULE_LICENSE("GPL");