err.no Git - linux-2.6/blob - 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    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/linkage.h>
  38 #include <linux/raid/md.h>
  39 #include <linux/raid/bitmap.h>
  40 #include <linux/sysctl.h>
  41 #include <linux/devfs_fs_kernel.h>
  42 #include <linux/buffer_head.h> /* for invalidate_bdev */
  43 #include <linux/suspend.h>
  44
  45 #include <linux/init.h>
  46
  47 #include <linux/file.h>
  48
  49 #ifdef CONFIG_KMOD
  50 #include <linux/kmod.h>
  51 #endif
  52
  53 #include <asm/unaligned.h>
  54
  55 #define MAJOR_NR MD_MAJOR
  56 #define MD_DRIVER
  57
  58 /* 63 partitions with the alternate major number (mdp) */
  59 #define MdpMinorShift 6
  60
  61 #define DEBUG 0
  62 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  63
  64
  65 #ifndef MODULE
  66 static void autostart_arrays (int part);
  67 #endif
  68
  69 static mdk_personality_t *pers[MAX_PERSONALITY];
  70 static DEFINE_SPINLOCK(pers_lock);
  71
  72 /*
  73  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  74  * is 1000 KB/sec, so the extra system load does not show up that much.
  75  * Increase it if you want to have more _guaranteed_ speed. Note that
  76  * the RAID driver will use the maximum available bandwith if the IO
  77  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  78  * speed limit - in case reconstruction slows down your system despite
  79  * idle IO detection.
  80  *
  81  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  82  */
  83
  84 static int sysctl_speed_limit_min = 1000;
  85 static int sysctl_speed_limit_max = 200000;
  86
  87 static struct ctl_table_header *raid_table_header;
  88
  89 static ctl_table raid_table[] = {
  90         {
  91                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  92                 .procname       = "speed_limit_min",
  93                 .data           = &sysctl_speed_limit_min,
  94                 .maxlen         = sizeof(int),
  95                 .mode           = 0644,
  96                 .proc_handler   = &proc_dointvec,
  97         },
  98         {
  99                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 100                 .procname       = "speed_limit_max",
 101                 .data           = &sysctl_speed_limit_max,
 102                 .maxlen         = sizeof(int),
 103                 .mode           = 0644,
 104                 .proc_handler   = &proc_dointvec,
 105         },
 106         { .ctl_name = 0 }
 107 };
 108
 109 static ctl_table raid_dir_table[] = {
 110         {
 111                 .ctl_name       = DEV_RAID,
 112                 .procname       = "raid",
 113                 .maxlen         = 0,
 114                 .mode           = 0555,
 115                 .child          = raid_table,
 116         },
 117         { .ctl_name = 0 }
 118 };
 119
 120 static ctl_table raid_root_table[] = {
 121         {
 122                 .ctl_name       = CTL_DEV,
 123                 .procname       = "dev",
 124                 .maxlen         = 0,
 125                 .mode           = 0555,
 126                 .child          = raid_dir_table,
 127         },
 128         { .ctl_name = 0 }
 129 };
 130
 131 static struct block_device_operations md_fops;
 132
 133 /*
 134  * Enables to iterate over all existing md arrays
 135  * all_mddevs_lock protects this list.
 136  */
 137 static LIST_HEAD(all_mddevs);
 138 static DEFINE_SPINLOCK(all_mddevs_lock);
 139
 140
 141 /*
 142  * iterates through all used mddevs in the system.
 143  * We take care to grab the all_mddevs_lock whenever navigating
 144  * the list, and to always hold a refcount when unlocked.
 145  * Any code which breaks out of this loop while own
 146  * a reference to the current mddev and must mddev_put it.
 147  */
 148 #define ITERATE_MDDEV(mddev,tmp)                                        \
 149                                                                         \
 150         for (({ spin_lock(&all_mddevs_lock);                            \
 151                 tmp = all_mddevs.next;                                  \
 152                 mddev = NULL;});                                        \
 153              ({ if (tmp != &all_mddevs)                                 \
 154                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 155                 spin_unlock(&all_mddevs_lock);                          \
 156                 if (mddev) mddev_put(mddev);                            \
 157                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 158                 tmp != &all_mddevs;});                                  \
 159              ({ spin_lock(&all_mddevs_lock);                            \
 160                 tmp = tmp->next;})                                      \
 161                 )
 162
 163
 164 static int md_fail_request (request_queue_t *q, struct bio *bio)
 165 {
 166         bio_io_error(bio, bio->bi_size);
 167         return 0;
 168 }
 169
 170 static inline mddev_t *mddev_get(mddev_t *mddev)
 171 {
 172         atomic_inc(&mddev->active);
 173         return mddev;
 174 }
 175
 176 static void mddev_put(mddev_t *mddev)
 177 {
 178         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 179                 return;
 180         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 181                 list_del(&mddev->all_mddevs);
 182                 blk_put_queue(mddev->queue);
 183                 kfree(mddev);
 184         }
 185         spin_unlock(&all_mddevs_lock);
 186 }
 187
 188 static mddev_t * mddev_find(dev_t unit)
 189 {
 190         mddev_t *mddev, *new = NULL;
 191
 192  retry:
 193         spin_lock(&all_mddevs_lock);
 194         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 195                 if (mddev->unit == unit) {
 196                         mddev_get(mddev);
 197                         spin_unlock(&all_mddevs_lock);
 198                         if (new)
 199                                 kfree(new);
 200                         return mddev;
 201                 }
 202
 203         if (new) {
 204                 list_add(&new->all_mddevs, &all_mddevs);
 205                 spin_unlock(&all_mddevs_lock);
 206                 return new;
 207         }
 208         spin_unlock(&all_mddevs_lock);
 209
 210         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 211         if (!new)
 212                 return NULL;
 213
 214         memset(new, 0, sizeof(*new));
 215
 216         new->unit = unit;
 217         if (MAJOR(unit) == MD_MAJOR)
 218                 new->md_minor = MINOR(unit);
 219         else
 220                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 221
 222         init_MUTEX(&new->reconfig_sem);
 223         INIT_LIST_HEAD(&new->disks);
 224         INIT_LIST_HEAD(&new->all_mddevs);
 225         init_timer(&new->safemode_timer);
 226         atomic_set(&new->active, 1);
 227         bio_list_init(&new->write_list);
 228         spin_lock_init(&new->write_lock);
 229
 230         new->queue = blk_alloc_queue(GFP_KERNEL);
 231         if (!new->queue) {
 232                 kfree(new);
 233                 return NULL;
 234         }
 235
 236         blk_queue_make_request(new->queue, md_fail_request);
 237
 238         goto retry;
 239 }
 240
 241 static inline int mddev_lock(mddev_t * mddev)
 242 {
 243         return down_interruptible(&mddev->reconfig_sem);
 244 }
 245
 246 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 247 {
 248         down(&mddev->reconfig_sem);
 249 }
 250
 251 static inline int mddev_trylock(mddev_t * mddev)
 252 {
 253         return down_trylock(&mddev->reconfig_sem);
 254 }
 255
 256 static inline void mddev_unlock(mddev_t * mddev)
 257 {
 258         up(&mddev->reconfig_sem);
 259
 260         if (mddev->thread)
 261                 md_wakeup_thread(mddev->thread);
 262 }
 263
 264 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 265 {
 266         mdk_rdev_t * rdev;
 267         struct list_head *tmp;
 268
 269         ITERATE_RDEV(mddev,rdev,tmp) {
 270                 if (rdev->desc_nr == nr)
 271                         return rdev;
 272         }
 273         return NULL;
 274 }
 275
 276 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 277 {
 278         struct list_head *tmp;
 279         mdk_rdev_t *rdev;
 280
 281         ITERATE_RDEV(mddev,rdev,tmp) {
 282                 if (rdev->bdev->bd_dev == dev)
 283                         return rdev;
 284         }
 285         return NULL;
 286 }
 287
 288 inline static sector_t calc_dev_sboffset(struct block_device *bdev)
 289 {
 290         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 291         return MD_NEW_SIZE_BLOCKS(size);
 292 }
 293
 294 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 295 {
 296         sector_t size;
 297
 298         size = rdev->sb_offset;
 299
 300         if (chunk_size)
 301                 size &= ~((sector_t)chunk_size/1024 - 1);
 302         return size;
 303 }
 304
 305 static int alloc_disk_sb(mdk_rdev_t * rdev)
 306 {
 307         if (rdev->sb_page)
 308                 MD_BUG();
 309
 310         rdev->sb_page = alloc_page(GFP_KERNEL);
 311         if (!rdev->sb_page) {
 312                 printk(KERN_ALERT "md: out of memory.\n");
 313                 return -EINVAL;
 314         }
 315
 316         return 0;
 317 }
 318
 319 static void free_disk_sb(mdk_rdev_t * rdev)
 320 {
 321         if (rdev->sb_page) {
 322                 page_cache_release(rdev->sb_page);
 323                 rdev->sb_loaded = 0;
 324                 rdev->sb_page = NULL;
 325                 rdev->sb_offset = 0;
 326                 rdev->size = 0;
 327         }
 328 }
 329
 330
 331 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 332 {
 333         if (bio->bi_size)
 334                 return 1;
 335
 336         complete((struct completion*)bio->bi_private);
 337         return 0;
 338 }
 339
 340 static int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 341                    struct page *page, int rw)
 342 {
 343         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 344         struct completion event;
 345         int ret;
 346
 347         rw |= (1 << BIO_RW_SYNC);
 348
 349         bio->bi_bdev = bdev;
 350         bio->bi_sector = sector;
 351         bio_add_page(bio, page, size, 0);
 352         init_completion(&event);
 353         bio->bi_private = &event;
 354         bio->bi_end_io = bi_complete;
 355         submit_bio(rw, bio);
 356         wait_for_completion(&event);
 357
 358         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 359         bio_put(bio);
 360         return ret;
 361 }
 362
 363 static int read_disk_sb(mdk_rdev_t * rdev)
 364 {
 365         char b[BDEVNAME_SIZE];
 366         if (!rdev->sb_page) {
 367                 MD_BUG();
 368                 return -EINVAL;
 369         }
 370         if (rdev->sb_loaded)
 371                 return 0;
 372
 373
 374         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
 375                 goto fail;
 376         rdev->sb_loaded = 1;
 377         return 0;
 378
 379 fail:
 380         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 381                 bdevname(rdev->bdev,b));
 382         return -EINVAL;
 383 }
 384
 385 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 386 {
 387         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 388                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 389                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 390                 (sb1->set_uuid3 == sb2->set_uuid3))
 391
 392                 return 1;
 393
 394         return 0;
 395 }
 396
 397
 398 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 399 {
 400         int ret;
 401         mdp_super_t *tmp1, *tmp2;
 402
 403         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 404         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 405
 406         if (!tmp1 || !tmp2) {
 407                 ret = 0;
 408                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 409                 goto abort;
 410         }
 411
 412         *tmp1 = *sb1;
 413         *tmp2 = *sb2;
 414
 415         /*
 416          * nr_disks is not constant
 417          */
 418         tmp1->nr_disks = 0;
 419         tmp2->nr_disks = 0;
 420
 421         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 422                 ret = 0;
 423         else
 424                 ret = 1;
 425
 426 abort:
 427         if (tmp1)
 428                 kfree(tmp1);
 429         if (tmp2)
 430                 kfree(tmp2);
 431
 432         return ret;
 433 }
 434
 435 static unsigned int calc_sb_csum(mdp_super_t * sb)
 436 {
 437         unsigned int disk_csum, csum;
 438
 439         disk_csum = sb->sb_csum;
 440         sb->sb_csum = 0;
 441         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 442         sb->sb_csum = disk_csum;
 443         return csum;
 444 }
 445
 446
 447 /*
 448  * Handle superblock details.
 449  * We want to be able to handle multiple superblock formats
 450  * so we have a common interface to them all, and an array of
 451  * different handlers.
 452  * We rely on user-space to write the initial superblock, and support
 453  * reading and updating of superblocks.
 454  * Interface methods are:
 455  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 456  *      loads and validates a superblock on dev.
 457  *      if refdev != NULL, compare superblocks on both devices
 458  *    Return:
 459  *      0 - dev has a superblock that is compatible with refdev
 460  *      1 - dev has a superblock that is compatible and newer than refdev
 461  *          so dev should be used as the refdev in future
 462  *     -EINVAL superblock incompatible or invalid
 463  *     -othererror e.g. -EIO
 464  *
 465  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 466  *      Verify that dev is acceptable into mddev.
 467  *       The first time, mddev->raid_disks will be 0, and data from
 468  *       dev should be merged in.  Subsequent calls check that dev
 469  *       is new enough.  Return 0 or -EINVAL
 470  *
 471  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 472  *     Update the superblock for rdev with data in mddev
 473  *     This does not write to disc.
 474  *
 475  */
 476
 477 struct super_type  {
 478         char            *name;
 479         struct module   *owner;
 480         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 481         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 482         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 483 };
 484
 485 /*
 486  * load_super for 0.90.0
 487  */
 488 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 489 {
 490         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 491         mdp_super_t *sb;
 492         int ret;
 493         sector_t sb_offset;
 494
 495         /*
 496          * Calculate the position of the superblock,
 497          * it's at the end of the disk.
 498          *
 499          * It also happens to be a multiple of 4Kb.
 500          */
 501         sb_offset = calc_dev_sboffset(rdev->bdev);
 502         rdev->sb_offset = sb_offset;
 503
 504         ret = read_disk_sb(rdev);
 505         if (ret) return ret;
 506
 507         ret = -EINVAL;
 508
 509         bdevname(rdev->bdev, b);
 510         sb = (mdp_super_t*)page_address(rdev->sb_page);
 511
 512         if (sb->md_magic != MD_SB_MAGIC) {
 513                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 514                        b);
 515                 goto abort;
 516         }
 517
 518         if (sb->major_version != 0 ||
 519             sb->minor_version != 90) {
 520                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 521                         sb->major_version, sb->minor_version,
 522                         b);
 523                 goto abort;
 524         }
 525
 526         if (sb->raid_disks <= 0)
 527                 goto abort;
 528
 529         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 530                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 531                         b);
 532                 goto abort;
 533         }
 534
 535         rdev->preferred_minor = sb->md_minor;
 536         rdev->data_offset = 0;
 537
 538         if (sb->level == LEVEL_MULTIPATH)
 539                 rdev->desc_nr = -1;
 540         else
 541                 rdev->desc_nr = sb->this_disk.number;
 542
 543         if (refdev == 0)
 544                 ret = 1;
 545         else {
 546                 __u64 ev1, ev2;
 547                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 548                 if (!uuid_equal(refsb, sb)) {
 549                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 550                                 b, bdevname(refdev->bdev,b2));
 551                         goto abort;
 552                 }
 553                 if (!sb_equal(refsb, sb)) {
 554                         printk(KERN_WARNING "md: %s has same UUID"
 555                                " but different superblock to %s\n",
 556                                b, bdevname(refdev->bdev, b2));
 557                         goto abort;
 558                 }
 559                 ev1 = md_event(sb);
 560                 ev2 = md_event(refsb);
 561                 if (ev1 > ev2)
 562                         ret = 1;
 563                 else
 564                         ret = 0;
 565         }
 566         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 567
 568  abort:
 569         return ret;
 570 }
 571
 572 /*
 573  * validate_super for 0.90.0
 574  */
 575 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 576 {
 577         mdp_disk_t *desc;
 578         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 579
 580         rdev->raid_disk = -1;
 581         rdev->in_sync = 0;
 582         if (mddev->raid_disks == 0) {
 583                 mddev->major_version = 0;
 584                 mddev->minor_version = sb->minor_version;
 585                 mddev->patch_version = sb->patch_version;
 586                 mddev->persistent = ! sb->not_persistent;
 587                 mddev->chunk_size = sb->chunk_size;
 588                 mddev->ctime = sb->ctime;
 589                 mddev->utime = sb->utime;
 590                 mddev->level = sb->level;
 591                 mddev->layout = sb->layout;
 592                 mddev->raid_disks = sb->raid_disks;
 593                 mddev->size = sb->size;
 594                 mddev->events = md_event(sb);
 595
 596                 if (sb->state & (1<<MD_SB_CLEAN))
 597                         mddev->recovery_cp = MaxSector;
 598                 else {
 599                         if (sb->events_hi == sb->cp_events_hi &&
 600                                 sb->events_lo == sb->cp_events_lo) {
 601                                 mddev->recovery_cp = sb->recovery_cp;
 602                         } else
 603                                 mddev->recovery_cp = 0;
 604                 }
 605
 606                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 607                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 608                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 609                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 610
 611                 mddev->max_disks = MD_SB_DISKS;
 612         } else if (mddev->pers == NULL) {
 613                 /* Insist on good event counter while assembling */
 614                 __u64 ev1 = md_event(sb);
 615                 ++ev1;
 616                 if (ev1 < mddev->events)
 617                         return -EINVAL;
 618         } else if (mddev->bitmap) {
 619                 /* if adding to array with a bitmap, then we can accept an
 620                  * older device ... but not too old.
 621                  */
 622                 __u64 ev1 = md_event(sb);
 623                 if (ev1 < mddev->bitmap->events_cleared)
 624                         return 0;
 625         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 626                 return 0;
 627
 628         if (mddev->level != LEVEL_MULTIPATH) {
 629                 rdev->faulty = 0;
 630                 desc = sb->disks + rdev->desc_nr;
 631
 632                 if (desc->state & (1<<MD_DISK_FAULTY))
 633                         rdev->faulty = 1;
 634                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 635                          desc->raid_disk < mddev->raid_disks) {
 636                         rdev->in_sync = 1;
 637                         rdev->raid_disk = desc->raid_disk;
 638                 }
 639         } else /* MULTIPATH are always insync */
 640                 rdev->in_sync = 1;
 641         return 0;
 642 }
 643
 644 /*
 645  * sync_super for 0.90.0
 646  */
 647 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 648 {
 649         mdp_super_t *sb;
 650         struct list_head *tmp;
 651         mdk_rdev_t *rdev2;
 652         int next_spare = mddev->raid_disks;
 653
 654         /* make rdev->sb match mddev data..
 655          *
 656          * 1/ zero out disks
 657          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 658          * 3/ any empty disks < next_spare become removed
 659          *
 660          * disks[0] gets initialised to REMOVED because
 661          * we cannot be sure from other fields if it has
 662          * been initialised or not.
 663          */
 664         int i;
 665         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 666
 667         sb = (mdp_super_t*)page_address(rdev->sb_page);
 668
 669         memset(sb, 0, sizeof(*sb));
 670
 671         sb->md_magic = MD_SB_MAGIC;
 672         sb->major_version = mddev->major_version;
 673         sb->minor_version = mddev->minor_version;
 674         sb->patch_version = mddev->patch_version;
 675         sb->gvalid_words  = 0; /* ignored */
 676         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 677         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 678         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 679         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 680
 681         sb->ctime = mddev->ctime;
 682         sb->level = mddev->level;
 683         sb->size  = mddev->size;
 684         sb->raid_disks = mddev->raid_disks;
 685         sb->md_minor = mddev->md_minor;
 686         sb->not_persistent = !mddev->persistent;
 687         sb->utime = mddev->utime;
 688         sb->state = 0;
 689         sb->events_hi = (mddev->events>>32);
 690         sb->events_lo = (u32)mddev->events;
 691
 692         if (mddev->in_sync)
 693         {
 694                 sb->recovery_cp = mddev->recovery_cp;
 695                 sb->cp_events_hi = (mddev->events>>32);
 696                 sb->cp_events_lo = (u32)mddev->events;
 697                 if (mddev->recovery_cp == MaxSector)
 698                         sb->state = (1<< MD_SB_CLEAN);
 699         } else
 700                 sb->recovery_cp = 0;
 701
 702         sb->layout = mddev->layout;
 703         sb->chunk_size = mddev->chunk_size;
 704
 705         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 706         ITERATE_RDEV(mddev,rdev2,tmp) {
 707                 mdp_disk_t *d;
 708                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 709                         rdev2->desc_nr = rdev2->raid_disk;
 710                 else
 711                         rdev2->desc_nr = next_spare++;
 712                 d = &sb->disks[rdev2->desc_nr];
 713                 nr_disks++;
 714                 d->number = rdev2->desc_nr;
 715                 d->major = MAJOR(rdev2->bdev->bd_dev);
 716                 d->minor = MINOR(rdev2->bdev->bd_dev);
 717                 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
 718                         d->raid_disk = rdev2->raid_disk;
 719                 else
 720                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 721                 if (rdev2->faulty) {
 722                         d->state = (1<<MD_DISK_FAULTY);
 723                         failed++;
 724                 } else if (rdev2->in_sync) {
 725                         d->state = (1<<MD_DISK_ACTIVE);
 726                         d->state |= (1<<MD_DISK_SYNC);
 727                         active++;
 728                         working++;
 729                 } else {
 730                         d->state = 0;
 731                         spare++;
 732                         working++;
 733                 }
 734         }
 735
 736         /* now set the "removed" and "faulty" bits on any missing devices */
 737         for (i=0 ; i < mddev->raid_disks ; i++) {
 738                 mdp_disk_t *d = &sb->disks[i];
 739                 if (d->state == 0 && d->number == 0) {
 740                         d->number = i;
 741                         d->raid_disk = i;
 742                         d->state = (1<<MD_DISK_REMOVED);
 743                         d->state |= (1<<MD_DISK_FAULTY);
 744                         failed++;
 745                 }
 746         }
 747         sb->nr_disks = nr_disks;
 748         sb->active_disks = active;
 749         sb->working_disks = working;
 750         sb->failed_disks = failed;
 751         sb->spare_disks = spare;
 752
 753         sb->this_disk = sb->disks[rdev->desc_nr];
 754         sb->sb_csum = calc_sb_csum(sb);
 755 }
 756
 757 /*
 758  * version 1 superblock
 759  */
 760
 761 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 762 {
 763         unsigned int disk_csum, csum;
 764         unsigned long long newcsum;
 765         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 766         unsigned int *isuper = (unsigned int*)sb;
 767         int i;
 768
 769         disk_csum = sb->sb_csum;
 770         sb->sb_csum = 0;
 771         newcsum = 0;
 772         for (i=0; size>=4; size -= 4 )
 773                 newcsum += le32_to_cpu(*isuper++);
 774
 775         if (size == 2)
 776                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 777
 778         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 779         sb->sb_csum = disk_csum;
 780         return cpu_to_le32(csum);
 781 }
 782
 783 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 784 {
 785         struct mdp_superblock_1 *sb;
 786         int ret;
 787         sector_t sb_offset;
 788         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 789
 790         /*
 791          * Calculate the position of the superblock.
 792          * It is always aligned to a 4K boundary and
 793          * depeding on minor_version, it can be:
 794          * 0: At least 8K, but less than 12K, from end of device
 795          * 1: At start of device
 796          * 2: 4K from start of device.
 797          */
 798         switch(minor_version) {
 799         case 0:
 800                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 801                 sb_offset -= 8*2;
 802                 sb_offset &= ~(4*2-1);
 803                 /* convert from sectors to K */
 804                 sb_offset /= 2;
 805                 break;
 806         case 1:
 807                 sb_offset = 0;
 808                 break;
 809         case 2:
 810                 sb_offset = 4;
 811                 break;
 812         default:
 813                 return -EINVAL;
 814         }
 815         rdev->sb_offset = sb_offset;
 816
 817         ret = read_disk_sb(rdev);
 818         if (ret) return ret;
 819
 820
 821         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 822
 823         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 824             sb->major_version != cpu_to_le32(1) ||
 825             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 826             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 827             sb->feature_map != 0)
 828                 return -EINVAL;
 829
 830         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 831                 printk("md: invalid superblock checksum on %s\n",
 832                         bdevname(rdev->bdev,b));
 833                 return -EINVAL;
 834         }
 835         if (le64_to_cpu(sb->data_size) < 10) {
 836                 printk("md: data_size too small on %s\n",
 837                        bdevname(rdev->bdev,b));
 838                 return -EINVAL;
 839         }
 840         rdev->preferred_minor = 0xffff;
 841         rdev->data_offset = le64_to_cpu(sb->data_offset);
 842
 843         if (refdev == 0)
 844                 return 1;
 845         else {
 846                 __u64 ev1, ev2;
 847                 struct mdp_superblock_1 *refsb =
 848                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 849
 850                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 851                     sb->level != refsb->level ||
 852                     sb->layout != refsb->layout ||
 853                     sb->chunksize != refsb->chunksize) {
 854                         printk(KERN_WARNING "md: %s has strangely different"
 855                                 " superblock to %s\n",
 856                                 bdevname(rdev->bdev,b),
 857                                 bdevname(refdev->bdev,b2));
 858                         return -EINVAL;
 859                 }
 860                 ev1 = le64_to_cpu(sb->events);
 861                 ev2 = le64_to_cpu(refsb->events);
 862
 863                 if (ev1 > ev2)
 864                         return 1;
 865         }
 866         if (minor_version)
 867                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 868         else
 869                 rdev->size = rdev->sb_offset;
 870         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 871                 return -EINVAL;
 872         rdev->size = le64_to_cpu(sb->data_size)/2;
 873         if (le32_to_cpu(sb->chunksize))
 874                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 875         return 0;
 876 }
 877
 878 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 879 {
 880         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 881
 882         rdev->raid_disk = -1;
 883         rdev->in_sync = 0;
 884         if (mddev->raid_disks == 0) {
 885                 mddev->major_version = 1;
 886                 mddev->patch_version = 0;
 887                 mddev->persistent = 1;
 888                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 889                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 890                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 891                 mddev->level = le32_to_cpu(sb->level);
 892                 mddev->layout = le32_to_cpu(sb->layout);
 893                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 894                 mddev->size = le64_to_cpu(sb->size)/2;
 895                 mddev->events = le64_to_cpu(sb->events);
 896
 897                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 898                 memcpy(mddev->uuid, sb->set_uuid, 16);
 899
 900                 mddev->max_disks =  (4096-256)/2;
 901         } else if (mddev->pers == NULL) {
 902                 /* Insist of good event counter while assembling */
 903                 __u64 ev1 = le64_to_cpu(sb->events);
 904                 ++ev1;
 905                 if (ev1 < mddev->events)
 906                         return -EINVAL;
 907         } else if (mddev->bitmap) {
 908                 /* If adding to array with a bitmap, then we can accept an
 909                  * older device, but not too old.
 910                  */
 911                 __u64 ev1 = le64_to_cpu(sb->events);
 912                 if (ev1 < mddev->bitmap->events_cleared)
 913                         return 0;
 914         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 915                 return 0;
 916
 917         if (mddev->level != LEVEL_MULTIPATH) {
 918                 int role;
 919                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
 920                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
 921                 switch(role) {
 922                 case 0xffff: /* spare */
 923                         rdev->faulty = 0;
 924                         break;
 925                 case 0xfffe: /* faulty */
 926                         rdev->faulty = 1;
 927                         break;
 928                 default:
 929                         rdev->in_sync = 1;
 930                         rdev->faulty = 0;
 931                         rdev->raid_disk = role;
 932                         break;
 933                 }
 934         } else /* MULTIPATH are always insync */
 935                 rdev->in_sync = 1;
 936
 937         return 0;
 938 }
 939
 940 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 941 {
 942         struct mdp_superblock_1 *sb;
 943         struct list_head *tmp;
 944         mdk_rdev_t *rdev2;
 945         int max_dev, i;
 946         /* make rdev->sb match mddev and rdev data. */
 947
 948         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 949
 950         sb->feature_map = 0;
 951         sb->pad0 = 0;
 952         memset(sb->pad1, 0, sizeof(sb->pad1));
 953         memset(sb->pad2, 0, sizeof(sb->pad2));
 954         memset(sb->pad3, 0, sizeof(sb->pad3));
 955
 956         sb->utime = cpu_to_le64((__u64)mddev->utime);
 957         sb->events = cpu_to_le64(mddev->events);
 958         if (mddev->in_sync)
 959                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
 960         else
 961                 sb->resync_offset = cpu_to_le64(0);
 962
 963         max_dev = 0;
 964         ITERATE_RDEV(mddev,rdev2,tmp)
 965                 if (rdev2->desc_nr+1 > max_dev)
 966                         max_dev = rdev2->desc_nr+1;
 967
 968         sb->max_dev = cpu_to_le32(max_dev);
 969         for (i=0; i<max_dev;i++)
 970                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
 971
 972         ITERATE_RDEV(mddev,rdev2,tmp) {
 973                 i = rdev2->desc_nr;
 974                 if (rdev2->faulty)
 975                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
 976                 else if (rdev2->in_sync)
 977                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
 978                 else
 979                         sb->dev_roles[i] = cpu_to_le16(0xffff);
 980         }
 981
 982         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
 983         sb->sb_csum = calc_sb_1_csum(sb);
 984 }
 985
 986
 987 static struct super_type super_types[] = {
 988         [0] = {
 989                 .name   = "0.90.0",
 990                 .owner  = THIS_MODULE,
 991                 .load_super     = super_90_load,
 992                 .validate_super = super_90_validate,
 993                 .sync_super     = super_90_sync,
 994         },
 995         [1] = {
 996                 .name   = "md-1",
 997                 .owner  = THIS_MODULE,
 998                 .load_super     = super_1_load,
 999                 .validate_super = super_1_validate,
1000                 .sync_super     = super_1_sync,
1001         },
1002 };
1003
1004 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1005 {
1006         struct list_head *tmp;
1007         mdk_rdev_t *rdev;
1008
1009         ITERATE_RDEV(mddev,rdev,tmp)
1010                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1011                         return rdev;
1012
1013         return NULL;
1014 }
1015
1016 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1017 {
1018         struct list_head *tmp;
1019         mdk_rdev_t *rdev;
1020
1021         ITERATE_RDEV(mddev1,rdev,tmp)
1022                 if (match_dev_unit(mddev2, rdev))
1023                         return 1;
1024
1025         return 0;
1026 }
1027
1028 static LIST_HEAD(pending_raid_disks);
1029
1030 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1031 {
1032         mdk_rdev_t *same_pdev;
1033         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1034
1035         if (rdev->mddev) {
1036                 MD_BUG();
1037                 return -EINVAL;
1038         }
1039         same_pdev = match_dev_unit(mddev, rdev);
1040         if (same_pdev)
1041                 printk(KERN_WARNING
1042                         "%s: WARNING: %s appears to be on the same physical"
1043                         " disk as %s. True\n     protection against single-disk"
1044                         " failure might be compromised.\n",
1045                         mdname(mddev), bdevname(rdev->bdev,b),
1046                         bdevname(same_pdev->bdev,b2));
1047
1048         /* Verify rdev->desc_nr is unique.
1049          * If it is -1, assign a free number, else
1050          * check number is not in use
1051          */
1052         if (rdev->desc_nr < 0) {
1053                 int choice = 0;
1054                 if (mddev->pers) choice = mddev->raid_disks;
1055                 while (find_rdev_nr(mddev, choice))
1056                         choice++;
1057                 rdev->desc_nr = choice;
1058         } else {
1059                 if (find_rdev_nr(mddev, rdev->desc_nr))
1060                         return -EBUSY;
1061         }
1062
1063         list_add(&rdev->same_set, &mddev->disks);
1064         rdev->mddev = mddev;
1065         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1066         return 0;
1067 }
1068
1069 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1070 {
1071         char b[BDEVNAME_SIZE];
1072         if (!rdev->mddev) {
1073                 MD_BUG();
1074                 return;
1075         }
1076         list_del_init(&rdev->same_set);
1077         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1078         rdev->mddev = NULL;
1079 }
1080
1081 /*
1082  * prevent the device from being mounted, repartitioned or
1083  * otherwise reused by a RAID array (or any other kernel
1084  * subsystem), by bd_claiming the device.
1085  */
1086 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1087 {
1088         int err = 0;
1089         struct block_device *bdev;
1090         char b[BDEVNAME_SIZE];
1091
1092         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1093         if (IS_ERR(bdev)) {
1094                 printk(KERN_ERR "md: could not open %s.\n",
1095                         __bdevname(dev, b));
1096                 return PTR_ERR(bdev);
1097         }
1098         err = bd_claim(bdev, rdev);
1099         if (err) {
1100                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1101                         bdevname(bdev, b));
1102                 blkdev_put(bdev);
1103                 return err;
1104         }
1105         rdev->bdev = bdev;
1106         return err;
1107 }
1108
1109 static void unlock_rdev(mdk_rdev_t *rdev)
1110 {
1111         struct block_device *bdev = rdev->bdev;
1112         rdev->bdev = NULL;
1113         if (!bdev)
1114                 MD_BUG();
1115         bd_release(bdev);
1116         blkdev_put(bdev);
1117 }
1118
1119 void md_autodetect_dev(dev_t dev);
1120
1121 static void export_rdev(mdk_rdev_t * rdev)
1122 {
1123         char b[BDEVNAME_SIZE];
1124         printk(KERN_INFO "md: export_rdev(%s)\n",
1125                 bdevname(rdev->bdev,b));
1126         if (rdev->mddev)
1127                 MD_BUG();
1128         free_disk_sb(rdev);
1129         list_del_init(&rdev->same_set);
1130 #ifndef MODULE
1131         md_autodetect_dev(rdev->bdev->bd_dev);
1132 #endif
1133         unlock_rdev(rdev);
1134         kfree(rdev);
1135 }
1136
1137 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1138 {
1139         unbind_rdev_from_array(rdev);
1140         export_rdev(rdev);
1141 }
1142
1143 static void export_array(mddev_t *mddev)
1144 {
1145         struct list_head *tmp;
1146         mdk_rdev_t *rdev;
1147
1148         ITERATE_RDEV(mddev,rdev,tmp) {
1149                 if (!rdev->mddev) {
1150                         MD_BUG();
1151                         continue;
1152                 }
1153                 kick_rdev_from_array(rdev);
1154         }
1155         if (!list_empty(&mddev->disks))
1156                 MD_BUG();
1157         mddev->raid_disks = 0;
1158         mddev->major_version = 0;
1159 }
1160
1161 static void print_desc(mdp_disk_t *desc)
1162 {
1163         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1164                 desc->major,desc->minor,desc->raid_disk,desc->state);
1165 }
1166
1167 static void print_sb(mdp_super_t *sb)
1168 {
1169         int i;
1170
1171         printk(KERN_INFO
1172                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1173                 sb->major_version, sb->minor_version, sb->patch_version,
1174                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1175                 sb->ctime);
1176         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1177                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1178                 sb->md_minor, sb->layout, sb->chunk_size);
1179         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1180                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1181                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1182                 sb->failed_disks, sb->spare_disks,
1183                 sb->sb_csum, (unsigned long)sb->events_lo);
1184
1185         printk(KERN_INFO);
1186         for (i = 0; i < MD_SB_DISKS; i++) {
1187                 mdp_disk_t *desc;
1188
1189                 desc = sb->disks + i;
1190                 if (desc->number || desc->major || desc->minor ||
1191                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1192                         printk("     D %2d: ", i);
1193                         print_desc(desc);
1194                 }
1195         }
1196         printk(KERN_INFO "md:     THIS: ");
1197         print_desc(&sb->this_disk);
1198
1199 }
1200
1201 static void print_rdev(mdk_rdev_t *rdev)
1202 {
1203         char b[BDEVNAME_SIZE];
1204         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1205                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1206                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1207         if (rdev->sb_loaded) {
1208                 printk(KERN_INFO "md: rdev superblock:\n");
1209                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1210         } else
1211                 printk(KERN_INFO "md: no rdev superblock!\n");
1212 }
1213
1214 void md_print_devices(void)
1215 {
1216         struct list_head *tmp, *tmp2;
1217         mdk_rdev_t *rdev;
1218         mddev_t *mddev;
1219         char b[BDEVNAME_SIZE];
1220
1221         printk("\n");
1222         printk("md:     **********************************\n");
1223         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1224         printk("md:     **********************************\n");
1225         ITERATE_MDDEV(mddev,tmp) {
1226
1227                 if (mddev->bitmap)
1228                         bitmap_print_sb(mddev->bitmap);
1229                 else
1230                         printk("%s: ", mdname(mddev));
1231                 ITERATE_RDEV(mddev,rdev,tmp2)
1232                         printk("<%s>", bdevname(rdev->bdev,b));
1233                 printk("\n");
1234
1235                 ITERATE_RDEV(mddev,rdev,tmp2)
1236                         print_rdev(rdev);
1237         }
1238         printk("md:     **********************************\n");
1239         printk("\n");
1240 }
1241
1242
1243 static int write_disk_sb(mdk_rdev_t * rdev)
1244 {
1245         char b[BDEVNAME_SIZE];
1246         if (!rdev->sb_loaded) {
1247                 MD_BUG();
1248                 return 1;
1249         }
1250         if (rdev->faulty) {
1251                 MD_BUG();
1252                 return 1;
1253         }
1254
1255         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1256                 bdevname(rdev->bdev,b),
1257                (unsigned long long)rdev->sb_offset);
1258
1259         if (sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, WRITE))
1260                 return 0;
1261
1262         printk("md: write_disk_sb failed for device %s\n",
1263                 bdevname(rdev->bdev,b));
1264         return 1;
1265 }
1266
1267 static void sync_sbs(mddev_t * mddev)
1268 {
1269         mdk_rdev_t *rdev;
1270         struct list_head *tmp;
1271
1272         ITERATE_RDEV(mddev,rdev,tmp) {
1273                 super_types[mddev->major_version].
1274                         sync_super(mddev, rdev);
1275                 rdev->sb_loaded = 1;
1276         }
1277 }
1278
1279 static void md_update_sb(mddev_t * mddev)
1280 {
1281         int err, count = 100;
1282         struct list_head *tmp;
1283         mdk_rdev_t *rdev;
1284         int sync_req;
1285
1286 repeat:
1287         spin_lock(&mddev->write_lock);
1288         sync_req = mddev->in_sync;
1289         mddev->utime = get_seconds();
1290         mddev->events ++;
1291
1292         if (!mddev->events) {
1293                 /*
1294                  * oops, this 64-bit counter should never wrap.
1295                  * Either we are in around ~1 trillion A.C., assuming
1296                  * 1 reboot per second, or we have a bug:
1297                  */
1298                 MD_BUG();
1299                 mddev->events --;
1300         }
1301         sync_sbs(mddev);
1302
1303         /*
1304          * do not write anything to disk if using
1305          * nonpersistent superblocks
1306          */
1307         if (!mddev->persistent) {
1308                 mddev->sb_dirty = 0;
1309                 spin_unlock(&mddev->write_lock);
1310                 return;
1311         }
1312         spin_unlock(&mddev->write_lock);
1313
1314         dprintk(KERN_INFO
1315                 "md: updating %s RAID superblock on device (in sync %d)\n",
1316                 mdname(mddev),mddev->in_sync);
1317
1318         err = bitmap_update_sb(mddev->bitmap);
1319         ITERATE_RDEV(mddev,rdev,tmp) {
1320                 char b[BDEVNAME_SIZE];
1321                 dprintk(KERN_INFO "md: ");
1322                 if (rdev->faulty)
1323                         dprintk("(skipping faulty ");
1324
1325                 dprintk("%s ", bdevname(rdev->bdev,b));
1326                 if (!rdev->faulty) {
1327                         err += write_disk_sb(rdev);
1328                 } else
1329                         dprintk(")\n");
1330                 if (!err && mddev->level == LEVEL_MULTIPATH)
1331                         /* only need to write one superblock... */
1332                         break;
1333         }
1334         if (err) {
1335                 if (--count) {
1336                         printk(KERN_ERR "md: errors occurred during superblock"
1337                                 " update, repeating\n");
1338                         goto repeat;
1339                 }
1340                 printk(KERN_ERR \
1341                         "md: excessive errors occurred during superblock update, exiting\n");
1342         }
1343         spin_lock(&mddev->write_lock);
1344         if (mddev->in_sync != sync_req) {
1345                 /* have to write it out again */
1346                 spin_unlock(&mddev->write_lock);
1347                 goto repeat;
1348         }
1349         mddev->sb_dirty = 0;
1350         spin_unlock(&mddev->write_lock);
1351
1352 }
1353
1354 /*
1355  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1356  *
1357  * mark the device faulty if:
1358  *
1359  *   - the device is nonexistent (zero size)
1360  *   - the device has no valid superblock
1361  *
1362  * a faulty rdev _never_ has rdev->sb set.
1363  */
1364 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1365 {
1366         char b[BDEVNAME_SIZE];
1367         int err;
1368         mdk_rdev_t *rdev;
1369         sector_t size;
1370
1371         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1372         if (!rdev) {
1373                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1374                 return ERR_PTR(-ENOMEM);
1375         }
1376         memset(rdev, 0, sizeof(*rdev));
1377
1378         if ((err = alloc_disk_sb(rdev)))
1379                 goto abort_free;
1380
1381         err = lock_rdev(rdev, newdev);
1382         if (err)
1383                 goto abort_free;
1384
1385         rdev->desc_nr = -1;
1386         rdev->faulty = 0;
1387         rdev->in_sync = 0;
1388         rdev->data_offset = 0;
1389         atomic_set(&rdev->nr_pending, 0);
1390
1391         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1392         if (!size) {
1393                 printk(KERN_WARNING
1394                         "md: %s has zero or unknown size, marking faulty!\n",
1395                         bdevname(rdev->bdev,b));
1396                 err = -EINVAL;
1397                 goto abort_free;
1398         }
1399
1400         if (super_format >= 0) {
1401                 err = super_types[super_format].
1402                         load_super(rdev, NULL, super_minor);
1403                 if (err == -EINVAL) {
1404                         printk(KERN_WARNING
1405                                 "md: %s has invalid sb, not importing!\n",
1406                                 bdevname(rdev->bdev,b));
1407                         goto abort_free;
1408                 }
1409                 if (err < 0) {
1410                         printk(KERN_WARNING
1411                                 "md: could not read %s's sb, not importing!\n",
1412                                 bdevname(rdev->bdev,b));
1413                         goto abort_free;
1414                 }
1415         }
1416         INIT_LIST_HEAD(&rdev->same_set);
1417
1418         return rdev;
1419
1420 abort_free:
1421         if (rdev->sb_page) {
1422                 if (rdev->bdev)
1423                         unlock_rdev(rdev);
1424                 free_disk_sb(rdev);
1425         }
1426         kfree(rdev);
1427         return ERR_PTR(err);
1428 }
1429
1430 /*
1431  * Check a full RAID array for plausibility
1432  */
1433
1434
1435 static void analyze_sbs(mddev_t * mddev)
1436 {
1437         int i;
1438         struct list_head *tmp;
1439         mdk_rdev_t *rdev, *freshest;
1440         char b[BDEVNAME_SIZE];
1441
1442         freshest = NULL;
1443         ITERATE_RDEV(mddev,rdev,tmp)
1444                 switch (super_types[mddev->major_version].
1445                         load_super(rdev, freshest, mddev->minor_version)) {
1446                 case 1:
1447                         freshest = rdev;
1448                         break;
1449                 case 0:
1450                         break;
1451                 default:
1452                         printk( KERN_ERR \
1453                                 "md: fatal superblock inconsistency in %s"
1454                                 " -- removing from array\n",
1455                                 bdevname(rdev->bdev,b));
1456                         kick_rdev_from_array(rdev);
1457                 }
1458
1459
1460         super_types[mddev->major_version].
1461                 validate_super(mddev, freshest);
1462
1463         i = 0;
1464         ITERATE_RDEV(mddev,rdev,tmp) {
1465                 if (rdev != freshest)
1466                         if (super_types[mddev->major_version].
1467                             validate_super(mddev, rdev)) {
1468                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1469                                         " from array!\n",
1470                                         bdevname(rdev->bdev,b));
1471                                 kick_rdev_from_array(rdev);
1472                                 continue;
1473                         }
1474                 if (mddev->level == LEVEL_MULTIPATH) {
1475                         rdev->desc_nr = i++;
1476                         rdev->raid_disk = rdev->desc_nr;
1477                         rdev->in_sync = 1;
1478                 }
1479         }
1480
1481
1482
1483         if (mddev->recovery_cp != MaxSector &&
1484             mddev->level >= 1)
1485                 printk(KERN_ERR "md: %s: raid array is not clean"
1486                        " -- starting background reconstruction\n",
1487                        mdname(mddev));
1488
1489 }
1490
1491 int mdp_major = 0;
1492
1493 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1494 {
1495         static DECLARE_MUTEX(disks_sem);
1496         mddev_t *mddev = mddev_find(dev);
1497         struct gendisk *disk;
1498         int partitioned = (MAJOR(dev) != MD_MAJOR);
1499         int shift = partitioned ? MdpMinorShift : 0;
1500         int unit = MINOR(dev) >> shift;
1501
1502         if (!mddev)
1503                 return NULL;
1504
1505         down(&disks_sem);
1506         if (mddev->gendisk) {
1507                 up(&disks_sem);
1508                 mddev_put(mddev);
1509                 return NULL;
1510         }
1511         disk = alloc_disk(1 << shift);
1512         if (!disk) {
1513                 up(&disks_sem);
1514                 mddev_put(mddev);
1515                 return NULL;
1516         }
1517         disk->major = MAJOR(dev);
1518         disk->first_minor = unit << shift;
1519         if (partitioned) {
1520                 sprintf(disk->disk_name, "md_d%d", unit);
1521                 sprintf(disk->devfs_name, "md/d%d", unit);
1522         } else {
1523                 sprintf(disk->disk_name, "md%d", unit);
1524                 sprintf(disk->devfs_name, "md/%d", unit);
1525         }
1526         disk->fops = &md_fops;
1527         disk->private_data = mddev;
1528         disk->queue = mddev->queue;
1529         add_disk(disk);
1530         mddev->gendisk = disk;
1531         up(&disks_sem);
1532         return NULL;
1533 }
1534
1535 void md_wakeup_thread(mdk_thread_t *thread);
1536
1537 static void md_safemode_timeout(unsigned long data)
1538 {
1539         mddev_t *mddev = (mddev_t *) data;
1540
1541         mddev->safemode = 1;
1542         md_wakeup_thread(mddev->thread);
1543 }
1544
1545
1546 static int do_md_run(mddev_t * mddev)
1547 {
1548         int pnum, err;
1549         int chunk_size;
1550         struct list_head *tmp;
1551         mdk_rdev_t *rdev;
1552         struct gendisk *disk;
1553         char b[BDEVNAME_SIZE];
1554
1555         if (list_empty(&mddev->disks))
1556                 /* cannot run an array with no devices.. */
1557                 return -EINVAL;
1558
1559         if (mddev->pers)
1560                 return -EBUSY;
1561
1562         /*
1563          * Analyze all RAID superblock(s)
1564          */
1565         if (!mddev->raid_disks)
1566                 analyze_sbs(mddev);
1567
1568         chunk_size = mddev->chunk_size;
1569         pnum = level_to_pers(mddev->level);
1570
1571         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1572                 if (!chunk_size) {
1573                         /*
1574                          * 'default chunksize' in the old md code used to
1575                          * be PAGE_SIZE, baaad.
1576                          * we abort here to be on the safe side. We don't
1577                          * want to continue the bad practice.
1578                          */
1579                         printk(KERN_ERR
1580                                 "no chunksize specified, see 'man raidtab'\n");
1581                         return -EINVAL;
1582                 }
1583                 if (chunk_size > MAX_CHUNK_SIZE) {
1584                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1585                                 chunk_size, MAX_CHUNK_SIZE);
1586                         return -EINVAL;
1587                 }
1588                 /*
1589                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1590                  */
1591                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1592                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1593                         return -EINVAL;
1594                 }
1595                 if (chunk_size < PAGE_SIZE) {
1596                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1597                                 chunk_size, PAGE_SIZE);
1598                         return -EINVAL;
1599                 }
1600
1601                 /* devices must have minimum size of one chunk */
1602                 ITERATE_RDEV(mddev,rdev,tmp) {
1603                         if (rdev->faulty)
1604                                 continue;
1605                         if (rdev->size < chunk_size / 1024) {
1606                                 printk(KERN_WARNING
1607                                         "md: Dev %s smaller than chunk_size:"
1608                                         " %lluk < %dk\n",
1609                                         bdevname(rdev->bdev,b),
1610                                         (unsigned long long)rdev->size,
1611                                         chunk_size / 1024);
1612                                 return -EINVAL;
1613                         }
1614                 }
1615         }
1616
1617 #ifdef CONFIG_KMOD
1618         if (!pers[pnum])
1619         {
1620                 request_module("md-personality-%d", pnum);
1621         }
1622 #endif
1623
1624         /*
1625          * Drop all container device buffers, from now on
1626          * the only valid external interface is through the md
1627          * device.
1628          * Also find largest hardsector size
1629          */
1630         ITERATE_RDEV(mddev,rdev,tmp) {
1631                 if (rdev->faulty)
1632                         continue;
1633                 sync_blockdev(rdev->bdev);
1634                 invalidate_bdev(rdev->bdev, 0);
1635         }
1636
1637         md_probe(mddev->unit, NULL, NULL);
1638         disk = mddev->gendisk;
1639         if (!disk)
1640                 return -ENOMEM;
1641
1642         spin_lock(&pers_lock);
1643         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1644                 spin_unlock(&pers_lock);
1645                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1646                        pnum);
1647                 return -EINVAL;
1648         }
1649
1650         mddev->pers = pers[pnum];
1651         spin_unlock(&pers_lock);
1652
1653         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1654
1655         /* before we start the array running, initialise the bitmap */
1656         err = bitmap_create(mddev);
1657         if (err)
1658                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1659                         mdname(mddev), err);
1660         else
1661                 err = mddev->pers->run(mddev);
1662         if (err) {
1663                 printk(KERN_ERR "md: pers->run() failed ...\n");
1664                 module_put(mddev->pers->owner);
1665                 mddev->pers = NULL;
1666                 bitmap_destroy(mddev);
1667                 return err;
1668         }
1669         atomic_set(&mddev->writes_pending,0);
1670         mddev->safemode = 0;
1671         mddev->safemode_timer.function = md_safemode_timeout;
1672         mddev->safemode_timer.data = (unsigned long) mddev;
1673         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1674         mddev->in_sync = 1;
1675
1676         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1677
1678         if (mddev->sb_dirty)
1679                 md_update_sb(mddev);
1680
1681         set_capacity(disk, mddev->array_size<<1);
1682
1683         /* If we call blk_queue_make_request here, it will
1684          * re-initialise max_sectors etc which may have been
1685          * refined inside -> run.  So just set the bits we need to set.
1686          * Most initialisation happended when we called
1687          * blk_queue_make_request(..., md_fail_request)
1688          * earlier.
1689          */
1690         mddev->queue->queuedata = mddev;
1691         mddev->queue->make_request_fn = mddev->pers->make_request;
1692
1693         mddev->changed = 1;
1694         return 0;
1695 }
1696
1697 static int restart_array(mddev_t *mddev)
1698 {
1699         struct gendisk *disk = mddev->gendisk;
1700         int err;
1701
1702         /*
1703          * Complain if it has no devices
1704          */
1705         err = -ENXIO;
1706         if (list_empty(&mddev->disks))
1707                 goto out;
1708
1709         if (mddev->pers) {
1710                 err = -EBUSY;
1711                 if (!mddev->ro)
1712                         goto out;
1713
1714                 mddev->safemode = 0;
1715                 mddev->ro = 0;
1716                 set_disk_ro(disk, 0);
1717
1718                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1719                         mdname(mddev));
1720                 /*
1721                  * Kick recovery or resync if necessary
1722                  */
1723                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1724                 md_wakeup_thread(mddev->thread);
1725                 err = 0;
1726         } else {
1727                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1728                         mdname(mddev));
1729                 err = -EINVAL;
1730         }
1731
1732 out:
1733         return err;
1734 }
1735
1736 static int do_md_stop(mddev_t * mddev, int ro)
1737 {
1738         int err = 0;
1739         struct gendisk *disk = mddev->gendisk;
1740
1741         if (mddev->pers) {
1742                 if (atomic_read(&mddev->active)>2) {
1743                         printk("md: %s still in use.\n",mdname(mddev));
1744                         return -EBUSY;
1745                 }
1746
1747                 if (mddev->sync_thread) {
1748                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1749                         md_unregister_thread(mddev->sync_thread);
1750                         mddev->sync_thread = NULL;
1751                 }
1752
1753                 del_timer_sync(&mddev->safemode_timer);
1754
1755                 invalidate_partition(disk, 0);
1756
1757                 if (ro) {
1758                         err  = -ENXIO;
1759                         if (mddev->ro)
1760                                 goto out;
1761                         mddev->ro = 1;
1762                 } else {
1763                         if (mddev->ro)
1764                                 set_disk_ro(disk, 0);
1765                         blk_queue_make_request(mddev->queue, md_fail_request);
1766                         mddev->pers->stop(mddev);
1767                         module_put(mddev->pers->owner);
1768                         mddev->pers = NULL;
1769                         if (mddev->ro)
1770                                 mddev->ro = 0;
1771                 }
1772                 if (!mddev->in_sync) {
1773                         /* mark array as shutdown cleanly */
1774                         mddev->in_sync = 1;
1775                         md_update_sb(mddev);
1776                 }
1777                 if (ro)
1778                         set_disk_ro(disk, 1);
1779         }
1780
1781         bitmap_destroy(mddev);
1782         if (mddev->bitmap_file) {
1783                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1784                 fput(mddev->bitmap_file);
1785                 mddev->bitmap_file = NULL;
1786         }
1787
1788         /*
1789          * Free resources if final stop
1790          */
1791         if (!ro) {
1792                 struct gendisk *disk;
1793                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1794
1795                 export_array(mddev);
1796
1797                 mddev->array_size = 0;
1798                 disk = mddev->gendisk;
1799                 if (disk)
1800                         set_capacity(disk, 0);
1801                 mddev->changed = 1;
1802         } else
1803                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1804                         mdname(mddev));
1805         err = 0;
1806 out:
1807         return err;
1808 }
1809
1810 static void autorun_array(mddev_t *mddev)
1811 {
1812         mdk_rdev_t *rdev;
1813         struct list_head *tmp;
1814         int err;
1815
1816         if (list_empty(&mddev->disks))
1817                 return;
1818
1819         printk(KERN_INFO "md: running: ");
1820
1821         ITERATE_RDEV(mddev,rdev,tmp) {
1822                 char b[BDEVNAME_SIZE];
1823                 printk("<%s>", bdevname(rdev->bdev,b));
1824         }
1825         printk("\n");
1826
1827         err = do_md_run (mddev);
1828         if (err) {
1829                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1830                 do_md_stop (mddev, 0);
1831         }
1832 }
1833
1834 /*
1835  * lets try to run arrays based on all disks that have arrived
1836  * until now. (those are in pending_raid_disks)
1837  *
1838  * the method: pick the first pending disk, collect all disks with
1839  * the same UUID, remove all from the pending list and put them into
1840  * the 'same_array' list. Then order this list based on superblock
1841  * update time (freshest comes first), kick out 'old' disks and
1842  * compare superblocks. If everything's fine then run it.
1843  *
1844  * If "unit" is allocated, then bump its reference count
1845  */
1846 static void autorun_devices(int part)
1847 {
1848         struct list_head candidates;
1849         struct list_head *tmp;
1850         mdk_rdev_t *rdev0, *rdev;
1851         mddev_t *mddev;
1852         char b[BDEVNAME_SIZE];
1853
1854         printk(KERN_INFO "md: autorun ...\n");
1855         while (!list_empty(&pending_raid_disks)) {
1856                 dev_t dev;
1857                 rdev0 = list_entry(pending_raid_disks.next,
1858                                          mdk_rdev_t, same_set);
1859
1860                 printk(KERN_INFO "md: considering %s ...\n",
1861                         bdevname(rdev0->bdev,b));
1862                 INIT_LIST_HEAD(&candidates);
1863                 ITERATE_RDEV_PENDING(rdev,tmp)
1864                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1865                                 printk(KERN_INFO "md:  adding %s ...\n",
1866                                         bdevname(rdev->bdev,b));
1867                                 list_move(&rdev->same_set, &candidates);
1868                         }
1869                 /*
1870                  * now we have a set of devices, with all of them having
1871                  * mostly sane superblocks. It's time to allocate the
1872                  * mddev.
1873                  */
1874                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1875                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1876                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1877                         break;
1878                 }
1879                 if (part)
1880                         dev = MKDEV(mdp_major,
1881                                     rdev0->preferred_minor << MdpMinorShift);
1882                 else
1883                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1884
1885                 md_probe(dev, NULL, NULL);
1886                 mddev = mddev_find(dev);
1887                 if (!mddev) {
1888                         printk(KERN_ERR
1889                                 "md: cannot allocate memory for md drive.\n");
1890                         break;
1891                 }
1892                 if (mddev_lock(mddev))
1893                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1894                                mdname(mddev));
1895                 else if (mddev->raid_disks || mddev->major_version
1896                          || !list_empty(&mddev->disks)) {
1897                         printk(KERN_WARNING
1898                                 "md: %s already running, cannot run %s\n",
1899                                 mdname(mddev), bdevname(rdev0->bdev,b));
1900                         mddev_unlock(mddev);
1901                 } else {
1902                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1903                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1904                                 list_del_init(&rdev->same_set);
1905                                 if (bind_rdev_to_array(rdev, mddev))
1906                                         export_rdev(rdev);
1907                         }
1908                         autorun_array(mddev);
1909                         mddev_unlock(mddev);
1910                 }
1911                 /* on success, candidates will be empty, on error
1912                  * it won't...
1913                  */
1914                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1915                         export_rdev(rdev);
1916                 mddev_put(mddev);
1917         }
1918         printk(KERN_INFO "md: ... autorun DONE.\n");
1919 }
1920
1921 /*
1922  * import RAID devices based on one partition
1923  * if possible, the array gets run as well.
1924  */
1925
1926 static int autostart_array(dev_t startdev)
1927 {
1928         char b[BDEVNAME_SIZE];
1929         int err = -EINVAL, i;
1930         mdp_super_t *sb = NULL;
1931         mdk_rdev_t *start_rdev = NULL, *rdev;
1932
1933         start_rdev = md_import_device(startdev, 0, 0);
1934         if (IS_ERR(start_rdev))
1935                 return err;
1936
1937
1938         /* NOTE: this can only work for 0.90.0 superblocks */
1939         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1940         if (sb->major_version != 0 ||
1941             sb->minor_version != 90 ) {
1942                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1943                 export_rdev(start_rdev);
1944                 return err;
1945         }
1946
1947         if (start_rdev->faulty) {
1948                 printk(KERN_WARNING
1949                         "md: can not autostart based on faulty %s!\n",
1950                         bdevname(start_rdev->bdev,b));
1951                 export_rdev(start_rdev);
1952                 return err;
1953         }
1954         list_add(&start_rdev->same_set, &pending_raid_disks);
1955
1956         for (i = 0; i < MD_SB_DISKS; i++) {
1957                 mdp_disk_t *desc = sb->disks + i;
1958                 dev_t dev = MKDEV(desc->major, desc->minor);
1959
1960                 if (!dev)
1961                         continue;
1962                 if (dev == startdev)
1963                         continue;
1964                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
1965                         continue;
1966                 rdev = md_import_device(dev, 0, 0);
1967                 if (IS_ERR(rdev))
1968                         continue;
1969
1970                 list_add(&rdev->same_set, &pending_raid_disks);
1971         }
1972
1973         /*
1974          * possibly return codes
1975          */
1976         autorun_devices(0);
1977         return 0;
1978
1979 }
1980
1981
1982 static int get_version(void __user * arg)
1983 {
1984         mdu_version_t ver;
1985
1986         ver.major = MD_MAJOR_VERSION;
1987         ver.minor = MD_MINOR_VERSION;
1988         ver.patchlevel = MD_PATCHLEVEL_VERSION;
1989
1990         if (copy_to_user(arg, &ver, sizeof(ver)))
1991                 return -EFAULT;
1992
1993         return 0;
1994 }
1995
1996 static int get_array_info(mddev_t * mddev, void __user * arg)
1997 {
1998         mdu_array_info_t info;
1999         int nr,working,active,failed,spare;
2000         mdk_rdev_t *rdev;
2001         struct list_head *tmp;
2002
2003         nr=working=active=failed=spare=0;
2004         ITERATE_RDEV(mddev,rdev,tmp) {
2005                 nr++;
2006                 if (rdev->faulty)
2007                         failed++;
2008                 else {
2009                         working++;
2010                         if (rdev->in_sync)
2011                                 active++;
2012                         else
2013                                 spare++;
2014                 }
2015         }
2016
2017         info.major_version = mddev->major_version;
2018         info.minor_version = mddev->minor_version;
2019         info.patch_version = MD_PATCHLEVEL_VERSION;
2020         info.ctime         = mddev->ctime;
2021         info.level         = mddev->level;
2022         info.size          = mddev->size;
2023         info.nr_disks      = nr;
2024         info.raid_disks    = mddev->raid_disks;
2025         info.md_minor      = mddev->md_minor;
2026         info.not_persistent= !mddev->persistent;
2027
2028         info.utime         = mddev->utime;
2029         info.state         = 0;
2030         if (mddev->in_sync)
2031                 info.state = (1<<MD_SB_CLEAN);
2032         info.active_disks  = active;
2033         info.working_disks = working;
2034         info.failed_disks  = failed;
2035         info.spare_disks   = spare;
2036
2037         info.layout        = mddev->layout;
2038         info.chunk_size    = mddev->chunk_size;
2039
2040         if (copy_to_user(arg, &info, sizeof(info)))
2041                 return -EFAULT;
2042
2043         return 0;
2044 }
2045
2046 static int get_bitmap_file(mddev_t * mddev, void * arg)
2047 {
2048         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2049         char *ptr, *buf = NULL;
2050         int err = -ENOMEM;
2051
2052         file = kmalloc(sizeof(*file), GFP_KERNEL);
2053         if (!file)
2054                 goto out;
2055
2056         /* bitmap disabled, zero the first byte and copy out */
2057         if (!mddev->bitmap || !mddev->bitmap->file) {
2058                 file->pathname[0] = '\0';
2059                 goto copy_out;
2060         }
2061
2062         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2063         if (!buf)
2064                 goto out;
2065
2066         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2067         if (!ptr)
2068                 goto out;
2069
2070         strcpy(file->pathname, ptr);
2071
2072 copy_out:
2073         err = 0;
2074         if (copy_to_user(arg, file, sizeof(*file)))
2075                 err = -EFAULT;
2076 out:
2077         kfree(buf);
2078         kfree(file);
2079         return err;
2080 }
2081
2082 static int get_disk_info(mddev_t * mddev, void __user * arg)
2083 {
2084         mdu_disk_info_t info;
2085         unsigned int nr;
2086         mdk_rdev_t *rdev;
2087
2088         if (copy_from_user(&info, arg, sizeof(info)))
2089                 return -EFAULT;
2090
2091         nr = info.number;
2092
2093         rdev = find_rdev_nr(mddev, nr);
2094         if (rdev) {
2095                 info.major = MAJOR(rdev->bdev->bd_dev);
2096                 info.minor = MINOR(rdev->bdev->bd_dev);
2097                 info.raid_disk = rdev->raid_disk;
2098                 info.state = 0;
2099                 if (rdev->faulty)
2100                         info.state |= (1<<MD_DISK_FAULTY);
2101                 else if (rdev->in_sync) {
2102                         info.state |= (1<<MD_DISK_ACTIVE);
2103                         info.state |= (1<<MD_DISK_SYNC);
2104                 }
2105         } else {
2106                 info.major = info.minor = 0;
2107                 info.raid_disk = -1;
2108                 info.state = (1<<MD_DISK_REMOVED);
2109         }
2110
2111         if (copy_to_user(arg, &info, sizeof(info)))
2112                 return -EFAULT;
2113
2114         return 0;
2115 }
2116
2117 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2118 {
2119         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2120         mdk_rdev_t *rdev;
2121         dev_t dev = MKDEV(info->major,info->minor);
2122
2123         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2124                 return -EOVERFLOW;
2125
2126         if (!mddev->raid_disks) {
2127                 int err;
2128                 /* expecting a device which has a superblock */
2129                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2130                 if (IS_ERR(rdev)) {
2131                         printk(KERN_WARNING
2132                                 "md: md_import_device returned %ld\n",
2133                                 PTR_ERR(rdev));
2134                         return PTR_ERR(rdev);
2135                 }
2136                 if (!list_empty(&mddev->disks)) {
2137                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2138                                                         mdk_rdev_t, same_set);
2139                         int err = super_types[mddev->major_version]
2140                                 .load_super(rdev, rdev0, mddev->minor_version);
2141                         if (err < 0) {
2142                                 printk(KERN_WARNING
2143                                         "md: %s has different UUID to %s\n",
2144                                         bdevname(rdev->bdev,b),
2145                                         bdevname(rdev0->bdev,b2));
2146                                 export_rdev(rdev);
2147                                 return -EINVAL;
2148                         }
2149                 }
2150                 err = bind_rdev_to_array(rdev, mddev);
2151                 if (err)
2152                         export_rdev(rdev);
2153                 return err;
2154         }
2155
2156         /*
2157          * add_new_disk can be used once the array is assembled
2158          * to add "hot spares".  They must already have a superblock
2159          * written
2160          */
2161         if (mddev->pers) {
2162                 int err;
2163                 if (!mddev->pers->hot_add_disk) {
2164                         printk(KERN_WARNING
2165                                 "%s: personality does not support diskops!\n",
2166                                mdname(mddev));
2167                         return -EINVAL;
2168                 }
2169                 rdev = md_import_device(dev, mddev->major_version,
2170                                         mddev->minor_version);
2171                 if (IS_ERR(rdev)) {
2172                         printk(KERN_WARNING
2173                                 "md: md_import_device returned %ld\n",
2174                                 PTR_ERR(rdev));
2175                         return PTR_ERR(rdev);
2176                 }
2177                 /* set save_raid_disk if appropriate */
2178                 if (!mddev->persistent) {
2179                         if (info->state & (1<<MD_DISK_SYNC)  &&
2180                             info->raid_disk < mddev->raid_disks)
2181                                 rdev->raid_disk = info->raid_disk;
2182                         else
2183                                 rdev->raid_disk = -1;
2184                 } else
2185                         super_types[mddev->major_version].
2186                                 validate_super(mddev, rdev);
2187                 rdev->saved_raid_disk = rdev->raid_disk;
2188
2189                 rdev->in_sync = 0; /* just to be sure */
2190                 rdev->raid_disk = -1;
2191                 err = bind_rdev_to_array(rdev, mddev);
2192                 if (err)
2193                         export_rdev(rdev);
2194
2195                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2196                 if (mddev->thread)
2197                         md_wakeup_thread(mddev->thread);
2198                 return err;
2199         }
2200
2201         /* otherwise, add_new_disk is only allowed
2202          * for major_version==0 superblocks
2203          */
2204         if (mddev->major_version != 0) {
2205                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2206                        mdname(mddev));
2207                 return -EINVAL;
2208         }
2209
2210         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2211                 int err;
2212                 rdev = md_import_device (dev, -1, 0);
2213                 if (IS_ERR(rdev)) {
2214                         printk(KERN_WARNING
2215                                 "md: error, md_import_device() returned %ld\n",
2216                                 PTR_ERR(rdev));
2217                         return PTR_ERR(rdev);
2218                 }
2219                 rdev->desc_nr = info->number;
2220                 if (info->raid_disk < mddev->raid_disks)
2221                         rdev->raid_disk = info->raid_disk;
2222                 else
2223                         rdev->raid_disk = -1;
2224
2225                 rdev->faulty = 0;
2226                 if (rdev->raid_disk < mddev->raid_disks)
2227                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2228                 else
2229                         rdev->in_sync = 0;
2230
2231                 err = bind_rdev_to_array(rdev, mddev);
2232                 if (err) {
2233                         export_rdev(rdev);
2234                         return err;
2235                 }
2236
2237                 if (!mddev->persistent) {
2238                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2239                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2240                 } else
2241                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2242                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2243
2244                 if (!mddev->size || (mddev->size > rdev->size))
2245                         mddev->size = rdev->size;
2246         }
2247
2248         return 0;
2249 }
2250
2251 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2252 {
2253         char b[BDEVNAME_SIZE];
2254         mdk_rdev_t *rdev;
2255
2256         if (!mddev->pers)
2257                 return -ENODEV;
2258
2259         rdev = find_rdev(mddev, dev);
2260         if (!rdev)
2261                 return -ENXIO;
2262
2263         if (rdev->raid_disk >= 0)
2264                 goto busy;
2265
2266         kick_rdev_from_array(rdev);
2267         md_update_sb(mddev);
2268
2269         return 0;
2270 busy:
2271         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2272                 bdevname(rdev->bdev,b), mdname(mddev));
2273         return -EBUSY;
2274 }
2275
2276 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2277 {
2278         char b[BDEVNAME_SIZE];
2279         int err;
2280         unsigned int size;
2281         mdk_rdev_t *rdev;
2282
2283         if (!mddev->pers)
2284                 return -ENODEV;
2285
2286         if (mddev->major_version != 0) {
2287                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2288                         " version-0 superblocks.\n",
2289                         mdname(mddev));
2290                 return -EINVAL;
2291         }
2292         if (!mddev->pers->hot_add_disk) {
2293                 printk(KERN_WARNING
2294                         "%s: personality does not support diskops!\n",
2295                         mdname(mddev));
2296                 return -EINVAL;
2297         }
2298
2299         rdev = md_import_device (dev, -1, 0);
2300         if (IS_ERR(rdev)) {
2301                 printk(KERN_WARNING
2302                         "md: error, md_import_device() returned %ld\n",
2303                         PTR_ERR(rdev));
2304                 return -EINVAL;
2305         }
2306
2307         if (mddev->persistent)
2308                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2309         else
2310                 rdev->sb_offset =
2311                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2312
2313         size = calc_dev_size(rdev, mddev->chunk_size);
2314         rdev->size = size;
2315
2316         if (size < mddev->size) {
2317                 printk(KERN_WARNING
2318                         "%s: disk size %llu blocks < array size %llu\n",
2319                         mdname(mddev), (unsigned long long)size,
2320                         (unsigned long long)mddev->size);
2321                 err = -ENOSPC;
2322                 goto abort_export;
2323         }
2324
2325         if (rdev->faulty) {
2326                 printk(KERN_WARNING
2327                         "md: can not hot-add faulty %s disk to %s!\n",
2328                         bdevname(rdev->bdev,b), mdname(mddev));
2329                 err = -EINVAL;
2330                 goto abort_export;
2331         }
2332         rdev->in_sync = 0;
2333         rdev->desc_nr = -1;
2334         bind_rdev_to_array(rdev, mddev);
2335
2336         /*
2337          * The rest should better be atomic, we can have disk failures
2338          * noticed in interrupt contexts ...
2339          */
2340
2341         if (rdev->desc_nr == mddev->max_disks) {
2342                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2343                         mdname(mddev));
2344                 err = -EBUSY;
2345                 goto abort_unbind_export;
2346         }
2347
2348         rdev->raid_disk = -1;
2349
2350         md_update_sb(mddev);
2351
2352         /*
2353          * Kick recovery, maybe this spare has to be added to the
2354          * array immediately.
2355          */
2356         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2357         md_wakeup_thread(mddev->thread);
2358
2359         return 0;
2360
2361 abort_unbind_export:
2362         unbind_rdev_from_array(rdev);
2363
2364 abort_export:
2365         export_rdev(rdev);
2366         return err;
2367 }
2368
2369 /* similar to deny_write_access, but accounts for our holding a reference
2370  * to the file ourselves */
2371 static int deny_bitmap_write_access(struct file * file)
2372 {
2373         struct inode *inode = file->f_mapping->host;
2374
2375         spin_lock(&inode->i_lock);
2376         if (atomic_read(&inode->i_writecount) > 1) {
2377                 spin_unlock(&inode->i_lock);
2378                 return -ETXTBSY;
2379         }
2380         atomic_set(&inode->i_writecount, -1);
2381         spin_unlock(&inode->i_lock);
2382
2383         return 0;
2384 }
2385
2386 static int set_bitmap_file(mddev_t *mddev, int fd)
2387 {
2388         int err;
2389
2390         if (mddev->pers)
2391                 return -EBUSY;
2392
2393         mddev->bitmap_file = fget(fd);
2394
2395         if (mddev->bitmap_file == NULL) {
2396                 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2397                         mdname(mddev));
2398                 return -EBADF;
2399         }
2400
2401         err = deny_bitmap_write_access(mddev->bitmap_file);
2402         if (err) {
2403                 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2404                         mdname(mddev));
2405                 fput(mddev->bitmap_file);
2406                 mddev->bitmap_file = NULL;
2407         }
2408         return err;
2409 }
2410
2411 /*
2412  * set_array_info is used two different ways
2413  * The original usage is when creating a new array.
2414  * In this usage, raid_disks is > 0 and it together with
2415  *  level, size, not_persistent,layout,chunksize determine the
2416  *  shape of the array.
2417  *  This will always create an array with a type-0.90.0 superblock.
2418  * The newer usage is when assembling an array.
2419  *  In this case raid_disks will be 0, and the major_version field is
2420  *  use to determine which style super-blocks are to be found on the devices.
2421  *  The minor and patch _version numbers are also kept incase the
2422  *  super_block handler wishes to interpret them.
2423  */
2424 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2425 {
2426
2427         if (info->raid_disks == 0) {
2428                 /* just setting version number for superblock loading */
2429                 if (info->major_version < 0 ||
2430                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2431                     super_types[info->major_version].name == NULL) {
2432                         /* maybe try to auto-load a module? */
2433                         printk(KERN_INFO
2434                                 "md: superblock version %d not known\n",
2435                                 info->major_version);
2436                         return -EINVAL;
2437                 }
2438                 mddev->major_version = info->major_version;
2439                 mddev->minor_version = info->minor_version;
2440                 mddev->patch_version = info->patch_version;
2441                 return 0;
2442         }
2443         mddev->major_version = MD_MAJOR_VERSION;
2444         mddev->minor_version = MD_MINOR_VERSION;
2445         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2446         mddev->ctime         = get_seconds();
2447
2448         mddev->level         = info->level;
2449         mddev->size          = info->size;
2450         mddev->raid_disks    = info->raid_disks;
2451         /* don't set md_minor, it is determined by which /dev/md* was
2452          * openned
2453          */
2454         if (info->state & (1<<MD_SB_CLEAN))
2455                 mddev->recovery_cp = MaxSector;
2456         else
2457                 mddev->recovery_cp = 0;
2458         mddev->persistent    = ! info->not_persistent;
2459
2460         mddev->layout        = info->layout;
2461         mddev->chunk_size    = info->chunk_size;
2462
2463         mddev->max_disks     = MD_SB_DISKS;
2464
2465         mddev->sb_dirty      = 1;
2466
2467         /*
2468          * Generate a 128 bit UUID
2469          */
2470         get_random_bytes(mddev->uuid, 16);
2471
2472         return 0;
2473 }
2474
2475 /*
2476  * update_array_info is used to change the configuration of an
2477  * on-line array.
2478  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2479  * fields in the info are checked against the array.
2480  * Any differences that cannot be handled will cause an error.
2481  * Normally, only one change can be managed at a time.
2482  */
2483 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2484 {
2485         int rv = 0;
2486         int cnt = 0;
2487
2488         if (mddev->major_version != info->major_version ||
2489             mddev->minor_version != info->minor_version ||
2490 /*          mddev->patch_version != info->patch_version || */
2491             mddev->ctime         != info->ctime         ||
2492             mddev->level         != info->level         ||
2493 /*          mddev->layout        != info->layout        || */
2494             !mddev->persistent   != info->not_persistent||
2495             mddev->chunk_size    != info->chunk_size    )
2496                 return -EINVAL;
2497         /* Check there is only one change */
2498         if (mddev->size != info->size) cnt++;
2499         if (mddev->raid_disks != info->raid_disks) cnt++;
2500         if (mddev->layout != info->layout) cnt++;
2501         if (cnt == 0) return 0;
2502         if (cnt > 1) return -EINVAL;
2503
2504         if (mddev->layout != info->layout) {
2505                 /* Change layout
2506                  * we don't need to do anything at the md level, the
2507                  * personality will take care of it all.
2508                  */
2509                 if (mddev->pers->reconfig == NULL)
2510                         return -EINVAL;
2511                 else
2512                         return mddev->pers->reconfig(mddev, info->layout, -1);
2513         }
2514         if (mddev->size != info->size) {
2515                 mdk_rdev_t * rdev;
2516                 struct list_head *tmp;
2517                 if (mddev->pers->resize == NULL)
2518                         return -EINVAL;
2519                 /* The "size" is the amount of each device that is used.
2520                  * This can only make sense for arrays with redundancy.
2521                  * linear and raid0 always use whatever space is available
2522                  * We can only consider changing the size if no resync
2523                  * or reconstruction is happening, and if the new size
2524                  * is acceptable. It must fit before the sb_offset or,
2525                  * if that is <data_offset, it must fit before the
2526                  * size of each device.
2527                  * If size is zero, we find the largest size that fits.
2528                  */
2529                 if (mddev->sync_thread)
2530                         return -EBUSY;
2531                 ITERATE_RDEV(mddev,rdev,tmp) {
2532                         sector_t avail;
2533                         int fit = (info->size == 0);
2534                         if (rdev->sb_offset > rdev->data_offset)
2535                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2536                         else
2537                                 avail = get_capacity(rdev->bdev->bd_disk)
2538                                         - rdev->data_offset;
2539                         if (fit && (info->size == 0 || info->size > avail/2))
2540                                 info->size = avail/2;
2541                         if (avail < ((sector_t)info->size << 1))
2542                                 return -ENOSPC;
2543                 }
2544                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2545                 if (!rv) {
2546                         struct block_device *bdev;
2547
2548                         bdev = bdget_disk(mddev->gendisk, 0);
2549                         if (bdev) {
2550                                 down(&bdev->bd_inode->i_sem);
2551                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2552                                 up(&bdev->bd_inode->i_sem);
2553                                 bdput(bdev);
2554                         }
2555                 }
2556         }
2557         if (mddev->raid_disks    != info->raid_disks) {
2558                 /* change the number of raid disks */
2559                 if (mddev->pers->reshape == NULL)
2560                         return -EINVAL;
2561                 if (info->raid_disks <= 0 ||
2562                     info->raid_disks >= mddev->max_disks)
2563                         return -EINVAL;
2564                 if (mddev->sync_thread)
2565                         return -EBUSY;
2566                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2567                 if (!rv) {
2568                         struct block_device *bdev;
2569
2570                         bdev = bdget_disk(mddev->gendisk, 0);
2571                         if (bdev) {
2572                                 down(&bdev->bd_inode->i_sem);
2573                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2574                                 up(&bdev->bd_inode->i_sem);
2575                                 bdput(bdev);
2576                         }
2577                 }
2578         }
2579         md_update_sb(mddev);
2580         return rv;
2581 }
2582
2583 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2584 {
2585         mdk_rdev_t *rdev;
2586
2587         if (mddev->pers == NULL)
2588                 return -ENODEV;
2589
2590         rdev = find_rdev(mddev, dev);
2591         if (!rdev)
2592                 return -ENODEV;
2593
2594         md_error(mddev, rdev);
2595         return 0;
2596 }
2597
2598 static int md_ioctl(struct inode *inode, struct file *file,
2599                         unsigned int cmd, unsigned long arg)
2600 {
2601         int err = 0;
2602         void __user *argp = (void __user *)arg;
2603         struct hd_geometry __user *loc = argp;
2604         mddev_t *mddev = NULL;
2605
2606         if (!capable(CAP_SYS_ADMIN))
2607                 return -EACCES;
2608
2609         /*
2610          * Commands dealing with the RAID driver but not any
2611          * particular array:
2612          */
2613         switch (cmd)
2614         {
2615                 case RAID_VERSION:
2616                         err = get_version(argp);
2617                         goto done;
2618
2619                 case PRINT_RAID_DEBUG:
2620                         err = 0;
2621                         md_print_devices();
2622                         goto done;
2623
2624 #ifndef MODULE
2625                 case RAID_AUTORUN:
2626                         err = 0;
2627                         autostart_arrays(arg);
2628                         goto done;
2629 #endif
2630                 default:;
2631         }
2632
2633         /*
2634          * Commands creating/starting a new array:
2635          */
2636
2637         mddev = inode->i_bdev->bd_disk->private_data;
2638
2639         if (!mddev) {
2640                 BUG();
2641                 goto abort;
2642         }
2643
2644
2645         if (cmd == START_ARRAY) {
2646                 /* START_ARRAY doesn't need to lock the array as autostart_array
2647                  * does the locking, and it could even be a different array
2648                  */
2649                 static int cnt = 3;
2650                 if (cnt > 0 ) {
2651                         printk(KERN_WARNING
2652                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2653                                "This will not be supported beyond 2.6\n",
2654                                current->comm, current->pid);
2655                         cnt--;
2656                 }
2657                 err = autostart_array(new_decode_dev(arg));
2658                 if (err) {
2659                         printk(KERN_WARNING "md: autostart failed!\n");
2660                         goto abort;
2661                 }
2662                 goto done;
2663         }
2664
2665         err = mddev_lock(mddev);
2666         if (err) {
2667                 printk(KERN_INFO
2668                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2669                         err, cmd);
2670                 goto abort;
2671         }
2672
2673         switch (cmd)
2674         {
2675                 case SET_ARRAY_INFO:
2676                         {
2677                                 mdu_array_info_t info;
2678                                 if (!arg)
2679                                         memset(&info, 0, sizeof(info));
2680                                 else if (copy_from_user(&info, argp, sizeof(info))) {
2681                                         err = -EFAULT;
2682                                         goto abort_unlock;
2683                                 }
2684                                 if (mddev->pers) {
2685                                         err = update_array_info(mddev, &info);
2686                                         if (err) {
2687                                                 printk(KERN_WARNING "md: couldn't update"
2688                                                        " array info. %d\n", err);
2689                                                 goto abort_unlock;
2690                                         }
2691                                         goto done_unlock;
2692                                 }
2693                                 if (!list_empty(&mddev->disks)) {
2694                                         printk(KERN_WARNING
2695                                                "md: array %s already has disks!\n",
2696                                                mdname(mddev));
2697                                         err = -EBUSY;
2698                                         goto abort_unlock;
2699                                 }
2700                                 if (mddev->raid_disks) {
2701                                         printk(KERN_WARNING
2702                                                "md: array %s already initialised!\n",
2703                                                mdname(mddev));
2704                                         err = -EBUSY;
2705                                         goto abort_unlock;
2706                                 }
2707                                 err = set_array_info(mddev, &info);
2708                                 if (err) {
2709                                         printk(KERN_WARNING "md: couldn't set"
2710                                                " array info. %d\n", err);
2711                                         goto abort_unlock;
2712                                 }
2713                         }
2714                         goto done_unlock;
2715
2716                 default:;
2717         }
2718
2719         /*
2720          * Commands querying/configuring an existing array:
2721          */
2722         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2723          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2724         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2725                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2726                 err = -ENODEV;
2727                 goto abort_unlock;
2728         }
2729
2730         /*
2731          * Commands even a read-only array can execute:
2732          */
2733         switch (cmd)
2734         {
2735                 case GET_ARRAY_INFO:
2736                         err = get_array_info(mddev, argp);
2737                         goto done_unlock;
2738
2739                 case GET_BITMAP_FILE:
2740                         err = get_bitmap_file(mddev, (void *)arg);
2741                         goto done_unlock;
2742
2743                 case GET_DISK_INFO:
2744                         err = get_disk_info(mddev, argp);
2745                         goto done_unlock;
2746
2747                 case RESTART_ARRAY_RW:
2748                         err = restart_array(mddev);
2749                         goto done_unlock;
2750
2751                 case STOP_ARRAY:
2752                         err = do_md_stop (mddev, 0);
2753                         goto done_unlock;
2754
2755                 case STOP_ARRAY_RO:
2756                         err = do_md_stop (mddev, 1);
2757                         goto done_unlock;
2758
2759         /*
2760          * We have a problem here : there is no easy way to give a CHS
2761          * virtual geometry. We currently pretend that we have a 2 heads
2762          * 4 sectors (with a BIG number of cylinders...). This drives
2763          * dosfs just mad... ;-)
2764          */
2765                 case HDIO_GETGEO:
2766                         if (!loc) {
2767                                 err = -EINVAL;
2768                                 goto abort_unlock;
2769                         }
2770                         err = put_user (2, (char __user *) &loc->heads);
2771                         if (err)
2772                                 goto abort_unlock;
2773                         err = put_user (4, (char __user *) &loc->sectors);
2774                         if (err)
2775                                 goto abort_unlock;
2776                         err = put_user(get_capacity(mddev->gendisk)/8,
2777                                         (short __user *) &loc->cylinders);
2778                         if (err)
2779                                 goto abort_unlock;
2780                         err = put_user (get_start_sect(inode->i_bdev),
2781                                                 (long __user *) &loc->start);
2782                         goto done_unlock;
2783         }
2784
2785         /*
2786          * The remaining ioctls are changing the state of the
2787          * superblock, so we do not allow read-only arrays
2788          * here:
2789          */
2790         if (mddev->ro) {
2791                 err = -EROFS;
2792                 goto abort_unlock;
2793         }
2794
2795         switch (cmd)
2796         {
2797                 case ADD_NEW_DISK:
2798                 {
2799                         mdu_disk_info_t info;
2800                         if (copy_from_user(&info, argp, sizeof(info)))
2801                                 err = -EFAULT;
2802                         else
2803                                 err = add_new_disk(mddev, &info);
2804                         goto done_unlock;
2805                 }
2806
2807                 case HOT_REMOVE_DISK:
2808                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2809                         goto done_unlock;
2810
2811                 case HOT_ADD_DISK:
2812                         err = hot_add_disk(mddev, new_decode_dev(arg));
2813                         goto done_unlock;
2814
2815                 case SET_DISK_FAULTY:
2816                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2817                         goto done_unlock;
2818
2819                 case RUN_ARRAY:
2820                         err = do_md_run (mddev);
2821                         goto done_unlock;
2822
2823                 case SET_BITMAP_FILE:
2824                         err = set_bitmap_file(mddev, (int)arg);
2825                         goto done_unlock;
2826
2827                 default:
2828                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2829                                 printk(KERN_WARNING "md: %s(pid %d) used"
2830                                         " obsolete MD ioctl, upgrade your"
2831                                         " software to use new ictls.\n",
2832                                         current->comm, current->pid);
2833                         err = -EINVAL;
2834                         goto abort_unlock;
2835         }
2836
2837 done_unlock:
2838 abort_unlock:
2839         mddev_unlock(mddev);
2840
2841         return err;
2842 done:
2843         if (err)
2844                 MD_BUG();
2845 abort:
2846         return err;
2847 }
2848
2849 static int md_open(struct inode *inode, struct file *file)
2850 {
2851         /*
2852          * Succeed if we can lock the mddev, which confirms that
2853          * it isn't being stopped right now.
2854          */
2855         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2856         int err;
2857
2858         if ((err = mddev_lock(mddev)))
2859                 goto out;
2860
2861         err = 0;
2862         mddev_get(mddev);
2863         mddev_unlock(mddev);
2864
2865         check_disk_change(inode->i_bdev);
2866  out:
2867         return err;
2868 }
2869
2870 static int md_release(struct inode *inode, struct file * file)
2871 {
2872         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2873
2874         if (!mddev)
2875                 BUG();
2876         mddev_put(mddev);
2877
2878         return 0;
2879 }
2880
2881 static int md_media_changed(struct gendisk *disk)
2882 {
2883         mddev_t *mddev = disk->private_data;
2884
2885         return mddev->changed;
2886 }
2887
2888 static int md_revalidate(struct gendisk *disk)
2889 {
2890         mddev_t *mddev = disk->private_data;
2891
2892         mddev->changed = 0;
2893         return 0;
2894 }
2895 static struct block_device_operations md_fops =
2896 {
2897         .owner          = THIS_MODULE,
2898         .open           = md_open,
2899         .release        = md_release,
2900         .ioctl          = md_ioctl,
2901         .media_changed  = md_media_changed,
2902         .revalidate_disk= md_revalidate,
2903 };
2904
2905 static int md_thread(void * arg)
2906 {
2907         mdk_thread_t *thread = arg;
2908
2909         lock_kernel();
2910
2911         /*
2912          * Detach thread
2913          */
2914
2915         daemonize(thread->name, mdname(thread->mddev));
2916
2917         current->exit_signal = SIGCHLD;
2918         allow_signal(SIGKILL);
2919         thread->tsk = current;
2920
2921         /*
2922          * md_thread is a 'system-thread', it's priority should be very
2923          * high. We avoid resource deadlocks individually in each
2924          * raid personality. (RAID5 does preallocation) We also use RR and
2925          * the very same RT priority as kswapd, thus we will never get
2926          * into a priority inversion deadlock.
2927          *
2928          * we definitely have to have equal or higher priority than
2929          * bdflush, otherwise bdflush will deadlock if there are too
2930          * many dirty RAID5 blocks.
2931          */
2932         unlock_kernel();
2933
2934         complete(thread->event);
2935         while (thread->run) {
2936                 void (*run)(mddev_t *);
2937
2938                 wait_event_interruptible_timeout(thread->wqueue,
2939                                                  test_bit(THREAD_WAKEUP, &thread->flags),
2940                                                  thread->timeout);
2941                 if (current->flags & PF_FREEZE)
2942                         refrigerator(PF_FREEZE);
2943
2944                 clear_bit(THREAD_WAKEUP, &thread->flags);
2945
2946                 run = thread->run;
2947                 if (run)
2948                         run(thread->mddev);
2949
2950                 if (signal_pending(current))
2951                         flush_signals(current);
2952         }
2953         complete(thread->event);
2954         return 0;
2955 }
2956
2957 void md_wakeup_thread(mdk_thread_t *thread)
2958 {
2959         if (thread) {
2960                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
2961                 set_bit(THREAD_WAKEUP, &thread->flags);
2962                 wake_up(&thread->wqueue);
2963         }
2964 }
2965
2966 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
2967                                  const char *name)
2968 {
2969         mdk_thread_t *thread;
2970         int ret;
2971         struct completion event;
2972
2973         thread = (mdk_thread_t *) kmalloc
2974                                 (sizeof(mdk_thread_t), GFP_KERNEL);
2975         if (!thread)
2976                 return NULL;
2977
2978         memset(thread, 0, sizeof(mdk_thread_t));
2979         init_waitqueue_head(&thread->wqueue);
2980
2981         init_completion(&event);
2982         thread->event = &event;
2983         thread->run = run;
2984         thread->mddev = mddev;
2985         thread->name = name;
2986         thread->timeout = MAX_SCHEDULE_TIMEOUT;
2987         ret = kernel_thread(md_thread, thread, 0);
2988         if (ret < 0) {
2989                 kfree(thread);
2990                 return NULL;
2991         }
2992         wait_for_completion(&event);
2993         return thread;
2994 }
2995
2996 void md_unregister_thread(mdk_thread_t *thread)
2997 {
2998         struct completion event;
2999
3000         init_completion(&event);
3001
3002         thread->event = &event;
3003
3004         /* As soon as ->run is set to NULL, the task could disappear,
3005          * so we need to hold tasklist_lock until we have sent the signal
3006          */
3007         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3008         read_lock(&tasklist_lock);
3009         thread->run = NULL;
3010         send_sig(SIGKILL, thread->tsk, 1);
3011         read_unlock(&tasklist_lock);
3012         wait_for_completion(&event);
3013         kfree(thread);
3014 }
3015
3016 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3017 {
3018         if (!mddev) {
3019                 MD_BUG();
3020                 return;
3021         }
3022
3023         if (!rdev || rdev->faulty)
3024                 return;
3025 /*
3026         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3027                 mdname(mddev),
3028                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3029                 __builtin_return_address(0),__builtin_return_address(1),
3030                 __builtin_return_address(2),__builtin_return_address(3));
3031 */
3032         if (!mddev->pers->error_handler)
3033                 return;
3034         mddev->pers->error_handler(mddev,rdev);
3035         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3036         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3037         md_wakeup_thread(mddev->thread);
3038 }
3039
3040 /* seq_file implementation /proc/mdstat */
3041
3042 static void status_unused(struct seq_file *seq)
3043 {
3044         int i = 0;
3045         mdk_rdev_t *rdev;
3046         struct list_head *tmp;
3047
3048         seq_printf(seq, "unused devices: ");
3049
3050         ITERATE_RDEV_PENDING(rdev,tmp) {
3051                 char b[BDEVNAME_SIZE];
3052                 i++;
3053                 seq_printf(seq, "%s ",
3054                               bdevname(rdev->bdev,b));
3055         }
3056         if (!i)
3057                 seq_printf(seq, "<none>");
3058
3059         seq_printf(seq, "\n");
3060 }
3061
3062
3063 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3064 {
3065         unsigned long max_blocks, resync, res, dt, db, rt;
3066
3067         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3068
3069         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3070                 max_blocks = mddev->resync_max_sectors >> 1;
3071         else
3072                 max_blocks = mddev->size;
3073
3074         /*
3075          * Should not happen.
3076          */
3077         if (!max_blocks) {
3078                 MD_BUG();
3079                 return;
3080         }
3081         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3082         {
3083                 int i, x = res/50, y = 20-x;
3084                 seq_printf(seq, "[");
3085                 for (i = 0; i < x; i++)
3086                         seq_printf(seq, "=");
3087                 seq_printf(seq, ">");
3088                 for (i = 0; i < y; i++)
3089                         seq_printf(seq, ".");
3090                 seq_printf(seq, "] ");
3091         }
3092         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3093                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3094                        "resync" : "recovery"),
3095                       res/10, res % 10, resync, max_blocks);
3096
3097         /*
3098          * We do not want to overflow, so the order of operands and
3099          * the * 100 / 100 trick are important. We do a +1 to be
3100          * safe against division by zero. We only estimate anyway.
3101          *
3102          * dt: time from mark until now
3103          * db: blocks written from mark until now
3104          * rt: remaining time
3105          */
3106         dt = ((jiffies - mddev->resync_mark) / HZ);
3107         if (!dt) dt++;
3108         db = resync - (mddev->resync_mark_cnt/2);
3109         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3110
3111         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3112
3113         seq_printf(seq, " speed=%ldK/sec", db/dt);
3114 }
3115
3116 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3117 {
3118         struct list_head *tmp;
3119         loff_t l = *pos;
3120         mddev_t *mddev;
3121
3122         if (l >= 0x10000)
3123                 return NULL;
3124         if (!l--)
3125                 /* header */
3126                 return (void*)1;
3127
3128         spin_lock(&all_mddevs_lock);
3129         list_for_each(tmp,&all_mddevs)
3130                 if (!l--) {
3131                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3132                         mddev_get(mddev);
3133                         spin_unlock(&all_mddevs_lock);
3134                         return mddev;
3135                 }
3136         spin_unlock(&all_mddevs_lock);
3137         if (!l--)
3138                 return (void*)2;/* tail */
3139         return NULL;
3140 }
3141
3142 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3143 {
3144         struct list_head *tmp;
3145         mddev_t *next_mddev, *mddev = v;
3146
3147         ++*pos;
3148         if (v == (void*)2)
3149                 return NULL;
3150
3151         spin_lock(&all_mddevs_lock);
3152         if (v == (void*)1)
3153                 tmp = all_mddevs.next;
3154         else
3155                 tmp = mddev->all_mddevs.next;
3156         if (tmp != &all_mddevs)
3157                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3158         else {
3159                 next_mddev = (void*)2;
3160                 *pos = 0x10000;
3161         }
3162         spin_unlock(&all_mddevs_lock);
3163
3164         if (v != (void*)1)
3165                 mddev_put(mddev);
3166         return next_mddev;
3167
3168 }
3169
3170 static void md_seq_stop(struct seq_file *seq, void *v)
3171 {
3172         mddev_t *mddev = v;
3173
3174         if (mddev && v != (void*)1 && v != (void*)2)
3175                 mddev_put(mddev);
3176 }
3177
3178 static int md_seq_show(struct seq_file *seq, void *v)
3179 {
3180         mddev_t *mddev = v;
3181         sector_t size;
3182         struct list_head *tmp2;
3183         mdk_rdev_t *rdev;
3184         int i;
3185         struct bitmap *bitmap;
3186
3187         if (v == (void*)1) {
3188                 seq_printf(seq, "Personalities : ");
3189                 spin_lock(&pers_lock);
3190                 for (i = 0; i < MAX_PERSONALITY; i++)
3191                         if (pers[i])
3192                                 seq_printf(seq, "[%s] ", pers[i]->name);
3193
3194                 spin_unlock(&pers_lock);
3195                 seq_printf(seq, "\n");
3196                 return 0;
3197         }
3198         if (v == (void*)2) {
3199                 status_unused(seq);
3200                 return 0;
3201         }
3202
3203         if (mddev_lock(mddev)!=0)
3204                 return -EINTR;
3205         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3206                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3207                                                 mddev->pers ? "" : "in");
3208                 if (mddev->pers) {
3209                         if (mddev->ro)
3210                                 seq_printf(seq, " (read-only)");
3211                         seq_printf(seq, " %s", mddev->pers->name);
3212                 }
3213
3214                 size = 0;
3215                 ITERATE_RDEV(mddev,rdev,tmp2) {
3216                         char b[BDEVNAME_SIZE];
3217                         seq_printf(seq, " %s[%d]",
3218                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3219                         if (rdev->faulty) {
3220                                 seq_printf(seq, "(F)");
3221                                 continue;
3222                         }
3223                         size += rdev->size;
3224                 }
3225
3226                 if (!list_empty(&mddev->disks)) {
3227                         if (mddev->pers)
3228                                 seq_printf(seq, "\n      %llu blocks",
3229                                         (unsigned long long)mddev->array_size);
3230                         else
3231                                 seq_printf(seq, "\n      %llu blocks",
3232                                         (unsigned long long)size);
3233                 }
3234
3235                 if (mddev->pers) {
3236                         mddev->pers->status (seq, mddev);
3237                         seq_printf(seq, "\n      ");
3238                         if (mddev->curr_resync > 2) {
3239                                 status_resync (seq, mddev);
3240                                 seq_printf(seq, "\n      ");
3241                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3242                                 seq_printf(seq, "       resync=DELAYED\n      ");
3243                 } else
3244                         seq_printf(seq, "\n       ");
3245
3246                 if ((bitmap = mddev->bitmap)) {
3247                         unsigned long chunk_kb;
3248                         unsigned long flags;
3249                         spin_lock_irqsave(&bitmap->lock, flags);
3250                         chunk_kb = bitmap->chunksize >> 10;
3251                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3252                                 "%lu%s chunk",
3253                                 bitmap->pages - bitmap->missing_pages,
3254                                 bitmap->pages,
3255                                 (bitmap->pages - bitmap->missing_pages)
3256                                         << (PAGE_SHIFT - 10),
3257                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3258                                 chunk_kb ? "KB" : "B");
3259                         if (bitmap->file) {
3260                                 seq_printf(seq, ", file: ");
3261                                 seq_path(seq, bitmap->file->f_vfsmnt,
3262                                          bitmap->file->f_dentry," \t\n");
3263                         }
3264
3265                         seq_printf(seq, "\n");
3266                         spin_unlock_irqrestore(&bitmap->lock, flags);
3267                 }
3268
3269                 seq_printf(seq, "\n");
3270         }
3271         mddev_unlock(mddev);
3272
3273         return 0;
3274 }
3275
3276 static struct seq_operations md_seq_ops = {
3277         .start  = md_seq_start,
3278         .next   = md_seq_next,
3279         .stop   = md_seq_stop,
3280         .show   = md_seq_show,
3281 };
3282
3283 static int md_seq_open(struct inode *inode, struct file *file)
3284 {
3285         int error;
3286
3287         error = seq_open(file, &md_seq_ops);
3288         return error;
3289 }
3290
3291 static struct file_operations md_seq_fops = {
3292         .open           = md_seq_open,
3293         .read           = seq_read,
3294         .llseek         = seq_lseek,
3295         .release        = seq_release,
3296 };
3297
3298 int register_md_personality(int pnum, mdk_personality_t *p)
3299 {
3300         if (pnum >= MAX_PERSONALITY) {
3301                 printk(KERN_ERR
3302                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3303                        p->name, pnum, MAX_PERSONALITY-1);
3304                 return -EINVAL;
3305         }
3306
3307         spin_lock(&pers_lock);
3308         if (pers[pnum]) {
3309                 spin_unlock(&pers_lock);
3310                 return -EBUSY;
3311         }
3312
3313         pers[pnum] = p;
3314         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3315         spin_unlock(&pers_lock);
3316         return 0;
3317 }
3318
3319 int unregister_md_personality(int pnum)
3320 {
3321         if (pnum >= MAX_PERSONALITY)
3322                 return -EINVAL;
3323
3324         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3325         spin_lock(&pers_lock);
3326         pers[pnum] = NULL;
3327         spin_unlock(&pers_lock);
3328         return 0;
3329 }
3330
3331 static int is_mddev_idle(mddev_t *mddev)
3332 {
3333         mdk_rdev_t * rdev;
3334         struct list_head *tmp;
3335         int idle;
3336         unsigned long curr_events;
3337
3338         idle = 1;
3339         ITERATE_RDEV(mddev,rdev,tmp) {
3340                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3341                 curr_events = disk_stat_read(disk, read_sectors) +
3342                                 disk_stat_read(disk, write_sectors) -
3343                                 atomic_read(&disk->sync_io);
3344                 /* Allow some slack between valud of curr_events and last_events,
3345                  * as there are some uninteresting races.
3346                  * Note: the following is an unsigned comparison.
3347                  */
3348                 if ((curr_events - rdev->last_events + 32) > 64) {
3349                         rdev->last_events = curr_events;
3350                         idle = 0;
3351                 }
3352         }
3353         return idle;
3354 }
3355
3356 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3357 {
3358         /* another "blocks" (512byte) blocks have been synced */
3359         atomic_sub(blocks, &mddev->recovery_active);
3360         wake_up(&mddev->recovery_wait);
3361         if (!ok) {
3362                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3363                 md_wakeup_thread(mddev->thread);
3364                 // stop recovery, signal do_sync ....
3365         }
3366 }
3367
3368
3369 /* md_write_start(mddev, bi)
3370  * If we need to update some array metadata (e.g. 'active' flag
3371  * in superblock) before writing, queue bi for later writing
3372  * and return 0, else return 1 and it will be written now
3373  */
3374 int md_write_start(mddev_t *mddev, struct bio *bi)
3375 {
3376         if (bio_data_dir(bi) != WRITE)
3377                 return 1;
3378
3379         atomic_inc(&mddev->writes_pending);
3380         spin_lock(&mddev->write_lock);
3381         if (mddev->in_sync == 0 && mddev->sb_dirty == 0) {
3382                 spin_unlock(&mddev->write_lock);
3383                 return 1;
3384         }
3385         bio_list_add(&mddev->write_list, bi);
3386
3387         if (mddev->in_sync) {
3388                 mddev->in_sync = 0;
3389                 mddev->sb_dirty = 1;
3390         }
3391         spin_unlock(&mddev->write_lock);
3392         md_wakeup_thread(mddev->thread);
3393         return 0;
3394 }
3395
3396 void md_write_end(mddev_t *mddev)
3397 {
3398         if (atomic_dec_and_test(&mddev->writes_pending)) {
3399                 if (mddev->safemode == 2)
3400                         md_wakeup_thread(mddev->thread);
3401                 else
3402                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3403         }
3404 }
3405
3406 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3407
3408 #define SYNC_MARKS      10
3409 #define SYNC_MARK_STEP  (3*HZ)
3410 static void md_do_sync(mddev_t *mddev)
3411 {
3412         mddev_t *mddev2;
3413         unsigned int currspeed = 0,
3414                  window;
3415         sector_t max_sectors,j, io_sectors;
3416         unsigned long mark[SYNC_MARKS];
3417         sector_t mark_cnt[SYNC_MARKS];
3418         int last_mark,m;
3419         struct list_head *tmp;
3420         sector_t last_check;
3421         int skipped = 0;
3422
3423         /* just incase thread restarts... */
3424         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3425                 return;
3426
3427         /* we overload curr_resync somewhat here.
3428          * 0 == not engaged in resync at all
3429          * 2 == checking that there is no conflict with another sync
3430          * 1 == like 2, but have yielded to allow conflicting resync to
3431          *              commense
3432          * other == active in resync - this many blocks
3433          *
3434          * Before starting a resync we must have set curr_resync to
3435          * 2, and then checked that every "conflicting" array has curr_resync
3436          * less than ours.  When we find one that is the same or higher
3437          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3438          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3439          * This will mean we have to start checking from the beginning again.
3440          *
3441          */
3442
3443         do {
3444                 mddev->curr_resync = 2;
3445
3446         try_again:
3447                 if (signal_pending(current)) {
3448                         flush_signals(current);
3449                         goto skip;
3450                 }
3451                 ITERATE_MDDEV(mddev2,tmp) {
3452                         printk(".");
3453                         if (mddev2 == mddev)
3454                                 continue;
3455                         if (mddev2->curr_resync &&
3456                             match_mddev_units(mddev,mddev2)) {
3457                                 DEFINE_WAIT(wq);
3458                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3459                                         /* arbitrarily yield */
3460                                         mddev->curr_resync = 1;
3461                                         wake_up(&resync_wait);
3462                                 }
3463                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3464                                         /* no need to wait here, we can wait the next
3465                                          * time 'round when curr_resync == 2
3466                                          */
3467                                         continue;
3468                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3469                                 if (!signal_pending(current)
3470                                     && mddev2->curr_resync >= mddev->curr_resync) {
3471                                         printk(KERN_INFO "md: delaying resync of %s"
3472                                                " until %s has finished resync (they"
3473                                                " share one or more physical units)\n",
3474                                                mdname(mddev), mdname(mddev2));
3475                                         mddev_put(mddev2);
3476                                         schedule();
3477                                         finish_wait(&resync_wait, &wq);
3478                                         goto try_again;
3479                                 }
3480                                 finish_wait(&resync_wait, &wq);
3481                         }
3482                 }
3483         } while (mddev->curr_resync < 2);
3484
3485         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3486                 /* resync follows the size requested by the personality,
3487                  * which defaults to physical size, but can be virtual size
3488                  */
3489                 max_sectors = mddev->resync_max_sectors;
3490         else
3491                 /* recovery follows the physical size of devices */
3492                 max_sectors = mddev->size << 1;
3493
3494         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3495         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3496                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3497         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3498                "(but not more than %d KB/sec) for reconstruction.\n",
3499                sysctl_speed_limit_max);
3500
3501         is_mddev_idle(mddev); /* this also initializes IO event counters */
3502         /* we don't use the checkpoint if there's a bitmap */
3503         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3504                 j = mddev->recovery_cp;
3505         else
3506                 j = 0;
3507         io_sectors = 0;
3508         for (m = 0; m < SYNC_MARKS; m++) {
3509                 mark[m] = jiffies;
3510                 mark_cnt[m] = io_sectors;
3511         }
3512         last_mark = 0;
3513         mddev->resync_mark = mark[last_mark];
3514         mddev->resync_mark_cnt = mark_cnt[last_mark];
3515
3516         /*
3517          * Tune reconstruction:
3518          */
3519         window = 32*(PAGE_SIZE/512);
3520         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3521                 window/2,(unsigned long long) max_sectors/2);
3522
3523         atomic_set(&mddev->recovery_active, 0);
3524         init_waitqueue_head(&mddev->recovery_wait);
3525         last_check = 0;
3526
3527         if (j>2) {
3528                 printk(KERN_INFO
3529                         "md: resuming recovery of %s from checkpoint.\n",
3530                         mdname(mddev));
3531                 mddev->curr_resync = j;
3532         }
3533
3534         while (j < max_sectors) {
3535                 sector_t sectors;
3536
3537                 skipped = 0;
3538                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3539                                             currspeed < sysctl_speed_limit_min);
3540                 if (sectors == 0) {
3541                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3542                         goto out;
3543                 }
3544
3545                 if (!skipped) { /* actual IO requested */
3546                         io_sectors += sectors;
3547                         atomic_add(sectors, &mddev->recovery_active);
3548                 }
3549
3550                 j += sectors;
3551                 if (j>1) mddev->curr_resync = j;
3552
3553
3554                 if (last_check + window > io_sectors || j == max_sectors)
3555                         continue;
3556
3557                 last_check = io_sectors;
3558
3559                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3560                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3561                         break;
3562
3563         repeat:
3564                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3565                         /* step marks */
3566                         int next = (last_mark+1) % SYNC_MARKS;
3567
3568                         mddev->resync_mark = mark[next];
3569                         mddev->resync_mark_cnt = mark_cnt[next];
3570                         mark[next] = jiffies;
3571                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3572                         last_mark = next;
3573                 }
3574
3575
3576                 if (signal_pending(current)) {
3577                         /*
3578                          * got a signal, exit.
3579                          */
3580                         printk(KERN_INFO
3581                                 "md: md_do_sync() got signal ... exiting\n");
3582                         flush_signals(current);
3583                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3584                         goto out;
3585                 }
3586
3587                 /*
3588                  * this loop exits only if either when we are slower than
3589                  * the 'hard' speed limit, or the system was IO-idle for
3590                  * a jiffy.
3591                  * the system might be non-idle CPU-wise, but we only care
3592                  * about not overloading the IO subsystem. (things like an
3593                  * e2fsck being done on the RAID array should execute fast)
3594                  */
3595                 mddev->queue->unplug_fn(mddev->queue);
3596                 cond_resched();
3597
3598                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3599                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
3600
3601                 if (currspeed > sysctl_speed_limit_min) {
3602                         if ((currspeed > sysctl_speed_limit_max) ||
3603                                         !is_mddev_idle(mddev)) {
3604                                 msleep_interruptible(250);
3605                                 goto repeat;
3606                         }
3607                 }
3608         }
3609         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3610         /*
3611          * this also signals 'finished resyncing' to md_stop
3612          */
3613  out:
3614         mddev->queue->unplug_fn(mddev->queue);
3615
3616         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3617
3618         /* tell personality that we are finished */
3619         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3620
3621         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3622             mddev->curr_resync > 2 &&
3623             mddev->curr_resync >= mddev->recovery_cp) {
3624                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3625                         printk(KERN_INFO
3626                                 "md: checkpointing recovery of %s.\n",
3627                                 mdname(mddev));
3628                         mddev->recovery_cp = mddev->curr_resync;
3629                 } else
3630                         mddev->recovery_cp = MaxSector;
3631         }
3632
3633  skip:
3634         mddev->curr_resync = 0;
3635         wake_up(&resync_wait);
3636         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3637         md_wakeup_thread(mddev->thread);
3638 }
3639
3640
3641 /*
3642  * This routine is regularly called by all per-raid-array threads to
3643  * deal with generic issues like resync and super-block update.
3644  * Raid personalities that don't have a thread (linear/raid0) do not
3645  * need this as they never do any recovery or update the superblock.
3646  *
3647  * It does not do any resync itself, but rather "forks" off other threads
3648  * to do that as needed.
3649  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3650  * "->recovery" and create a thread at ->sync_thread.
3651  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3652  * and wakeups up this thread which will reap the thread and finish up.
3653  * This thread also removes any faulty devices (with nr_pending == 0).
3654  *
3655  * The overall approach is:
3656  *  1/ if the superblock needs updating, update it.
3657  *  2/ If a recovery thread is running, don't do anything else.
3658  *  3/ If recovery has finished, clean up, possibly marking spares active.
3659  *  4/ If there are any faulty devices, remove them.
3660  *  5/ If array is degraded, try to add spares devices
3661  *  6/ If array has spares or is not in-sync, start a resync thread.
3662  */
3663 void md_check_recovery(mddev_t *mddev)
3664 {
3665         mdk_rdev_t *rdev;
3666         struct list_head *rtmp;
3667
3668
3669         if (mddev->bitmap)
3670                 bitmap_daemon_work(mddev->bitmap);
3671
3672         if (mddev->ro)
3673                 return;
3674
3675         if (signal_pending(current)) {
3676                 if (mddev->pers->sync_request) {
3677                         printk(KERN_INFO "md: %s in immediate safe mode\n",
3678                                mdname(mddev));
3679                         mddev->safemode = 2;
3680                 }
3681                 flush_signals(current);
3682         }
3683
3684         if ( ! (
3685                 mddev->sb_dirty ||
3686                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3687                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3688                 mddev->write_list.head ||
3689                 (mddev->safemode == 1) ||
3690                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3691                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3692                 ))
3693                 return;
3694
3695         if (mddev_trylock(mddev)==0) {
3696                 int spares =0;
3697                 struct bio *blist;
3698
3699                 spin_lock(&mddev->write_lock);
3700                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3701                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3702                         mddev->in_sync = 1;
3703                         mddev->sb_dirty = 1;
3704                 }
3705                 if (mddev->safemode == 1)
3706                         mddev->safemode = 0;
3707                 blist = bio_list_get(&mddev->write_list);
3708                 spin_unlock(&mddev->write_lock);
3709
3710                 if (mddev->sb_dirty)
3711                         md_update_sb(mddev);
3712
3713                 while (blist) {
3714                         struct bio *b = blist;
3715                         blist = blist->bi_next;
3716                         b->bi_next = NULL;
3717                         generic_make_request(b);
3718                         /* we already counted this, so need to un-count */
3719                         md_write_end(mddev);
3720                 }
3721
3722
3723                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3724                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3725                         /* resync/recovery still happening */
3726                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3727                         goto unlock;
3728                 }
3729                 if (mddev->sync_thread) {
3730                         /* resync has finished, collect result */
3731                         md_unregister_thread(mddev->sync_thread);
3732                         mddev->sync_thread = NULL;
3733                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3734                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3735                                 /* success...*/
3736                                 /* activate any spares */
3737                                 mddev->pers->spare_active(mddev);
3738                         }
3739                         md_update_sb(mddev);
3740
3741                         /* if array is no-longer degraded, then any saved_raid_disk
3742                          * information must be scrapped
3743                          */
3744                         if (!mddev->degraded)
3745                                 ITERATE_RDEV(mddev,rdev,rtmp)
3746                                         rdev->saved_raid_disk = -1;
3747
3748                         mddev->recovery = 0;
3749                         /* flag recovery needed just to double check */
3750                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3751                         goto unlock;
3752                 }
3753                 if (mddev->recovery)
3754                         /* probably just the RECOVERY_NEEDED flag */
3755                         mddev->recovery = 0;
3756
3757                 /* no recovery is running.
3758                  * remove any failed drives, then
3759                  * add spares if possible.
3760                  * Spare are also removed and re-added, to allow
3761                  * the personality to fail the re-add.
3762                  */
3763                 ITERATE_RDEV(mddev,rdev,rtmp)
3764                         if (rdev->raid_disk >= 0 &&
3765                             (rdev->faulty || ! rdev->in_sync) &&
3766                             atomic_read(&rdev->nr_pending)==0) {
3767                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3768                                         rdev->raid_disk = -1;
3769                         }
3770
3771                 if (mddev->degraded) {
3772                         ITERATE_RDEV(mddev,rdev,rtmp)
3773                                 if (rdev->raid_disk < 0
3774                                     && !rdev->faulty) {
3775                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3776                                                 spares++;
3777                                         else
3778                                                 break;
3779                                 }
3780                 }
3781
3782                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3783                         /* nothing we can do ... */
3784                         goto unlock;
3785                 }
3786                 if (mddev->pers->sync_request) {
3787                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3788                         if (!spares)
3789                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3790                         mddev->sync_thread = md_register_thread(md_do_sync,
3791                                                                 mddev,
3792                                                                 "%s_resync");
3793                         if (!mddev->sync_thread) {
3794                                 printk(KERN_ERR "%s: could not start resync"
3795                                         " thread...\n",
3796                                         mdname(mddev));
3797                                 /* leave the spares where they are, it shouldn't hurt */
3798                                 mddev->recovery = 0;
3799                         } else {
3800                                 md_wakeup_thread(mddev->sync_thread);
3801                         }
3802                 }
3803         unlock:
3804                 mddev_unlock(mddev);
3805         }
3806 }
3807
3808 static int md_notify_reboot(struct notifier_block *this,
3809                             unsigned long code, void *x)
3810 {
3811         struct list_head *tmp;
3812         mddev_t *mddev;
3813
3814         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3815
3816                 printk(KERN_INFO "md: stopping all md devices.\n");
3817
3818                 ITERATE_MDDEV(mddev,tmp)
3819                         if (mddev_trylock(mddev)==0)
3820                                 do_md_stop (mddev, 1);
3821                 /*
3822                  * certain more exotic SCSI devices are known to be
3823                  * volatile wrt too early system reboots. While the
3824                  * right place to handle this issue is the given
3825                  * driver, we do want to have a safe RAID driver ...
3826                  */
3827                 mdelay(1000*1);
3828         }
3829         return NOTIFY_DONE;
3830 }
3831
3832 static struct notifier_block md_notifier = {
3833         .notifier_call  = md_notify_reboot,
3834         .next           = NULL,
3835         .priority       = INT_MAX, /* before any real devices */
3836 };
3837
3838 static void md_geninit(void)
3839 {
3840         struct proc_dir_entry *p;
3841
3842         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3843
3844         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3845         if (p)
3846                 p->proc_fops = &md_seq_fops;
3847 }
3848
3849 static int __init md_init(void)
3850 {
3851         int minor;
3852
3853         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3854                         " MD_SB_DISKS=%d\n",
3855                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3856                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3857         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3858                         BITMAP_MINOR);
3859
3860         if (register_blkdev(MAJOR_NR, "md"))
3861                 return -1;
3862         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3863                 unregister_blkdev(MAJOR_NR, "md");
3864                 return -1;
3865         }
3866         devfs_mk_dir("md");
3867         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3868                                 md_probe, NULL, NULL);
3869         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3870                             md_probe, NULL, NULL);
3871
3872         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3873                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3874                                 S_IFBLK|S_IRUSR|S_IWUSR,
3875                                 "md/%d", minor);
3876
3877         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3878                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3879                               S_IFBLK|S_IRUSR|S_IWUSR,
3880                               "md/mdp%d", minor);
3881
3882
3883         register_reboot_notifier(&md_notifier);
3884         raid_table_header = register_sysctl_table(raid_root_table, 1);
3885
3886         md_geninit();
3887         return (0);
3888 }
3889
3890
3891 #ifndef MODULE
3892
3893 /*
3894  * Searches all registered partitions for autorun RAID arrays
3895  * at boot time.
3896  */
3897 static dev_t detected_devices[128];
3898 static int dev_cnt;
3899
3900 void md_autodetect_dev(dev_t dev)
3901 {
3902         if (dev_cnt >= 0 && dev_cnt < 127)
3903                 detected_devices[dev_cnt++] = dev;
3904 }
3905
3906
3907 static void autostart_arrays(int part)
3908 {
3909         mdk_rdev_t *rdev;
3910         int i;
3911
3912         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3913
3914         for (i = 0; i < dev_cnt; i++) {
3915                 dev_t dev = detected_devices[i];
3916
3917                 rdev = md_import_device(dev,0, 0);
3918                 if (IS_ERR(rdev))
3919                         continue;
3920
3921                 if (rdev->faulty) {
3922                         MD_BUG();
3923                         continue;
3924                 }
3925                 list_add(&rdev->same_set, &pending_raid_disks);
3926         }
3927         dev_cnt = 0;
3928
3929         autorun_devices(part);
3930 }
3931
3932 #endif
3933
3934 static __exit void md_exit(void)
3935 {
3936         mddev_t *mddev;
3937         struct list_head *tmp;
3938         int i;
3939         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3940         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3941         for (i=0; i < MAX_MD_DEVS; i++)
3942                 devfs_remove("md/%d", i);
3943         for (i=0; i < MAX_MD_DEVS; i++)
3944                 devfs_remove("md/d%d", i);
3945
3946         devfs_remove("md");
3947
3948         unregister_blkdev(MAJOR_NR,"md");
3949         unregister_blkdev(mdp_major, "mdp");
3950         unregister_reboot_notifier(&md_notifier);
3951         unregister_sysctl_table(raid_table_header);
3952         remove_proc_entry("mdstat", NULL);
3953         ITERATE_MDDEV(mddev,tmp) {
3954                 struct gendisk *disk = mddev->gendisk;
3955                 if (!disk)
3956                         continue;
3957                 export_array(mddev);
3958                 del_gendisk(disk);
3959                 put_disk(disk);
3960                 mddev->gendisk = NULL;
3961                 mddev_put(mddev);
3962         }
3963 }
3964
3965 module_init(md_init)
3966 module_exit(md_exit)
3967
3968 EXPORT_SYMBOL(register_md_personality);
3969 EXPORT_SYMBOL(unregister_md_personality);
3970 EXPORT_SYMBOL(md_error);
3971 EXPORT_SYMBOL(md_done_sync);
3972 EXPORT_SYMBOL(md_write_start);
3973 EXPORT_SYMBOL(md_write_end);
3974 EXPORT_SYMBOL(md_register_thread);
3975 EXPORT_SYMBOL(md_unregister_thread);
3976 EXPORT_SYMBOL(md_wakeup_thread);
3977 EXPORT_SYMBOL(md_print_devices);
3978 EXPORT_SYMBOL(md_check_recovery);
3979 MODULE_LICENSE("GPL");