err.no Git - linux-2.6/blob - drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/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
1004         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1005         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1006         if (rdev->sb_size & bmask)
1007                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1008
1009         if (refdev == 0)
1010                 return 1;
1011         else {
1012                 __u64 ev1, ev2;
1013                 struct mdp_superblock_1 *refsb =
1014                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1015
1016                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1017                     sb->level != refsb->level ||
1018                     sb->layout != refsb->layout ||
1019                     sb->chunksize != refsb->chunksize) {
1020                         printk(KERN_WARNING "md: %s has strangely different"
1021                                 " superblock to %s\n",
1022                                 bdevname(rdev->bdev,b),
1023                                 bdevname(refdev->bdev,b2));
1024                         return -EINVAL;
1025                 }
1026                 ev1 = le64_to_cpu(sb->events);
1027                 ev2 = le64_to_cpu(refsb->events);
1028
1029                 if (ev1 > ev2)
1030                         return 1;
1031         }
1032         if (minor_version)
1033                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1034         else
1035                 rdev->size = rdev->sb_offset;
1036         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1037                 return -EINVAL;
1038         rdev->size = le64_to_cpu(sb->data_size)/2;
1039         if (le32_to_cpu(sb->chunksize))
1040                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1041         return 0;
1042 }
1043
1044 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1045 {
1046         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1047
1048         rdev->raid_disk = -1;
1049         rdev->flags = 0;
1050         if (mddev->raid_disks == 0) {
1051                 mddev->major_version = 1;
1052                 mddev->patch_version = 0;
1053                 mddev->persistent = 1;
1054                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1055                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1056                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1057                 mddev->level = le32_to_cpu(sb->level);
1058                 mddev->clevel[0] = 0;
1059                 mddev->layout = le32_to_cpu(sb->layout);
1060                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1061                 mddev->size = le64_to_cpu(sb->size)/2;
1062                 mddev->events = le64_to_cpu(sb->events);
1063                 mddev->bitmap_offset = 0;
1064                 mddev->default_bitmap_offset = 1024;
1065
1066                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1067                 memcpy(mddev->uuid, sb->set_uuid, 16);
1068
1069                 mddev->max_disks =  (4096-256)/2;
1070
1071                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1072                     mddev->bitmap_file == NULL ) {
1073                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1074                             && mddev->level != 10) {
1075                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1076                                 return -EINVAL;
1077                         }
1078                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1079                 }
1080         } else if (mddev->pers == NULL) {
1081                 /* Insist of good event counter while assembling */
1082                 __u64 ev1 = le64_to_cpu(sb->events);
1083                 ++ev1;
1084                 if (ev1 < mddev->events)
1085                         return -EINVAL;
1086         } else if (mddev->bitmap) {
1087                 /* If adding to array with a bitmap, then we can accept an
1088                  * older device, but not too old.
1089                  */
1090                 __u64 ev1 = le64_to_cpu(sb->events);
1091                 if (ev1 < mddev->bitmap->events_cleared)
1092                         return 0;
1093         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1094                 return 0;
1095
1096         if (mddev->level != LEVEL_MULTIPATH) {
1097                 int role;
1098                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1099                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1100                 switch(role) {
1101                 case 0xffff: /* spare */
1102                         break;
1103                 case 0xfffe: /* faulty */
1104                         set_bit(Faulty, &rdev->flags);
1105                         break;
1106                 default:
1107                         set_bit(In_sync, &rdev->flags);
1108                         rdev->raid_disk = role;
1109                         break;
1110                 }
1111                 if (sb->devflags & WriteMostly1)
1112                         set_bit(WriteMostly, &rdev->flags);
1113         } else /* MULTIPATH are always insync */
1114                 set_bit(In_sync, &rdev->flags);
1115
1116         return 0;
1117 }
1118
1119 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1120 {
1121         struct mdp_superblock_1 *sb;
1122         struct list_head *tmp;
1123         mdk_rdev_t *rdev2;
1124         int max_dev, i;
1125         /* make rdev->sb match mddev and rdev data. */
1126
1127         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1128
1129         sb->feature_map = 0;
1130         sb->pad0 = 0;
1131         memset(sb->pad1, 0, sizeof(sb->pad1));
1132         memset(sb->pad2, 0, sizeof(sb->pad2));
1133         memset(sb->pad3, 0, sizeof(sb->pad3));
1134
1135         sb->utime = cpu_to_le64((__u64)mddev->utime);
1136         sb->events = cpu_to_le64(mddev->events);
1137         if (mddev->in_sync)
1138                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1139         else
1140                 sb->resync_offset = cpu_to_le64(0);
1141
1142         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1143                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1144                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1145         }
1146
1147         max_dev = 0;
1148         ITERATE_RDEV(mddev,rdev2,tmp)
1149                 if (rdev2->desc_nr+1 > max_dev)
1150                         max_dev = rdev2->desc_nr+1;
1151
1152         sb->max_dev = cpu_to_le32(max_dev);
1153         for (i=0; i<max_dev;i++)
1154                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1155
1156         ITERATE_RDEV(mddev,rdev2,tmp) {
1157                 i = rdev2->desc_nr;
1158                 if (test_bit(Faulty, &rdev2->flags))
1159                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1160                 else if (test_bit(In_sync, &rdev2->flags))
1161                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1162                 else
1163                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1164         }
1165
1166         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1167         sb->sb_csum = calc_sb_1_csum(sb);
1168 }
1169
1170
1171 static struct super_type super_types[] = {
1172         [0] = {
1173                 .name   = "0.90.0",
1174                 .owner  = THIS_MODULE,
1175                 .load_super     = super_90_load,
1176                 .validate_super = super_90_validate,
1177                 .sync_super     = super_90_sync,
1178         },
1179         [1] = {
1180                 .name   = "md-1",
1181                 .owner  = THIS_MODULE,
1182                 .load_super     = super_1_load,
1183                 .validate_super = super_1_validate,
1184                 .sync_super     = super_1_sync,
1185         },
1186 };
1187
1188 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1189 {
1190         struct list_head *tmp;
1191         mdk_rdev_t *rdev;
1192
1193         ITERATE_RDEV(mddev,rdev,tmp)
1194                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1195                         return rdev;
1196
1197         return NULL;
1198 }
1199
1200 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1201 {
1202         struct list_head *tmp;
1203         mdk_rdev_t *rdev;
1204
1205         ITERATE_RDEV(mddev1,rdev,tmp)
1206                 if (match_dev_unit(mddev2, rdev))
1207                         return 1;
1208
1209         return 0;
1210 }
1211
1212 static LIST_HEAD(pending_raid_disks);
1213
1214 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1215 {
1216         mdk_rdev_t *same_pdev;
1217         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1218         struct kobject *ko;
1219
1220         if (rdev->mddev) {
1221                 MD_BUG();
1222                 return -EINVAL;
1223         }
1224         same_pdev = match_dev_unit(mddev, rdev);
1225         if (same_pdev)
1226                 printk(KERN_WARNING
1227                         "%s: WARNING: %s appears to be on the same physical"
1228                         " disk as %s. True\n     protection against single-disk"
1229                         " failure might be compromised.\n",
1230                         mdname(mddev), bdevname(rdev->bdev,b),
1231                         bdevname(same_pdev->bdev,b2));
1232
1233         /* Verify rdev->desc_nr is unique.
1234          * If it is -1, assign a free number, else
1235          * check number is not in use
1236          */
1237         if (rdev->desc_nr < 0) {
1238                 int choice = 0;
1239                 if (mddev->pers) choice = mddev->raid_disks;
1240                 while (find_rdev_nr(mddev, choice))
1241                         choice++;
1242                 rdev->desc_nr = choice;
1243         } else {
1244                 if (find_rdev_nr(mddev, rdev->desc_nr))
1245                         return -EBUSY;
1246         }
1247         bdevname(rdev->bdev,b);
1248         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1249                 return -ENOMEM;
1250
1251         list_add(&rdev->same_set, &mddev->disks);
1252         rdev->mddev = mddev;
1253         printk(KERN_INFO "md: bind<%s>\n", b);
1254
1255         rdev->kobj.parent = &mddev->kobj;
1256         kobject_add(&rdev->kobj);
1257
1258         if (rdev->bdev->bd_part)
1259                 ko = &rdev->bdev->bd_part->kobj;
1260         else
1261                 ko = &rdev->bdev->bd_disk->kobj;
1262         sysfs_create_link(&rdev->kobj, ko, "block");
1263         return 0;
1264 }
1265
1266 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1267 {
1268         char b[BDEVNAME_SIZE];
1269         if (!rdev->mddev) {
1270                 MD_BUG();
1271                 return;
1272         }
1273         list_del_init(&rdev->same_set);
1274         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1275         rdev->mddev = NULL;
1276         sysfs_remove_link(&rdev->kobj, "block");
1277         kobject_del(&rdev->kobj);
1278 }
1279
1280 /*
1281  * prevent the device from being mounted, repartitioned or
1282  * otherwise reused by a RAID array (or any other kernel
1283  * subsystem), by bd_claiming the device.
1284  */
1285 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1286 {
1287         int err = 0;
1288         struct block_device *bdev;
1289         char b[BDEVNAME_SIZE];
1290
1291         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1292         if (IS_ERR(bdev)) {
1293                 printk(KERN_ERR "md: could not open %s.\n",
1294                         __bdevname(dev, b));
1295                 return PTR_ERR(bdev);
1296         }
1297         err = bd_claim(bdev, rdev);
1298         if (err) {
1299                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1300                         bdevname(bdev, b));
1301                 blkdev_put(bdev);
1302                 return err;
1303         }
1304         rdev->bdev = bdev;
1305         return err;
1306 }
1307
1308 static void unlock_rdev(mdk_rdev_t *rdev)
1309 {
1310         struct block_device *bdev = rdev->bdev;
1311         rdev->bdev = NULL;
1312         if (!bdev)
1313                 MD_BUG();
1314         bd_release(bdev);
1315         blkdev_put(bdev);
1316 }
1317
1318 void md_autodetect_dev(dev_t dev);
1319
1320 static void export_rdev(mdk_rdev_t * rdev)
1321 {
1322         char b[BDEVNAME_SIZE];
1323         printk(KERN_INFO "md: export_rdev(%s)\n",
1324                 bdevname(rdev->bdev,b));
1325         if (rdev->mddev)
1326                 MD_BUG();
1327         free_disk_sb(rdev);
1328         list_del_init(&rdev->same_set);
1329 #ifndef MODULE
1330         md_autodetect_dev(rdev->bdev->bd_dev);
1331 #endif
1332         unlock_rdev(rdev);
1333         kobject_put(&rdev->kobj);
1334 }
1335
1336 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1337 {
1338         unbind_rdev_from_array(rdev);
1339         export_rdev(rdev);
1340 }
1341
1342 static void export_array(mddev_t *mddev)
1343 {
1344         struct list_head *tmp;
1345         mdk_rdev_t *rdev;
1346
1347         ITERATE_RDEV(mddev,rdev,tmp) {
1348                 if (!rdev->mddev) {
1349                         MD_BUG();
1350                         continue;
1351                 }
1352                 kick_rdev_from_array(rdev);
1353         }
1354         if (!list_empty(&mddev->disks))
1355                 MD_BUG();
1356         mddev->raid_disks = 0;
1357         mddev->major_version = 0;
1358 }
1359
1360 static void print_desc(mdp_disk_t *desc)
1361 {
1362         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1363                 desc->major,desc->minor,desc->raid_disk,desc->state);
1364 }
1365
1366 static void print_sb(mdp_super_t *sb)
1367 {
1368         int i;
1369
1370         printk(KERN_INFO
1371                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1372                 sb->major_version, sb->minor_version, sb->patch_version,
1373                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1374                 sb->ctime);
1375         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1376                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1377                 sb->md_minor, sb->layout, sb->chunk_size);
1378         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1379                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1380                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1381                 sb->failed_disks, sb->spare_disks,
1382                 sb->sb_csum, (unsigned long)sb->events_lo);
1383
1384         printk(KERN_INFO);
1385         for (i = 0; i < MD_SB_DISKS; i++) {
1386                 mdp_disk_t *desc;
1387
1388                 desc = sb->disks + i;
1389                 if (desc->number || desc->major || desc->minor ||
1390                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1391                         printk("     D %2d: ", i);
1392                         print_desc(desc);
1393                 }
1394         }
1395         printk(KERN_INFO "md:     THIS: ");
1396         print_desc(&sb->this_disk);
1397
1398 }
1399
1400 static void print_rdev(mdk_rdev_t *rdev)
1401 {
1402         char b[BDEVNAME_SIZE];
1403         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1404                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1405                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1406                 rdev->desc_nr);
1407         if (rdev->sb_loaded) {
1408                 printk(KERN_INFO "md: rdev superblock:\n");
1409                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1410         } else
1411                 printk(KERN_INFO "md: no rdev superblock!\n");
1412 }
1413
1414 void md_print_devices(void)
1415 {
1416         struct list_head *tmp, *tmp2;
1417         mdk_rdev_t *rdev;
1418         mddev_t *mddev;
1419         char b[BDEVNAME_SIZE];
1420
1421         printk("\n");
1422         printk("md:     **********************************\n");
1423         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1424         printk("md:     **********************************\n");
1425         ITERATE_MDDEV(mddev,tmp) {
1426
1427                 if (mddev->bitmap)
1428                         bitmap_print_sb(mddev->bitmap);
1429                 else
1430                         printk("%s: ", mdname(mddev));
1431                 ITERATE_RDEV(mddev,rdev,tmp2)
1432                         printk("<%s>", bdevname(rdev->bdev,b));
1433                 printk("\n");
1434
1435                 ITERATE_RDEV(mddev,rdev,tmp2)
1436                         print_rdev(rdev);
1437         }
1438         printk("md:     **********************************\n");
1439         printk("\n");
1440 }
1441
1442
1443 static void sync_sbs(mddev_t * mddev)
1444 {
1445         mdk_rdev_t *rdev;
1446         struct list_head *tmp;
1447
1448         ITERATE_RDEV(mddev,rdev,tmp) {
1449                 super_types[mddev->major_version].
1450                         sync_super(mddev, rdev);
1451                 rdev->sb_loaded = 1;
1452         }
1453 }
1454
1455 static void md_update_sb(mddev_t * mddev)
1456 {
1457         int err;
1458         struct list_head *tmp;
1459         mdk_rdev_t *rdev;
1460         int sync_req;
1461
1462 repeat:
1463         spin_lock_irq(&mddev->write_lock);
1464         sync_req = mddev->in_sync;
1465         mddev->utime = get_seconds();
1466         mddev->events ++;
1467
1468         if (!mddev->events) {
1469                 /*
1470                  * oops, this 64-bit counter should never wrap.
1471                  * Either we are in around ~1 trillion A.C., assuming
1472                  * 1 reboot per second, or we have a bug:
1473                  */
1474                 MD_BUG();
1475                 mddev->events --;
1476         }
1477         mddev->sb_dirty = 2;
1478         sync_sbs(mddev);
1479
1480         /*
1481          * do not write anything to disk if using
1482          * nonpersistent superblocks
1483          */
1484         if (!mddev->persistent) {
1485                 mddev->sb_dirty = 0;
1486                 spin_unlock_irq(&mddev->write_lock);
1487                 wake_up(&mddev->sb_wait);
1488                 return;
1489         }
1490         spin_unlock_irq(&mddev->write_lock);
1491
1492         dprintk(KERN_INFO
1493                 "md: updating %s RAID superblock on device (in sync %d)\n",
1494                 mdname(mddev),mddev->in_sync);
1495
1496         err = bitmap_update_sb(mddev->bitmap);
1497         ITERATE_RDEV(mddev,rdev,tmp) {
1498                 char b[BDEVNAME_SIZE];
1499                 dprintk(KERN_INFO "md: ");
1500                 if (test_bit(Faulty, &rdev->flags))
1501                         dprintk("(skipping faulty ");
1502
1503                 dprintk("%s ", bdevname(rdev->bdev,b));
1504                 if (!test_bit(Faulty, &rdev->flags)) {
1505                         md_super_write(mddev,rdev,
1506                                        rdev->sb_offset<<1, rdev->sb_size,
1507                                        rdev->sb_page);
1508                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1509                                 bdevname(rdev->bdev,b),
1510                                 (unsigned long long)rdev->sb_offset);
1511
1512                 } else
1513                         dprintk(")\n");
1514                 if (mddev->level == LEVEL_MULTIPATH)
1515                         /* only need to write one superblock... */
1516                         break;
1517         }
1518         md_super_wait(mddev);
1519         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1520
1521         spin_lock_irq(&mddev->write_lock);
1522         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1523                 /* have to write it out again */
1524                 spin_unlock_irq(&mddev->write_lock);
1525                 goto repeat;
1526         }
1527         mddev->sb_dirty = 0;
1528         spin_unlock_irq(&mddev->write_lock);
1529         wake_up(&mddev->sb_wait);
1530
1531 }
1532
1533 /* words written to sysfs files may, or my not, be \n terminated.
1534  * We want to accept with case. For this we use cmd_match.
1535  */
1536 static int cmd_match(const char *cmd, const char *str)
1537 {
1538         /* See if cmd, written into a sysfs file, matches
1539          * str.  They must either be the same, or cmd can
1540          * have a trailing newline
1541          */
1542         while (*cmd && *str && *cmd == *str) {
1543                 cmd++;
1544                 str++;
1545         }
1546         if (*cmd == '\n')
1547                 cmd++;
1548         if (*str || *cmd)
1549                 return 0;
1550         return 1;
1551 }
1552
1553 struct rdev_sysfs_entry {
1554         struct attribute attr;
1555         ssize_t (*show)(mdk_rdev_t *, char *);
1556         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1557 };
1558
1559 static ssize_t
1560 state_show(mdk_rdev_t *rdev, char *page)
1561 {
1562         char *sep = "";
1563         int len=0;
1564
1565         if (test_bit(Faulty, &rdev->flags)) {
1566                 len+= sprintf(page+len, "%sfaulty",sep);
1567                 sep = ",";
1568         }
1569         if (test_bit(In_sync, &rdev->flags)) {
1570                 len += sprintf(page+len, "%sin_sync",sep);
1571                 sep = ",";
1572         }
1573         if (!test_bit(Faulty, &rdev->flags) &&
1574             !test_bit(In_sync, &rdev->flags)) {
1575                 len += sprintf(page+len, "%sspare", sep);
1576                 sep = ",";
1577         }
1578         return len+sprintf(page+len, "\n");
1579 }
1580
1581 static struct rdev_sysfs_entry
1582 rdev_state = __ATTR_RO(state);
1583
1584 static ssize_t
1585 super_show(mdk_rdev_t *rdev, char *page)
1586 {
1587         if (rdev->sb_loaded && rdev->sb_size) {
1588                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1589                 return rdev->sb_size;
1590         } else
1591                 return 0;
1592 }
1593 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1594
1595 static struct attribute *rdev_default_attrs[] = {
1596         &rdev_state.attr,
1597         &rdev_super.attr,
1598         NULL,
1599 };
1600 static ssize_t
1601 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1602 {
1603         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1604         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1605
1606         if (!entry->show)
1607                 return -EIO;
1608         return entry->show(rdev, page);
1609 }
1610
1611 static ssize_t
1612 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1613               const char *page, size_t length)
1614 {
1615         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1616         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1617
1618         if (!entry->store)
1619                 return -EIO;
1620         return entry->store(rdev, page, length);
1621 }
1622
1623 static void rdev_free(struct kobject *ko)
1624 {
1625         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1626         kfree(rdev);
1627 }
1628 static struct sysfs_ops rdev_sysfs_ops = {
1629         .show           = rdev_attr_show,
1630         .store          = rdev_attr_store,
1631 };
1632 static struct kobj_type rdev_ktype = {
1633         .release        = rdev_free,
1634         .sysfs_ops      = &rdev_sysfs_ops,
1635         .default_attrs  = rdev_default_attrs,
1636 };
1637
1638 /*
1639  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1640  *
1641  * mark the device faulty if:
1642  *
1643  *   - the device is nonexistent (zero size)
1644  *   - the device has no valid superblock
1645  *
1646  * a faulty rdev _never_ has rdev->sb set.
1647  */
1648 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1649 {
1650         char b[BDEVNAME_SIZE];
1651         int err;
1652         mdk_rdev_t *rdev;
1653         sector_t size;
1654
1655         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1656         if (!rdev) {
1657                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1658                 return ERR_PTR(-ENOMEM);
1659         }
1660
1661         if ((err = alloc_disk_sb(rdev)))
1662                 goto abort_free;
1663
1664         err = lock_rdev(rdev, newdev);
1665         if (err)
1666                 goto abort_free;
1667
1668         rdev->kobj.parent = NULL;
1669         rdev->kobj.ktype = &rdev_ktype;
1670         kobject_init(&rdev->kobj);
1671
1672         rdev->desc_nr = -1;
1673         rdev->flags = 0;
1674         rdev->data_offset = 0;
1675         atomic_set(&rdev->nr_pending, 0);
1676         atomic_set(&rdev->read_errors, 0);
1677
1678         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1679         if (!size) {
1680                 printk(KERN_WARNING
1681                         "md: %s has zero or unknown size, marking faulty!\n",
1682                         bdevname(rdev->bdev,b));
1683                 err = -EINVAL;
1684                 goto abort_free;
1685         }
1686
1687         if (super_format >= 0) {
1688                 err = super_types[super_format].
1689                         load_super(rdev, NULL, super_minor);
1690                 if (err == -EINVAL) {
1691                         printk(KERN_WARNING
1692                                 "md: %s has invalid sb, not importing!\n",
1693                                 bdevname(rdev->bdev,b));
1694                         goto abort_free;
1695                 }
1696                 if (err < 0) {
1697                         printk(KERN_WARNING
1698                                 "md: could not read %s's sb, not importing!\n",
1699                                 bdevname(rdev->bdev,b));
1700                         goto abort_free;
1701                 }
1702         }
1703         INIT_LIST_HEAD(&rdev->same_set);
1704
1705         return rdev;
1706
1707 abort_free:
1708         if (rdev->sb_page) {
1709                 if (rdev->bdev)
1710                         unlock_rdev(rdev);
1711                 free_disk_sb(rdev);
1712         }
1713         kfree(rdev);
1714         return ERR_PTR(err);
1715 }
1716
1717 /*
1718  * Check a full RAID array for plausibility
1719  */
1720
1721
1722 static void analyze_sbs(mddev_t * mddev)
1723 {
1724         int i;
1725         struct list_head *tmp;
1726         mdk_rdev_t *rdev, *freshest;
1727         char b[BDEVNAME_SIZE];
1728
1729         freshest = NULL;
1730         ITERATE_RDEV(mddev,rdev,tmp)
1731                 switch (super_types[mddev->major_version].
1732                         load_super(rdev, freshest, mddev->minor_version)) {
1733                 case 1:
1734                         freshest = rdev;
1735                         break;
1736                 case 0:
1737                         break;
1738                 default:
1739                         printk( KERN_ERR \
1740                                 "md: fatal superblock inconsistency in %s"
1741                                 " -- removing from array\n",
1742                                 bdevname(rdev->bdev,b));
1743                         kick_rdev_from_array(rdev);
1744                 }
1745
1746
1747         super_types[mddev->major_version].
1748                 validate_super(mddev, freshest);
1749
1750         i = 0;
1751         ITERATE_RDEV(mddev,rdev,tmp) {
1752                 if (rdev != freshest)
1753                         if (super_types[mddev->major_version].
1754                             validate_super(mddev, rdev)) {
1755                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1756                                         " from array!\n",
1757                                         bdevname(rdev->bdev,b));
1758                                 kick_rdev_from_array(rdev);
1759                                 continue;
1760                         }
1761                 if (mddev->level == LEVEL_MULTIPATH) {
1762                         rdev->desc_nr = i++;
1763                         rdev->raid_disk = rdev->desc_nr;
1764                         set_bit(In_sync, &rdev->flags);
1765                 }
1766         }
1767
1768
1769
1770         if (mddev->recovery_cp != MaxSector &&
1771             mddev->level >= 1)
1772                 printk(KERN_ERR "md: %s: raid array is not clean"
1773                        " -- starting background reconstruction\n",
1774                        mdname(mddev));
1775
1776 }
1777
1778 static ssize_t
1779 level_show(mddev_t *mddev, char *page)
1780 {
1781         struct mdk_personality *p = mddev->pers;
1782         if (p)
1783                 return sprintf(page, "%s\n", p->name);
1784         else if (mddev->clevel[0])
1785                 return sprintf(page, "%s\n", mddev->clevel);
1786         else if (mddev->level != LEVEL_NONE)
1787                 return sprintf(page, "%d\n", mddev->level);
1788         else
1789                 return 0;
1790 }
1791
1792 static ssize_t
1793 level_store(mddev_t *mddev, const char *buf, size_t len)
1794 {
1795         int rv = len;
1796         if (mddev->pers)
1797                 return -EBUSY;
1798         if (len == 0)
1799                 return 0;
1800         if (len >= sizeof(mddev->clevel))
1801                 return -ENOSPC;
1802         strncpy(mddev->clevel, buf, len);
1803         if (mddev->clevel[len-1] == '\n')
1804                 len--;
1805         mddev->clevel[len] = 0;
1806         mddev->level = LEVEL_NONE;
1807         return rv;
1808 }
1809
1810 static struct md_sysfs_entry md_level =
1811 __ATTR(level, 0644, level_show, level_store);
1812
1813 static ssize_t
1814 raid_disks_show(mddev_t *mddev, char *page)
1815 {
1816         if (mddev->raid_disks == 0)
1817                 return 0;
1818         return sprintf(page, "%d\n", mddev->raid_disks);
1819 }
1820
1821 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1822
1823 static ssize_t
1824 chunk_size_show(mddev_t *mddev, char *page)
1825 {
1826         return sprintf(page, "%d\n", mddev->chunk_size);
1827 }
1828
1829 static ssize_t
1830 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1831 {
1832         /* can only set chunk_size if array is not yet active */
1833         char *e;
1834         unsigned long n = simple_strtoul(buf, &e, 10);
1835
1836         if (mddev->pers)
1837                 return -EBUSY;
1838         if (!*buf || (*e && *e != '\n'))
1839                 return -EINVAL;
1840
1841         mddev->chunk_size = n;
1842         return len;
1843 }
1844 static struct md_sysfs_entry md_chunk_size =
1845 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
1846
1847
1848 static ssize_t
1849 size_show(mddev_t *mddev, char *page)
1850 {
1851         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
1852 }
1853
1854 static int update_size(mddev_t *mddev, unsigned long size);
1855
1856 static ssize_t
1857 size_store(mddev_t *mddev, const char *buf, size_t len)
1858 {
1859         /* If array is inactive, we can reduce the component size, but
1860          * not increase it (except from 0).
1861          * If array is active, we can try an on-line resize
1862          */
1863         char *e;
1864         int err = 0;
1865         unsigned long long size = simple_strtoull(buf, &e, 10);
1866         if (!*buf || *buf == '\n' ||
1867             (*e && *e != '\n'))
1868                 return -EINVAL;
1869
1870         if (mddev->pers) {
1871                 err = update_size(mddev, size);
1872                 md_update_sb(mddev);
1873         } else {
1874                 if (mddev->size == 0 ||
1875                     mddev->size > size)
1876                         mddev->size = size;
1877                 else
1878                         err = -ENOSPC;
1879         }
1880         return err ? err : len;
1881 }
1882
1883 static struct md_sysfs_entry md_size =
1884 __ATTR(component_size, 0644, size_show, size_store);
1885
1886
1887 /* Metdata version.
1888  * This is either 'none' for arrays with externally managed metadata,
1889  * or N.M for internally known formats
1890  */
1891 static ssize_t
1892 metadata_show(mddev_t *mddev, char *page)
1893 {
1894         if (mddev->persistent)
1895                 return sprintf(page, "%d.%d\n",
1896                                mddev->major_version, mddev->minor_version);
1897         else
1898                 return sprintf(page, "none\n");
1899 }
1900
1901 static ssize_t
1902 metadata_store(mddev_t *mddev, const char *buf, size_t len)
1903 {
1904         int major, minor;
1905         char *e;
1906         if (!list_empty(&mddev->disks))
1907                 return -EBUSY;
1908
1909         if (cmd_match(buf, "none")) {
1910                 mddev->persistent = 0;
1911                 mddev->major_version = 0;
1912                 mddev->minor_version = 90;
1913                 return len;
1914         }
1915         major = simple_strtoul(buf, &e, 10);
1916         if (e==buf || *e != '.')
1917                 return -EINVAL;
1918         buf = e+1;
1919         minor = simple_strtoul(buf, &e, 10);
1920         if (e==buf || *e != '\n')
1921                 return -EINVAL;
1922         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
1923             super_types[major].name == NULL)
1924                 return -ENOENT;
1925         mddev->major_version = major;
1926         mddev->minor_version = minor;
1927         mddev->persistent = 1;
1928         return len;
1929 }
1930
1931 static struct md_sysfs_entry md_metadata =
1932 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
1933
1934 static ssize_t
1935 action_show(mddev_t *mddev, char *page)
1936 {
1937         char *type = "idle";
1938         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1939             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1940                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1941                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1942                                 type = "resync";
1943                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1944                                 type = "check";
1945                         else
1946                                 type = "repair";
1947                 } else
1948                         type = "recover";
1949         }
1950         return sprintf(page, "%s\n", type);
1951 }
1952
1953 static ssize_t
1954 action_store(mddev_t *mddev, const char *page, size_t len)
1955 {
1956         if (!mddev->pers || !mddev->pers->sync_request)
1957                 return -EINVAL;
1958
1959         if (cmd_match(page, "idle")) {
1960                 if (mddev->sync_thread) {
1961                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1962                         md_unregister_thread(mddev->sync_thread);
1963                         mddev->sync_thread = NULL;
1964                         mddev->recovery = 0;
1965                 }
1966         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1967                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1968                 return -EBUSY;
1969         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
1970                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1971         else {
1972                 if (cmd_match(page, "check"))
1973                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1974                 else if (cmd_match(page, "repair"))
1975                         return -EINVAL;
1976                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1977                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1978         }
1979         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1980         md_wakeup_thread(mddev->thread);
1981         return len;
1982 }
1983
1984 static ssize_t
1985 mismatch_cnt_show(mddev_t *mddev, char *page)
1986 {
1987         return sprintf(page, "%llu\n",
1988                        (unsigned long long) mddev->resync_mismatches);
1989 }
1990
1991 static struct md_sysfs_entry
1992 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1993
1994
1995 static struct md_sysfs_entry
1996 md_mismatches = __ATTR_RO(mismatch_cnt);
1997
1998 static struct attribute *md_default_attrs[] = {
1999         &md_level.attr,
2000         &md_raid_disks.attr,
2001         &md_chunk_size.attr,
2002         &md_size.attr,
2003         &md_metadata.attr,
2004         NULL,
2005 };
2006
2007 static struct attribute *md_redundancy_attrs[] = {
2008         &md_scan_mode.attr,
2009         &md_mismatches.attr,
2010         NULL,
2011 };
2012 static struct attribute_group md_redundancy_group = {
2013         .name = NULL,
2014         .attrs = md_redundancy_attrs,
2015 };
2016
2017
2018 static ssize_t
2019 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2020 {
2021         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2022         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2023         ssize_t rv;
2024
2025         if (!entry->show)
2026                 return -EIO;
2027         mddev_lock(mddev);
2028         rv = entry->show(mddev, page);
2029         mddev_unlock(mddev);
2030         return rv;
2031 }
2032
2033 static ssize_t
2034 md_attr_store(struct kobject *kobj, struct attribute *attr,
2035               const char *page, size_t length)
2036 {
2037         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2038         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2039         ssize_t rv;
2040
2041         if (!entry->store)
2042                 return -EIO;
2043         mddev_lock(mddev);
2044         rv = entry->store(mddev, page, length);
2045         mddev_unlock(mddev);
2046         return rv;
2047 }
2048
2049 static void md_free(struct kobject *ko)
2050 {
2051         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2052         kfree(mddev);
2053 }
2054
2055 static struct sysfs_ops md_sysfs_ops = {
2056         .show   = md_attr_show,
2057         .store  = md_attr_store,
2058 };
2059 static struct kobj_type md_ktype = {
2060         .release        = md_free,
2061         .sysfs_ops      = &md_sysfs_ops,
2062         .default_attrs  = md_default_attrs,
2063 };
2064
2065 int mdp_major = 0;
2066
2067 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2068 {
2069         static DECLARE_MUTEX(disks_sem);
2070         mddev_t *mddev = mddev_find(dev);
2071         struct gendisk *disk;
2072         int partitioned = (MAJOR(dev) != MD_MAJOR);
2073         int shift = partitioned ? MdpMinorShift : 0;
2074         int unit = MINOR(dev) >> shift;
2075
2076         if (!mddev)
2077                 return NULL;
2078
2079         down(&disks_sem);
2080         if (mddev->gendisk) {
2081                 up(&disks_sem);
2082                 mddev_put(mddev);
2083                 return NULL;
2084         }
2085         disk = alloc_disk(1 << shift);
2086         if (!disk) {
2087                 up(&disks_sem);
2088                 mddev_put(mddev);
2089                 return NULL;
2090         }
2091         disk->major = MAJOR(dev);
2092         disk->first_minor = unit << shift;
2093         if (partitioned) {
2094                 sprintf(disk->disk_name, "md_d%d", unit);
2095                 sprintf(disk->devfs_name, "md/d%d", unit);
2096         } else {
2097                 sprintf(disk->disk_name, "md%d", unit);
2098                 sprintf(disk->devfs_name, "md/%d", unit);
2099         }
2100         disk->fops = &md_fops;
2101         disk->private_data = mddev;
2102         disk->queue = mddev->queue;
2103         add_disk(disk);
2104         mddev->gendisk = disk;
2105         up(&disks_sem);
2106         mddev->kobj.parent = &disk->kobj;
2107         mddev->kobj.k_name = NULL;
2108         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2109         mddev->kobj.ktype = &md_ktype;
2110         kobject_register(&mddev->kobj);
2111         return NULL;
2112 }
2113
2114 void md_wakeup_thread(mdk_thread_t *thread);
2115
2116 static void md_safemode_timeout(unsigned long data)
2117 {
2118         mddev_t *mddev = (mddev_t *) data;
2119
2120         mddev->safemode = 1;
2121         md_wakeup_thread(mddev->thread);
2122 }
2123
2124 static int start_dirty_degraded;
2125
2126 static int do_md_run(mddev_t * mddev)
2127 {
2128         int err;
2129         int chunk_size;
2130         struct list_head *tmp;
2131         mdk_rdev_t *rdev;
2132         struct gendisk *disk;
2133         struct mdk_personality *pers;
2134         char b[BDEVNAME_SIZE];
2135
2136         if (list_empty(&mddev->disks))
2137                 /* cannot run an array with no devices.. */
2138                 return -EINVAL;
2139
2140         if (mddev->pers)
2141                 return -EBUSY;
2142
2143         /*
2144          * Analyze all RAID superblock(s)
2145          */
2146         if (!mddev->raid_disks)
2147                 analyze_sbs(mddev);
2148
2149         chunk_size = mddev->chunk_size;
2150
2151         if (chunk_size) {
2152                 if (chunk_size > MAX_CHUNK_SIZE) {
2153                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2154                                 chunk_size, MAX_CHUNK_SIZE);
2155                         return -EINVAL;
2156                 }
2157                 /*
2158                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2159                  */
2160                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2161                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2162                         return -EINVAL;
2163                 }
2164                 if (chunk_size < PAGE_SIZE) {
2165                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2166                                 chunk_size, PAGE_SIZE);
2167                         return -EINVAL;
2168                 }
2169
2170                 /* devices must have minimum size of one chunk */
2171                 ITERATE_RDEV(mddev,rdev,tmp) {
2172                         if (test_bit(Faulty, &rdev->flags))
2173                                 continue;
2174                         if (rdev->size < chunk_size / 1024) {
2175                                 printk(KERN_WARNING
2176                                         "md: Dev %s smaller than chunk_size:"
2177                                         " %lluk < %dk\n",
2178                                         bdevname(rdev->bdev,b),
2179                                         (unsigned long long)rdev->size,
2180                                         chunk_size / 1024);
2181                                 return -EINVAL;
2182                         }
2183                 }
2184         }
2185
2186 #ifdef CONFIG_KMOD
2187         if (mddev->level != LEVEL_NONE)
2188                 request_module("md-level-%d", mddev->level);
2189         else if (mddev->clevel[0])
2190                 request_module("md-%s", mddev->clevel);
2191 #endif
2192
2193         /*
2194          * Drop all container device buffers, from now on
2195          * the only valid external interface is through the md
2196          * device.
2197          * Also find largest hardsector size
2198          */
2199         ITERATE_RDEV(mddev,rdev,tmp) {
2200                 if (test_bit(Faulty, &rdev->flags))
2201                         continue;
2202                 sync_blockdev(rdev->bdev);
2203                 invalidate_bdev(rdev->bdev, 0);
2204         }
2205
2206         md_probe(mddev->unit, NULL, NULL);
2207         disk = mddev->gendisk;
2208         if (!disk)
2209                 return -ENOMEM;
2210
2211         spin_lock(&pers_lock);
2212         pers = find_pers(mddev->level, mddev->clevel);
2213         if (!pers || !try_module_get(pers->owner)) {
2214                 spin_unlock(&pers_lock);
2215                 if (mddev->level != LEVEL_NONE)
2216                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2217                                mddev->level);
2218                 else
2219                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2220                                mddev->clevel);
2221                 return -EINVAL;
2222         }
2223         mddev->pers = pers;
2224         spin_unlock(&pers_lock);
2225         mddev->level = pers->level;
2226         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2227
2228         mddev->recovery = 0;
2229         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2230         mddev->barriers_work = 1;
2231         mddev->ok_start_degraded = start_dirty_degraded;
2232
2233         if (start_readonly)
2234                 mddev->ro = 2; /* read-only, but switch on first write */
2235
2236         err = mddev->pers->run(mddev);
2237         if (!err && mddev->pers->sync_request) {
2238                 err = bitmap_create(mddev);
2239                 if (err) {
2240                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2241                                mdname(mddev), err);
2242                         mddev->pers->stop(mddev);
2243                 }
2244         }
2245         if (err) {
2246                 printk(KERN_ERR "md: pers->run() failed ...\n");
2247                 module_put(mddev->pers->owner);
2248                 mddev->pers = NULL;
2249                 bitmap_destroy(mddev);
2250                 return err;
2251         }
2252         if (mddev->pers->sync_request)
2253                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2254         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2255                 mddev->ro = 0;
2256
2257         atomic_set(&mddev->writes_pending,0);
2258         mddev->safemode = 0;
2259         mddev->safemode_timer.function = md_safemode_timeout;
2260         mddev->safemode_timer.data = (unsigned long) mddev;
2261         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2262         mddev->in_sync = 1;
2263
2264         ITERATE_RDEV(mddev,rdev,tmp)
2265                 if (rdev->raid_disk >= 0) {
2266                         char nm[20];
2267                         sprintf(nm, "rd%d", rdev->raid_disk);
2268                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2269                 }
2270
2271         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2272         md_wakeup_thread(mddev->thread);
2273
2274         if (mddev->sb_dirty)
2275                 md_update_sb(mddev);
2276
2277         set_capacity(disk, mddev->array_size<<1);
2278
2279         /* If we call blk_queue_make_request here, it will
2280          * re-initialise max_sectors etc which may have been
2281          * refined inside -> run.  So just set the bits we need to set.
2282          * Most initialisation happended when we called
2283          * blk_queue_make_request(..., md_fail_request)
2284          * earlier.
2285          */
2286         mddev->queue->queuedata = mddev;
2287         mddev->queue->make_request_fn = mddev->pers->make_request;
2288
2289         mddev->changed = 1;
2290         md_new_event(mddev);
2291         return 0;
2292 }
2293
2294 static int restart_array(mddev_t *mddev)
2295 {
2296         struct gendisk *disk = mddev->gendisk;
2297         int err;
2298
2299         /*
2300          * Complain if it has no devices
2301          */
2302         err = -ENXIO;
2303         if (list_empty(&mddev->disks))
2304                 goto out;
2305
2306         if (mddev->pers) {
2307                 err = -EBUSY;
2308                 if (!mddev->ro)
2309                         goto out;
2310
2311                 mddev->safemode = 0;
2312                 mddev->ro = 0;
2313                 set_disk_ro(disk, 0);
2314
2315                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2316                         mdname(mddev));
2317                 /*
2318                  * Kick recovery or resync if necessary
2319                  */
2320                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2321                 md_wakeup_thread(mddev->thread);
2322                 err = 0;
2323         } else {
2324                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2325                         mdname(mddev));
2326                 err = -EINVAL;
2327         }
2328
2329 out:
2330         return err;
2331 }
2332
2333 static int do_md_stop(mddev_t * mddev, int ro)
2334 {
2335         int err = 0;
2336         struct gendisk *disk = mddev->gendisk;
2337
2338         if (mddev->pers) {
2339                 if (atomic_read(&mddev->active)>2) {
2340                         printk("md: %s still in use.\n",mdname(mddev));
2341                         return -EBUSY;
2342                 }
2343
2344                 if (mddev->sync_thread) {
2345                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2346                         md_unregister_thread(mddev->sync_thread);
2347                         mddev->sync_thread = NULL;
2348                 }
2349
2350                 del_timer_sync(&mddev->safemode_timer);
2351
2352                 invalidate_partition(disk, 0);
2353
2354                 if (ro) {
2355                         err  = -ENXIO;
2356                         if (mddev->ro==1)
2357                                 goto out;
2358                         mddev->ro = 1;
2359                 } else {
2360                         bitmap_flush(mddev);
2361                         md_super_wait(mddev);
2362                         if (mddev->ro)
2363                                 set_disk_ro(disk, 0);
2364                         blk_queue_make_request(mddev->queue, md_fail_request);
2365                         mddev->pers->stop(mddev);
2366                         if (mddev->pers->sync_request)
2367                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2368
2369                         module_put(mddev->pers->owner);
2370                         mddev->pers = NULL;
2371                         if (mddev->ro)
2372                                 mddev->ro = 0;
2373                 }
2374                 if (!mddev->in_sync) {
2375                         /* mark array as shutdown cleanly */
2376                         mddev->in_sync = 1;
2377                         md_update_sb(mddev);
2378                 }
2379                 if (ro)
2380                         set_disk_ro(disk, 1);
2381         }
2382
2383         bitmap_destroy(mddev);
2384         if (mddev->bitmap_file) {
2385                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2386                 fput(mddev->bitmap_file);
2387                 mddev->bitmap_file = NULL;
2388         }
2389         mddev->bitmap_offset = 0;
2390
2391         /*
2392          * Free resources if final stop
2393          */
2394         if (!ro) {
2395                 mdk_rdev_t *rdev;
2396                 struct list_head *tmp;
2397                 struct gendisk *disk;
2398                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2399
2400                 ITERATE_RDEV(mddev,rdev,tmp)
2401                         if (rdev->raid_disk >= 0) {
2402                                 char nm[20];
2403                                 sprintf(nm, "rd%d", rdev->raid_disk);
2404                                 sysfs_remove_link(&mddev->kobj, nm);
2405                         }
2406
2407                 export_array(mddev);
2408
2409                 mddev->array_size = 0;
2410                 disk = mddev->gendisk;
2411                 if (disk)
2412                         set_capacity(disk, 0);
2413                 mddev->changed = 1;
2414         } else
2415                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2416                         mdname(mddev));
2417         err = 0;
2418         md_new_event(mddev);
2419 out:
2420         return err;
2421 }
2422
2423 static void autorun_array(mddev_t *mddev)
2424 {
2425         mdk_rdev_t *rdev;
2426         struct list_head *tmp;
2427         int err;
2428
2429         if (list_empty(&mddev->disks))
2430                 return;
2431
2432         printk(KERN_INFO "md: running: ");
2433
2434         ITERATE_RDEV(mddev,rdev,tmp) {
2435                 char b[BDEVNAME_SIZE];
2436                 printk("<%s>", bdevname(rdev->bdev,b));
2437         }
2438         printk("\n");
2439
2440         err = do_md_run (mddev);
2441         if (err) {
2442                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2443                 do_md_stop (mddev, 0);
2444         }
2445 }
2446
2447 /*
2448  * lets try to run arrays based on all disks that have arrived
2449  * until now. (those are in pending_raid_disks)
2450  *
2451  * the method: pick the first pending disk, collect all disks with
2452  * the same UUID, remove all from the pending list and put them into
2453  * the 'same_array' list. Then order this list based on superblock
2454  * update time (freshest comes first), kick out 'old' disks and
2455  * compare superblocks. If everything's fine then run it.
2456  *
2457  * If "unit" is allocated, then bump its reference count
2458  */
2459 static void autorun_devices(int part)
2460 {
2461         struct list_head candidates;
2462         struct list_head *tmp;
2463         mdk_rdev_t *rdev0, *rdev;
2464         mddev_t *mddev;
2465         char b[BDEVNAME_SIZE];
2466
2467         printk(KERN_INFO "md: autorun ...\n");
2468         while (!list_empty(&pending_raid_disks)) {
2469                 dev_t dev;
2470                 rdev0 = list_entry(pending_raid_disks.next,
2471                                          mdk_rdev_t, same_set);
2472
2473                 printk(KERN_INFO "md: considering %s ...\n",
2474                         bdevname(rdev0->bdev,b));
2475                 INIT_LIST_HEAD(&candidates);
2476                 ITERATE_RDEV_PENDING(rdev,tmp)
2477                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2478                                 printk(KERN_INFO "md:  adding %s ...\n",
2479                                         bdevname(rdev->bdev,b));
2480                                 list_move(&rdev->same_set, &candidates);
2481                         }
2482                 /*
2483                  * now we have a set of devices, with all of them having
2484                  * mostly sane superblocks. It's time to allocate the
2485                  * mddev.
2486                  */
2487                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2488                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2489                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2490                         break;
2491                 }
2492                 if (part)
2493                         dev = MKDEV(mdp_major,
2494                                     rdev0->preferred_minor << MdpMinorShift);
2495                 else
2496                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2497
2498                 md_probe(dev, NULL, NULL);
2499                 mddev = mddev_find(dev);
2500                 if (!mddev) {
2501                         printk(KERN_ERR
2502                                 "md: cannot allocate memory for md drive.\n");
2503                         break;
2504                 }
2505                 if (mddev_lock(mddev))
2506                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2507                                mdname(mddev));
2508                 else if (mddev->raid_disks || mddev->major_version
2509                          || !list_empty(&mddev->disks)) {
2510                         printk(KERN_WARNING
2511                                 "md: %s already running, cannot run %s\n",
2512                                 mdname(mddev), bdevname(rdev0->bdev,b));
2513                         mddev_unlock(mddev);
2514                 } else {
2515                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2516                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2517                                 list_del_init(&rdev->same_set);
2518                                 if (bind_rdev_to_array(rdev, mddev))
2519                                         export_rdev(rdev);
2520                         }
2521                         autorun_array(mddev);
2522                         mddev_unlock(mddev);
2523                 }
2524                 /* on success, candidates will be empty, on error
2525                  * it won't...
2526                  */
2527                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2528                         export_rdev(rdev);
2529                 mddev_put(mddev);
2530         }
2531         printk(KERN_INFO "md: ... autorun DONE.\n");
2532 }
2533
2534 /*
2535  * import RAID devices based on one partition
2536  * if possible, the array gets run as well.
2537  */
2538
2539 static int autostart_array(dev_t startdev)
2540 {
2541         char b[BDEVNAME_SIZE];
2542         int err = -EINVAL, i;
2543         mdp_super_t *sb = NULL;
2544         mdk_rdev_t *start_rdev = NULL, *rdev;
2545
2546         start_rdev = md_import_device(startdev, 0, 0);
2547         if (IS_ERR(start_rdev))
2548                 return err;
2549
2550
2551         /* NOTE: this can only work for 0.90.0 superblocks */
2552         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2553         if (sb->major_version != 0 ||
2554             sb->minor_version != 90 ) {
2555                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2556                 export_rdev(start_rdev);
2557                 return err;
2558         }
2559
2560         if (test_bit(Faulty, &start_rdev->flags)) {
2561                 printk(KERN_WARNING
2562                         "md: can not autostart based on faulty %s!\n",
2563                         bdevname(start_rdev->bdev,b));
2564                 export_rdev(start_rdev);
2565                 return err;
2566         }
2567         list_add(&start_rdev->same_set, &pending_raid_disks);
2568
2569         for (i = 0; i < MD_SB_DISKS; i++) {
2570                 mdp_disk_t *desc = sb->disks + i;
2571                 dev_t dev = MKDEV(desc->major, desc->minor);
2572
2573                 if (!dev)
2574                         continue;
2575                 if (dev == startdev)
2576                         continue;
2577                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2578                         continue;
2579                 rdev = md_import_device(dev, 0, 0);
2580                 if (IS_ERR(rdev))
2581                         continue;
2582
2583                 list_add(&rdev->same_set, &pending_raid_disks);
2584         }
2585
2586         /*
2587          * possibly return codes
2588          */
2589         autorun_devices(0);
2590         return 0;
2591
2592 }
2593
2594
2595 static int get_version(void __user * arg)
2596 {
2597         mdu_version_t ver;
2598
2599         ver.major = MD_MAJOR_VERSION;
2600         ver.minor = MD_MINOR_VERSION;
2601         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2602
2603         if (copy_to_user(arg, &ver, sizeof(ver)))
2604                 return -EFAULT;
2605
2606         return 0;
2607 }
2608
2609 static int get_array_info(mddev_t * mddev, void __user * arg)
2610 {
2611         mdu_array_info_t info;
2612         int nr,working,active,failed,spare;
2613         mdk_rdev_t *rdev;
2614         struct list_head *tmp;
2615
2616         nr=working=active=failed=spare=0;
2617         ITERATE_RDEV(mddev,rdev,tmp) {
2618                 nr++;
2619                 if (test_bit(Faulty, &rdev->flags))
2620                         failed++;
2621                 else {
2622                         working++;
2623                         if (test_bit(In_sync, &rdev->flags))
2624                                 active++;
2625                         else
2626                                 spare++;
2627                 }
2628         }
2629
2630         info.major_version = mddev->major_version;
2631         info.minor_version = mddev->minor_version;
2632         info.patch_version = MD_PATCHLEVEL_VERSION;
2633         info.ctime         = mddev->ctime;
2634         info.level         = mddev->level;
2635         info.size          = mddev->size;
2636         info.nr_disks      = nr;
2637         info.raid_disks    = mddev->raid_disks;
2638         info.md_minor      = mddev->md_minor;
2639         info.not_persistent= !mddev->persistent;
2640
2641         info.utime         = mddev->utime;
2642         info.state         = 0;
2643         if (mddev->in_sync)
2644                 info.state = (1<<MD_SB_CLEAN);
2645         if (mddev->bitmap && mddev->bitmap_offset)
2646                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2647         info.active_disks  = active;
2648         info.working_disks = working;
2649         info.failed_disks  = failed;
2650         info.spare_disks   = spare;
2651
2652         info.layout        = mddev->layout;
2653         info.chunk_size    = mddev->chunk_size;
2654
2655         if (copy_to_user(arg, &info, sizeof(info)))
2656                 return -EFAULT;
2657
2658         return 0;
2659 }
2660
2661 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2662 {
2663         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2664         char *ptr, *buf = NULL;
2665         int err = -ENOMEM;
2666
2667         file = kmalloc(sizeof(*file), GFP_KERNEL);
2668         if (!file)
2669                 goto out;
2670
2671         /* bitmap disabled, zero the first byte and copy out */
2672         if (!mddev->bitmap || !mddev->bitmap->file) {
2673                 file->pathname[0] = '\0';
2674                 goto copy_out;
2675         }
2676
2677         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2678         if (!buf)
2679                 goto out;
2680
2681         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2682         if (!ptr)
2683                 goto out;
2684
2685         strcpy(file->pathname, ptr);
2686
2687 copy_out:
2688         err = 0;
2689         if (copy_to_user(arg, file, sizeof(*file)))
2690                 err = -EFAULT;
2691 out:
2692         kfree(buf);
2693         kfree(file);
2694         return err;
2695 }
2696
2697 static int get_disk_info(mddev_t * mddev, void __user * arg)
2698 {
2699         mdu_disk_info_t info;
2700         unsigned int nr;
2701         mdk_rdev_t *rdev;
2702
2703         if (copy_from_user(&info, arg, sizeof(info)))
2704                 return -EFAULT;
2705
2706         nr = info.number;
2707
2708         rdev = find_rdev_nr(mddev, nr);
2709         if (rdev) {
2710                 info.major = MAJOR(rdev->bdev->bd_dev);
2711                 info.minor = MINOR(rdev->bdev->bd_dev);
2712                 info.raid_disk = rdev->raid_disk;
2713                 info.state = 0;
2714                 if (test_bit(Faulty, &rdev->flags))
2715                         info.state |= (1<<MD_DISK_FAULTY);
2716                 else if (test_bit(In_sync, &rdev->flags)) {
2717                         info.state |= (1<<MD_DISK_ACTIVE);
2718                         info.state |= (1<<MD_DISK_SYNC);
2719                 }
2720                 if (test_bit(WriteMostly, &rdev->flags))
2721                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2722         } else {
2723                 info.major = info.minor = 0;
2724                 info.raid_disk = -1;
2725                 info.state = (1<<MD_DISK_REMOVED);
2726         }
2727
2728         if (copy_to_user(arg, &info, sizeof(info)))
2729                 return -EFAULT;
2730
2731         return 0;
2732 }
2733
2734 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2735 {
2736         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2737         mdk_rdev_t *rdev;
2738         dev_t dev = MKDEV(info->major,info->minor);
2739
2740         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2741                 return -EOVERFLOW;
2742
2743         if (!mddev->raid_disks) {
2744                 int err;
2745                 /* expecting a device which has a superblock */
2746                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2747                 if (IS_ERR(rdev)) {
2748                         printk(KERN_WARNING
2749                                 "md: md_import_device returned %ld\n",
2750                                 PTR_ERR(rdev));
2751                         return PTR_ERR(rdev);
2752                 }
2753                 if (!list_empty(&mddev->disks)) {
2754                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2755                                                         mdk_rdev_t, same_set);
2756                         int err = super_types[mddev->major_version]
2757                                 .load_super(rdev, rdev0, mddev->minor_version);
2758                         if (err < 0) {
2759                                 printk(KERN_WARNING
2760                                         "md: %s has different UUID to %s\n",
2761                                         bdevname(rdev->bdev,b),
2762                                         bdevname(rdev0->bdev,b2));
2763                                 export_rdev(rdev);
2764                                 return -EINVAL;
2765                         }
2766                 }
2767                 err = bind_rdev_to_array(rdev, mddev);
2768                 if (err)
2769                         export_rdev(rdev);
2770                 return err;
2771         }
2772
2773         /*
2774          * add_new_disk can be used once the array is assembled
2775          * to add "hot spares".  They must already have a superblock
2776          * written
2777          */
2778         if (mddev->pers) {
2779                 int err;
2780                 if (!mddev->pers->hot_add_disk) {
2781                         printk(KERN_WARNING
2782                                 "%s: personality does not support diskops!\n",
2783                                mdname(mddev));
2784                         return -EINVAL;
2785                 }
2786                 if (mddev->persistent)
2787                         rdev = md_import_device(dev, mddev->major_version,
2788                                                 mddev->minor_version);
2789                 else
2790                         rdev = md_import_device(dev, -1, -1);
2791                 if (IS_ERR(rdev)) {
2792                         printk(KERN_WARNING
2793                                 "md: md_import_device returned %ld\n",
2794                                 PTR_ERR(rdev));
2795                         return PTR_ERR(rdev);
2796                 }
2797                 /* set save_raid_disk if appropriate */
2798                 if (!mddev->persistent) {
2799                         if (info->state & (1<<MD_DISK_SYNC)  &&
2800                             info->raid_disk < mddev->raid_disks)
2801                                 rdev->raid_disk = info->raid_disk;
2802                         else
2803                                 rdev->raid_disk = -1;
2804                 } else
2805                         super_types[mddev->major_version].
2806                                 validate_super(mddev, rdev);
2807                 rdev->saved_raid_disk = rdev->raid_disk;
2808
2809                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2810                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2811                         set_bit(WriteMostly, &rdev->flags);
2812
2813                 rdev->raid_disk = -1;
2814                 err = bind_rdev_to_array(rdev, mddev);
2815                 if (err)
2816                         export_rdev(rdev);
2817
2818                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2819                 md_wakeup_thread(mddev->thread);
2820                 return err;
2821         }
2822
2823         /* otherwise, add_new_disk is only allowed
2824          * for major_version==0 superblocks
2825          */
2826         if (mddev->major_version != 0) {
2827                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2828                        mdname(mddev));
2829                 return -EINVAL;
2830         }
2831
2832         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2833                 int err;
2834                 rdev = md_import_device (dev, -1, 0);
2835                 if (IS_ERR(rdev)) {
2836                         printk(KERN_WARNING
2837                                 "md: error, md_import_device() returned %ld\n",
2838                                 PTR_ERR(rdev));
2839                         return PTR_ERR(rdev);
2840                 }
2841                 rdev->desc_nr = info->number;
2842                 if (info->raid_disk < mddev->raid_disks)
2843                         rdev->raid_disk = info->raid_disk;
2844                 else
2845                         rdev->raid_disk = -1;
2846
2847                 rdev->flags = 0;
2848
2849                 if (rdev->raid_disk < mddev->raid_disks)
2850                         if (info->state & (1<<MD_DISK_SYNC))
2851                                 set_bit(In_sync, &rdev->flags);
2852
2853                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2854                         set_bit(WriteMostly, &rdev->flags);
2855
2856                 err = bind_rdev_to_array(rdev, mddev);
2857                 if (err) {
2858                         export_rdev(rdev);
2859                         return err;
2860                 }
2861
2862                 if (!mddev->persistent) {
2863                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2864                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2865                 } else
2866                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2867                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2868
2869                 if (!mddev->size || (mddev->size > rdev->size))
2870                         mddev->size = rdev->size;
2871         }
2872
2873         return 0;
2874 }
2875
2876 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2877 {
2878         char b[BDEVNAME_SIZE];
2879         mdk_rdev_t *rdev;
2880
2881         if (!mddev->pers)
2882                 return -ENODEV;
2883
2884         rdev = find_rdev(mddev, dev);
2885         if (!rdev)
2886                 return -ENXIO;
2887
2888         if (rdev->raid_disk >= 0)
2889                 goto busy;
2890
2891         kick_rdev_from_array(rdev);
2892         md_update_sb(mddev);
2893         md_new_event(mddev);
2894
2895         return 0;
2896 busy:
2897         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2898                 bdevname(rdev->bdev,b), mdname(mddev));
2899         return -EBUSY;
2900 }
2901
2902 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2903 {
2904         char b[BDEVNAME_SIZE];
2905         int err;
2906         unsigned int size;
2907         mdk_rdev_t *rdev;
2908
2909         if (!mddev->pers)
2910                 return -ENODEV;
2911
2912         if (mddev->major_version != 0) {
2913                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2914                         " version-0 superblocks.\n",
2915                         mdname(mddev));
2916                 return -EINVAL;
2917         }
2918         if (!mddev->pers->hot_add_disk) {
2919                 printk(KERN_WARNING
2920                         "%s: personality does not support diskops!\n",
2921                         mdname(mddev));
2922                 return -EINVAL;
2923         }
2924
2925         rdev = md_import_device (dev, -1, 0);
2926         if (IS_ERR(rdev)) {
2927                 printk(KERN_WARNING
2928                         "md: error, md_import_device() returned %ld\n",
2929                         PTR_ERR(rdev));
2930                 return -EINVAL;
2931         }
2932
2933         if (mddev->persistent)
2934                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2935         else
2936                 rdev->sb_offset =
2937                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2938
2939         size = calc_dev_size(rdev, mddev->chunk_size);
2940         rdev->size = size;
2941
2942         if (size < mddev->size) {
2943                 printk(KERN_WARNING
2944                         "%s: disk size %llu blocks < array size %llu\n",
2945                         mdname(mddev), (unsigned long long)size,
2946                         (unsigned long long)mddev->size);
2947                 err = -ENOSPC;
2948                 goto abort_export;
2949         }
2950
2951         if (test_bit(Faulty, &rdev->flags)) {
2952                 printk(KERN_WARNING
2953                         "md: can not hot-add faulty %s disk to %s!\n",
2954                         bdevname(rdev->bdev,b), mdname(mddev));
2955                 err = -EINVAL;
2956                 goto abort_export;
2957         }
2958         clear_bit(In_sync, &rdev->flags);
2959         rdev->desc_nr = -1;
2960         bind_rdev_to_array(rdev, mddev);
2961
2962         /*
2963          * The rest should better be atomic, we can have disk failures
2964          * noticed in interrupt contexts ...
2965          */
2966
2967         if (rdev->desc_nr == mddev->max_disks) {
2968                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2969                         mdname(mddev));
2970                 err = -EBUSY;
2971                 goto abort_unbind_export;
2972         }
2973
2974         rdev->raid_disk = -1;
2975
2976         md_update_sb(mddev);
2977
2978         /*
2979          * Kick recovery, maybe this spare has to be added to the
2980          * array immediately.
2981          */
2982         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2983         md_wakeup_thread(mddev->thread);
2984         md_new_event(mddev);
2985         return 0;
2986
2987 abort_unbind_export:
2988         unbind_rdev_from_array(rdev);
2989
2990 abort_export:
2991         export_rdev(rdev);
2992         return err;
2993 }
2994
2995 /* similar to deny_write_access, but accounts for our holding a reference
2996  * to the file ourselves */
2997 static int deny_bitmap_write_access(struct file * file)
2998 {
2999         struct inode *inode = file->f_mapping->host;
3000
3001         spin_lock(&inode->i_lock);
3002         if (atomic_read(&inode->i_writecount) > 1) {
3003                 spin_unlock(&inode->i_lock);
3004                 return -ETXTBSY;
3005         }
3006         atomic_set(&inode->i_writecount, -1);
3007         spin_unlock(&inode->i_lock);
3008
3009         return 0;
3010 }
3011
3012 static int set_bitmap_file(mddev_t *mddev, int fd)
3013 {
3014         int err;
3015
3016         if (mddev->pers) {
3017                 if (!mddev->pers->quiesce)
3018                         return -EBUSY;
3019                 if (mddev->recovery || mddev->sync_thread)
3020                         return -EBUSY;
3021                 /* we should be able to change the bitmap.. */
3022         }
3023
3024
3025         if (fd >= 0) {
3026                 if (mddev->bitmap)
3027                         return -EEXIST; /* cannot add when bitmap is present */
3028                 mddev->bitmap_file = fget(fd);
3029
3030                 if (mddev->bitmap_file == NULL) {
3031                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3032                                mdname(mddev));
3033                         return -EBADF;
3034                 }
3035
3036                 err = deny_bitmap_write_access(mddev->bitmap_file);
3037                 if (err) {
3038                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3039                                mdname(mddev));
3040                         fput(mddev->bitmap_file);
3041                         mddev->bitmap_file = NULL;
3042                         return err;
3043                 }
3044                 mddev->bitmap_offset = 0; /* file overrides offset */
3045         } else if (mddev->bitmap == NULL)
3046                 return -ENOENT; /* cannot remove what isn't there */
3047         err = 0;
3048         if (mddev->pers) {
3049                 mddev->pers->quiesce(mddev, 1);
3050                 if (fd >= 0)
3051                         err = bitmap_create(mddev);
3052                 if (fd < 0 || err)
3053                         bitmap_destroy(mddev);
3054                 mddev->pers->quiesce(mddev, 0);
3055         } else if (fd < 0) {
3056                 if (mddev->bitmap_file)
3057                         fput(mddev->bitmap_file);
3058                 mddev->bitmap_file = NULL;
3059         }
3060
3061         return err;
3062 }
3063
3064 /*
3065  * set_array_info is used two different ways
3066  * The original usage is when creating a new array.
3067  * In this usage, raid_disks is > 0 and it together with
3068  *  level, size, not_persistent,layout,chunksize determine the
3069  *  shape of the array.
3070  *  This will always create an array with a type-0.90.0 superblock.
3071  * The newer usage is when assembling an array.
3072  *  In this case raid_disks will be 0, and the major_version field is
3073  *  use to determine which style super-blocks are to be found on the devices.
3074  *  The minor and patch _version numbers are also kept incase the
3075  *  super_block handler wishes to interpret them.
3076  */
3077 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3078 {
3079
3080         if (info->raid_disks == 0) {
3081                 /* just setting version number for superblock loading */
3082                 if (info->major_version < 0 ||
3083                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3084                     super_types[info->major_version].name == NULL) {
3085                         /* maybe try to auto-load a module? */
3086                         printk(KERN_INFO
3087                                 "md: superblock version %d not known\n",
3088                                 info->major_version);
3089                         return -EINVAL;
3090                 }
3091                 mddev->major_version = info->major_version;
3092                 mddev->minor_version = info->minor_version;
3093                 mddev->patch_version = info->patch_version;
3094                 return 0;
3095         }
3096         mddev->major_version = MD_MAJOR_VERSION;
3097         mddev->minor_version = MD_MINOR_VERSION;
3098         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3099         mddev->ctime         = get_seconds();
3100
3101         mddev->level         = info->level;
3102         mddev->size          = info->size;
3103         mddev->raid_disks    = info->raid_disks;
3104         /* don't set md_minor, it is determined by which /dev/md* was
3105          * openned
3106          */
3107         if (info->state & (1<<MD_SB_CLEAN))
3108                 mddev->recovery_cp = MaxSector;
3109         else
3110                 mddev->recovery_cp = 0;
3111         mddev->persistent    = ! info->not_persistent;
3112
3113         mddev->layout        = info->layout;
3114         mddev->chunk_size    = info->chunk_size;
3115
3116         mddev->max_disks     = MD_SB_DISKS;
3117
3118         mddev->sb_dirty      = 1;
3119
3120         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3121         mddev->bitmap_offset = 0;
3122
3123         /*
3124          * Generate a 128 bit UUID
3125          */
3126         get_random_bytes(mddev->uuid, 16);
3127
3128         return 0;
3129 }
3130
3131 static int update_size(mddev_t *mddev, unsigned long size)
3132 {
3133         mdk_rdev_t * rdev;
3134         int rv;
3135         struct list_head *tmp;
3136
3137         if (mddev->pers->resize == NULL)
3138                 return -EINVAL;
3139         /* The "size" is the amount of each device that is used.
3140          * This can only make sense for arrays with redundancy.
3141          * linear and raid0 always use whatever space is available
3142          * We can only consider changing the size if no resync
3143          * or reconstruction is happening, and if the new size
3144          * is acceptable. It must fit before the sb_offset or,
3145          * if that is <data_offset, it must fit before the
3146          * size of each device.
3147          * If size is zero, we find the largest size that fits.
3148          */
3149         if (mddev->sync_thread)
3150                 return -EBUSY;
3151         ITERATE_RDEV(mddev,rdev,tmp) {
3152                 sector_t avail;
3153                 int fit = (size == 0);
3154                 if (rdev->sb_offset > rdev->data_offset)
3155                         avail = (rdev->sb_offset*2) - rdev->data_offset;
3156                 else
3157                         avail = get_capacity(rdev->bdev->bd_disk)
3158                                 - rdev->data_offset;
3159                 if (fit && (size == 0 || size > avail/2))
3160                         size = avail/2;
3161                 if (avail < ((sector_t)size << 1))
3162                         return -ENOSPC;
3163         }
3164         rv = mddev->pers->resize(mddev, (sector_t)size *2);
3165         if (!rv) {
3166                 struct block_device *bdev;
3167
3168                 bdev = bdget_disk(mddev->gendisk, 0);
3169                 if (bdev) {
3170                         down(&bdev->bd_inode->i_sem);
3171                         i_size_write(bdev->bd_inode, mddev->array_size << 10);
3172                         up(&bdev->bd_inode->i_sem);
3173                         bdput(bdev);
3174                 }
3175         }
3176         return rv;
3177 }
3178
3179 /*
3180  * update_array_info is used to change the configuration of an
3181  * on-line array.
3182  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3183  * fields in the info are checked against the array.
3184  * Any differences that cannot be handled will cause an error.
3185  * Normally, only one change can be managed at a time.
3186  */
3187 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3188 {
3189         int rv = 0;
3190         int cnt = 0;
3191         int state = 0;
3192
3193         /* calculate expected state,ignoring low bits */
3194         if (mddev->bitmap && mddev->bitmap_offset)
3195                 state |= (1 << MD_SB_BITMAP_PRESENT);
3196
3197         if (mddev->major_version != info->major_version ||
3198             mddev->minor_version != info->minor_version ||
3199 /*          mddev->patch_version != info->patch_version || */
3200             mddev->ctime         != info->ctime         ||
3201             mddev->level         != info->level         ||
3202 /*          mddev->layout        != info->layout        || */
3203             !mddev->persistent   != info->not_persistent||
3204             mddev->chunk_size    != info->chunk_size    ||
3205             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3206             ((state^info->state) & 0xfffffe00)
3207                 )
3208                 return -EINVAL;
3209         /* Check there is only one change */
3210         if (mddev->size != info->size) cnt++;
3211         if (mddev->raid_disks != info->raid_disks) cnt++;
3212         if (mddev->layout != info->layout) cnt++;
3213         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3214         if (cnt == 0) return 0;
3215         if (cnt > 1) return -EINVAL;
3216
3217         if (mddev->layout != info->layout) {
3218                 /* Change layout
3219                  * we don't need to do anything at the md level, the
3220                  * personality will take care of it all.
3221                  */
3222                 if (mddev->pers->reconfig == NULL)
3223                         return -EINVAL;
3224                 else
3225                         return mddev->pers->reconfig(mddev, info->layout, -1);
3226         }
3227         if (mddev->size != info->size)
3228                 rv = update_size(mddev, info->size);
3229
3230         if (mddev->raid_disks    != info->raid_disks) {
3231                 /* change the number of raid disks */
3232                 if (mddev->pers->reshape == NULL)
3233                         return -EINVAL;
3234                 if (info->raid_disks <= 0 ||
3235                     info->raid_disks >= mddev->max_disks)
3236                         return -EINVAL;
3237                 if (mddev->sync_thread)
3238                         return -EBUSY;
3239                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3240                 if (!rv) {
3241                         struct block_device *bdev;
3242
3243                         bdev = bdget_disk(mddev->gendisk, 0);
3244                         if (bdev) {
3245                                 down(&bdev->bd_inode->i_sem);
3246                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3247                                 up(&bdev->bd_inode->i_sem);
3248                                 bdput(bdev);
3249                         }
3250                 }
3251         }
3252         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3253                 if (mddev->pers->quiesce == NULL)
3254                         return -EINVAL;
3255                 if (mddev->recovery || mddev->sync_thread)
3256                         return -EBUSY;
3257                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3258                         /* add the bitmap */
3259                         if (mddev->bitmap)
3260                                 return -EEXIST;
3261                         if (mddev->default_bitmap_offset == 0)
3262                                 return -EINVAL;
3263                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3264                         mddev->pers->quiesce(mddev, 1);
3265                         rv = bitmap_create(mddev);
3266                         if (rv)
3267                                 bitmap_destroy(mddev);
3268                         mddev->pers->quiesce(mddev, 0);
3269                 } else {
3270                         /* remove the bitmap */
3271                         if (!mddev->bitmap)
3272                                 return -ENOENT;
3273                         if (mddev->bitmap->file)
3274                                 return -EINVAL;
3275                         mddev->pers->quiesce(mddev, 1);
3276                         bitmap_destroy(mddev);
3277                         mddev->pers->quiesce(mddev, 0);
3278                         mddev->bitmap_offset = 0;
3279                 }
3280         }
3281         md_update_sb(mddev);
3282         return rv;
3283 }
3284
3285 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3286 {
3287         mdk_rdev_t *rdev;
3288
3289         if (mddev->pers == NULL)
3290                 return -ENODEV;
3291
3292         rdev = find_rdev(mddev, dev);
3293         if (!rdev)
3294                 return -ENODEV;
3295
3296         md_error(mddev, rdev);
3297         return 0;
3298 }
3299
3300 static int md_ioctl(struct inode *inode, struct file *file,
3301                         unsigned int cmd, unsigned long arg)
3302 {
3303         int err = 0;
3304         void __user *argp = (void __user *)arg;
3305         struct hd_geometry __user *loc = argp;
3306         mddev_t *mddev = NULL;
3307
3308         if (!capable(CAP_SYS_ADMIN))
3309                 return -EACCES;
3310
3311         /*
3312          * Commands dealing with the RAID driver but not any
3313          * particular array:
3314          */
3315         switch (cmd)
3316         {
3317                 case RAID_VERSION:
3318                         err = get_version(argp);
3319                         goto done;
3320
3321                 case PRINT_RAID_DEBUG:
3322                         err = 0;
3323                         md_print_devices();
3324                         goto done;
3325
3326 #ifndef MODULE
3327                 case RAID_AUTORUN:
3328                         err = 0;
3329                         autostart_arrays(arg);
3330                         goto done;
3331 #endif
3332                 default:;
3333         }
3334
3335         /*
3336          * Commands creating/starting a new array:
3337          */
3338
3339         mddev = inode->i_bdev->bd_disk->private_data;
3340
3341         if (!mddev) {
3342                 BUG();
3343                 goto abort;
3344         }
3345
3346
3347         if (cmd == START_ARRAY) {
3348                 /* START_ARRAY doesn't need to lock the array as autostart_array
3349                  * does the locking, and it could even be a different array
3350                  */
3351                 static int cnt = 3;
3352                 if (cnt > 0 ) {
3353                         printk(KERN_WARNING
3354                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3355                                "This will not be supported beyond July 2006\n",
3356                                current->comm, current->pid);
3357                         cnt--;
3358                 }
3359                 err = autostart_array(new_decode_dev(arg));
3360                 if (err) {
3361                         printk(KERN_WARNING "md: autostart failed!\n");
3362                         goto abort;
3363                 }
3364                 goto done;
3365         }
3366
3367         err = mddev_lock(mddev);
3368         if (err) {
3369                 printk(KERN_INFO
3370                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3371                         err, cmd);
3372                 goto abort;
3373         }
3374
3375         switch (cmd)
3376         {
3377                 case SET_ARRAY_INFO:
3378                         {
3379                                 mdu_array_info_t info;
3380                                 if (!arg)
3381                                         memset(&info, 0, sizeof(info));
3382                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3383                                         err = -EFAULT;
3384                                         goto abort_unlock;
3385                                 }
3386                                 if (mddev->pers) {
3387                                         err = update_array_info(mddev, &info);
3388                                         if (err) {
3389                                                 printk(KERN_WARNING "md: couldn't update"
3390                                                        " array info. %d\n", err);
3391                                                 goto abort_unlock;
3392                                         }
3393                                         goto done_unlock;
3394                                 }
3395                                 if (!list_empty(&mddev->disks)) {
3396                                         printk(KERN_WARNING
3397                                                "md: array %s already has disks!\n",
3398                                                mdname(mddev));
3399                                         err = -EBUSY;
3400                                         goto abort_unlock;
3401                                 }
3402                                 if (mddev->raid_disks) {
3403                                         printk(KERN_WARNING
3404                                                "md: array %s already initialised!\n",
3405                                                mdname(mddev));
3406                                         err = -EBUSY;
3407                                         goto abort_unlock;
3408                                 }
3409                                 err = set_array_info(mddev, &info);
3410                                 if (err) {
3411                                         printk(KERN_WARNING "md: couldn't set"
3412                                                " array info. %d\n", err);
3413                                         goto abort_unlock;
3414                                 }
3415                         }
3416                         goto done_unlock;
3417
3418                 default:;
3419         }
3420
3421         /*
3422          * Commands querying/configuring an existing array:
3423          */
3424         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3425          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3426         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3427                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3428                 err = -ENODEV;
3429                 goto abort_unlock;
3430         }
3431
3432         /*
3433          * Commands even a read-only array can execute:
3434          */
3435         switch (cmd)
3436         {
3437                 case GET_ARRAY_INFO:
3438                         err = get_array_info(mddev, argp);
3439                         goto done_unlock;
3440
3441                 case GET_BITMAP_FILE:
3442                         err = get_bitmap_file(mddev, argp);
3443                         goto done_unlock;
3444
3445                 case GET_DISK_INFO:
3446                         err = get_disk_info(mddev, argp);
3447                         goto done_unlock;
3448
3449                 case RESTART_ARRAY_RW:
3450                         err = restart_array(mddev);
3451                         goto done_unlock;
3452
3453                 case STOP_ARRAY:
3454                         err = do_md_stop (mddev, 0);
3455                         goto done_unlock;
3456
3457                 case STOP_ARRAY_RO:
3458                         err = do_md_stop (mddev, 1);
3459                         goto done_unlock;
3460
3461         /*
3462          * We have a problem here : there is no easy way to give a CHS
3463          * virtual geometry. We currently pretend that we have a 2 heads
3464          * 4 sectors (with a BIG number of cylinders...). This drives
3465          * dosfs just mad... ;-)
3466          */
3467                 case HDIO_GETGEO:
3468                         if (!loc) {
3469                                 err = -EINVAL;
3470                                 goto abort_unlock;
3471                         }
3472                         err = put_user (2, (char __user *) &loc->heads);
3473                         if (err)
3474                                 goto abort_unlock;
3475                         err = put_user (4, (char __user *) &loc->sectors);
3476                         if (err)
3477                                 goto abort_unlock;
3478                         err = put_user(get_capacity(mddev->gendisk)/8,
3479                                         (short __user *) &loc->cylinders);
3480                         if (err)
3481                                 goto abort_unlock;
3482                         err = put_user (get_start_sect(inode->i_bdev),
3483                                                 (long __user *) &loc->start);
3484                         goto done_unlock;
3485         }
3486
3487         /*
3488          * The remaining ioctls are changing the state of the
3489          * superblock, so we do not allow them on read-only arrays.
3490          * However non-MD ioctls (e.g. get-size) will still come through
3491          * here and hit the 'default' below, so only disallow
3492          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3493          */
3494         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3495             mddev->ro && mddev->pers) {
3496                 if (mddev->ro == 2) {
3497                         mddev->ro = 0;
3498                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3499                 md_wakeup_thread(mddev->thread);
3500
3501                 } else {
3502                         err = -EROFS;
3503                         goto abort_unlock;
3504                 }
3505         }
3506
3507         switch (cmd)
3508         {
3509                 case ADD_NEW_DISK:
3510                 {
3511                         mdu_disk_info_t info;
3512                         if (copy_from_user(&info, argp, sizeof(info)))
3513                                 err = -EFAULT;
3514                         else
3515                                 err = add_new_disk(mddev, &info);
3516                         goto done_unlock;
3517                 }
3518
3519                 case HOT_REMOVE_DISK:
3520                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3521                         goto done_unlock;
3522
3523                 case HOT_ADD_DISK:
3524                         err = hot_add_disk(mddev, new_decode_dev(arg));
3525                         goto done_unlock;
3526
3527                 case SET_DISK_FAULTY:
3528                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3529                         goto done_unlock;
3530
3531                 case RUN_ARRAY:
3532                         err = do_md_run (mddev);
3533                         goto done_unlock;
3534
3535                 case SET_BITMAP_FILE:
3536                         err = set_bitmap_file(mddev, (int)arg);
3537                         goto done_unlock;
3538
3539                 default:
3540                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3541                                 printk(KERN_WARNING "md: %s(pid %d) used"
3542                                         " obsolete MD ioctl, upgrade your"
3543                                         " software to use new ictls.\n",
3544                                         current->comm, current->pid);
3545                         err = -EINVAL;
3546                         goto abort_unlock;
3547         }
3548
3549 done_unlock:
3550 abort_unlock:
3551         mddev_unlock(mddev);
3552
3553         return err;
3554 done:
3555         if (err)
3556                 MD_BUG();
3557 abort:
3558         return err;
3559 }
3560
3561 static int md_open(struct inode *inode, struct file *file)
3562 {
3563         /*
3564          * Succeed if we can lock the mddev, which confirms that
3565          * it isn't being stopped right now.
3566          */
3567         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3568         int err;
3569
3570         if ((err = mddev_lock(mddev)))
3571                 goto out;
3572
3573         err = 0;
3574         mddev_get(mddev);
3575         mddev_unlock(mddev);
3576
3577         check_disk_change(inode->i_bdev);
3578  out:
3579         return err;
3580 }
3581
3582 static int md_release(struct inode *inode, struct file * file)
3583 {
3584         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3585
3586         if (!mddev)
3587                 BUG();
3588         mddev_put(mddev);
3589
3590         return 0;
3591 }
3592
3593 static int md_media_changed(struct gendisk *disk)
3594 {
3595         mddev_t *mddev = disk->private_data;
3596
3597         return mddev->changed;
3598 }
3599
3600 static int md_revalidate(struct gendisk *disk)
3601 {
3602         mddev_t *mddev = disk->private_data;
3603
3604         mddev->changed = 0;
3605         return 0;
3606 }
3607 static struct block_device_operations md_fops =
3608 {
3609         .owner          = THIS_MODULE,
3610         .open           = md_open,
3611         .release        = md_release,
3612         .ioctl          = md_ioctl,
3613         .media_changed  = md_media_changed,
3614         .revalidate_disk= md_revalidate,
3615 };
3616
3617 static int md_thread(void * arg)
3618 {
3619         mdk_thread_t *thread = arg;
3620
3621         /*
3622          * md_thread is a 'system-thread', it's priority should be very
3623          * high. We avoid resource deadlocks individually in each
3624          * raid personality. (RAID5 does preallocation) We also use RR and
3625          * the very same RT priority as kswapd, thus we will never get
3626          * into a priority inversion deadlock.
3627          *
3628          * we definitely have to have equal or higher priority than
3629          * bdflush, otherwise bdflush will deadlock if there are too
3630          * many dirty RAID5 blocks.
3631          */
3632
3633         allow_signal(SIGKILL);
3634         while (!kthread_should_stop()) {
3635
3636                 /* We need to wait INTERRUPTIBLE so that
3637                  * we don't add to the load-average.
3638                  * That means we need to be sure no signals are
3639                  * pending
3640                  */
3641                 if (signal_pending(current))
3642                         flush_signals(current);
3643
3644                 wait_event_interruptible_timeout
3645                         (thread->wqueue,
3646                          test_bit(THREAD_WAKEUP, &thread->flags)
3647                          || kthread_should_stop(),
3648                          thread->timeout);
3649                 try_to_freeze();
3650
3651                 clear_bit(THREAD_WAKEUP, &thread->flags);
3652
3653                 thread->run(thread->mddev);
3654         }
3655
3656         return 0;
3657 }
3658
3659 void md_wakeup_thread(mdk_thread_t *thread)
3660 {
3661         if (thread) {
3662                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3663                 set_bit(THREAD_WAKEUP, &thread->flags);
3664                 wake_up(&thread->wqueue);
3665         }
3666 }
3667
3668 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3669                                  const char *name)
3670 {
3671         mdk_thread_t *thread;
3672
3673         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3674         if (!thread)
3675                 return NULL;
3676
3677         init_waitqueue_head(&thread->wqueue);
3678
3679         thread->run = run;
3680         thread->mddev = mddev;
3681         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3682         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3683         if (IS_ERR(thread->tsk)) {
3684                 kfree(thread);
3685                 return NULL;
3686         }
3687         return thread;
3688 }
3689
3690 void md_unregister_thread(mdk_thread_t *thread)
3691 {
3692         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3693
3694         kthread_stop(thread->tsk);
3695         kfree(thread);
3696 }
3697
3698 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3699 {
3700         if (!mddev) {
3701                 MD_BUG();
3702                 return;
3703         }
3704
3705         if (!rdev || test_bit(Faulty, &rdev->flags))
3706                 return;
3707 /*
3708         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3709                 mdname(mddev),
3710                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3711                 __builtin_return_address(0),__builtin_return_address(1),
3712                 __builtin_return_address(2),__builtin_return_address(3));
3713 */
3714         if (!mddev->pers->error_handler)
3715                 return;
3716         mddev->pers->error_handler(mddev,rdev);
3717         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3718         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3719         md_wakeup_thread(mddev->thread);
3720         md_new_event(mddev);
3721 }
3722
3723 /* seq_file implementation /proc/mdstat */
3724
3725 static void status_unused(struct seq_file *seq)
3726 {
3727         int i = 0;
3728         mdk_rdev_t *rdev;
3729         struct list_head *tmp;
3730
3731         seq_printf(seq, "unused devices: ");
3732
3733         ITERATE_RDEV_PENDING(rdev,tmp) {
3734                 char b[BDEVNAME_SIZE];
3735                 i++;
3736                 seq_printf(seq, "%s ",
3737                               bdevname(rdev->bdev,b));
3738         }
3739         if (!i)
3740                 seq_printf(seq, "<none>");
3741
3742         seq_printf(seq, "\n");
3743 }
3744
3745
3746 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3747 {
3748         unsigned long max_blocks, resync, res, dt, db, rt;
3749
3750         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3751
3752         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3753                 max_blocks = mddev->resync_max_sectors >> 1;
3754         else
3755                 max_blocks = mddev->size;
3756
3757         /*
3758          * Should not happen.
3759          */
3760         if (!max_blocks) {
3761                 MD_BUG();
3762                 return;
3763         }
3764         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3765         {
3766                 int i, x = res/50, y = 20-x;
3767                 seq_printf(seq, "[");
3768                 for (i = 0; i < x; i++)
3769                         seq_printf(seq, "=");
3770                 seq_printf(seq, ">");
3771                 for (i = 0; i < y; i++)
3772                         seq_printf(seq, ".");
3773                 seq_printf(seq, "] ");
3774         }
3775         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3776                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3777                        "resync" : "recovery"),
3778                       res/10, res % 10, resync, max_blocks);
3779
3780         /*
3781          * We do not want to overflow, so the order of operands and
3782          * the * 100 / 100 trick are important. We do a +1 to be
3783          * safe against division by zero. We only estimate anyway.
3784          *
3785          * dt: time from mark until now
3786          * db: blocks written from mark until now
3787          * rt: remaining time
3788          */
3789         dt = ((jiffies - mddev->resync_mark) / HZ);
3790         if (!dt) dt++;
3791         db = resync - (mddev->resync_mark_cnt/2);
3792         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3793
3794         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3795
3796         seq_printf(seq, " speed=%ldK/sec", db/dt);
3797 }
3798
3799 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3800 {
3801         struct list_head *tmp;
3802         loff_t l = *pos;
3803         mddev_t *mddev;
3804
3805         if (l >= 0x10000)
3806                 return NULL;
3807         if (!l--)
3808                 /* header */
3809                 return (void*)1;
3810
3811         spin_lock(&all_mddevs_lock);
3812         list_for_each(tmp,&all_mddevs)
3813                 if (!l--) {
3814                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3815                         mddev_get(mddev);
3816                         spin_unlock(&all_mddevs_lock);
3817                         return mddev;
3818                 }
3819         spin_unlock(&all_mddevs_lock);
3820         if (!l--)
3821                 return (void*)2;/* tail */
3822         return NULL;
3823 }
3824
3825 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3826 {
3827         struct list_head *tmp;
3828         mddev_t *next_mddev, *mddev = v;
3829
3830         ++*pos;
3831         if (v == (void*)2)
3832                 return NULL;
3833
3834         spin_lock(&all_mddevs_lock);
3835         if (v == (void*)1)
3836                 tmp = all_mddevs.next;
3837         else
3838                 tmp = mddev->all_mddevs.next;
3839         if (tmp != &all_mddevs)
3840                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3841         else {
3842                 next_mddev = (void*)2;
3843                 *pos = 0x10000;
3844         }
3845         spin_unlock(&all_mddevs_lock);
3846
3847         if (v != (void*)1)
3848                 mddev_put(mddev);
3849         return next_mddev;
3850
3851 }
3852
3853 static void md_seq_stop(struct seq_file *seq, void *v)
3854 {
3855         mddev_t *mddev = v;
3856
3857         if (mddev && v != (void*)1 && v != (void*)2)
3858                 mddev_put(mddev);
3859 }
3860
3861 struct mdstat_info {
3862         int event;
3863 };
3864
3865 static int md_seq_show(struct seq_file *seq, void *v)
3866 {
3867         mddev_t *mddev = v;
3868         sector_t size;
3869         struct list_head *tmp2;
3870         mdk_rdev_t *rdev;
3871         struct mdstat_info *mi = seq->private;
3872         struct bitmap *bitmap;
3873
3874         if (v == (void*)1) {
3875                 struct mdk_personality *pers;
3876                 seq_printf(seq, "Personalities : ");
3877                 spin_lock(&pers_lock);
3878                 list_for_each_entry(pers, &pers_list, list)
3879                         seq_printf(seq, "[%s] ", pers->name);
3880
3881                 spin_unlock(&pers_lock);
3882                 seq_printf(seq, "\n");
3883                 mi->event = atomic_read(&md_event_count);
3884                 return 0;
3885         }
3886         if (v == (void*)2) {
3887                 status_unused(seq);
3888                 return 0;
3889         }
3890
3891         if (mddev_lock(mddev)!=0)
3892                 return -EINTR;
3893         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3894                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3895                                                 mddev->pers ? "" : "in");
3896                 if (mddev->pers) {
3897                         if (mddev->ro==1)
3898                                 seq_printf(seq, " (read-only)");
3899                         if (mddev->ro==2)
3900                                 seq_printf(seq, "(auto-read-only)");
3901                         seq_printf(seq, " %s", mddev->pers->name);
3902                 }
3903
3904                 size = 0;
3905                 ITERATE_RDEV(mddev,rdev,tmp2) {
3906                         char b[BDEVNAME_SIZE];
3907                         seq_printf(seq, " %s[%d]",
3908                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3909                         if (test_bit(WriteMostly, &rdev->flags))
3910                                 seq_printf(seq, "(W)");
3911                         if (test_bit(Faulty, &rdev->flags)) {
3912                                 seq_printf(seq, "(F)");
3913                                 continue;
3914                         } else if (rdev->raid_disk < 0)
3915                                 seq_printf(seq, "(S)"); /* spare */
3916                         size += rdev->size;
3917                 }
3918
3919                 if (!list_empty(&mddev->disks)) {
3920                         if (mddev->pers)
3921                                 seq_printf(seq, "\n      %llu blocks",
3922                                         (unsigned long long)mddev->array_size);
3923                         else
3924                                 seq_printf(seq, "\n      %llu blocks",
3925                                         (unsigned long long)size);
3926                 }
3927                 if (mddev->persistent) {
3928                         if (mddev->major_version != 0 ||
3929                             mddev->minor_version != 90) {
3930                                 seq_printf(seq," super %d.%d",
3931                                            mddev->major_version,
3932                                            mddev->minor_version);
3933                         }
3934                 } else
3935                         seq_printf(seq, " super non-persistent");
3936
3937                 if (mddev->pers) {
3938                         mddev->pers->status (seq, mddev);
3939                         seq_printf(seq, "\n      ");
3940                         if (mddev->pers->sync_request) {
3941                                 if (mddev->curr_resync > 2) {
3942                                         status_resync (seq, mddev);
3943                                         seq_printf(seq, "\n      ");
3944                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3945                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3946                                 else if (mddev->recovery_cp < MaxSector)
3947                                         seq_printf(seq, "\tresync=PENDING\n      ");
3948                         }
3949                 } else
3950                         seq_printf(seq, "\n       ");
3951
3952                 if ((bitmap = mddev->bitmap)) {
3953                         unsigned long chunk_kb;
3954                         unsigned long flags;
3955                         spin_lock_irqsave(&bitmap->lock, flags);
3956                         chunk_kb = bitmap->chunksize >> 10;
3957                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3958                                 "%lu%s chunk",
3959                                 bitmap->pages - bitmap->missing_pages,
3960                                 bitmap->pages,
3961                                 (bitmap->pages - bitmap->missing_pages)
3962                                         << (PAGE_SHIFT - 10),
3963                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3964                                 chunk_kb ? "KB" : "B");
3965                         if (bitmap->file) {
3966                                 seq_printf(seq, ", file: ");
3967                                 seq_path(seq, bitmap->file->f_vfsmnt,
3968                                          bitmap->file->f_dentry," \t\n");
3969                         }
3970
3971                         seq_printf(seq, "\n");
3972                         spin_unlock_irqrestore(&bitmap->lock, flags);
3973                 }
3974
3975                 seq_printf(seq, "\n");
3976         }
3977         mddev_unlock(mddev);
3978
3979         return 0;
3980 }
3981
3982 static struct seq_operations md_seq_ops = {
3983         .start  = md_seq_start,
3984         .next   = md_seq_next,
3985         .stop   = md_seq_stop,
3986         .show   = md_seq_show,
3987 };
3988
3989 static int md_seq_open(struct inode *inode, struct file *file)
3990 {
3991         int error;
3992         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
3993         if (mi == NULL)
3994                 return -ENOMEM;
3995
3996         error = seq_open(file, &md_seq_ops);
3997         if (error)
3998                 kfree(mi);
3999         else {
4000                 struct seq_file *p = file->private_data;
4001                 p->private = mi;
4002                 mi->event = atomic_read(&md_event_count);
4003         }
4004         return error;
4005 }
4006
4007 static int md_seq_release(struct inode *inode, struct file *file)
4008 {
4009         struct seq_file *m = file->private_data;
4010         struct mdstat_info *mi = m->private;
4011         m->private = NULL;
4012         kfree(mi);
4013         return seq_release(inode, file);
4014 }
4015
4016 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4017 {
4018         struct seq_file *m = filp->private_data;
4019         struct mdstat_info *mi = m->private;
4020         int mask;
4021
4022         poll_wait(filp, &md_event_waiters, wait);
4023
4024         /* always allow read */
4025         mask = POLLIN | POLLRDNORM;
4026
4027         if (mi->event != atomic_read(&md_event_count))
4028                 mask |= POLLERR | POLLPRI;
4029         return mask;
4030 }
4031
4032 static struct file_operations md_seq_fops = {
4033         .open           = md_seq_open,
4034         .read           = seq_read,
4035         .llseek         = seq_lseek,
4036         .release        = md_seq_release,
4037         .poll           = mdstat_poll,
4038 };
4039
4040 int register_md_personality(struct mdk_personality *p)
4041 {
4042         spin_lock(&pers_lock);
4043         list_add_tail(&p->list, &pers_list);
4044         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4045         spin_unlock(&pers_lock);
4046         return 0;
4047 }
4048
4049 int unregister_md_personality(struct mdk_personality *p)
4050 {
4051         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4052         spin_lock(&pers_lock);
4053         list_del_init(&p->list);
4054         spin_unlock(&pers_lock);
4055         return 0;
4056 }
4057
4058 static int is_mddev_idle(mddev_t *mddev)
4059 {
4060         mdk_rdev_t * rdev;
4061         struct list_head *tmp;
4062         int idle;
4063         unsigned long curr_events;
4064
4065         idle = 1;
4066         ITERATE_RDEV(mddev,rdev,tmp) {
4067                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4068                 curr_events = disk_stat_read(disk, sectors[0]) +
4069                                 disk_stat_read(disk, sectors[1]) -
4070                                 atomic_read(&disk->sync_io);
4071                 /* The difference between curr_events and last_events
4072                  * will be affected by any new non-sync IO (making
4073                  * curr_events bigger) and any difference in the amount of
4074                  * in-flight syncio (making current_events bigger or smaller)
4075                  * The amount in-flight is currently limited to
4076                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4077                  * which is at most 4096 sectors.
4078                  * These numbers are fairly fragile and should be made
4079                  * more robust, probably by enforcing the
4080                  * 'window size' that md_do_sync sort-of uses.
4081                  *
4082                  * Note: the following is an unsigned comparison.
4083                  */
4084                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4085                         rdev->last_events = curr_events;
4086                         idle = 0;
4087                 }
4088         }
4089         return idle;
4090 }
4091
4092 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4093 {
4094         /* another "blocks" (512byte) blocks have been synced */
4095         atomic_sub(blocks, &mddev->recovery_active);
4096         wake_up(&mddev->recovery_wait);
4097         if (!ok) {
4098                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4099                 md_wakeup_thread(mddev->thread);
4100                 // stop recovery, signal do_sync ....
4101         }
4102 }
4103
4104
4105 /* md_write_start(mddev, bi)
4106  * If we need to update some array metadata (e.g. 'active' flag
4107  * in superblock) before writing, schedule a superblock update
4108  * and wait for it to complete.
4109  */
4110 void md_write_start(mddev_t *mddev, struct bio *bi)
4111 {
4112         if (bio_data_dir(bi) != WRITE)
4113                 return;
4114
4115         BUG_ON(mddev->ro == 1);
4116         if (mddev->ro == 2) {
4117                 /* need to switch to read/write */
4118                 mddev->ro = 0;
4119                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4120                 md_wakeup_thread(mddev->thread);
4121         }
4122         atomic_inc(&mddev->writes_pending);
4123         if (mddev->in_sync) {
4124                 spin_lock_irq(&mddev->write_lock);
4125                 if (mddev->in_sync) {
4126                         mddev->in_sync = 0;
4127                         mddev->sb_dirty = 1;
4128                         md_wakeup_thread(mddev->thread);
4129                 }
4130                 spin_unlock_irq(&mddev->write_lock);
4131         }
4132         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4133 }
4134
4135 void md_write_end(mddev_t *mddev)
4136 {
4137         if (atomic_dec_and_test(&mddev->writes_pending)) {
4138                 if (mddev->safemode == 2)
4139                         md_wakeup_thread(mddev->thread);
4140                 else
4141                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4142         }
4143 }
4144
4145 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4146
4147 #define SYNC_MARKS      10
4148 #define SYNC_MARK_STEP  (3*HZ)
4149 static void md_do_sync(mddev_t *mddev)
4150 {
4151         mddev_t *mddev2;
4152         unsigned int currspeed = 0,
4153                  window;
4154         sector_t max_sectors,j, io_sectors;
4155         unsigned long mark[SYNC_MARKS];
4156         sector_t mark_cnt[SYNC_MARKS];
4157         int last_mark,m;
4158         struct list_head *tmp;
4159         sector_t last_check;
4160         int skipped = 0;
4161
4162         /* just incase thread restarts... */
4163         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4164                 return;
4165
4166         /* we overload curr_resync somewhat here.
4167          * 0 == not engaged in resync at all
4168          * 2 == checking that there is no conflict with another sync
4169          * 1 == like 2, but have yielded to allow conflicting resync to
4170          *              commense
4171          * other == active in resync - this many blocks
4172          *
4173          * Before starting a resync we must have set curr_resync to
4174          * 2, and then checked that every "conflicting" array has curr_resync
4175          * less than ours.  When we find one that is the same or higher
4176          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4177          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4178          * This will mean we have to start checking from the beginning again.
4179          *
4180          */
4181
4182         do {
4183                 mddev->curr_resync = 2;
4184
4185         try_again:
4186                 if (kthread_should_stop()) {
4187                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4188                         goto skip;
4189                 }
4190                 ITERATE_MDDEV(mddev2,tmp) {
4191                         if (mddev2 == mddev)
4192                                 continue;
4193                         if (mddev2->curr_resync &&
4194                             match_mddev_units(mddev,mddev2)) {
4195                                 DEFINE_WAIT(wq);
4196                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4197                                         /* arbitrarily yield */
4198                                         mddev->curr_resync = 1;
4199                                         wake_up(&resync_wait);
4200                                 }
4201                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4202                                         /* no need to wait here, we can wait the next
4203                                          * time 'round when curr_resync == 2
4204                                          */
4205                                         continue;
4206                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4207                                 if (!kthread_should_stop() &&
4208                                     mddev2->curr_resync >= mddev->curr_resync) {
4209                                         printk(KERN_INFO "md: delaying resync of %s"
4210                                                " until %s has finished resync (they"
4211                                                " share one or more physical units)\n",
4212                                                mdname(mddev), mdname(mddev2));
4213                                         mddev_put(mddev2);
4214                                         schedule();
4215                                         finish_wait(&resync_wait, &wq);
4216                                         goto try_again;
4217                                 }
4218                                 finish_wait(&resync_wait, &wq);
4219                         }
4220                 }
4221         } while (mddev->curr_resync < 2);
4222
4223         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4224                 /* resync follows the size requested by the personality,
4225                  * which defaults to physical size, but can be virtual size
4226                  */
4227                 max_sectors = mddev->resync_max_sectors;
4228                 mddev->resync_mismatches = 0;
4229         } else
4230                 /* recovery follows the physical size of devices */
4231                 max_sectors = mddev->size << 1;
4232
4233         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4234         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4235                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4236         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4237                "(but not more than %d KB/sec) for reconstruction.\n",
4238                sysctl_speed_limit_max);
4239
4240         is_mddev_idle(mddev); /* this also initializes IO event counters */
4241         /* we don't use the checkpoint if there's a bitmap */
4242         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4243             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4244                 j = mddev->recovery_cp;
4245         else
4246                 j = 0;
4247         io_sectors = 0;
4248         for (m = 0; m < SYNC_MARKS; m++) {
4249                 mark[m] = jiffies;
4250                 mark_cnt[m] = io_sectors;
4251         }
4252         last_mark = 0;
4253         mddev->resync_mark = mark[last_mark];
4254         mddev->resync_mark_cnt = mark_cnt[last_mark];
4255
4256         /*
4257          * Tune reconstruction:
4258          */
4259         window = 32*(PAGE_SIZE/512);
4260         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4261                 window/2,(unsigned long long) max_sectors/2);
4262
4263         atomic_set(&mddev->recovery_active, 0);
4264         init_waitqueue_head(&mddev->recovery_wait);
4265         last_check = 0;
4266
4267         if (j>2) {
4268                 printk(KERN_INFO
4269                         "md: resuming recovery of %s from checkpoint.\n",
4270                         mdname(mddev));
4271                 mddev->curr_resync = j;
4272         }
4273
4274         while (j < max_sectors) {
4275                 sector_t sectors;
4276
4277                 skipped = 0;
4278                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4279                                             currspeed < sysctl_speed_limit_min);
4280                 if (sectors == 0) {
4281                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4282                         goto out;
4283                 }
4284
4285                 if (!skipped) { /* actual IO requested */
4286                         io_sectors += sectors;
4287                         atomic_add(sectors, &mddev->recovery_active);
4288                 }
4289
4290                 j += sectors;
4291                 if (j>1) mddev->curr_resync = j;
4292                 if (last_check == 0)
4293                         /* this is the earliers that rebuilt will be
4294                          * visible in /proc/mdstat
4295                          */
4296                         md_new_event(mddev);
4297
4298                 if (last_check + window > io_sectors || j == max_sectors)
4299                         continue;
4300
4301                 last_check = io_sectors;
4302
4303                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4304                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4305                         break;
4306
4307         repeat:
4308                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4309                         /* step marks */
4310                         int next = (last_mark+1) % SYNC_MARKS;
4311
4312                         mddev->resync_mark = mark[next];
4313                         mddev->resync_mark_cnt = mark_cnt[next];
4314                         mark[next] = jiffies;
4315                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4316                         last_mark = next;
4317                 }
4318
4319
4320                 if (kthread_should_stop()) {
4321                         /*
4322                          * got a signal, exit.
4323                          */
4324                         printk(KERN_INFO
4325                                 "md: md_do_sync() got signal ... exiting\n");
4326                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4327                         goto out;
4328                 }
4329
4330                 /*
4331                  * this loop exits only if either when we are slower than
4332                  * the 'hard' speed limit, or the system was IO-idle for
4333                  * a jiffy.
4334                  * the system might be non-idle CPU-wise, but we only care
4335                  * about not overloading the IO subsystem. (things like an
4336                  * e2fsck being done on the RAID array should execute fast)
4337                  */
4338                 mddev->queue->unplug_fn(mddev->queue);
4339                 cond_resched();
4340
4341                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4342                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4343
4344                 if (currspeed > sysctl_speed_limit_min) {
4345                         if ((currspeed > sysctl_speed_limit_max) ||
4346                                         !is_mddev_idle(mddev)) {
4347                                 msleep(500);
4348                                 goto repeat;
4349                         }
4350                 }
4351         }
4352         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4353         /*
4354          * this also signals 'finished resyncing' to md_stop
4355          */
4356  out:
4357         mddev->queue->unplug_fn(mddev->queue);
4358
4359         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4360
4361         /* tell personality that we are finished */
4362         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4363
4364         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4365             mddev->curr_resync > 2 &&
4366             mddev->curr_resync >= mddev->recovery_cp) {
4367                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4368                         printk(KERN_INFO
4369                                 "md: checkpointing recovery of %s.\n",
4370                                 mdname(mddev));
4371                         mddev->recovery_cp = mddev->curr_resync;
4372                 } else
4373                         mddev->recovery_cp = MaxSector;
4374         }
4375
4376  skip:
4377         mddev->curr_resync = 0;
4378         wake_up(&resync_wait);
4379         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4380         md_wakeup_thread(mddev->thread);
4381 }
4382
4383
4384 /*
4385  * This routine is regularly called by all per-raid-array threads to
4386  * deal with generic issues like resync and super-block update.
4387  * Raid personalities that don't have a thread (linear/raid0) do not
4388  * need this as they never do any recovery or update the superblock.
4389  *
4390  * It does not do any resync itself, but rather "forks" off other threads
4391  * to do that as needed.
4392  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4393  * "->recovery" and create a thread at ->sync_thread.
4394  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4395  * and wakeups up this thread which will reap the thread and finish up.
4396  * This thread also removes any faulty devices (with nr_pending == 0).
4397  *
4398  * The overall approach is:
4399  *  1/ if the superblock needs updating, update it.
4400  *  2/ If a recovery thread is running, don't do anything else.
4401  *  3/ If recovery has finished, clean up, possibly marking spares active.
4402  *  4/ If there are any faulty devices, remove them.
4403  *  5/ If array is degraded, try to add spares devices
4404  *  6/ If array has spares or is not in-sync, start a resync thread.
4405  */
4406 void md_check_recovery(mddev_t *mddev)
4407 {
4408         mdk_rdev_t *rdev;
4409         struct list_head *rtmp;
4410
4411
4412         if (mddev->bitmap)
4413                 bitmap_daemon_work(mddev->bitmap);
4414
4415         if (mddev->ro)
4416                 return;
4417
4418         if (signal_pending(current)) {
4419                 if (mddev->pers->sync_request) {
4420                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4421                                mdname(mddev));
4422                         mddev->safemode = 2;
4423                 }
4424                 flush_signals(current);
4425         }
4426
4427         if ( ! (
4428                 mddev->sb_dirty ||
4429                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4430                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4431                 (mddev->safemode == 1) ||
4432                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4433                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4434                 ))
4435                 return;
4436
4437         if (mddev_trylock(mddev)==0) {
4438                 int spares =0;
4439
4440                 spin_lock_irq(&mddev->write_lock);
4441                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4442                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4443                         mddev->in_sync = 1;
4444                         mddev->sb_dirty = 1;
4445                 }
4446                 if (mddev->safemode == 1)
4447                         mddev->safemode = 0;
4448                 spin_unlock_irq(&mddev->write_lock);
4449
4450                 if (mddev->sb_dirty)
4451                         md_update_sb(mddev);
4452
4453
4454                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4455                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4456                         /* resync/recovery still happening */
4457                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4458                         goto unlock;
4459                 }
4460                 if (mddev->sync_thread) {
4461                         /* resync has finished, collect result */
4462                         md_unregister_thread(mddev->sync_thread);
4463                         mddev->sync_thread = NULL;
4464                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4465                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4466                                 /* success...*/
4467                                 /* activate any spares */
4468                                 mddev->pers->spare_active(mddev);
4469                         }
4470                         md_update_sb(mddev);
4471
4472                         /* if array is no-longer degraded, then any saved_raid_disk
4473                          * information must be scrapped
4474                          */
4475                         if (!mddev->degraded)
4476                                 ITERATE_RDEV(mddev,rdev,rtmp)
4477                                         rdev->saved_raid_disk = -1;
4478
4479                         mddev->recovery = 0;
4480                         /* flag recovery needed just to double check */
4481                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4482                         md_new_event(mddev);
4483                         goto unlock;
4484                 }
4485                 /* Clear some bits that don't mean anything, but
4486                  * might be left set
4487                  */
4488                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4489                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4490                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4491                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4492
4493                 /* no recovery is running.
4494                  * remove any failed drives, then
4495                  * add spares if possible.
4496                  * Spare are also removed and re-added, to allow
4497                  * the personality to fail the re-add.
4498                  */
4499                 ITERATE_RDEV(mddev,rdev,rtmp)
4500                         if (rdev->raid_disk >= 0 &&
4501                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4502                             atomic_read(&rdev->nr_pending)==0) {
4503                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4504                                         char nm[20];
4505                                         sprintf(nm,"rd%d", rdev->raid_disk);
4506                                         sysfs_remove_link(&mddev->kobj, nm);
4507                                         rdev->raid_disk = -1;
4508                                 }
4509                         }
4510
4511                 if (mddev->degraded) {
4512                         ITERATE_RDEV(mddev,rdev,rtmp)
4513                                 if (rdev->raid_disk < 0
4514                                     && !test_bit(Faulty, &rdev->flags)) {
4515                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4516                                                 char nm[20];
4517                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4518                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4519                                                 spares++;
4520                                                 md_new_event(mddev);
4521                                         } else
4522                                                 break;
4523                                 }
4524                 }
4525
4526                 if (spares) {
4527                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4528                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4529                 } else if (mddev->recovery_cp < MaxSector) {
4530                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4531                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4532                         /* nothing to be done ... */
4533                         goto unlock;
4534
4535                 if (mddev->pers->sync_request) {
4536                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4537                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4538                                 /* We are adding a device or devices to an array
4539                                  * which has the bitmap stored on all devices.
4540                                  * So make sure all bitmap pages get written
4541                                  */
4542                                 bitmap_write_all(mddev->bitmap);
4543                         }
4544                         mddev->sync_thread = md_register_thread(md_do_sync,
4545                                                                 mddev,
4546                                                                 "%s_resync");
4547                         if (!mddev->sync_thread) {
4548                                 printk(KERN_ERR "%s: could not start resync"
4549                                         " thread...\n",
4550                                         mdname(mddev));
4551                                 /* leave the spares where they are, it shouldn't hurt */
4552                                 mddev->recovery = 0;
4553                         } else
4554                                 md_wakeup_thread(mddev->sync_thread);
4555                         md_new_event(mddev);
4556                 }
4557         unlock:
4558                 mddev_unlock(mddev);
4559         }
4560 }
4561
4562 static int md_notify_reboot(struct notifier_block *this,
4563                             unsigned long code, void *x)
4564 {
4565         struct list_head *tmp;
4566         mddev_t *mddev;
4567
4568         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4569
4570                 printk(KERN_INFO "md: stopping all md devices.\n");
4571
4572                 ITERATE_MDDEV(mddev,tmp)
4573                         if (mddev_trylock(mddev)==0)
4574                                 do_md_stop (mddev, 1);
4575                 /*
4576                  * certain more exotic SCSI devices are known to be
4577                  * volatile wrt too early system reboots. While the
4578                  * right place to handle this issue is the given
4579                  * driver, we do want to have a safe RAID driver ...
4580                  */
4581                 mdelay(1000*1);
4582         }
4583         return NOTIFY_DONE;
4584 }
4585
4586 static struct notifier_block md_notifier = {
4587         .notifier_call  = md_notify_reboot,
4588         .next           = NULL,
4589         .priority       = INT_MAX, /* before any real devices */
4590 };
4591
4592 static void md_geninit(void)
4593 {
4594         struct proc_dir_entry *p;
4595
4596         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4597
4598         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4599         if (p)
4600                 p->proc_fops = &md_seq_fops;
4601 }
4602
4603 static int __init md_init(void)
4604 {
4605         int minor;
4606
4607         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4608                         " MD_SB_DISKS=%d\n",
4609                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4610                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4611         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4612                         BITMAP_MINOR);
4613
4614         if (register_blkdev(MAJOR_NR, "md"))
4615                 return -1;
4616         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4617                 unregister_blkdev(MAJOR_NR, "md");
4618                 return -1;
4619         }
4620         devfs_mk_dir("md");
4621         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4622                                 md_probe, NULL, NULL);
4623         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4624                             md_probe, NULL, NULL);
4625
4626         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4627                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4628                                 S_IFBLK|S_IRUSR|S_IWUSR,
4629                                 "md/%d", minor);
4630
4631         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4632                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4633                               S_IFBLK|S_IRUSR|S_IWUSR,
4634                               "md/mdp%d", minor);
4635
4636
4637         register_reboot_notifier(&md_notifier);
4638         raid_table_header = register_sysctl_table(raid_root_table, 1);
4639
4640         md_geninit();
4641         return (0);
4642 }
4643
4644
4645 #ifndef MODULE
4646
4647 /*
4648  * Searches all registered partitions for autorun RAID arrays
4649  * at boot time.
4650  */
4651 static dev_t detected_devices[128];
4652 static int dev_cnt;
4653
4654 void md_autodetect_dev(dev_t dev)
4655 {
4656         if (dev_cnt >= 0 && dev_cnt < 127)
4657                 detected_devices[dev_cnt++] = dev;
4658 }
4659
4660
4661 static void autostart_arrays(int part)
4662 {
4663         mdk_rdev_t *rdev;
4664         int i;
4665
4666         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4667
4668         for (i = 0; i < dev_cnt; i++) {
4669                 dev_t dev = detected_devices[i];
4670
4671                 rdev = md_import_device(dev,0, 0);
4672                 if (IS_ERR(rdev))
4673                         continue;
4674
4675                 if (test_bit(Faulty, &rdev->flags)) {
4676                         MD_BUG();
4677                         continue;
4678                 }
4679                 list_add(&rdev->same_set, &pending_raid_disks);
4680         }
4681         dev_cnt = 0;
4682
4683         autorun_devices(part);
4684 }
4685
4686 #endif
4687
4688 static __exit void md_exit(void)
4689 {
4690         mddev_t *mddev;
4691         struct list_head *tmp;
4692         int i;
4693         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4694         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4695         for (i=0; i < MAX_MD_DEVS; i++)
4696                 devfs_remove("md/%d", i);
4697         for (i=0; i < MAX_MD_DEVS; i++)
4698                 devfs_remove("md/d%d", i);
4699
4700         devfs_remove("md");
4701
4702         unregister_blkdev(MAJOR_NR,"md");
4703         unregister_blkdev(mdp_major, "mdp");
4704         unregister_reboot_notifier(&md_notifier);
4705         unregister_sysctl_table(raid_table_header);
4706         remove_proc_entry("mdstat", NULL);
4707         ITERATE_MDDEV(mddev,tmp) {
4708                 struct gendisk *disk = mddev->gendisk;
4709                 if (!disk)
4710                         continue;
4711                 export_array(mddev);
4712                 del_gendisk(disk);
4713                 put_disk(disk);
4714                 mddev->gendisk = NULL;
4715                 mddev_put(mddev);
4716         }
4717 }
4718
4719 module_init(md_init)
4720 module_exit(md_exit)
4721
4722 static int get_ro(char *buffer, struct kernel_param *kp)
4723 {
4724         return sprintf(buffer, "%d", start_readonly);
4725 }
4726 static int set_ro(const char *val, struct kernel_param *kp)
4727 {
4728         char *e;
4729         int num = simple_strtoul(val, &e, 10);
4730         if (*val && (*e == '\0' || *e == '\n')) {
4731                 start_readonly = num;
4732                 return 0;;
4733         }
4734         return -EINVAL;
4735 }
4736
4737 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4738 module_param(start_dirty_degraded, int, 0644);
4739
4740
4741 EXPORT_SYMBOL(register_md_personality);
4742 EXPORT_SYMBOL(unregister_md_personality);
4743 EXPORT_SYMBOL(md_error);
4744 EXPORT_SYMBOL(md_done_sync);
4745 EXPORT_SYMBOL(md_write_start);
4746 EXPORT_SYMBOL(md_write_end);
4747 EXPORT_SYMBOL(md_register_thread);
4748 EXPORT_SYMBOL(md_unregister_thread);
4749 EXPORT_SYMBOL(md_wakeup_thread);
4750 EXPORT_SYMBOL(md_print_devices);
4751 EXPORT_SYMBOL(md_check_recovery);
4752 MODULE_LICENSE("GPL");
4753 MODULE_ALIAS("md");
4754 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);