drivers/md/md.c

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