Pileus Git - ~andy/linux/blob - drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/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->flags = 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                 desc = sb->disks + rdev->desc_nr;
 675
 676                 if (desc->state & (1<<MD_DISK_FAULTY))
 677                         set_bit(Faulty, &rdev->flags);
 678                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 679                          desc->raid_disk < mddev->raid_disks) {
 680                         set_bit(In_sync, &rdev->flags);
 681                         rdev->raid_disk = desc->raid_disk;
 682                 }
 683                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 684                         set_bit(WriteMostly, &rdev->flags);
 685         } else /* MULTIPATH are always insync */
 686                 set_bit(In_sync, &rdev->flags);
 687         return 0;
 688 }
 689
 690 /*
 691  * sync_super for 0.90.0
 692  */
 693 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 694 {
 695         mdp_super_t *sb;
 696         struct list_head *tmp;
 697         mdk_rdev_t *rdev2;
 698         int next_spare = mddev->raid_disks;
 699         char nm[20];
 700
 701         /* make rdev->sb match mddev data..
 702          *
 703          * 1/ zero out disks
 704          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 705          * 3/ any empty disks < next_spare become removed
 706          *
 707          * disks[0] gets initialised to REMOVED because
 708          * we cannot be sure from other fields if it has
 709          * been initialised or not.
 710          */
 711         int i;
 712         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 713         unsigned int fixdesc=0;
 714
 715         rdev->sb_size = MD_SB_BYTES;
 716
 717         sb = (mdp_super_t*)page_address(rdev->sb_page);
 718
 719         memset(sb, 0, sizeof(*sb));
 720
 721         sb->md_magic = MD_SB_MAGIC;
 722         sb->major_version = mddev->major_version;
 723         sb->minor_version = mddev->minor_version;
 724         sb->patch_version = mddev->patch_version;
 725         sb->gvalid_words  = 0; /* ignored */
 726         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 727         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 728         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 729         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 730
 731         sb->ctime = mddev->ctime;
 732         sb->level = mddev->level;
 733         sb->size  = mddev->size;
 734         sb->raid_disks = mddev->raid_disks;
 735         sb->md_minor = mddev->md_minor;
 736         sb->not_persistent = !mddev->persistent;
 737         sb->utime = mddev->utime;
 738         sb->state = 0;
 739         sb->events_hi = (mddev->events>>32);
 740         sb->events_lo = (u32)mddev->events;
 741
 742         if (mddev->in_sync)
 743         {
 744                 sb->recovery_cp = mddev->recovery_cp;
 745                 sb->cp_events_hi = (mddev->events>>32);
 746                 sb->cp_events_lo = (u32)mddev->events;
 747                 if (mddev->recovery_cp == MaxSector)
 748                         sb->state = (1<< MD_SB_CLEAN);
 749         } else
 750                 sb->recovery_cp = 0;
 751
 752         sb->layout = mddev->layout;
 753         sb->chunk_size = mddev->chunk_size;
 754
 755         if (mddev->bitmap && mddev->bitmap_file == NULL)
 756                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 757
 758         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 759         ITERATE_RDEV(mddev,rdev2,tmp) {
 760                 mdp_disk_t *d;
 761                 int desc_nr;
 762                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 763                     && !test_bit(Faulty, &rdev2->flags))
 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                                 sprintf(nm, "rd%d", rdev2->raid_disk);
 772                                 sysfs_remove_link(&mddev->kobj, nm);
 773                         }
 774                         sysfs_remove_link(&rdev2->kobj, "block");
 775                         kobject_del(&rdev2->kobj);
 776                 }
 777                 d = &sb->disks[rdev2->desc_nr];
 778                 nr_disks++;
 779                 d->number = rdev2->desc_nr;
 780                 d->major = MAJOR(rdev2->bdev->bd_dev);
 781                 d->minor = MINOR(rdev2->bdev->bd_dev);
 782                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 783                     && !test_bit(Faulty, &rdev2->flags))
 784                         d->raid_disk = rdev2->raid_disk;
 785                 else
 786                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 787                 if (test_bit(Faulty, &rdev2->flags)) {
 788                         d->state = (1<<MD_DISK_FAULTY);
 789                         failed++;
 790                 } else if (test_bit(In_sync, &rdev2->flags)) {
 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 gets a ref on the parent, so
 809                                  * we have to drop the one we already have
 810                                  */
 811                                 kobject_add(&rdev2->kobj);
 812                                 kobject_put(rdev->kobj.parent);
 813                                 sysfs_create_link(&rdev2->kobj,
 814                                                   &rdev2->bdev->bd_disk->kobj,
 815                                                   "block");
 816                                 if (rdev2->raid_disk >= 0) {
 817                                         sprintf(nm, "rd%d", rdev2->raid_disk);
 818                                         sysfs_create_link(&mddev->kobj,
 819                                                           &rdev2->kobj, nm);
 820                                 }
 821                         }
 822         /* now set the "removed" and "faulty" bits on any missing devices */
 823         for (i=0 ; i < mddev->raid_disks ; i++) {
 824                 mdp_disk_t *d = &sb->disks[i];
 825                 if (d->state == 0 && d->number == 0) {
 826                         d->number = i;
 827                         d->raid_disk = i;
 828                         d->state = (1<<MD_DISK_REMOVED);
 829                         d->state |= (1<<MD_DISK_FAULTY);
 830                         failed++;
 831                 }
 832         }
 833         sb->nr_disks = nr_disks;
 834         sb->active_disks = active;
 835         sb->working_disks = working;
 836         sb->failed_disks = failed;
 837         sb->spare_disks = spare;
 838
 839         sb->this_disk = sb->disks[rdev->desc_nr];
 840         sb->sb_csum = calc_sb_csum(sb);
 841 }
 842
 843 /*
 844  * version 1 superblock
 845  */
 846
 847 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 848 {
 849         unsigned int disk_csum, csum;
 850         unsigned long long newcsum;
 851         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 852         unsigned int *isuper = (unsigned int*)sb;
 853         int i;
 854
 855         disk_csum = sb->sb_csum;
 856         sb->sb_csum = 0;
 857         newcsum = 0;
 858         for (i=0; size>=4; size -= 4 )
 859                 newcsum += le32_to_cpu(*isuper++);
 860
 861         if (size == 2)
 862                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 863
 864         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 865         sb->sb_csum = disk_csum;
 866         return cpu_to_le32(csum);
 867 }
 868
 869 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 870 {
 871         struct mdp_superblock_1 *sb;
 872         int ret;
 873         sector_t sb_offset;
 874         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 875         int bmask;
 876
 877         /*
 878          * Calculate the position of the superblock.
 879          * It is always aligned to a 4K boundary and
 880          * depeding on minor_version, it can be:
 881          * 0: At least 8K, but less than 12K, from end of device
 882          * 1: At start of device
 883          * 2: 4K from start of device.
 884          */
 885         switch(minor_version) {
 886         case 0:
 887                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 888                 sb_offset -= 8*2;
 889                 sb_offset &= ~(sector_t)(4*2-1);
 890                 /* convert from sectors to K */
 891                 sb_offset /= 2;
 892                 break;
 893         case 1:
 894                 sb_offset = 0;
 895                 break;
 896         case 2:
 897                 sb_offset = 4;
 898                 break;
 899         default:
 900                 return -EINVAL;
 901         }
 902         rdev->sb_offset = sb_offset;
 903
 904         /* superblock is rarely larger than 1K, but it can be larger,
 905          * and it is safe to read 4k, so we do that
 906          */
 907         ret = read_disk_sb(rdev, 4096);
 908         if (ret) return ret;
 909
 910
 911         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 912
 913         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 914             sb->major_version != cpu_to_le32(1) ||
 915             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 916             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 917             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 918                 return -EINVAL;
 919
 920         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 921                 printk("md: invalid superblock checksum on %s\n",
 922                         bdevname(rdev->bdev,b));
 923                 return -EINVAL;
 924         }
 925         if (le64_to_cpu(sb->data_size) < 10) {
 926                 printk("md: data_size too small on %s\n",
 927                        bdevname(rdev->bdev,b));
 928                 return -EINVAL;
 929         }
 930         rdev->preferred_minor = 0xffff;
 931         rdev->data_offset = le64_to_cpu(sb->data_offset);
 932
 933         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 934         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 935         if (rdev->sb_size & bmask)
 936                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 937
 938         if (refdev == 0)
 939                 return 1;
 940         else {
 941                 __u64 ev1, ev2;
 942                 struct mdp_superblock_1 *refsb =
 943                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 944
 945                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 946                     sb->level != refsb->level ||
 947                     sb->layout != refsb->layout ||
 948                     sb->chunksize != refsb->chunksize) {
 949                         printk(KERN_WARNING "md: %s has strangely different"
 950                                 " superblock to %s\n",
 951                                 bdevname(rdev->bdev,b),
 952                                 bdevname(refdev->bdev,b2));
 953                         return -EINVAL;
 954                 }
 955                 ev1 = le64_to_cpu(sb->events);
 956                 ev2 = le64_to_cpu(refsb->events);
 957
 958                 if (ev1 > ev2)
 959                         return 1;
 960         }
 961         if (minor_version)
 962                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 963         else
 964                 rdev->size = rdev->sb_offset;
 965         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 966                 return -EINVAL;
 967         rdev->size = le64_to_cpu(sb->data_size)/2;
 968         if (le32_to_cpu(sb->chunksize))
 969                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 970         return 0;
 971 }
 972
 973 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 974 {
 975         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 976
 977         rdev->raid_disk = -1;
 978         rdev->flags = 0;
 979         if (mddev->raid_disks == 0) {
 980                 mddev->major_version = 1;
 981                 mddev->patch_version = 0;
 982                 mddev->persistent = 1;
 983                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 984                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 985                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 986                 mddev->level = le32_to_cpu(sb->level);
 987                 mddev->layout = le32_to_cpu(sb->layout);
 988                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 989                 mddev->size = le64_to_cpu(sb->size)/2;
 990                 mddev->events = le64_to_cpu(sb->events);
 991                 mddev->bitmap_offset = 0;
 992                 mddev->default_bitmap_offset = 0;
 993                 mddev->default_bitmap_offset = 1024;
 994
 995                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 996                 memcpy(mddev->uuid, sb->set_uuid, 16);
 997
 998                 mddev->max_disks =  (4096-256)/2;
 999
1000                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1001                     mddev->bitmap_file == NULL ) {
1002                         if (mddev->level != 1) {
1003                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1004                                 return -EINVAL;
1005                         }
1006                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1007                 }
1008         } else if (mddev->pers == NULL) {
1009                 /* Insist of good event counter while assembling */
1010                 __u64 ev1 = le64_to_cpu(sb->events);
1011                 ++ev1;
1012                 if (ev1 < mddev->events)
1013                         return -EINVAL;
1014         } else if (mddev->bitmap) {
1015                 /* If adding to array with a bitmap, then we can accept an
1016                  * older device, but not too old.
1017                  */
1018                 __u64 ev1 = le64_to_cpu(sb->events);
1019                 if (ev1 < mddev->bitmap->events_cleared)
1020                         return 0;
1021         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1022                 return 0;
1023
1024         if (mddev->level != LEVEL_MULTIPATH) {
1025                 int role;
1026                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1027                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1028                 switch(role) {
1029                 case 0xffff: /* spare */
1030                         break;
1031                 case 0xfffe: /* faulty */
1032                         set_bit(Faulty, &rdev->flags);
1033                         break;
1034                 default:
1035                         set_bit(In_sync, &rdev->flags);
1036                         rdev->raid_disk = role;
1037                         break;
1038                 }
1039                 if (sb->devflags & WriteMostly1)
1040                         set_bit(WriteMostly, &rdev->flags);
1041         } else /* MULTIPATH are always insync */
1042                 set_bit(In_sync, &rdev->flags);
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 (test_bit(Faulty, &rdev2->flags))
1087                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1088                 else if (test_bit(In_sync, &rdev2->flags))
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 = &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                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1328                 rdev->desc_nr);
1329         if (rdev->sb_loaded) {
1330                 printk(KERN_INFO "md: rdev superblock:\n");
1331                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1332         } else
1333                 printk(KERN_INFO "md: no rdev superblock!\n");
1334 }
1335
1336 void md_print_devices(void)
1337 {
1338         struct list_head *tmp, *tmp2;
1339         mdk_rdev_t *rdev;
1340         mddev_t *mddev;
1341         char b[BDEVNAME_SIZE];
1342
1343         printk("\n");
1344         printk("md:     **********************************\n");
1345         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1346         printk("md:     **********************************\n");
1347         ITERATE_MDDEV(mddev,tmp) {
1348
1349                 if (mddev->bitmap)
1350                         bitmap_print_sb(mddev->bitmap);
1351                 else
1352                         printk("%s: ", mdname(mddev));
1353                 ITERATE_RDEV(mddev,rdev,tmp2)
1354                         printk("<%s>", bdevname(rdev->bdev,b));
1355                 printk("\n");
1356
1357                 ITERATE_RDEV(mddev,rdev,tmp2)
1358                         print_rdev(rdev);
1359         }
1360         printk("md:     **********************************\n");
1361         printk("\n");
1362 }
1363
1364
1365 static void sync_sbs(mddev_t * mddev)
1366 {
1367         mdk_rdev_t *rdev;
1368         struct list_head *tmp;
1369
1370         ITERATE_RDEV(mddev,rdev,tmp) {
1371                 super_types[mddev->major_version].
1372                         sync_super(mddev, rdev);
1373                 rdev->sb_loaded = 1;
1374         }
1375 }
1376
1377 static void md_update_sb(mddev_t * mddev)
1378 {
1379         int err;
1380         struct list_head *tmp;
1381         mdk_rdev_t *rdev;
1382         int sync_req;
1383
1384 repeat:
1385         spin_lock(&mddev->write_lock);
1386         sync_req = mddev->in_sync;
1387         mddev->utime = get_seconds();
1388         mddev->events ++;
1389
1390         if (!mddev->events) {
1391                 /*
1392                  * oops, this 64-bit counter should never wrap.
1393                  * Either we are in around ~1 trillion A.C., assuming
1394                  * 1 reboot per second, or we have a bug:
1395                  */
1396                 MD_BUG();
1397                 mddev->events --;
1398         }
1399         mddev->sb_dirty = 2;
1400         sync_sbs(mddev);
1401
1402         /*
1403          * do not write anything to disk if using
1404          * nonpersistent superblocks
1405          */
1406         if (!mddev->persistent) {
1407                 mddev->sb_dirty = 0;
1408                 spin_unlock(&mddev->write_lock);
1409                 wake_up(&mddev->sb_wait);
1410                 return;
1411         }
1412         spin_unlock(&mddev->write_lock);
1413
1414         dprintk(KERN_INFO
1415                 "md: updating %s RAID superblock on device (in sync %d)\n",
1416                 mdname(mddev),mddev->in_sync);
1417
1418         err = bitmap_update_sb(mddev->bitmap);
1419         ITERATE_RDEV(mddev,rdev,tmp) {
1420                 char b[BDEVNAME_SIZE];
1421                 dprintk(KERN_INFO "md: ");
1422                 if (test_bit(Faulty, &rdev->flags))
1423                         dprintk("(skipping faulty ");
1424
1425                 dprintk("%s ", bdevname(rdev->bdev,b));
1426                 if (!test_bit(Faulty, &rdev->flags)) {
1427                         md_super_write(mddev,rdev,
1428                                        rdev->sb_offset<<1, rdev->sb_size,
1429                                        rdev->sb_page);
1430                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1431                                 bdevname(rdev->bdev,b),
1432                                 (unsigned long long)rdev->sb_offset);
1433
1434                 } else
1435                         dprintk(")\n");
1436                 if (mddev->level == LEVEL_MULTIPATH)
1437                         /* only need to write one superblock... */
1438                         break;
1439         }
1440         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1441         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1442
1443         spin_lock(&mddev->write_lock);
1444         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1445                 /* have to write it out again */
1446                 spin_unlock(&mddev->write_lock);
1447                 goto repeat;
1448         }
1449         mddev->sb_dirty = 0;
1450         spin_unlock(&mddev->write_lock);
1451         wake_up(&mddev->sb_wait);
1452
1453 }
1454
1455 struct rdev_sysfs_entry {
1456         struct attribute attr;
1457         ssize_t (*show)(mdk_rdev_t *, char *);
1458         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1459 };
1460
1461 static ssize_t
1462 rdev_show_state(mdk_rdev_t *rdev, char *page)
1463 {
1464         char *sep = "";
1465         int len=0;
1466
1467         if (test_bit(Faulty, &rdev->flags)) {
1468                 len+= sprintf(page+len, "%sfaulty",sep);
1469                 sep = ",";
1470         }
1471         if (test_bit(In_sync, &rdev->flags)) {
1472                 len += sprintf(page+len, "%sin_sync",sep);
1473                 sep = ",";
1474         }
1475         if (!test_bit(Faulty, &rdev->flags) &&
1476             !test_bit(In_sync, &rdev->flags)) {
1477                 len += sprintf(page+len, "%sspare", sep);
1478                 sep = ",";
1479         }
1480         return len+sprintf(page+len, "\n");
1481 }
1482
1483 static struct rdev_sysfs_entry rdev_state = {
1484         .attr = {.name = "state", .mode = S_IRUGO },
1485         .show = rdev_show_state,
1486 };
1487
1488 static ssize_t
1489 rdev_show_super(mdk_rdev_t *rdev, char *page)
1490 {
1491         if (rdev->sb_loaded && rdev->sb_size) {
1492                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1493                 return rdev->sb_size;
1494         } else
1495                 return 0;
1496 }
1497 static struct rdev_sysfs_entry rdev_super = {
1498         .attr = {.name = "super", .mode = S_IRUGO },
1499         .show = rdev_show_super,
1500 };
1501 static struct attribute *rdev_default_attrs[] = {
1502         &rdev_state.attr,
1503         &rdev_super.attr,
1504         NULL,
1505 };
1506 static ssize_t
1507 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1508 {
1509         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1510         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1511
1512         if (!entry->show)
1513                 return -EIO;
1514         return entry->show(rdev, page);
1515 }
1516
1517 static ssize_t
1518 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1519               const char *page, size_t length)
1520 {
1521         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1522         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1523
1524         if (!entry->store)
1525                 return -EIO;
1526         return entry->store(rdev, page, length);
1527 }
1528
1529 static void rdev_free(struct kobject *ko)
1530 {
1531         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1532         kfree(rdev);
1533 }
1534 static struct sysfs_ops rdev_sysfs_ops = {
1535         .show           = rdev_attr_show,
1536         .store          = rdev_attr_store,
1537 };
1538 static struct kobj_type rdev_ktype = {
1539         .release        = rdev_free,
1540         .sysfs_ops      = &rdev_sysfs_ops,
1541         .default_attrs  = rdev_default_attrs,
1542 };
1543
1544 /*
1545  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1546  *
1547  * mark the device faulty if:
1548  *
1549  *   - the device is nonexistent (zero size)
1550  *   - the device has no valid superblock
1551  *
1552  * a faulty rdev _never_ has rdev->sb set.
1553  */
1554 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1555 {
1556         char b[BDEVNAME_SIZE];
1557         int err;
1558         mdk_rdev_t *rdev;
1559         sector_t size;
1560
1561         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1562         if (!rdev) {
1563                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1564                 return ERR_PTR(-ENOMEM);
1565         }
1566         memset(rdev, 0, sizeof(*rdev));
1567
1568         if ((err = alloc_disk_sb(rdev)))
1569                 goto abort_free;
1570
1571         err = lock_rdev(rdev, newdev);
1572         if (err)
1573                 goto abort_free;
1574
1575         rdev->kobj.parent = NULL;
1576         rdev->kobj.ktype = &rdev_ktype;
1577         kobject_init(&rdev->kobj);
1578
1579         rdev->desc_nr = -1;
1580         rdev->flags = 0;
1581         rdev->data_offset = 0;
1582         atomic_set(&rdev->nr_pending, 0);
1583         atomic_set(&rdev->read_errors, 0);
1584
1585         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1586         if (!size) {
1587                 printk(KERN_WARNING
1588                         "md: %s has zero or unknown size, marking faulty!\n",
1589                         bdevname(rdev->bdev,b));
1590                 err = -EINVAL;
1591                 goto abort_free;
1592         }
1593
1594         if (super_format >= 0) {
1595                 err = super_types[super_format].
1596                         load_super(rdev, NULL, super_minor);
1597                 if (err == -EINVAL) {
1598                         printk(KERN_WARNING
1599                                 "md: %s has invalid sb, not importing!\n",
1600                                 bdevname(rdev->bdev,b));
1601                         goto abort_free;
1602                 }
1603                 if (err < 0) {
1604                         printk(KERN_WARNING
1605                                 "md: could not read %s's sb, not importing!\n",
1606                                 bdevname(rdev->bdev,b));
1607                         goto abort_free;
1608                 }
1609         }
1610         INIT_LIST_HEAD(&rdev->same_set);
1611
1612         return rdev;
1613
1614 abort_free:
1615         if (rdev->sb_page) {
1616                 if (rdev->bdev)
1617                         unlock_rdev(rdev);
1618                 free_disk_sb(rdev);
1619         }
1620         kfree(rdev);
1621         return ERR_PTR(err);
1622 }
1623
1624 /*
1625  * Check a full RAID array for plausibility
1626  */
1627
1628
1629 static void analyze_sbs(mddev_t * mddev)
1630 {
1631         int i;
1632         struct list_head *tmp;
1633         mdk_rdev_t *rdev, *freshest;
1634         char b[BDEVNAME_SIZE];
1635
1636         freshest = NULL;
1637         ITERATE_RDEV(mddev,rdev,tmp)
1638                 switch (super_types[mddev->major_version].
1639                         load_super(rdev, freshest, mddev->minor_version)) {
1640                 case 1:
1641                         freshest = rdev;
1642                         break;
1643                 case 0:
1644                         break;
1645                 default:
1646                         printk( KERN_ERR \
1647                                 "md: fatal superblock inconsistency in %s"
1648                                 " -- removing from array\n",
1649                                 bdevname(rdev->bdev,b));
1650                         kick_rdev_from_array(rdev);
1651                 }
1652
1653
1654         super_types[mddev->major_version].
1655                 validate_super(mddev, freshest);
1656
1657         i = 0;
1658         ITERATE_RDEV(mddev,rdev,tmp) {
1659                 if (rdev != freshest)
1660                         if (super_types[mddev->major_version].
1661                             validate_super(mddev, rdev)) {
1662                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1663                                         " from array!\n",
1664                                         bdevname(rdev->bdev,b));
1665                                 kick_rdev_from_array(rdev);
1666                                 continue;
1667                         }
1668                 if (mddev->level == LEVEL_MULTIPATH) {
1669                         rdev->desc_nr = i++;
1670                         rdev->raid_disk = rdev->desc_nr;
1671                         set_bit(In_sync, &rdev->flags);
1672                 }
1673         }
1674
1675
1676
1677         if (mddev->recovery_cp != MaxSector &&
1678             mddev->level >= 1)
1679                 printk(KERN_ERR "md: %s: raid array is not clean"
1680                        " -- starting background reconstruction\n",
1681                        mdname(mddev));
1682
1683 }
1684
1685 static ssize_t
1686 md_show_level(mddev_t *mddev, char *page)
1687 {
1688         mdk_personality_t *p = mddev->pers;
1689         if (p == NULL)
1690                 return 0;
1691         if (mddev->level >= 0)
1692                 return sprintf(page, "RAID-%d\n", mddev->level);
1693         else
1694                 return sprintf(page, "%s\n", p->name);
1695 }
1696
1697 static struct md_sysfs_entry md_level = {
1698         .attr = {.name = "level", .mode = S_IRUGO },
1699         .show = md_show_level,
1700 };
1701
1702 static ssize_t
1703 md_show_rdisks(mddev_t *mddev, char *page)
1704 {
1705         return sprintf(page, "%d\n", mddev->raid_disks);
1706 }
1707
1708 static struct md_sysfs_entry md_raid_disks = {
1709         .attr = {.name = "raid_disks", .mode = S_IRUGO },
1710         .show = md_show_rdisks,
1711 };
1712
1713 static ssize_t
1714 md_show_scan(mddev_t *mddev, char *page)
1715 {
1716         char *type = "none";
1717         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1718             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1719                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1720                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1721                                 type = "resync";
1722                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1723                                 type = "check";
1724                         else
1725                                 type = "repair";
1726                 } else
1727                         type = "recover";
1728         }
1729         return sprintf(page, "%s\n", type);
1730 }
1731
1732 static ssize_t
1733 md_store_scan(mddev_t *mddev, const char *page, size_t len)
1734 {
1735         int canscan=0;
1736
1737         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1738             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1739                 return -EBUSY;
1740         down(&mddev->reconfig_sem);
1741         if (mddev->pers && mddev->pers->sync_request)
1742                 canscan=1;
1743         up(&mddev->reconfig_sem);
1744         if (!canscan)
1745                 return -EINVAL;
1746
1747         if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1748                 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1749         else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1750                 return -EINVAL;
1751         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1752         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1753         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1754         md_wakeup_thread(mddev->thread);
1755         return len;
1756 }
1757
1758 static ssize_t
1759 md_show_mismatch(mddev_t *mddev, char *page)
1760 {
1761         return sprintf(page, "%llu\n",
1762                        (unsigned long long) mddev->resync_mismatches);
1763 }
1764
1765 static struct md_sysfs_entry md_scan_mode = {
1766         .attr = {.name = "scan_mode", .mode = S_IRUGO|S_IWUSR },
1767         .show = md_show_scan,
1768         .store = md_store_scan,
1769 };
1770
1771 static struct md_sysfs_entry md_mismatches = {
1772         .attr = {.name = "mismatch_cnt", .mode = S_IRUGO },
1773         .show = md_show_mismatch,
1774 };
1775
1776 static struct attribute *md_default_attrs[] = {
1777         &md_level.attr,
1778         &md_raid_disks.attr,
1779         &md_scan_mode.attr,
1780         &md_mismatches.attr,
1781         NULL,
1782 };
1783
1784 static ssize_t
1785 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1786 {
1787         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1788         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1789
1790         if (!entry->show)
1791                 return -EIO;
1792         return entry->show(mddev, page);
1793 }
1794
1795 static ssize_t
1796 md_attr_store(struct kobject *kobj, struct attribute *attr,
1797               const char *page, size_t length)
1798 {
1799         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1800         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1801
1802         if (!entry->store)
1803                 return -EIO;
1804         return entry->store(mddev, page, length);
1805 }
1806
1807 static void md_free(struct kobject *ko)
1808 {
1809         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1810         kfree(mddev);
1811 }
1812
1813 static struct sysfs_ops md_sysfs_ops = {
1814         .show   = md_attr_show,
1815         .store  = md_attr_store,
1816 };
1817 static struct kobj_type md_ktype = {
1818         .release        = md_free,
1819         .sysfs_ops      = &md_sysfs_ops,
1820         .default_attrs  = md_default_attrs,
1821 };
1822
1823 int mdp_major = 0;
1824
1825 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1826 {
1827         static DECLARE_MUTEX(disks_sem);
1828         mddev_t *mddev = mddev_find(dev);
1829         struct gendisk *disk;
1830         int partitioned = (MAJOR(dev) != MD_MAJOR);
1831         int shift = partitioned ? MdpMinorShift : 0;
1832         int unit = MINOR(dev) >> shift;
1833
1834         if (!mddev)
1835                 return NULL;
1836
1837         down(&disks_sem);
1838         if (mddev->gendisk) {
1839                 up(&disks_sem);
1840                 mddev_put(mddev);
1841                 return NULL;
1842         }
1843         disk = alloc_disk(1 << shift);
1844         if (!disk) {
1845                 up(&disks_sem);
1846                 mddev_put(mddev);
1847                 return NULL;
1848         }
1849         disk->major = MAJOR(dev);
1850         disk->first_minor = unit << shift;
1851         if (partitioned) {
1852                 sprintf(disk->disk_name, "md_d%d", unit);
1853                 sprintf(disk->devfs_name, "md/d%d", unit);
1854         } else {
1855                 sprintf(disk->disk_name, "md%d", unit);
1856                 sprintf(disk->devfs_name, "md/%d", unit);
1857         }
1858         disk->fops = &md_fops;
1859         disk->private_data = mddev;
1860         disk->queue = mddev->queue;
1861         add_disk(disk);
1862         mddev->gendisk = disk;
1863         up(&disks_sem);
1864         mddev->kobj.parent = &disk->kobj;
1865         mddev->kobj.k_name = NULL;
1866         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1867         mddev->kobj.ktype = &md_ktype;
1868         kobject_register(&mddev->kobj);
1869         return NULL;
1870 }
1871
1872 void md_wakeup_thread(mdk_thread_t *thread);
1873
1874 static void md_safemode_timeout(unsigned long data)
1875 {
1876         mddev_t *mddev = (mddev_t *) data;
1877
1878         mddev->safemode = 1;
1879         md_wakeup_thread(mddev->thread);
1880 }
1881
1882
1883 static int do_md_run(mddev_t * mddev)
1884 {
1885         int pnum, err;
1886         int chunk_size;
1887         struct list_head *tmp;
1888         mdk_rdev_t *rdev;
1889         struct gendisk *disk;
1890         char b[BDEVNAME_SIZE];
1891
1892         if (list_empty(&mddev->disks))
1893                 /* cannot run an array with no devices.. */
1894                 return -EINVAL;
1895
1896         if (mddev->pers)
1897                 return -EBUSY;
1898
1899         /*
1900          * Analyze all RAID superblock(s)
1901          */
1902         if (!mddev->raid_disks)
1903                 analyze_sbs(mddev);
1904
1905         chunk_size = mddev->chunk_size;
1906         pnum = level_to_pers(mddev->level);
1907
1908         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1909                 if (!chunk_size) {
1910                         /*
1911                          * 'default chunksize' in the old md code used to
1912                          * be PAGE_SIZE, baaad.
1913                          * we abort here to be on the safe side. We don't
1914                          * want to continue the bad practice.
1915                          */
1916                         printk(KERN_ERR
1917                                 "no chunksize specified, see 'man raidtab'\n");
1918                         return -EINVAL;
1919                 }
1920                 if (chunk_size > MAX_CHUNK_SIZE) {
1921                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1922                                 chunk_size, MAX_CHUNK_SIZE);
1923                         return -EINVAL;
1924                 }
1925                 /*
1926                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1927                  */
1928                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1929                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1930                         return -EINVAL;
1931                 }
1932                 if (chunk_size < PAGE_SIZE) {
1933                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1934                                 chunk_size, PAGE_SIZE);
1935                         return -EINVAL;
1936                 }
1937
1938                 /* devices must have minimum size of one chunk */
1939                 ITERATE_RDEV(mddev,rdev,tmp) {
1940                         if (test_bit(Faulty, &rdev->flags))
1941                                 continue;
1942                         if (rdev->size < chunk_size / 1024) {
1943                                 printk(KERN_WARNING
1944                                         "md: Dev %s smaller than chunk_size:"
1945                                         " %lluk < %dk\n",
1946                                         bdevname(rdev->bdev,b),
1947                                         (unsigned long long)rdev->size,
1948                                         chunk_size / 1024);
1949                                 return -EINVAL;
1950                         }
1951                 }
1952         }
1953
1954 #ifdef CONFIG_KMOD
1955         if (!pers[pnum])
1956         {
1957                 request_module("md-personality-%d", pnum);
1958         }
1959 #endif
1960
1961         /*
1962          * Drop all container device buffers, from now on
1963          * the only valid external interface is through the md
1964          * device.
1965          * Also find largest hardsector size
1966          */
1967         ITERATE_RDEV(mddev,rdev,tmp) {
1968                 if (test_bit(Faulty, &rdev->flags))
1969                         continue;
1970                 sync_blockdev(rdev->bdev);
1971                 invalidate_bdev(rdev->bdev, 0);
1972         }
1973
1974         md_probe(mddev->unit, NULL, NULL);
1975         disk = mddev->gendisk;
1976         if (!disk)
1977                 return -ENOMEM;
1978
1979         spin_lock(&pers_lock);
1980         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1981                 spin_unlock(&pers_lock);
1982                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1983                        pnum);
1984                 return -EINVAL;
1985         }
1986
1987         mddev->pers = pers[pnum];
1988         spin_unlock(&pers_lock);
1989
1990         mddev->recovery = 0;
1991         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1992
1993         /* before we start the array running, initialise the bitmap */
1994         err = bitmap_create(mddev);
1995         if (err)
1996                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1997                         mdname(mddev), err);
1998         else
1999                 err = mddev->pers->run(mddev);
2000         if (err) {
2001                 printk(KERN_ERR "md: pers->run() failed ...\n");
2002                 module_put(mddev->pers->owner);
2003                 mddev->pers = NULL;
2004                 bitmap_destroy(mddev);
2005                 return err;
2006         }
2007         atomic_set(&mddev->writes_pending,0);
2008         mddev->safemode = 0;
2009         mddev->safemode_timer.function = md_safemode_timeout;
2010         mddev->safemode_timer.data = (unsigned long) mddev;
2011         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2012         mddev->in_sync = 1;
2013
2014         ITERATE_RDEV(mddev,rdev,tmp)
2015                 if (rdev->raid_disk >= 0) {
2016                         char nm[20];
2017                         sprintf(nm, "rd%d", rdev->raid_disk);
2018                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2019                 }
2020
2021         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2022         md_wakeup_thread(mddev->thread);
2023
2024         if (mddev->sb_dirty)
2025                 md_update_sb(mddev);
2026
2027         set_capacity(disk, mddev->array_size<<1);
2028
2029         /* If we call blk_queue_make_request here, it will
2030          * re-initialise max_sectors etc which may have been
2031          * refined inside -> run.  So just set the bits we need to set.
2032          * Most initialisation happended when we called
2033          * blk_queue_make_request(..., md_fail_request)
2034          * earlier.
2035          */
2036         mddev->queue->queuedata = mddev;
2037         mddev->queue->make_request_fn = mddev->pers->make_request;
2038
2039         mddev->changed = 1;
2040         return 0;
2041 }
2042
2043 static int restart_array(mddev_t *mddev)
2044 {
2045         struct gendisk *disk = mddev->gendisk;
2046         int err;
2047
2048         /*
2049          * Complain if it has no devices
2050          */
2051         err = -ENXIO;
2052         if (list_empty(&mddev->disks))
2053                 goto out;
2054
2055         if (mddev->pers) {
2056                 err = -EBUSY;
2057                 if (!mddev->ro)
2058                         goto out;
2059
2060                 mddev->safemode = 0;
2061                 mddev->ro = 0;
2062                 set_disk_ro(disk, 0);
2063
2064                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2065                         mdname(mddev));
2066                 /*
2067                  * Kick recovery or resync if necessary
2068                  */
2069                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2070                 md_wakeup_thread(mddev->thread);
2071                 err = 0;
2072         } else {
2073                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2074                         mdname(mddev));
2075                 err = -EINVAL;
2076         }
2077
2078 out:
2079         return err;
2080 }
2081
2082 static int do_md_stop(mddev_t * mddev, int ro)
2083 {
2084         int err = 0;
2085         struct gendisk *disk = mddev->gendisk;
2086
2087         if (mddev->pers) {
2088                 if (atomic_read(&mddev->active)>2) {
2089                         printk("md: %s still in use.\n",mdname(mddev));
2090                         return -EBUSY;
2091                 }
2092
2093                 if (mddev->sync_thread) {
2094                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2095                         md_unregister_thread(mddev->sync_thread);
2096                         mddev->sync_thread = NULL;
2097                 }
2098
2099                 del_timer_sync(&mddev->safemode_timer);
2100
2101                 invalidate_partition(disk, 0);
2102
2103                 if (ro) {
2104                         err  = -ENXIO;
2105                         if (mddev->ro)
2106                                 goto out;
2107                         mddev->ro = 1;
2108                 } else {
2109                         bitmap_flush(mddev);
2110                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
2111                         if (mddev->ro)
2112                                 set_disk_ro(disk, 0);
2113                         blk_queue_make_request(mddev->queue, md_fail_request);
2114                         mddev->pers->stop(mddev);
2115                         module_put(mddev->pers->owner);
2116                         mddev->pers = NULL;
2117                         if (mddev->ro)
2118                                 mddev->ro = 0;
2119                 }
2120                 if (!mddev->in_sync) {
2121                         /* mark array as shutdown cleanly */
2122                         mddev->in_sync = 1;
2123                         md_update_sb(mddev);
2124                 }
2125                 if (ro)
2126                         set_disk_ro(disk, 1);
2127         }
2128
2129         bitmap_destroy(mddev);
2130         if (mddev->bitmap_file) {
2131                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2132                 fput(mddev->bitmap_file);
2133                 mddev->bitmap_file = NULL;
2134         }
2135         mddev->bitmap_offset = 0;
2136
2137         /*
2138          * Free resources if final stop
2139          */
2140         if (!ro) {
2141                 mdk_rdev_t *rdev;
2142                 struct list_head *tmp;
2143                 struct gendisk *disk;
2144                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2145
2146                 ITERATE_RDEV(mddev,rdev,tmp)
2147                         if (rdev->raid_disk >= 0) {
2148                                 char nm[20];
2149                                 sprintf(nm, "rd%d", rdev->raid_disk);
2150                                 sysfs_remove_link(&mddev->kobj, nm);
2151                         }
2152
2153                 export_array(mddev);
2154
2155                 mddev->array_size = 0;
2156                 disk = mddev->gendisk;
2157                 if (disk)
2158                         set_capacity(disk, 0);
2159                 mddev->changed = 1;
2160         } else
2161                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2162                         mdname(mddev));
2163         err = 0;
2164 out:
2165         return err;
2166 }
2167
2168 static void autorun_array(mddev_t *mddev)
2169 {
2170         mdk_rdev_t *rdev;
2171         struct list_head *tmp;
2172         int err;
2173
2174         if (list_empty(&mddev->disks))
2175                 return;
2176
2177         printk(KERN_INFO "md: running: ");
2178
2179         ITERATE_RDEV(mddev,rdev,tmp) {
2180                 char b[BDEVNAME_SIZE];
2181                 printk("<%s>", bdevname(rdev->bdev,b));
2182         }
2183         printk("\n");
2184
2185         err = do_md_run (mddev);
2186         if (err) {
2187                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2188                 do_md_stop (mddev, 0);
2189         }
2190 }
2191
2192 /*
2193  * lets try to run arrays based on all disks that have arrived
2194  * until now. (those are in pending_raid_disks)
2195  *
2196  * the method: pick the first pending disk, collect all disks with
2197  * the same UUID, remove all from the pending list and put them into
2198  * the 'same_array' list. Then order this list based on superblock
2199  * update time (freshest comes first), kick out 'old' disks and
2200  * compare superblocks. If everything's fine then run it.
2201  *
2202  * If "unit" is allocated, then bump its reference count
2203  */
2204 static void autorun_devices(int part)
2205 {
2206         struct list_head candidates;
2207         struct list_head *tmp;
2208         mdk_rdev_t *rdev0, *rdev;
2209         mddev_t *mddev;
2210         char b[BDEVNAME_SIZE];
2211
2212         printk(KERN_INFO "md: autorun ...\n");
2213         while (!list_empty(&pending_raid_disks)) {
2214                 dev_t dev;
2215                 rdev0 = list_entry(pending_raid_disks.next,
2216                                          mdk_rdev_t, same_set);
2217
2218                 printk(KERN_INFO "md: considering %s ...\n",
2219                         bdevname(rdev0->bdev,b));
2220                 INIT_LIST_HEAD(&candidates);
2221                 ITERATE_RDEV_PENDING(rdev,tmp)
2222                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2223                                 printk(KERN_INFO "md:  adding %s ...\n",
2224                                         bdevname(rdev->bdev,b));
2225                                 list_move(&rdev->same_set, &candidates);
2226                         }
2227                 /*
2228                  * now we have a set of devices, with all of them having
2229                  * mostly sane superblocks. It's time to allocate the
2230                  * mddev.
2231                  */
2232                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2233                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2234                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2235                         break;
2236                 }
2237                 if (part)
2238                         dev = MKDEV(mdp_major,
2239                                     rdev0->preferred_minor << MdpMinorShift);
2240                 else
2241                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2242
2243                 md_probe(dev, NULL, NULL);
2244                 mddev = mddev_find(dev);
2245                 if (!mddev) {
2246                         printk(KERN_ERR
2247                                 "md: cannot allocate memory for md drive.\n");
2248                         break;
2249                 }
2250                 if (mddev_lock(mddev))
2251                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2252                                mdname(mddev));
2253                 else if (mddev->raid_disks || mddev->major_version
2254                          || !list_empty(&mddev->disks)) {
2255                         printk(KERN_WARNING
2256                                 "md: %s already running, cannot run %s\n",
2257                                 mdname(mddev), bdevname(rdev0->bdev,b));
2258                         mddev_unlock(mddev);
2259                 } else {
2260                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2261                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2262                                 list_del_init(&rdev->same_set);
2263                                 if (bind_rdev_to_array(rdev, mddev))
2264                                         export_rdev(rdev);
2265                         }
2266                         autorun_array(mddev);
2267                         mddev_unlock(mddev);
2268                 }
2269                 /* on success, candidates will be empty, on error
2270                  * it won't...
2271                  */
2272                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2273                         export_rdev(rdev);
2274                 mddev_put(mddev);
2275         }
2276         printk(KERN_INFO "md: ... autorun DONE.\n");
2277 }
2278
2279 /*
2280  * import RAID devices based on one partition
2281  * if possible, the array gets run as well.
2282  */
2283
2284 static int autostart_array(dev_t startdev)
2285 {
2286         char b[BDEVNAME_SIZE];
2287         int err = -EINVAL, i;
2288         mdp_super_t *sb = NULL;
2289         mdk_rdev_t *start_rdev = NULL, *rdev;
2290
2291         start_rdev = md_import_device(startdev, 0, 0);
2292         if (IS_ERR(start_rdev))
2293                 return err;
2294
2295
2296         /* NOTE: this can only work for 0.90.0 superblocks */
2297         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2298         if (sb->major_version != 0 ||
2299             sb->minor_version != 90 ) {
2300                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2301                 export_rdev(start_rdev);
2302                 return err;
2303         }
2304
2305         if (test_bit(Faulty, &start_rdev->flags)) {
2306                 printk(KERN_WARNING
2307                         "md: can not autostart based on faulty %s!\n",
2308                         bdevname(start_rdev->bdev,b));
2309                 export_rdev(start_rdev);
2310                 return err;
2311         }
2312         list_add(&start_rdev->same_set, &pending_raid_disks);
2313
2314         for (i = 0; i < MD_SB_DISKS; i++) {
2315                 mdp_disk_t *desc = sb->disks + i;
2316                 dev_t dev = MKDEV(desc->major, desc->minor);
2317
2318                 if (!dev)
2319                         continue;
2320                 if (dev == startdev)
2321                         continue;
2322                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2323                         continue;
2324                 rdev = md_import_device(dev, 0, 0);
2325                 if (IS_ERR(rdev))
2326                         continue;
2327
2328                 list_add(&rdev->same_set, &pending_raid_disks);
2329         }
2330
2331         /*
2332          * possibly return codes
2333          */
2334         autorun_devices(0);
2335         return 0;
2336
2337 }
2338
2339
2340 static int get_version(void __user * arg)
2341 {
2342         mdu_version_t ver;
2343
2344         ver.major = MD_MAJOR_VERSION;
2345         ver.minor = MD_MINOR_VERSION;
2346         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2347
2348         if (copy_to_user(arg, &ver, sizeof(ver)))
2349                 return -EFAULT;
2350
2351         return 0;
2352 }
2353
2354 static int get_array_info(mddev_t * mddev, void __user * arg)
2355 {
2356         mdu_array_info_t info;
2357         int nr,working,active,failed,spare;
2358         mdk_rdev_t *rdev;
2359         struct list_head *tmp;
2360
2361         nr=working=active=failed=spare=0;
2362         ITERATE_RDEV(mddev,rdev,tmp) {
2363                 nr++;
2364                 if (test_bit(Faulty, &rdev->flags))
2365                         failed++;
2366                 else {
2367                         working++;
2368                         if (test_bit(In_sync, &rdev->flags))
2369                                 active++;
2370                         else
2371                                 spare++;
2372                 }
2373         }
2374
2375         info.major_version = mddev->major_version;
2376         info.minor_version = mddev->minor_version;
2377         info.patch_version = MD_PATCHLEVEL_VERSION;
2378         info.ctime         = mddev->ctime;
2379         info.level         = mddev->level;
2380         info.size          = mddev->size;
2381         info.nr_disks      = nr;
2382         info.raid_disks    = mddev->raid_disks;
2383         info.md_minor      = mddev->md_minor;
2384         info.not_persistent= !mddev->persistent;
2385
2386         info.utime         = mddev->utime;
2387         info.state         = 0;
2388         if (mddev->in_sync)
2389                 info.state = (1<<MD_SB_CLEAN);
2390         if (mddev->bitmap && mddev->bitmap_offset)
2391                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2392         info.active_disks  = active;
2393         info.working_disks = working;
2394         info.failed_disks  = failed;
2395         info.spare_disks   = spare;
2396
2397         info.layout        = mddev->layout;
2398         info.chunk_size    = mddev->chunk_size;
2399
2400         if (copy_to_user(arg, &info, sizeof(info)))
2401                 return -EFAULT;
2402
2403         return 0;
2404 }
2405
2406 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2407 {
2408         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2409         char *ptr, *buf = NULL;
2410         int err = -ENOMEM;
2411
2412         file = kmalloc(sizeof(*file), GFP_KERNEL);
2413         if (!file)
2414                 goto out;
2415
2416         /* bitmap disabled, zero the first byte and copy out */
2417         if (!mddev->bitmap || !mddev->bitmap->file) {
2418                 file->pathname[0] = '\0';
2419                 goto copy_out;
2420         }
2421
2422         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2423         if (!buf)
2424                 goto out;
2425
2426         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2427         if (!ptr)
2428                 goto out;
2429
2430         strcpy(file->pathname, ptr);
2431
2432 copy_out:
2433         err = 0;
2434         if (copy_to_user(arg, file, sizeof(*file)))
2435                 err = -EFAULT;
2436 out:
2437         kfree(buf);
2438         kfree(file);
2439         return err;
2440 }
2441
2442 static int get_disk_info(mddev_t * mddev, void __user * arg)
2443 {
2444         mdu_disk_info_t info;
2445         unsigned int nr;
2446         mdk_rdev_t *rdev;
2447
2448         if (copy_from_user(&info, arg, sizeof(info)))
2449                 return -EFAULT;
2450
2451         nr = info.number;
2452
2453         rdev = find_rdev_nr(mddev, nr);
2454         if (rdev) {
2455                 info.major = MAJOR(rdev->bdev->bd_dev);
2456                 info.minor = MINOR(rdev->bdev->bd_dev);
2457                 info.raid_disk = rdev->raid_disk;
2458                 info.state = 0;
2459                 if (test_bit(Faulty, &rdev->flags))
2460                         info.state |= (1<<MD_DISK_FAULTY);
2461                 else if (test_bit(In_sync, &rdev->flags)) {
2462                         info.state |= (1<<MD_DISK_ACTIVE);
2463                         info.state |= (1<<MD_DISK_SYNC);
2464                 }
2465                 if (test_bit(WriteMostly, &rdev->flags))
2466                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2467         } else {
2468                 info.major = info.minor = 0;
2469                 info.raid_disk = -1;
2470                 info.state = (1<<MD_DISK_REMOVED);
2471         }
2472
2473         if (copy_to_user(arg, &info, sizeof(info)))
2474                 return -EFAULT;
2475
2476         return 0;
2477 }
2478
2479 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2480 {
2481         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2482         mdk_rdev_t *rdev;
2483         dev_t dev = MKDEV(info->major,info->minor);
2484
2485         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2486                 return -EOVERFLOW;
2487
2488         if (!mddev->raid_disks) {
2489                 int err;
2490                 /* expecting a device which has a superblock */
2491                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2492                 if (IS_ERR(rdev)) {
2493                         printk(KERN_WARNING
2494                                 "md: md_import_device returned %ld\n",
2495                                 PTR_ERR(rdev));
2496                         return PTR_ERR(rdev);
2497                 }
2498                 if (!list_empty(&mddev->disks)) {
2499                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2500                                                         mdk_rdev_t, same_set);
2501                         int err = super_types[mddev->major_version]
2502                                 .load_super(rdev, rdev0, mddev->minor_version);
2503                         if (err < 0) {
2504                                 printk(KERN_WARNING
2505                                         "md: %s has different UUID to %s\n",
2506                                         bdevname(rdev->bdev,b),
2507                                         bdevname(rdev0->bdev,b2));
2508                                 export_rdev(rdev);
2509                                 return -EINVAL;
2510                         }
2511                 }
2512                 err = bind_rdev_to_array(rdev, mddev);
2513                 if (err)
2514                         export_rdev(rdev);
2515                 return err;
2516         }
2517
2518         /*
2519          * add_new_disk can be used once the array is assembled
2520          * to add "hot spares".  They must already have a superblock
2521          * written
2522          */
2523         if (mddev->pers) {
2524                 int err;
2525                 if (!mddev->pers->hot_add_disk) {
2526                         printk(KERN_WARNING
2527                                 "%s: personality does not support diskops!\n",
2528                                mdname(mddev));
2529                         return -EINVAL;
2530                 }
2531                 if (mddev->persistent)
2532                         rdev = md_import_device(dev, mddev->major_version,
2533                                                 mddev->minor_version);
2534                 else
2535                         rdev = md_import_device(dev, -1, -1);
2536                 if (IS_ERR(rdev)) {
2537                         printk(KERN_WARNING
2538                                 "md: md_import_device returned %ld\n",
2539                                 PTR_ERR(rdev));
2540                         return PTR_ERR(rdev);
2541                 }
2542                 /* set save_raid_disk if appropriate */
2543                 if (!mddev->persistent) {
2544                         if (info->state & (1<<MD_DISK_SYNC)  &&
2545                             info->raid_disk < mddev->raid_disks)
2546                                 rdev->raid_disk = info->raid_disk;
2547                         else
2548                                 rdev->raid_disk = -1;
2549                 } else
2550                         super_types[mddev->major_version].
2551                                 validate_super(mddev, rdev);
2552                 rdev->saved_raid_disk = rdev->raid_disk;
2553
2554                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2555                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2556                         set_bit(WriteMostly, &rdev->flags);
2557
2558                 rdev->raid_disk = -1;
2559                 err = bind_rdev_to_array(rdev, mddev);
2560                 if (err)
2561                         export_rdev(rdev);
2562
2563                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2564                 md_wakeup_thread(mddev->thread);
2565                 return err;
2566         }
2567
2568         /* otherwise, add_new_disk is only allowed
2569          * for major_version==0 superblocks
2570          */
2571         if (mddev->major_version != 0) {
2572                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2573                        mdname(mddev));
2574                 return -EINVAL;
2575         }
2576
2577         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2578                 int err;
2579                 rdev = md_import_device (dev, -1, 0);
2580                 if (IS_ERR(rdev)) {
2581                         printk(KERN_WARNING
2582                                 "md: error, md_import_device() returned %ld\n",
2583                                 PTR_ERR(rdev));
2584                         return PTR_ERR(rdev);
2585                 }
2586                 rdev->desc_nr = info->number;
2587                 if (info->raid_disk < mddev->raid_disks)
2588                         rdev->raid_disk = info->raid_disk;
2589                 else
2590                         rdev->raid_disk = -1;
2591
2592                 rdev->flags = 0;
2593
2594                 if (rdev->raid_disk < mddev->raid_disks)
2595                         if (info->state & (1<<MD_DISK_SYNC))
2596                                 set_bit(In_sync, &rdev->flags);
2597
2598                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2599                         set_bit(WriteMostly, &rdev->flags);
2600
2601                 err = bind_rdev_to_array(rdev, mddev);
2602                 if (err) {
2603                         export_rdev(rdev);
2604                         return err;
2605                 }
2606
2607                 if (!mddev->persistent) {
2608                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2609                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2610                 } else
2611                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2612                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2613
2614                 if (!mddev->size || (mddev->size > rdev->size))
2615                         mddev->size = rdev->size;
2616         }
2617
2618         return 0;
2619 }
2620
2621 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2622 {
2623         char b[BDEVNAME_SIZE];
2624         mdk_rdev_t *rdev;
2625
2626         if (!mddev->pers)
2627                 return -ENODEV;
2628
2629         rdev = find_rdev(mddev, dev);
2630         if (!rdev)
2631                 return -ENXIO;
2632
2633         if (rdev->raid_disk >= 0)
2634                 goto busy;
2635
2636         kick_rdev_from_array(rdev);
2637         md_update_sb(mddev);
2638
2639         return 0;
2640 busy:
2641         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2642                 bdevname(rdev->bdev,b), mdname(mddev));
2643         return -EBUSY;
2644 }
2645
2646 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2647 {
2648         char b[BDEVNAME_SIZE];
2649         int err;
2650         unsigned int size;
2651         mdk_rdev_t *rdev;
2652
2653         if (!mddev->pers)
2654                 return -ENODEV;
2655
2656         if (mddev->major_version != 0) {
2657                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2658                         " version-0 superblocks.\n",
2659                         mdname(mddev));
2660                 return -EINVAL;
2661         }
2662         if (!mddev->pers->hot_add_disk) {
2663                 printk(KERN_WARNING
2664                         "%s: personality does not support diskops!\n",
2665                         mdname(mddev));
2666                 return -EINVAL;
2667         }
2668
2669         rdev = md_import_device (dev, -1, 0);
2670         if (IS_ERR(rdev)) {
2671                 printk(KERN_WARNING
2672                         "md: error, md_import_device() returned %ld\n",
2673                         PTR_ERR(rdev));
2674                 return -EINVAL;
2675         }
2676
2677         if (mddev->persistent)
2678                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2679         else
2680                 rdev->sb_offset =
2681                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2682
2683         size = calc_dev_size(rdev, mddev->chunk_size);
2684         rdev->size = size;
2685
2686         if (size < mddev->size) {
2687                 printk(KERN_WARNING
2688                         "%s: disk size %llu blocks < array size %llu\n",
2689                         mdname(mddev), (unsigned long long)size,
2690                         (unsigned long long)mddev->size);
2691                 err = -ENOSPC;
2692                 goto abort_export;
2693         }
2694
2695         if (test_bit(Faulty, &rdev->flags)) {
2696                 printk(KERN_WARNING
2697                         "md: can not hot-add faulty %s disk to %s!\n",
2698                         bdevname(rdev->bdev,b), mdname(mddev));
2699                 err = -EINVAL;
2700                 goto abort_export;
2701         }
2702         clear_bit(In_sync, &rdev->flags);
2703         rdev->desc_nr = -1;
2704         bind_rdev_to_array(rdev, mddev);
2705
2706         /*
2707          * The rest should better be atomic, we can have disk failures
2708          * noticed in interrupt contexts ...
2709          */
2710
2711         if (rdev->desc_nr == mddev->max_disks) {
2712                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2713                         mdname(mddev));
2714                 err = -EBUSY;
2715                 goto abort_unbind_export;
2716         }
2717
2718         rdev->raid_disk = -1;
2719
2720         md_update_sb(mddev);
2721
2722         /*
2723          * Kick recovery, maybe this spare has to be added to the
2724          * array immediately.
2725          */
2726         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2727         md_wakeup_thread(mddev->thread);
2728
2729         return 0;
2730
2731 abort_unbind_export:
2732         unbind_rdev_from_array(rdev);
2733
2734 abort_export:
2735         export_rdev(rdev);
2736         return err;
2737 }
2738
2739 /* similar to deny_write_access, but accounts for our holding a reference
2740  * to the file ourselves */
2741 static int deny_bitmap_write_access(struct file * file)
2742 {
2743         struct inode *inode = file->f_mapping->host;
2744
2745         spin_lock(&inode->i_lock);
2746         if (atomic_read(&inode->i_writecount) > 1) {
2747                 spin_unlock(&inode->i_lock);
2748                 return -ETXTBSY;
2749         }
2750         atomic_set(&inode->i_writecount, -1);
2751         spin_unlock(&inode->i_lock);
2752
2753         return 0;
2754 }
2755
2756 static int set_bitmap_file(mddev_t *mddev, int fd)
2757 {
2758         int err;
2759
2760         if (mddev->pers) {
2761                 if (!mddev->pers->quiesce)
2762                         return -EBUSY;
2763                 if (mddev->recovery || mddev->sync_thread)
2764                         return -EBUSY;
2765                 /* we should be able to change the bitmap.. */
2766         }
2767
2768
2769         if (fd >= 0) {
2770                 if (mddev->bitmap)
2771                         return -EEXIST; /* cannot add when bitmap is present */
2772                 mddev->bitmap_file = fget(fd);
2773
2774                 if (mddev->bitmap_file == NULL) {
2775                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2776                                mdname(mddev));
2777                         return -EBADF;
2778                 }
2779
2780                 err = deny_bitmap_write_access(mddev->bitmap_file);
2781                 if (err) {
2782                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2783                                mdname(mddev));
2784                         fput(mddev->bitmap_file);
2785                         mddev->bitmap_file = NULL;
2786                         return err;
2787                 }
2788                 mddev->bitmap_offset = 0; /* file overrides offset */
2789         } else if (mddev->bitmap == NULL)
2790                 return -ENOENT; /* cannot remove what isn't there */
2791         err = 0;
2792         if (mddev->pers) {
2793                 mddev->pers->quiesce(mddev, 1);
2794                 if (fd >= 0)
2795                         err = bitmap_create(mddev);
2796                 if (fd < 0 || err)
2797                         bitmap_destroy(mddev);
2798                 mddev->pers->quiesce(mddev, 0);
2799         } else if (fd < 0) {
2800                 if (mddev->bitmap_file)
2801                         fput(mddev->bitmap_file);
2802                 mddev->bitmap_file = NULL;
2803         }
2804
2805         return err;
2806 }
2807
2808 /*
2809  * set_array_info is used two different ways
2810  * The original usage is when creating a new array.
2811  * In this usage, raid_disks is > 0 and it together with
2812  *  level, size, not_persistent,layout,chunksize determine the
2813  *  shape of the array.
2814  *  This will always create an array with a type-0.90.0 superblock.
2815  * The newer usage is when assembling an array.
2816  *  In this case raid_disks will be 0, and the major_version field is
2817  *  use to determine which style super-blocks are to be found on the devices.
2818  *  The minor and patch _version numbers are also kept incase the
2819  *  super_block handler wishes to interpret them.
2820  */
2821 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2822 {
2823
2824         if (info->raid_disks == 0) {
2825                 /* just setting version number for superblock loading */
2826                 if (info->major_version < 0 ||
2827                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2828                     super_types[info->major_version].name == NULL) {
2829                         /* maybe try to auto-load a module? */
2830                         printk(KERN_INFO
2831                                 "md: superblock version %d not known\n",
2832                                 info->major_version);
2833                         return -EINVAL;
2834                 }
2835                 mddev->major_version = info->major_version;
2836                 mddev->minor_version = info->minor_version;
2837                 mddev->patch_version = info->patch_version;
2838                 return 0;
2839         }
2840         mddev->major_version = MD_MAJOR_VERSION;
2841         mddev->minor_version = MD_MINOR_VERSION;
2842         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2843         mddev->ctime         = get_seconds();
2844
2845         mddev->level         = info->level;
2846         mddev->size          = info->size;
2847         mddev->raid_disks    = info->raid_disks;
2848         /* don't set md_minor, it is determined by which /dev/md* was
2849          * openned
2850          */
2851         if (info->state & (1<<MD_SB_CLEAN))
2852                 mddev->recovery_cp = MaxSector;
2853         else
2854                 mddev->recovery_cp = 0;
2855         mddev->persistent    = ! info->not_persistent;
2856
2857         mddev->layout        = info->layout;
2858         mddev->chunk_size    = info->chunk_size;
2859
2860         mddev->max_disks     = MD_SB_DISKS;
2861
2862         mddev->sb_dirty      = 1;
2863
2864         /*
2865          * Generate a 128 bit UUID
2866          */
2867         get_random_bytes(mddev->uuid, 16);
2868
2869         return 0;
2870 }
2871
2872 /*
2873  * update_array_info is used to change the configuration of an
2874  * on-line array.
2875  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2876  * fields in the info are checked against the array.
2877  * Any differences that cannot be handled will cause an error.
2878  * Normally, only one change can be managed at a time.
2879  */
2880 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2881 {
2882         int rv = 0;
2883         int cnt = 0;
2884         int state = 0;
2885
2886         /* calculate expected state,ignoring low bits */
2887         if (mddev->bitmap && mddev->bitmap_offset)
2888                 state |= (1 << MD_SB_BITMAP_PRESENT);
2889
2890         if (mddev->major_version != info->major_version ||
2891             mddev->minor_version != info->minor_version ||
2892 /*          mddev->patch_version != info->patch_version || */
2893             mddev->ctime         != info->ctime         ||
2894             mddev->level         != info->level         ||
2895 /*          mddev->layout        != info->layout        || */
2896             !mddev->persistent   != info->not_persistent||
2897             mddev->chunk_size    != info->chunk_size    ||
2898             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2899             ((state^info->state) & 0xfffffe00)
2900                 )
2901                 return -EINVAL;
2902         /* Check there is only one change */
2903         if (mddev->size != info->size) cnt++;
2904         if (mddev->raid_disks != info->raid_disks) cnt++;
2905         if (mddev->layout != info->layout) cnt++;
2906         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2907         if (cnt == 0) return 0;
2908         if (cnt > 1) return -EINVAL;
2909
2910         if (mddev->layout != info->layout) {
2911                 /* Change layout
2912                  * we don't need to do anything at the md level, the
2913                  * personality will take care of it all.
2914                  */
2915                 if (mddev->pers->reconfig == NULL)
2916                         return -EINVAL;
2917                 else
2918                         return mddev->pers->reconfig(mddev, info->layout, -1);
2919         }
2920         if (mddev->size != info->size) {
2921                 mdk_rdev_t * rdev;
2922                 struct list_head *tmp;
2923                 if (mddev->pers->resize == NULL)
2924                         return -EINVAL;
2925                 /* The "size" is the amount of each device that is used.
2926                  * This can only make sense for arrays with redundancy.
2927                  * linear and raid0 always use whatever space is available
2928                  * We can only consider changing the size if no resync
2929                  * or reconstruction is happening, and if the new size
2930                  * is acceptable. It must fit before the sb_offset or,
2931                  * if that is <data_offset, it must fit before the
2932                  * size of each device.
2933                  * If size is zero, we find the largest size that fits.
2934                  */
2935                 if (mddev->sync_thread)
2936                         return -EBUSY;
2937                 ITERATE_RDEV(mddev,rdev,tmp) {
2938                         sector_t avail;
2939                         int fit = (info->size == 0);
2940                         if (rdev->sb_offset > rdev->data_offset)
2941                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2942                         else
2943                                 avail = get_capacity(rdev->bdev->bd_disk)
2944                                         - rdev->data_offset;
2945                         if (fit && (info->size == 0 || info->size > avail/2))
2946                                 info->size = avail/2;
2947                         if (avail < ((sector_t)info->size << 1))
2948                                 return -ENOSPC;
2949                 }
2950                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2951                 if (!rv) {
2952                         struct block_device *bdev;
2953
2954                         bdev = bdget_disk(mddev->gendisk, 0);
2955                         if (bdev) {
2956                                 down(&bdev->bd_inode->i_sem);
2957                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2958                                 up(&bdev->bd_inode->i_sem);
2959                                 bdput(bdev);
2960                         }
2961                 }
2962         }
2963         if (mddev->raid_disks    != info->raid_disks) {
2964                 /* change the number of raid disks */
2965                 if (mddev->pers->reshape == NULL)
2966                         return -EINVAL;
2967                 if (info->raid_disks <= 0 ||
2968                     info->raid_disks >= mddev->max_disks)
2969                         return -EINVAL;
2970                 if (mddev->sync_thread)
2971                         return -EBUSY;
2972                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2973                 if (!rv) {
2974                         struct block_device *bdev;
2975
2976                         bdev = bdget_disk(mddev->gendisk, 0);
2977                         if (bdev) {
2978                                 down(&bdev->bd_inode->i_sem);
2979                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2980                                 up(&bdev->bd_inode->i_sem);
2981                                 bdput(bdev);
2982                         }
2983                 }
2984         }
2985         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2986                 if (mddev->pers->quiesce == NULL)
2987                         return -EINVAL;
2988                 if (mddev->recovery || mddev->sync_thread)
2989                         return -EBUSY;
2990                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2991                         /* add the bitmap */
2992                         if (mddev->bitmap)
2993                                 return -EEXIST;
2994                         if (mddev->default_bitmap_offset == 0)
2995                                 return -EINVAL;
2996                         mddev->bitmap_offset = mddev->default_bitmap_offset;
2997                         mddev->pers->quiesce(mddev, 1);
2998                         rv = bitmap_create(mddev);
2999                         if (rv)
3000                                 bitmap_destroy(mddev);
3001                         mddev->pers->quiesce(mddev, 0);
3002                 } else {
3003                         /* remove the bitmap */
3004                         if (!mddev->bitmap)
3005                                 return -ENOENT;
3006                         if (mddev->bitmap->file)
3007                                 return -EINVAL;
3008                         mddev->pers->quiesce(mddev, 1);
3009                         bitmap_destroy(mddev);
3010                         mddev->pers->quiesce(mddev, 0);
3011                         mddev->bitmap_offset = 0;
3012                 }
3013         }
3014         md_update_sb(mddev);
3015         return rv;
3016 }
3017
3018 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3019 {
3020         mdk_rdev_t *rdev;
3021
3022         if (mddev->pers == NULL)
3023                 return -ENODEV;
3024
3025         rdev = find_rdev(mddev, dev);
3026         if (!rdev)
3027                 return -ENODEV;
3028
3029         md_error(mddev, rdev);
3030         return 0;
3031 }
3032
3033 static int md_ioctl(struct inode *inode, struct file *file,
3034                         unsigned int cmd, unsigned long arg)
3035 {
3036         int err = 0;
3037         void __user *argp = (void __user *)arg;
3038         struct hd_geometry __user *loc = argp;
3039         mddev_t *mddev = NULL;
3040
3041         if (!capable(CAP_SYS_ADMIN))
3042                 return -EACCES;
3043
3044         /*
3045          * Commands dealing with the RAID driver but not any
3046          * particular array:
3047          */
3048         switch (cmd)
3049         {
3050                 case RAID_VERSION:
3051                         err = get_version(argp);
3052                         goto done;
3053
3054                 case PRINT_RAID_DEBUG:
3055                         err = 0;
3056                         md_print_devices();
3057                         goto done;
3058
3059 #ifndef MODULE
3060                 case RAID_AUTORUN:
3061                         err = 0;
3062                         autostart_arrays(arg);
3063                         goto done;
3064 #endif
3065                 default:;
3066         }
3067
3068         /*
3069          * Commands creating/starting a new array:
3070          */
3071
3072         mddev = inode->i_bdev->bd_disk->private_data;
3073
3074         if (!mddev) {
3075                 BUG();
3076                 goto abort;
3077         }
3078
3079
3080         if (cmd == START_ARRAY) {
3081                 /* START_ARRAY doesn't need to lock the array as autostart_array
3082                  * does the locking, and it could even be a different array
3083                  */
3084                 static int cnt = 3;
3085                 if (cnt > 0 ) {
3086                         printk(KERN_WARNING
3087                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3088                                "This will not be supported beyond 2.6\n",
3089                                current->comm, current->pid);
3090                         cnt--;
3091                 }
3092                 err = autostart_array(new_decode_dev(arg));
3093                 if (err) {
3094                         printk(KERN_WARNING "md: autostart failed!\n");
3095                         goto abort;
3096                 }
3097                 goto done;
3098         }
3099
3100         err = mddev_lock(mddev);
3101         if (err) {
3102                 printk(KERN_INFO
3103                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3104                         err, cmd);
3105                 goto abort;
3106         }
3107
3108         switch (cmd)
3109         {
3110                 case SET_ARRAY_INFO:
3111                         {
3112                                 mdu_array_info_t info;
3113                                 if (!arg)
3114                                         memset(&info, 0, sizeof(info));
3115                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3116                                         err = -EFAULT;
3117                                         goto abort_unlock;
3118                                 }
3119                                 if (mddev->pers) {
3120                                         err = update_array_info(mddev, &info);
3121                                         if (err) {
3122                                                 printk(KERN_WARNING "md: couldn't update"
3123                                                        " array info. %d\n", err);
3124                                                 goto abort_unlock;
3125                                         }
3126                                         goto done_unlock;
3127                                 }
3128                                 if (!list_empty(&mddev->disks)) {
3129                                         printk(KERN_WARNING
3130                                                "md: array %s already has disks!\n",
3131                                                mdname(mddev));
3132                                         err = -EBUSY;
3133                                         goto abort_unlock;
3134                                 }
3135                                 if (mddev->raid_disks) {
3136                                         printk(KERN_WARNING
3137                                                "md: array %s already initialised!\n",
3138                                                mdname(mddev));
3139                                         err = -EBUSY;
3140                                         goto abort_unlock;
3141                                 }
3142                                 err = set_array_info(mddev, &info);
3143                                 if (err) {
3144                                         printk(KERN_WARNING "md: couldn't set"
3145                                                " array info. %d\n", err);
3146                                         goto abort_unlock;
3147                                 }
3148                         }
3149                         goto done_unlock;
3150
3151                 default:;
3152         }
3153
3154         /*
3155          * Commands querying/configuring an existing array:
3156          */
3157         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3158          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3159         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3160                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3161                 err = -ENODEV;
3162                 goto abort_unlock;
3163         }
3164
3165         /*
3166          * Commands even a read-only array can execute:
3167          */
3168         switch (cmd)
3169         {
3170                 case GET_ARRAY_INFO:
3171                         err = get_array_info(mddev, argp);
3172                         goto done_unlock;
3173
3174                 case GET_BITMAP_FILE:
3175                         err = get_bitmap_file(mddev, argp);
3176                         goto done_unlock;
3177
3178                 case GET_DISK_INFO:
3179                         err = get_disk_info(mddev, argp);
3180                         goto done_unlock;
3181
3182                 case RESTART_ARRAY_RW:
3183                         err = restart_array(mddev);
3184                         goto done_unlock;
3185
3186                 case STOP_ARRAY:
3187                         err = do_md_stop (mddev, 0);
3188                         goto done_unlock;
3189
3190                 case STOP_ARRAY_RO:
3191                         err = do_md_stop (mddev, 1);
3192                         goto done_unlock;
3193
3194         /*
3195          * We have a problem here : there is no easy way to give a CHS
3196          * virtual geometry. We currently pretend that we have a 2 heads
3197          * 4 sectors (with a BIG number of cylinders...). This drives
3198          * dosfs just mad... ;-)
3199          */
3200                 case HDIO_GETGEO:
3201                         if (!loc) {
3202                                 err = -EINVAL;
3203                                 goto abort_unlock;
3204                         }
3205                         err = put_user (2, (char __user *) &loc->heads);
3206                         if (err)
3207                                 goto abort_unlock;
3208                         err = put_user (4, (char __user *) &loc->sectors);
3209                         if (err)
3210                                 goto abort_unlock;
3211                         err = put_user(get_capacity(mddev->gendisk)/8,
3212                                         (short __user *) &loc->cylinders);
3213                         if (err)
3214                                 goto abort_unlock;
3215                         err = put_user (get_start_sect(inode->i_bdev),
3216                                                 (long __user *) &loc->start);
3217                         goto done_unlock;
3218         }
3219
3220         /*
3221          * The remaining ioctls are changing the state of the
3222          * superblock, so we do not allow read-only arrays
3223          * here:
3224          */
3225         if (mddev->ro) {
3226                 err = -EROFS;
3227                 goto abort_unlock;
3228         }
3229
3230         switch (cmd)
3231         {
3232                 case ADD_NEW_DISK:
3233                 {
3234                         mdu_disk_info_t info;
3235                         if (copy_from_user(&info, argp, sizeof(info)))
3236                                 err = -EFAULT;
3237                         else
3238                                 err = add_new_disk(mddev, &info);
3239                         goto done_unlock;
3240                 }
3241
3242                 case HOT_REMOVE_DISK:
3243                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3244                         goto done_unlock;
3245
3246                 case HOT_ADD_DISK:
3247                         err = hot_add_disk(mddev, new_decode_dev(arg));
3248                         goto done_unlock;
3249
3250                 case SET_DISK_FAULTY:
3251                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3252                         goto done_unlock;
3253
3254                 case RUN_ARRAY:
3255                         err = do_md_run (mddev);
3256                         goto done_unlock;
3257
3258                 case SET_BITMAP_FILE:
3259                         err = set_bitmap_file(mddev, (int)arg);
3260                         goto done_unlock;
3261
3262                 default:
3263                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3264                                 printk(KERN_WARNING "md: %s(pid %d) used"
3265                                         " obsolete MD ioctl, upgrade your"
3266                                         " software to use new ictls.\n",
3267                                         current->comm, current->pid);
3268                         err = -EINVAL;
3269                         goto abort_unlock;
3270         }
3271
3272 done_unlock:
3273 abort_unlock:
3274         mddev_unlock(mddev);
3275
3276         return err;
3277 done:
3278         if (err)
3279                 MD_BUG();
3280 abort:
3281         return err;
3282 }
3283
3284 static int md_open(struct inode *inode, struct file *file)
3285 {
3286         /*
3287          * Succeed if we can lock the mddev, which confirms that
3288          * it isn't being stopped right now.
3289          */
3290         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3291         int err;
3292
3293         if ((err = mddev_lock(mddev)))
3294                 goto out;
3295
3296         err = 0;
3297         mddev_get(mddev);
3298         mddev_unlock(mddev);
3299
3300         check_disk_change(inode->i_bdev);
3301  out:
3302         return err;
3303 }
3304
3305 static int md_release(struct inode *inode, struct file * file)
3306 {
3307         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3308
3309         if (!mddev)
3310                 BUG();
3311         mddev_put(mddev);
3312
3313         return 0;
3314 }
3315
3316 static int md_media_changed(struct gendisk *disk)
3317 {
3318         mddev_t *mddev = disk->private_data;
3319
3320         return mddev->changed;
3321 }
3322
3323 static int md_revalidate(struct gendisk *disk)
3324 {
3325         mddev_t *mddev = disk->private_data;
3326
3327         mddev->changed = 0;
3328         return 0;
3329 }
3330 static struct block_device_operations md_fops =
3331 {
3332         .owner          = THIS_MODULE,
3333         .open           = md_open,
3334         .release        = md_release,
3335         .ioctl          = md_ioctl,
3336         .media_changed  = md_media_changed,
3337         .revalidate_disk= md_revalidate,
3338 };
3339
3340 static int md_thread(void * arg)
3341 {
3342         mdk_thread_t *thread = arg;
3343
3344         /*
3345          * md_thread is a 'system-thread', it's priority should be very
3346          * high. We avoid resource deadlocks individually in each
3347          * raid personality. (RAID5 does preallocation) We also use RR and
3348          * the very same RT priority as kswapd, thus we will never get
3349          * into a priority inversion deadlock.
3350          *
3351          * we definitely have to have equal or higher priority than
3352          * bdflush, otherwise bdflush will deadlock if there are too
3353          * many dirty RAID5 blocks.
3354          */
3355
3356         allow_signal(SIGKILL);
3357         complete(thread->event);
3358         while (!kthread_should_stop()) {
3359                 void (*run)(mddev_t *);
3360
3361                 wait_event_interruptible_timeout(thread->wqueue,
3362                                                  test_bit(THREAD_WAKEUP, &thread->flags)
3363                                                  || kthread_should_stop(),
3364                                                  thread->timeout);
3365                 try_to_freeze();
3366
3367                 clear_bit(THREAD_WAKEUP, &thread->flags);
3368
3369                 run = thread->run;
3370                 if (run)
3371                         run(thread->mddev);
3372         }
3373
3374         return 0;
3375 }
3376
3377 void md_wakeup_thread(mdk_thread_t *thread)
3378 {
3379         if (thread) {
3380                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3381                 set_bit(THREAD_WAKEUP, &thread->flags);
3382                 wake_up(&thread->wqueue);
3383         }
3384 }
3385
3386 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3387                                  const char *name)
3388 {
3389         mdk_thread_t *thread;
3390         struct completion event;
3391
3392         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3393         if (!thread)
3394                 return NULL;
3395
3396         memset(thread, 0, sizeof(mdk_thread_t));
3397         init_waitqueue_head(&thread->wqueue);
3398
3399         init_completion(&event);
3400         thread->event = &event;
3401         thread->run = run;
3402         thread->mddev = mddev;
3403         thread->name = name;
3404         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3405         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3406         if (IS_ERR(thread->tsk)) {
3407                 kfree(thread);
3408                 return NULL;
3409         }
3410         wait_for_completion(&event);
3411         return thread;
3412 }
3413
3414 void md_unregister_thread(mdk_thread_t *thread)
3415 {
3416         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3417
3418         kthread_stop(thread->tsk);
3419         kfree(thread);
3420 }
3421
3422 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3423 {
3424         if (!mddev) {
3425                 MD_BUG();
3426                 return;
3427         }
3428
3429         if (!rdev || test_bit(Faulty, &rdev->flags))
3430                 return;
3431 /*
3432         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3433                 mdname(mddev),
3434                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3435                 __builtin_return_address(0),__builtin_return_address(1),
3436                 __builtin_return_address(2),__builtin_return_address(3));
3437 */
3438         if (!mddev->pers->error_handler)
3439                 return;
3440         mddev->pers->error_handler(mddev,rdev);
3441         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3442         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3443         md_wakeup_thread(mddev->thread);
3444 }
3445
3446 /* seq_file implementation /proc/mdstat */
3447
3448 static void status_unused(struct seq_file *seq)
3449 {
3450         int i = 0;
3451         mdk_rdev_t *rdev;
3452         struct list_head *tmp;
3453
3454         seq_printf(seq, "unused devices: ");
3455
3456         ITERATE_RDEV_PENDING(rdev,tmp) {
3457                 char b[BDEVNAME_SIZE];
3458                 i++;
3459                 seq_printf(seq, "%s ",
3460                               bdevname(rdev->bdev,b));
3461         }
3462         if (!i)
3463                 seq_printf(seq, "<none>");
3464
3465         seq_printf(seq, "\n");
3466 }
3467
3468
3469 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3470 {
3471         unsigned long max_blocks, resync, res, dt, db, rt;
3472
3473         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3474
3475         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3476                 max_blocks = mddev->resync_max_sectors >> 1;
3477         else
3478                 max_blocks = mddev->size;
3479
3480         /*
3481          * Should not happen.
3482          */
3483         if (!max_blocks) {
3484                 MD_BUG();
3485                 return;
3486         }
3487         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3488         {
3489                 int i, x = res/50, y = 20-x;
3490                 seq_printf(seq, "[");
3491                 for (i = 0; i < x; i++)
3492                         seq_printf(seq, "=");
3493                 seq_printf(seq, ">");
3494                 for (i = 0; i < y; i++)
3495                         seq_printf(seq, ".");
3496                 seq_printf(seq, "] ");
3497         }
3498         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3499                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3500                        "resync" : "recovery"),
3501                       res/10, res % 10, resync, max_blocks);
3502
3503         /*
3504          * We do not want to overflow, so the order of operands and
3505          * the * 100 / 100 trick are important. We do a +1 to be
3506          * safe against division by zero. We only estimate anyway.
3507          *
3508          * dt: time from mark until now
3509          * db: blocks written from mark until now
3510          * rt: remaining time
3511          */
3512         dt = ((jiffies - mddev->resync_mark) / HZ);
3513         if (!dt) dt++;
3514         db = resync - (mddev->resync_mark_cnt/2);
3515         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3516
3517         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3518
3519         seq_printf(seq, " speed=%ldK/sec", db/dt);
3520 }
3521
3522 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3523 {
3524         struct list_head *tmp;
3525         loff_t l = *pos;
3526         mddev_t *mddev;
3527
3528         if (l >= 0x10000)
3529                 return NULL;
3530         if (!l--)
3531                 /* header */
3532                 return (void*)1;
3533
3534         spin_lock(&all_mddevs_lock);
3535         list_for_each(tmp,&all_mddevs)
3536                 if (!l--) {
3537                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3538                         mddev_get(mddev);
3539                         spin_unlock(&all_mddevs_lock);
3540                         return mddev;
3541                 }
3542         spin_unlock(&all_mddevs_lock);
3543         if (!l--)
3544                 return (void*)2;/* tail */
3545         return NULL;
3546 }
3547
3548 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3549 {
3550         struct list_head *tmp;
3551         mddev_t *next_mddev, *mddev = v;
3552
3553         ++*pos;
3554         if (v == (void*)2)
3555                 return NULL;
3556
3557         spin_lock(&all_mddevs_lock);
3558         if (v == (void*)1)
3559                 tmp = all_mddevs.next;
3560         else
3561                 tmp = mddev->all_mddevs.next;
3562         if (tmp != &all_mddevs)
3563                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3564         else {
3565                 next_mddev = (void*)2;
3566                 *pos = 0x10000;
3567         }
3568         spin_unlock(&all_mddevs_lock);
3569
3570         if (v != (void*)1)
3571                 mddev_put(mddev);
3572         return next_mddev;
3573
3574 }
3575
3576 static void md_seq_stop(struct seq_file *seq, void *v)
3577 {
3578         mddev_t *mddev = v;
3579
3580         if (mddev && v != (void*)1 && v != (void*)2)
3581                 mddev_put(mddev);
3582 }
3583
3584 static int md_seq_show(struct seq_file *seq, void *v)
3585 {
3586         mddev_t *mddev = v;
3587         sector_t size;
3588         struct list_head *tmp2;
3589         mdk_rdev_t *rdev;
3590         int i;
3591         struct bitmap *bitmap;
3592
3593         if (v == (void*)1) {
3594                 seq_printf(seq, "Personalities : ");
3595                 spin_lock(&pers_lock);
3596                 for (i = 0; i < MAX_PERSONALITY; i++)
3597                         if (pers[i])
3598                                 seq_printf(seq, "[%s] ", pers[i]->name);
3599
3600                 spin_unlock(&pers_lock);
3601                 seq_printf(seq, "\n");
3602                 return 0;
3603         }
3604         if (v == (void*)2) {
3605                 status_unused(seq);
3606                 return 0;
3607         }
3608
3609         if (mddev_lock(mddev)!=0)
3610                 return -EINTR;
3611         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3612                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3613                                                 mddev->pers ? "" : "in");
3614                 if (mddev->pers) {
3615                         if (mddev->ro)
3616                                 seq_printf(seq, " (read-only)");
3617                         seq_printf(seq, " %s", mddev->pers->name);
3618                 }
3619
3620                 size = 0;
3621                 ITERATE_RDEV(mddev,rdev,tmp2) {
3622                         char b[BDEVNAME_SIZE];
3623                         seq_printf(seq, " %s[%d]",
3624                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3625                         if (test_bit(WriteMostly, &rdev->flags))
3626                                 seq_printf(seq, "(W)");
3627                         if (test_bit(Faulty, &rdev->flags)) {
3628                                 seq_printf(seq, "(F)");
3629                                 continue;
3630                         } else if (rdev->raid_disk < 0)
3631                                 seq_printf(seq, "(S)"); /* spare */
3632                         size += rdev->size;
3633                 }
3634
3635                 if (!list_empty(&mddev->disks)) {
3636                         if (mddev->pers)
3637                                 seq_printf(seq, "\n      %llu blocks",
3638                                         (unsigned long long)mddev->array_size);
3639                         else
3640                                 seq_printf(seq, "\n      %llu blocks",
3641                                         (unsigned long long)size);
3642                 }
3643                 if (mddev->persistent) {
3644                         if (mddev->major_version != 0 ||
3645                             mddev->minor_version != 90) {
3646                                 seq_printf(seq," super %d.%d",
3647                                            mddev->major_version,
3648                                            mddev->minor_version);
3649                         }
3650                 } else
3651                         seq_printf(seq, " super non-persistent");
3652
3653                 if (mddev->pers) {
3654                         mddev->pers->status (seq, mddev);
3655                         seq_printf(seq, "\n      ");
3656                         if (mddev->curr_resync > 2) {
3657                                 status_resync (seq, mddev);
3658                                 seq_printf(seq, "\n      ");
3659                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3660                                 seq_printf(seq, "       resync=DELAYED\n      ");
3661                 } else
3662                         seq_printf(seq, "\n       ");
3663
3664                 if ((bitmap = mddev->bitmap)) {
3665                         unsigned long chunk_kb;
3666                         unsigned long flags;
3667                         spin_lock_irqsave(&bitmap->lock, flags);
3668                         chunk_kb = bitmap->chunksize >> 10;
3669                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3670                                 "%lu%s chunk",
3671                                 bitmap->pages - bitmap->missing_pages,
3672                                 bitmap->pages,
3673                                 (bitmap->pages - bitmap->missing_pages)
3674                                         << (PAGE_SHIFT - 10),
3675                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3676                                 chunk_kb ? "KB" : "B");
3677                         if (bitmap->file) {
3678                                 seq_printf(seq, ", file: ");
3679                                 seq_path(seq, bitmap->file->f_vfsmnt,
3680                                          bitmap->file->f_dentry," \t\n");
3681                         }
3682
3683                         seq_printf(seq, "\n");
3684                         spin_unlock_irqrestore(&bitmap->lock, flags);
3685                 }
3686
3687                 seq_printf(seq, "\n");
3688         }
3689         mddev_unlock(mddev);
3690
3691         return 0;
3692 }
3693
3694 static struct seq_operations md_seq_ops = {
3695         .start  = md_seq_start,
3696         .next   = md_seq_next,
3697         .stop   = md_seq_stop,
3698         .show   = md_seq_show,
3699 };
3700
3701 static int md_seq_open(struct inode *inode, struct file *file)
3702 {
3703         int error;
3704
3705         error = seq_open(file, &md_seq_ops);
3706         return error;
3707 }
3708
3709 static struct file_operations md_seq_fops = {
3710         .open           = md_seq_open,
3711         .read           = seq_read,
3712         .llseek         = seq_lseek,
3713         .release        = seq_release,
3714 };
3715
3716 int register_md_personality(int pnum, mdk_personality_t *p)
3717 {
3718         if (pnum >= MAX_PERSONALITY) {
3719                 printk(KERN_ERR
3720                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3721                        p->name, pnum, MAX_PERSONALITY-1);
3722                 return -EINVAL;
3723         }
3724
3725         spin_lock(&pers_lock);
3726         if (pers[pnum]) {
3727                 spin_unlock(&pers_lock);
3728                 return -EBUSY;
3729         }
3730
3731         pers[pnum] = p;
3732         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3733         spin_unlock(&pers_lock);
3734         return 0;
3735 }
3736
3737 int unregister_md_personality(int pnum)
3738 {
3739         if (pnum >= MAX_PERSONALITY)
3740                 return -EINVAL;
3741
3742         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3743         spin_lock(&pers_lock);
3744         pers[pnum] = NULL;
3745         spin_unlock(&pers_lock);
3746         return 0;
3747 }
3748
3749 static int is_mddev_idle(mddev_t *mddev)
3750 {
3751         mdk_rdev_t * rdev;
3752         struct list_head *tmp;
3753         int idle;
3754         unsigned long curr_events;
3755
3756         idle = 1;
3757         ITERATE_RDEV(mddev,rdev,tmp) {
3758                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3759                 curr_events = disk_stat_read(disk, sectors[0]) +
3760                                 disk_stat_read(disk, sectors[1]) -
3761                                 atomic_read(&disk->sync_io);
3762                 /* Allow some slack between valud of curr_events and last_events,
3763                  * as there are some uninteresting races.
3764                  * Note: the following is an unsigned comparison.
3765                  */
3766                 if ((curr_events - rdev->last_events + 32) > 64) {
3767                         rdev->last_events = curr_events;
3768                         idle = 0;
3769                 }
3770         }
3771         return idle;
3772 }
3773
3774 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3775 {
3776         /* another "blocks" (512byte) blocks have been synced */
3777         atomic_sub(blocks, &mddev->recovery_active);
3778         wake_up(&mddev->recovery_wait);
3779         if (!ok) {
3780                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3781                 md_wakeup_thread(mddev->thread);
3782                 // stop recovery, signal do_sync ....
3783         }
3784 }
3785
3786
3787 /* md_write_start(mddev, bi)
3788  * If we need to update some array metadata (e.g. 'active' flag
3789  * in superblock) before writing, schedule a superblock update
3790  * and wait for it to complete.
3791  */
3792 void md_write_start(mddev_t *mddev, struct bio *bi)
3793 {
3794         if (bio_data_dir(bi) != WRITE)
3795                 return;
3796
3797         atomic_inc(&mddev->writes_pending);
3798         if (mddev->in_sync) {
3799                 spin_lock(&mddev->write_lock);
3800                 if (mddev->in_sync) {
3801                         mddev->in_sync = 0;
3802                         mddev->sb_dirty = 1;
3803                         md_wakeup_thread(mddev->thread);
3804                 }
3805                 spin_unlock(&mddev->write_lock);
3806         }
3807         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3808 }
3809
3810 void md_write_end(mddev_t *mddev)
3811 {
3812         if (atomic_dec_and_test(&mddev->writes_pending)) {
3813                 if (mddev->safemode == 2)
3814                         md_wakeup_thread(mddev->thread);
3815                 else
3816                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3817         }
3818 }
3819
3820 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3821
3822 #define SYNC_MARKS      10
3823 #define SYNC_MARK_STEP  (3*HZ)
3824 static void md_do_sync(mddev_t *mddev)
3825 {
3826         mddev_t *mddev2;
3827         unsigned int currspeed = 0,
3828                  window;
3829         sector_t max_sectors,j, io_sectors;
3830         unsigned long mark[SYNC_MARKS];
3831         sector_t mark_cnt[SYNC_MARKS];
3832         int last_mark,m;
3833         struct list_head *tmp;
3834         sector_t last_check;
3835         int skipped = 0;
3836
3837         /* just incase thread restarts... */
3838         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3839                 return;
3840
3841         /* we overload curr_resync somewhat here.
3842          * 0 == not engaged in resync at all
3843          * 2 == checking that there is no conflict with another sync
3844          * 1 == like 2, but have yielded to allow conflicting resync to
3845          *              commense
3846          * other == active in resync - this many blocks
3847          *
3848          * Before starting a resync we must have set curr_resync to
3849          * 2, and then checked that every "conflicting" array has curr_resync
3850          * less than ours.  When we find one that is the same or higher
3851          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3852          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3853          * This will mean we have to start checking from the beginning again.
3854          *
3855          */
3856
3857         do {
3858                 mddev->curr_resync = 2;
3859
3860         try_again:
3861                 if (signal_pending(current) ||
3862                     kthread_should_stop()) {
3863                         flush_signals(current);
3864                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3865                         goto skip;
3866                 }
3867                 ITERATE_MDDEV(mddev2,tmp) {
3868                         if (mddev2 == mddev)
3869                                 continue;
3870                         if (mddev2->curr_resync &&
3871                             match_mddev_units(mddev,mddev2)) {
3872                                 DEFINE_WAIT(wq);
3873                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3874                                         /* arbitrarily yield */
3875                                         mddev->curr_resync = 1;
3876                                         wake_up(&resync_wait);
3877                                 }
3878                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3879                                         /* no need to wait here, we can wait the next
3880                                          * time 'round when curr_resync == 2
3881                                          */
3882                                         continue;
3883                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3884                                 if (!signal_pending(current) &&
3885                                     !kthread_should_stop() &&
3886                                     mddev2->curr_resync >= mddev->curr_resync) {
3887                                         printk(KERN_INFO "md: delaying resync of %s"
3888                                                " until %s has finished resync (they"
3889                                                " share one or more physical units)\n",
3890                                                mdname(mddev), mdname(mddev2));
3891                                         mddev_put(mddev2);
3892                                         schedule();
3893                                         finish_wait(&resync_wait, &wq);
3894                                         goto try_again;
3895                                 }
3896                                 finish_wait(&resync_wait, &wq);
3897                         }
3898                 }
3899         } while (mddev->curr_resync < 2);
3900
3901         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3902                 /* resync follows the size requested by the personality,
3903                  * which defaults to physical size, but can be virtual size
3904                  */
3905                 max_sectors = mddev->resync_max_sectors;
3906                 mddev->resync_mismatches = 0;
3907         } else
3908                 /* recovery follows the physical size of devices */
3909                 max_sectors = mddev->size << 1;
3910
3911         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3912         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3913                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3914         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3915                "(but not more than %d KB/sec) for reconstruction.\n",
3916                sysctl_speed_limit_max);
3917
3918         is_mddev_idle(mddev); /* this also initializes IO event counters */
3919         /* we don't use the checkpoint if there's a bitmap */
3920         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
3921             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3922                 j = mddev->recovery_cp;
3923         else
3924                 j = 0;
3925         io_sectors = 0;
3926         for (m = 0; m < SYNC_MARKS; m++) {
3927                 mark[m] = jiffies;
3928                 mark_cnt[m] = io_sectors;
3929         }
3930         last_mark = 0;
3931         mddev->resync_mark = mark[last_mark];
3932         mddev->resync_mark_cnt = mark_cnt[last_mark];
3933
3934         /*
3935          * Tune reconstruction:
3936          */
3937         window = 32*(PAGE_SIZE/512);
3938         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3939                 window/2,(unsigned long long) max_sectors/2);
3940
3941         atomic_set(&mddev->recovery_active, 0);
3942         init_waitqueue_head(&mddev->recovery_wait);
3943         last_check = 0;
3944
3945         if (j>2) {
3946                 printk(KERN_INFO
3947                         "md: resuming recovery of %s from checkpoint.\n",
3948                         mdname(mddev));
3949                 mddev->curr_resync = j;
3950         }
3951
3952         while (j < max_sectors) {
3953                 sector_t sectors;
3954
3955                 skipped = 0;
3956                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3957                                             currspeed < sysctl_speed_limit_min);
3958                 if (sectors == 0) {
3959                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3960                         goto out;
3961                 }
3962
3963                 if (!skipped) { /* actual IO requested */
3964                         io_sectors += sectors;
3965                         atomic_add(sectors, &mddev->recovery_active);
3966                 }
3967
3968                 j += sectors;
3969                 if (j>1) mddev->curr_resync = j;
3970
3971
3972                 if (last_check + window > io_sectors || j == max_sectors)
3973                         continue;
3974
3975                 last_check = io_sectors;
3976
3977                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3978                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3979                         break;
3980
3981         repeat:
3982                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3983                         /* step marks */
3984                         int next = (last_mark+1) % SYNC_MARKS;
3985
3986                         mddev->resync_mark = mark[next];
3987                         mddev->resync_mark_cnt = mark_cnt[next];
3988                         mark[next] = jiffies;
3989                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3990                         last_mark = next;
3991                 }
3992
3993
3994                 if (signal_pending(current) || kthread_should_stop()) {
3995                         /*
3996                          * got a signal, exit.
3997                          */
3998                         printk(KERN_INFO
3999                                 "md: md_do_sync() got signal ... exiting\n");
4000                         flush_signals(current);
4001                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4002                         goto out;
4003                 }
4004
4005                 /*
4006                  * this loop exits only if either when we are slower than
4007                  * the 'hard' speed limit, or the system was IO-idle for
4008                  * a jiffy.
4009                  * the system might be non-idle CPU-wise, but we only care
4010                  * about not overloading the IO subsystem. (things like an
4011                  * e2fsck being done on the RAID array should execute fast)
4012                  */
4013                 mddev->queue->unplug_fn(mddev->queue);
4014                 cond_resched();
4015
4016                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4017                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4018
4019                 if (currspeed > sysctl_speed_limit_min) {
4020                         if ((currspeed > sysctl_speed_limit_max) ||
4021                                         !is_mddev_idle(mddev)) {
4022                                 msleep_interruptible(250);
4023                                 goto repeat;
4024                         }
4025                 }
4026         }
4027         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4028         /*
4029          * this also signals 'finished resyncing' to md_stop
4030          */
4031  out:
4032         mddev->queue->unplug_fn(mddev->queue);
4033
4034         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4035
4036         /* tell personality that we are finished */
4037         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4038
4039         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4040             mddev->curr_resync > 2 &&
4041             mddev->curr_resync >= mddev->recovery_cp) {
4042                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4043                         printk(KERN_INFO
4044                                 "md: checkpointing recovery of %s.\n",
4045                                 mdname(mddev));
4046                         mddev->recovery_cp = mddev->curr_resync;
4047                 } else
4048                         mddev->recovery_cp = MaxSector;
4049         }
4050
4051  skip:
4052         mddev->curr_resync = 0;
4053         wake_up(&resync_wait);
4054         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4055         md_wakeup_thread(mddev->thread);
4056 }
4057
4058
4059 /*
4060  * This routine is regularly called by all per-raid-array threads to
4061  * deal with generic issues like resync and super-block update.
4062  * Raid personalities that don't have a thread (linear/raid0) do not
4063  * need this as they never do any recovery or update the superblock.
4064  *
4065  * It does not do any resync itself, but rather "forks" off other threads
4066  * to do that as needed.
4067  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4068  * "->recovery" and create a thread at ->sync_thread.
4069  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4070  * and wakeups up this thread which will reap the thread and finish up.
4071  * This thread also removes any faulty devices (with nr_pending == 0).
4072  *
4073  * The overall approach is:
4074  *  1/ if the superblock needs updating, update it.
4075  *  2/ If a recovery thread is running, don't do anything else.
4076  *  3/ If recovery has finished, clean up, possibly marking spares active.
4077  *  4/ If there are any faulty devices, remove them.
4078  *  5/ If array is degraded, try to add spares devices
4079  *  6/ If array has spares or is not in-sync, start a resync thread.
4080  */
4081 void md_check_recovery(mddev_t *mddev)
4082 {
4083         mdk_rdev_t *rdev;
4084         struct list_head *rtmp;
4085
4086
4087         if (mddev->bitmap)
4088                 bitmap_daemon_work(mddev->bitmap);
4089
4090         if (mddev->ro)
4091                 return;
4092
4093         if (signal_pending(current)) {
4094                 if (mddev->pers->sync_request) {
4095                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4096                                mdname(mddev));
4097                         mddev->safemode = 2;
4098                 }
4099                 flush_signals(current);
4100         }
4101
4102         if ( ! (
4103                 mddev->sb_dirty ||
4104                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4105                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4106                 (mddev->safemode == 1) ||
4107                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4108                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4109                 ))
4110                 return;
4111
4112         if (mddev_trylock(mddev)==0) {
4113                 int spares =0;
4114
4115                 spin_lock(&mddev->write_lock);
4116                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4117                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4118                         mddev->in_sync = 1;
4119                         mddev->sb_dirty = 1;
4120                 }
4121                 if (mddev->safemode == 1)
4122                         mddev->safemode = 0;
4123                 spin_unlock(&mddev->write_lock);
4124
4125                 if (mddev->sb_dirty)
4126                         md_update_sb(mddev);
4127
4128
4129                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4130                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4131                         /* resync/recovery still happening */
4132                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4133                         goto unlock;
4134                 }
4135                 if (mddev->sync_thread) {
4136                         /* resync has finished, collect result */
4137                         md_unregister_thread(mddev->sync_thread);
4138                         mddev->sync_thread = NULL;
4139                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4140                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4141                                 /* success...*/
4142                                 /* activate any spares */
4143                                 mddev->pers->spare_active(mddev);
4144                         }
4145                         md_update_sb(mddev);
4146
4147                         /* if array is no-longer degraded, then any saved_raid_disk
4148                          * information must be scrapped
4149                          */
4150                         if (!mddev->degraded)
4151                                 ITERATE_RDEV(mddev,rdev,rtmp)
4152                                         rdev->saved_raid_disk = -1;
4153
4154                         mddev->recovery = 0;
4155                         /* flag recovery needed just to double check */
4156                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4157                         goto unlock;
4158                 }
4159                 /* Clear some bits that don't mean anything, but
4160                  * might be left set
4161                  */
4162                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4163                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4164                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4165                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4166
4167                 /* no recovery is running.
4168                  * remove any failed drives, then
4169                  * add spares if possible.
4170                  * Spare are also removed and re-added, to allow
4171                  * the personality to fail the re-add.
4172                  */
4173                 ITERATE_RDEV(mddev,rdev,rtmp)
4174                         if (rdev->raid_disk >= 0 &&
4175                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4176                             atomic_read(&rdev->nr_pending)==0) {
4177                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4178                                         char nm[20];
4179                                         sprintf(nm,"rd%d", rdev->raid_disk);
4180                                         sysfs_remove_link(&mddev->kobj, nm);
4181                                         rdev->raid_disk = -1;
4182                                 }
4183                         }
4184
4185                 if (mddev->degraded) {
4186                         ITERATE_RDEV(mddev,rdev,rtmp)
4187                                 if (rdev->raid_disk < 0
4188                                     && !test_bit(Faulty, &rdev->flags)) {
4189                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4190                                                 char nm[20];
4191                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4192                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4193                                                 spares++;
4194                                         } else
4195                                                 break;
4196                                 }
4197                 }
4198
4199                 if (spares) {
4200                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4201                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4202                 } else if (mddev->recovery_cp < MaxSector) {
4203                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4204                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4205                         /* nothing to be done ... */
4206                         goto unlock;
4207
4208                 if (mddev->pers->sync_request) {
4209                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4210                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4211                                 /* We are adding a device or devices to an array
4212                                  * which has the bitmap stored on all devices.
4213                                  * So make sure all bitmap pages get written
4214                                  */
4215                                 bitmap_write_all(mddev->bitmap);
4216                         }
4217                         mddev->sync_thread = md_register_thread(md_do_sync,
4218                                                                 mddev,
4219                                                                 "%s_resync");
4220                         if (!mddev->sync_thread) {
4221                                 printk(KERN_ERR "%s: could not start resync"
4222                                         " thread...\n",
4223                                         mdname(mddev));
4224                                 /* leave the spares where they are, it shouldn't hurt */
4225                                 mddev->recovery = 0;
4226                         } else {
4227                                 md_wakeup_thread(mddev->sync_thread);
4228                         }
4229                 }
4230         unlock:
4231                 mddev_unlock(mddev);
4232         }
4233 }
4234
4235 static int md_notify_reboot(struct notifier_block *this,
4236                             unsigned long code, void *x)
4237 {
4238         struct list_head *tmp;
4239         mddev_t *mddev;
4240
4241         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4242
4243                 printk(KERN_INFO "md: stopping all md devices.\n");
4244
4245                 ITERATE_MDDEV(mddev,tmp)
4246                         if (mddev_trylock(mddev)==0)
4247                                 do_md_stop (mddev, 1);
4248                 /*
4249                  * certain more exotic SCSI devices are known to be
4250                  * volatile wrt too early system reboots. While the
4251                  * right place to handle this issue is the given
4252                  * driver, we do want to have a safe RAID driver ...
4253                  */
4254                 mdelay(1000*1);
4255         }
4256         return NOTIFY_DONE;
4257 }
4258
4259 static struct notifier_block md_notifier = {
4260         .notifier_call  = md_notify_reboot,
4261         .next           = NULL,
4262         .priority       = INT_MAX, /* before any real devices */
4263 };
4264
4265 static void md_geninit(void)
4266 {
4267         struct proc_dir_entry *p;
4268
4269         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4270
4271         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4272         if (p)
4273                 p->proc_fops = &md_seq_fops;
4274 }
4275
4276 static int __init md_init(void)
4277 {
4278         int minor;
4279
4280         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4281                         " MD_SB_DISKS=%d\n",
4282                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4283                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4284         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
4285                         BITMAP_MINOR);
4286
4287         if (register_blkdev(MAJOR_NR, "md"))
4288                 return -1;
4289         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4290                 unregister_blkdev(MAJOR_NR, "md");
4291                 return -1;
4292         }
4293         devfs_mk_dir("md");
4294         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4295                                 md_probe, NULL, NULL);
4296         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4297                             md_probe, NULL, NULL);
4298
4299         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4300                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4301                                 S_IFBLK|S_IRUSR|S_IWUSR,
4302                                 "md/%d", minor);
4303
4304         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4305                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4306                               S_IFBLK|S_IRUSR|S_IWUSR,
4307                               "md/mdp%d", minor);
4308
4309
4310         register_reboot_notifier(&md_notifier);
4311         raid_table_header = register_sysctl_table(raid_root_table, 1);
4312
4313         md_geninit();
4314         return (0);
4315 }
4316
4317
4318 #ifndef MODULE
4319
4320 /*
4321  * Searches all registered partitions for autorun RAID arrays
4322  * at boot time.
4323  */
4324 static dev_t detected_devices[128];
4325 static int dev_cnt;
4326
4327 void md_autodetect_dev(dev_t dev)
4328 {
4329         if (dev_cnt >= 0 && dev_cnt < 127)
4330                 detected_devices[dev_cnt++] = dev;
4331 }
4332
4333
4334 static void autostart_arrays(int part)
4335 {
4336         mdk_rdev_t *rdev;
4337         int i;
4338
4339         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4340
4341         for (i = 0; i < dev_cnt; i++) {
4342                 dev_t dev = detected_devices[i];
4343
4344                 rdev = md_import_device(dev,0, 0);
4345                 if (IS_ERR(rdev))
4346                         continue;
4347
4348                 if (test_bit(Faulty, &rdev->flags)) {
4349                         MD_BUG();
4350                         continue;
4351                 }
4352                 list_add(&rdev->same_set, &pending_raid_disks);
4353         }
4354         dev_cnt = 0;
4355
4356         autorun_devices(part);
4357 }
4358
4359 #endif
4360
4361 static __exit void md_exit(void)
4362 {
4363         mddev_t *mddev;
4364         struct list_head *tmp;
4365         int i;
4366         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4367         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4368         for (i=0; i < MAX_MD_DEVS; i++)
4369                 devfs_remove("md/%d", i);
4370         for (i=0; i < MAX_MD_DEVS; i++)
4371                 devfs_remove("md/d%d", i);
4372
4373         devfs_remove("md");
4374
4375         unregister_blkdev(MAJOR_NR,"md");
4376         unregister_blkdev(mdp_major, "mdp");
4377         unregister_reboot_notifier(&md_notifier);
4378         unregister_sysctl_table(raid_table_header);
4379         remove_proc_entry("mdstat", NULL);
4380         ITERATE_MDDEV(mddev,tmp) {
4381                 struct gendisk *disk = mddev->gendisk;
4382                 if (!disk)
4383                         continue;
4384                 export_array(mddev);
4385                 del_gendisk(disk);
4386                 put_disk(disk);
4387                 mddev->gendisk = NULL;
4388                 mddev_put(mddev);
4389         }
4390 }
4391
4392 module_init(md_init)
4393 module_exit(md_exit)
4394
4395 EXPORT_SYMBOL(register_md_personality);
4396 EXPORT_SYMBOL(unregister_md_personality);
4397 EXPORT_SYMBOL(md_error);
4398 EXPORT_SYMBOL(md_done_sync);
4399 EXPORT_SYMBOL(md_write_start);
4400 EXPORT_SYMBOL(md_write_end);
4401 EXPORT_SYMBOL(md_register_thread);
4402 EXPORT_SYMBOL(md_unregister_thread);
4403 EXPORT_SYMBOL(md_wakeup_thread);
4404 EXPORT_SYMBOL(md_print_devices);
4405 EXPORT_SYMBOL(md_check_recovery);
4406 MODULE_LICENSE("GPL");
4407 MODULE_ALIAS("md");
4408 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);