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 #include <linux/poll.h>
  46
  47 #include <linux/init.h>
  48
  49 #include <linux/file.h>
  50
  51 #ifdef CONFIG_KMOD
  52 #include <linux/kmod.h>
  53 #endif
  54
  55 #include <asm/unaligned.h>
  56
  57 #define MAJOR_NR MD_MAJOR
  58 #define MD_DRIVER
  59
  60 /* 63 partitions with the alternate major number (mdp) */
  61 #define MdpMinorShift 6
  62
  63 #define DEBUG 0
  64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  65
  66
  67 #ifndef MODULE
  68 static void autostart_arrays (int part);
  69 #endif
  70
  71 static LIST_HEAD(pers_list);
  72 static DEFINE_SPINLOCK(pers_lock);
  73
  74 /*
  75  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  76  * is 1000 KB/sec, so the extra system load does not show up that much.
  77  * Increase it if you want to have more _guaranteed_ speed. Note that
  78  * the RAID driver will use the maximum available bandwidth if the IO
  79  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  80  * speed limit - in case reconstruction slows down your system despite
  81  * idle IO detection.
  82  *
  83  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  84  */
  85
  86 static int sysctl_speed_limit_min = 1000;
  87 static int sysctl_speed_limit_max = 200000;
  88
  89 static struct ctl_table_header *raid_table_header;
  90
  91 static ctl_table raid_table[] = {
  92         {
  93                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  94                 .procname       = "speed_limit_min",
  95                 .data           = &sysctl_speed_limit_min,
  96                 .maxlen         = sizeof(int),
  97                 .mode           = 0644,
  98                 .proc_handler   = &proc_dointvec,
  99         },
 100         {
 101                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 102                 .procname       = "speed_limit_max",
 103                 .data           = &sysctl_speed_limit_max,
 104                 .maxlen         = sizeof(int),
 105                 .mode           = 0644,
 106                 .proc_handler   = &proc_dointvec,
 107         },
 108         { .ctl_name = 0 }
 109 };
 110
 111 static ctl_table raid_dir_table[] = {
 112         {
 113                 .ctl_name       = DEV_RAID,
 114                 .procname       = "raid",
 115                 .maxlen         = 0,
 116                 .mode           = 0555,
 117                 .child          = raid_table,
 118         },
 119         { .ctl_name = 0 }
 120 };
 121
 122 static ctl_table raid_root_table[] = {
 123         {
 124                 .ctl_name       = CTL_DEV,
 125                 .procname       = "dev",
 126                 .maxlen         = 0,
 127                 .mode           = 0555,
 128                 .child          = raid_dir_table,
 129         },
 130         { .ctl_name = 0 }
 131 };
 132
 133 static struct block_device_operations md_fops;
 134
 135 static int start_readonly;
 136
 137 /*
 138  * We have a system wide 'event count' that is incremented
 139  * on any 'interesting' event, and readers of /proc/mdstat
 140  * can use 'poll' or 'select' to find out when the event
 141  * count increases.
 142  *
 143  * Events are:
 144  *  start array, stop array, error, add device, remove device,
 145  *  start build, activate spare
 146  */
 147 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 148 static atomic_t md_event_count;
 149 static void md_new_event(mddev_t *mddev)
 150 {
 151         atomic_inc(&md_event_count);
 152         wake_up(&md_event_waiters);
 153 }
 154
 155 /*
 156  * Enables to iterate over all existing md arrays
 157  * all_mddevs_lock protects this list.
 158  */
 159 static LIST_HEAD(all_mddevs);
 160 static DEFINE_SPINLOCK(all_mddevs_lock);
 161
 162
 163 /*
 164  * iterates through all used mddevs in the system.
 165  * We take care to grab the all_mddevs_lock whenever navigating
 166  * the list, and to always hold a refcount when unlocked.
 167  * Any code which breaks out of this loop while own
 168  * a reference to the current mddev and must mddev_put it.
 169  */
 170 #define ITERATE_MDDEV(mddev,tmp)                                        \
 171                                                                         \
 172         for (({ spin_lock(&all_mddevs_lock);                            \
 173                 tmp = all_mddevs.next;                                  \
 174                 mddev = NULL;});                                        \
 175              ({ if (tmp != &all_mddevs)                                 \
 176                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 177                 spin_unlock(&all_mddevs_lock);                          \
 178                 if (mddev) mddev_put(mddev);                            \
 179                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 180                 tmp != &all_mddevs;});                                  \
 181              ({ spin_lock(&all_mddevs_lock);                            \
 182                 tmp = tmp->next;})                                      \
 183                 )
 184
 185
 186 static int md_fail_request (request_queue_t *q, struct bio *bio)
 187 {
 188         bio_io_error(bio, bio->bi_size);
 189         return 0;
 190 }
 191
 192 static inline mddev_t *mddev_get(mddev_t *mddev)
 193 {
 194         atomic_inc(&mddev->active);
 195         return mddev;
 196 }
 197
 198 static void mddev_put(mddev_t *mddev)
 199 {
 200         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 201                 return;
 202         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 203                 list_del(&mddev->all_mddevs);
 204                 blk_put_queue(mddev->queue);
 205                 kobject_unregister(&mddev->kobj);
 206         }
 207         spin_unlock(&all_mddevs_lock);
 208 }
 209
 210 static mddev_t * mddev_find(dev_t unit)
 211 {
 212         mddev_t *mddev, *new = NULL;
 213
 214  retry:
 215         spin_lock(&all_mddevs_lock);
 216         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 217                 if (mddev->unit == unit) {
 218                         mddev_get(mddev);
 219                         spin_unlock(&all_mddevs_lock);
 220                         kfree(new);
 221                         return mddev;
 222                 }
 223
 224         if (new) {
 225                 list_add(&new->all_mddevs, &all_mddevs);
 226                 spin_unlock(&all_mddevs_lock);
 227                 return new;
 228         }
 229         spin_unlock(&all_mddevs_lock);
 230
 231         new = kzalloc(sizeof(*new), GFP_KERNEL);
 232         if (!new)
 233                 return NULL;
 234
 235         new->unit = unit;
 236         if (MAJOR(unit) == MD_MAJOR)
 237                 new->md_minor = MINOR(unit);
 238         else
 239                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 240
 241         init_MUTEX(&new->reconfig_sem);
 242         INIT_LIST_HEAD(&new->disks);
 243         INIT_LIST_HEAD(&new->all_mddevs);
 244         init_timer(&new->safemode_timer);
 245         atomic_set(&new->active, 1);
 246         spin_lock_init(&new->write_lock);
 247         init_waitqueue_head(&new->sb_wait);
 248
 249         new->queue = blk_alloc_queue(GFP_KERNEL);
 250         if (!new->queue) {
 251                 kfree(new);
 252                 return NULL;
 253         }
 254
 255         blk_queue_make_request(new->queue, md_fail_request);
 256
 257         goto retry;
 258 }
 259
 260 static inline int mddev_lock(mddev_t * mddev)
 261 {
 262         return down_interruptible(&mddev->reconfig_sem);
 263 }
 264
 265 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 266 {
 267         down(&mddev->reconfig_sem);
 268 }
 269
 270 static inline int mddev_trylock(mddev_t * mddev)
 271 {
 272         return down_trylock(&mddev->reconfig_sem);
 273 }
 274
 275 static inline void mddev_unlock(mddev_t * mddev)
 276 {
 277         up(&mddev->reconfig_sem);
 278
 279         md_wakeup_thread(mddev->thread);
 280 }
 281
 282 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 283 {
 284         mdk_rdev_t * rdev;
 285         struct list_head *tmp;
 286
 287         ITERATE_RDEV(mddev,rdev,tmp) {
 288                 if (rdev->desc_nr == nr)
 289                         return rdev;
 290         }
 291         return NULL;
 292 }
 293
 294 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 295 {
 296         struct list_head *tmp;
 297         mdk_rdev_t *rdev;
 298
 299         ITERATE_RDEV(mddev,rdev,tmp) {
 300                 if (rdev->bdev->bd_dev == dev)
 301                         return rdev;
 302         }
 303         return NULL;
 304 }
 305
 306 static struct mdk_personality *find_pers(int level, char *clevel)
 307 {
 308         struct mdk_personality *pers;
 309         list_for_each_entry(pers, &pers_list, list) {
 310                 if (level != LEVEL_NONE && pers->level == level)
 311                         return pers;
 312                 if (strcmp(pers->name, clevel)==0)
 313                         return pers;
 314         }
 315         return NULL;
 316 }
 317
 318 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 319 {
 320         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 321         return MD_NEW_SIZE_BLOCKS(size);
 322 }
 323
 324 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 325 {
 326         sector_t size;
 327
 328         size = rdev->sb_offset;
 329
 330         if (chunk_size)
 331                 size &= ~((sector_t)chunk_size/1024 - 1);
 332         return size;
 333 }
 334
 335 static int alloc_disk_sb(mdk_rdev_t * rdev)
 336 {
 337         if (rdev->sb_page)
 338                 MD_BUG();
 339
 340         rdev->sb_page = alloc_page(GFP_KERNEL);
 341         if (!rdev->sb_page) {
 342                 printk(KERN_ALERT "md: out of memory.\n");
 343                 return -EINVAL;
 344         }
 345
 346         return 0;
 347 }
 348
 349 static void free_disk_sb(mdk_rdev_t * rdev)
 350 {
 351         if (rdev->sb_page) {
 352                 put_page(rdev->sb_page);
 353                 rdev->sb_loaded = 0;
 354                 rdev->sb_page = NULL;
 355                 rdev->sb_offset = 0;
 356                 rdev->size = 0;
 357         }
 358 }
 359
 360
 361 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 362 {
 363         mdk_rdev_t *rdev = bio->bi_private;
 364         mddev_t *mddev = rdev->mddev;
 365         if (bio->bi_size)
 366                 return 1;
 367
 368         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 369                 md_error(mddev, rdev);
 370
 371         if (atomic_dec_and_test(&mddev->pending_writes))
 372                 wake_up(&mddev->sb_wait);
 373         bio_put(bio);
 374         return 0;
 375 }
 376
 377 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 378 {
 379         struct bio *bio2 = bio->bi_private;
 380         mdk_rdev_t *rdev = bio2->bi_private;
 381         mddev_t *mddev = rdev->mddev;
 382         if (bio->bi_size)
 383                 return 1;
 384
 385         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 386             error == -EOPNOTSUPP) {
 387                 unsigned long flags;
 388                 /* barriers don't appear to be supported :-( */
 389                 set_bit(BarriersNotsupp, &rdev->flags);
 390                 mddev->barriers_work = 0;
 391                 spin_lock_irqsave(&mddev->write_lock, flags);
 392                 bio2->bi_next = mddev->biolist;
 393                 mddev->biolist = bio2;
 394                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 395                 wake_up(&mddev->sb_wait);
 396                 bio_put(bio);
 397                 return 0;
 398         }
 399         bio_put(bio2);
 400         bio->bi_private = rdev;
 401         return super_written(bio, bytes_done, error);
 402 }
 403
 404 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 405                    sector_t sector, int size, struct page *page)
 406 {
 407         /* write first size bytes of page to sector of rdev
 408          * Increment mddev->pending_writes before returning
 409          * and decrement it on completion, waking up sb_wait
 410          * if zero is reached.
 411          * If an error occurred, call md_error
 412          *
 413          * As we might need to resubmit the request if BIO_RW_BARRIER
 414          * causes ENOTSUPP, we allocate a spare bio...
 415          */
 416         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 417         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 418
 419         bio->bi_bdev = rdev->bdev;
 420         bio->bi_sector = sector;
 421         bio_add_page(bio, page, size, 0);
 422         bio->bi_private = rdev;
 423         bio->bi_end_io = super_written;
 424         bio->bi_rw = rw;
 425
 426         atomic_inc(&mddev->pending_writes);
 427         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 428                 struct bio *rbio;
 429                 rw |= (1<<BIO_RW_BARRIER);
 430                 rbio = bio_clone(bio, GFP_NOIO);
 431                 rbio->bi_private = bio;
 432                 rbio->bi_end_io = super_written_barrier;
 433                 submit_bio(rw, rbio);
 434         } else
 435                 submit_bio(rw, bio);
 436 }
 437
 438 void md_super_wait(mddev_t *mddev)
 439 {
 440         /* wait for all superblock writes that were scheduled to complete.
 441          * if any had to be retried (due to BARRIER problems), retry them
 442          */
 443         DEFINE_WAIT(wq);
 444         for(;;) {
 445                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 446                 if (atomic_read(&mddev->pending_writes)==0)
 447                         break;
 448                 while (mddev->biolist) {
 449                         struct bio *bio;
 450                         spin_lock_irq(&mddev->write_lock);
 451                         bio = mddev->biolist;
 452                         mddev->biolist = bio->bi_next ;
 453                         bio->bi_next = NULL;
 454                         spin_unlock_irq(&mddev->write_lock);
 455                         submit_bio(bio->bi_rw, bio);
 456                 }
 457                 schedule();
 458         }
 459         finish_wait(&mddev->sb_wait, &wq);
 460 }
 461
 462 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 463 {
 464         if (bio->bi_size)
 465                 return 1;
 466
 467         complete((struct completion*)bio->bi_private);
 468         return 0;
 469 }
 470
 471 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 472                    struct page *page, int rw)
 473 {
 474         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 475         struct completion event;
 476         int ret;
 477
 478         rw |= (1 << BIO_RW_SYNC);
 479
 480         bio->bi_bdev = bdev;
 481         bio->bi_sector = sector;
 482         bio_add_page(bio, page, size, 0);
 483         init_completion(&event);
 484         bio->bi_private = &event;
 485         bio->bi_end_io = bi_complete;
 486         submit_bio(rw, bio);
 487         wait_for_completion(&event);
 488
 489         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 490         bio_put(bio);
 491         return ret;
 492 }
 493 EXPORT_SYMBOL_GPL(sync_page_io);
 494
 495 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 496 {
 497         char b[BDEVNAME_SIZE];
 498         if (!rdev->sb_page) {
 499                 MD_BUG();
 500                 return -EINVAL;
 501         }
 502         if (rdev->sb_loaded)
 503                 return 0;
 504
 505
 506         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 507                 goto fail;
 508         rdev->sb_loaded = 1;
 509         return 0;
 510
 511 fail:
 512         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 513                 bdevname(rdev->bdev,b));
 514         return -EINVAL;
 515 }
 516
 517 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 518 {
 519         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 520                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 521                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 522                 (sb1->set_uuid3 == sb2->set_uuid3))
 523
 524                 return 1;
 525
 526         return 0;
 527 }
 528
 529
 530 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 531 {
 532         int ret;
 533         mdp_super_t *tmp1, *tmp2;
 534
 535         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 536         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 537
 538         if (!tmp1 || !tmp2) {
 539                 ret = 0;
 540                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 541                 goto abort;
 542         }
 543
 544         *tmp1 = *sb1;
 545         *tmp2 = *sb2;
 546
 547         /*
 548          * nr_disks is not constant
 549          */
 550         tmp1->nr_disks = 0;
 551         tmp2->nr_disks = 0;
 552
 553         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 554                 ret = 0;
 555         else
 556                 ret = 1;
 557
 558 abort:
 559         kfree(tmp1);
 560         kfree(tmp2);
 561         return ret;
 562 }
 563
 564 static unsigned int calc_sb_csum(mdp_super_t * sb)
 565 {
 566         unsigned int disk_csum, csum;
 567
 568         disk_csum = sb->sb_csum;
 569         sb->sb_csum = 0;
 570         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 571         sb->sb_csum = disk_csum;
 572         return csum;
 573 }
 574
 575
 576 /*
 577  * Handle superblock details.
 578  * We want to be able to handle multiple superblock formats
 579  * so we have a common interface to them all, and an array of
 580  * different handlers.
 581  * We rely on user-space to write the initial superblock, and support
 582  * reading and updating of superblocks.
 583  * Interface methods are:
 584  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 585  *      loads and validates a superblock on dev.
 586  *      if refdev != NULL, compare superblocks on both devices
 587  *    Return:
 588  *      0 - dev has a superblock that is compatible with refdev
 589  *      1 - dev has a superblock that is compatible and newer than refdev
 590  *          so dev should be used as the refdev in future
 591  *     -EINVAL superblock incompatible or invalid
 592  *     -othererror e.g. -EIO
 593  *
 594  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 595  *      Verify that dev is acceptable into mddev.
 596  *       The first time, mddev->raid_disks will be 0, and data from
 597  *       dev should be merged in.  Subsequent calls check that dev
 598  *       is new enough.  Return 0 or -EINVAL
 599  *
 600  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 601  *     Update the superblock for rdev with data in mddev
 602  *     This does not write to disc.
 603  *
 604  */
 605
 606 struct super_type  {
 607         char            *name;
 608         struct module   *owner;
 609         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 610         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 611         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 612 };
 613
 614 /*
 615  * load_super for 0.90.0
 616  */
 617 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 618 {
 619         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 620         mdp_super_t *sb;
 621         int ret;
 622         sector_t sb_offset;
 623
 624         /*
 625          * Calculate the position of the superblock,
 626          * it's at the end of the disk.
 627          *
 628          * It also happens to be a multiple of 4Kb.
 629          */
 630         sb_offset = calc_dev_sboffset(rdev->bdev);
 631         rdev->sb_offset = sb_offset;
 632
 633         ret = read_disk_sb(rdev, MD_SB_BYTES);
 634         if (ret) return ret;
 635
 636         ret = -EINVAL;
 637
 638         bdevname(rdev->bdev, b);
 639         sb = (mdp_super_t*)page_address(rdev->sb_page);
 640
 641         if (sb->md_magic != MD_SB_MAGIC) {
 642                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 643                        b);
 644                 goto abort;
 645         }
 646
 647         if (sb->major_version != 0 ||
 648             sb->minor_version != 90) {
 649                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 650                         sb->major_version, sb->minor_version,
 651                         b);
 652                 goto abort;
 653         }
 654
 655         if (sb->raid_disks <= 0)
 656                 goto abort;
 657
 658         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 659                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 660                         b);
 661                 goto abort;
 662         }
 663
 664         rdev->preferred_minor = sb->md_minor;
 665         rdev->data_offset = 0;
 666         rdev->sb_size = MD_SB_BYTES;
 667
 668         if (sb->level == LEVEL_MULTIPATH)
 669                 rdev->desc_nr = -1;
 670         else
 671                 rdev->desc_nr = sb->this_disk.number;
 672
 673         if (refdev == 0)
 674                 ret = 1;
 675         else {
 676                 __u64 ev1, ev2;
 677                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 678                 if (!uuid_equal(refsb, sb)) {
 679                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 680                                 b, bdevname(refdev->bdev,b2));
 681                         goto abort;
 682                 }
 683                 if (!sb_equal(refsb, sb)) {
 684                         printk(KERN_WARNING "md: %s has same UUID"
 685                                " but different superblock to %s\n",
 686                                b, bdevname(refdev->bdev, b2));
 687                         goto abort;
 688                 }
 689                 ev1 = md_event(sb);
 690                 ev2 = md_event(refsb);
 691                 if (ev1 > ev2)
 692                         ret = 1;
 693                 else
 694                         ret = 0;
 695         }
 696         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 697
 698  abort:
 699         return ret;
 700 }
 701
 702 /*
 703  * validate_super for 0.90.0
 704  */
 705 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 706 {
 707         mdp_disk_t *desc;
 708         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 709
 710         rdev->raid_disk = -1;
 711         rdev->flags = 0;
 712         if (mddev->raid_disks == 0) {
 713                 mddev->major_version = 0;
 714                 mddev->minor_version = sb->minor_version;
 715                 mddev->patch_version = sb->patch_version;
 716                 mddev->persistent = ! sb->not_persistent;
 717                 mddev->chunk_size = sb->chunk_size;
 718                 mddev->ctime = sb->ctime;
 719                 mddev->utime = sb->utime;
 720                 mddev->level = sb->level;
 721                 mddev->clevel[0] = 0;
 722                 mddev->layout = sb->layout;
 723                 mddev->raid_disks = sb->raid_disks;
 724                 mddev->size = sb->size;
 725                 mddev->events = md_event(sb);
 726                 mddev->bitmap_offset = 0;
 727                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 728
 729                 if (sb->state & (1<<MD_SB_CLEAN))
 730                         mddev->recovery_cp = MaxSector;
 731                 else {
 732                         if (sb->events_hi == sb->cp_events_hi &&
 733                                 sb->events_lo == sb->cp_events_lo) {
 734                                 mddev->recovery_cp = sb->recovery_cp;
 735                         } else
 736                                 mddev->recovery_cp = 0;
 737                 }
 738
 739                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 740                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 741                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 742                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 743
 744                 mddev->max_disks = MD_SB_DISKS;
 745
 746                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 747                     mddev->bitmap_file == NULL) {
 748                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
 749                             && mddev->level != 10) {
 750                                 /* FIXME use a better test */
 751                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
 752                                 return -EINVAL;
 753                         }
 754                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 755                 }
 756
 757         } else if (mddev->pers == NULL) {
 758                 /* Insist on good event counter while assembling */
 759                 __u64 ev1 = md_event(sb);
 760                 ++ev1;
 761                 if (ev1 < mddev->events)
 762                         return -EINVAL;
 763         } else if (mddev->bitmap) {
 764                 /* if adding to array with a bitmap, then we can accept an
 765                  * older device ... but not too old.
 766                  */
 767                 __u64 ev1 = md_event(sb);
 768                 if (ev1 < mddev->bitmap->events_cleared)
 769                         return 0;
 770         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 771                 return 0;
 772
 773         if (mddev->level != LEVEL_MULTIPATH) {
 774                 desc = sb->disks + rdev->desc_nr;
 775
 776                 if (desc->state & (1<<MD_DISK_FAULTY))
 777                         set_bit(Faulty, &rdev->flags);
 778                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 779                          desc->raid_disk < mddev->raid_disks) {
 780                         set_bit(In_sync, &rdev->flags);
 781                         rdev->raid_disk = desc->raid_disk;
 782                 }
 783                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 784                         set_bit(WriteMostly, &rdev->flags);
 785         } else /* MULTIPATH are always insync */
 786                 set_bit(In_sync, &rdev->flags);
 787         return 0;
 788 }
 789
 790 /*
 791  * sync_super for 0.90.0
 792  */
 793 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 794 {
 795         mdp_super_t *sb;
 796         struct list_head *tmp;
 797         mdk_rdev_t *rdev2;
 798         int next_spare = mddev->raid_disks;
 799
 800
 801         /* make rdev->sb match mddev data..
 802          *
 803          * 1/ zero out disks
 804          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 805          * 3/ any empty disks < next_spare become removed
 806          *
 807          * disks[0] gets initialised to REMOVED because
 808          * we cannot be sure from other fields if it has
 809          * been initialised or not.
 810          */
 811         int i;
 812         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 813
 814         rdev->sb_size = MD_SB_BYTES;
 815
 816         sb = (mdp_super_t*)page_address(rdev->sb_page);
 817
 818         memset(sb, 0, sizeof(*sb));
 819
 820         sb->md_magic = MD_SB_MAGIC;
 821         sb->major_version = mddev->major_version;
 822         sb->minor_version = mddev->minor_version;
 823         sb->patch_version = mddev->patch_version;
 824         sb->gvalid_words  = 0; /* ignored */
 825         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 826         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 827         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 828         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 829
 830         sb->ctime = mddev->ctime;
 831         sb->level = mddev->level;
 832         sb->size  = mddev->size;
 833         sb->raid_disks = mddev->raid_disks;
 834         sb->md_minor = mddev->md_minor;
 835         sb->not_persistent = !mddev->persistent;
 836         sb->utime = mddev->utime;
 837         sb->state = 0;
 838         sb->events_hi = (mddev->events>>32);
 839         sb->events_lo = (u32)mddev->events;
 840
 841         if (mddev->in_sync)
 842         {
 843                 sb->recovery_cp = mddev->recovery_cp;
 844                 sb->cp_events_hi = (mddev->events>>32);
 845                 sb->cp_events_lo = (u32)mddev->events;
 846                 if (mddev->recovery_cp == MaxSector)
 847                         sb->state = (1<< MD_SB_CLEAN);
 848         } else
 849                 sb->recovery_cp = 0;
 850
 851         sb->layout = mddev->layout;
 852         sb->chunk_size = mddev->chunk_size;
 853
 854         if (mddev->bitmap && mddev->bitmap_file == NULL)
 855                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 856
 857         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 858         ITERATE_RDEV(mddev,rdev2,tmp) {
 859                 mdp_disk_t *d;
 860                 int desc_nr;
 861                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 862                     && !test_bit(Faulty, &rdev2->flags))
 863                         desc_nr = rdev2->raid_disk;
 864                 else
 865                         desc_nr = next_spare++;
 866                 rdev2->desc_nr = desc_nr;
 867                 d = &sb->disks[rdev2->desc_nr];
 868                 nr_disks++;
 869                 d->number = rdev2->desc_nr;
 870                 d->major = MAJOR(rdev2->bdev->bd_dev);
 871                 d->minor = MINOR(rdev2->bdev->bd_dev);
 872                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 873                     && !test_bit(Faulty, &rdev2->flags))
 874                         d->raid_disk = rdev2->raid_disk;
 875                 else
 876                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 877                 if (test_bit(Faulty, &rdev2->flags)) {
 878                         d->state = (1<<MD_DISK_FAULTY);
 879                         failed++;
 880                 } else if (test_bit(In_sync, &rdev2->flags)) {
 881                         d->state = (1<<MD_DISK_ACTIVE);
 882                         d->state |= (1<<MD_DISK_SYNC);
 883                         active++;
 884                         working++;
 885                 } else {
 886                         d->state = 0;
 887                         spare++;
 888                         working++;
 889                 }
 890                 if (test_bit(WriteMostly, &rdev2->flags))
 891                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 892         }
 893         /* now set the "removed" and "faulty" bits on any missing devices */
 894         for (i=0 ; i < mddev->raid_disks ; i++) {
 895                 mdp_disk_t *d = &sb->disks[i];
 896                 if (d->state == 0 && d->number == 0) {
 897                         d->number = i;
 898                         d->raid_disk = i;
 899                         d->state = (1<<MD_DISK_REMOVED);
 900                         d->state |= (1<<MD_DISK_FAULTY);
 901                         failed++;
 902                 }
 903         }
 904         sb->nr_disks = nr_disks;
 905         sb->active_disks = active;
 906         sb->working_disks = working;
 907         sb->failed_disks = failed;
 908         sb->spare_disks = spare;
 909
 910         sb->this_disk = sb->disks[rdev->desc_nr];
 911         sb->sb_csum = calc_sb_csum(sb);
 912 }
 913
 914 /*
 915  * version 1 superblock
 916  */
 917
 918 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 919 {
 920         unsigned int disk_csum, csum;
 921         unsigned long long newcsum;
 922         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 923         unsigned int *isuper = (unsigned int*)sb;
 924         int i;
 925
 926         disk_csum = sb->sb_csum;
 927         sb->sb_csum = 0;
 928         newcsum = 0;
 929         for (i=0; size>=4; size -= 4 )
 930                 newcsum += le32_to_cpu(*isuper++);
 931
 932         if (size == 2)
 933                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 934
 935         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 936         sb->sb_csum = disk_csum;
 937         return cpu_to_le32(csum);
 938 }
 939
 940 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 941 {
 942         struct mdp_superblock_1 *sb;
 943         int ret;
 944         sector_t sb_offset;
 945         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 946         int bmask;
 947
 948         /*
 949          * Calculate the position of the superblock.
 950          * It is always aligned to a 4K boundary and
 951          * depeding on minor_version, it can be:
 952          * 0: At least 8K, but less than 12K, from end of device
 953          * 1: At start of device
 954          * 2: 4K from start of device.
 955          */
 956         switch(minor_version) {
 957         case 0:
 958                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 959                 sb_offset -= 8*2;
 960                 sb_offset &= ~(sector_t)(4*2-1);
 961                 /* convert from sectors to K */
 962                 sb_offset /= 2;
 963                 break;
 964         case 1:
 965                 sb_offset = 0;
 966                 break;
 967         case 2:
 968                 sb_offset = 4;
 969                 break;
 970         default:
 971                 return -EINVAL;
 972         }
 973         rdev->sb_offset = sb_offset;
 974
 975         /* superblock is rarely larger than 1K, but it can be larger,
 976          * and it is safe to read 4k, so we do that
 977          */
 978         ret = read_disk_sb(rdev, 4096);
 979         if (ret) return ret;
 980
 981
 982         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 983
 984         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 985             sb->major_version != cpu_to_le32(1) ||
 986             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 987             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 988             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 989                 return -EINVAL;
 990
 991         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 992                 printk("md: invalid superblock checksum on %s\n",
 993                         bdevname(rdev->bdev,b));
 994                 return -EINVAL;
 995         }
 996         if (le64_to_cpu(sb->data_size) < 10) {
 997                 printk("md: data_size too small on %s\n",
 998                        bdevname(rdev->bdev,b));
 999                 return -EINVAL;
1000         }
1001         rdev->preferred_minor = 0xffff;
1002         rdev->data_offset = le64_to_cpu(sb->data_offset);
1003         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1004
1005         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1006         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1007         if (rdev->sb_size & bmask)
1008                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1009
1010         if (refdev == 0)
1011                 return 1;
1012         else {
1013                 __u64 ev1, ev2;
1014                 struct mdp_superblock_1 *refsb =
1015                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1016
1017                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1018                     sb->level != refsb->level ||
1019                     sb->layout != refsb->layout ||
1020                     sb->chunksize != refsb->chunksize) {
1021                         printk(KERN_WARNING "md: %s has strangely different"
1022                                 " superblock to %s\n",
1023                                 bdevname(rdev->bdev,b),
1024                                 bdevname(refdev->bdev,b2));
1025                         return -EINVAL;
1026                 }
1027                 ev1 = le64_to_cpu(sb->events);
1028                 ev2 = le64_to_cpu(refsb->events);
1029
1030                 if (ev1 > ev2)
1031                         return 1;
1032         }
1033         if (minor_version)
1034                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1035         else
1036                 rdev->size = rdev->sb_offset;
1037         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1038                 return -EINVAL;
1039         rdev->size = le64_to_cpu(sb->data_size)/2;
1040         if (le32_to_cpu(sb->chunksize))
1041                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1042         return 0;
1043 }
1044
1045 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1046 {
1047         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1048
1049         rdev->raid_disk = -1;
1050         rdev->flags = 0;
1051         if (mddev->raid_disks == 0) {
1052                 mddev->major_version = 1;
1053                 mddev->patch_version = 0;
1054                 mddev->persistent = 1;
1055                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1056                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1057                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1058                 mddev->level = le32_to_cpu(sb->level);
1059                 mddev->clevel[0] = 0;
1060                 mddev->layout = le32_to_cpu(sb->layout);
1061                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1062                 mddev->size = le64_to_cpu(sb->size)/2;
1063                 mddev->events = le64_to_cpu(sb->events);
1064                 mddev->bitmap_offset = 0;
1065                 mddev->default_bitmap_offset = 1024;
1066
1067                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1068                 memcpy(mddev->uuid, sb->set_uuid, 16);
1069
1070                 mddev->max_disks =  (4096-256)/2;
1071
1072                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1073                     mddev->bitmap_file == NULL ) {
1074                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1075                             && mddev->level != 10) {
1076                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1077                                 return -EINVAL;
1078                         }
1079                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1080                 }
1081         } else if (mddev->pers == NULL) {
1082                 /* Insist of good event counter while assembling */
1083                 __u64 ev1 = le64_to_cpu(sb->events);
1084                 ++ev1;
1085                 if (ev1 < mddev->events)
1086                         return -EINVAL;
1087         } else if (mddev->bitmap) {
1088                 /* If adding to array with a bitmap, then we can accept an
1089                  * older device, but not too old.
1090                  */
1091                 __u64 ev1 = le64_to_cpu(sb->events);
1092                 if (ev1 < mddev->bitmap->events_cleared)
1093                         return 0;
1094         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1095                 return 0;
1096
1097         if (mddev->level != LEVEL_MULTIPATH) {
1098                 int role;
1099                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1100                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1101                 switch(role) {
1102                 case 0xffff: /* spare */
1103                         break;
1104                 case 0xfffe: /* faulty */
1105                         set_bit(Faulty, &rdev->flags);
1106                         break;
1107                 default:
1108                         set_bit(In_sync, &rdev->flags);
1109                         rdev->raid_disk = role;
1110                         break;
1111                 }
1112                 if (sb->devflags & WriteMostly1)
1113                         set_bit(WriteMostly, &rdev->flags);
1114         } else /* MULTIPATH are always insync */
1115                 set_bit(In_sync, &rdev->flags);
1116
1117         return 0;
1118 }
1119
1120 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1121 {
1122         struct mdp_superblock_1 *sb;
1123         struct list_head *tmp;
1124         mdk_rdev_t *rdev2;
1125         int max_dev, i;
1126         /* make rdev->sb match mddev and rdev data. */
1127
1128         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1129
1130         sb->feature_map = 0;
1131         sb->pad0 = 0;
1132         memset(sb->pad1, 0, sizeof(sb->pad1));
1133         memset(sb->pad2, 0, sizeof(sb->pad2));
1134         memset(sb->pad3, 0, sizeof(sb->pad3));
1135
1136         sb->utime = cpu_to_le64((__u64)mddev->utime);
1137         sb->events = cpu_to_le64(mddev->events);
1138         if (mddev->in_sync)
1139                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1140         else
1141                 sb->resync_offset = cpu_to_le64(0);
1142
1143         sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1144
1145         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1146                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1147                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1148         }
1149
1150         max_dev = 0;
1151         ITERATE_RDEV(mddev,rdev2,tmp)
1152                 if (rdev2->desc_nr+1 > max_dev)
1153                         max_dev = rdev2->desc_nr+1;
1154
1155         sb->max_dev = cpu_to_le32(max_dev);
1156         for (i=0; i<max_dev;i++)
1157                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1158
1159         ITERATE_RDEV(mddev,rdev2,tmp) {
1160                 i = rdev2->desc_nr;
1161                 if (test_bit(Faulty, &rdev2->flags))
1162                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1163                 else if (test_bit(In_sync, &rdev2->flags))
1164                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1165                 else
1166                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1167         }
1168
1169         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1170         sb->sb_csum = calc_sb_1_csum(sb);
1171 }
1172
1173
1174 static struct super_type super_types[] = {
1175         [0] = {
1176                 .name   = "0.90.0",
1177                 .owner  = THIS_MODULE,
1178                 .load_super     = super_90_load,
1179                 .validate_super = super_90_validate,
1180                 .sync_super     = super_90_sync,
1181         },
1182         [1] = {
1183                 .name   = "md-1",
1184                 .owner  = THIS_MODULE,
1185                 .load_super     = super_1_load,
1186                 .validate_super = super_1_validate,
1187                 .sync_super     = super_1_sync,
1188         },
1189 };
1190
1191 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1192 {
1193         struct list_head *tmp;
1194         mdk_rdev_t *rdev;
1195
1196         ITERATE_RDEV(mddev,rdev,tmp)
1197                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1198                         return rdev;
1199
1200         return NULL;
1201 }
1202
1203 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1204 {
1205         struct list_head *tmp;
1206         mdk_rdev_t *rdev;
1207
1208         ITERATE_RDEV(mddev1,rdev,tmp)
1209                 if (match_dev_unit(mddev2, rdev))
1210                         return 1;
1211
1212         return 0;
1213 }
1214
1215 static LIST_HEAD(pending_raid_disks);
1216
1217 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1218 {
1219         mdk_rdev_t *same_pdev;
1220         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1221         struct kobject *ko;
1222
1223         if (rdev->mddev) {
1224                 MD_BUG();
1225                 return -EINVAL;
1226         }
1227         same_pdev = match_dev_unit(mddev, rdev);
1228         if (same_pdev)
1229                 printk(KERN_WARNING
1230                         "%s: WARNING: %s appears to be on the same physical"
1231                         " disk as %s. True\n     protection against single-disk"
1232                         " failure might be compromised.\n",
1233                         mdname(mddev), bdevname(rdev->bdev,b),
1234                         bdevname(same_pdev->bdev,b2));
1235
1236         /* Verify rdev->desc_nr is unique.
1237          * If it is -1, assign a free number, else
1238          * check number is not in use
1239          */
1240         if (rdev->desc_nr < 0) {
1241                 int choice = 0;
1242                 if (mddev->pers) choice = mddev->raid_disks;
1243                 while (find_rdev_nr(mddev, choice))
1244                         choice++;
1245                 rdev->desc_nr = choice;
1246         } else {
1247                 if (find_rdev_nr(mddev, rdev->desc_nr))
1248                         return -EBUSY;
1249         }
1250         bdevname(rdev->bdev,b);
1251         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1252                 return -ENOMEM;
1253
1254         list_add(&rdev->same_set, &mddev->disks);
1255         rdev->mddev = mddev;
1256         printk(KERN_INFO "md: bind<%s>\n", b);
1257
1258         rdev->kobj.parent = &mddev->kobj;
1259         kobject_add(&rdev->kobj);
1260
1261         if (rdev->bdev->bd_part)
1262                 ko = &rdev->bdev->bd_part->kobj;
1263         else
1264                 ko = &rdev->bdev->bd_disk->kobj;
1265         sysfs_create_link(&rdev->kobj, ko, "block");
1266         return 0;
1267 }
1268
1269 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1270 {
1271         char b[BDEVNAME_SIZE];
1272         if (!rdev->mddev) {
1273                 MD_BUG();
1274                 return;
1275         }
1276         list_del_init(&rdev->same_set);
1277         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1278         rdev->mddev = NULL;
1279         sysfs_remove_link(&rdev->kobj, "block");
1280         kobject_del(&rdev->kobj);
1281 }
1282
1283 /*
1284  * prevent the device from being mounted, repartitioned or
1285  * otherwise reused by a RAID array (or any other kernel
1286  * subsystem), by bd_claiming the device.
1287  */
1288 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1289 {
1290         int err = 0;
1291         struct block_device *bdev;
1292         char b[BDEVNAME_SIZE];
1293
1294         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1295         if (IS_ERR(bdev)) {
1296                 printk(KERN_ERR "md: could not open %s.\n",
1297                         __bdevname(dev, b));
1298                 return PTR_ERR(bdev);
1299         }
1300         err = bd_claim(bdev, rdev);
1301         if (err) {
1302                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1303                         bdevname(bdev, b));
1304                 blkdev_put(bdev);
1305                 return err;
1306         }
1307         rdev->bdev = bdev;
1308         return err;
1309 }
1310
1311 static void unlock_rdev(mdk_rdev_t *rdev)
1312 {
1313         struct block_device *bdev = rdev->bdev;
1314         rdev->bdev = NULL;
1315         if (!bdev)
1316                 MD_BUG();
1317         bd_release(bdev);
1318         blkdev_put(bdev);
1319 }
1320
1321 void md_autodetect_dev(dev_t dev);
1322
1323 static void export_rdev(mdk_rdev_t * rdev)
1324 {
1325         char b[BDEVNAME_SIZE];
1326         printk(KERN_INFO "md: export_rdev(%s)\n",
1327                 bdevname(rdev->bdev,b));
1328         if (rdev->mddev)
1329                 MD_BUG();
1330         free_disk_sb(rdev);
1331         list_del_init(&rdev->same_set);
1332 #ifndef MODULE
1333         md_autodetect_dev(rdev->bdev->bd_dev);
1334 #endif
1335         unlock_rdev(rdev);
1336         kobject_put(&rdev->kobj);
1337 }
1338
1339 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1340 {
1341         unbind_rdev_from_array(rdev);
1342         export_rdev(rdev);
1343 }
1344
1345 static void export_array(mddev_t *mddev)
1346 {
1347         struct list_head *tmp;
1348         mdk_rdev_t *rdev;
1349
1350         ITERATE_RDEV(mddev,rdev,tmp) {
1351                 if (!rdev->mddev) {
1352                         MD_BUG();
1353                         continue;
1354                 }
1355                 kick_rdev_from_array(rdev);
1356         }
1357         if (!list_empty(&mddev->disks))
1358                 MD_BUG();
1359         mddev->raid_disks = 0;
1360         mddev->major_version = 0;
1361 }
1362
1363 static void print_desc(mdp_disk_t *desc)
1364 {
1365         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1366                 desc->major,desc->minor,desc->raid_disk,desc->state);
1367 }
1368
1369 static void print_sb(mdp_super_t *sb)
1370 {
1371         int i;
1372
1373         printk(KERN_INFO
1374                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1375                 sb->major_version, sb->minor_version, sb->patch_version,
1376                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1377                 sb->ctime);
1378         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1379                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1380                 sb->md_minor, sb->layout, sb->chunk_size);
1381         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1382                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1383                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1384                 sb->failed_disks, sb->spare_disks,
1385                 sb->sb_csum, (unsigned long)sb->events_lo);
1386
1387         printk(KERN_INFO);
1388         for (i = 0; i < MD_SB_DISKS; i++) {
1389                 mdp_disk_t *desc;
1390
1391                 desc = sb->disks + i;
1392                 if (desc->number || desc->major || desc->minor ||
1393                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1394                         printk("     D %2d: ", i);
1395                         print_desc(desc);
1396                 }
1397         }
1398         printk(KERN_INFO "md:     THIS: ");
1399         print_desc(&sb->this_disk);
1400
1401 }
1402
1403 static void print_rdev(mdk_rdev_t *rdev)
1404 {
1405         char b[BDEVNAME_SIZE];
1406         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1407                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1408                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1409                 rdev->desc_nr);
1410         if (rdev->sb_loaded) {
1411                 printk(KERN_INFO "md: rdev superblock:\n");
1412                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1413         } else
1414                 printk(KERN_INFO "md: no rdev superblock!\n");
1415 }
1416
1417 void md_print_devices(void)
1418 {
1419         struct list_head *tmp, *tmp2;
1420         mdk_rdev_t *rdev;
1421         mddev_t *mddev;
1422         char b[BDEVNAME_SIZE];
1423
1424         printk("\n");
1425         printk("md:     **********************************\n");
1426         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1427         printk("md:     **********************************\n");
1428         ITERATE_MDDEV(mddev,tmp) {
1429
1430                 if (mddev->bitmap)
1431                         bitmap_print_sb(mddev->bitmap);
1432                 else
1433                         printk("%s: ", mdname(mddev));
1434                 ITERATE_RDEV(mddev,rdev,tmp2)
1435                         printk("<%s>", bdevname(rdev->bdev,b));
1436                 printk("\n");
1437
1438                 ITERATE_RDEV(mddev,rdev,tmp2)
1439                         print_rdev(rdev);
1440         }
1441         printk("md:     **********************************\n");
1442         printk("\n");
1443 }
1444
1445
1446 static void sync_sbs(mddev_t * mddev)
1447 {
1448         mdk_rdev_t *rdev;
1449         struct list_head *tmp;
1450
1451         ITERATE_RDEV(mddev,rdev,tmp) {
1452                 super_types[mddev->major_version].
1453                         sync_super(mddev, rdev);
1454                 rdev->sb_loaded = 1;
1455         }
1456 }
1457
1458 static void md_update_sb(mddev_t * mddev)
1459 {
1460         int err;
1461         struct list_head *tmp;
1462         mdk_rdev_t *rdev;
1463         int sync_req;
1464
1465 repeat:
1466         spin_lock_irq(&mddev->write_lock);
1467         sync_req = mddev->in_sync;
1468         mddev->utime = get_seconds();
1469         mddev->events ++;
1470
1471         if (!mddev->events) {
1472                 /*
1473                  * oops, this 64-bit counter should never wrap.
1474                  * Either we are in around ~1 trillion A.C., assuming
1475                  * 1 reboot per second, or we have a bug:
1476                  */
1477                 MD_BUG();
1478                 mddev->events --;
1479         }
1480         mddev->sb_dirty = 2;
1481         sync_sbs(mddev);
1482
1483         /*
1484          * do not write anything to disk if using
1485          * nonpersistent superblocks
1486          */
1487         if (!mddev->persistent) {
1488                 mddev->sb_dirty = 0;
1489                 spin_unlock_irq(&mddev->write_lock);
1490                 wake_up(&mddev->sb_wait);
1491                 return;
1492         }
1493         spin_unlock_irq(&mddev->write_lock);
1494
1495         dprintk(KERN_INFO
1496                 "md: updating %s RAID superblock on device (in sync %d)\n",
1497                 mdname(mddev),mddev->in_sync);
1498
1499         err = bitmap_update_sb(mddev->bitmap);
1500         ITERATE_RDEV(mddev,rdev,tmp) {
1501                 char b[BDEVNAME_SIZE];
1502                 dprintk(KERN_INFO "md: ");
1503                 if (test_bit(Faulty, &rdev->flags))
1504                         dprintk("(skipping faulty ");
1505
1506                 dprintk("%s ", bdevname(rdev->bdev,b));
1507                 if (!test_bit(Faulty, &rdev->flags)) {
1508                         md_super_write(mddev,rdev,
1509                                        rdev->sb_offset<<1, rdev->sb_size,
1510                                        rdev->sb_page);
1511                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1512                                 bdevname(rdev->bdev,b),
1513                                 (unsigned long long)rdev->sb_offset);
1514
1515                 } else
1516                         dprintk(")\n");
1517                 if (mddev->level == LEVEL_MULTIPATH)
1518                         /* only need to write one superblock... */
1519                         break;
1520         }
1521         md_super_wait(mddev);
1522         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1523
1524         spin_lock_irq(&mddev->write_lock);
1525         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1526                 /* have to write it out again */
1527                 spin_unlock_irq(&mddev->write_lock);
1528                 goto repeat;
1529         }
1530         mddev->sb_dirty = 0;
1531         spin_unlock_irq(&mddev->write_lock);
1532         wake_up(&mddev->sb_wait);
1533
1534 }
1535
1536 /* words written to sysfs files may, or my not, be \n terminated.
1537  * We want to accept with case. For this we use cmd_match.
1538  */
1539 static int cmd_match(const char *cmd, const char *str)
1540 {
1541         /* See if cmd, written into a sysfs file, matches
1542          * str.  They must either be the same, or cmd can
1543          * have a trailing newline
1544          */
1545         while (*cmd && *str && *cmd == *str) {
1546                 cmd++;
1547                 str++;
1548         }
1549         if (*cmd == '\n')
1550                 cmd++;
1551         if (*str || *cmd)
1552                 return 0;
1553         return 1;
1554 }
1555
1556 struct rdev_sysfs_entry {
1557         struct attribute attr;
1558         ssize_t (*show)(mdk_rdev_t *, char *);
1559         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1560 };
1561
1562 static ssize_t
1563 state_show(mdk_rdev_t *rdev, char *page)
1564 {
1565         char *sep = "";
1566         int len=0;
1567
1568         if (test_bit(Faulty, &rdev->flags)) {
1569                 len+= sprintf(page+len, "%sfaulty",sep);
1570                 sep = ",";
1571         }
1572         if (test_bit(In_sync, &rdev->flags)) {
1573                 len += sprintf(page+len, "%sin_sync",sep);
1574                 sep = ",";
1575         }
1576         if (!test_bit(Faulty, &rdev->flags) &&
1577             !test_bit(In_sync, &rdev->flags)) {
1578                 len += sprintf(page+len, "%sspare", sep);
1579                 sep = ",";
1580         }
1581         return len+sprintf(page+len, "\n");
1582 }
1583
1584 static struct rdev_sysfs_entry
1585 rdev_state = __ATTR_RO(state);
1586
1587 static ssize_t
1588 super_show(mdk_rdev_t *rdev, char *page)
1589 {
1590         if (rdev->sb_loaded && rdev->sb_size) {
1591                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1592                 return rdev->sb_size;
1593         } else
1594                 return 0;
1595 }
1596 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1597
1598 static ssize_t
1599 errors_show(mdk_rdev_t *rdev, char *page)
1600 {
1601         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1602 }
1603
1604 static ssize_t
1605 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1606 {
1607         char *e;
1608         unsigned long n = simple_strtoul(buf, &e, 10);
1609         if (*buf && (*e == 0 || *e == '\n')) {
1610                 atomic_set(&rdev->corrected_errors, n);
1611                 return len;
1612         }
1613         return -EINVAL;
1614 }
1615 static struct rdev_sysfs_entry rdev_errors =
1616 __ATTR(errors, 0644, errors_show, errors_store);
1617
1618 static struct attribute *rdev_default_attrs[] = {
1619         &rdev_state.attr,
1620         &rdev_super.attr,
1621         &rdev_errors.attr,
1622         NULL,
1623 };
1624 static ssize_t
1625 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1626 {
1627         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1628         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1629
1630         if (!entry->show)
1631                 return -EIO;
1632         return entry->show(rdev, page);
1633 }
1634
1635 static ssize_t
1636 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1637               const char *page, size_t length)
1638 {
1639         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1640         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1641
1642         if (!entry->store)
1643                 return -EIO;
1644         return entry->store(rdev, page, length);
1645 }
1646
1647 static void rdev_free(struct kobject *ko)
1648 {
1649         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1650         kfree(rdev);
1651 }
1652 static struct sysfs_ops rdev_sysfs_ops = {
1653         .show           = rdev_attr_show,
1654         .store          = rdev_attr_store,
1655 };
1656 static struct kobj_type rdev_ktype = {
1657         .release        = rdev_free,
1658         .sysfs_ops      = &rdev_sysfs_ops,
1659         .default_attrs  = rdev_default_attrs,
1660 };
1661
1662 /*
1663  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1664  *
1665  * mark the device faulty if:
1666  *
1667  *   - the device is nonexistent (zero size)
1668  *   - the device has no valid superblock
1669  *
1670  * a faulty rdev _never_ has rdev->sb set.
1671  */
1672 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1673 {
1674         char b[BDEVNAME_SIZE];
1675         int err;
1676         mdk_rdev_t *rdev;
1677         sector_t size;
1678
1679         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1680         if (!rdev) {
1681                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1682                 return ERR_PTR(-ENOMEM);
1683         }
1684
1685         if ((err = alloc_disk_sb(rdev)))
1686                 goto abort_free;
1687
1688         err = lock_rdev(rdev, newdev);
1689         if (err)
1690                 goto abort_free;
1691
1692         rdev->kobj.parent = NULL;
1693         rdev->kobj.ktype = &rdev_ktype;
1694         kobject_init(&rdev->kobj);
1695
1696         rdev->desc_nr = -1;
1697         rdev->flags = 0;
1698         rdev->data_offset = 0;
1699         atomic_set(&rdev->nr_pending, 0);
1700         atomic_set(&rdev->read_errors, 0);
1701         atomic_set(&rdev->corrected_errors, 0);
1702
1703         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1704         if (!size) {
1705                 printk(KERN_WARNING
1706                         "md: %s has zero or unknown size, marking faulty!\n",
1707                         bdevname(rdev->bdev,b));
1708                 err = -EINVAL;
1709                 goto abort_free;
1710         }
1711
1712         if (super_format >= 0) {
1713                 err = super_types[super_format].
1714                         load_super(rdev, NULL, super_minor);
1715                 if (err == -EINVAL) {
1716                         printk(KERN_WARNING
1717                                 "md: %s has invalid sb, not importing!\n",
1718                                 bdevname(rdev->bdev,b));
1719                         goto abort_free;
1720                 }
1721                 if (err < 0) {
1722                         printk(KERN_WARNING
1723                                 "md: could not read %s's sb, not importing!\n",
1724                                 bdevname(rdev->bdev,b));
1725                         goto abort_free;
1726                 }
1727         }
1728         INIT_LIST_HEAD(&rdev->same_set);
1729
1730         return rdev;
1731
1732 abort_free:
1733         if (rdev->sb_page) {
1734                 if (rdev->bdev)
1735                         unlock_rdev(rdev);
1736                 free_disk_sb(rdev);
1737         }
1738         kfree(rdev);
1739         return ERR_PTR(err);
1740 }
1741
1742 /*
1743  * Check a full RAID array for plausibility
1744  */
1745
1746
1747 static void analyze_sbs(mddev_t * mddev)
1748 {
1749         int i;
1750         struct list_head *tmp;
1751         mdk_rdev_t *rdev, *freshest;
1752         char b[BDEVNAME_SIZE];
1753
1754         freshest = NULL;
1755         ITERATE_RDEV(mddev,rdev,tmp)
1756                 switch (super_types[mddev->major_version].
1757                         load_super(rdev, freshest, mddev->minor_version)) {
1758                 case 1:
1759                         freshest = rdev;
1760                         break;
1761                 case 0:
1762                         break;
1763                 default:
1764                         printk( KERN_ERR \
1765                                 "md: fatal superblock inconsistency in %s"
1766                                 " -- removing from array\n",
1767                                 bdevname(rdev->bdev,b));
1768                         kick_rdev_from_array(rdev);
1769                 }
1770
1771
1772         super_types[mddev->major_version].
1773                 validate_super(mddev, freshest);
1774
1775         i = 0;
1776         ITERATE_RDEV(mddev,rdev,tmp) {
1777                 if (rdev != freshest)
1778                         if (super_types[mddev->major_version].
1779                             validate_super(mddev, rdev)) {
1780                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1781                                         " from array!\n",
1782                                         bdevname(rdev->bdev,b));
1783                                 kick_rdev_from_array(rdev);
1784                                 continue;
1785                         }
1786                 if (mddev->level == LEVEL_MULTIPATH) {
1787                         rdev->desc_nr = i++;
1788                         rdev->raid_disk = rdev->desc_nr;
1789                         set_bit(In_sync, &rdev->flags);
1790                 }
1791         }
1792
1793
1794
1795         if (mddev->recovery_cp != MaxSector &&
1796             mddev->level >= 1)
1797                 printk(KERN_ERR "md: %s: raid array is not clean"
1798                        " -- starting background reconstruction\n",
1799                        mdname(mddev));
1800
1801 }
1802
1803 static ssize_t
1804 level_show(mddev_t *mddev, char *page)
1805 {
1806         struct mdk_personality *p = mddev->pers;
1807         if (p)
1808                 return sprintf(page, "%s\n", p->name);
1809         else if (mddev->clevel[0])
1810                 return sprintf(page, "%s\n", mddev->clevel);
1811         else if (mddev->level != LEVEL_NONE)
1812                 return sprintf(page, "%d\n", mddev->level);
1813         else
1814                 return 0;
1815 }
1816
1817 static ssize_t
1818 level_store(mddev_t *mddev, const char *buf, size_t len)
1819 {
1820         int rv = len;
1821         if (mddev->pers)
1822                 return -EBUSY;
1823         if (len == 0)
1824                 return 0;
1825         if (len >= sizeof(mddev->clevel))
1826                 return -ENOSPC;
1827         strncpy(mddev->clevel, buf, len);
1828         if (mddev->clevel[len-1] == '\n')
1829                 len--;
1830         mddev->clevel[len] = 0;
1831         mddev->level = LEVEL_NONE;
1832         return rv;
1833 }
1834
1835 static struct md_sysfs_entry md_level =
1836 __ATTR(level, 0644, level_show, level_store);
1837
1838 static ssize_t
1839 raid_disks_show(mddev_t *mddev, char *page)
1840 {
1841         if (mddev->raid_disks == 0)
1842                 return 0;
1843         return sprintf(page, "%d\n", mddev->raid_disks);
1844 }
1845
1846 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1847
1848 static ssize_t
1849 chunk_size_show(mddev_t *mddev, char *page)
1850 {
1851         return sprintf(page, "%d\n", mddev->chunk_size);
1852 }
1853
1854 static ssize_t
1855 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1856 {
1857         /* can only set chunk_size if array is not yet active */
1858         char *e;
1859         unsigned long n = simple_strtoul(buf, &e, 10);
1860
1861         if (mddev->pers)
1862                 return -EBUSY;
1863         if (!*buf || (*e && *e != '\n'))
1864                 return -EINVAL;
1865
1866         mddev->chunk_size = n;
1867         return len;
1868 }
1869 static struct md_sysfs_entry md_chunk_size =
1870 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
1871
1872
1873 static ssize_t
1874 size_show(mddev_t *mddev, char *page)
1875 {
1876         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
1877 }
1878
1879 static int update_size(mddev_t *mddev, unsigned long size);
1880
1881 static ssize_t
1882 size_store(mddev_t *mddev, const char *buf, size_t len)
1883 {
1884         /* If array is inactive, we can reduce the component size, but
1885          * not increase it (except from 0).
1886          * If array is active, we can try an on-line resize
1887          */
1888         char *e;
1889         int err = 0;
1890         unsigned long long size = simple_strtoull(buf, &e, 10);
1891         if (!*buf || *buf == '\n' ||
1892             (*e && *e != '\n'))
1893                 return -EINVAL;
1894
1895         if (mddev->pers) {
1896                 err = update_size(mddev, size);
1897                 md_update_sb(mddev);
1898         } else {
1899                 if (mddev->size == 0 ||
1900                     mddev->size > size)
1901                         mddev->size = size;
1902                 else
1903                         err = -ENOSPC;
1904         }
1905         return err ? err : len;
1906 }
1907
1908 static struct md_sysfs_entry md_size =
1909 __ATTR(component_size, 0644, size_show, size_store);
1910
1911
1912 /* Metdata version.
1913  * This is either 'none' for arrays with externally managed metadata,
1914  * or N.M for internally known formats
1915  */
1916 static ssize_t
1917 metadata_show(mddev_t *mddev, char *page)
1918 {
1919         if (mddev->persistent)
1920                 return sprintf(page, "%d.%d\n",
1921                                mddev->major_version, mddev->minor_version);
1922         else
1923                 return sprintf(page, "none\n");
1924 }
1925
1926 static ssize_t
1927 metadata_store(mddev_t *mddev, const char *buf, size_t len)
1928 {
1929         int major, minor;
1930         char *e;
1931         if (!list_empty(&mddev->disks))
1932                 return -EBUSY;
1933
1934         if (cmd_match(buf, "none")) {
1935                 mddev->persistent = 0;
1936                 mddev->major_version = 0;
1937                 mddev->minor_version = 90;
1938                 return len;
1939         }
1940         major = simple_strtoul(buf, &e, 10);
1941         if (e==buf || *e != '.')
1942                 return -EINVAL;
1943         buf = e+1;
1944         minor = simple_strtoul(buf, &e, 10);
1945         if (e==buf || *e != '\n')
1946                 return -EINVAL;
1947         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
1948             super_types[major].name == NULL)
1949                 return -ENOENT;
1950         mddev->major_version = major;
1951         mddev->minor_version = minor;
1952         mddev->persistent = 1;
1953         return len;
1954 }
1955
1956 static struct md_sysfs_entry md_metadata =
1957 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
1958
1959 static ssize_t
1960 action_show(mddev_t *mddev, char *page)
1961 {
1962         char *type = "idle";
1963         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1964             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1965                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1966                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1967                                 type = "resync";
1968                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1969                                 type = "check";
1970                         else
1971                                 type = "repair";
1972                 } else
1973                         type = "recover";
1974         }
1975         return sprintf(page, "%s\n", type);
1976 }
1977
1978 static ssize_t
1979 action_store(mddev_t *mddev, const char *page, size_t len)
1980 {
1981         if (!mddev->pers || !mddev->pers->sync_request)
1982                 return -EINVAL;
1983
1984         if (cmd_match(page, "idle")) {
1985                 if (mddev->sync_thread) {
1986                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1987                         md_unregister_thread(mddev->sync_thread);
1988                         mddev->sync_thread = NULL;
1989                         mddev->recovery = 0;
1990                 }
1991         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1992                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1993                 return -EBUSY;
1994         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
1995                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1996         else {
1997                 if (cmd_match(page, "check"))
1998                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1999                 else if (cmd_match(page, "repair"))
2000                         return -EINVAL;
2001                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2002                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2003         }
2004         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2005         md_wakeup_thread(mddev->thread);
2006         return len;
2007 }
2008
2009 static ssize_t
2010 mismatch_cnt_show(mddev_t *mddev, char *page)
2011 {
2012         return sprintf(page, "%llu\n",
2013                        (unsigned long long) mddev->resync_mismatches);
2014 }
2015
2016 static struct md_sysfs_entry
2017 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2018
2019
2020 static struct md_sysfs_entry
2021 md_mismatches = __ATTR_RO(mismatch_cnt);
2022
2023 static struct attribute *md_default_attrs[] = {
2024         &md_level.attr,
2025         &md_raid_disks.attr,
2026         &md_chunk_size.attr,
2027         &md_size.attr,
2028         &md_metadata.attr,
2029         NULL,
2030 };
2031
2032 static struct attribute *md_redundancy_attrs[] = {
2033         &md_scan_mode.attr,
2034         &md_mismatches.attr,
2035         NULL,
2036 };
2037 static struct attribute_group md_redundancy_group = {
2038         .name = NULL,
2039         .attrs = md_redundancy_attrs,
2040 };
2041
2042
2043 static ssize_t
2044 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2045 {
2046         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2047         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2048         ssize_t rv;
2049
2050         if (!entry->show)
2051                 return -EIO;
2052         mddev_lock(mddev);
2053         rv = entry->show(mddev, page);
2054         mddev_unlock(mddev);
2055         return rv;
2056 }
2057
2058 static ssize_t
2059 md_attr_store(struct kobject *kobj, struct attribute *attr,
2060               const char *page, size_t length)
2061 {
2062         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2063         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2064         ssize_t rv;
2065
2066         if (!entry->store)
2067                 return -EIO;
2068         mddev_lock(mddev);
2069         rv = entry->store(mddev, page, length);
2070         mddev_unlock(mddev);
2071         return rv;
2072 }
2073
2074 static void md_free(struct kobject *ko)
2075 {
2076         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2077         kfree(mddev);
2078 }
2079
2080 static struct sysfs_ops md_sysfs_ops = {
2081         .show   = md_attr_show,
2082         .store  = md_attr_store,
2083 };
2084 static struct kobj_type md_ktype = {
2085         .release        = md_free,
2086         .sysfs_ops      = &md_sysfs_ops,
2087         .default_attrs  = md_default_attrs,
2088 };
2089
2090 int mdp_major = 0;
2091
2092 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2093 {
2094         static DECLARE_MUTEX(disks_sem);
2095         mddev_t *mddev = mddev_find(dev);
2096         struct gendisk *disk;
2097         int partitioned = (MAJOR(dev) != MD_MAJOR);
2098         int shift = partitioned ? MdpMinorShift : 0;
2099         int unit = MINOR(dev) >> shift;
2100
2101         if (!mddev)
2102                 return NULL;
2103
2104         down(&disks_sem);
2105         if (mddev->gendisk) {
2106                 up(&disks_sem);
2107                 mddev_put(mddev);
2108                 return NULL;
2109         }
2110         disk = alloc_disk(1 << shift);
2111         if (!disk) {
2112                 up(&disks_sem);
2113                 mddev_put(mddev);
2114                 return NULL;
2115         }
2116         disk->major = MAJOR(dev);
2117         disk->first_minor = unit << shift;
2118         if (partitioned) {
2119                 sprintf(disk->disk_name, "md_d%d", unit);
2120                 sprintf(disk->devfs_name, "md/d%d", unit);
2121         } else {
2122                 sprintf(disk->disk_name, "md%d", unit);
2123                 sprintf(disk->devfs_name, "md/%d", unit);
2124         }
2125         disk->fops = &md_fops;
2126         disk->private_data = mddev;
2127         disk->queue = mddev->queue;
2128         add_disk(disk);
2129         mddev->gendisk = disk;
2130         up(&disks_sem);
2131         mddev->kobj.parent = &disk->kobj;
2132         mddev->kobj.k_name = NULL;
2133         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2134         mddev->kobj.ktype = &md_ktype;
2135         kobject_register(&mddev->kobj);
2136         return NULL;
2137 }
2138
2139 void md_wakeup_thread(mdk_thread_t *thread);
2140
2141 static void md_safemode_timeout(unsigned long data)
2142 {
2143         mddev_t *mddev = (mddev_t *) data;
2144
2145         mddev->safemode = 1;
2146         md_wakeup_thread(mddev->thread);
2147 }
2148
2149 static int start_dirty_degraded;
2150
2151 static int do_md_run(mddev_t * mddev)
2152 {
2153         int err;
2154         int chunk_size;
2155         struct list_head *tmp;
2156         mdk_rdev_t *rdev;
2157         struct gendisk *disk;
2158         struct mdk_personality *pers;
2159         char b[BDEVNAME_SIZE];
2160
2161         if (list_empty(&mddev->disks))
2162                 /* cannot run an array with no devices.. */
2163                 return -EINVAL;
2164
2165         if (mddev->pers)
2166                 return -EBUSY;
2167
2168         /*
2169          * Analyze all RAID superblock(s)
2170          */
2171         if (!mddev->raid_disks)
2172                 analyze_sbs(mddev);
2173
2174         chunk_size = mddev->chunk_size;
2175
2176         if (chunk_size) {
2177                 if (chunk_size > MAX_CHUNK_SIZE) {
2178                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2179                                 chunk_size, MAX_CHUNK_SIZE);
2180                         return -EINVAL;
2181                 }
2182                 /*
2183                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2184                  */
2185                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2186                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2187                         return -EINVAL;
2188                 }
2189                 if (chunk_size < PAGE_SIZE) {
2190                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2191                                 chunk_size, PAGE_SIZE);
2192                         return -EINVAL;
2193                 }
2194
2195                 /* devices must have minimum size of one chunk */
2196                 ITERATE_RDEV(mddev,rdev,tmp) {
2197                         if (test_bit(Faulty, &rdev->flags))
2198                                 continue;
2199                         if (rdev->size < chunk_size / 1024) {
2200                                 printk(KERN_WARNING
2201                                         "md: Dev %s smaller than chunk_size:"
2202                                         " %lluk < %dk\n",
2203                                         bdevname(rdev->bdev,b),
2204                                         (unsigned long long)rdev->size,
2205                                         chunk_size / 1024);
2206                                 return -EINVAL;
2207                         }
2208                 }
2209         }
2210
2211 #ifdef CONFIG_KMOD
2212         if (mddev->level != LEVEL_NONE)
2213                 request_module("md-level-%d", mddev->level);
2214         else if (mddev->clevel[0])
2215                 request_module("md-%s", mddev->clevel);
2216 #endif
2217
2218         /*
2219          * Drop all container device buffers, from now on
2220          * the only valid external interface is through the md
2221          * device.
2222          * Also find largest hardsector size
2223          */
2224         ITERATE_RDEV(mddev,rdev,tmp) {
2225                 if (test_bit(Faulty, &rdev->flags))
2226                         continue;
2227                 sync_blockdev(rdev->bdev);
2228                 invalidate_bdev(rdev->bdev, 0);
2229         }
2230
2231         md_probe(mddev->unit, NULL, NULL);
2232         disk = mddev->gendisk;
2233         if (!disk)
2234                 return -ENOMEM;
2235
2236         spin_lock(&pers_lock);
2237         pers = find_pers(mddev->level, mddev->clevel);
2238         if (!pers || !try_module_get(pers->owner)) {
2239                 spin_unlock(&pers_lock);
2240                 if (mddev->level != LEVEL_NONE)
2241                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2242                                mddev->level);
2243                 else
2244                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2245                                mddev->clevel);
2246                 return -EINVAL;
2247         }
2248         mddev->pers = pers;
2249         spin_unlock(&pers_lock);
2250         mddev->level = pers->level;
2251         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2252
2253         mddev->recovery = 0;
2254         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2255         mddev->barriers_work = 1;
2256         mddev->ok_start_degraded = start_dirty_degraded;
2257
2258         if (start_readonly)
2259                 mddev->ro = 2; /* read-only, but switch on first write */
2260
2261         err = mddev->pers->run(mddev);
2262         if (!err && mddev->pers->sync_request) {
2263                 err = bitmap_create(mddev);
2264                 if (err) {
2265                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2266                                mdname(mddev), err);
2267                         mddev->pers->stop(mddev);
2268                 }
2269         }
2270         if (err) {
2271                 printk(KERN_ERR "md: pers->run() failed ...\n");
2272                 module_put(mddev->pers->owner);
2273                 mddev->pers = NULL;
2274                 bitmap_destroy(mddev);
2275                 return err;
2276         }
2277         if (mddev->pers->sync_request)
2278                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2279         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2280                 mddev->ro = 0;
2281
2282         atomic_set(&mddev->writes_pending,0);
2283         mddev->safemode = 0;
2284         mddev->safemode_timer.function = md_safemode_timeout;
2285         mddev->safemode_timer.data = (unsigned long) mddev;
2286         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2287         mddev->in_sync = 1;
2288
2289         ITERATE_RDEV(mddev,rdev,tmp)
2290                 if (rdev->raid_disk >= 0) {
2291                         char nm[20];
2292                         sprintf(nm, "rd%d", rdev->raid_disk);
2293                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2294                 }
2295
2296         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2297         md_wakeup_thread(mddev->thread);
2298
2299         if (mddev->sb_dirty)
2300                 md_update_sb(mddev);
2301
2302         set_capacity(disk, mddev->array_size<<1);
2303
2304         /* If we call blk_queue_make_request here, it will
2305          * re-initialise max_sectors etc which may have been
2306          * refined inside -> run.  So just set the bits we need to set.
2307          * Most initialisation happended when we called
2308          * blk_queue_make_request(..., md_fail_request)
2309          * earlier.
2310          */
2311         mddev->queue->queuedata = mddev;
2312         mddev->queue->make_request_fn = mddev->pers->make_request;
2313
2314         mddev->changed = 1;
2315         md_new_event(mddev);
2316         return 0;
2317 }
2318
2319 static int restart_array(mddev_t *mddev)
2320 {
2321         struct gendisk *disk = mddev->gendisk;
2322         int err;
2323
2324         /*
2325          * Complain if it has no devices
2326          */
2327         err = -ENXIO;
2328         if (list_empty(&mddev->disks))
2329                 goto out;
2330
2331         if (mddev->pers) {
2332                 err = -EBUSY;
2333                 if (!mddev->ro)
2334                         goto out;
2335
2336                 mddev->safemode = 0;
2337                 mddev->ro = 0;
2338                 set_disk_ro(disk, 0);
2339
2340                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2341                         mdname(mddev));
2342                 /*
2343                  * Kick recovery or resync if necessary
2344                  */
2345                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2346                 md_wakeup_thread(mddev->thread);
2347                 err = 0;
2348         } else {
2349                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2350                         mdname(mddev));
2351                 err = -EINVAL;
2352         }
2353
2354 out:
2355         return err;
2356 }
2357
2358 static int do_md_stop(mddev_t * mddev, int ro)
2359 {
2360         int err = 0;
2361         struct gendisk *disk = mddev->gendisk;
2362
2363         if (mddev->pers) {
2364                 if (atomic_read(&mddev->active)>2) {
2365                         printk("md: %s still in use.\n",mdname(mddev));
2366                         return -EBUSY;
2367                 }
2368
2369                 if (mddev->sync_thread) {
2370                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2371                         md_unregister_thread(mddev->sync_thread);
2372                         mddev->sync_thread = NULL;
2373                 }
2374
2375                 del_timer_sync(&mddev->safemode_timer);
2376
2377                 invalidate_partition(disk, 0);
2378
2379                 if (ro) {
2380                         err  = -ENXIO;
2381                         if (mddev->ro==1)
2382                                 goto out;
2383                         mddev->ro = 1;
2384                 } else {
2385                         bitmap_flush(mddev);
2386                         md_super_wait(mddev);
2387                         if (mddev->ro)
2388                                 set_disk_ro(disk, 0);
2389                         blk_queue_make_request(mddev->queue, md_fail_request);
2390                         mddev->pers->stop(mddev);
2391                         if (mddev->pers->sync_request)
2392                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2393
2394                         module_put(mddev->pers->owner);
2395                         mddev->pers = NULL;
2396                         if (mddev->ro)
2397                                 mddev->ro = 0;
2398                 }
2399                 if (!mddev->in_sync) {
2400                         /* mark array as shutdown cleanly */
2401                         mddev->in_sync = 1;
2402                         md_update_sb(mddev);
2403                 }
2404                 if (ro)
2405                         set_disk_ro(disk, 1);
2406         }
2407
2408         bitmap_destroy(mddev);
2409         if (mddev->bitmap_file) {
2410                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2411                 fput(mddev->bitmap_file);
2412                 mddev->bitmap_file = NULL;
2413         }
2414         mddev->bitmap_offset = 0;
2415
2416         /*
2417          * Free resources if final stop
2418          */
2419         if (!ro) {
2420                 mdk_rdev_t *rdev;
2421                 struct list_head *tmp;
2422                 struct gendisk *disk;
2423                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2424
2425                 ITERATE_RDEV(mddev,rdev,tmp)
2426                         if (rdev->raid_disk >= 0) {
2427                                 char nm[20];
2428                                 sprintf(nm, "rd%d", rdev->raid_disk);
2429                                 sysfs_remove_link(&mddev->kobj, nm);
2430                         }
2431
2432                 export_array(mddev);
2433
2434                 mddev->array_size = 0;
2435                 disk = mddev->gendisk;
2436                 if (disk)
2437                         set_capacity(disk, 0);
2438                 mddev->changed = 1;
2439         } else
2440                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2441                         mdname(mddev));
2442         err = 0;
2443         md_new_event(mddev);
2444 out:
2445         return err;
2446 }
2447
2448 static void autorun_array(mddev_t *mddev)
2449 {
2450         mdk_rdev_t *rdev;
2451         struct list_head *tmp;
2452         int err;
2453
2454         if (list_empty(&mddev->disks))
2455                 return;
2456
2457         printk(KERN_INFO "md: running: ");
2458
2459         ITERATE_RDEV(mddev,rdev,tmp) {
2460                 char b[BDEVNAME_SIZE];
2461                 printk("<%s>", bdevname(rdev->bdev,b));
2462         }
2463         printk("\n");
2464
2465         err = do_md_run (mddev);
2466         if (err) {
2467                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2468                 do_md_stop (mddev, 0);
2469         }
2470 }
2471
2472 /*
2473  * lets try to run arrays based on all disks that have arrived
2474  * until now. (those are in pending_raid_disks)
2475  *
2476  * the method: pick the first pending disk, collect all disks with
2477  * the same UUID, remove all from the pending list and put them into
2478  * the 'same_array' list. Then order this list based on superblock
2479  * update time (freshest comes first), kick out 'old' disks and
2480  * compare superblocks. If everything's fine then run it.
2481  *
2482  * If "unit" is allocated, then bump its reference count
2483  */
2484 static void autorun_devices(int part)
2485 {
2486         struct list_head candidates;
2487         struct list_head *tmp;
2488         mdk_rdev_t *rdev0, *rdev;
2489         mddev_t *mddev;
2490         char b[BDEVNAME_SIZE];
2491
2492         printk(KERN_INFO "md: autorun ...\n");
2493         while (!list_empty(&pending_raid_disks)) {
2494                 dev_t dev;
2495                 rdev0 = list_entry(pending_raid_disks.next,
2496                                          mdk_rdev_t, same_set);
2497
2498                 printk(KERN_INFO "md: considering %s ...\n",
2499                         bdevname(rdev0->bdev,b));
2500                 INIT_LIST_HEAD(&candidates);
2501                 ITERATE_RDEV_PENDING(rdev,tmp)
2502                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2503                                 printk(KERN_INFO "md:  adding %s ...\n",
2504                                         bdevname(rdev->bdev,b));
2505                                 list_move(&rdev->same_set, &candidates);
2506                         }
2507                 /*
2508                  * now we have a set of devices, with all of them having
2509                  * mostly sane superblocks. It's time to allocate the
2510                  * mddev.
2511                  */
2512                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2513                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2514                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2515                         break;
2516                 }
2517                 if (part)
2518                         dev = MKDEV(mdp_major,
2519                                     rdev0->preferred_minor << MdpMinorShift);
2520                 else
2521                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2522
2523                 md_probe(dev, NULL, NULL);
2524                 mddev = mddev_find(dev);
2525                 if (!mddev) {
2526                         printk(KERN_ERR
2527                                 "md: cannot allocate memory for md drive.\n");
2528                         break;
2529                 }
2530                 if (mddev_lock(mddev))
2531                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2532                                mdname(mddev));
2533                 else if (mddev->raid_disks || mddev->major_version
2534                          || !list_empty(&mddev->disks)) {
2535                         printk(KERN_WARNING
2536                                 "md: %s already running, cannot run %s\n",
2537                                 mdname(mddev), bdevname(rdev0->bdev,b));
2538                         mddev_unlock(mddev);
2539                 } else {
2540                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2541                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2542                                 list_del_init(&rdev->same_set);
2543                                 if (bind_rdev_to_array(rdev, mddev))
2544                                         export_rdev(rdev);
2545                         }
2546                         autorun_array(mddev);
2547                         mddev_unlock(mddev);
2548                 }
2549                 /* on success, candidates will be empty, on error
2550                  * it won't...
2551                  */
2552                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2553                         export_rdev(rdev);
2554                 mddev_put(mddev);
2555         }
2556         printk(KERN_INFO "md: ... autorun DONE.\n");
2557 }
2558
2559 /*
2560  * import RAID devices based on one partition
2561  * if possible, the array gets run as well.
2562  */
2563
2564 static int autostart_array(dev_t startdev)
2565 {
2566         char b[BDEVNAME_SIZE];
2567         int err = -EINVAL, i;
2568         mdp_super_t *sb = NULL;
2569         mdk_rdev_t *start_rdev = NULL, *rdev;
2570
2571         start_rdev = md_import_device(startdev, 0, 0);
2572         if (IS_ERR(start_rdev))
2573                 return err;
2574
2575
2576         /* NOTE: this can only work for 0.90.0 superblocks */
2577         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2578         if (sb->major_version != 0 ||
2579             sb->minor_version != 90 ) {
2580                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2581                 export_rdev(start_rdev);
2582                 return err;
2583         }
2584
2585         if (test_bit(Faulty, &start_rdev->flags)) {
2586                 printk(KERN_WARNING
2587                         "md: can not autostart based on faulty %s!\n",
2588                         bdevname(start_rdev->bdev,b));
2589                 export_rdev(start_rdev);
2590                 return err;
2591         }
2592         list_add(&start_rdev->same_set, &pending_raid_disks);
2593
2594         for (i = 0; i < MD_SB_DISKS; i++) {
2595                 mdp_disk_t *desc = sb->disks + i;
2596                 dev_t dev = MKDEV(desc->major, desc->minor);
2597
2598                 if (!dev)
2599                         continue;
2600                 if (dev == startdev)
2601                         continue;
2602                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2603                         continue;
2604                 rdev = md_import_device(dev, 0, 0);
2605                 if (IS_ERR(rdev))
2606                         continue;
2607
2608                 list_add(&rdev->same_set, &pending_raid_disks);
2609         }
2610
2611         /*
2612          * possibly return codes
2613          */
2614         autorun_devices(0);
2615         return 0;
2616
2617 }
2618
2619
2620 static int get_version(void __user * arg)
2621 {
2622         mdu_version_t ver;
2623
2624         ver.major = MD_MAJOR_VERSION;
2625         ver.minor = MD_MINOR_VERSION;
2626         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2627
2628         if (copy_to_user(arg, &ver, sizeof(ver)))
2629                 return -EFAULT;
2630
2631         return 0;
2632 }
2633
2634 static int get_array_info(mddev_t * mddev, void __user * arg)
2635 {
2636         mdu_array_info_t info;
2637         int nr,working,active,failed,spare;
2638         mdk_rdev_t *rdev;
2639         struct list_head *tmp;
2640
2641         nr=working=active=failed=spare=0;
2642         ITERATE_RDEV(mddev,rdev,tmp) {
2643                 nr++;
2644                 if (test_bit(Faulty, &rdev->flags))
2645                         failed++;
2646                 else {
2647                         working++;
2648                         if (test_bit(In_sync, &rdev->flags))
2649                                 active++;
2650                         else
2651                                 spare++;
2652                 }
2653         }
2654
2655         info.major_version = mddev->major_version;
2656         info.minor_version = mddev->minor_version;
2657         info.patch_version = MD_PATCHLEVEL_VERSION;
2658         info.ctime         = mddev->ctime;
2659         info.level         = mddev->level;
2660         info.size          = mddev->size;
2661         info.nr_disks      = nr;
2662         info.raid_disks    = mddev->raid_disks;
2663         info.md_minor      = mddev->md_minor;
2664         info.not_persistent= !mddev->persistent;
2665
2666         info.utime         = mddev->utime;
2667         info.state         = 0;
2668         if (mddev->in_sync)
2669                 info.state = (1<<MD_SB_CLEAN);
2670         if (mddev->bitmap && mddev->bitmap_offset)
2671                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2672         info.active_disks  = active;
2673         info.working_disks = working;
2674         info.failed_disks  = failed;
2675         info.spare_disks   = spare;
2676
2677         info.layout        = mddev->layout;
2678         info.chunk_size    = mddev->chunk_size;
2679
2680         if (copy_to_user(arg, &info, sizeof(info)))
2681                 return -EFAULT;
2682
2683         return 0;
2684 }
2685
2686 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2687 {
2688         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2689         char *ptr, *buf = NULL;
2690         int err = -ENOMEM;
2691
2692         file = kmalloc(sizeof(*file), GFP_KERNEL);
2693         if (!file)
2694                 goto out;
2695
2696         /* bitmap disabled, zero the first byte and copy out */
2697         if (!mddev->bitmap || !mddev->bitmap->file) {
2698                 file->pathname[0] = '\0';
2699                 goto copy_out;
2700         }
2701
2702         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2703         if (!buf)
2704                 goto out;
2705
2706         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2707         if (!ptr)
2708                 goto out;
2709
2710         strcpy(file->pathname, ptr);
2711
2712 copy_out:
2713         err = 0;
2714         if (copy_to_user(arg, file, sizeof(*file)))
2715                 err = -EFAULT;
2716 out:
2717         kfree(buf);
2718         kfree(file);
2719         return err;
2720 }
2721
2722 static int get_disk_info(mddev_t * mddev, void __user * arg)
2723 {
2724         mdu_disk_info_t info;
2725         unsigned int nr;
2726         mdk_rdev_t *rdev;
2727
2728         if (copy_from_user(&info, arg, sizeof(info)))
2729                 return -EFAULT;
2730
2731         nr = info.number;
2732
2733         rdev = find_rdev_nr(mddev, nr);
2734         if (rdev) {
2735                 info.major = MAJOR(rdev->bdev->bd_dev);
2736                 info.minor = MINOR(rdev->bdev->bd_dev);
2737                 info.raid_disk = rdev->raid_disk;
2738                 info.state = 0;
2739                 if (test_bit(Faulty, &rdev->flags))
2740                         info.state |= (1<<MD_DISK_FAULTY);
2741                 else if (test_bit(In_sync, &rdev->flags)) {
2742                         info.state |= (1<<MD_DISK_ACTIVE);
2743                         info.state |= (1<<MD_DISK_SYNC);
2744                 }
2745                 if (test_bit(WriteMostly, &rdev->flags))
2746                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2747         } else {
2748                 info.major = info.minor = 0;
2749                 info.raid_disk = -1;
2750                 info.state = (1<<MD_DISK_REMOVED);
2751         }
2752
2753         if (copy_to_user(arg, &info, sizeof(info)))
2754                 return -EFAULT;
2755
2756         return 0;
2757 }
2758
2759 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2760 {
2761         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2762         mdk_rdev_t *rdev;
2763         dev_t dev = MKDEV(info->major,info->minor);
2764
2765         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2766                 return -EOVERFLOW;
2767
2768         if (!mddev->raid_disks) {
2769                 int err;
2770                 /* expecting a device which has a superblock */
2771                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2772                 if (IS_ERR(rdev)) {
2773                         printk(KERN_WARNING
2774                                 "md: md_import_device returned %ld\n",
2775                                 PTR_ERR(rdev));
2776                         return PTR_ERR(rdev);
2777                 }
2778                 if (!list_empty(&mddev->disks)) {
2779                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2780                                                         mdk_rdev_t, same_set);
2781                         int err = super_types[mddev->major_version]
2782                                 .load_super(rdev, rdev0, mddev->minor_version);
2783                         if (err < 0) {
2784                                 printk(KERN_WARNING
2785                                         "md: %s has different UUID to %s\n",
2786                                         bdevname(rdev->bdev,b),
2787                                         bdevname(rdev0->bdev,b2));
2788                                 export_rdev(rdev);
2789                                 return -EINVAL;
2790                         }
2791                 }
2792                 err = bind_rdev_to_array(rdev, mddev);
2793                 if (err)
2794                         export_rdev(rdev);
2795                 return err;
2796         }
2797
2798         /*
2799          * add_new_disk can be used once the array is assembled
2800          * to add "hot spares".  They must already have a superblock
2801          * written
2802          */
2803         if (mddev->pers) {
2804                 int err;
2805                 if (!mddev->pers->hot_add_disk) {
2806                         printk(KERN_WARNING
2807                                 "%s: personality does not support diskops!\n",
2808                                mdname(mddev));
2809                         return -EINVAL;
2810                 }
2811                 if (mddev->persistent)
2812                         rdev = md_import_device(dev, mddev->major_version,
2813                                                 mddev->minor_version);
2814                 else
2815                         rdev = md_import_device(dev, -1, -1);
2816                 if (IS_ERR(rdev)) {
2817                         printk(KERN_WARNING
2818                                 "md: md_import_device returned %ld\n",
2819                                 PTR_ERR(rdev));
2820                         return PTR_ERR(rdev);
2821                 }
2822                 /* set save_raid_disk if appropriate */
2823                 if (!mddev->persistent) {
2824                         if (info->state & (1<<MD_DISK_SYNC)  &&
2825                             info->raid_disk < mddev->raid_disks)
2826                                 rdev->raid_disk = info->raid_disk;
2827                         else
2828                                 rdev->raid_disk = -1;
2829                 } else
2830                         super_types[mddev->major_version].
2831                                 validate_super(mddev, rdev);
2832                 rdev->saved_raid_disk = rdev->raid_disk;
2833
2834                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2835                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2836                         set_bit(WriteMostly, &rdev->flags);
2837
2838                 rdev->raid_disk = -1;
2839                 err = bind_rdev_to_array(rdev, mddev);
2840                 if (err)
2841                         export_rdev(rdev);
2842
2843                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2844                 md_wakeup_thread(mddev->thread);
2845                 return err;
2846         }
2847
2848         /* otherwise, add_new_disk is only allowed
2849          * for major_version==0 superblocks
2850          */
2851         if (mddev->major_version != 0) {
2852                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2853                        mdname(mddev));
2854                 return -EINVAL;
2855         }
2856
2857         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2858                 int err;
2859                 rdev = md_import_device (dev, -1, 0);
2860                 if (IS_ERR(rdev)) {
2861                         printk(KERN_WARNING
2862                                 "md: error, md_import_device() returned %ld\n",
2863                                 PTR_ERR(rdev));
2864                         return PTR_ERR(rdev);
2865                 }
2866                 rdev->desc_nr = info->number;
2867                 if (info->raid_disk < mddev->raid_disks)
2868                         rdev->raid_disk = info->raid_disk;
2869                 else
2870                         rdev->raid_disk = -1;
2871
2872                 rdev->flags = 0;
2873
2874                 if (rdev->raid_disk < mddev->raid_disks)
2875                         if (info->state & (1<<MD_DISK_SYNC))
2876                                 set_bit(In_sync, &rdev->flags);
2877
2878                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2879                         set_bit(WriteMostly, &rdev->flags);
2880
2881                 err = bind_rdev_to_array(rdev, mddev);
2882                 if (err) {
2883                         export_rdev(rdev);
2884                         return err;
2885                 }
2886
2887                 if (!mddev->persistent) {
2888                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2889                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2890                 } else
2891                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2892                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2893
2894                 if (!mddev->size || (mddev->size > rdev->size))
2895                         mddev->size = rdev->size;
2896         }
2897
2898         return 0;
2899 }
2900
2901 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2902 {
2903         char b[BDEVNAME_SIZE];
2904         mdk_rdev_t *rdev;
2905
2906         if (!mddev->pers)
2907                 return -ENODEV;
2908
2909         rdev = find_rdev(mddev, dev);
2910         if (!rdev)
2911                 return -ENXIO;
2912
2913         if (rdev->raid_disk >= 0)
2914                 goto busy;
2915
2916         kick_rdev_from_array(rdev);
2917         md_update_sb(mddev);
2918         md_new_event(mddev);
2919
2920         return 0;
2921 busy:
2922         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2923                 bdevname(rdev->bdev,b), mdname(mddev));
2924         return -EBUSY;
2925 }
2926
2927 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2928 {
2929         char b[BDEVNAME_SIZE];
2930         int err;
2931         unsigned int size;
2932         mdk_rdev_t *rdev;
2933
2934         if (!mddev->pers)
2935                 return -ENODEV;
2936
2937         if (mddev->major_version != 0) {
2938                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2939                         " version-0 superblocks.\n",
2940                         mdname(mddev));
2941                 return -EINVAL;
2942         }
2943         if (!mddev->pers->hot_add_disk) {
2944                 printk(KERN_WARNING
2945                         "%s: personality does not support diskops!\n",
2946                         mdname(mddev));
2947                 return -EINVAL;
2948         }
2949
2950         rdev = md_import_device (dev, -1, 0);
2951         if (IS_ERR(rdev)) {
2952                 printk(KERN_WARNING
2953                         "md: error, md_import_device() returned %ld\n",
2954                         PTR_ERR(rdev));
2955                 return -EINVAL;
2956         }
2957
2958         if (mddev->persistent)
2959                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2960         else
2961                 rdev->sb_offset =
2962                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2963
2964         size = calc_dev_size(rdev, mddev->chunk_size);
2965         rdev->size = size;
2966
2967         if (size < mddev->size) {
2968                 printk(KERN_WARNING
2969                         "%s: disk size %llu blocks < array size %llu\n",
2970                         mdname(mddev), (unsigned long long)size,
2971                         (unsigned long long)mddev->size);
2972                 err = -ENOSPC;
2973                 goto abort_export;
2974         }
2975
2976         if (test_bit(Faulty, &rdev->flags)) {
2977                 printk(KERN_WARNING
2978                         "md: can not hot-add faulty %s disk to %s!\n",
2979                         bdevname(rdev->bdev,b), mdname(mddev));
2980                 err = -EINVAL;
2981                 goto abort_export;
2982         }
2983         clear_bit(In_sync, &rdev->flags);
2984         rdev->desc_nr = -1;
2985         bind_rdev_to_array(rdev, mddev);
2986
2987         /*
2988          * The rest should better be atomic, we can have disk failures
2989          * noticed in interrupt contexts ...
2990          */
2991
2992         if (rdev->desc_nr == mddev->max_disks) {
2993                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2994                         mdname(mddev));
2995                 err = -EBUSY;
2996                 goto abort_unbind_export;
2997         }
2998
2999         rdev->raid_disk = -1;
3000
3001         md_update_sb(mddev);
3002
3003         /*
3004          * Kick recovery, maybe this spare has to be added to the
3005          * array immediately.
3006          */
3007         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3008         md_wakeup_thread(mddev->thread);
3009         md_new_event(mddev);
3010         return 0;
3011
3012 abort_unbind_export:
3013         unbind_rdev_from_array(rdev);
3014
3015 abort_export:
3016         export_rdev(rdev);
3017         return err;
3018 }
3019
3020 /* similar to deny_write_access, but accounts for our holding a reference
3021  * to the file ourselves */
3022 static int deny_bitmap_write_access(struct file * file)
3023 {
3024         struct inode *inode = file->f_mapping->host;
3025
3026         spin_lock(&inode->i_lock);
3027         if (atomic_read(&inode->i_writecount) > 1) {
3028                 spin_unlock(&inode->i_lock);
3029                 return -ETXTBSY;
3030         }
3031         atomic_set(&inode->i_writecount, -1);
3032         spin_unlock(&inode->i_lock);
3033
3034         return 0;
3035 }
3036
3037 static int set_bitmap_file(mddev_t *mddev, int fd)
3038 {
3039         int err;
3040
3041         if (mddev->pers) {
3042                 if (!mddev->pers->quiesce)
3043                         return -EBUSY;
3044                 if (mddev->recovery || mddev->sync_thread)
3045                         return -EBUSY;
3046                 /* we should be able to change the bitmap.. */
3047         }
3048
3049
3050         if (fd >= 0) {
3051                 if (mddev->bitmap)
3052                         return -EEXIST; /* cannot add when bitmap is present */
3053                 mddev->bitmap_file = fget(fd);
3054
3055                 if (mddev->bitmap_file == NULL) {
3056                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3057                                mdname(mddev));
3058                         return -EBADF;
3059                 }
3060
3061                 err = deny_bitmap_write_access(mddev->bitmap_file);
3062                 if (err) {
3063                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3064                                mdname(mddev));
3065                         fput(mddev->bitmap_file);
3066                         mddev->bitmap_file = NULL;
3067                         return err;
3068                 }
3069                 mddev->bitmap_offset = 0; /* file overrides offset */
3070         } else if (mddev->bitmap == NULL)
3071                 return -ENOENT; /* cannot remove what isn't there */
3072         err = 0;
3073         if (mddev->pers) {
3074                 mddev->pers->quiesce(mddev, 1);
3075                 if (fd >= 0)
3076                         err = bitmap_create(mddev);
3077                 if (fd < 0 || err)
3078                         bitmap_destroy(mddev);
3079                 mddev->pers->quiesce(mddev, 0);
3080         } else if (fd < 0) {
3081                 if (mddev->bitmap_file)
3082                         fput(mddev->bitmap_file);
3083                 mddev->bitmap_file = NULL;
3084         }
3085
3086         return err;
3087 }
3088
3089 /*
3090  * set_array_info is used two different ways
3091  * The original usage is when creating a new array.
3092  * In this usage, raid_disks is > 0 and it together with
3093  *  level, size, not_persistent,layout,chunksize determine the
3094  *  shape of the array.
3095  *  This will always create an array with a type-0.90.0 superblock.
3096  * The newer usage is when assembling an array.
3097  *  In this case raid_disks will be 0, and the major_version field is
3098  *  use to determine which style super-blocks are to be found on the devices.
3099  *  The minor and patch _version numbers are also kept incase the
3100  *  super_block handler wishes to interpret them.
3101  */
3102 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3103 {
3104
3105         if (info->raid_disks == 0) {
3106                 /* just setting version number for superblock loading */
3107                 if (info->major_version < 0 ||
3108                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3109                     super_types[info->major_version].name == NULL) {
3110                         /* maybe try to auto-load a module? */
3111                         printk(KERN_INFO
3112                                 "md: superblock version %d not known\n",
3113                                 info->major_version);
3114                         return -EINVAL;
3115                 }
3116                 mddev->major_version = info->major_version;
3117                 mddev->minor_version = info->minor_version;
3118                 mddev->patch_version = info->patch_version;
3119                 return 0;
3120         }
3121         mddev->major_version = MD_MAJOR_VERSION;
3122         mddev->minor_version = MD_MINOR_VERSION;
3123         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3124         mddev->ctime         = get_seconds();
3125
3126         mddev->level         = info->level;
3127         mddev->size          = info->size;
3128         mddev->raid_disks    = info->raid_disks;
3129         /* don't set md_minor, it is determined by which /dev/md* was
3130          * openned
3131          */
3132         if (info->state & (1<<MD_SB_CLEAN))
3133                 mddev->recovery_cp = MaxSector;
3134         else
3135                 mddev->recovery_cp = 0;
3136         mddev->persistent    = ! info->not_persistent;
3137
3138         mddev->layout        = info->layout;
3139         mddev->chunk_size    = info->chunk_size;
3140
3141         mddev->max_disks     = MD_SB_DISKS;
3142
3143         mddev->sb_dirty      = 1;
3144
3145         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3146         mddev->bitmap_offset = 0;
3147
3148         /*
3149          * Generate a 128 bit UUID
3150          */
3151         get_random_bytes(mddev->uuid, 16);
3152
3153         return 0;
3154 }
3155
3156 static int update_size(mddev_t *mddev, unsigned long size)
3157 {
3158         mdk_rdev_t * rdev;
3159         int rv;
3160         struct list_head *tmp;
3161
3162         if (mddev->pers->resize == NULL)
3163                 return -EINVAL;
3164         /* The "size" is the amount of each device that is used.
3165          * This can only make sense for arrays with redundancy.
3166          * linear and raid0 always use whatever space is available
3167          * We can only consider changing the size if no resync
3168          * or reconstruction is happening, and if the new size
3169          * is acceptable. It must fit before the sb_offset or,
3170          * if that is <data_offset, it must fit before the
3171          * size of each device.
3172          * If size is zero, we find the largest size that fits.
3173          */
3174         if (mddev->sync_thread)
3175                 return -EBUSY;
3176         ITERATE_RDEV(mddev,rdev,tmp) {
3177                 sector_t avail;
3178                 int fit = (size == 0);
3179                 if (rdev->sb_offset > rdev->data_offset)
3180                         avail = (rdev->sb_offset*2) - rdev->data_offset;
3181                 else
3182                         avail = get_capacity(rdev->bdev->bd_disk)
3183                                 - rdev->data_offset;
3184                 if (fit && (size == 0 || size > avail/2))
3185                         size = avail/2;
3186                 if (avail < ((sector_t)size << 1))
3187                         return -ENOSPC;
3188         }
3189         rv = mddev->pers->resize(mddev, (sector_t)size *2);
3190         if (!rv) {
3191                 struct block_device *bdev;
3192
3193                 bdev = bdget_disk(mddev->gendisk, 0);
3194                 if (bdev) {
3195                         down(&bdev->bd_inode->i_sem);
3196                         i_size_write(bdev->bd_inode, mddev->array_size << 10);
3197                         up(&bdev->bd_inode->i_sem);
3198                         bdput(bdev);
3199                 }
3200         }
3201         return rv;
3202 }
3203
3204 /*
3205  * update_array_info is used to change the configuration of an
3206  * on-line array.
3207  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3208  * fields in the info are checked against the array.
3209  * Any differences that cannot be handled will cause an error.
3210  * Normally, only one change can be managed at a time.
3211  */
3212 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3213 {
3214         int rv = 0;
3215         int cnt = 0;
3216         int state = 0;
3217
3218         /* calculate expected state,ignoring low bits */
3219         if (mddev->bitmap && mddev->bitmap_offset)
3220                 state |= (1 << MD_SB_BITMAP_PRESENT);
3221
3222         if (mddev->major_version != info->major_version ||
3223             mddev->minor_version != info->minor_version ||
3224 /*          mddev->patch_version != info->patch_version || */
3225             mddev->ctime         != info->ctime         ||
3226             mddev->level         != info->level         ||
3227 /*          mddev->layout        != info->layout        || */
3228             !mddev->persistent   != info->not_persistent||
3229             mddev->chunk_size    != info->chunk_size    ||
3230             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3231             ((state^info->state) & 0xfffffe00)
3232                 )
3233                 return -EINVAL;
3234         /* Check there is only one change */
3235         if (mddev->size != info->size) cnt++;
3236         if (mddev->raid_disks != info->raid_disks) cnt++;
3237         if (mddev->layout != info->layout) cnt++;
3238         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3239         if (cnt == 0) return 0;
3240         if (cnt > 1) return -EINVAL;
3241
3242         if (mddev->layout != info->layout) {
3243                 /* Change layout
3244                  * we don't need to do anything at the md level, the
3245                  * personality will take care of it all.
3246                  */
3247                 if (mddev->pers->reconfig == NULL)
3248                         return -EINVAL;
3249                 else
3250                         return mddev->pers->reconfig(mddev, info->layout, -1);
3251         }
3252         if (mddev->size != info->size)
3253                 rv = update_size(mddev, info->size);
3254
3255         if (mddev->raid_disks    != info->raid_disks) {
3256                 /* change the number of raid disks */
3257                 if (mddev->pers->reshape == NULL)
3258                         return -EINVAL;
3259                 if (info->raid_disks <= 0 ||
3260                     info->raid_disks >= mddev->max_disks)
3261                         return -EINVAL;
3262                 if (mddev->sync_thread)
3263                         return -EBUSY;
3264                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3265                 if (!rv) {
3266                         struct block_device *bdev;
3267
3268                         bdev = bdget_disk(mddev->gendisk, 0);
3269                         if (bdev) {
3270                                 down(&bdev->bd_inode->i_sem);
3271                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3272                                 up(&bdev->bd_inode->i_sem);
3273                                 bdput(bdev);
3274                         }
3275                 }
3276         }
3277         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3278                 if (mddev->pers->quiesce == NULL)
3279                         return -EINVAL;
3280                 if (mddev->recovery || mddev->sync_thread)
3281                         return -EBUSY;
3282                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3283                         /* add the bitmap */
3284                         if (mddev->bitmap)
3285                                 return -EEXIST;
3286                         if (mddev->default_bitmap_offset == 0)
3287                                 return -EINVAL;
3288                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3289                         mddev->pers->quiesce(mddev, 1);
3290                         rv = bitmap_create(mddev);
3291                         if (rv)
3292                                 bitmap_destroy(mddev);
3293                         mddev->pers->quiesce(mddev, 0);
3294                 } else {
3295                         /* remove the bitmap */
3296                         if (!mddev->bitmap)
3297                                 return -ENOENT;
3298                         if (mddev->bitmap->file)
3299                                 return -EINVAL;
3300                         mddev->pers->quiesce(mddev, 1);
3301                         bitmap_destroy(mddev);
3302                         mddev->pers->quiesce(mddev, 0);
3303                         mddev->bitmap_offset = 0;
3304                 }
3305         }
3306         md_update_sb(mddev);
3307         return rv;
3308 }
3309
3310 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3311 {
3312         mdk_rdev_t *rdev;
3313
3314         if (mddev->pers == NULL)
3315                 return -ENODEV;
3316
3317         rdev = find_rdev(mddev, dev);
3318         if (!rdev)
3319                 return -ENODEV;
3320
3321         md_error(mddev, rdev);
3322         return 0;
3323 }
3324
3325 static int md_ioctl(struct inode *inode, struct file *file,
3326                         unsigned int cmd, unsigned long arg)
3327 {
3328         int err = 0;
3329         void __user *argp = (void __user *)arg;
3330         struct hd_geometry __user *loc = argp;
3331         mddev_t *mddev = NULL;
3332
3333         if (!capable(CAP_SYS_ADMIN))
3334                 return -EACCES;
3335
3336         /*
3337          * Commands dealing with the RAID driver but not any
3338          * particular array:
3339          */
3340         switch (cmd)
3341         {
3342                 case RAID_VERSION:
3343                         err = get_version(argp);
3344                         goto done;
3345
3346                 case PRINT_RAID_DEBUG:
3347                         err = 0;
3348                         md_print_devices();
3349                         goto done;
3350
3351 #ifndef MODULE
3352                 case RAID_AUTORUN:
3353                         err = 0;
3354                         autostart_arrays(arg);
3355                         goto done;
3356 #endif
3357                 default:;
3358         }
3359
3360         /*
3361          * Commands creating/starting a new array:
3362          */
3363
3364         mddev = inode->i_bdev->bd_disk->private_data;
3365
3366         if (!mddev) {
3367                 BUG();
3368                 goto abort;
3369         }
3370
3371
3372         if (cmd == START_ARRAY) {
3373                 /* START_ARRAY doesn't need to lock the array as autostart_array
3374                  * does the locking, and it could even be a different array
3375                  */
3376                 static int cnt = 3;
3377                 if (cnt > 0 ) {
3378                         printk(KERN_WARNING
3379                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3380                                "This will not be supported beyond July 2006\n",
3381                                current->comm, current->pid);
3382                         cnt--;
3383                 }
3384                 err = autostart_array(new_decode_dev(arg));
3385                 if (err) {
3386                         printk(KERN_WARNING "md: autostart failed!\n");
3387                         goto abort;
3388                 }
3389                 goto done;
3390         }
3391
3392         err = mddev_lock(mddev);
3393         if (err) {
3394                 printk(KERN_INFO
3395                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3396                         err, cmd);
3397                 goto abort;
3398         }
3399
3400         switch (cmd)
3401         {
3402                 case SET_ARRAY_INFO:
3403                         {
3404                                 mdu_array_info_t info;
3405                                 if (!arg)
3406                                         memset(&info, 0, sizeof(info));
3407                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3408                                         err = -EFAULT;
3409                                         goto abort_unlock;
3410                                 }
3411                                 if (mddev->pers) {
3412                                         err = update_array_info(mddev, &info);
3413                                         if (err) {
3414                                                 printk(KERN_WARNING "md: couldn't update"
3415                                                        " array info. %d\n", err);
3416                                                 goto abort_unlock;
3417                                         }
3418                                         goto done_unlock;
3419                                 }
3420                                 if (!list_empty(&mddev->disks)) {
3421                                         printk(KERN_WARNING
3422                                                "md: array %s already has disks!\n",
3423                                                mdname(mddev));
3424                                         err = -EBUSY;
3425                                         goto abort_unlock;
3426                                 }
3427                                 if (mddev->raid_disks) {
3428                                         printk(KERN_WARNING
3429                                                "md: array %s already initialised!\n",
3430                                                mdname(mddev));
3431                                         err = -EBUSY;
3432                                         goto abort_unlock;
3433                                 }
3434                                 err = set_array_info(mddev, &info);
3435                                 if (err) {
3436                                         printk(KERN_WARNING "md: couldn't set"
3437                                                " array info. %d\n", err);
3438                                         goto abort_unlock;
3439                                 }
3440                         }
3441                         goto done_unlock;
3442
3443                 default:;
3444         }
3445
3446         /*
3447          * Commands querying/configuring an existing array:
3448          */
3449         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3450          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3451         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3452                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3453                 err = -ENODEV;
3454                 goto abort_unlock;
3455         }
3456
3457         /*
3458          * Commands even a read-only array can execute:
3459          */
3460         switch (cmd)
3461         {
3462                 case GET_ARRAY_INFO:
3463                         err = get_array_info(mddev, argp);
3464                         goto done_unlock;
3465
3466                 case GET_BITMAP_FILE:
3467                         err = get_bitmap_file(mddev, argp);
3468                         goto done_unlock;
3469
3470                 case GET_DISK_INFO:
3471                         err = get_disk_info(mddev, argp);
3472                         goto done_unlock;
3473
3474                 case RESTART_ARRAY_RW:
3475                         err = restart_array(mddev);
3476                         goto done_unlock;
3477
3478                 case STOP_ARRAY:
3479                         err = do_md_stop (mddev, 0);
3480                         goto done_unlock;
3481
3482                 case STOP_ARRAY_RO:
3483                         err = do_md_stop (mddev, 1);
3484                         goto done_unlock;
3485
3486         /*
3487          * We have a problem here : there is no easy way to give a CHS
3488          * virtual geometry. We currently pretend that we have a 2 heads
3489          * 4 sectors (with a BIG number of cylinders...). This drives
3490          * dosfs just mad... ;-)
3491          */
3492                 case HDIO_GETGEO:
3493                         if (!loc) {
3494                                 err = -EINVAL;
3495                                 goto abort_unlock;
3496                         }
3497                         err = put_user (2, (char __user *) &loc->heads);
3498                         if (err)
3499                                 goto abort_unlock;
3500                         err = put_user (4, (char __user *) &loc->sectors);
3501                         if (err)
3502                                 goto abort_unlock;
3503                         err = put_user(get_capacity(mddev->gendisk)/8,
3504                                         (short __user *) &loc->cylinders);
3505                         if (err)
3506                                 goto abort_unlock;
3507                         err = put_user (get_start_sect(inode->i_bdev),
3508                                                 (long __user *) &loc->start);
3509                         goto done_unlock;
3510         }
3511
3512         /*
3513          * The remaining ioctls are changing the state of the
3514          * superblock, so we do not allow them on read-only arrays.
3515          * However non-MD ioctls (e.g. get-size) will still come through
3516          * here and hit the 'default' below, so only disallow
3517          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3518          */
3519         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3520             mddev->ro && mddev->pers) {
3521                 if (mddev->ro == 2) {
3522                         mddev->ro = 0;
3523                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3524                 md_wakeup_thread(mddev->thread);
3525
3526                 } else {
3527                         err = -EROFS;
3528                         goto abort_unlock;
3529                 }
3530         }
3531
3532         switch (cmd)
3533         {
3534                 case ADD_NEW_DISK:
3535                 {
3536                         mdu_disk_info_t info;
3537                         if (copy_from_user(&info, argp, sizeof(info)))
3538                                 err = -EFAULT;
3539                         else
3540                                 err = add_new_disk(mddev, &info);
3541                         goto done_unlock;
3542                 }
3543
3544                 case HOT_REMOVE_DISK:
3545                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3546                         goto done_unlock;
3547
3548                 case HOT_ADD_DISK:
3549                         err = hot_add_disk(mddev, new_decode_dev(arg));
3550                         goto done_unlock;
3551
3552                 case SET_DISK_FAULTY:
3553                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3554                         goto done_unlock;
3555
3556                 case RUN_ARRAY:
3557                         err = do_md_run (mddev);
3558                         goto done_unlock;
3559
3560                 case SET_BITMAP_FILE:
3561                         err = set_bitmap_file(mddev, (int)arg);
3562                         goto done_unlock;
3563
3564                 default:
3565                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3566                                 printk(KERN_WARNING "md: %s(pid %d) used"
3567                                         " obsolete MD ioctl, upgrade your"
3568                                         " software to use new ictls.\n",
3569                                         current->comm, current->pid);
3570                         err = -EINVAL;
3571                         goto abort_unlock;
3572         }
3573
3574 done_unlock:
3575 abort_unlock:
3576         mddev_unlock(mddev);
3577
3578         return err;
3579 done:
3580         if (err)
3581                 MD_BUG();
3582 abort:
3583         return err;
3584 }
3585
3586 static int md_open(struct inode *inode, struct file *file)
3587 {
3588         /*
3589          * Succeed if we can lock the mddev, which confirms that
3590          * it isn't being stopped right now.
3591          */
3592         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3593         int err;
3594
3595         if ((err = mddev_lock(mddev)))
3596                 goto out;
3597
3598         err = 0;
3599         mddev_get(mddev);
3600         mddev_unlock(mddev);
3601
3602         check_disk_change(inode->i_bdev);
3603  out:
3604         return err;
3605 }
3606
3607 static int md_release(struct inode *inode, struct file * file)
3608 {
3609         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3610
3611         if (!mddev)
3612                 BUG();
3613         mddev_put(mddev);
3614
3615         return 0;
3616 }
3617
3618 static int md_media_changed(struct gendisk *disk)
3619 {
3620         mddev_t *mddev = disk->private_data;
3621
3622         return mddev->changed;
3623 }
3624
3625 static int md_revalidate(struct gendisk *disk)
3626 {
3627         mddev_t *mddev = disk->private_data;
3628
3629         mddev->changed = 0;
3630         return 0;
3631 }
3632 static struct block_device_operations md_fops =
3633 {
3634         .owner          = THIS_MODULE,
3635         .open           = md_open,
3636         .release        = md_release,
3637         .ioctl          = md_ioctl,
3638         .media_changed  = md_media_changed,
3639         .revalidate_disk= md_revalidate,
3640 };
3641
3642 static int md_thread(void * arg)
3643 {
3644         mdk_thread_t *thread = arg;
3645
3646         /*
3647          * md_thread is a 'system-thread', it's priority should be very
3648          * high. We avoid resource deadlocks individually in each
3649          * raid personality. (RAID5 does preallocation) We also use RR and
3650          * the very same RT priority as kswapd, thus we will never get
3651          * into a priority inversion deadlock.
3652          *
3653          * we definitely have to have equal or higher priority than
3654          * bdflush, otherwise bdflush will deadlock if there are too
3655          * many dirty RAID5 blocks.
3656          */
3657
3658         allow_signal(SIGKILL);
3659         while (!kthread_should_stop()) {
3660
3661                 /* We need to wait INTERRUPTIBLE so that
3662                  * we don't add to the load-average.
3663                  * That means we need to be sure no signals are
3664                  * pending
3665                  */
3666                 if (signal_pending(current))
3667                         flush_signals(current);
3668
3669                 wait_event_interruptible_timeout
3670                         (thread->wqueue,
3671                          test_bit(THREAD_WAKEUP, &thread->flags)
3672                          || kthread_should_stop(),
3673                          thread->timeout);
3674                 try_to_freeze();
3675
3676                 clear_bit(THREAD_WAKEUP, &thread->flags);
3677
3678                 thread->run(thread->mddev);
3679         }
3680
3681         return 0;
3682 }
3683
3684 void md_wakeup_thread(mdk_thread_t *thread)
3685 {
3686         if (thread) {
3687                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3688                 set_bit(THREAD_WAKEUP, &thread->flags);
3689                 wake_up(&thread->wqueue);
3690         }
3691 }
3692
3693 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3694                                  const char *name)
3695 {
3696         mdk_thread_t *thread;
3697
3698         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3699         if (!thread)
3700                 return NULL;
3701
3702         init_waitqueue_head(&thread->wqueue);
3703
3704         thread->run = run;
3705         thread->mddev = mddev;
3706         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3707         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3708         if (IS_ERR(thread->tsk)) {
3709                 kfree(thread);
3710                 return NULL;
3711         }
3712         return thread;
3713 }
3714
3715 void md_unregister_thread(mdk_thread_t *thread)
3716 {
3717         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3718
3719         kthread_stop(thread->tsk);
3720         kfree(thread);
3721 }
3722
3723 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3724 {
3725         if (!mddev) {
3726                 MD_BUG();
3727                 return;
3728         }
3729
3730         if (!rdev || test_bit(Faulty, &rdev->flags))
3731                 return;
3732 /*
3733         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3734                 mdname(mddev),
3735                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3736                 __builtin_return_address(0),__builtin_return_address(1),
3737                 __builtin_return_address(2),__builtin_return_address(3));
3738 */
3739         if (!mddev->pers->error_handler)
3740                 return;
3741         mddev->pers->error_handler(mddev,rdev);
3742         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3743         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3744         md_wakeup_thread(mddev->thread);
3745         md_new_event(mddev);
3746 }
3747
3748 /* seq_file implementation /proc/mdstat */
3749
3750 static void status_unused(struct seq_file *seq)
3751 {
3752         int i = 0;
3753         mdk_rdev_t *rdev;
3754         struct list_head *tmp;
3755
3756         seq_printf(seq, "unused devices: ");
3757
3758         ITERATE_RDEV_PENDING(rdev,tmp) {
3759                 char b[BDEVNAME_SIZE];
3760                 i++;
3761                 seq_printf(seq, "%s ",
3762                               bdevname(rdev->bdev,b));
3763         }
3764         if (!i)
3765                 seq_printf(seq, "<none>");
3766
3767         seq_printf(seq, "\n");
3768 }
3769
3770
3771 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3772 {
3773         unsigned long max_blocks, resync, res, dt, db, rt;
3774
3775         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3776
3777         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3778                 max_blocks = mddev->resync_max_sectors >> 1;
3779         else
3780                 max_blocks = mddev->size;
3781
3782         /*
3783          * Should not happen.
3784          */
3785         if (!max_blocks) {
3786                 MD_BUG();
3787                 return;
3788         }
3789         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3790         {
3791                 int i, x = res/50, y = 20-x;
3792                 seq_printf(seq, "[");
3793                 for (i = 0; i < x; i++)
3794                         seq_printf(seq, "=");
3795                 seq_printf(seq, ">");
3796                 for (i = 0; i < y; i++)
3797                         seq_printf(seq, ".");
3798                 seq_printf(seq, "] ");
3799         }
3800         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3801                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3802                        "resync" : "recovery"),
3803                       res/10, res % 10, resync, max_blocks);
3804
3805         /*
3806          * We do not want to overflow, so the order of operands and
3807          * the * 100 / 100 trick are important. We do a +1 to be
3808          * safe against division by zero. We only estimate anyway.
3809          *
3810          * dt: time from mark until now
3811          * db: blocks written from mark until now
3812          * rt: remaining time
3813          */
3814         dt = ((jiffies - mddev->resync_mark) / HZ);
3815         if (!dt) dt++;
3816         db = resync - (mddev->resync_mark_cnt/2);
3817         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3818
3819         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3820
3821         seq_printf(seq, " speed=%ldK/sec", db/dt);
3822 }
3823
3824 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3825 {
3826         struct list_head *tmp;
3827         loff_t l = *pos;
3828         mddev_t *mddev;
3829
3830         if (l >= 0x10000)
3831                 return NULL;
3832         if (!l--)
3833                 /* header */
3834                 return (void*)1;
3835
3836         spin_lock(&all_mddevs_lock);
3837         list_for_each(tmp,&all_mddevs)
3838                 if (!l--) {
3839                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3840                         mddev_get(mddev);
3841                         spin_unlock(&all_mddevs_lock);
3842                         return mddev;
3843                 }
3844         spin_unlock(&all_mddevs_lock);
3845         if (!l--)
3846                 return (void*)2;/* tail */
3847         return NULL;
3848 }
3849
3850 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3851 {
3852         struct list_head *tmp;
3853         mddev_t *next_mddev, *mddev = v;
3854
3855         ++*pos;
3856         if (v == (void*)2)
3857                 return NULL;
3858
3859         spin_lock(&all_mddevs_lock);
3860         if (v == (void*)1)
3861                 tmp = all_mddevs.next;
3862         else
3863                 tmp = mddev->all_mddevs.next;
3864         if (tmp != &all_mddevs)
3865                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3866         else {
3867                 next_mddev = (void*)2;
3868                 *pos = 0x10000;
3869         }
3870         spin_unlock(&all_mddevs_lock);
3871
3872         if (v != (void*)1)
3873                 mddev_put(mddev);
3874         return next_mddev;
3875
3876 }
3877
3878 static void md_seq_stop(struct seq_file *seq, void *v)
3879 {
3880         mddev_t *mddev = v;
3881
3882         if (mddev && v != (void*)1 && v != (void*)2)
3883                 mddev_put(mddev);
3884 }
3885
3886 struct mdstat_info {
3887         int event;
3888 };
3889
3890 static int md_seq_show(struct seq_file *seq, void *v)
3891 {
3892         mddev_t *mddev = v;
3893         sector_t size;
3894         struct list_head *tmp2;
3895         mdk_rdev_t *rdev;
3896         struct mdstat_info *mi = seq->private;
3897         struct bitmap *bitmap;
3898
3899         if (v == (void*)1) {
3900                 struct mdk_personality *pers;
3901                 seq_printf(seq, "Personalities : ");
3902                 spin_lock(&pers_lock);
3903                 list_for_each_entry(pers, &pers_list, list)
3904                         seq_printf(seq, "[%s] ", pers->name);
3905
3906                 spin_unlock(&pers_lock);
3907                 seq_printf(seq, "\n");
3908                 mi->event = atomic_read(&md_event_count);
3909                 return 0;
3910         }
3911         if (v == (void*)2) {
3912                 status_unused(seq);
3913                 return 0;
3914         }
3915
3916         if (mddev_lock(mddev)!=0)
3917                 return -EINTR;
3918         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3919                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3920                                                 mddev->pers ? "" : "in");
3921                 if (mddev->pers) {
3922                         if (mddev->ro==1)
3923                                 seq_printf(seq, " (read-only)");
3924                         if (mddev->ro==2)
3925                                 seq_printf(seq, "(auto-read-only)");
3926                         seq_printf(seq, " %s", mddev->pers->name);
3927                 }
3928
3929                 size = 0;
3930                 ITERATE_RDEV(mddev,rdev,tmp2) {
3931                         char b[BDEVNAME_SIZE];
3932                         seq_printf(seq, " %s[%d]",
3933                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3934                         if (test_bit(WriteMostly, &rdev->flags))
3935                                 seq_printf(seq, "(W)");
3936                         if (test_bit(Faulty, &rdev->flags)) {
3937                                 seq_printf(seq, "(F)");
3938                                 continue;
3939                         } else if (rdev->raid_disk < 0)
3940                                 seq_printf(seq, "(S)"); /* spare */
3941                         size += rdev->size;
3942                 }
3943
3944                 if (!list_empty(&mddev->disks)) {
3945                         if (mddev->pers)
3946                                 seq_printf(seq, "\n      %llu blocks",
3947                                         (unsigned long long)mddev->array_size);
3948                         else
3949                                 seq_printf(seq, "\n      %llu blocks",
3950                                         (unsigned long long)size);
3951                 }
3952                 if (mddev->persistent) {
3953                         if (mddev->major_version != 0 ||
3954                             mddev->minor_version != 90) {
3955                                 seq_printf(seq," super %d.%d",
3956                                            mddev->major_version,
3957                                            mddev->minor_version);
3958                         }
3959                 } else
3960                         seq_printf(seq, " super non-persistent");
3961
3962                 if (mddev->pers) {
3963                         mddev->pers->status (seq, mddev);
3964                         seq_printf(seq, "\n      ");
3965                         if (mddev->pers->sync_request) {
3966                                 if (mddev->curr_resync > 2) {
3967                                         status_resync (seq, mddev);
3968                                         seq_printf(seq, "\n      ");
3969                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3970                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3971                                 else if (mddev->recovery_cp < MaxSector)
3972                                         seq_printf(seq, "\tresync=PENDING\n      ");
3973                         }
3974                 } else
3975                         seq_printf(seq, "\n       ");
3976
3977                 if ((bitmap = mddev->bitmap)) {
3978                         unsigned long chunk_kb;
3979                         unsigned long flags;
3980                         spin_lock_irqsave(&bitmap->lock, flags);
3981                         chunk_kb = bitmap->chunksize >> 10;
3982                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3983                                 "%lu%s chunk",
3984                                 bitmap->pages - bitmap->missing_pages,
3985                                 bitmap->pages,
3986                                 (bitmap->pages - bitmap->missing_pages)
3987                                         << (PAGE_SHIFT - 10),
3988                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3989                                 chunk_kb ? "KB" : "B");
3990                         if (bitmap->file) {
3991                                 seq_printf(seq, ", file: ");
3992                                 seq_path(seq, bitmap->file->f_vfsmnt,
3993                                          bitmap->file->f_dentry," \t\n");
3994                         }
3995
3996                         seq_printf(seq, "\n");
3997                         spin_unlock_irqrestore(&bitmap->lock, flags);
3998                 }
3999
4000                 seq_printf(seq, "\n");
4001         }
4002         mddev_unlock(mddev);
4003
4004         return 0;
4005 }
4006
4007 static struct seq_operations md_seq_ops = {
4008         .start  = md_seq_start,
4009         .next   = md_seq_next,
4010         .stop   = md_seq_stop,
4011         .show   = md_seq_show,
4012 };
4013
4014 static int md_seq_open(struct inode *inode, struct file *file)
4015 {
4016         int error;
4017         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4018         if (mi == NULL)
4019                 return -ENOMEM;
4020
4021         error = seq_open(file, &md_seq_ops);
4022         if (error)
4023                 kfree(mi);
4024         else {
4025                 struct seq_file *p = file->private_data;
4026                 p->private = mi;
4027                 mi->event = atomic_read(&md_event_count);
4028         }
4029         return error;
4030 }
4031
4032 static int md_seq_release(struct inode *inode, struct file *file)
4033 {
4034         struct seq_file *m = file->private_data;
4035         struct mdstat_info *mi = m->private;
4036         m->private = NULL;
4037         kfree(mi);
4038         return seq_release(inode, file);
4039 }
4040
4041 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4042 {
4043         struct seq_file *m = filp->private_data;
4044         struct mdstat_info *mi = m->private;
4045         int mask;
4046
4047         poll_wait(filp, &md_event_waiters, wait);
4048
4049         /* always allow read */
4050         mask = POLLIN | POLLRDNORM;
4051
4052         if (mi->event != atomic_read(&md_event_count))
4053                 mask |= POLLERR | POLLPRI;
4054         return mask;
4055 }
4056
4057 static struct file_operations md_seq_fops = {
4058         .open           = md_seq_open,
4059         .read           = seq_read,
4060         .llseek         = seq_lseek,
4061         .release        = md_seq_release,
4062         .poll           = mdstat_poll,
4063 };
4064
4065 int register_md_personality(struct mdk_personality *p)
4066 {
4067         spin_lock(&pers_lock);
4068         list_add_tail(&p->list, &pers_list);
4069         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4070         spin_unlock(&pers_lock);
4071         return 0;
4072 }
4073
4074 int unregister_md_personality(struct mdk_personality *p)
4075 {
4076         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4077         spin_lock(&pers_lock);
4078         list_del_init(&p->list);
4079         spin_unlock(&pers_lock);
4080         return 0;
4081 }
4082
4083 static int is_mddev_idle(mddev_t *mddev)
4084 {
4085         mdk_rdev_t * rdev;
4086         struct list_head *tmp;
4087         int idle;
4088         unsigned long curr_events;
4089
4090         idle = 1;
4091         ITERATE_RDEV(mddev,rdev,tmp) {
4092                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4093                 curr_events = disk_stat_read(disk, sectors[0]) +
4094                                 disk_stat_read(disk, sectors[1]) -
4095                                 atomic_read(&disk->sync_io);
4096                 /* The difference between curr_events and last_events
4097                  * will be affected by any new non-sync IO (making
4098                  * curr_events bigger) and any difference in the amount of
4099                  * in-flight syncio (making current_events bigger or smaller)
4100                  * The amount in-flight is currently limited to
4101                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4102                  * which is at most 4096 sectors.
4103                  * These numbers are fairly fragile and should be made
4104                  * more robust, probably by enforcing the
4105                  * 'window size' that md_do_sync sort-of uses.
4106                  *
4107                  * Note: the following is an unsigned comparison.
4108                  */
4109                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4110                         rdev->last_events = curr_events;
4111                         idle = 0;
4112                 }
4113         }
4114         return idle;
4115 }
4116
4117 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4118 {
4119         /* another "blocks" (512byte) blocks have been synced */
4120         atomic_sub(blocks, &mddev->recovery_active);
4121         wake_up(&mddev->recovery_wait);
4122         if (!ok) {
4123                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4124                 md_wakeup_thread(mddev->thread);
4125                 // stop recovery, signal do_sync ....
4126         }
4127 }
4128
4129
4130 /* md_write_start(mddev, bi)
4131  * If we need to update some array metadata (e.g. 'active' flag
4132  * in superblock) before writing, schedule a superblock update
4133  * and wait for it to complete.
4134  */
4135 void md_write_start(mddev_t *mddev, struct bio *bi)
4136 {
4137         if (bio_data_dir(bi) != WRITE)
4138                 return;
4139
4140         BUG_ON(mddev->ro == 1);
4141         if (mddev->ro == 2) {
4142                 /* need to switch to read/write */
4143                 mddev->ro = 0;
4144                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4145                 md_wakeup_thread(mddev->thread);
4146         }
4147         atomic_inc(&mddev->writes_pending);
4148         if (mddev->in_sync) {
4149                 spin_lock_irq(&mddev->write_lock);
4150                 if (mddev->in_sync) {
4151                         mddev->in_sync = 0;
4152                         mddev->sb_dirty = 1;
4153                         md_wakeup_thread(mddev->thread);
4154                 }
4155                 spin_unlock_irq(&mddev->write_lock);
4156         }
4157         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4158 }
4159
4160 void md_write_end(mddev_t *mddev)
4161 {
4162         if (atomic_dec_and_test(&mddev->writes_pending)) {
4163                 if (mddev->safemode == 2)
4164                         md_wakeup_thread(mddev->thread);
4165                 else
4166                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4167         }
4168 }
4169
4170 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4171
4172 #define SYNC_MARKS      10
4173 #define SYNC_MARK_STEP  (3*HZ)
4174 static void md_do_sync(mddev_t *mddev)
4175 {
4176         mddev_t *mddev2;
4177         unsigned int currspeed = 0,
4178                  window;
4179         sector_t max_sectors,j, io_sectors;
4180         unsigned long mark[SYNC_MARKS];
4181         sector_t mark_cnt[SYNC_MARKS];
4182         int last_mark,m;
4183         struct list_head *tmp;
4184         sector_t last_check;
4185         int skipped = 0;
4186
4187         /* just incase thread restarts... */
4188         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4189                 return;
4190
4191         /* we overload curr_resync somewhat here.
4192          * 0 == not engaged in resync at all
4193          * 2 == checking that there is no conflict with another sync
4194          * 1 == like 2, but have yielded to allow conflicting resync to
4195          *              commense
4196          * other == active in resync - this many blocks
4197          *
4198          * Before starting a resync we must have set curr_resync to
4199          * 2, and then checked that every "conflicting" array has curr_resync
4200          * less than ours.  When we find one that is the same or higher
4201          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4202          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4203          * This will mean we have to start checking from the beginning again.
4204          *
4205          */
4206
4207         do {
4208                 mddev->curr_resync = 2;
4209
4210         try_again:
4211                 if (kthread_should_stop()) {
4212                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4213                         goto skip;
4214                 }
4215                 ITERATE_MDDEV(mddev2,tmp) {
4216                         if (mddev2 == mddev)
4217                                 continue;
4218                         if (mddev2->curr_resync &&
4219                             match_mddev_units(mddev,mddev2)) {
4220                                 DEFINE_WAIT(wq);
4221                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4222                                         /* arbitrarily yield */
4223                                         mddev->curr_resync = 1;
4224                                         wake_up(&resync_wait);
4225                                 }
4226                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4227                                         /* no need to wait here, we can wait the next
4228                                          * time 'round when curr_resync == 2
4229                                          */
4230                                         continue;
4231                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4232                                 if (!kthread_should_stop() &&
4233                                     mddev2->curr_resync >= mddev->curr_resync) {
4234                                         printk(KERN_INFO "md: delaying resync of %s"
4235                                                " until %s has finished resync (they"
4236                                                " share one or more physical units)\n",
4237                                                mdname(mddev), mdname(mddev2));
4238                                         mddev_put(mddev2);
4239                                         schedule();
4240                                         finish_wait(&resync_wait, &wq);
4241                                         goto try_again;
4242                                 }
4243                                 finish_wait(&resync_wait, &wq);
4244                         }
4245                 }
4246         } while (mddev->curr_resync < 2);
4247
4248         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4249                 /* resync follows the size requested by the personality,
4250                  * which defaults to physical size, but can be virtual size
4251                  */
4252                 max_sectors = mddev->resync_max_sectors;
4253                 mddev->resync_mismatches = 0;
4254         } else
4255                 /* recovery follows the physical size of devices */
4256                 max_sectors = mddev->size << 1;
4257
4258         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4259         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4260                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4261         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4262                "(but not more than %d KB/sec) for reconstruction.\n",
4263                sysctl_speed_limit_max);
4264
4265         is_mddev_idle(mddev); /* this also initializes IO event counters */
4266         /* we don't use the checkpoint if there's a bitmap */
4267         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4268             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4269                 j = mddev->recovery_cp;
4270         else
4271                 j = 0;
4272         io_sectors = 0;
4273         for (m = 0; m < SYNC_MARKS; m++) {
4274                 mark[m] = jiffies;
4275                 mark_cnt[m] = io_sectors;
4276         }
4277         last_mark = 0;
4278         mddev->resync_mark = mark[last_mark];
4279         mddev->resync_mark_cnt = mark_cnt[last_mark];
4280
4281         /*
4282          * Tune reconstruction:
4283          */
4284         window = 32*(PAGE_SIZE/512);
4285         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4286                 window/2,(unsigned long long) max_sectors/2);
4287
4288         atomic_set(&mddev->recovery_active, 0);
4289         init_waitqueue_head(&mddev->recovery_wait);
4290         last_check = 0;
4291
4292         if (j>2) {
4293                 printk(KERN_INFO
4294                         "md: resuming recovery of %s from checkpoint.\n",
4295                         mdname(mddev));
4296                 mddev->curr_resync = j;
4297         }
4298
4299         while (j < max_sectors) {
4300                 sector_t sectors;
4301
4302                 skipped = 0;
4303                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4304                                             currspeed < sysctl_speed_limit_min);
4305                 if (sectors == 0) {
4306                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4307                         goto out;
4308                 }
4309
4310                 if (!skipped) { /* actual IO requested */
4311                         io_sectors += sectors;
4312                         atomic_add(sectors, &mddev->recovery_active);
4313                 }
4314
4315                 j += sectors;
4316                 if (j>1) mddev->curr_resync = j;
4317                 if (last_check == 0)
4318                         /* this is the earliers that rebuilt will be
4319                          * visible in /proc/mdstat
4320                          */
4321                         md_new_event(mddev);
4322
4323                 if (last_check + window > io_sectors || j == max_sectors)
4324                         continue;
4325
4326                 last_check = io_sectors;
4327
4328                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4329                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4330                         break;
4331
4332         repeat:
4333                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4334                         /* step marks */
4335                         int next = (last_mark+1) % SYNC_MARKS;
4336
4337                         mddev->resync_mark = mark[next];
4338                         mddev->resync_mark_cnt = mark_cnt[next];
4339                         mark[next] = jiffies;
4340                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4341                         last_mark = next;
4342                 }
4343
4344
4345                 if (kthread_should_stop()) {
4346                         /*
4347                          * got a signal, exit.
4348                          */
4349                         printk(KERN_INFO
4350                                 "md: md_do_sync() got signal ... exiting\n");
4351                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4352                         goto out;
4353                 }
4354
4355                 /*
4356                  * this loop exits only if either when we are slower than
4357                  * the 'hard' speed limit, or the system was IO-idle for
4358                  * a jiffy.
4359                  * the system might be non-idle CPU-wise, but we only care
4360                  * about not overloading the IO subsystem. (things like an
4361                  * e2fsck being done on the RAID array should execute fast)
4362                  */
4363                 mddev->queue->unplug_fn(mddev->queue);
4364                 cond_resched();
4365
4366                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4367                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4368
4369                 if (currspeed > sysctl_speed_limit_min) {
4370                         if ((currspeed > sysctl_speed_limit_max) ||
4371                                         !is_mddev_idle(mddev)) {
4372                                 msleep(500);
4373                                 goto repeat;
4374                         }
4375                 }
4376         }
4377         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4378         /*
4379          * this also signals 'finished resyncing' to md_stop
4380          */
4381  out:
4382         mddev->queue->unplug_fn(mddev->queue);
4383
4384         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4385
4386         /* tell personality that we are finished */
4387         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4388
4389         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4390             mddev->curr_resync > 2 &&
4391             mddev->curr_resync >= mddev->recovery_cp) {
4392                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4393                         printk(KERN_INFO
4394                                 "md: checkpointing recovery of %s.\n",
4395                                 mdname(mddev));
4396                         mddev->recovery_cp = mddev->curr_resync;
4397                 } else
4398                         mddev->recovery_cp = MaxSector;
4399         }
4400
4401  skip:
4402         mddev->curr_resync = 0;
4403         wake_up(&resync_wait);
4404         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4405         md_wakeup_thread(mddev->thread);
4406 }
4407
4408
4409 /*
4410  * This routine is regularly called by all per-raid-array threads to
4411  * deal with generic issues like resync and super-block update.
4412  * Raid personalities that don't have a thread (linear/raid0) do not
4413  * need this as they never do any recovery or update the superblock.
4414  *
4415  * It does not do any resync itself, but rather "forks" off other threads
4416  * to do that as needed.
4417  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4418  * "->recovery" and create a thread at ->sync_thread.
4419  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4420  * and wakeups up this thread which will reap the thread and finish up.
4421  * This thread also removes any faulty devices (with nr_pending == 0).
4422  *
4423  * The overall approach is:
4424  *  1/ if the superblock needs updating, update it.
4425  *  2/ If a recovery thread is running, don't do anything else.
4426  *  3/ If recovery has finished, clean up, possibly marking spares active.
4427  *  4/ If there are any faulty devices, remove them.
4428  *  5/ If array is degraded, try to add spares devices
4429  *  6/ If array has spares or is not in-sync, start a resync thread.
4430  */
4431 void md_check_recovery(mddev_t *mddev)
4432 {
4433         mdk_rdev_t *rdev;
4434         struct list_head *rtmp;
4435
4436
4437         if (mddev->bitmap)
4438                 bitmap_daemon_work(mddev->bitmap);
4439
4440         if (mddev->ro)
4441                 return;
4442
4443         if (signal_pending(current)) {
4444                 if (mddev->pers->sync_request) {
4445                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4446                                mdname(mddev));
4447                         mddev->safemode = 2;
4448                 }
4449                 flush_signals(current);
4450         }
4451
4452         if ( ! (
4453                 mddev->sb_dirty ||
4454                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4455                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4456                 (mddev->safemode == 1) ||
4457                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4458                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4459                 ))
4460                 return;
4461
4462         if (mddev_trylock(mddev)==0) {
4463                 int spares =0;
4464
4465                 spin_lock_irq(&mddev->write_lock);
4466                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4467                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4468                         mddev->in_sync = 1;
4469                         mddev->sb_dirty = 1;
4470                 }
4471                 if (mddev->safemode == 1)
4472                         mddev->safemode = 0;
4473                 spin_unlock_irq(&mddev->write_lock);
4474
4475                 if (mddev->sb_dirty)
4476                         md_update_sb(mddev);
4477
4478
4479                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4480                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4481                         /* resync/recovery still happening */
4482                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4483                         goto unlock;
4484                 }
4485                 if (mddev->sync_thread) {
4486                         /* resync has finished, collect result */
4487                         md_unregister_thread(mddev->sync_thread);
4488                         mddev->sync_thread = NULL;
4489                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4490                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4491                                 /* success...*/
4492                                 /* activate any spares */
4493                                 mddev->pers->spare_active(mddev);
4494                         }
4495                         md_update_sb(mddev);
4496
4497                         /* if array is no-longer degraded, then any saved_raid_disk
4498                          * information must be scrapped
4499                          */
4500                         if (!mddev->degraded)
4501                                 ITERATE_RDEV(mddev,rdev,rtmp)
4502                                         rdev->saved_raid_disk = -1;
4503
4504                         mddev->recovery = 0;
4505                         /* flag recovery needed just to double check */
4506                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4507                         md_new_event(mddev);
4508                         goto unlock;
4509                 }
4510                 /* Clear some bits that don't mean anything, but
4511                  * might be left set
4512                  */
4513                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4514                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4515                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4516                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4517
4518                 /* no recovery is running.
4519                  * remove any failed drives, then
4520                  * add spares if possible.
4521                  * Spare are also removed and re-added, to allow
4522                  * the personality to fail the re-add.
4523                  */
4524                 ITERATE_RDEV(mddev,rdev,rtmp)
4525                         if (rdev->raid_disk >= 0 &&
4526                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4527                             atomic_read(&rdev->nr_pending)==0) {
4528                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4529                                         char nm[20];
4530                                         sprintf(nm,"rd%d", rdev->raid_disk);
4531                                         sysfs_remove_link(&mddev->kobj, nm);
4532                                         rdev->raid_disk = -1;
4533                                 }
4534                         }
4535
4536                 if (mddev->degraded) {
4537                         ITERATE_RDEV(mddev,rdev,rtmp)
4538                                 if (rdev->raid_disk < 0
4539                                     && !test_bit(Faulty, &rdev->flags)) {
4540                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4541                                                 char nm[20];
4542                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4543                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4544                                                 spares++;
4545                                                 md_new_event(mddev);
4546                                         } else
4547                                                 break;
4548                                 }
4549                 }
4550
4551                 if (spares) {
4552                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4553                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4554                 } else if (mddev->recovery_cp < MaxSector) {
4555                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4556                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4557                         /* nothing to be done ... */
4558                         goto unlock;
4559
4560                 if (mddev->pers->sync_request) {
4561                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4562                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4563                                 /* We are adding a device or devices to an array
4564                                  * which has the bitmap stored on all devices.
4565                                  * So make sure all bitmap pages get written
4566                                  */
4567                                 bitmap_write_all(mddev->bitmap);
4568                         }
4569                         mddev->sync_thread = md_register_thread(md_do_sync,
4570                                                                 mddev,
4571                                                                 "%s_resync");
4572                         if (!mddev->sync_thread) {
4573                                 printk(KERN_ERR "%s: could not start resync"
4574                                         " thread...\n",
4575                                         mdname(mddev));
4576                                 /* leave the spares where they are, it shouldn't hurt */
4577                                 mddev->recovery = 0;
4578                         } else
4579                                 md_wakeup_thread(mddev->sync_thread);
4580                         md_new_event(mddev);
4581                 }
4582         unlock:
4583                 mddev_unlock(mddev);
4584         }
4585 }
4586
4587 static int md_notify_reboot(struct notifier_block *this,
4588                             unsigned long code, void *x)
4589 {
4590         struct list_head *tmp;
4591         mddev_t *mddev;
4592
4593         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4594
4595                 printk(KERN_INFO "md: stopping all md devices.\n");
4596
4597                 ITERATE_MDDEV(mddev,tmp)
4598                         if (mddev_trylock(mddev)==0)
4599                                 do_md_stop (mddev, 1);
4600                 /*
4601                  * certain more exotic SCSI devices are known to be
4602                  * volatile wrt too early system reboots. While the
4603                  * right place to handle this issue is the given
4604                  * driver, we do want to have a safe RAID driver ...
4605                  */
4606                 mdelay(1000*1);
4607         }
4608         return NOTIFY_DONE;
4609 }
4610
4611 static struct notifier_block md_notifier = {
4612         .notifier_call  = md_notify_reboot,
4613         .next           = NULL,
4614         .priority       = INT_MAX, /* before any real devices */
4615 };
4616
4617 static void md_geninit(void)
4618 {
4619         struct proc_dir_entry *p;
4620
4621         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4622
4623         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4624         if (p)
4625                 p->proc_fops = &md_seq_fops;
4626 }
4627
4628 static int __init md_init(void)
4629 {
4630         int minor;
4631
4632         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4633                         " MD_SB_DISKS=%d\n",
4634                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4635                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4636         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4637                         BITMAP_MINOR);
4638
4639         if (register_blkdev(MAJOR_NR, "md"))
4640                 return -1;
4641         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4642                 unregister_blkdev(MAJOR_NR, "md");
4643                 return -1;
4644         }
4645         devfs_mk_dir("md");
4646         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4647                                 md_probe, NULL, NULL);
4648         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4649                             md_probe, NULL, NULL);
4650
4651         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4652                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4653                                 S_IFBLK|S_IRUSR|S_IWUSR,
4654                                 "md/%d", minor);
4655
4656         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4657                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4658                               S_IFBLK|S_IRUSR|S_IWUSR,
4659                               "md/mdp%d", minor);
4660
4661
4662         register_reboot_notifier(&md_notifier);
4663         raid_table_header = register_sysctl_table(raid_root_table, 1);
4664
4665         md_geninit();
4666         return (0);
4667 }
4668
4669
4670 #ifndef MODULE
4671
4672 /*
4673  * Searches all registered partitions for autorun RAID arrays
4674  * at boot time.
4675  */
4676 static dev_t detected_devices[128];
4677 static int dev_cnt;
4678
4679 void md_autodetect_dev(dev_t dev)
4680 {
4681         if (dev_cnt >= 0 && dev_cnt < 127)
4682                 detected_devices[dev_cnt++] = dev;
4683 }
4684
4685
4686 static void autostart_arrays(int part)
4687 {
4688         mdk_rdev_t *rdev;
4689         int i;
4690
4691         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4692
4693         for (i = 0; i < dev_cnt; i++) {
4694                 dev_t dev = detected_devices[i];
4695
4696                 rdev = md_import_device(dev,0, 0);
4697                 if (IS_ERR(rdev))
4698                         continue;
4699
4700                 if (test_bit(Faulty, &rdev->flags)) {
4701                         MD_BUG();
4702                         continue;
4703                 }
4704                 list_add(&rdev->same_set, &pending_raid_disks);
4705         }
4706         dev_cnt = 0;
4707
4708         autorun_devices(part);
4709 }
4710
4711 #endif
4712
4713 static __exit void md_exit(void)
4714 {
4715         mddev_t *mddev;
4716         struct list_head *tmp;
4717         int i;
4718         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4719         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4720         for (i=0; i < MAX_MD_DEVS; i++)
4721                 devfs_remove("md/%d", i);
4722         for (i=0; i < MAX_MD_DEVS; i++)
4723                 devfs_remove("md/d%d", i);
4724
4725         devfs_remove("md");
4726
4727         unregister_blkdev(MAJOR_NR,"md");
4728         unregister_blkdev(mdp_major, "mdp");
4729         unregister_reboot_notifier(&md_notifier);
4730         unregister_sysctl_table(raid_table_header);
4731         remove_proc_entry("mdstat", NULL);
4732         ITERATE_MDDEV(mddev,tmp) {
4733                 struct gendisk *disk = mddev->gendisk;
4734                 if (!disk)
4735                         continue;
4736                 export_array(mddev);
4737                 del_gendisk(disk);
4738                 put_disk(disk);
4739                 mddev->gendisk = NULL;
4740                 mddev_put(mddev);
4741         }
4742 }
4743
4744 module_init(md_init)
4745 module_exit(md_exit)
4746
4747 static int get_ro(char *buffer, struct kernel_param *kp)
4748 {
4749         return sprintf(buffer, "%d", start_readonly);
4750 }
4751 static int set_ro(const char *val, struct kernel_param *kp)
4752 {
4753         char *e;
4754         int num = simple_strtoul(val, &e, 10);
4755         if (*val && (*e == '\0' || *e == '\n')) {
4756                 start_readonly = num;
4757                 return 0;
4758         }
4759         return -EINVAL;
4760 }
4761
4762 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4763 module_param(start_dirty_degraded, int, 0644);
4764
4765
4766 EXPORT_SYMBOL(register_md_personality);
4767 EXPORT_SYMBOL(unregister_md_personality);
4768 EXPORT_SYMBOL(md_error);
4769 EXPORT_SYMBOL(md_done_sync);
4770 EXPORT_SYMBOL(md_write_start);
4771 EXPORT_SYMBOL(md_write_end);
4772 EXPORT_SYMBOL(md_register_thread);
4773 EXPORT_SYMBOL(md_unregister_thread);
4774 EXPORT_SYMBOL(md_wakeup_thread);
4775 EXPORT_SYMBOL(md_print_devices);
4776 EXPORT_SYMBOL(md_check_recovery);
4777 MODULE_LICENSE("GPL");
4778 MODULE_ALIAS("md");
4779 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);