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/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45
  46 #include <linux/init.h>
  47
  48 #include <linux/file.h>
  49
  50 #ifdef CONFIG_KMOD
  51 #include <linux/kmod.h>
  52 #endif
  53
  54 #include <asm/unaligned.h>
  55
  56 #define MAJOR_NR MD_MAJOR
  57 #define MD_DRIVER
  58
  59 /* 63 partitions with the alternate major number (mdp) */
  60 #define MdpMinorShift 6
  61
  62 #define DEBUG 0
  63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  64
  65
  66 #ifndef MODULE
  67 static void autostart_arrays (int part);
  68 #endif
  69
  70 static mdk_personality_t *pers[MAX_PERSONALITY];
  71 static DEFINE_SPINLOCK(pers_lock);
  72
  73 /*
  74  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  75  * is 1000 KB/sec, so the extra system load does not show up that much.
  76  * Increase it if you want to have more _guaranteed_ speed. Note that
  77  * the RAID driver will use the maximum available bandwidth if the IO
  78  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  79  * speed limit - in case reconstruction slows down your system despite
  80  * idle IO detection.
  81  *
  82  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  83  */
  84
  85 static int sysctl_speed_limit_min = 1000;
  86 static int sysctl_speed_limit_max = 200000;
  87
  88 static struct ctl_table_header *raid_table_header;
  89
  90 static ctl_table raid_table[] = {
  91         {
  92                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  93                 .procname       = "speed_limit_min",
  94                 .data           = &sysctl_speed_limit_min,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec,
  98         },
  99         {
 100                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 101                 .procname       = "speed_limit_max",
 102                 .data           = &sysctl_speed_limit_max,
 103                 .maxlen         = sizeof(int),
 104                 .mode           = 0644,
 105                 .proc_handler   = &proc_dointvec,
 106         },
 107         { .ctl_name = 0 }
 108 };
 109
 110 static ctl_table raid_dir_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID,
 113                 .procname       = "raid",
 114                 .maxlen         = 0,
 115                 .mode           = 0555,
 116                 .child          = raid_table,
 117         },
 118         { .ctl_name = 0 }
 119 };
 120
 121 static ctl_table raid_root_table[] = {
 122         {
 123                 .ctl_name       = CTL_DEV,
 124                 .procname       = "dev",
 125                 .maxlen         = 0,
 126                 .mode           = 0555,
 127                 .child          = raid_dir_table,
 128         },
 129         { .ctl_name = 0 }
 130 };
 131
 132 static struct block_device_operations md_fops;
 133
 134 /*
 135  * Enables to iterate over all existing md arrays
 136  * all_mddevs_lock protects this list.
 137  */
 138 static LIST_HEAD(all_mddevs);
 139 static DEFINE_SPINLOCK(all_mddevs_lock);
 140
 141
 142 /*
 143  * iterates through all used mddevs in the system.
 144  * We take care to grab the all_mddevs_lock whenever navigating
 145  * the list, and to always hold a refcount when unlocked.
 146  * Any code which breaks out of this loop while own
 147  * a reference to the current mddev and must mddev_put it.
 148  */
 149 #define ITERATE_MDDEV(mddev,tmp)                                        \
 150                                                                         \
 151         for (({ spin_lock(&all_mddevs_lock);                            \
 152                 tmp = all_mddevs.next;                                  \
 153                 mddev = NULL;});                                        \
 154              ({ if (tmp != &all_mddevs)                                 \
 155                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 156                 spin_unlock(&all_mddevs_lock);                          \
 157                 if (mddev) mddev_put(mddev);                            \
 158                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 159                 tmp != &all_mddevs;});                                  \
 160              ({ spin_lock(&all_mddevs_lock);                            \
 161                 tmp = tmp->next;})                                      \
 162                 )
 163
 164
 165 static int md_fail_request (request_queue_t *q, struct bio *bio)
 166 {
 167         bio_io_error(bio, bio->bi_size);
 168         return 0;
 169 }
 170
 171 static inline mddev_t *mddev_get(mddev_t *mddev)
 172 {
 173         atomic_inc(&mddev->active);
 174         return mddev;
 175 }
 176
 177 static void mddev_put(mddev_t *mddev)
 178 {
 179         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 180                 return;
 181         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 182                 list_del(&mddev->all_mddevs);
 183                 blk_put_queue(mddev->queue);
 184                 kobject_unregister(&mddev->kobj);
 185         }
 186         spin_unlock(&all_mddevs_lock);
 187 }
 188
 189 static mddev_t * mddev_find(dev_t unit)
 190 {
 191         mddev_t *mddev, *new = NULL;
 192
 193  retry:
 194         spin_lock(&all_mddevs_lock);
 195         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 196                 if (mddev->unit == unit) {
 197                         mddev_get(mddev);
 198                         spin_unlock(&all_mddevs_lock);
 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         spin_lock_init(&new->write_lock);
 228         init_waitqueue_head(&new->sb_wait);
 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         md_wakeup_thread(mddev->thread);
 261 }
 262
 263 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 264 {
 265         mdk_rdev_t * rdev;
 266         struct list_head *tmp;
 267
 268         ITERATE_RDEV(mddev,rdev,tmp) {
 269                 if (rdev->desc_nr == nr)
 270                         return rdev;
 271         }
 272         return NULL;
 273 }
 274
 275 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 276 {
 277         struct list_head *tmp;
 278         mdk_rdev_t *rdev;
 279
 280         ITERATE_RDEV(mddev,rdev,tmp) {
 281                 if (rdev->bdev->bd_dev == dev)
 282                         return rdev;
 283         }
 284         return NULL;
 285 }
 286
 287 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 288 {
 289         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 290         return MD_NEW_SIZE_BLOCKS(size);
 291 }
 292
 293 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 294 {
 295         sector_t size;
 296
 297         size = rdev->sb_offset;
 298
 299         if (chunk_size)
 300                 size &= ~((sector_t)chunk_size/1024 - 1);
 301         return size;
 302 }
 303
 304 static int alloc_disk_sb(mdk_rdev_t * rdev)
 305 {
 306         if (rdev->sb_page)
 307                 MD_BUG();
 308
 309         rdev->sb_page = alloc_page(GFP_KERNEL);
 310         if (!rdev->sb_page) {
 311                 printk(KERN_ALERT "md: out of memory.\n");
 312                 return -EINVAL;
 313         }
 314
 315         return 0;
 316 }
 317
 318 static void free_disk_sb(mdk_rdev_t * rdev)
 319 {
 320         if (rdev->sb_page) {
 321                 page_cache_release(rdev->sb_page);
 322                 rdev->sb_loaded = 0;
 323                 rdev->sb_page = NULL;
 324                 rdev->sb_offset = 0;
 325                 rdev->size = 0;
 326         }
 327 }
 328
 329
 330 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 331 {
 332         mdk_rdev_t *rdev = bio->bi_private;
 333         if (bio->bi_size)
 334                 return 1;
 335
 336         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 337                 md_error(rdev->mddev, rdev);
 338
 339         if (atomic_dec_and_test(&rdev->mddev->pending_writes))
 340                 wake_up(&rdev->mddev->sb_wait);
 341         bio_put(bio);
 342         return 0;
 343 }
 344
 345 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 346                    sector_t sector, int size, struct page *page)
 347 {
 348         /* write first size bytes of page to sector of rdev
 349          * Increment mddev->pending_writes before returning
 350          * and decrement it on completion, waking up sb_wait
 351          * if zero is reached.
 352          * If an error occurred, call md_error
 353          */
 354         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 355
 356         bio->bi_bdev = rdev->bdev;
 357         bio->bi_sector = sector;
 358         bio_add_page(bio, page, size, 0);
 359         bio->bi_private = rdev;
 360         bio->bi_end_io = super_written;
 361         atomic_inc(&mddev->pending_writes);
 362         submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
 363 }
 364
 365 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 366 {
 367         if (bio->bi_size)
 368                 return 1;
 369
 370         complete((struct completion*)bio->bi_private);
 371         return 0;
 372 }
 373
 374 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 375                    struct page *page, int rw)
 376 {
 377         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 378         struct completion event;
 379         int ret;
 380
 381         rw |= (1 << BIO_RW_SYNC);
 382
 383         bio->bi_bdev = bdev;
 384         bio->bi_sector = sector;
 385         bio_add_page(bio, page, size, 0);
 386         init_completion(&event);
 387         bio->bi_private = &event;
 388         bio->bi_end_io = bi_complete;
 389         submit_bio(rw, bio);
 390         wait_for_completion(&event);
 391
 392         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 393         bio_put(bio);
 394         return ret;
 395 }
 396
 397 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 398 {
 399         char b[BDEVNAME_SIZE];
 400         if (!rdev->sb_page) {
 401                 MD_BUG();
 402                 return -EINVAL;
 403         }
 404         if (rdev->sb_loaded)
 405                 return 0;
 406
 407
 408         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 409                 goto fail;
 410         rdev->sb_loaded = 1;
 411         return 0;
 412
 413 fail:
 414         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 415                 bdevname(rdev->bdev,b));
 416         return -EINVAL;
 417 }
 418
 419 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 420 {
 421         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 422                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 423                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 424                 (sb1->set_uuid3 == sb2->set_uuid3))
 425
 426                 return 1;
 427
 428         return 0;
 429 }
 430
 431
 432 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 433 {
 434         int ret;
 435         mdp_super_t *tmp1, *tmp2;
 436
 437         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 438         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 439
 440         if (!tmp1 || !tmp2) {
 441                 ret = 0;
 442                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 443                 goto abort;
 444         }
 445
 446         *tmp1 = *sb1;
 447         *tmp2 = *sb2;
 448
 449         /*
 450          * nr_disks is not constant
 451          */
 452         tmp1->nr_disks = 0;
 453         tmp2->nr_disks = 0;
 454
 455         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 456                 ret = 0;
 457         else
 458                 ret = 1;
 459
 460 abort:
 461         kfree(tmp1);
 462         kfree(tmp2);
 463         return ret;
 464 }
 465
 466 static unsigned int calc_sb_csum(mdp_super_t * sb)
 467 {
 468         unsigned int disk_csum, csum;
 469
 470         disk_csum = sb->sb_csum;
 471         sb->sb_csum = 0;
 472         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 473         sb->sb_csum = disk_csum;
 474         return csum;
 475 }
 476
 477
 478 /*
 479  * Handle superblock details.
 480  * We want to be able to handle multiple superblock formats
 481  * so we have a common interface to them all, and an array of
 482  * different handlers.
 483  * We rely on user-space to write the initial superblock, and support
 484  * reading and updating of superblocks.
 485  * Interface methods are:
 486  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 487  *      loads and validates a superblock on dev.
 488  *      if refdev != NULL, compare superblocks on both devices
 489  *    Return:
 490  *      0 - dev has a superblock that is compatible with refdev
 491  *      1 - dev has a superblock that is compatible and newer than refdev
 492  *          so dev should be used as the refdev in future
 493  *     -EINVAL superblock incompatible or invalid
 494  *     -othererror e.g. -EIO
 495  *
 496  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 497  *      Verify that dev is acceptable into mddev.
 498  *       The first time, mddev->raid_disks will be 0, and data from
 499  *       dev should be merged in.  Subsequent calls check that dev
 500  *       is new enough.  Return 0 or -EINVAL
 501  *
 502  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 503  *     Update the superblock for rdev with data in mddev
 504  *     This does not write to disc.
 505  *
 506  */
 507
 508 struct super_type  {
 509         char            *name;
 510         struct module   *owner;
 511         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 512         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 513         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 514 };
 515
 516 /*
 517  * load_super for 0.90.0
 518  */
 519 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 520 {
 521         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 522         mdp_super_t *sb;
 523         int ret;
 524         sector_t sb_offset;
 525
 526         /*
 527          * Calculate the position of the superblock,
 528          * it's at the end of the disk.
 529          *
 530          * It also happens to be a multiple of 4Kb.
 531          */
 532         sb_offset = calc_dev_sboffset(rdev->bdev);
 533         rdev->sb_offset = sb_offset;
 534
 535         ret = read_disk_sb(rdev, MD_SB_BYTES);
 536         if (ret) return ret;
 537
 538         ret = -EINVAL;
 539
 540         bdevname(rdev->bdev, b);
 541         sb = (mdp_super_t*)page_address(rdev->sb_page);
 542
 543         if (sb->md_magic != MD_SB_MAGIC) {
 544                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 545                        b);
 546                 goto abort;
 547         }
 548
 549         if (sb->major_version != 0 ||
 550             sb->minor_version != 90) {
 551                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 552                         sb->major_version, sb->minor_version,
 553                         b);
 554                 goto abort;
 555         }
 556
 557         if (sb->raid_disks <= 0)
 558                 goto abort;
 559
 560         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 561                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 562                         b);
 563                 goto abort;
 564         }
 565
 566         rdev->preferred_minor = sb->md_minor;
 567         rdev->data_offset = 0;
 568         rdev->sb_size = MD_SB_BYTES;
 569
 570         if (sb->level == LEVEL_MULTIPATH)
 571                 rdev->desc_nr = -1;
 572         else
 573                 rdev->desc_nr = sb->this_disk.number;
 574
 575         if (refdev == 0)
 576                 ret = 1;
 577         else {
 578                 __u64 ev1, ev2;
 579                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 580                 if (!uuid_equal(refsb, sb)) {
 581                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 582                                 b, bdevname(refdev->bdev,b2));
 583                         goto abort;
 584                 }
 585                 if (!sb_equal(refsb, sb)) {
 586                         printk(KERN_WARNING "md: %s has same UUID"
 587                                " but different superblock to %s\n",
 588                                b, bdevname(refdev->bdev, b2));
 589                         goto abort;
 590                 }
 591                 ev1 = md_event(sb);
 592                 ev2 = md_event(refsb);
 593                 if (ev1 > ev2)
 594                         ret = 1;
 595                 else
 596                         ret = 0;
 597         }
 598         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 599
 600  abort:
 601         return ret;
 602 }
 603
 604 /*
 605  * validate_super for 0.90.0
 606  */
 607 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 608 {
 609         mdp_disk_t *desc;
 610         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 611
 612         rdev->raid_disk = -1;
 613         rdev->in_sync = 0;
 614         if (mddev->raid_disks == 0) {
 615                 mddev->major_version = 0;
 616                 mddev->minor_version = sb->minor_version;
 617                 mddev->patch_version = sb->patch_version;
 618                 mddev->persistent = ! sb->not_persistent;
 619                 mddev->chunk_size = sb->chunk_size;
 620                 mddev->ctime = sb->ctime;
 621                 mddev->utime = sb->utime;
 622                 mddev->level = sb->level;
 623                 mddev->layout = sb->layout;
 624                 mddev->raid_disks = sb->raid_disks;
 625                 mddev->size = sb->size;
 626                 mddev->events = md_event(sb);
 627                 mddev->bitmap_offset = 0;
 628                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 629
 630                 if (sb->state & (1<<MD_SB_CLEAN))
 631                         mddev->recovery_cp = MaxSector;
 632                 else {
 633                         if (sb->events_hi == sb->cp_events_hi &&
 634                                 sb->events_lo == sb->cp_events_lo) {
 635                                 mddev->recovery_cp = sb->recovery_cp;
 636                         } else
 637                                 mddev->recovery_cp = 0;
 638                 }
 639
 640                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 641                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 642                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 643                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 644
 645                 mddev->max_disks = MD_SB_DISKS;
 646
 647                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 648                     mddev->bitmap_file == NULL) {
 649                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
 650                                 /* FIXME use a better test */
 651                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
 652                                 return -EINVAL;
 653                         }
 654                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 655                 }
 656
 657         } else if (mddev->pers == NULL) {
 658                 /* Insist on good event counter while assembling */
 659                 __u64 ev1 = md_event(sb);
 660                 ++ev1;
 661                 if (ev1 < mddev->events)
 662                         return -EINVAL;
 663         } else if (mddev->bitmap) {
 664                 /* if adding to array with a bitmap, then we can accept an
 665                  * older device ... but not too old.
 666                  */
 667                 __u64 ev1 = md_event(sb);
 668                 if (ev1 < mddev->bitmap->events_cleared)
 669                         return 0;
 670         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 671                 return 0;
 672
 673         if (mddev->level != LEVEL_MULTIPATH) {
 674                 rdev->faulty = 0;
 675                 rdev->flags = 0;
 676                 desc = sb->disks + rdev->desc_nr;
 677
 678                 if (desc->state & (1<<MD_DISK_FAULTY))
 679                         rdev->faulty = 1;
 680                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 681                          desc->raid_disk < mddev->raid_disks) {
 682                         rdev->in_sync = 1;
 683                         rdev->raid_disk = desc->raid_disk;
 684                 }
 685                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 686                         set_bit(WriteMostly, &rdev->flags);
 687         } else /* MULTIPATH are always insync */
 688                 rdev->in_sync = 1;
 689         return 0;
 690 }
 691
 692 /*
 693  * sync_super for 0.90.0
 694  */
 695 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 696 {
 697         mdp_super_t *sb;
 698         struct list_head *tmp;
 699         mdk_rdev_t *rdev2;
 700         int next_spare = mddev->raid_disks;
 701
 702         /* make rdev->sb match mddev data..
 703          *
 704          * 1/ zero out disks
 705          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 706          * 3/ any empty disks < next_spare become removed
 707          *
 708          * disks[0] gets initialised to REMOVED because
 709          * we cannot be sure from other fields if it has
 710          * been initialised or not.
 711          */
 712         int i;
 713         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 714         unsigned int fixdesc=0;
 715
 716         rdev->sb_size = MD_SB_BYTES;
 717
 718         sb = (mdp_super_t*)page_address(rdev->sb_page);
 719
 720         memset(sb, 0, sizeof(*sb));
 721
 722         sb->md_magic = MD_SB_MAGIC;
 723         sb->major_version = mddev->major_version;
 724         sb->minor_version = mddev->minor_version;
 725         sb->patch_version = mddev->patch_version;
 726         sb->gvalid_words  = 0; /* ignored */
 727         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 728         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 729         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 730         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 731
 732         sb->ctime = mddev->ctime;
 733         sb->level = mddev->level;
 734         sb->size  = mddev->size;
 735         sb->raid_disks = mddev->raid_disks;
 736         sb->md_minor = mddev->md_minor;
 737         sb->not_persistent = !mddev->persistent;
 738         sb->utime = mddev->utime;
 739         sb->state = 0;
 740         sb->events_hi = (mddev->events>>32);
 741         sb->events_lo = (u32)mddev->events;
 742
 743         if (mddev->in_sync)
 744         {
 745                 sb->recovery_cp = mddev->recovery_cp;
 746                 sb->cp_events_hi = (mddev->events>>32);
 747                 sb->cp_events_lo = (u32)mddev->events;
 748                 if (mddev->recovery_cp == MaxSector)
 749                         sb->state = (1<< MD_SB_CLEAN);
 750         } else
 751                 sb->recovery_cp = 0;
 752
 753         sb->layout = mddev->layout;
 754         sb->chunk_size = mddev->chunk_size;
 755
 756         if (mddev->bitmap && mddev->bitmap_file == NULL)
 757                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 758
 759         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 760         ITERATE_RDEV(mddev,rdev2,tmp) {
 761                 mdp_disk_t *d;
 762                 int desc_nr;
 763                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 764                         desc_nr = rdev2->raid_disk;
 765                 else
 766                         desc_nr = next_spare++;
 767                 if (desc_nr != rdev2->desc_nr) {
 768                         fixdesc |= (1 << desc_nr);
 769                         rdev2->desc_nr = desc_nr;
 770                         if (rdev2->raid_disk >= 0) {
 771                                 char nm[20];
 772                                 sprintf(nm, "rd%d", rdev2->raid_disk);
 773                                 sysfs_remove_link(&mddev->kobj, nm);
 774                         }
 775                         sysfs_remove_link(&rdev2->kobj, "block");
 776                         kobject_del(&rdev2->kobj);
 777                 }
 778                 d = &sb->disks[rdev2->desc_nr];
 779                 nr_disks++;
 780                 d->number = rdev2->desc_nr;
 781                 d->major = MAJOR(rdev2->bdev->bd_dev);
 782                 d->minor = MINOR(rdev2->bdev->bd_dev);
 783                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 784                         d->raid_disk = rdev2->raid_disk;
 785                 else
 786                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 787                 if (rdev2->faulty) {
 788                         d->state = (1<<MD_DISK_FAULTY);
 789                         failed++;
 790                 } else if (rdev2->in_sync) {
 791                         d->state = (1<<MD_DISK_ACTIVE);
 792                         d->state |= (1<<MD_DISK_SYNC);
 793                         active++;
 794                         working++;
 795                 } else {
 796                         d->state = 0;
 797                         spare++;
 798                         working++;
 799                 }
 800                 if (test_bit(WriteMostly, &rdev2->flags))
 801                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 802         }
 803         if (fixdesc)
 804                 ITERATE_RDEV(mddev,rdev2,tmp)
 805                         if (fixdesc & (1<<rdev2->desc_nr)) {
 806                                 snprintf(rdev2->kobj.name, KOBJ_NAME_LEN, "dev%d",
 807                                          rdev2->desc_nr);
 808                                 kobject_add(&rdev2->kobj);
 809                                 sysfs_create_link(&rdev2->kobj,
 810                                                   &rdev2->bdev->bd_disk->kobj,
 811                                                   "block");
 812                                 if (rdev2->raid_disk >= 0) {
 813                                         char nm[20];
 814                                         sprintf(nm, "rd%d", rdev2->raid_disk);
 815                                         sysfs_create_link(&mddev->kobj,
 816                                                           &rdev2->kobj, nm);
 817                                 }
 818                         }
 819         /* now set the "removed" and "faulty" bits on any missing devices */
 820         for (i=0 ; i < mddev->raid_disks ; i++) {
 821                 mdp_disk_t *d = &sb->disks[i];
 822                 if (d->state == 0 && d->number == 0) {
 823                         d->number = i;
 824                         d->raid_disk = i;
 825                         d->state = (1<<MD_DISK_REMOVED);
 826                         d->state |= (1<<MD_DISK_FAULTY);
 827                         failed++;
 828                 }
 829         }
 830         sb->nr_disks = nr_disks;
 831         sb->active_disks = active;
 832         sb->working_disks = working;
 833         sb->failed_disks = failed;
 834         sb->spare_disks = spare;
 835
 836         sb->this_disk = sb->disks[rdev->desc_nr];
 837         sb->sb_csum = calc_sb_csum(sb);
 838 }
 839
 840 /*
 841  * version 1 superblock
 842  */
 843
 844 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 845 {
 846         unsigned int disk_csum, csum;
 847         unsigned long long newcsum;
 848         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 849         unsigned int *isuper = (unsigned int*)sb;
 850         int i;
 851
 852         disk_csum = sb->sb_csum;
 853         sb->sb_csum = 0;
 854         newcsum = 0;
 855         for (i=0; size>=4; size -= 4 )
 856                 newcsum += le32_to_cpu(*isuper++);
 857
 858         if (size == 2)
 859                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 860
 861         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 862         sb->sb_csum = disk_csum;
 863         return cpu_to_le32(csum);
 864 }
 865
 866 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 867 {
 868         struct mdp_superblock_1 *sb;
 869         int ret;
 870         sector_t sb_offset;
 871         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 872         int bmask;
 873
 874         /*
 875          * Calculate the position of the superblock.
 876          * It is always aligned to a 4K boundary and
 877          * depeding on minor_version, it can be:
 878          * 0: At least 8K, but less than 12K, from end of device
 879          * 1: At start of device
 880          * 2: 4K from start of device.
 881          */
 882         switch(minor_version) {
 883         case 0:
 884                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 885                 sb_offset -= 8*2;
 886                 sb_offset &= ~(sector_t)(4*2-1);
 887                 /* convert from sectors to K */
 888                 sb_offset /= 2;
 889                 break;
 890         case 1:
 891                 sb_offset = 0;
 892                 break;
 893         case 2:
 894                 sb_offset = 4;
 895                 break;
 896         default:
 897                 return -EINVAL;
 898         }
 899         rdev->sb_offset = sb_offset;
 900
 901         /* superblock is rarely larger than 1K, but it can be larger,
 902          * and it is safe to read 4k, so we do that
 903          */
 904         ret = read_disk_sb(rdev, 4096);
 905         if (ret) return ret;
 906
 907
 908         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 909
 910         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 911             sb->major_version != cpu_to_le32(1) ||
 912             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 913             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 914             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 915                 return -EINVAL;
 916
 917         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 918                 printk("md: invalid superblock checksum on %s\n",
 919                         bdevname(rdev->bdev,b));
 920                 return -EINVAL;
 921         }
 922         if (le64_to_cpu(sb->data_size) < 10) {
 923                 printk("md: data_size too small on %s\n",
 924                        bdevname(rdev->bdev,b));
 925                 return -EINVAL;
 926         }
 927         rdev->preferred_minor = 0xffff;
 928         rdev->data_offset = le64_to_cpu(sb->data_offset);
 929
 930         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 931         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 932         if (rdev->sb_size & bmask)
 933                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 934
 935         if (refdev == 0)
 936                 return 1;
 937         else {
 938                 __u64 ev1, ev2;
 939                 struct mdp_superblock_1 *refsb =
 940                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 941
 942                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 943                     sb->level != refsb->level ||
 944                     sb->layout != refsb->layout ||
 945                     sb->chunksize != refsb->chunksize) {
 946                         printk(KERN_WARNING "md: %s has strangely different"
 947                                 " superblock to %s\n",
 948                                 bdevname(rdev->bdev,b),
 949                                 bdevname(refdev->bdev,b2));
 950                         return -EINVAL;
 951                 }
 952                 ev1 = le64_to_cpu(sb->events);
 953                 ev2 = le64_to_cpu(refsb->events);
 954
 955                 if (ev1 > ev2)
 956                         return 1;
 957         }
 958         if (minor_version)
 959                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 960         else
 961                 rdev->size = rdev->sb_offset;
 962         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 963                 return -EINVAL;
 964         rdev->size = le64_to_cpu(sb->data_size)/2;
 965         if (le32_to_cpu(sb->chunksize))
 966                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 967         return 0;
 968 }
 969
 970 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 971 {
 972         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 973
 974         rdev->raid_disk = -1;
 975         rdev->in_sync = 0;
 976         if (mddev->raid_disks == 0) {
 977                 mddev->major_version = 1;
 978                 mddev->patch_version = 0;
 979                 mddev->persistent = 1;
 980                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 981                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 982                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 983                 mddev->level = le32_to_cpu(sb->level);
 984                 mddev->layout = le32_to_cpu(sb->layout);
 985                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 986                 mddev->size = le64_to_cpu(sb->size)/2;
 987                 mddev->events = le64_to_cpu(sb->events);
 988                 mddev->bitmap_offset = 0;
 989                 mddev->default_bitmap_offset = 0;
 990                 mddev->default_bitmap_offset = 1024;
 991
 992                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 993                 memcpy(mddev->uuid, sb->set_uuid, 16);
 994
 995                 mddev->max_disks =  (4096-256)/2;
 996
 997                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
 998                     mddev->bitmap_file == NULL ) {
 999                         if (mddev->level != 1) {
1000                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1001                                 return -EINVAL;
1002                         }
1003                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1004                 }
1005         } else if (mddev->pers == NULL) {
1006                 /* Insist of good event counter while assembling */
1007                 __u64 ev1 = le64_to_cpu(sb->events);
1008                 ++ev1;
1009                 if (ev1 < mddev->events)
1010                         return -EINVAL;
1011         } else if (mddev->bitmap) {
1012                 /* If adding to array with a bitmap, then we can accept an
1013                  * older device, but not too old.
1014                  */
1015                 __u64 ev1 = le64_to_cpu(sb->events);
1016                 if (ev1 < mddev->bitmap->events_cleared)
1017                         return 0;
1018         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1019                 return 0;
1020
1021         if (mddev->level != LEVEL_MULTIPATH) {
1022                 int role;
1023                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1024                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1025                 switch(role) {
1026                 case 0xffff: /* spare */
1027                         rdev->faulty = 0;
1028                         break;
1029                 case 0xfffe: /* faulty */
1030                         rdev->faulty = 1;
1031                         break;
1032                 default:
1033                         rdev->in_sync = 1;
1034                         rdev->faulty = 0;
1035                         rdev->raid_disk = role;
1036                         break;
1037                 }
1038                 rdev->flags = 0;
1039                 if (sb->devflags & WriteMostly1)
1040                         set_bit(WriteMostly, &rdev->flags);
1041         } else /* MULTIPATH are always insync */
1042                 rdev->in_sync = 1;
1043
1044         return 0;
1045 }
1046
1047 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1048 {
1049         struct mdp_superblock_1 *sb;
1050         struct list_head *tmp;
1051         mdk_rdev_t *rdev2;
1052         int max_dev, i;
1053         /* make rdev->sb match mddev and rdev data. */
1054
1055         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1056
1057         sb->feature_map = 0;
1058         sb->pad0 = 0;
1059         memset(sb->pad1, 0, sizeof(sb->pad1));
1060         memset(sb->pad2, 0, sizeof(sb->pad2));
1061         memset(sb->pad3, 0, sizeof(sb->pad3));
1062
1063         sb->utime = cpu_to_le64((__u64)mddev->utime);
1064         sb->events = cpu_to_le64(mddev->events);
1065         if (mddev->in_sync)
1066                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1067         else
1068                 sb->resync_offset = cpu_to_le64(0);
1069
1070         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1071                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1072                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1073         }
1074
1075         max_dev = 0;
1076         ITERATE_RDEV(mddev,rdev2,tmp)
1077                 if (rdev2->desc_nr+1 > max_dev)
1078                         max_dev = rdev2->desc_nr+1;
1079
1080         sb->max_dev = cpu_to_le32(max_dev);
1081         for (i=0; i<max_dev;i++)
1082                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1083
1084         ITERATE_RDEV(mddev,rdev2,tmp) {
1085                 i = rdev2->desc_nr;
1086                 if (rdev2->faulty)
1087                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1088                 else if (rdev2->in_sync)
1089                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1090                 else
1091                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1092         }
1093
1094         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1095         sb->sb_csum = calc_sb_1_csum(sb);
1096 }
1097
1098
1099 static struct super_type super_types[] = {
1100         [0] = {
1101                 .name   = "0.90.0",
1102                 .owner  = THIS_MODULE,
1103                 .load_super     = super_90_load,
1104                 .validate_super = super_90_validate,
1105                 .sync_super     = super_90_sync,
1106         },
1107         [1] = {
1108                 .name   = "md-1",
1109                 .owner  = THIS_MODULE,
1110                 .load_super     = super_1_load,
1111                 .validate_super = super_1_validate,
1112                 .sync_super     = super_1_sync,
1113         },
1114 };
1115
1116 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1117 {
1118         struct list_head *tmp;
1119         mdk_rdev_t *rdev;
1120
1121         ITERATE_RDEV(mddev,rdev,tmp)
1122                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1123                         return rdev;
1124
1125         return NULL;
1126 }
1127
1128 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1129 {
1130         struct list_head *tmp;
1131         mdk_rdev_t *rdev;
1132
1133         ITERATE_RDEV(mddev1,rdev,tmp)
1134                 if (match_dev_unit(mddev2, rdev))
1135                         return 1;
1136
1137         return 0;
1138 }
1139
1140 static LIST_HEAD(pending_raid_disks);
1141
1142 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1143 {
1144         mdk_rdev_t *same_pdev;
1145         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1146
1147         if (rdev->mddev) {
1148                 MD_BUG();
1149                 return -EINVAL;
1150         }
1151         same_pdev = match_dev_unit(mddev, rdev);
1152         if (same_pdev)
1153                 printk(KERN_WARNING
1154                         "%s: WARNING: %s appears to be on the same physical"
1155                         " disk as %s. True\n     protection against single-disk"
1156                         " failure might be compromised.\n",
1157                         mdname(mddev), bdevname(rdev->bdev,b),
1158                         bdevname(same_pdev->bdev,b2));
1159
1160         /* Verify rdev->desc_nr is unique.
1161          * If it is -1, assign a free number, else
1162          * check number is not in use
1163          */
1164         if (rdev->desc_nr < 0) {
1165                 int choice = 0;
1166                 if (mddev->pers) choice = mddev->raid_disks;
1167                 while (find_rdev_nr(mddev, choice))
1168                         choice++;
1169                 rdev->desc_nr = choice;
1170         } else {
1171                 if (find_rdev_nr(mddev, rdev->desc_nr))
1172                         return -EBUSY;
1173         }
1174
1175         list_add(&rdev->same_set, &mddev->disks);
1176         rdev->mddev = mddev;
1177         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1178
1179         rdev->kobj.k_name = NULL;
1180         snprintf(rdev->kobj.name, KOBJ_NAME_LEN, "dev%d", rdev->desc_nr);
1181         rdev->kobj.parent = kobject_get(&mddev->kobj);
1182         kobject_add(&rdev->kobj);
1183
1184         sysfs_create_link(&rdev->kobj, &rdev->bdev->bd_disk->kobj, "block");
1185         return 0;
1186 }
1187
1188 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1189 {
1190         char b[BDEVNAME_SIZE];
1191         if (!rdev->mddev) {
1192                 MD_BUG();
1193                 return;
1194         }
1195         list_del_init(&rdev->same_set);
1196         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1197         rdev->mddev = NULL;
1198         sysfs_remove_link(&rdev->kobj, "block");
1199         kobject_del(&rdev->kobj);
1200 }
1201
1202 /*
1203  * prevent the device from being mounted, repartitioned or
1204  * otherwise reused by a RAID array (or any other kernel
1205  * subsystem), by bd_claiming the device.
1206  */
1207 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1208 {
1209         int err = 0;
1210         struct block_device *bdev;
1211         char b[BDEVNAME_SIZE];
1212
1213         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1214         if (IS_ERR(bdev)) {
1215                 printk(KERN_ERR "md: could not open %s.\n",
1216                         __bdevname(dev, b));
1217                 return PTR_ERR(bdev);
1218         }
1219         err = bd_claim(bdev, rdev);
1220         if (err) {
1221                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1222                         bdevname(bdev, b));
1223                 blkdev_put(bdev);
1224                 return err;
1225         }
1226         rdev->bdev = bdev;
1227         return err;
1228 }
1229
1230 static void unlock_rdev(mdk_rdev_t *rdev)
1231 {
1232         struct block_device *bdev = rdev->bdev;
1233         rdev->bdev = NULL;
1234         if (!bdev)
1235                 MD_BUG();
1236         bd_release(bdev);
1237         blkdev_put(bdev);
1238 }
1239
1240 void md_autodetect_dev(dev_t dev);
1241
1242 static void export_rdev(mdk_rdev_t * rdev)
1243 {
1244         char b[BDEVNAME_SIZE];
1245         printk(KERN_INFO "md: export_rdev(%s)\n",
1246                 bdevname(rdev->bdev,b));
1247         if (rdev->mddev)
1248                 MD_BUG();
1249         free_disk_sb(rdev);
1250         list_del_init(&rdev->same_set);
1251 #ifndef MODULE
1252         md_autodetect_dev(rdev->bdev->bd_dev);
1253 #endif
1254         unlock_rdev(rdev);
1255         kobject_put(&rdev->kobj);
1256 }
1257
1258 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1259 {
1260         unbind_rdev_from_array(rdev);
1261         export_rdev(rdev);
1262 }
1263
1264 static void export_array(mddev_t *mddev)
1265 {
1266         struct list_head *tmp;
1267         mdk_rdev_t *rdev;
1268
1269         ITERATE_RDEV(mddev,rdev,tmp) {
1270                 if (!rdev->mddev) {
1271                         MD_BUG();
1272                         continue;
1273                 }
1274                 kick_rdev_from_array(rdev);
1275         }
1276         if (!list_empty(&mddev->disks))
1277                 MD_BUG();
1278         mddev->raid_disks = 0;
1279         mddev->major_version = 0;
1280 }
1281
1282 static void print_desc(mdp_disk_t *desc)
1283 {
1284         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1285                 desc->major,desc->minor,desc->raid_disk,desc->state);
1286 }
1287
1288 static void print_sb(mdp_super_t *sb)
1289 {
1290         int i;
1291
1292         printk(KERN_INFO
1293                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1294                 sb->major_version, sb->minor_version, sb->patch_version,
1295                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1296                 sb->ctime);
1297         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1298                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1299                 sb->md_minor, sb->layout, sb->chunk_size);
1300         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1301                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1302                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1303                 sb->failed_disks, sb->spare_disks,
1304                 sb->sb_csum, (unsigned long)sb->events_lo);
1305
1306         printk(KERN_INFO);
1307         for (i = 0; i < MD_SB_DISKS; i++) {
1308                 mdp_disk_t *desc;
1309
1310                 desc = sb->disks + i;
1311                 if (desc->number || desc->major || desc->minor ||
1312                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1313                         printk("     D %2d: ", i);
1314                         print_desc(desc);
1315                 }
1316         }
1317         printk(KERN_INFO "md:     THIS: ");
1318         print_desc(&sb->this_disk);
1319
1320 }
1321
1322 static void print_rdev(mdk_rdev_t *rdev)
1323 {
1324         char b[BDEVNAME_SIZE];
1325         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1326                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1327                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1328         if (rdev->sb_loaded) {
1329                 printk(KERN_INFO "md: rdev superblock:\n");
1330                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1331         } else
1332                 printk(KERN_INFO "md: no rdev superblock!\n");
1333 }
1334
1335 void md_print_devices(void)
1336 {
1337         struct list_head *tmp, *tmp2;
1338         mdk_rdev_t *rdev;
1339         mddev_t *mddev;
1340         char b[BDEVNAME_SIZE];
1341
1342         printk("\n");
1343         printk("md:     **********************************\n");
1344         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1345         printk("md:     **********************************\n");
1346         ITERATE_MDDEV(mddev,tmp) {
1347
1348                 if (mddev->bitmap)
1349                         bitmap_print_sb(mddev->bitmap);
1350                 else
1351                         printk("%s: ", mdname(mddev));
1352                 ITERATE_RDEV(mddev,rdev,tmp2)
1353                         printk("<%s>", bdevname(rdev->bdev,b));
1354                 printk("\n");
1355
1356                 ITERATE_RDEV(mddev,rdev,tmp2)
1357                         print_rdev(rdev);
1358         }
1359         printk("md:     **********************************\n");
1360         printk("\n");
1361 }
1362
1363
1364 static void sync_sbs(mddev_t * mddev)
1365 {
1366         mdk_rdev_t *rdev;
1367         struct list_head *tmp;
1368
1369         ITERATE_RDEV(mddev,rdev,tmp) {
1370                 super_types[mddev->major_version].
1371                         sync_super(mddev, rdev);
1372                 rdev->sb_loaded = 1;
1373         }
1374 }
1375
1376 static void md_update_sb(mddev_t * mddev)
1377 {
1378         int err;
1379         struct list_head *tmp;
1380         mdk_rdev_t *rdev;
1381         int sync_req;
1382
1383 repeat:
1384         spin_lock(&mddev->write_lock);
1385         sync_req = mddev->in_sync;
1386         mddev->utime = get_seconds();
1387         mddev->events ++;
1388
1389         if (!mddev->events) {
1390                 /*
1391                  * oops, this 64-bit counter should never wrap.
1392                  * Either we are in around ~1 trillion A.C., assuming
1393                  * 1 reboot per second, or we have a bug:
1394                  */
1395                 MD_BUG();
1396                 mddev->events --;
1397         }
1398         mddev->sb_dirty = 2;
1399         sync_sbs(mddev);
1400
1401         /*
1402          * do not write anything to disk if using
1403          * nonpersistent superblocks
1404          */
1405         if (!mddev->persistent) {
1406                 mddev->sb_dirty = 0;
1407                 spin_unlock(&mddev->write_lock);
1408                 wake_up(&mddev->sb_wait);
1409                 return;
1410         }
1411         spin_unlock(&mddev->write_lock);
1412
1413         dprintk(KERN_INFO
1414                 "md: updating %s RAID superblock on device (in sync %d)\n",
1415                 mdname(mddev),mddev->in_sync);
1416
1417         err = bitmap_update_sb(mddev->bitmap);
1418         ITERATE_RDEV(mddev,rdev,tmp) {
1419                 char b[BDEVNAME_SIZE];
1420                 dprintk(KERN_INFO "md: ");
1421                 if (rdev->faulty)
1422                         dprintk("(skipping faulty ");
1423
1424                 dprintk("%s ", bdevname(rdev->bdev,b));
1425                 if (!rdev->faulty) {
1426                         md_super_write(mddev,rdev,
1427                                        rdev->sb_offset<<1, rdev->sb_size,
1428                                        rdev->sb_page);
1429                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1430                                 bdevname(rdev->bdev,b),
1431                                 (unsigned long long)rdev->sb_offset);
1432
1433                 } else
1434                         dprintk(")\n");
1435                 if (mddev->level == LEVEL_MULTIPATH)
1436                         /* only need to write one superblock... */
1437                         break;
1438         }
1439         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1440         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1441
1442         spin_lock(&mddev->write_lock);
1443         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1444                 /* have to write it out again */
1445                 spin_unlock(&mddev->write_lock);
1446                 goto repeat;
1447         }
1448         mddev->sb_dirty = 0;
1449         spin_unlock(&mddev->write_lock);
1450         wake_up(&mddev->sb_wait);
1451
1452 }
1453
1454 struct rdev_sysfs_entry {
1455         struct attribute attr;
1456         ssize_t (*show)(mdk_rdev_t *, char *);
1457         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1458 };
1459
1460 static ssize_t
1461 rdev_show_state(mdk_rdev_t *rdev, char *page)
1462 {
1463         char *sep = "";
1464         int len=0;
1465
1466         if (rdev->faulty) {
1467                 len+= sprintf(page+len, "%sfaulty",sep);
1468                 sep = ",";
1469         }
1470         if (rdev->in_sync) {
1471                 len += sprintf(page+len, "%sin_sync",sep);
1472                 sep = ",";
1473         }
1474         if (!rdev->faulty && !rdev->in_sync) {
1475                 len += sprintf(page+len, "%sspare", sep);
1476                 sep = ",";
1477         }
1478         return len+sprintf(page+len, "\n");
1479 }
1480
1481 static struct rdev_sysfs_entry rdev_state = {
1482         .attr = {.name = "state", .mode = S_IRUGO },
1483         .show = rdev_show_state,
1484 };
1485
1486 static ssize_t
1487 rdev_show_super(mdk_rdev_t *rdev, char *page)
1488 {
1489         if (rdev->sb_loaded && rdev->sb_size) {
1490                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1491                 return rdev->sb_size;
1492         } else
1493                 return 0;
1494 }
1495 static struct rdev_sysfs_entry rdev_super = {
1496         .attr = {.name = "super", .mode = S_IRUGO },
1497         .show = rdev_show_super,
1498 };
1499 static struct attribute *rdev_default_attrs[] = {
1500         &rdev_state.attr,
1501         &rdev_super.attr,
1502         NULL,
1503 };
1504 static ssize_t
1505 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1506 {
1507         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1508         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1509
1510         if (!entry->show)
1511                 return -EIO;
1512         return entry->show(rdev, page);
1513 }
1514
1515 static ssize_t
1516 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1517               const char *page, size_t length)
1518 {
1519         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1520         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1521
1522         if (!entry->store)
1523                 return -EIO;
1524         return entry->store(rdev, page, length);
1525 }
1526
1527 static void rdev_free(struct kobject *ko)
1528 {
1529         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1530         kfree(rdev);
1531 }
1532 static struct sysfs_ops rdev_sysfs_ops = {
1533         .show           = rdev_attr_show,
1534         .store          = rdev_attr_store,
1535 };
1536 static struct kobj_type rdev_ktype = {
1537         .release        = rdev_free,
1538         .sysfs_ops      = &rdev_sysfs_ops,
1539         .default_attrs  = rdev_default_attrs,
1540 };
1541
1542 /*
1543  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1544  *
1545  * mark the device faulty if:
1546  *
1547  *   - the device is nonexistent (zero size)
1548  *   - the device has no valid superblock
1549  *
1550  * a faulty rdev _never_ has rdev->sb set.
1551  */
1552 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1553 {
1554         char b[BDEVNAME_SIZE];
1555         int err;
1556         mdk_rdev_t *rdev;
1557         sector_t size;
1558
1559         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1560         if (!rdev) {
1561                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1562                 return ERR_PTR(-ENOMEM);
1563         }
1564         memset(rdev, 0, sizeof(*rdev));
1565
1566         if ((err = alloc_disk_sb(rdev)))
1567                 goto abort_free;
1568
1569         err = lock_rdev(rdev, newdev);
1570         if (err)
1571                 goto abort_free;
1572
1573         rdev->kobj.parent = NULL;
1574         rdev->kobj.ktype = &rdev_ktype;
1575         kobject_init(&rdev->kobj);
1576
1577         rdev->desc_nr = -1;
1578         rdev->faulty = 0;
1579         rdev->in_sync = 0;
1580         rdev->data_offset = 0;
1581         atomic_set(&rdev->nr_pending, 0);
1582
1583         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1584         if (!size) {
1585                 printk(KERN_WARNING
1586                         "md: %s has zero or unknown size, marking faulty!\n",
1587                         bdevname(rdev->bdev,b));
1588                 err = -EINVAL;
1589                 goto abort_free;
1590         }
1591
1592         if (super_format >= 0) {
1593                 err = super_types[super_format].
1594                         load_super(rdev, NULL, super_minor);
1595                 if (err == -EINVAL) {
1596                         printk(KERN_WARNING
1597                                 "md: %s has invalid sb, not importing!\n",
1598                                 bdevname(rdev->bdev,b));
1599                         goto abort_free;
1600                 }
1601                 if (err < 0) {
1602                         printk(KERN_WARNING
1603                                 "md: could not read %s's sb, not importing!\n",
1604                                 bdevname(rdev->bdev,b));
1605                         goto abort_free;
1606                 }
1607         }
1608         INIT_LIST_HEAD(&rdev->same_set);
1609
1610         return rdev;
1611
1612 abort_free:
1613         if (rdev->sb_page) {
1614                 if (rdev->bdev)
1615                         unlock_rdev(rdev);
1616                 free_disk_sb(rdev);
1617         }
1618         kfree(rdev);
1619         return ERR_PTR(err);
1620 }
1621
1622 /*
1623  * Check a full RAID array for plausibility
1624  */
1625
1626
1627 static void analyze_sbs(mddev_t * mddev)
1628 {
1629         int i;
1630         struct list_head *tmp;
1631         mdk_rdev_t *rdev, *freshest;
1632         char b[BDEVNAME_SIZE];
1633
1634         freshest = NULL;
1635         ITERATE_RDEV(mddev,rdev,tmp)
1636                 switch (super_types[mddev->major_version].
1637                         load_super(rdev, freshest, mddev->minor_version)) {
1638                 case 1:
1639                         freshest = rdev;
1640                         break;
1641                 case 0:
1642                         break;
1643                 default:
1644                         printk( KERN_ERR \
1645                                 "md: fatal superblock inconsistency in %s"
1646                                 " -- removing from array\n",
1647                                 bdevname(rdev->bdev,b));
1648                         kick_rdev_from_array(rdev);
1649                 }
1650
1651
1652         super_types[mddev->major_version].
1653                 validate_super(mddev, freshest);
1654
1655         i = 0;
1656         ITERATE_RDEV(mddev,rdev,tmp) {
1657                 if (rdev != freshest)
1658                         if (super_types[mddev->major_version].
1659                             validate_super(mddev, rdev)) {
1660                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1661                                         " from array!\n",
1662                                         bdevname(rdev->bdev,b));
1663                                 kick_rdev_from_array(rdev);
1664                                 continue;
1665                         }
1666                 if (mddev->level == LEVEL_MULTIPATH) {
1667                         rdev->desc_nr = i++;
1668                         rdev->raid_disk = rdev->desc_nr;
1669                         rdev->in_sync = 1;
1670                 }
1671         }
1672
1673
1674
1675         if (mddev->recovery_cp != MaxSector &&
1676             mddev->level >= 1)
1677                 printk(KERN_ERR "md: %s: raid array is not clean"
1678                        " -- starting background reconstruction\n",
1679                        mdname(mddev));
1680
1681 }
1682
1683 struct md_sysfs_entry {
1684         struct attribute attr;
1685         ssize_t (*show)(mddev_t *, char *);
1686         ssize_t (*store)(mddev_t *, const char *, size_t);
1687 };
1688
1689 static ssize_t
1690 md_show_level(mddev_t *mddev, char *page)
1691 {
1692         mdk_personality_t *p = mddev->pers;
1693         if (p == NULL)
1694                 return 0;
1695         if (mddev->level >= 0)
1696                 return sprintf(page, "RAID-%d\n", mddev->level);
1697         else
1698                 return sprintf(page, "%s\n", p->name);
1699 }
1700
1701 static struct md_sysfs_entry md_level = {
1702         .attr = {.name = "level", .mode = S_IRUGO },
1703         .show = md_show_level,
1704 };
1705
1706 static ssize_t
1707 md_show_rdisks(mddev_t *mddev, char *page)
1708 {
1709         return sprintf(page, "%d\n", mddev->raid_disks);
1710 }
1711
1712 static struct md_sysfs_entry md_raid_disks = {
1713         .attr = {.name = "raid_disks", .mode = S_IRUGO },
1714         .show = md_show_rdisks,
1715 };
1716
1717 static struct attribute *md_default_attrs[] = {
1718         &md_level.attr,
1719         &md_raid_disks.attr,
1720         NULL,
1721 };
1722
1723 static ssize_t
1724 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1725 {
1726         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1727         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1728
1729         if (!entry->show)
1730                 return -EIO;
1731         return entry->show(mddev, page);
1732 }
1733
1734 static ssize_t
1735 md_attr_store(struct kobject *kobj, struct attribute *attr,
1736               const char *page, size_t length)
1737 {
1738         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1739         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1740
1741         if (!entry->store)
1742                 return -EIO;
1743         return entry->store(mddev, page, length);
1744 }
1745
1746 static void md_free(struct kobject *ko)
1747 {
1748         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1749         kfree(mddev);
1750 }
1751
1752 static struct sysfs_ops md_sysfs_ops = {
1753         .show   = md_attr_show,
1754         .store  = md_attr_store,
1755 };
1756 static struct kobj_type md_ktype = {
1757         .release        = md_free,
1758         .sysfs_ops      = &md_sysfs_ops,
1759         .default_attrs  = md_default_attrs,
1760 };
1761
1762 int mdp_major = 0;
1763
1764 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1765 {
1766         static DECLARE_MUTEX(disks_sem);
1767         mddev_t *mddev = mddev_find(dev);
1768         struct gendisk *disk;
1769         int partitioned = (MAJOR(dev) != MD_MAJOR);
1770         int shift = partitioned ? MdpMinorShift : 0;
1771         int unit = MINOR(dev) >> shift;
1772
1773         if (!mddev)
1774                 return NULL;
1775
1776         down(&disks_sem);
1777         if (mddev->gendisk) {
1778                 up(&disks_sem);
1779                 mddev_put(mddev);
1780                 return NULL;
1781         }
1782         disk = alloc_disk(1 << shift);
1783         if (!disk) {
1784                 up(&disks_sem);
1785                 mddev_put(mddev);
1786                 return NULL;
1787         }
1788         disk->major = MAJOR(dev);
1789         disk->first_minor = unit << shift;
1790         if (partitioned) {
1791                 sprintf(disk->disk_name, "md_d%d", unit);
1792                 sprintf(disk->devfs_name, "md/d%d", unit);
1793         } else {
1794                 sprintf(disk->disk_name, "md%d", unit);
1795                 sprintf(disk->devfs_name, "md/%d", unit);
1796         }
1797         disk->fops = &md_fops;
1798         disk->private_data = mddev;
1799         disk->queue = mddev->queue;
1800         add_disk(disk);
1801         mddev->gendisk = disk;
1802         up(&disks_sem);
1803         mddev->kobj.parent = kobject_get(&disk->kobj);
1804         mddev->kobj.k_name = NULL;
1805         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1806         mddev->kobj.ktype = &md_ktype;
1807         kobject_register(&mddev->kobj);
1808         return NULL;
1809 }
1810
1811 void md_wakeup_thread(mdk_thread_t *thread);
1812
1813 static void md_safemode_timeout(unsigned long data)
1814 {
1815         mddev_t *mddev = (mddev_t *) data;
1816
1817         mddev->safemode = 1;
1818         md_wakeup_thread(mddev->thread);
1819 }
1820
1821
1822 static int do_md_run(mddev_t * mddev)
1823 {
1824         int pnum, err;
1825         int chunk_size;
1826         struct list_head *tmp;
1827         mdk_rdev_t *rdev;
1828         struct gendisk *disk;
1829         char b[BDEVNAME_SIZE];
1830
1831         if (list_empty(&mddev->disks))
1832                 /* cannot run an array with no devices.. */
1833                 return -EINVAL;
1834
1835         if (mddev->pers)
1836                 return -EBUSY;
1837
1838         /*
1839          * Analyze all RAID superblock(s)
1840          */
1841         if (!mddev->raid_disks)
1842                 analyze_sbs(mddev);
1843
1844         chunk_size = mddev->chunk_size;
1845         pnum = level_to_pers(mddev->level);
1846
1847         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1848                 if (!chunk_size) {
1849                         /*
1850                          * 'default chunksize' in the old md code used to
1851                          * be PAGE_SIZE, baaad.
1852                          * we abort here to be on the safe side. We don't
1853                          * want to continue the bad practice.
1854                          */
1855                         printk(KERN_ERR
1856                                 "no chunksize specified, see 'man raidtab'\n");
1857                         return -EINVAL;
1858                 }
1859                 if (chunk_size > MAX_CHUNK_SIZE) {
1860                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1861                                 chunk_size, MAX_CHUNK_SIZE);
1862                         return -EINVAL;
1863                 }
1864                 /*
1865                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1866                  */
1867                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1868                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1869                         return -EINVAL;
1870                 }
1871                 if (chunk_size < PAGE_SIZE) {
1872                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1873                                 chunk_size, PAGE_SIZE);
1874                         return -EINVAL;
1875                 }
1876
1877                 /* devices must have minimum size of one chunk */
1878                 ITERATE_RDEV(mddev,rdev,tmp) {
1879                         if (rdev->faulty)
1880                                 continue;
1881                         if (rdev->size < chunk_size / 1024) {
1882                                 printk(KERN_WARNING
1883                                         "md: Dev %s smaller than chunk_size:"
1884                                         " %lluk < %dk\n",
1885                                         bdevname(rdev->bdev,b),
1886                                         (unsigned long long)rdev->size,
1887                                         chunk_size / 1024);
1888                                 return -EINVAL;
1889                         }
1890                 }
1891         }
1892
1893 #ifdef CONFIG_KMOD
1894         if (!pers[pnum])
1895         {
1896                 request_module("md-personality-%d", pnum);
1897         }
1898 #endif
1899
1900         /*
1901          * Drop all container device buffers, from now on
1902          * the only valid external interface is through the md
1903          * device.
1904          * Also find largest hardsector size
1905          */
1906         ITERATE_RDEV(mddev,rdev,tmp) {
1907                 if (rdev->faulty)
1908                         continue;
1909                 sync_blockdev(rdev->bdev);
1910                 invalidate_bdev(rdev->bdev, 0);
1911         }
1912
1913         md_probe(mddev->unit, NULL, NULL);
1914         disk = mddev->gendisk;
1915         if (!disk)
1916                 return -ENOMEM;
1917
1918         spin_lock(&pers_lock);
1919         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1920                 spin_unlock(&pers_lock);
1921                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1922                        pnum);
1923                 return -EINVAL;
1924         }
1925
1926         mddev->pers = pers[pnum];
1927         spin_unlock(&pers_lock);
1928
1929         mddev->recovery = 0;
1930         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1931
1932         /* before we start the array running, initialise the bitmap */
1933         err = bitmap_create(mddev);
1934         if (err)
1935                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1936                         mdname(mddev), err);
1937         else
1938                 err = mddev->pers->run(mddev);
1939         if (err) {
1940                 printk(KERN_ERR "md: pers->run() failed ...\n");
1941                 module_put(mddev->pers->owner);
1942                 mddev->pers = NULL;
1943                 bitmap_destroy(mddev);
1944                 return err;
1945         }
1946         atomic_set(&mddev->writes_pending,0);
1947         mddev->safemode = 0;
1948         mddev->safemode_timer.function = md_safemode_timeout;
1949         mddev->safemode_timer.data = (unsigned long) mddev;
1950         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1951         mddev->in_sync = 1;
1952
1953         ITERATE_RDEV(mddev,rdev,tmp)
1954                 if (rdev->raid_disk >= 0) {
1955                         char nm[20];
1956                         sprintf(nm, "rd%d", rdev->raid_disk);
1957                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
1958                 }
1959
1960         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1961         md_wakeup_thread(mddev->thread);
1962
1963         if (mddev->sb_dirty)
1964                 md_update_sb(mddev);
1965
1966         set_capacity(disk, mddev->array_size<<1);
1967
1968         /* If we call blk_queue_make_request here, it will
1969          * re-initialise max_sectors etc which may have been
1970          * refined inside -> run.  So just set the bits we need to set.
1971          * Most initialisation happended when we called
1972          * blk_queue_make_request(..., md_fail_request)
1973          * earlier.
1974          */
1975         mddev->queue->queuedata = mddev;
1976         mddev->queue->make_request_fn = mddev->pers->make_request;
1977
1978         mddev->changed = 1;
1979         return 0;
1980 }
1981
1982 static int restart_array(mddev_t *mddev)
1983 {
1984         struct gendisk *disk = mddev->gendisk;
1985         int err;
1986
1987         /*
1988          * Complain if it has no devices
1989          */
1990         err = -ENXIO;
1991         if (list_empty(&mddev->disks))
1992                 goto out;
1993
1994         if (mddev->pers) {
1995                 err = -EBUSY;
1996                 if (!mddev->ro)
1997                         goto out;
1998
1999                 mddev->safemode = 0;
2000                 mddev->ro = 0;
2001                 set_disk_ro(disk, 0);
2002
2003                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2004                         mdname(mddev));
2005                 /*
2006                  * Kick recovery or resync if necessary
2007                  */
2008                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2009                 md_wakeup_thread(mddev->thread);
2010                 err = 0;
2011         } else {
2012                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2013                         mdname(mddev));
2014                 err = -EINVAL;
2015         }
2016
2017 out:
2018         return err;
2019 }
2020
2021 static int do_md_stop(mddev_t * mddev, int ro)
2022 {
2023         int err = 0;
2024         struct gendisk *disk = mddev->gendisk;
2025
2026         if (mddev->pers) {
2027                 if (atomic_read(&mddev->active)>2) {
2028                         printk("md: %s still in use.\n",mdname(mddev));
2029                         return -EBUSY;
2030                 }
2031
2032                 if (mddev->sync_thread) {
2033                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2034                         md_unregister_thread(mddev->sync_thread);
2035                         mddev->sync_thread = NULL;
2036                 }
2037
2038                 del_timer_sync(&mddev->safemode_timer);
2039
2040                 invalidate_partition(disk, 0);
2041
2042                 if (ro) {
2043                         err  = -ENXIO;
2044                         if (mddev->ro)
2045                                 goto out;
2046                         mddev->ro = 1;
2047                 } else {
2048                         bitmap_flush(mddev);
2049                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
2050                         if (mddev->ro)
2051                                 set_disk_ro(disk, 0);
2052                         blk_queue_make_request(mddev->queue, md_fail_request);
2053                         mddev->pers->stop(mddev);
2054                         module_put(mddev->pers->owner);
2055                         mddev->pers = NULL;
2056                         if (mddev->ro)
2057                                 mddev->ro = 0;
2058                 }
2059                 if (!mddev->in_sync) {
2060                         /* mark array as shutdown cleanly */
2061                         mddev->in_sync = 1;
2062                         md_update_sb(mddev);
2063                 }
2064                 if (ro)
2065                         set_disk_ro(disk, 1);
2066         }
2067
2068         bitmap_destroy(mddev);
2069         if (mddev->bitmap_file) {
2070                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2071                 fput(mddev->bitmap_file);
2072                 mddev->bitmap_file = NULL;
2073         }
2074         mddev->bitmap_offset = 0;
2075
2076         /*
2077          * Free resources if final stop
2078          */
2079         if (!ro) {
2080                 mdk_rdev_t *rdev;
2081                 struct list_head *tmp;
2082                 struct gendisk *disk;
2083                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2084
2085                 ITERATE_RDEV(mddev,rdev,tmp)
2086                         if (rdev->raid_disk >= 0) {
2087                                 char nm[20];
2088                                 sprintf(nm, "rd%d", rdev->raid_disk);
2089                                 sysfs_remove_link(&mddev->kobj, nm);
2090                         }
2091
2092                 export_array(mddev);
2093
2094                 mddev->array_size = 0;
2095                 disk = mddev->gendisk;
2096                 if (disk)
2097                         set_capacity(disk, 0);
2098                 mddev->changed = 1;
2099         } else
2100                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2101                         mdname(mddev));
2102         err = 0;
2103 out:
2104         return err;
2105 }
2106
2107 static void autorun_array(mddev_t *mddev)
2108 {
2109         mdk_rdev_t *rdev;
2110         struct list_head *tmp;
2111         int err;
2112
2113         if (list_empty(&mddev->disks))
2114                 return;
2115
2116         printk(KERN_INFO "md: running: ");
2117
2118         ITERATE_RDEV(mddev,rdev,tmp) {
2119                 char b[BDEVNAME_SIZE];
2120                 printk("<%s>", bdevname(rdev->bdev,b));
2121         }
2122         printk("\n");
2123
2124         err = do_md_run (mddev);
2125         if (err) {
2126                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2127                 do_md_stop (mddev, 0);
2128         }
2129 }
2130
2131 /*
2132  * lets try to run arrays based on all disks that have arrived
2133  * until now. (those are in pending_raid_disks)
2134  *
2135  * the method: pick the first pending disk, collect all disks with
2136  * the same UUID, remove all from the pending list and put them into
2137  * the 'same_array' list. Then order this list based on superblock
2138  * update time (freshest comes first), kick out 'old' disks and
2139  * compare superblocks. If everything's fine then run it.
2140  *
2141  * If "unit" is allocated, then bump its reference count
2142  */
2143 static void autorun_devices(int part)
2144 {
2145         struct list_head candidates;
2146         struct list_head *tmp;
2147         mdk_rdev_t *rdev0, *rdev;
2148         mddev_t *mddev;
2149         char b[BDEVNAME_SIZE];
2150
2151         printk(KERN_INFO "md: autorun ...\n");
2152         while (!list_empty(&pending_raid_disks)) {
2153                 dev_t dev;
2154                 rdev0 = list_entry(pending_raid_disks.next,
2155                                          mdk_rdev_t, same_set);
2156
2157                 printk(KERN_INFO "md: considering %s ...\n",
2158                         bdevname(rdev0->bdev,b));
2159                 INIT_LIST_HEAD(&candidates);
2160                 ITERATE_RDEV_PENDING(rdev,tmp)
2161                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2162                                 printk(KERN_INFO "md:  adding %s ...\n",
2163                                         bdevname(rdev->bdev,b));
2164                                 list_move(&rdev->same_set, &candidates);
2165                         }
2166                 /*
2167                  * now we have a set of devices, with all of them having
2168                  * mostly sane superblocks. It's time to allocate the
2169                  * mddev.
2170                  */
2171                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2172                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2173                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2174                         break;
2175                 }
2176                 if (part)
2177                         dev = MKDEV(mdp_major,
2178                                     rdev0->preferred_minor << MdpMinorShift);
2179                 else
2180                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2181
2182                 md_probe(dev, NULL, NULL);
2183                 mddev = mddev_find(dev);
2184                 if (!mddev) {
2185                         printk(KERN_ERR
2186                                 "md: cannot allocate memory for md drive.\n");
2187                         break;
2188                 }
2189                 if (mddev_lock(mddev))
2190                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2191                                mdname(mddev));
2192                 else if (mddev->raid_disks || mddev->major_version
2193                          || !list_empty(&mddev->disks)) {
2194                         printk(KERN_WARNING
2195                                 "md: %s already running, cannot run %s\n",
2196                                 mdname(mddev), bdevname(rdev0->bdev,b));
2197                         mddev_unlock(mddev);
2198                 } else {
2199                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2200                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2201                                 list_del_init(&rdev->same_set);
2202                                 if (bind_rdev_to_array(rdev, mddev))
2203                                         export_rdev(rdev);
2204                         }
2205                         autorun_array(mddev);
2206                         mddev_unlock(mddev);
2207                 }
2208                 /* on success, candidates will be empty, on error
2209                  * it won't...
2210                  */
2211                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2212                         export_rdev(rdev);
2213                 mddev_put(mddev);
2214         }
2215         printk(KERN_INFO "md: ... autorun DONE.\n");
2216 }
2217
2218 /*
2219  * import RAID devices based on one partition
2220  * if possible, the array gets run as well.
2221  */
2222
2223 static int autostart_array(dev_t startdev)
2224 {
2225         char b[BDEVNAME_SIZE];
2226         int err = -EINVAL, i;
2227         mdp_super_t *sb = NULL;
2228         mdk_rdev_t *start_rdev = NULL, *rdev;
2229
2230         start_rdev = md_import_device(startdev, 0, 0);
2231         if (IS_ERR(start_rdev))
2232                 return err;
2233
2234
2235         /* NOTE: this can only work for 0.90.0 superblocks */
2236         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2237         if (sb->major_version != 0 ||
2238             sb->minor_version != 90 ) {
2239                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2240                 export_rdev(start_rdev);
2241                 return err;
2242         }
2243
2244         if (start_rdev->faulty) {
2245                 printk(KERN_WARNING
2246                         "md: can not autostart based on faulty %s!\n",
2247                         bdevname(start_rdev->bdev,b));
2248                 export_rdev(start_rdev);
2249                 return err;
2250         }
2251         list_add(&start_rdev->same_set, &pending_raid_disks);
2252
2253         for (i = 0; i < MD_SB_DISKS; i++) {
2254                 mdp_disk_t *desc = sb->disks + i;
2255                 dev_t dev = MKDEV(desc->major, desc->minor);
2256
2257                 if (!dev)
2258                         continue;
2259                 if (dev == startdev)
2260                         continue;
2261                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2262                         continue;
2263                 rdev = md_import_device(dev, 0, 0);
2264                 if (IS_ERR(rdev))
2265                         continue;
2266
2267                 list_add(&rdev->same_set, &pending_raid_disks);
2268         }
2269
2270         /*
2271          * possibly return codes
2272          */
2273         autorun_devices(0);
2274         return 0;
2275
2276 }
2277
2278
2279 static int get_version(void __user * arg)
2280 {
2281         mdu_version_t ver;
2282
2283         ver.major = MD_MAJOR_VERSION;
2284         ver.minor = MD_MINOR_VERSION;
2285         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2286
2287         if (copy_to_user(arg, &ver, sizeof(ver)))
2288                 return -EFAULT;
2289
2290         return 0;
2291 }
2292
2293 static int get_array_info(mddev_t * mddev, void __user * arg)
2294 {
2295         mdu_array_info_t info;
2296         int nr,working,active,failed,spare;
2297         mdk_rdev_t *rdev;
2298         struct list_head *tmp;
2299
2300         nr=working=active=failed=spare=0;
2301         ITERATE_RDEV(mddev,rdev,tmp) {
2302                 nr++;
2303                 if (rdev->faulty)
2304                         failed++;
2305                 else {
2306                         working++;
2307                         if (rdev->in_sync)
2308                                 active++;
2309                         else
2310                                 spare++;
2311                 }
2312         }
2313
2314         info.major_version = mddev->major_version;
2315         info.minor_version = mddev->minor_version;
2316         info.patch_version = MD_PATCHLEVEL_VERSION;
2317         info.ctime         = mddev->ctime;
2318         info.level         = mddev->level;
2319         info.size          = mddev->size;
2320         info.nr_disks      = nr;
2321         info.raid_disks    = mddev->raid_disks;
2322         info.md_minor      = mddev->md_minor;
2323         info.not_persistent= !mddev->persistent;
2324
2325         info.utime         = mddev->utime;
2326         info.state         = 0;
2327         if (mddev->in_sync)
2328                 info.state = (1<<MD_SB_CLEAN);
2329         if (mddev->bitmap && mddev->bitmap_offset)
2330                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2331         info.active_disks  = active;
2332         info.working_disks = working;
2333         info.failed_disks  = failed;
2334         info.spare_disks   = spare;
2335
2336         info.layout        = mddev->layout;
2337         info.chunk_size    = mddev->chunk_size;
2338
2339         if (copy_to_user(arg, &info, sizeof(info)))
2340                 return -EFAULT;
2341
2342         return 0;
2343 }
2344
2345 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2346 {
2347         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2348         char *ptr, *buf = NULL;
2349         int err = -ENOMEM;
2350
2351         file = kmalloc(sizeof(*file), GFP_KERNEL);
2352         if (!file)
2353                 goto out;
2354
2355         /* bitmap disabled, zero the first byte and copy out */
2356         if (!mddev->bitmap || !mddev->bitmap->file) {
2357                 file->pathname[0] = '\0';
2358                 goto copy_out;
2359         }
2360
2361         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2362         if (!buf)
2363                 goto out;
2364
2365         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2366         if (!ptr)
2367                 goto out;
2368
2369         strcpy(file->pathname, ptr);
2370
2371 copy_out:
2372         err = 0;
2373         if (copy_to_user(arg, file, sizeof(*file)))
2374                 err = -EFAULT;
2375 out:
2376         kfree(buf);
2377         kfree(file);
2378         return err;
2379 }
2380
2381 static int get_disk_info(mddev_t * mddev, void __user * arg)
2382 {
2383         mdu_disk_info_t info;
2384         unsigned int nr;
2385         mdk_rdev_t *rdev;
2386
2387         if (copy_from_user(&info, arg, sizeof(info)))
2388                 return -EFAULT;
2389
2390         nr = info.number;
2391
2392         rdev = find_rdev_nr(mddev, nr);
2393         if (rdev) {
2394                 info.major = MAJOR(rdev->bdev->bd_dev);
2395                 info.minor = MINOR(rdev->bdev->bd_dev);
2396                 info.raid_disk = rdev->raid_disk;
2397                 info.state = 0;
2398                 if (rdev->faulty)
2399                         info.state |= (1<<MD_DISK_FAULTY);
2400                 else if (rdev->in_sync) {
2401                         info.state |= (1<<MD_DISK_ACTIVE);
2402                         info.state |= (1<<MD_DISK_SYNC);
2403                 }
2404                 if (test_bit(WriteMostly, &rdev->flags))
2405                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2406         } else {
2407                 info.major = info.minor = 0;
2408                 info.raid_disk = -1;
2409                 info.state = (1<<MD_DISK_REMOVED);
2410         }
2411
2412         if (copy_to_user(arg, &info, sizeof(info)))
2413                 return -EFAULT;
2414
2415         return 0;
2416 }
2417
2418 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2419 {
2420         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2421         mdk_rdev_t *rdev;
2422         dev_t dev = MKDEV(info->major,info->minor);
2423
2424         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2425                 return -EOVERFLOW;
2426
2427         if (!mddev->raid_disks) {
2428                 int err;
2429                 /* expecting a device which has a superblock */
2430                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2431                 if (IS_ERR(rdev)) {
2432                         printk(KERN_WARNING
2433                                 "md: md_import_device returned %ld\n",
2434                                 PTR_ERR(rdev));
2435                         return PTR_ERR(rdev);
2436                 }
2437                 if (!list_empty(&mddev->disks)) {
2438                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2439                                                         mdk_rdev_t, same_set);
2440                         int err = super_types[mddev->major_version]
2441                                 .load_super(rdev, rdev0, mddev->minor_version);
2442                         if (err < 0) {
2443                                 printk(KERN_WARNING
2444                                         "md: %s has different UUID to %s\n",
2445                                         bdevname(rdev->bdev,b),
2446                                         bdevname(rdev0->bdev,b2));
2447                                 export_rdev(rdev);
2448                                 return -EINVAL;
2449                         }
2450                 }
2451                 err = bind_rdev_to_array(rdev, mddev);
2452                 if (err)
2453                         export_rdev(rdev);
2454                 return err;
2455         }
2456
2457         /*
2458          * add_new_disk can be used once the array is assembled
2459          * to add "hot spares".  They must already have a superblock
2460          * written
2461          */
2462         if (mddev->pers) {
2463                 int err;
2464                 if (!mddev->pers->hot_add_disk) {
2465                         printk(KERN_WARNING
2466                                 "%s: personality does not support diskops!\n",
2467                                mdname(mddev));
2468                         return -EINVAL;
2469                 }
2470                 if (mddev->persistent)
2471                         rdev = md_import_device(dev, mddev->major_version,
2472                                                 mddev->minor_version);
2473                 else
2474                         rdev = md_import_device(dev, -1, -1);
2475                 if (IS_ERR(rdev)) {
2476                         printk(KERN_WARNING
2477                                 "md: md_import_device returned %ld\n",
2478                                 PTR_ERR(rdev));
2479                         return PTR_ERR(rdev);
2480                 }
2481                 /* set save_raid_disk if appropriate */
2482                 if (!mddev->persistent) {
2483                         if (info->state & (1<<MD_DISK_SYNC)  &&
2484                             info->raid_disk < mddev->raid_disks)
2485                                 rdev->raid_disk = info->raid_disk;
2486                         else
2487                                 rdev->raid_disk = -1;
2488                 } else
2489                         super_types[mddev->major_version].
2490                                 validate_super(mddev, rdev);
2491                 rdev->saved_raid_disk = rdev->raid_disk;
2492
2493                 rdev->in_sync = 0; /* just to be sure */
2494                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2495                         set_bit(WriteMostly, &rdev->flags);
2496
2497                 rdev->raid_disk = -1;
2498                 err = bind_rdev_to_array(rdev, mddev);
2499                 if (err)
2500                         export_rdev(rdev);
2501
2502                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2503                 md_wakeup_thread(mddev->thread);
2504                 return err;
2505         }
2506
2507         /* otherwise, add_new_disk is only allowed
2508          * for major_version==0 superblocks
2509          */
2510         if (mddev->major_version != 0) {
2511                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2512                        mdname(mddev));
2513                 return -EINVAL;
2514         }
2515
2516         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2517                 int err;
2518                 rdev = md_import_device (dev, -1, 0);
2519                 if (IS_ERR(rdev)) {
2520                         printk(KERN_WARNING
2521                                 "md: error, md_import_device() returned %ld\n",
2522                                 PTR_ERR(rdev));
2523                         return PTR_ERR(rdev);
2524                 }
2525                 rdev->desc_nr = info->number;
2526                 if (info->raid_disk < mddev->raid_disks)
2527                         rdev->raid_disk = info->raid_disk;
2528                 else
2529                         rdev->raid_disk = -1;
2530
2531                 rdev->faulty = 0;
2532                 if (rdev->raid_disk < mddev->raid_disks)
2533                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2534                 else
2535                         rdev->in_sync = 0;
2536
2537                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2538                         set_bit(WriteMostly, &rdev->flags);
2539
2540                 err = bind_rdev_to_array(rdev, mddev);
2541                 if (err) {
2542                         export_rdev(rdev);
2543                         return err;
2544                 }
2545
2546                 if (!mddev->persistent) {
2547                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2548                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2549                 } else
2550                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2551                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2552
2553                 if (!mddev->size || (mddev->size > rdev->size))
2554                         mddev->size = rdev->size;
2555         }
2556
2557         return 0;
2558 }
2559
2560 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2561 {
2562         char b[BDEVNAME_SIZE];
2563         mdk_rdev_t *rdev;
2564
2565         if (!mddev->pers)
2566                 return -ENODEV;
2567
2568         rdev = find_rdev(mddev, dev);
2569         if (!rdev)
2570                 return -ENXIO;
2571
2572         if (rdev->raid_disk >= 0)
2573                 goto busy;
2574
2575         kick_rdev_from_array(rdev);
2576         md_update_sb(mddev);
2577
2578         return 0;
2579 busy:
2580         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2581                 bdevname(rdev->bdev,b), mdname(mddev));
2582         return -EBUSY;
2583 }
2584
2585 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2586 {
2587         char b[BDEVNAME_SIZE];
2588         int err;
2589         unsigned int size;
2590         mdk_rdev_t *rdev;
2591
2592         if (!mddev->pers)
2593                 return -ENODEV;
2594
2595         if (mddev->major_version != 0) {
2596                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2597                         " version-0 superblocks.\n",
2598                         mdname(mddev));
2599                 return -EINVAL;
2600         }
2601         if (!mddev->pers->hot_add_disk) {
2602                 printk(KERN_WARNING
2603                         "%s: personality does not support diskops!\n",
2604                         mdname(mddev));
2605                 return -EINVAL;
2606         }
2607
2608         rdev = md_import_device (dev, -1, 0);
2609         if (IS_ERR(rdev)) {
2610                 printk(KERN_WARNING
2611                         "md: error, md_import_device() returned %ld\n",
2612                         PTR_ERR(rdev));
2613                 return -EINVAL;
2614         }
2615
2616         if (mddev->persistent)
2617                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2618         else
2619                 rdev->sb_offset =
2620                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2621
2622         size = calc_dev_size(rdev, mddev->chunk_size);
2623         rdev->size = size;
2624
2625         if (size < mddev->size) {
2626                 printk(KERN_WARNING
2627                         "%s: disk size %llu blocks < array size %llu\n",
2628                         mdname(mddev), (unsigned long long)size,
2629                         (unsigned long long)mddev->size);
2630                 err = -ENOSPC;
2631                 goto abort_export;
2632         }
2633
2634         if (rdev->faulty) {
2635                 printk(KERN_WARNING
2636                         "md: can not hot-add faulty %s disk to %s!\n",
2637                         bdevname(rdev->bdev,b), mdname(mddev));
2638                 err = -EINVAL;
2639                 goto abort_export;
2640         }
2641         rdev->in_sync = 0;
2642         rdev->desc_nr = -1;
2643         bind_rdev_to_array(rdev, mddev);
2644
2645         /*
2646          * The rest should better be atomic, we can have disk failures
2647          * noticed in interrupt contexts ...
2648          */
2649
2650         if (rdev->desc_nr == mddev->max_disks) {
2651                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2652                         mdname(mddev));
2653                 err = -EBUSY;
2654                 goto abort_unbind_export;
2655         }
2656
2657         rdev->raid_disk = -1;
2658
2659         md_update_sb(mddev);
2660
2661         /*
2662          * Kick recovery, maybe this spare has to be added to the
2663          * array immediately.
2664          */
2665         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2666         md_wakeup_thread(mddev->thread);
2667
2668         return 0;
2669
2670 abort_unbind_export:
2671         unbind_rdev_from_array(rdev);
2672
2673 abort_export:
2674         export_rdev(rdev);
2675         return err;
2676 }
2677
2678 /* similar to deny_write_access, but accounts for our holding a reference
2679  * to the file ourselves */
2680 static int deny_bitmap_write_access(struct file * file)
2681 {
2682         struct inode *inode = file->f_mapping->host;
2683
2684         spin_lock(&inode->i_lock);
2685         if (atomic_read(&inode->i_writecount) > 1) {
2686                 spin_unlock(&inode->i_lock);
2687                 return -ETXTBSY;
2688         }
2689         atomic_set(&inode->i_writecount, -1);
2690         spin_unlock(&inode->i_lock);
2691
2692         return 0;
2693 }
2694
2695 static int set_bitmap_file(mddev_t *mddev, int fd)
2696 {
2697         int err;
2698
2699         if (mddev->pers) {
2700                 if (!mddev->pers->quiesce)
2701                         return -EBUSY;
2702                 if (mddev->recovery || mddev->sync_thread)
2703                         return -EBUSY;
2704                 /* we should be able to change the bitmap.. */
2705         }
2706
2707
2708         if (fd >= 0) {
2709                 if (mddev->bitmap)
2710                         return -EEXIST; /* cannot add when bitmap is present */
2711                 mddev->bitmap_file = fget(fd);
2712
2713                 if (mddev->bitmap_file == NULL) {
2714                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2715                                mdname(mddev));
2716                         return -EBADF;
2717                 }
2718
2719                 err = deny_bitmap_write_access(mddev->bitmap_file);
2720                 if (err) {
2721                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2722                                mdname(mddev));
2723                         fput(mddev->bitmap_file);
2724                         mddev->bitmap_file = NULL;
2725                         return err;
2726                 }
2727                 mddev->bitmap_offset = 0; /* file overrides offset */
2728         } else if (mddev->bitmap == NULL)
2729                 return -ENOENT; /* cannot remove what isn't there */
2730         err = 0;
2731         if (mddev->pers) {
2732                 mddev->pers->quiesce(mddev, 1);
2733                 if (fd >= 0)
2734                         err = bitmap_create(mddev);
2735                 if (fd < 0 || err)
2736                         bitmap_destroy(mddev);
2737                 mddev->pers->quiesce(mddev, 0);
2738         } else if (fd < 0) {
2739                 if (mddev->bitmap_file)
2740                         fput(mddev->bitmap_file);
2741                 mddev->bitmap_file = NULL;
2742         }
2743
2744         return err;
2745 }
2746
2747 /*
2748  * set_array_info is used two different ways
2749  * The original usage is when creating a new array.
2750  * In this usage, raid_disks is > 0 and it together with
2751  *  level, size, not_persistent,layout,chunksize determine the
2752  *  shape of the array.
2753  *  This will always create an array with a type-0.90.0 superblock.
2754  * The newer usage is when assembling an array.
2755  *  In this case raid_disks will be 0, and the major_version field is
2756  *  use to determine which style super-blocks are to be found on the devices.
2757  *  The minor and patch _version numbers are also kept incase the
2758  *  super_block handler wishes to interpret them.
2759  */
2760 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2761 {
2762
2763         if (info->raid_disks == 0) {
2764                 /* just setting version number for superblock loading */
2765                 if (info->major_version < 0 ||
2766                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2767                     super_types[info->major_version].name == NULL) {
2768                         /* maybe try to auto-load a module? */
2769                         printk(KERN_INFO
2770                                 "md: superblock version %d not known\n",
2771                                 info->major_version);
2772                         return -EINVAL;
2773                 }
2774                 mddev->major_version = info->major_version;
2775                 mddev->minor_version = info->minor_version;
2776                 mddev->patch_version = info->patch_version;
2777                 return 0;
2778         }
2779         mddev->major_version = MD_MAJOR_VERSION;
2780         mddev->minor_version = MD_MINOR_VERSION;
2781         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2782         mddev->ctime         = get_seconds();
2783
2784         mddev->level         = info->level;
2785         mddev->size          = info->size;
2786         mddev->raid_disks    = info->raid_disks;
2787         /* don't set md_minor, it is determined by which /dev/md* was
2788          * openned
2789          */
2790         if (info->state & (1<<MD_SB_CLEAN))
2791                 mddev->recovery_cp = MaxSector;
2792         else
2793                 mddev->recovery_cp = 0;
2794         mddev->persistent    = ! info->not_persistent;
2795
2796         mddev->layout        = info->layout;
2797         mddev->chunk_size    = info->chunk_size;
2798
2799         mddev->max_disks     = MD_SB_DISKS;
2800
2801         mddev->sb_dirty      = 1;
2802
2803         /*
2804          * Generate a 128 bit UUID
2805          */
2806         get_random_bytes(mddev->uuid, 16);
2807
2808         return 0;
2809 }
2810
2811 /*
2812  * update_array_info is used to change the configuration of an
2813  * on-line array.
2814  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2815  * fields in the info are checked against the array.
2816  * Any differences that cannot be handled will cause an error.
2817  * Normally, only one change can be managed at a time.
2818  */
2819 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2820 {
2821         int rv = 0;
2822         int cnt = 0;
2823         int state = 0;
2824
2825         /* calculate expected state,ignoring low bits */
2826         if (mddev->bitmap && mddev->bitmap_offset)
2827                 state |= (1 << MD_SB_BITMAP_PRESENT);
2828
2829         if (mddev->major_version != info->major_version ||
2830             mddev->minor_version != info->minor_version ||
2831 /*          mddev->patch_version != info->patch_version || */
2832             mddev->ctime         != info->ctime         ||
2833             mddev->level         != info->level         ||
2834 /*          mddev->layout        != info->layout        || */
2835             !mddev->persistent   != info->not_persistent||
2836             mddev->chunk_size    != info->chunk_size    ||
2837             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2838             ((state^info->state) & 0xfffffe00)
2839                 )
2840                 return -EINVAL;
2841         /* Check there is only one change */
2842         if (mddev->size != info->size) cnt++;
2843         if (mddev->raid_disks != info->raid_disks) cnt++;
2844         if (mddev->layout != info->layout) cnt++;
2845         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2846         if (cnt == 0) return 0;
2847         if (cnt > 1) return -EINVAL;
2848
2849         if (mddev->layout != info->layout) {
2850                 /* Change layout
2851                  * we don't need to do anything at the md level, the
2852                  * personality will take care of it all.
2853                  */
2854                 if (mddev->pers->reconfig == NULL)
2855                         return -EINVAL;
2856                 else
2857                         return mddev->pers->reconfig(mddev, info->layout, -1);
2858         }
2859         if (mddev->size != info->size) {
2860                 mdk_rdev_t * rdev;
2861                 struct list_head *tmp;
2862                 if (mddev->pers->resize == NULL)
2863                         return -EINVAL;
2864                 /* The "size" is the amount of each device that is used.
2865                  * This can only make sense for arrays with redundancy.
2866                  * linear and raid0 always use whatever space is available
2867                  * We can only consider changing the size if no resync
2868                  * or reconstruction is happening, and if the new size
2869                  * is acceptable. It must fit before the sb_offset or,
2870                  * if that is <data_offset, it must fit before the
2871                  * size of each device.
2872                  * If size is zero, we find the largest size that fits.
2873                  */
2874                 if (mddev->sync_thread)
2875                         return -EBUSY;
2876                 ITERATE_RDEV(mddev,rdev,tmp) {
2877                         sector_t avail;
2878                         int fit = (info->size == 0);
2879                         if (rdev->sb_offset > rdev->data_offset)
2880                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2881                         else
2882                                 avail = get_capacity(rdev->bdev->bd_disk)
2883                                         - rdev->data_offset;
2884                         if (fit && (info->size == 0 || info->size > avail/2))
2885                                 info->size = avail/2;
2886                         if (avail < ((sector_t)info->size << 1))
2887                                 return -ENOSPC;
2888                 }
2889                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2890                 if (!rv) {
2891                         struct block_device *bdev;
2892
2893                         bdev = bdget_disk(mddev->gendisk, 0);
2894                         if (bdev) {
2895                                 down(&bdev->bd_inode->i_sem);
2896                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2897                                 up(&bdev->bd_inode->i_sem);
2898                                 bdput(bdev);
2899                         }
2900                 }
2901         }
2902         if (mddev->raid_disks    != info->raid_disks) {
2903                 /* change the number of raid disks */
2904                 if (mddev->pers->reshape == NULL)
2905                         return -EINVAL;
2906                 if (info->raid_disks <= 0 ||
2907                     info->raid_disks >= mddev->max_disks)
2908                         return -EINVAL;
2909                 if (mddev->sync_thread)
2910                         return -EBUSY;
2911                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2912                 if (!rv) {
2913                         struct block_device *bdev;
2914
2915                         bdev = bdget_disk(mddev->gendisk, 0);
2916                         if (bdev) {
2917                                 down(&bdev->bd_inode->i_sem);
2918                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2919                                 up(&bdev->bd_inode->i_sem);
2920                                 bdput(bdev);
2921                         }
2922                 }
2923         }
2924         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2925                 if (mddev->pers->quiesce == NULL)
2926                         return -EINVAL;
2927                 if (mddev->recovery || mddev->sync_thread)
2928                         return -EBUSY;
2929                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2930                         /* add the bitmap */
2931                         if (mddev->bitmap)
2932                                 return -EEXIST;
2933                         if (mddev->default_bitmap_offset == 0)
2934                                 return -EINVAL;
2935                         mddev->bitmap_offset = mddev->default_bitmap_offset;
2936                         mddev->pers->quiesce(mddev, 1);
2937                         rv = bitmap_create(mddev);
2938                         if (rv)
2939                                 bitmap_destroy(mddev);
2940                         mddev->pers->quiesce(mddev, 0);
2941                 } else {
2942                         /* remove the bitmap */
2943                         if (!mddev->bitmap)
2944                                 return -ENOENT;
2945                         if (mddev->bitmap->file)
2946                                 return -EINVAL;
2947                         mddev->pers->quiesce(mddev, 1);
2948                         bitmap_destroy(mddev);
2949                         mddev->pers->quiesce(mddev, 0);
2950                         mddev->bitmap_offset = 0;
2951                 }
2952         }
2953         md_update_sb(mddev);
2954         return rv;
2955 }
2956
2957 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2958 {
2959         mdk_rdev_t *rdev;
2960
2961         if (mddev->pers == NULL)
2962                 return -ENODEV;
2963
2964         rdev = find_rdev(mddev, dev);
2965         if (!rdev)
2966                 return -ENODEV;
2967
2968         md_error(mddev, rdev);
2969         return 0;
2970 }
2971
2972 static int md_ioctl(struct inode *inode, struct file *file,
2973                         unsigned int cmd, unsigned long arg)
2974 {
2975         int err = 0;
2976         void __user *argp = (void __user *)arg;
2977         struct hd_geometry __user *loc = argp;
2978         mddev_t *mddev = NULL;
2979
2980         if (!capable(CAP_SYS_ADMIN))
2981                 return -EACCES;
2982
2983         /*
2984          * Commands dealing with the RAID driver but not any
2985          * particular array:
2986          */
2987         switch (cmd)
2988         {
2989                 case RAID_VERSION:
2990                         err = get_version(argp);
2991                         goto done;
2992
2993                 case PRINT_RAID_DEBUG:
2994                         err = 0;
2995                         md_print_devices();
2996                         goto done;
2997
2998 #ifndef MODULE
2999                 case RAID_AUTORUN:
3000                         err = 0;
3001                         autostart_arrays(arg);
3002                         goto done;
3003 #endif
3004                 default:;
3005         }
3006
3007         /*
3008          * Commands creating/starting a new array:
3009          */
3010
3011         mddev = inode->i_bdev->bd_disk->private_data;
3012
3013         if (!mddev) {
3014                 BUG();
3015                 goto abort;
3016         }
3017
3018
3019         if (cmd == START_ARRAY) {
3020                 /* START_ARRAY doesn't need to lock the array as autostart_array
3021                  * does the locking, and it could even be a different array
3022                  */
3023                 static int cnt = 3;
3024                 if (cnt > 0 ) {
3025                         printk(KERN_WARNING
3026                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3027                                "This will not be supported beyond 2.6\n",
3028                                current->comm, current->pid);
3029                         cnt--;
3030                 }
3031                 err = autostart_array(new_decode_dev(arg));
3032                 if (err) {
3033                         printk(KERN_WARNING "md: autostart failed!\n");
3034                         goto abort;
3035                 }
3036                 goto done;
3037         }
3038
3039         err = mddev_lock(mddev);
3040         if (err) {
3041                 printk(KERN_INFO
3042                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3043                         err, cmd);
3044                 goto abort;
3045         }
3046
3047         switch (cmd)
3048         {
3049                 case SET_ARRAY_INFO:
3050                         {
3051                                 mdu_array_info_t info;
3052                                 if (!arg)
3053                                         memset(&info, 0, sizeof(info));
3054                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3055                                         err = -EFAULT;
3056                                         goto abort_unlock;
3057                                 }
3058                                 if (mddev->pers) {
3059                                         err = update_array_info(mddev, &info);
3060                                         if (err) {
3061                                                 printk(KERN_WARNING "md: couldn't update"
3062                                                        " array info. %d\n", err);
3063                                                 goto abort_unlock;
3064                                         }
3065                                         goto done_unlock;
3066                                 }
3067                                 if (!list_empty(&mddev->disks)) {
3068                                         printk(KERN_WARNING
3069                                                "md: array %s already has disks!\n",
3070                                                mdname(mddev));
3071                                         err = -EBUSY;
3072                                         goto abort_unlock;
3073                                 }
3074                                 if (mddev->raid_disks) {
3075                                         printk(KERN_WARNING
3076                                                "md: array %s already initialised!\n",
3077                                                mdname(mddev));
3078                                         err = -EBUSY;
3079                                         goto abort_unlock;
3080                                 }
3081                                 err = set_array_info(mddev, &info);
3082                                 if (err) {
3083                                         printk(KERN_WARNING "md: couldn't set"
3084                                                " array info. %d\n", err);
3085                                         goto abort_unlock;
3086                                 }
3087                         }
3088                         goto done_unlock;
3089
3090                 default:;
3091         }
3092
3093         /*
3094          * Commands querying/configuring an existing array:
3095          */
3096         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3097          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3098         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3099                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3100                 err = -ENODEV;
3101                 goto abort_unlock;
3102         }
3103
3104         /*
3105          * Commands even a read-only array can execute:
3106          */
3107         switch (cmd)
3108         {
3109                 case GET_ARRAY_INFO:
3110                         err = get_array_info(mddev, argp);
3111                         goto done_unlock;
3112
3113                 case GET_BITMAP_FILE:
3114                         err = get_bitmap_file(mddev, argp);
3115                         goto done_unlock;
3116
3117                 case GET_DISK_INFO:
3118                         err = get_disk_info(mddev, argp);
3119                         goto done_unlock;
3120
3121                 case RESTART_ARRAY_RW:
3122                         err = restart_array(mddev);
3123                         goto done_unlock;
3124
3125                 case STOP_ARRAY:
3126                         err = do_md_stop (mddev, 0);
3127                         goto done_unlock;
3128
3129                 case STOP_ARRAY_RO:
3130                         err = do_md_stop (mddev, 1);
3131                         goto done_unlock;
3132
3133         /*
3134          * We have a problem here : there is no easy way to give a CHS
3135          * virtual geometry. We currently pretend that we have a 2 heads
3136          * 4 sectors (with a BIG number of cylinders...). This drives
3137          * dosfs just mad... ;-)
3138          */
3139                 case HDIO_GETGEO:
3140                         if (!loc) {
3141                                 err = -EINVAL;
3142                                 goto abort_unlock;
3143                         }
3144                         err = put_user (2, (char __user *) &loc->heads);
3145                         if (err)
3146                                 goto abort_unlock;
3147                         err = put_user (4, (char __user *) &loc->sectors);
3148                         if (err)
3149                                 goto abort_unlock;
3150                         err = put_user(get_capacity(mddev->gendisk)/8,
3151                                         (short __user *) &loc->cylinders);
3152                         if (err)
3153                                 goto abort_unlock;
3154                         err = put_user (get_start_sect(inode->i_bdev),
3155                                                 (long __user *) &loc->start);
3156                         goto done_unlock;
3157         }
3158
3159         /*
3160          * The remaining ioctls are changing the state of the
3161          * superblock, so we do not allow read-only arrays
3162          * here:
3163          */
3164         if (mddev->ro) {
3165                 err = -EROFS;
3166                 goto abort_unlock;
3167         }
3168
3169         switch (cmd)
3170         {
3171                 case ADD_NEW_DISK:
3172                 {
3173                         mdu_disk_info_t info;
3174                         if (copy_from_user(&info, argp, sizeof(info)))
3175                                 err = -EFAULT;
3176                         else
3177                                 err = add_new_disk(mddev, &info);
3178                         goto done_unlock;
3179                 }
3180
3181                 case HOT_REMOVE_DISK:
3182                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3183                         goto done_unlock;
3184
3185                 case HOT_ADD_DISK:
3186                         err = hot_add_disk(mddev, new_decode_dev(arg));
3187                         goto done_unlock;
3188
3189                 case SET_DISK_FAULTY:
3190                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3191                         goto done_unlock;
3192
3193                 case RUN_ARRAY:
3194                         err = do_md_run (mddev);
3195                         goto done_unlock;
3196
3197                 case SET_BITMAP_FILE:
3198                         err = set_bitmap_file(mddev, (int)arg);
3199                         goto done_unlock;
3200
3201                 default:
3202                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3203                                 printk(KERN_WARNING "md: %s(pid %d) used"
3204                                         " obsolete MD ioctl, upgrade your"
3205                                         " software to use new ictls.\n",
3206                                         current->comm, current->pid);
3207                         err = -EINVAL;
3208                         goto abort_unlock;
3209         }
3210
3211 done_unlock:
3212 abort_unlock:
3213         mddev_unlock(mddev);
3214
3215         return err;
3216 done:
3217         if (err)
3218                 MD_BUG();
3219 abort:
3220         return err;
3221 }
3222
3223 static int md_open(struct inode *inode, struct file *file)
3224 {
3225         /*
3226          * Succeed if we can lock the mddev, which confirms that
3227          * it isn't being stopped right now.
3228          */
3229         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3230         int err;
3231
3232         if ((err = mddev_lock(mddev)))
3233                 goto out;
3234
3235         err = 0;
3236         mddev_get(mddev);
3237         mddev_unlock(mddev);
3238
3239         check_disk_change(inode->i_bdev);
3240  out:
3241         return err;
3242 }
3243
3244 static int md_release(struct inode *inode, struct file * file)
3245 {
3246         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3247
3248         if (!mddev)
3249                 BUG();
3250         mddev_put(mddev);
3251
3252         return 0;
3253 }
3254
3255 static int md_media_changed(struct gendisk *disk)
3256 {
3257         mddev_t *mddev = disk->private_data;
3258
3259         return mddev->changed;
3260 }
3261
3262 static int md_revalidate(struct gendisk *disk)
3263 {
3264         mddev_t *mddev = disk->private_data;
3265
3266         mddev->changed = 0;
3267         return 0;
3268 }
3269 static struct block_device_operations md_fops =
3270 {
3271         .owner          = THIS_MODULE,
3272         .open           = md_open,
3273         .release        = md_release,
3274         .ioctl          = md_ioctl,
3275         .media_changed  = md_media_changed,
3276         .revalidate_disk= md_revalidate,
3277 };
3278
3279 static int md_thread(void * arg)
3280 {
3281         mdk_thread_t *thread = arg;
3282
3283         /*
3284          * md_thread is a 'system-thread', it's priority should be very
3285          * high. We avoid resource deadlocks individually in each
3286          * raid personality. (RAID5 does preallocation) We also use RR and
3287          * the very same RT priority as kswapd, thus we will never get
3288          * into a priority inversion deadlock.
3289          *
3290          * we definitely have to have equal or higher priority than
3291          * bdflush, otherwise bdflush will deadlock if there are too
3292          * many dirty RAID5 blocks.
3293          */
3294
3295         allow_signal(SIGKILL);
3296         complete(thread->event);
3297         while (!kthread_should_stop()) {
3298                 void (*run)(mddev_t *);
3299
3300                 wait_event_interruptible_timeout(thread->wqueue,
3301                                                  test_bit(THREAD_WAKEUP, &thread->flags)
3302                                                  || kthread_should_stop(),
3303                                                  thread->timeout);
3304                 try_to_freeze();
3305
3306                 clear_bit(THREAD_WAKEUP, &thread->flags);
3307
3308                 run = thread->run;
3309                 if (run)
3310                         run(thread->mddev);
3311         }
3312
3313         return 0;
3314 }
3315
3316 void md_wakeup_thread(mdk_thread_t *thread)
3317 {
3318         if (thread) {
3319                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3320                 set_bit(THREAD_WAKEUP, &thread->flags);
3321                 wake_up(&thread->wqueue);
3322         }
3323 }
3324
3325 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3326                                  const char *name)
3327 {
3328         mdk_thread_t *thread;
3329         struct completion event;
3330
3331         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3332         if (!thread)
3333                 return NULL;
3334
3335         memset(thread, 0, sizeof(mdk_thread_t));
3336         init_waitqueue_head(&thread->wqueue);
3337
3338         init_completion(&event);
3339         thread->event = &event;
3340         thread->run = run;
3341         thread->mddev = mddev;
3342         thread->name = name;
3343         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3344         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3345         if (IS_ERR(thread->tsk)) {
3346                 kfree(thread);
3347                 return NULL;
3348         }
3349         wait_for_completion(&event);
3350         return thread;
3351 }
3352
3353 void md_unregister_thread(mdk_thread_t *thread)
3354 {
3355         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3356
3357         kthread_stop(thread->tsk);
3358         kfree(thread);
3359 }
3360
3361 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3362 {
3363         if (!mddev) {
3364                 MD_BUG();
3365                 return;
3366         }
3367
3368         if (!rdev || rdev->faulty)
3369                 return;
3370 /*
3371         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3372                 mdname(mddev),
3373                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3374                 __builtin_return_address(0),__builtin_return_address(1),
3375                 __builtin_return_address(2),__builtin_return_address(3));
3376 */
3377         if (!mddev->pers->error_handler)
3378                 return;
3379         mddev->pers->error_handler(mddev,rdev);
3380         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3381         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3382         md_wakeup_thread(mddev->thread);
3383 }
3384
3385 /* seq_file implementation /proc/mdstat */
3386
3387 static void status_unused(struct seq_file *seq)
3388 {
3389         int i = 0;
3390         mdk_rdev_t *rdev;
3391         struct list_head *tmp;
3392
3393         seq_printf(seq, "unused devices: ");
3394
3395         ITERATE_RDEV_PENDING(rdev,tmp) {
3396                 char b[BDEVNAME_SIZE];
3397                 i++;
3398                 seq_printf(seq, "%s ",
3399                               bdevname(rdev->bdev,b));
3400         }
3401         if (!i)
3402                 seq_printf(seq, "<none>");
3403
3404         seq_printf(seq, "\n");
3405 }
3406
3407
3408 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3409 {
3410         unsigned long max_blocks, resync, res, dt, db, rt;
3411
3412         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3413
3414         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3415                 max_blocks = mddev->resync_max_sectors >> 1;
3416         else
3417                 max_blocks = mddev->size;
3418
3419         /*
3420          * Should not happen.
3421          */
3422         if (!max_blocks) {
3423                 MD_BUG();
3424                 return;
3425         }
3426         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3427         {
3428                 int i, x = res/50, y = 20-x;
3429                 seq_printf(seq, "[");
3430                 for (i = 0; i < x; i++)
3431                         seq_printf(seq, "=");
3432                 seq_printf(seq, ">");
3433                 for (i = 0; i < y; i++)
3434                         seq_printf(seq, ".");
3435                 seq_printf(seq, "] ");
3436         }
3437         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3438                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3439                        "resync" : "recovery"),
3440                       res/10, res % 10, resync, max_blocks);
3441
3442         /*
3443          * We do not want to overflow, so the order of operands and
3444          * the * 100 / 100 trick are important. We do a +1 to be
3445          * safe against division by zero. We only estimate anyway.
3446          *
3447          * dt: time from mark until now
3448          * db: blocks written from mark until now
3449          * rt: remaining time
3450          */
3451         dt = ((jiffies - mddev->resync_mark) / HZ);
3452         if (!dt) dt++;
3453         db = resync - (mddev->resync_mark_cnt/2);
3454         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3455
3456         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3457
3458         seq_printf(seq, " speed=%ldK/sec", db/dt);
3459 }
3460
3461 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3462 {
3463         struct list_head *tmp;
3464         loff_t l = *pos;
3465         mddev_t *mddev;
3466
3467         if (l >= 0x10000)
3468                 return NULL;
3469         if (!l--)
3470                 /* header */
3471                 return (void*)1;
3472
3473         spin_lock(&all_mddevs_lock);
3474         list_for_each(tmp,&all_mddevs)
3475                 if (!l--) {
3476                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3477                         mddev_get(mddev);
3478                         spin_unlock(&all_mddevs_lock);
3479                         return mddev;
3480                 }
3481         spin_unlock(&all_mddevs_lock);
3482         if (!l--)
3483                 return (void*)2;/* tail */
3484         return NULL;
3485 }
3486
3487 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3488 {
3489         struct list_head *tmp;
3490         mddev_t *next_mddev, *mddev = v;
3491
3492         ++*pos;
3493         if (v == (void*)2)
3494                 return NULL;
3495
3496         spin_lock(&all_mddevs_lock);
3497         if (v == (void*)1)
3498                 tmp = all_mddevs.next;
3499         else
3500                 tmp = mddev->all_mddevs.next;
3501         if (tmp != &all_mddevs)
3502                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3503         else {
3504                 next_mddev = (void*)2;
3505                 *pos = 0x10000;
3506         }
3507         spin_unlock(&all_mddevs_lock);
3508
3509         if (v != (void*)1)
3510                 mddev_put(mddev);
3511         return next_mddev;
3512
3513 }
3514
3515 static void md_seq_stop(struct seq_file *seq, void *v)
3516 {
3517         mddev_t *mddev = v;
3518
3519         if (mddev && v != (void*)1 && v != (void*)2)
3520                 mddev_put(mddev);
3521 }
3522
3523 static int md_seq_show(struct seq_file *seq, void *v)
3524 {
3525         mddev_t *mddev = v;
3526         sector_t size;
3527         struct list_head *tmp2;
3528         mdk_rdev_t *rdev;
3529         int i;
3530         struct bitmap *bitmap;
3531
3532         if (v == (void*)1) {
3533                 seq_printf(seq, "Personalities : ");
3534                 spin_lock(&pers_lock);
3535                 for (i = 0; i < MAX_PERSONALITY; i++)
3536                         if (pers[i])
3537                                 seq_printf(seq, "[%s] ", pers[i]->name);
3538
3539                 spin_unlock(&pers_lock);
3540                 seq_printf(seq, "\n");
3541                 return 0;
3542         }
3543         if (v == (void*)2) {
3544                 status_unused(seq);
3545                 return 0;
3546         }
3547
3548         if (mddev_lock(mddev)!=0)
3549                 return -EINTR;
3550         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3551                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3552                                                 mddev->pers ? "" : "in");
3553                 if (mddev->pers) {
3554                         if (mddev->ro)
3555                                 seq_printf(seq, " (read-only)");
3556                         seq_printf(seq, " %s", mddev->pers->name);
3557                 }
3558
3559                 size = 0;
3560                 ITERATE_RDEV(mddev,rdev,tmp2) {
3561                         char b[BDEVNAME_SIZE];
3562                         seq_printf(seq, " %s[%d]",
3563                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3564                         if (test_bit(WriteMostly, &rdev->flags))
3565                                 seq_printf(seq, "(W)");
3566                         if (rdev->faulty) {
3567                                 seq_printf(seq, "(F)");
3568                                 continue;
3569                         } else if (rdev->raid_disk < 0)
3570                                 seq_printf(seq, "(S)"); /* spare */
3571                         size += rdev->size;
3572                 }
3573
3574                 if (!list_empty(&mddev->disks)) {
3575                         if (mddev->pers)
3576                                 seq_printf(seq, "\n      %llu blocks",
3577                                         (unsigned long long)mddev->array_size);
3578                         else
3579                                 seq_printf(seq, "\n      %llu blocks",
3580                                         (unsigned long long)size);
3581                 }
3582                 if (mddev->persistent) {
3583                         if (mddev->major_version != 0 ||
3584                             mddev->minor_version != 90) {
3585                                 seq_printf(seq," super %d.%d",
3586                                            mddev->major_version,
3587                                            mddev->minor_version);
3588                         }
3589                 } else
3590                         seq_printf(seq, " super non-persistent");
3591
3592                 if (mddev->pers) {
3593                         mddev->pers->status (seq, mddev);
3594                         seq_printf(seq, "\n      ");
3595                         if (mddev->curr_resync > 2) {
3596                                 status_resync (seq, mddev);
3597                                 seq_printf(seq, "\n      ");
3598                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3599                                 seq_printf(seq, "       resync=DELAYED\n      ");
3600                 } else
3601                         seq_printf(seq, "\n       ");
3602
3603                 if ((bitmap = mddev->bitmap)) {
3604                         unsigned long chunk_kb;
3605                         unsigned long flags;
3606                         spin_lock_irqsave(&bitmap->lock, flags);
3607                         chunk_kb = bitmap->chunksize >> 10;
3608                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3609                                 "%lu%s chunk",
3610                                 bitmap->pages - bitmap->missing_pages,
3611                                 bitmap->pages,
3612                                 (bitmap->pages - bitmap->missing_pages)
3613                                         << (PAGE_SHIFT - 10),
3614                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3615                                 chunk_kb ? "KB" : "B");
3616                         if (bitmap->file) {
3617                                 seq_printf(seq, ", file: ");
3618                                 seq_path(seq, bitmap->file->f_vfsmnt,
3619                                          bitmap->file->f_dentry," \t\n");
3620                         }
3621
3622                         seq_printf(seq, "\n");
3623                         spin_unlock_irqrestore(&bitmap->lock, flags);
3624                 }
3625
3626                 seq_printf(seq, "\n");
3627         }
3628         mddev_unlock(mddev);
3629
3630         return 0;
3631 }
3632
3633 static struct seq_operations md_seq_ops = {
3634         .start  = md_seq_start,
3635         .next   = md_seq_next,
3636         .stop   = md_seq_stop,
3637         .show   = md_seq_show,
3638 };
3639
3640 static int md_seq_open(struct inode *inode, struct file *file)
3641 {
3642         int error;
3643
3644         error = seq_open(file, &md_seq_ops);
3645         return error;
3646 }
3647
3648 static struct file_operations md_seq_fops = {
3649         .open           = md_seq_open,
3650         .read           = seq_read,
3651         .llseek         = seq_lseek,
3652         .release        = seq_release,
3653 };
3654
3655 int register_md_personality(int pnum, mdk_personality_t *p)
3656 {
3657         if (pnum >= MAX_PERSONALITY) {
3658                 printk(KERN_ERR
3659                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3660                        p->name, pnum, MAX_PERSONALITY-1);
3661                 return -EINVAL;
3662         }
3663
3664         spin_lock(&pers_lock);
3665         if (pers[pnum]) {
3666                 spin_unlock(&pers_lock);
3667                 return -EBUSY;
3668         }
3669
3670         pers[pnum] = p;
3671         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3672         spin_unlock(&pers_lock);
3673         return 0;
3674 }
3675
3676 int unregister_md_personality(int pnum)
3677 {
3678         if (pnum >= MAX_PERSONALITY)
3679                 return -EINVAL;
3680
3681         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3682         spin_lock(&pers_lock);
3683         pers[pnum] = NULL;
3684         spin_unlock(&pers_lock);
3685         return 0;
3686 }
3687
3688 static int is_mddev_idle(mddev_t *mddev)
3689 {
3690         mdk_rdev_t * rdev;
3691         struct list_head *tmp;
3692         int idle;
3693         unsigned long curr_events;
3694
3695         idle = 1;
3696         ITERATE_RDEV(mddev,rdev,tmp) {
3697                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3698                 curr_events = disk_stat_read(disk, sectors[0]) +
3699                                 disk_stat_read(disk, sectors[1]) -
3700                                 atomic_read(&disk->sync_io);
3701                 /* Allow some slack between valud of curr_events and last_events,
3702                  * as there are some uninteresting races.
3703                  * Note: the following is an unsigned comparison.
3704                  */
3705                 if ((curr_events - rdev->last_events + 32) > 64) {
3706                         rdev->last_events = curr_events;
3707                         idle = 0;
3708                 }
3709         }
3710         return idle;
3711 }
3712
3713 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3714 {
3715         /* another "blocks" (512byte) blocks have been synced */
3716         atomic_sub(blocks, &mddev->recovery_active);
3717         wake_up(&mddev->recovery_wait);
3718         if (!ok) {
3719                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3720                 md_wakeup_thread(mddev->thread);
3721                 // stop recovery, signal do_sync ....
3722         }
3723 }
3724
3725
3726 /* md_write_start(mddev, bi)
3727  * If we need to update some array metadata (e.g. 'active' flag
3728  * in superblock) before writing, schedule a superblock update
3729  * and wait for it to complete.
3730  */
3731 void md_write_start(mddev_t *mddev, struct bio *bi)
3732 {
3733         if (bio_data_dir(bi) != WRITE)
3734                 return;
3735
3736         atomic_inc(&mddev->writes_pending);
3737         if (mddev->in_sync) {
3738                 spin_lock(&mddev->write_lock);
3739                 if (mddev->in_sync) {
3740                         mddev->in_sync = 0;
3741                         mddev->sb_dirty = 1;
3742                         md_wakeup_thread(mddev->thread);
3743                 }
3744                 spin_unlock(&mddev->write_lock);
3745         }
3746         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3747 }
3748
3749 void md_write_end(mddev_t *mddev)
3750 {
3751         if (atomic_dec_and_test(&mddev->writes_pending)) {
3752                 if (mddev->safemode == 2)
3753                         md_wakeup_thread(mddev->thread);
3754                 else
3755                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3756         }
3757 }
3758
3759 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3760
3761 #define SYNC_MARKS      10
3762 #define SYNC_MARK_STEP  (3*HZ)
3763 static void md_do_sync(mddev_t *mddev)
3764 {
3765         mddev_t *mddev2;
3766         unsigned int currspeed = 0,
3767                  window;
3768         sector_t max_sectors,j, io_sectors;
3769         unsigned long mark[SYNC_MARKS];
3770         sector_t mark_cnt[SYNC_MARKS];
3771         int last_mark,m;
3772         struct list_head *tmp;
3773         sector_t last_check;
3774         int skipped = 0;
3775
3776         /* just incase thread restarts... */
3777         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3778                 return;
3779
3780         /* we overload curr_resync somewhat here.
3781          * 0 == not engaged in resync at all
3782          * 2 == checking that there is no conflict with another sync
3783          * 1 == like 2, but have yielded to allow conflicting resync to
3784          *              commense
3785          * other == active in resync - this many blocks
3786          *
3787          * Before starting a resync we must have set curr_resync to
3788          * 2, and then checked that every "conflicting" array has curr_resync
3789          * less than ours.  When we find one that is the same or higher
3790          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3791          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3792          * This will mean we have to start checking from the beginning again.
3793          *
3794          */
3795
3796         do {
3797                 mddev->curr_resync = 2;
3798
3799         try_again:
3800                 if (signal_pending(current) ||
3801                     kthread_should_stop()) {
3802                         flush_signals(current);
3803                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3804                         goto skip;
3805                 }
3806                 ITERATE_MDDEV(mddev2,tmp) {
3807                         if (mddev2 == mddev)
3808                                 continue;
3809                         if (mddev2->curr_resync &&
3810                             match_mddev_units(mddev,mddev2)) {
3811                                 DEFINE_WAIT(wq);
3812                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3813                                         /* arbitrarily yield */
3814                                         mddev->curr_resync = 1;
3815                                         wake_up(&resync_wait);
3816                                 }
3817                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3818                                         /* no need to wait here, we can wait the next
3819                                          * time 'round when curr_resync == 2
3820                                          */
3821                                         continue;
3822                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3823                                 if (!signal_pending(current) &&
3824                                     !kthread_should_stop() &&
3825                                     mddev2->curr_resync >= mddev->curr_resync) {
3826                                         printk(KERN_INFO "md: delaying resync of %s"
3827                                                " until %s has finished resync (they"
3828                                                " share one or more physical units)\n",
3829                                                mdname(mddev), mdname(mddev2));
3830                                         mddev_put(mddev2);
3831                                         schedule();
3832                                         finish_wait(&resync_wait, &wq);
3833                                         goto try_again;
3834                                 }
3835                                 finish_wait(&resync_wait, &wq);
3836                         }
3837                 }
3838         } while (mddev->curr_resync < 2);
3839
3840         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3841                 /* resync follows the size requested by the personality,
3842                  * which defaults to physical size, but can be virtual size
3843                  */
3844                 max_sectors = mddev->resync_max_sectors;
3845         else
3846                 /* recovery follows the physical size of devices */
3847                 max_sectors = mddev->size << 1;
3848
3849         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3850         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3851                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3852         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3853                "(but not more than %d KB/sec) for reconstruction.\n",
3854                sysctl_speed_limit_max);
3855
3856         is_mddev_idle(mddev); /* this also initializes IO event counters */
3857         /* we don't use the checkpoint if there's a bitmap */
3858         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3859                 j = mddev->recovery_cp;
3860         else
3861                 j = 0;
3862         io_sectors = 0;
3863         for (m = 0; m < SYNC_MARKS; m++) {
3864                 mark[m] = jiffies;
3865                 mark_cnt[m] = io_sectors;
3866         }
3867         last_mark = 0;
3868         mddev->resync_mark = mark[last_mark];
3869         mddev->resync_mark_cnt = mark_cnt[last_mark];
3870
3871         /*
3872          * Tune reconstruction:
3873          */
3874         window = 32*(PAGE_SIZE/512);
3875         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3876                 window/2,(unsigned long long) max_sectors/2);
3877
3878         atomic_set(&mddev->recovery_active, 0);
3879         init_waitqueue_head(&mddev->recovery_wait);
3880         last_check = 0;
3881
3882         if (j>2) {
3883                 printk(KERN_INFO
3884                         "md: resuming recovery of %s from checkpoint.\n",
3885                         mdname(mddev));
3886                 mddev->curr_resync = j;
3887         }
3888
3889         while (j < max_sectors) {
3890                 sector_t sectors;
3891
3892                 skipped = 0;
3893                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3894                                             currspeed < sysctl_speed_limit_min);
3895                 if (sectors == 0) {
3896                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3897                         goto out;
3898                 }
3899
3900                 if (!skipped) { /* actual IO requested */
3901                         io_sectors += sectors;
3902                         atomic_add(sectors, &mddev->recovery_active);
3903                 }
3904
3905                 j += sectors;
3906                 if (j>1) mddev->curr_resync = j;
3907
3908
3909                 if (last_check + window > io_sectors || j == max_sectors)
3910                         continue;
3911
3912                 last_check = io_sectors;
3913
3914                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3915                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3916                         break;
3917
3918         repeat:
3919                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3920                         /* step marks */
3921                         int next = (last_mark+1) % SYNC_MARKS;
3922
3923                         mddev->resync_mark = mark[next];
3924                         mddev->resync_mark_cnt = mark_cnt[next];
3925                         mark[next] = jiffies;
3926                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3927                         last_mark = next;
3928                 }
3929
3930
3931                 if (signal_pending(current) || kthread_should_stop()) {
3932                         /*
3933                          * got a signal, exit.
3934                          */
3935                         printk(KERN_INFO
3936                                 "md: md_do_sync() got signal ... exiting\n");
3937                         flush_signals(current);
3938                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3939                         goto out;
3940                 }
3941
3942                 /*
3943                  * this loop exits only if either when we are slower than
3944                  * the 'hard' speed limit, or the system was IO-idle for
3945                  * a jiffy.
3946                  * the system might be non-idle CPU-wise, but we only care
3947                  * about not overloading the IO subsystem. (things like an
3948                  * e2fsck being done on the RAID array should execute fast)
3949                  */
3950                 mddev->queue->unplug_fn(mddev->queue);
3951                 cond_resched();
3952
3953                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3954                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
3955
3956                 if (currspeed > sysctl_speed_limit_min) {
3957                         if ((currspeed > sysctl_speed_limit_max) ||
3958                                         !is_mddev_idle(mddev)) {
3959                                 msleep_interruptible(250);
3960                                 goto repeat;
3961                         }
3962                 }
3963         }
3964         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3965         /*
3966          * this also signals 'finished resyncing' to md_stop
3967          */
3968  out:
3969         mddev->queue->unplug_fn(mddev->queue);
3970
3971         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3972
3973         /* tell personality that we are finished */
3974         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3975
3976         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3977             mddev->curr_resync > 2 &&
3978             mddev->curr_resync >= mddev->recovery_cp) {
3979                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3980                         printk(KERN_INFO
3981                                 "md: checkpointing recovery of %s.\n",
3982                                 mdname(mddev));
3983                         mddev->recovery_cp = mddev->curr_resync;
3984                 } else
3985                         mddev->recovery_cp = MaxSector;
3986         }
3987
3988  skip:
3989         mddev->curr_resync = 0;
3990         wake_up(&resync_wait);
3991         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3992         md_wakeup_thread(mddev->thread);
3993 }
3994
3995
3996 /*
3997  * This routine is regularly called by all per-raid-array threads to
3998  * deal with generic issues like resync and super-block update.
3999  * Raid personalities that don't have a thread (linear/raid0) do not
4000  * need this as they never do any recovery or update the superblock.
4001  *
4002  * It does not do any resync itself, but rather "forks" off other threads
4003  * to do that as needed.
4004  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4005  * "->recovery" and create a thread at ->sync_thread.
4006  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4007  * and wakeups up this thread which will reap the thread and finish up.
4008  * This thread also removes any faulty devices (with nr_pending == 0).
4009  *
4010  * The overall approach is:
4011  *  1/ if the superblock needs updating, update it.
4012  *  2/ If a recovery thread is running, don't do anything else.
4013  *  3/ If recovery has finished, clean up, possibly marking spares active.
4014  *  4/ If there are any faulty devices, remove them.
4015  *  5/ If array is degraded, try to add spares devices
4016  *  6/ If array has spares or is not in-sync, start a resync thread.
4017  */
4018 void md_check_recovery(mddev_t *mddev)
4019 {
4020         mdk_rdev_t *rdev;
4021         struct list_head *rtmp;
4022
4023
4024         if (mddev->bitmap)
4025                 bitmap_daemon_work(mddev->bitmap);
4026
4027         if (mddev->ro)
4028                 return;
4029
4030         if (signal_pending(current)) {
4031                 if (mddev->pers->sync_request) {
4032                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4033                                mdname(mddev));
4034                         mddev->safemode = 2;
4035                 }
4036                 flush_signals(current);
4037         }
4038
4039         if ( ! (
4040                 mddev->sb_dirty ||
4041                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4042                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4043                 (mddev->safemode == 1) ||
4044                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4045                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4046                 ))
4047                 return;
4048
4049         if (mddev_trylock(mddev)==0) {
4050                 int spares =0;
4051
4052                 spin_lock(&mddev->write_lock);
4053                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4054                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4055                         mddev->in_sync = 1;
4056                         mddev->sb_dirty = 1;
4057                 }
4058                 if (mddev->safemode == 1)
4059                         mddev->safemode = 0;
4060                 spin_unlock(&mddev->write_lock);
4061
4062                 if (mddev->sb_dirty)
4063                         md_update_sb(mddev);
4064
4065
4066                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4067                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4068                         /* resync/recovery still happening */
4069                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4070                         goto unlock;
4071                 }
4072                 if (mddev->sync_thread) {
4073                         /* resync has finished, collect result */
4074                         md_unregister_thread(mddev->sync_thread);
4075                         mddev->sync_thread = NULL;
4076                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4077                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4078                                 /* success...*/
4079                                 /* activate any spares */
4080                                 mddev->pers->spare_active(mddev);
4081                         }
4082                         md_update_sb(mddev);
4083
4084                         /* if array is no-longer degraded, then any saved_raid_disk
4085                          * information must be scrapped
4086                          */
4087                         if (!mddev->degraded)
4088                                 ITERATE_RDEV(mddev,rdev,rtmp)
4089                                         rdev->saved_raid_disk = -1;
4090
4091                         mddev->recovery = 0;
4092                         /* flag recovery needed just to double check */
4093                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4094                         goto unlock;
4095                 }
4096                 if (mddev->recovery)
4097                         /* probably just the RECOVERY_NEEDED flag */
4098                         mddev->recovery = 0;
4099
4100                 /* no recovery is running.
4101                  * remove any failed drives, then
4102                  * add spares if possible.
4103                  * Spare are also removed and re-added, to allow
4104                  * the personality to fail the re-add.
4105                  */
4106                 ITERATE_RDEV(mddev,rdev,rtmp)
4107                         if (rdev->raid_disk >= 0 &&
4108                             (rdev->faulty || ! rdev->in_sync) &&
4109                             atomic_read(&rdev->nr_pending)==0) {
4110                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4111                                         char nm[20];
4112                                         sprintf(nm,"rd%d", rdev->raid_disk);
4113                                         sysfs_remove_link(&mddev->kobj, nm);
4114                                         rdev->raid_disk = -1;
4115                                 }
4116                         }
4117
4118                 if (mddev->degraded) {
4119                         ITERATE_RDEV(mddev,rdev,rtmp)
4120                                 if (rdev->raid_disk < 0
4121                                     && !rdev->faulty) {
4122                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4123                                                 char nm[20];
4124                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4125                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4126                                                 spares++;
4127                                         } else
4128                                                 break;
4129                                 }
4130                 }
4131
4132                 if (!spares && (mddev->recovery_cp == MaxSector )) {
4133                         /* nothing we can do ... */
4134                         goto unlock;
4135                 }
4136                 if (mddev->pers->sync_request) {
4137                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4138                         if (!spares)
4139                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4140                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4141                                 /* We are adding a device or devices to an array
4142                                  * which has the bitmap stored on all devices.
4143                                  * So make sure all bitmap pages get written
4144                                  */
4145                                 bitmap_write_all(mddev->bitmap);
4146                         }
4147                         mddev->sync_thread = md_register_thread(md_do_sync,
4148                                                                 mddev,
4149                                                                 "%s_resync");
4150                         if (!mddev->sync_thread) {
4151                                 printk(KERN_ERR "%s: could not start resync"
4152                                         " thread...\n",
4153                                         mdname(mddev));
4154                                 /* leave the spares where they are, it shouldn't hurt */
4155                                 mddev->recovery = 0;
4156                         } else {
4157                                 md_wakeup_thread(mddev->sync_thread);
4158                         }
4159                 }
4160         unlock:
4161                 mddev_unlock(mddev);
4162         }
4163 }
4164
4165 static int md_notify_reboot(struct notifier_block *this,
4166                             unsigned long code, void *x)
4167 {
4168         struct list_head *tmp;
4169         mddev_t *mddev;
4170
4171         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4172
4173                 printk(KERN_INFO "md: stopping all md devices.\n");
4174
4175                 ITERATE_MDDEV(mddev,tmp)
4176                         if (mddev_trylock(mddev)==0)
4177                                 do_md_stop (mddev, 1);
4178                 /*
4179                  * certain more exotic SCSI devices are known to be
4180                  * volatile wrt too early system reboots. While the
4181                  * right place to handle this issue is the given
4182                  * driver, we do want to have a safe RAID driver ...
4183                  */
4184                 mdelay(1000*1);
4185         }
4186         return NOTIFY_DONE;
4187 }
4188
4189 static struct notifier_block md_notifier = {
4190         .notifier_call  = md_notify_reboot,
4191         .next           = NULL,
4192         .priority       = INT_MAX, /* before any real devices */
4193 };
4194
4195 static void md_geninit(void)
4196 {
4197         struct proc_dir_entry *p;
4198
4199         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4200
4201         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4202         if (p)
4203                 p->proc_fops = &md_seq_fops;
4204 }
4205
4206 static int __init md_init(void)
4207 {
4208         int minor;
4209
4210         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4211                         " MD_SB_DISKS=%d\n",
4212                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4213                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4214         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
4215                         BITMAP_MINOR);
4216
4217         if (register_blkdev(MAJOR_NR, "md"))
4218                 return -1;
4219         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4220                 unregister_blkdev(MAJOR_NR, "md");
4221                 return -1;
4222         }
4223         devfs_mk_dir("md");
4224         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4225                                 md_probe, NULL, NULL);
4226         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4227                             md_probe, NULL, NULL);
4228
4229         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4230                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4231                                 S_IFBLK|S_IRUSR|S_IWUSR,
4232                                 "md/%d", minor);
4233
4234         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4235                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4236                               S_IFBLK|S_IRUSR|S_IWUSR,
4237                               "md/mdp%d", minor);
4238
4239
4240         register_reboot_notifier(&md_notifier);
4241         raid_table_header = register_sysctl_table(raid_root_table, 1);
4242
4243         md_geninit();
4244         return (0);
4245 }
4246
4247
4248 #ifndef MODULE
4249
4250 /*
4251  * Searches all registered partitions for autorun RAID arrays
4252  * at boot time.
4253  */
4254 static dev_t detected_devices[128];
4255 static int dev_cnt;
4256
4257 void md_autodetect_dev(dev_t dev)
4258 {
4259         if (dev_cnt >= 0 && dev_cnt < 127)
4260                 detected_devices[dev_cnt++] = dev;
4261 }
4262
4263
4264 static void autostart_arrays(int part)
4265 {
4266         mdk_rdev_t *rdev;
4267         int i;
4268
4269         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4270
4271         for (i = 0; i < dev_cnt; i++) {
4272                 dev_t dev = detected_devices[i];
4273
4274                 rdev = md_import_device(dev,0, 0);
4275                 if (IS_ERR(rdev))
4276                         continue;
4277
4278                 if (rdev->faulty) {
4279                         MD_BUG();
4280                         continue;
4281                 }
4282                 list_add(&rdev->same_set, &pending_raid_disks);
4283         }
4284         dev_cnt = 0;
4285
4286         autorun_devices(part);
4287 }
4288
4289 #endif
4290
4291 static __exit void md_exit(void)
4292 {
4293         mddev_t *mddev;
4294         struct list_head *tmp;
4295         int i;
4296         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4297         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4298         for (i=0; i < MAX_MD_DEVS; i++)
4299                 devfs_remove("md/%d", i);
4300         for (i=0; i < MAX_MD_DEVS; i++)
4301                 devfs_remove("md/d%d", i);
4302
4303         devfs_remove("md");
4304
4305         unregister_blkdev(MAJOR_NR,"md");
4306         unregister_blkdev(mdp_major, "mdp");
4307         unregister_reboot_notifier(&md_notifier);
4308         unregister_sysctl_table(raid_table_header);
4309         remove_proc_entry("mdstat", NULL);
4310         ITERATE_MDDEV(mddev,tmp) {
4311                 struct gendisk *disk = mddev->gendisk;
4312                 if (!disk)
4313                         continue;
4314                 export_array(mddev);
4315                 del_gendisk(disk);
4316                 put_disk(disk);
4317                 mddev->gendisk = NULL;
4318                 mddev_put(mddev);
4319         }
4320 }
4321
4322 module_init(md_init)
4323 module_exit(md_exit)
4324
4325 EXPORT_SYMBOL(register_md_personality);
4326 EXPORT_SYMBOL(unregister_md_personality);
4327 EXPORT_SYMBOL(md_error);
4328 EXPORT_SYMBOL(md_done_sync);
4329 EXPORT_SYMBOL(md_write_start);
4330 EXPORT_SYMBOL(md_write_end);
4331 EXPORT_SYMBOL(md_register_thread);
4332 EXPORT_SYMBOL(md_unregister_thread);
4333 EXPORT_SYMBOL(md_wakeup_thread);
4334 EXPORT_SYMBOL(md_print_devices);
4335 EXPORT_SYMBOL(md_check_recovery);
4336 MODULE_LICENSE("GPL");
4337 MODULE_ALIAS("md");
4338 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);