drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45 #include <linux/poll.h>
  46
  47 #include <linux/init.h>
  48
  49 #include <linux/file.h>
  50
  51 #ifdef CONFIG_KMOD
  52 #include <linux/kmod.h>
  53 #endif
  54
  55 #include <asm/unaligned.h>
  56
  57 #define MAJOR_NR MD_MAJOR
  58 #define MD_DRIVER
  59
  60 /* 63 partitions with the alternate major number (mdp) */
  61 #define MdpMinorShift 6
  62
  63 #define DEBUG 0
  64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  65
  66
  67 #ifndef MODULE
  68 static void autostart_arrays (int part);
  69 #endif
  70
  71 static LIST_HEAD(pers_list);
  72 static DEFINE_SPINLOCK(pers_lock);
  73
  74 /*
  75  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  76  * is 1000 KB/sec, so the extra system load does not show up that much.
  77  * Increase it if you want to have more _guaranteed_ speed. Note that
  78  * the RAID driver will use the maximum available bandwidth if the IO
  79  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  80  * speed limit - in case reconstruction slows down your system despite
  81  * idle IO detection.
  82  *
  83  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  84  */
  85
  86 static int sysctl_speed_limit_min = 1000;
  87 static int sysctl_speed_limit_max = 200000;
  88
  89 static struct ctl_table_header *raid_table_header;
  90
  91 static ctl_table raid_table[] = {
  92         {
  93                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  94                 .procname       = "speed_limit_min",
  95                 .data           = &sysctl_speed_limit_min,
  96                 .maxlen         = sizeof(int),
  97                 .mode           = 0644,
  98                 .proc_handler   = &proc_dointvec,
  99         },
 100         {
 101                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 102                 .procname       = "speed_limit_max",
 103                 .data           = &sysctl_speed_limit_max,
 104                 .maxlen         = sizeof(int),
 105                 .mode           = 0644,
 106                 .proc_handler   = &proc_dointvec,
 107         },
 108         { .ctl_name = 0 }
 109 };
 110
 111 static ctl_table raid_dir_table[] = {
 112         {
 113                 .ctl_name       = DEV_RAID,
 114                 .procname       = "raid",
 115                 .maxlen         = 0,
 116                 .mode           = 0555,
 117                 .child          = raid_table,
 118         },
 119         { .ctl_name = 0 }
 120 };
 121
 122 static ctl_table raid_root_table[] = {
 123         {
 124                 .ctl_name       = CTL_DEV,
 125                 .procname       = "dev",
 126                 .maxlen         = 0,
 127                 .mode           = 0555,
 128                 .child          = raid_dir_table,
 129         },
 130         { .ctl_name = 0 }
 131 };
 132
 133 static struct block_device_operations md_fops;
 134
 135 static int start_readonly;
 136
 137 /*
 138  * We have a system wide 'event count' that is incremented
 139  * on any 'interesting' event, and readers of /proc/mdstat
 140  * can use 'poll' or 'select' to find out when the event
 141  * count increases.
 142  *
 143  * Events are:
 144  *  start array, stop array, error, add device, remove device,
 145  *  start build, activate spare
 146  */
 147 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 148 static atomic_t md_event_count;
 149 static void md_new_event(mddev_t *mddev)
 150 {
 151         atomic_inc(&md_event_count);
 152         wake_up(&md_event_waiters);
 153 }
 154
 155 /*
 156  * Enables to iterate over all existing md arrays
 157  * all_mddevs_lock protects this list.
 158  */
 159 static LIST_HEAD(all_mddevs);
 160 static DEFINE_SPINLOCK(all_mddevs_lock);
 161
 162
 163 /*
 164  * iterates through all used mddevs in the system.
 165  * We take care to grab the all_mddevs_lock whenever navigating
 166  * the list, and to always hold a refcount when unlocked.
 167  * Any code which breaks out of this loop while own
 168  * a reference to the current mddev and must mddev_put it.
 169  */
 170 #define ITERATE_MDDEV(mddev,tmp)                                        \
 171                                                                         \
 172         for (({ spin_lock(&all_mddevs_lock);                            \
 173                 tmp = all_mddevs.next;                                  \
 174                 mddev = NULL;});                                        \
 175              ({ if (tmp != &all_mddevs)                                 \
 176                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 177                 spin_unlock(&all_mddevs_lock);                          \
 178                 if (mddev) mddev_put(mddev);                            \
 179                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 180                 tmp != &all_mddevs;});                                  \
 181              ({ spin_lock(&all_mddevs_lock);                            \
 182                 tmp = tmp->next;})                                      \
 183                 )
 184
 185
 186 static int md_fail_request (request_queue_t *q, struct bio *bio)
 187 {
 188         bio_io_error(bio, bio->bi_size);
 189         return 0;
 190 }
 191
 192 static inline mddev_t *mddev_get(mddev_t *mddev)
 193 {
 194         atomic_inc(&mddev->active);
 195         return mddev;
 196 }
 197
 198 static void mddev_put(mddev_t *mddev)
 199 {
 200         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 201                 return;
 202         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 203                 list_del(&mddev->all_mddevs);
 204                 blk_put_queue(mddev->queue);
 205                 kobject_unregister(&mddev->kobj);
 206         }
 207         spin_unlock(&all_mddevs_lock);
 208 }
 209
 210 static mddev_t * mddev_find(dev_t unit)
 211 {
 212         mddev_t *mddev, *new = NULL;
 213
 214  retry:
 215         spin_lock(&all_mddevs_lock);
 216         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 217                 if (mddev->unit == unit) {
 218                         mddev_get(mddev);
 219                         spin_unlock(&all_mddevs_lock);
 220                         kfree(new);
 221                         return mddev;
 222                 }
 223
 224         if (new) {
 225                 list_add(&new->all_mddevs, &all_mddevs);
 226                 spin_unlock(&all_mddevs_lock);
 227                 return new;
 228         }
 229         spin_unlock(&all_mddevs_lock);
 230
 231         new = kzalloc(sizeof(*new), GFP_KERNEL);
 232         if (!new)
 233                 return NULL;
 234
 235         new->unit = unit;
 236         if (MAJOR(unit) == MD_MAJOR)
 237                 new->md_minor = MINOR(unit);
 238         else
 239                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 240
 241         init_MUTEX(&new->reconfig_sem);
 242         INIT_LIST_HEAD(&new->disks);
 243         INIT_LIST_HEAD(&new->all_mddevs);
 244         init_timer(&new->safemode_timer);
 245         atomic_set(&new->active, 1);
 246         spin_lock_init(&new->write_lock);
 247         init_waitqueue_head(&new->sb_wait);
 248
 249         new->queue = blk_alloc_queue(GFP_KERNEL);
 250         if (!new->queue) {
 251                 kfree(new);
 252                 return NULL;
 253         }
 254
 255         blk_queue_make_request(new->queue, md_fail_request);
 256
 257         goto retry;
 258 }
 259
 260 static inline int mddev_lock(mddev_t * mddev)
 261 {
 262         return down_interruptible(&mddev->reconfig_sem);
 263 }
 264
 265 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 266 {
 267         down(&mddev->reconfig_sem);
 268 }
 269
 270 static inline int mddev_trylock(mddev_t * mddev)
 271 {
 272         return down_trylock(&mddev->reconfig_sem);
 273 }
 274
 275 static inline void mddev_unlock(mddev_t * mddev)
 276 {
 277         up(&mddev->reconfig_sem);
 278
 279         md_wakeup_thread(mddev->thread);
 280 }
 281
 282 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 283 {
 284         mdk_rdev_t * rdev;
 285         struct list_head *tmp;
 286
 287         ITERATE_RDEV(mddev,rdev,tmp) {
 288                 if (rdev->desc_nr == nr)
 289                         return rdev;
 290         }
 291         return NULL;
 292 }
 293
 294 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 295 {
 296         struct list_head *tmp;
 297         mdk_rdev_t *rdev;
 298
 299         ITERATE_RDEV(mddev,rdev,tmp) {
 300                 if (rdev->bdev->bd_dev == dev)
 301                         return rdev;
 302         }
 303         return NULL;
 304 }
 305
 306 static struct mdk_personality *find_pers(int level, char *clevel)
 307 {
 308         struct mdk_personality *pers;
 309         list_for_each_entry(pers, &pers_list, list) {
 310                 if (level != LEVEL_NONE && pers->level == level)
 311                         return pers;
 312                 if (strcmp(pers->name, clevel)==0)
 313                         return pers;
 314         }
 315         return NULL;
 316 }
 317
 318 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 319 {
 320         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 321         return MD_NEW_SIZE_BLOCKS(size);
 322 }
 323
 324 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 325 {
 326         sector_t size;
 327
 328         size = rdev->sb_offset;
 329
 330         if (chunk_size)
 331                 size &= ~((sector_t)chunk_size/1024 - 1);
 332         return size;
 333 }
 334
 335 static int alloc_disk_sb(mdk_rdev_t * rdev)
 336 {
 337         if (rdev->sb_page)
 338                 MD_BUG();
 339
 340         rdev->sb_page = alloc_page(GFP_KERNEL);
 341         if (!rdev->sb_page) {
 342                 printk(KERN_ALERT "md: out of memory.\n");
 343                 return -EINVAL;
 344         }
 345
 346         return 0;
 347 }
 348
 349 static void free_disk_sb(mdk_rdev_t * rdev)
 350 {
 351         if (rdev->sb_page) {
 352                 put_page(rdev->sb_page);
 353                 rdev->sb_loaded = 0;
 354                 rdev->sb_page = NULL;
 355                 rdev->sb_offset = 0;
 356                 rdev->size = 0;
 357         }
 358 }
 359
 360
 361 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 362 {
 363         mdk_rdev_t *rdev = bio->bi_private;
 364         mddev_t *mddev = rdev->mddev;
 365         if (bio->bi_size)
 366                 return 1;
 367
 368         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 369                 md_error(mddev, rdev);
 370
 371         if (atomic_dec_and_test(&mddev->pending_writes))
 372                 wake_up(&mddev->sb_wait);
 373         bio_put(bio);
 374         return 0;
 375 }
 376
 377 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 378 {
 379         struct bio *bio2 = bio->bi_private;
 380         mdk_rdev_t *rdev = bio2->bi_private;
 381         mddev_t *mddev = rdev->mddev;
 382         if (bio->bi_size)
 383                 return 1;
 384
 385         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 386             error == -EOPNOTSUPP) {
 387                 unsigned long flags;
 388                 /* barriers don't appear to be supported :-( */
 389                 set_bit(BarriersNotsupp, &rdev->flags);
 390                 mddev->barriers_work = 0;
 391                 spin_lock_irqsave(&mddev->write_lock, flags);
 392                 bio2->bi_next = mddev->biolist;
 393                 mddev->biolist = bio2;
 394                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 395                 wake_up(&mddev->sb_wait);
 396                 bio_put(bio);
 397                 return 0;
 398         }
 399         bio_put(bio2);
 400         bio->bi_private = rdev;
 401         return super_written(bio, bytes_done, error);
 402 }
 403
 404 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 405                    sector_t sector, int size, struct page *page)
 406 {
 407         /* write first size bytes of page to sector of rdev
 408          * Increment mddev->pending_writes before returning
 409          * and decrement it on completion, waking up sb_wait
 410          * if zero is reached.
 411          * If an error occurred, call md_error
 412          *
 413          * As we might need to resubmit the request if BIO_RW_BARRIER
 414          * causes ENOTSUPP, we allocate a spare bio...
 415          */
 416         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 417         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 418
 419         bio->bi_bdev = rdev->bdev;
 420         bio->bi_sector = sector;
 421         bio_add_page(bio, page, size, 0);
 422         bio->bi_private = rdev;
 423         bio->bi_end_io = super_written;
 424         bio->bi_rw = rw;
 425
 426         atomic_inc(&mddev->pending_writes);
 427         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 428                 struct bio *rbio;
 429                 rw |= (1<<BIO_RW_BARRIER);
 430                 rbio = bio_clone(bio, GFP_NOIO);
 431                 rbio->bi_private = bio;
 432                 rbio->bi_end_io = super_written_barrier;
 433                 submit_bio(rw, rbio);
 434         } else
 435                 submit_bio(rw, bio);
 436 }
 437
 438 void md_super_wait(mddev_t *mddev)
 439 {
 440         /* wait for all superblock writes that were scheduled to complete.
 441          * if any had to be retried (due to BARRIER problems), retry them
 442          */
 443         DEFINE_WAIT(wq);
 444         for(;;) {
 445                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 446                 if (atomic_read(&mddev->pending_writes)==0)
 447                         break;
 448                 while (mddev->biolist) {
 449                         struct bio *bio;
 450                         spin_lock_irq(&mddev->write_lock);
 451                         bio = mddev->biolist;
 452                         mddev->biolist = bio->bi_next ;
 453                         bio->bi_next = NULL;
 454                         spin_unlock_irq(&mddev->write_lock);
 455                         submit_bio(bio->bi_rw, bio);
 456                 }
 457                 schedule();
 458         }
 459         finish_wait(&mddev->sb_wait, &wq);
 460 }
 461
 462 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 463 {
 464         if (bio->bi_size)
 465                 return 1;
 466
 467         complete((struct completion*)bio->bi_private);
 468         return 0;
 469 }
 470
 471 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 472                    struct page *page, int rw)
 473 {
 474         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 475         struct completion event;
 476         int ret;
 477
 478         rw |= (1 << BIO_RW_SYNC);
 479
 480         bio->bi_bdev = bdev;
 481         bio->bi_sector = sector;
 482         bio_add_page(bio, page, size, 0);
 483         init_completion(&event);
 484         bio->bi_private = &event;
 485         bio->bi_end_io = bi_complete;
 486         submit_bio(rw, bio);
 487         wait_for_completion(&event);
 488
 489         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 490         bio_put(bio);
 491         return ret;
 492 }
 493 EXPORT_SYMBOL_GPL(sync_page_io);
 494
 495 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 496 {
 497         char b[BDEVNAME_SIZE];
 498         if (!rdev->sb_page) {
 499                 MD_BUG();
 500                 return -EINVAL;
 501         }
 502         if (rdev->sb_loaded)
 503                 return 0;
 504
 505
 506         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 507                 goto fail;
 508         rdev->sb_loaded = 1;
 509         return 0;
 510
 511 fail:
 512         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 513                 bdevname(rdev->bdev,b));
 514         return -EINVAL;
 515 }
 516
 517 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 518 {
 519         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 520                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 521                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 522                 (sb1->set_uuid3 == sb2->set_uuid3))
 523
 524                 return 1;
 525
 526         return 0;
 527 }
 528
 529
 530 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 531 {
 532         int ret;
 533         mdp_super_t *tmp1, *tmp2;
 534
 535         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 536         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 537
 538         if (!tmp1 || !tmp2) {
 539                 ret = 0;
 540                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 541                 goto abort;
 542         }
 543
 544         *tmp1 = *sb1;
 545         *tmp2 = *sb2;
 546
 547         /*
 548          * nr_disks is not constant
 549          */
 550         tmp1->nr_disks = 0;
 551         tmp2->nr_disks = 0;
 552
 553         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 554                 ret = 0;
 555         else
 556                 ret = 1;
 557
 558 abort:
 559         kfree(tmp1);
 560         kfree(tmp2);
 561         return ret;
 562 }
 563
 564 static unsigned int calc_sb_csum(mdp_super_t * sb)
 565 {
 566         unsigned int disk_csum, csum;
 567
 568         disk_csum = sb->sb_csum;
 569         sb->sb_csum = 0;
 570         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 571         sb->sb_csum = disk_csum;
 572         return csum;
 573 }
 574
 575
 576 /*
 577  * Handle superblock details.
 578  * We want to be able to handle multiple superblock formats
 579  * so we have a common interface to them all, and an array of
 580  * different handlers.
 581  * We rely on user-space to write the initial superblock, and support
 582  * reading and updating of superblocks.
 583  * Interface methods are:
 584  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 585  *      loads and validates a superblock on dev.
 586  *      if refdev != NULL, compare superblocks on both devices
 587  *    Return:
 588  *      0 - dev has a superblock that is compatible with refdev
 589  *      1 - dev has a superblock that is compatible and newer than refdev
 590  *          so dev should be used as the refdev in future
 591  *     -EINVAL superblock incompatible or invalid
 592  *     -othererror e.g. -EIO
 593  *
 594  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 595  *      Verify that dev is acceptable into mddev.
 596  *       The first time, mddev->raid_disks will be 0, and data from
 597  *       dev should be merged in.  Subsequent calls check that dev
 598  *       is new enough.  Return 0 or -EINVAL
 599  *
 600  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 601  *     Update the superblock for rdev with data in mddev
 602  *     This does not write to disc.
 603  *
 604  */
 605
 606 struct super_type  {
 607         char            *name;
 608         struct module   *owner;
 609         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 610         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 611         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 612 };
 613
 614 /*
 615  * load_super for 0.90.0
 616  */
 617 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 618 {
 619         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 620         mdp_super_t *sb;
 621         int ret;
 622         sector_t sb_offset;
 623
 624         /*
 625          * Calculate the position of the superblock,
 626          * it's at the end of the disk.
 627          *
 628          * It also happens to be a multiple of 4Kb.
 629          */
 630         sb_offset = calc_dev_sboffset(rdev->bdev);
 631         rdev->sb_offset = sb_offset;
 632
 633         ret = read_disk_sb(rdev, MD_SB_BYTES);
 634         if (ret) return ret;
 635
 636         ret = -EINVAL;
 637
 638         bdevname(rdev->bdev, b);
 639         sb = (mdp_super_t*)page_address(rdev->sb_page);
 640
 641         if (sb->md_magic != MD_SB_MAGIC) {
 642                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 643                        b);
 644                 goto abort;
 645         }
 646
 647         if (sb->major_version != 0 ||
 648             sb->minor_version != 90) {
 649                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 650                         sb->major_version, sb->minor_version,
 651                         b);
 652                 goto abort;
 653         }
 654
 655         if (sb->raid_disks <= 0)
 656                 goto abort;
 657
 658         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 659                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 660                         b);
 661                 goto abort;
 662         }
 663
 664         rdev->preferred_minor = sb->md_minor;
 665         rdev->data_offset = 0;
 666         rdev->sb_size = MD_SB_BYTES;
 667
 668         if (sb->level == LEVEL_MULTIPATH)
 669                 rdev->desc_nr = -1;
 670         else
 671                 rdev->desc_nr = sb->this_disk.number;
 672
 673         if (refdev == 0)
 674                 ret = 1;
 675         else {
 676                 __u64 ev1, ev2;
 677                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 678                 if (!uuid_equal(refsb, sb)) {
 679                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 680                                 b, bdevname(refdev->bdev,b2));
 681                         goto abort;
 682                 }
 683                 if (!sb_equal(refsb, sb)) {
 684                         printk(KERN_WARNING "md: %s has same UUID"
 685                                " but different superblock to %s\n",
 686                                b, bdevname(refdev->bdev, b2));
 687                         goto abort;
 688                 }
 689                 ev1 = md_event(sb);
 690                 ev2 = md_event(refsb);
 691                 if (ev1 > ev2)
 692                         ret = 1;
 693                 else
 694                         ret = 0;
 695         }
 696         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 697
 698  abort:
 699         return ret;
 700 }
 701
 702 /*
 703  * validate_super for 0.90.0
 704  */
 705 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 706 {
 707         mdp_disk_t *desc;
 708         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 709
 710         rdev->raid_disk = -1;
 711         rdev->flags = 0;
 712         if (mddev->raid_disks == 0) {
 713                 mddev->major_version = 0;
 714                 mddev->minor_version = sb->minor_version;
 715                 mddev->patch_version = sb->patch_version;
 716                 mddev->persistent = ! sb->not_persistent;
 717                 mddev->chunk_size = sb->chunk_size;
 718                 mddev->ctime = sb->ctime;
 719                 mddev->utime = sb->utime;
 720                 mddev->level = sb->level;
 721                 mddev->clevel[0] = 0;
 722                 mddev->layout = sb->layout;
 723                 mddev->raid_disks = sb->raid_disks;
 724                 mddev->size = sb->size;
 725                 mddev->events = md_event(sb);
 726                 mddev->bitmap_offset = 0;
 727                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 728
 729                 if (sb->state & (1<<MD_SB_CLEAN))
 730                         mddev->recovery_cp = MaxSector;
 731                 else {
 732                         if (sb->events_hi == sb->cp_events_hi &&
 733                                 sb->events_lo == sb->cp_events_lo) {
 734                                 mddev->recovery_cp = sb->recovery_cp;
 735                         } else
 736                                 mddev->recovery_cp = 0;
 737                 }
 738
 739                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 740                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 741                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 742                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 743
 744                 mddev->max_disks = MD_SB_DISKS;
 745
 746                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 747                     mddev->bitmap_file == NULL) {
 748                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
 749                             && mddev->level != 10) {
 750                                 /* FIXME use a better test */
 751                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
 752                                 return -EINVAL;
 753                         }
 754                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 755                 }
 756
 757         } else if (mddev->pers == NULL) {
 758                 /* Insist on good event counter while assembling */
 759                 __u64 ev1 = md_event(sb);
 760                 ++ev1;
 761                 if (ev1 < mddev->events)
 762                         return -EINVAL;
 763         } else if (mddev->bitmap) {
 764                 /* if adding to array with a bitmap, then we can accept an
 765                  * older device ... but not too old.
 766                  */
 767                 __u64 ev1 = md_event(sb);
 768                 if (ev1 < mddev->bitmap->events_cleared)
 769                         return 0;
 770         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 771                 return 0;
 772
 773         if (mddev->level != LEVEL_MULTIPATH) {
 774                 desc = sb->disks + rdev->desc_nr;
 775
 776                 if (desc->state & (1<<MD_DISK_FAULTY))
 777                         set_bit(Faulty, &rdev->flags);
 778                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 779                          desc->raid_disk < mddev->raid_disks) {
 780                         set_bit(In_sync, &rdev->flags);
 781                         rdev->raid_disk = desc->raid_disk;
 782                 }
 783                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 784                         set_bit(WriteMostly, &rdev->flags);
 785         } else /* MULTIPATH are always insync */
 786                 set_bit(In_sync, &rdev->flags);
 787         return 0;
 788 }
 789
 790 /*
 791  * sync_super for 0.90.0
 792  */
 793 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 794 {
 795         mdp_super_t *sb;
 796         struct list_head *tmp;
 797         mdk_rdev_t *rdev2;
 798         int next_spare = mddev->raid_disks;
 799
 800
 801         /* make rdev->sb match mddev data..
 802          *
 803          * 1/ zero out disks
 804          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 805          * 3/ any empty disks < next_spare become removed
 806          *
 807          * disks[0] gets initialised to REMOVED because
 808          * we cannot be sure from other fields if it has
 809          * been initialised or not.
 810          */
 811         int i;
 812         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 813
 814         rdev->sb_size = MD_SB_BYTES;
 815
 816         sb = (mdp_super_t*)page_address(rdev->sb_page);
 817
 818         memset(sb, 0, sizeof(*sb));
 819
 820         sb->md_magic = MD_SB_MAGIC;
 821         sb->major_version = mddev->major_version;
 822         sb->minor_version = mddev->minor_version;
 823         sb->patch_version = mddev->patch_version;
 824         sb->gvalid_words  = 0; /* ignored */
 825         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 826         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 827         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 828         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 829
 830         sb->ctime = mddev->ctime;
 831         sb->level = mddev->level;
 832         sb->size  = mddev->size;
 833         sb->raid_disks = mddev->raid_disks;
 834         sb->md_minor = mddev->md_minor;
 835         sb->not_persistent = !mddev->persistent;
 836         sb->utime = mddev->utime;
 837         sb->state = 0;
 838         sb->events_hi = (mddev->events>>32);
 839         sb->events_lo = (u32)mddev->events;
 840
 841         if (mddev->in_sync)
 842         {
 843                 sb->recovery_cp = mddev->recovery_cp;
 844                 sb->cp_events_hi = (mddev->events>>32);
 845                 sb->cp_events_lo = (u32)mddev->events;
 846                 if (mddev->recovery_cp == MaxSector)
 847                         sb->state = (1<< MD_SB_CLEAN);
 848         } else
 849                 sb->recovery_cp = 0;
 850
 851         sb->layout = mddev->layout;
 852         sb->chunk_size = mddev->chunk_size;
 853
 854         if (mddev->bitmap && mddev->bitmap_file == NULL)
 855                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 856
 857         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 858         ITERATE_RDEV(mddev,rdev2,tmp) {
 859                 mdp_disk_t *d;
 860                 int desc_nr;
 861                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 862                     && !test_bit(Faulty, &rdev2->flags))
 863                         desc_nr = rdev2->raid_disk;
 864                 else
 865                         desc_nr = next_spare++;
 866                 rdev2->desc_nr = desc_nr;
 867                 d = &sb->disks[rdev2->desc_nr];
 868                 nr_disks++;
 869                 d->number = rdev2->desc_nr;
 870                 d->major = MAJOR(rdev2->bdev->bd_dev);
 871                 d->minor = MINOR(rdev2->bdev->bd_dev);
 872                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 873                     && !test_bit(Faulty, &rdev2->flags))
 874                         d->raid_disk = rdev2->raid_disk;
 875                 else
 876                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 877                 if (test_bit(Faulty, &rdev2->flags)) {
 878                         d->state = (1<<MD_DISK_FAULTY);
 879                         failed++;
 880                 } else if (test_bit(In_sync, &rdev2->flags)) {
 881                         d->state = (1<<MD_DISK_ACTIVE);
 882                         d->state |= (1<<MD_DISK_SYNC);
 883                         active++;
 884                         working++;
 885                 } else {
 886                         d->state = 0;
 887                         spare++;
 888                         working++;
 889                 }
 890                 if (test_bit(WriteMostly, &rdev2->flags))
 891                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 892         }
 893         /* now set the "removed" and "faulty" bits on any missing devices */
 894         for (i=0 ; i < mddev->raid_disks ; i++) {
 895                 mdp_disk_t *d = &sb->disks[i];
 896                 if (d->state == 0 && d->number == 0) {
 897                         d->number = i;
 898                         d->raid_disk = i;
 899                         d->state = (1<<MD_DISK_REMOVED);
 900                         d->state |= (1<<MD_DISK_FAULTY);
 901                         failed++;
 902                 }
 903         }
 904         sb->nr_disks = nr_disks;
 905         sb->active_disks = active;
 906         sb->working_disks = working;
 907         sb->failed_disks = failed;
 908         sb->spare_disks = spare;
 909
 910         sb->this_disk = sb->disks[rdev->desc_nr];
 911         sb->sb_csum = calc_sb_csum(sb);
 912 }
 913
 914 /*
 915  * version 1 superblock
 916  */
 917
 918 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 919 {
 920         unsigned int disk_csum, csum;
 921         unsigned long long newcsum;
 922         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 923         unsigned int *isuper = (unsigned int*)sb;
 924         int i;
 925
 926         disk_csum = sb->sb_csum;
 927         sb->sb_csum = 0;
 928         newcsum = 0;
 929         for (i=0; size>=4; size -= 4 )
 930                 newcsum += le32_to_cpu(*isuper++);
 931
 932         if (size == 2)
 933                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 934
 935         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 936         sb->sb_csum = disk_csum;
 937         return cpu_to_le32(csum);
 938 }
 939
 940 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 941 {
 942         struct mdp_superblock_1 *sb;
 943         int ret;
 944         sector_t sb_offset;
 945         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 946         int bmask;
 947
 948         /*
 949          * Calculate the position of the superblock.
 950          * It is always aligned to a 4K boundary and
 951          * depeding on minor_version, it can be:
 952          * 0: At least 8K, but less than 12K, from end of device
 953          * 1: At start of device
 954          * 2: 4K from start of device.
 955          */
 956         switch(minor_version) {
 957         case 0:
 958                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 959                 sb_offset -= 8*2;
 960                 sb_offset &= ~(sector_t)(4*2-1);
 961                 /* convert from sectors to K */
 962                 sb_offset /= 2;
 963                 break;
 964         case 1:
 965                 sb_offset = 0;
 966                 break;
 967         case 2:
 968                 sb_offset = 4;
 969                 break;
 970         default:
 971                 return -EINVAL;
 972         }
 973         rdev->sb_offset = sb_offset;
 974
 975         /* superblock is rarely larger than 1K, but it can be larger,
 976          * and it is safe to read 4k, so we do that
 977          */
 978         ret = read_disk_sb(rdev, 4096);
 979         if (ret) return ret;
 980
 981
 982         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 983
 984         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 985             sb->major_version != cpu_to_le32(1) ||
 986             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 987             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 988             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 989                 return -EINVAL;
 990
 991         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 992                 printk("md: invalid superblock checksum on %s\n",
 993                         bdevname(rdev->bdev,b));
 994                 return -EINVAL;
 995         }
 996         if (le64_to_cpu(sb->data_size) < 10) {
 997                 printk("md: data_size too small on %s\n",
 998                        bdevname(rdev->bdev,b));
 999                 return -EINVAL;
1000         }
1001         rdev->preferred_minor = 0xffff;
1002         rdev->data_offset = le64_to_cpu(sb->data_offset);
1003         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1004
1005         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1006         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1007         if (rdev->sb_size & bmask)
1008                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1009
1010         if (refdev == 0)
1011                 return 1;
1012         else {
1013                 __u64 ev1, ev2;
1014                 struct mdp_superblock_1 *refsb =
1015                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1016
1017                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1018                     sb->level != refsb->level ||
1019                     sb->layout != refsb->layout ||
1020                     sb->chunksize != refsb->chunksize) {
1021                         printk(KERN_WARNING "md: %s has strangely different"
1022                                 " superblock to %s\n",
1023                                 bdevname(rdev->bdev,b),
1024                                 bdevname(refdev->bdev,b2));
1025                         return -EINVAL;
1026                 }
1027                 ev1 = le64_to_cpu(sb->events);
1028                 ev2 = le64_to_cpu(refsb->events);
1029
1030                 if (ev1 > ev2)
1031                         return 1;
1032         }
1033         if (minor_version)
1034                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1035         else
1036                 rdev->size = rdev->sb_offset;
1037         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1038                 return -EINVAL;
1039         rdev->size = le64_to_cpu(sb->data_size)/2;
1040         if (le32_to_cpu(sb->chunksize))
1041                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1042         return 0;
1043 }
1044
1045 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1046 {
1047         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1048
1049         rdev->raid_disk = -1;
1050         rdev->flags = 0;
1051         if (mddev->raid_disks == 0) {
1052                 mddev->major_version = 1;
1053                 mddev->patch_version = 0;
1054                 mddev->persistent = 1;
1055                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1056                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1057                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1058                 mddev->level = le32_to_cpu(sb->level);
1059                 mddev->clevel[0] = 0;
1060                 mddev->layout = le32_to_cpu(sb->layout);
1061                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1062                 mddev->size = le64_to_cpu(sb->size)/2;
1063                 mddev->events = le64_to_cpu(sb->events);
1064                 mddev->bitmap_offset = 0;
1065                 mddev->default_bitmap_offset = 1024;
1066
1067                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1068                 memcpy(mddev->uuid, sb->set_uuid, 16);
1069
1070                 mddev->max_disks =  (4096-256)/2;
1071
1072                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1073                     mddev->bitmap_file == NULL ) {
1074                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1075                             && mddev->level != 10) {
1076                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1077                                 return -EINVAL;
1078                         }
1079                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1080                 }
1081         } else if (mddev->pers == NULL) {
1082                 /* Insist of good event counter while assembling */
1083                 __u64 ev1 = le64_to_cpu(sb->events);
1084                 ++ev1;
1085                 if (ev1 < mddev->events)
1086                         return -EINVAL;
1087         } else if (mddev->bitmap) {
1088                 /* If adding to array with a bitmap, then we can accept an
1089                  * older device, but not too old.
1090                  */
1091                 __u64 ev1 = le64_to_cpu(sb->events);
1092                 if (ev1 < mddev->bitmap->events_cleared)
1093                         return 0;
1094         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1095                 return 0;
1096
1097         if (mddev->level != LEVEL_MULTIPATH) {
1098                 int role;
1099                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1100                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1101                 switch(role) {
1102                 case 0xffff: /* spare */
1103                         break;
1104                 case 0xfffe: /* faulty */
1105                         set_bit(Faulty, &rdev->flags);
1106                         break;
1107                 default:
1108                         set_bit(In_sync, &rdev->flags);
1109                         rdev->raid_disk = role;
1110                         break;
1111                 }
1112                 if (sb->devflags & WriteMostly1)
1113                         set_bit(WriteMostly, &rdev->flags);
1114         } else /* MULTIPATH are always insync */
1115                 set_bit(In_sync, &rdev->flags);
1116
1117         return 0;
1118 }
1119
1120 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1121 {
1122         struct mdp_superblock_1 *sb;
1123         struct list_head *tmp;
1124         mdk_rdev_t *rdev2;
1125         int max_dev, i;
1126         /* make rdev->sb match mddev and rdev data. */
1127
1128         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1129
1130         sb->feature_map = 0;
1131         sb->pad0 = 0;
1132         memset(sb->pad1, 0, sizeof(sb->pad1));
1133         memset(sb->pad2, 0, sizeof(sb->pad2));
1134         memset(sb->pad3, 0, sizeof(sb->pad3));
1135
1136         sb->utime = cpu_to_le64((__u64)mddev->utime);
1137         sb->events = cpu_to_le64(mddev->events);
1138         if (mddev->in_sync)
1139                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1140         else
1141                 sb->resync_offset = cpu_to_le64(0);
1142
1143         sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1144
1145         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1146                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1147                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1148         }
1149
1150         max_dev = 0;
1151         ITERATE_RDEV(mddev,rdev2,tmp)
1152                 if (rdev2->desc_nr+1 > max_dev)
1153                         max_dev = rdev2->desc_nr+1;
1154
1155         sb->max_dev = cpu_to_le32(max_dev);
1156         for (i=0; i<max_dev;i++)
1157                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1158
1159         ITERATE_RDEV(mddev,rdev2,tmp) {
1160                 i = rdev2->desc_nr;
1161                 if (test_bit(Faulty, &rdev2->flags))
1162                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1163                 else if (test_bit(In_sync, &rdev2->flags))
1164                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1165                 else
1166                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1167         }
1168
1169         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1170         sb->sb_csum = calc_sb_1_csum(sb);
1171 }
1172
1173
1174 static struct super_type super_types[] = {
1175         [0] = {
1176                 .name   = "0.90.0",
1177                 .owner  = THIS_MODULE,
1178                 .load_super     = super_90_load,
1179                 .validate_super = super_90_validate,
1180                 .sync_super     = super_90_sync,
1181         },
1182         [1] = {
1183                 .name   = "md-1",
1184                 .owner  = THIS_MODULE,
1185                 .load_super     = super_1_load,
1186                 .validate_super = super_1_validate,
1187                 .sync_super     = super_1_sync,
1188         },
1189 };
1190
1191 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1192 {
1193         struct list_head *tmp;
1194         mdk_rdev_t *rdev;
1195
1196         ITERATE_RDEV(mddev,rdev,tmp)
1197                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1198                         return rdev;
1199
1200         return NULL;
1201 }
1202
1203 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1204 {
1205         struct list_head *tmp;
1206         mdk_rdev_t *rdev;
1207
1208         ITERATE_RDEV(mddev1,rdev,tmp)
1209                 if (match_dev_unit(mddev2, rdev))
1210                         return 1;
1211
1212         return 0;
1213 }
1214
1215 static LIST_HEAD(pending_raid_disks);
1216
1217 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1218 {
1219         mdk_rdev_t *same_pdev;
1220         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1221         struct kobject *ko;
1222
1223         if (rdev->mddev) {
1224                 MD_BUG();
1225                 return -EINVAL;
1226         }
1227         same_pdev = match_dev_unit(mddev, rdev);
1228         if (same_pdev)
1229                 printk(KERN_WARNING
1230                         "%s: WARNING: %s appears to be on the same physical"
1231                         " disk as %s. True\n     protection against single-disk"
1232                         " failure might be compromised.\n",
1233                         mdname(mddev), bdevname(rdev->bdev,b),
1234                         bdevname(same_pdev->bdev,b2));
1235
1236         /* Verify rdev->desc_nr is unique.
1237          * If it is -1, assign a free number, else
1238          * check number is not in use
1239          */
1240         if (rdev->desc_nr < 0) {
1241                 int choice = 0;
1242                 if (mddev->pers) choice = mddev->raid_disks;
1243                 while (find_rdev_nr(mddev, choice))
1244                         choice++;
1245                 rdev->desc_nr = choice;
1246         } else {
1247                 if (find_rdev_nr(mddev, rdev->desc_nr))
1248                         return -EBUSY;
1249         }
1250         bdevname(rdev->bdev,b);
1251         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1252                 return -ENOMEM;
1253
1254         list_add(&rdev->same_set, &mddev->disks);
1255         rdev->mddev = mddev;
1256         printk(KERN_INFO "md: bind<%s>\n", b);
1257
1258         rdev->kobj.parent = &mddev->kobj;
1259         kobject_add(&rdev->kobj);
1260
1261         if (rdev->bdev->bd_part)
1262                 ko = &rdev->bdev->bd_part->kobj;
1263         else
1264                 ko = &rdev->bdev->bd_disk->kobj;
1265         sysfs_create_link(&rdev->kobj, ko, "block");
1266         return 0;
1267 }
1268
1269 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1270 {
1271         char b[BDEVNAME_SIZE];
1272         if (!rdev->mddev) {
1273                 MD_BUG();
1274                 return;
1275         }
1276         list_del_init(&rdev->same_set);
1277         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1278         rdev->mddev = NULL;
1279         sysfs_remove_link(&rdev->kobj, "block");
1280         kobject_del(&rdev->kobj);
1281 }
1282
1283 /*
1284  * prevent the device from being mounted, repartitioned or
1285  * otherwise reused by a RAID array (or any other kernel
1286  * subsystem), by bd_claiming the device.
1287  */
1288 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1289 {
1290         int err = 0;
1291         struct block_device *bdev;
1292         char b[BDEVNAME_SIZE];
1293
1294         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1295         if (IS_ERR(bdev)) {
1296                 printk(KERN_ERR "md: could not open %s.\n",
1297                         __bdevname(dev, b));
1298                 return PTR_ERR(bdev);
1299         }
1300         err = bd_claim(bdev, rdev);
1301         if (err) {
1302                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1303                         bdevname(bdev, b));
1304                 blkdev_put(bdev);
1305                 return err;
1306         }
1307         rdev->bdev = bdev;
1308         return err;
1309 }
1310
1311 static void unlock_rdev(mdk_rdev_t *rdev)
1312 {
1313         struct block_device *bdev = rdev->bdev;
1314         rdev->bdev = NULL;
1315         if (!bdev)
1316                 MD_BUG();
1317         bd_release(bdev);
1318         blkdev_put(bdev);
1319 }
1320
1321 void md_autodetect_dev(dev_t dev);
1322
1323 static void export_rdev(mdk_rdev_t * rdev)
1324 {
1325         char b[BDEVNAME_SIZE];
1326         printk(KERN_INFO "md: export_rdev(%s)\n",
1327                 bdevname(rdev->bdev,b));
1328         if (rdev->mddev)
1329                 MD_BUG();
1330         free_disk_sb(rdev);
1331         list_del_init(&rdev->same_set);
1332 #ifndef MODULE
1333         md_autodetect_dev(rdev->bdev->bd_dev);
1334 #endif
1335         unlock_rdev(rdev);
1336         kobject_put(&rdev->kobj);
1337 }
1338
1339 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1340 {
1341         unbind_rdev_from_array(rdev);
1342         export_rdev(rdev);
1343 }
1344
1345 static void export_array(mddev_t *mddev)
1346 {
1347         struct list_head *tmp;
1348         mdk_rdev_t *rdev;
1349
1350         ITERATE_RDEV(mddev,rdev,tmp) {
1351                 if (!rdev->mddev) {
1352                         MD_BUG();
1353                         continue;
1354                 }
1355                 kick_rdev_from_array(rdev);
1356         }
1357         if (!list_empty(&mddev->disks))
1358                 MD_BUG();
1359         mddev->raid_disks = 0;
1360         mddev->major_version = 0;
1361 }
1362
1363 static void print_desc(mdp_disk_t *desc)
1364 {
1365         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1366                 desc->major,desc->minor,desc->raid_disk,desc->state);
1367 }
1368
1369 static void print_sb(mdp_super_t *sb)
1370 {
1371         int i;
1372
1373         printk(KERN_INFO
1374                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1375                 sb->major_version, sb->minor_version, sb->patch_version,
1376                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1377                 sb->ctime);
1378         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1379                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1380                 sb->md_minor, sb->layout, sb->chunk_size);
1381         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1382                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1383                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1384                 sb->failed_disks, sb->spare_disks,
1385                 sb->sb_csum, (unsigned long)sb->events_lo);
1386
1387         printk(KERN_INFO);
1388         for (i = 0; i < MD_SB_DISKS; i++) {
1389                 mdp_disk_t *desc;
1390
1391                 desc = sb->disks + i;
1392                 if (desc->number || desc->major || desc->minor ||
1393                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1394                         printk("     D %2d: ", i);
1395                         print_desc(desc);
1396                 }
1397         }
1398         printk(KERN_INFO "md:     THIS: ");
1399         print_desc(&sb->this_disk);
1400
1401 }
1402
1403 static void print_rdev(mdk_rdev_t *rdev)
1404 {
1405         char b[BDEVNAME_SIZE];
1406         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1407                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1408                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1409                 rdev->desc_nr);
1410         if (rdev->sb_loaded) {
1411                 printk(KERN_INFO "md: rdev superblock:\n");
1412                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1413         } else
1414                 printk(KERN_INFO "md: no rdev superblock!\n");
1415 }
1416
1417 void md_print_devices(void)
1418 {
1419         struct list_head *tmp, *tmp2;
1420         mdk_rdev_t *rdev;
1421         mddev_t *mddev;
1422         char b[BDEVNAME_SIZE];
1423
1424         printk("\n");
1425         printk("md:     **********************************\n");
1426         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1427         printk("md:     **********************************\n");
1428         ITERATE_MDDEV(mddev,tmp) {
1429
1430                 if (mddev->bitmap)
1431                         bitmap_print_sb(mddev->bitmap);
1432                 else
1433                         printk("%s: ", mdname(mddev));
1434                 ITERATE_RDEV(mddev,rdev,tmp2)
1435                         printk("<%s>", bdevname(rdev->bdev,b));
1436                 printk("\n");
1437
1438                 ITERATE_RDEV(mddev,rdev,tmp2)
1439                         print_rdev(rdev);
1440         }
1441         printk("md:     **********************************\n");
1442         printk("\n");
1443 }
1444
1445
1446 static void sync_sbs(mddev_t * mddev)
1447 {
1448         mdk_rdev_t *rdev;
1449         struct list_head *tmp;
1450
1451         ITERATE_RDEV(mddev,rdev,tmp) {
1452                 super_types[mddev->major_version].
1453                         sync_super(mddev, rdev);
1454                 rdev->sb_loaded = 1;
1455         }
1456 }
1457
1458 static void md_update_sb(mddev_t * mddev)
1459 {
1460         int err;
1461         struct list_head *tmp;
1462         mdk_rdev_t *rdev;
1463         int sync_req;
1464
1465 repeat:
1466         spin_lock_irq(&mddev->write_lock);
1467         sync_req = mddev->in_sync;
1468         mddev->utime = get_seconds();
1469         mddev->events ++;
1470
1471         if (!mddev->events) {
1472                 /*
1473                  * oops, this 64-bit counter should never wrap.
1474                  * Either we are in around ~1 trillion A.C., assuming
1475                  * 1 reboot per second, or we have a bug:
1476                  */
1477                 MD_BUG();
1478                 mddev->events --;
1479         }
1480         mddev->sb_dirty = 2;
1481         sync_sbs(mddev);
1482
1483         /*
1484          * do not write anything to disk if using
1485          * nonpersistent superblocks
1486          */
1487         if (!mddev->persistent) {
1488                 mddev->sb_dirty = 0;
1489                 spin_unlock_irq(&mddev->write_lock);
1490                 wake_up(&mddev->sb_wait);
1491                 return;
1492         }
1493         spin_unlock_irq(&mddev->write_lock);
1494
1495         dprintk(KERN_INFO
1496                 "md: updating %s RAID superblock on device (in sync %d)\n",
1497                 mdname(mddev),mddev->in_sync);
1498
1499         err = bitmap_update_sb(mddev->bitmap);
1500         ITERATE_RDEV(mddev,rdev,tmp) {
1501                 char b[BDEVNAME_SIZE];
1502                 dprintk(KERN_INFO "md: ");
1503                 if (test_bit(Faulty, &rdev->flags))
1504                         dprintk("(skipping faulty ");
1505
1506                 dprintk("%s ", bdevname(rdev->bdev,b));
1507                 if (!test_bit(Faulty, &rdev->flags)) {
1508                         md_super_write(mddev,rdev,
1509                                        rdev->sb_offset<<1, rdev->sb_size,
1510                                        rdev->sb_page);
1511                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1512                                 bdevname(rdev->bdev,b),
1513                                 (unsigned long long)rdev->sb_offset);
1514
1515                 } else
1516                         dprintk(")\n");
1517                 if (mddev->level == LEVEL_MULTIPATH)
1518                         /* only need to write one superblock... */
1519                         break;
1520         }
1521         md_super_wait(mddev);
1522         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1523
1524         spin_lock_irq(&mddev->write_lock);
1525         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1526                 /* have to write it out again */
1527                 spin_unlock_irq(&mddev->write_lock);
1528                 goto repeat;
1529         }
1530         mddev->sb_dirty = 0;
1531         spin_unlock_irq(&mddev->write_lock);
1532         wake_up(&mddev->sb_wait);
1533
1534 }
1535
1536 /* words written to sysfs files may, or my not, be \n terminated.
1537  * We want to accept with case. For this we use cmd_match.
1538  */
1539 static int cmd_match(const char *cmd, const char *str)
1540 {
1541         /* See if cmd, written into a sysfs file, matches
1542          * str.  They must either be the same, or cmd can
1543          * have a trailing newline
1544          */
1545         while (*cmd && *str && *cmd == *str) {
1546                 cmd++;
1547                 str++;
1548         }
1549         if (*cmd == '\n')
1550                 cmd++;
1551         if (*str || *cmd)
1552                 return 0;
1553         return 1;
1554 }
1555
1556 struct rdev_sysfs_entry {
1557         struct attribute attr;
1558         ssize_t (*show)(mdk_rdev_t *, char *);
1559         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1560 };
1561
1562 static ssize_t
1563 state_show(mdk_rdev_t *rdev, char *page)
1564 {
1565         char *sep = "";
1566         int len=0;
1567
1568         if (test_bit(Faulty, &rdev->flags)) {
1569                 len+= sprintf(page+len, "%sfaulty",sep);
1570                 sep = ",";
1571         }
1572         if (test_bit(In_sync, &rdev->flags)) {
1573                 len += sprintf(page+len, "%sin_sync",sep);
1574                 sep = ",";
1575         }
1576         if (!test_bit(Faulty, &rdev->flags) &&
1577             !test_bit(In_sync, &rdev->flags)) {
1578                 len += sprintf(page+len, "%sspare", sep);
1579                 sep = ",";
1580         }
1581         return len+sprintf(page+len, "\n");
1582 }
1583
1584 static struct rdev_sysfs_entry
1585 rdev_state = __ATTR_RO(state);
1586
1587 static ssize_t
1588 super_show(mdk_rdev_t *rdev, char *page)
1589 {
1590         if (rdev->sb_loaded && rdev->sb_size) {
1591                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1592                 return rdev->sb_size;
1593         } else
1594                 return 0;
1595 }
1596 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1597
1598 static ssize_t
1599 errors_show(mdk_rdev_t *rdev, char *page)
1600 {
1601         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1602 }
1603
1604 static ssize_t
1605 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1606 {
1607         char *e;
1608         unsigned long n = simple_strtoul(buf, &e, 10);
1609         if (*buf && (*e == 0 || *e == '\n')) {
1610                 atomic_set(&rdev->corrected_errors, n);
1611                 return len;
1612         }
1613         return -EINVAL;
1614 }
1615 static struct rdev_sysfs_entry rdev_errors =
1616 __ATTR(errors, 0644, errors_show, errors_store);
1617
1618 static struct attribute *rdev_default_attrs[] = {
1619         &rdev_state.attr,
1620         &rdev_super.attr,
1621         &rdev_errors.attr,
1622         NULL,
1623 };
1624 static ssize_t
1625 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1626 {
1627         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1628         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1629
1630         if (!entry->show)
1631                 return -EIO;
1632         return entry->show(rdev, page);
1633 }
1634
1635 static ssize_t
1636 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1637               const char *page, size_t length)
1638 {
1639         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1640         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1641
1642         if (!entry->store)
1643                 return -EIO;
1644         return entry->store(rdev, page, length);
1645 }
1646
1647 static void rdev_free(struct kobject *ko)
1648 {
1649         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1650         kfree(rdev);
1651 }
1652 static struct sysfs_ops rdev_sysfs_ops = {
1653         .show           = rdev_attr_show,
1654         .store          = rdev_attr_store,
1655 };
1656 static struct kobj_type rdev_ktype = {
1657         .release        = rdev_free,
1658         .sysfs_ops      = &rdev_sysfs_ops,
1659         .default_attrs  = rdev_default_attrs,
1660 };
1661
1662 /*
1663  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1664  *
1665  * mark the device faulty if:
1666  *
1667  *   - the device is nonexistent (zero size)
1668  *   - the device has no valid superblock
1669  *
1670  * a faulty rdev _never_ has rdev->sb set.
1671  */
1672 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1673 {
1674         char b[BDEVNAME_SIZE];
1675         int err;
1676         mdk_rdev_t *rdev;
1677         sector_t size;
1678
1679         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1680         if (!rdev) {
1681                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1682                 return ERR_PTR(-ENOMEM);
1683         }
1684
1685         if ((err = alloc_disk_sb(rdev)))
1686                 goto abort_free;
1687
1688         err = lock_rdev(rdev, newdev);
1689         if (err)
1690                 goto abort_free;
1691
1692         rdev->kobj.parent = NULL;
1693         rdev->kobj.ktype = &rdev_ktype;
1694         kobject_init(&rdev->kobj);
1695
1696         rdev->desc_nr = -1;
1697         rdev->flags = 0;
1698         rdev->data_offset = 0;
1699         atomic_set(&rdev->nr_pending, 0);
1700         atomic_set(&rdev->read_errors, 0);
1701         atomic_set(&rdev->corrected_errors, 0);
1702
1703         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1704         if (!size) {
1705                 printk(KERN_WARNING
1706                         "md: %s has zero or unknown size, marking faulty!\n",
1707                         bdevname(rdev->bdev,b));
1708                 err = -EINVAL;
1709                 goto abort_free;
1710         }
1711
1712         if (super_format >= 0) {
1713                 err = super_types[super_format].
1714                         load_super(rdev, NULL, super_minor);
1715                 if (err == -EINVAL) {
1716                         printk(KERN_WARNING
1717                                 "md: %s has invalid sb, not importing!\n",
1718                                 bdevname(rdev->bdev,b));
1719                         goto abort_free;
1720                 }
1721                 if (err < 0) {
1722                         printk(KERN_WARNING
1723                                 "md: could not read %s's sb, not importing!\n",
1724                                 bdevname(rdev->bdev,b));
1725                         goto abort_free;
1726                 }
1727         }
1728         INIT_LIST_HEAD(&rdev->same_set);
1729
1730         return rdev;
1731
1732 abort_free:
1733         if (rdev->sb_page) {
1734                 if (rdev->bdev)
1735                         unlock_rdev(rdev);
1736                 free_disk_sb(rdev);
1737         }
1738         kfree(rdev);
1739         return ERR_PTR(err);
1740 }
1741
1742 /*
1743  * Check a full RAID array for plausibility
1744  */
1745
1746
1747 static void analyze_sbs(mddev_t * mddev)
1748 {
1749         int i;
1750         struct list_head *tmp;
1751         mdk_rdev_t *rdev, *freshest;
1752         char b[BDEVNAME_SIZE];
1753
1754         freshest = NULL;
1755         ITERATE_RDEV(mddev,rdev,tmp)
1756                 switch (super_types[mddev->major_version].
1757                         load_super(rdev, freshest, mddev->minor_version)) {
1758                 case 1:
1759                         freshest = rdev;
1760                         break;
1761                 case 0:
1762                         break;
1763                 default:
1764                         printk( KERN_ERR \
1765                                 "md: fatal superblock inconsistency in %s"
1766                                 " -- removing from array\n",
1767                                 bdevname(rdev->bdev,b));
1768                         kick_rdev_from_array(rdev);
1769                 }
1770
1771
1772         super_types[mddev->major_version].
1773                 validate_super(mddev, freshest);
1774
1775         i = 0;
1776         ITERATE_RDEV(mddev,rdev,tmp) {
1777                 if (rdev != freshest)
1778                         if (super_types[mddev->major_version].
1779                             validate_super(mddev, rdev)) {
1780                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1781                                         " from array!\n",
1782                                         bdevname(rdev->bdev,b));
1783                                 kick_rdev_from_array(rdev);
1784                                 continue;
1785                         }
1786                 if (mddev->level == LEVEL_MULTIPATH) {
1787                         rdev->desc_nr = i++;
1788                         rdev->raid_disk = rdev->desc_nr;
1789                         set_bit(In_sync, &rdev->flags);
1790                 }
1791         }
1792
1793
1794
1795         if (mddev->recovery_cp != MaxSector &&
1796             mddev->level >= 1)
1797                 printk(KERN_ERR "md: %s: raid array is not clean"
1798                        " -- starting background reconstruction\n",
1799                        mdname(mddev));
1800
1801 }
1802
1803 static ssize_t
1804 level_show(mddev_t *mddev, char *page)
1805 {
1806         struct mdk_personality *p = mddev->pers;
1807         if (p)
1808                 return sprintf(page, "%s\n", p->name);
1809         else if (mddev->clevel[0])
1810                 return sprintf(page, "%s\n", mddev->clevel);
1811         else if (mddev->level != LEVEL_NONE)
1812                 return sprintf(page, "%d\n", mddev->level);
1813         else
1814                 return 0;
1815 }
1816
1817 static ssize_t
1818 level_store(mddev_t *mddev, const char *buf, size_t len)
1819 {
1820         int rv = len;
1821         if (mddev->pers)
1822                 return -EBUSY;
1823         if (len == 0)
1824                 return 0;
1825         if (len >= sizeof(mddev->clevel))
1826                 return -ENOSPC;
1827         strncpy(mddev->clevel, buf, len);
1828         if (mddev->clevel[len-1] == '\n')
1829                 len--;
1830         mddev->clevel[len] = 0;
1831         mddev->level = LEVEL_NONE;
1832         return rv;
1833 }
1834
1835 static struct md_sysfs_entry md_level =
1836 __ATTR(level, 0644, level_show, level_store);
1837
1838 static ssize_t
1839 raid_disks_show(mddev_t *mddev, char *page)
1840 {
1841         if (mddev->raid_disks == 0)
1842                 return 0;
1843         return sprintf(page, "%d\n", mddev->raid_disks);
1844 }
1845
1846 static int update_raid_disks(mddev_t *mddev, int raid_disks);
1847
1848 static ssize_t
1849 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
1850 {
1851         /* can only set raid_disks if array is not yet active */
1852         char *e;
1853         int rv = 0;
1854         unsigned long n = simple_strtoul(buf, &e, 10);
1855
1856         if (!*buf || (*e && *e != '\n'))
1857                 return -EINVAL;
1858
1859         if (mddev->pers)
1860                 rv = update_raid_disks(mddev, n);
1861         else
1862                 mddev->raid_disks = n;
1863         return rv ? rv : len;
1864 }
1865 static struct md_sysfs_entry md_raid_disks =
1866 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
1867
1868 static ssize_t
1869 chunk_size_show(mddev_t *mddev, char *page)
1870 {
1871         return sprintf(page, "%d\n", mddev->chunk_size);
1872 }
1873
1874 static ssize_t
1875 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1876 {
1877         /* can only set chunk_size if array is not yet active */
1878         char *e;
1879         unsigned long n = simple_strtoul(buf, &e, 10);
1880
1881         if (mddev->pers)
1882                 return -EBUSY;
1883         if (!*buf || (*e && *e != '\n'))
1884                 return -EINVAL;
1885
1886         mddev->chunk_size = n;
1887         return len;
1888 }
1889 static struct md_sysfs_entry md_chunk_size =
1890 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
1891
1892
1893 static ssize_t
1894 size_show(mddev_t *mddev, char *page)
1895 {
1896         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
1897 }
1898
1899 static int update_size(mddev_t *mddev, unsigned long size);
1900
1901 static ssize_t
1902 size_store(mddev_t *mddev, const char *buf, size_t len)
1903 {
1904         /* If array is inactive, we can reduce the component size, but
1905          * not increase it (except from 0).
1906          * If array is active, we can try an on-line resize
1907          */
1908         char *e;
1909         int err = 0;
1910         unsigned long long size = simple_strtoull(buf, &e, 10);
1911         if (!*buf || *buf == '\n' ||
1912             (*e && *e != '\n'))
1913                 return -EINVAL;
1914
1915         if (mddev->pers) {
1916                 err = update_size(mddev, size);
1917                 md_update_sb(mddev);
1918         } else {
1919                 if (mddev->size == 0 ||
1920                     mddev->size > size)
1921                         mddev->size = size;
1922                 else
1923                         err = -ENOSPC;
1924         }
1925         return err ? err : len;
1926 }
1927
1928 static struct md_sysfs_entry md_size =
1929 __ATTR(component_size, 0644, size_show, size_store);
1930
1931
1932 /* Metdata version.
1933  * This is either 'none' for arrays with externally managed metadata,
1934  * or N.M for internally known formats
1935  */
1936 static ssize_t
1937 metadata_show(mddev_t *mddev, char *page)
1938 {
1939         if (mddev->persistent)
1940                 return sprintf(page, "%d.%d\n",
1941                                mddev->major_version, mddev->minor_version);
1942         else
1943                 return sprintf(page, "none\n");
1944 }
1945
1946 static ssize_t
1947 metadata_store(mddev_t *mddev, const char *buf, size_t len)
1948 {
1949         int major, minor;
1950         char *e;
1951         if (!list_empty(&mddev->disks))
1952                 return -EBUSY;
1953
1954         if (cmd_match(buf, "none")) {
1955                 mddev->persistent = 0;
1956                 mddev->major_version = 0;
1957                 mddev->minor_version = 90;
1958                 return len;
1959         }
1960         major = simple_strtoul(buf, &e, 10);
1961         if (e==buf || *e != '.')
1962                 return -EINVAL;
1963         buf = e+1;
1964         minor = simple_strtoul(buf, &e, 10);
1965         if (e==buf || *e != '\n')
1966                 return -EINVAL;
1967         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
1968             super_types[major].name == NULL)
1969                 return -ENOENT;
1970         mddev->major_version = major;
1971         mddev->minor_version = minor;
1972         mddev->persistent = 1;
1973         return len;
1974 }
1975
1976 static struct md_sysfs_entry md_metadata =
1977 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
1978
1979 static ssize_t
1980 action_show(mddev_t *mddev, char *page)
1981 {
1982         char *type = "idle";
1983         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1984             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1985                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1986                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1987                                 type = "resync";
1988                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1989                                 type = "check";
1990                         else
1991                                 type = "repair";
1992                 } else
1993                         type = "recover";
1994         }
1995         return sprintf(page, "%s\n", type);
1996 }
1997
1998 static ssize_t
1999 action_store(mddev_t *mddev, const char *page, size_t len)
2000 {
2001         if (!mddev->pers || !mddev->pers->sync_request)
2002                 return -EINVAL;
2003
2004         if (cmd_match(page, "idle")) {
2005                 if (mddev->sync_thread) {
2006                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2007                         md_unregister_thread(mddev->sync_thread);
2008                         mddev->sync_thread = NULL;
2009                         mddev->recovery = 0;
2010                 }
2011         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2012                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2013                 return -EBUSY;
2014         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2015                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2016         else {
2017                 if (cmd_match(page, "check"))
2018                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2019                 else if (cmd_match(page, "repair"))
2020                         return -EINVAL;
2021                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2022                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2023         }
2024         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2025         md_wakeup_thread(mddev->thread);
2026         return len;
2027 }
2028
2029 static ssize_t
2030 mismatch_cnt_show(mddev_t *mddev, char *page)
2031 {
2032         return sprintf(page, "%llu\n",
2033                        (unsigned long long) mddev->resync_mismatches);
2034 }
2035
2036 static struct md_sysfs_entry
2037 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2038
2039
2040 static struct md_sysfs_entry
2041 md_mismatches = __ATTR_RO(mismatch_cnt);
2042
2043 static struct attribute *md_default_attrs[] = {
2044         &md_level.attr,
2045         &md_raid_disks.attr,
2046         &md_chunk_size.attr,
2047         &md_size.attr,
2048         &md_metadata.attr,
2049         NULL,
2050 };
2051
2052 static struct attribute *md_redundancy_attrs[] = {
2053         &md_scan_mode.attr,
2054         &md_mismatches.attr,
2055         NULL,
2056 };
2057 static struct attribute_group md_redundancy_group = {
2058         .name = NULL,
2059         .attrs = md_redundancy_attrs,
2060 };
2061
2062
2063 static ssize_t
2064 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2065 {
2066         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2067         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2068         ssize_t rv;
2069
2070         if (!entry->show)
2071                 return -EIO;
2072         mddev_lock(mddev);
2073         rv = entry->show(mddev, page);
2074         mddev_unlock(mddev);
2075         return rv;
2076 }
2077
2078 static ssize_t
2079 md_attr_store(struct kobject *kobj, struct attribute *attr,
2080               const char *page, size_t length)
2081 {
2082         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2083         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2084         ssize_t rv;
2085
2086         if (!entry->store)
2087                 return -EIO;
2088         mddev_lock(mddev);
2089         rv = entry->store(mddev, page, length);
2090         mddev_unlock(mddev);
2091         return rv;
2092 }
2093
2094 static void md_free(struct kobject *ko)
2095 {
2096         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2097         kfree(mddev);
2098 }
2099
2100 static struct sysfs_ops md_sysfs_ops = {
2101         .show   = md_attr_show,
2102         .store  = md_attr_store,
2103 };
2104 static struct kobj_type md_ktype = {
2105         .release        = md_free,
2106         .sysfs_ops      = &md_sysfs_ops,
2107         .default_attrs  = md_default_attrs,
2108 };
2109
2110 int mdp_major = 0;
2111
2112 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2113 {
2114         static DECLARE_MUTEX(disks_sem);
2115         mddev_t *mddev = mddev_find(dev);
2116         struct gendisk *disk;
2117         int partitioned = (MAJOR(dev) != MD_MAJOR);
2118         int shift = partitioned ? MdpMinorShift : 0;
2119         int unit = MINOR(dev) >> shift;
2120
2121         if (!mddev)
2122                 return NULL;
2123
2124         down(&disks_sem);
2125         if (mddev->gendisk) {
2126                 up(&disks_sem);
2127                 mddev_put(mddev);
2128                 return NULL;
2129         }
2130         disk = alloc_disk(1 << shift);
2131         if (!disk) {
2132                 up(&disks_sem);
2133                 mddev_put(mddev);
2134                 return NULL;
2135         }
2136         disk->major = MAJOR(dev);
2137         disk->first_minor = unit << shift;
2138         if (partitioned) {
2139                 sprintf(disk->disk_name, "md_d%d", unit);
2140                 sprintf(disk->devfs_name, "md/d%d", unit);
2141         } else {
2142                 sprintf(disk->disk_name, "md%d", unit);
2143                 sprintf(disk->devfs_name, "md/%d", unit);
2144         }
2145         disk->fops = &md_fops;
2146         disk->private_data = mddev;
2147         disk->queue = mddev->queue;
2148         add_disk(disk);
2149         mddev->gendisk = disk;
2150         up(&disks_sem);
2151         mddev->kobj.parent = &disk->kobj;
2152         mddev->kobj.k_name = NULL;
2153         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2154         mddev->kobj.ktype = &md_ktype;
2155         kobject_register(&mddev->kobj);
2156         return NULL;
2157 }
2158
2159 void md_wakeup_thread(mdk_thread_t *thread);
2160
2161 static void md_safemode_timeout(unsigned long data)
2162 {
2163         mddev_t *mddev = (mddev_t *) data;
2164
2165         mddev->safemode = 1;
2166         md_wakeup_thread(mddev->thread);
2167 }
2168
2169 static int start_dirty_degraded;
2170
2171 static int do_md_run(mddev_t * mddev)
2172 {
2173         int err;
2174         int chunk_size;
2175         struct list_head *tmp;
2176         mdk_rdev_t *rdev;
2177         struct gendisk *disk;
2178         struct mdk_personality *pers;
2179         char b[BDEVNAME_SIZE];
2180
2181         if (list_empty(&mddev->disks))
2182                 /* cannot run an array with no devices.. */
2183                 return -EINVAL;
2184
2185         if (mddev->pers)
2186                 return -EBUSY;
2187
2188         /*
2189          * Analyze all RAID superblock(s)
2190          */
2191         if (!mddev->raid_disks)
2192                 analyze_sbs(mddev);
2193
2194         chunk_size = mddev->chunk_size;
2195
2196         if (chunk_size) {
2197                 if (chunk_size > MAX_CHUNK_SIZE) {
2198                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2199                                 chunk_size, MAX_CHUNK_SIZE);
2200                         return -EINVAL;
2201                 }
2202                 /*
2203                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2204                  */
2205                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2206                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2207                         return -EINVAL;
2208                 }
2209                 if (chunk_size < PAGE_SIZE) {
2210                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2211                                 chunk_size, PAGE_SIZE);
2212                         return -EINVAL;
2213                 }
2214
2215                 /* devices must have minimum size of one chunk */
2216                 ITERATE_RDEV(mddev,rdev,tmp) {
2217                         if (test_bit(Faulty, &rdev->flags))
2218                                 continue;
2219                         if (rdev->size < chunk_size / 1024) {
2220                                 printk(KERN_WARNING
2221                                         "md: Dev %s smaller than chunk_size:"
2222                                         " %lluk < %dk\n",
2223                                         bdevname(rdev->bdev,b),
2224                                         (unsigned long long)rdev->size,
2225                                         chunk_size / 1024);
2226                                 return -EINVAL;
2227                         }
2228                 }
2229         }
2230
2231 #ifdef CONFIG_KMOD
2232         if (mddev->level != LEVEL_NONE)
2233                 request_module("md-level-%d", mddev->level);
2234         else if (mddev->clevel[0])
2235                 request_module("md-%s", mddev->clevel);
2236 #endif
2237
2238         /*
2239          * Drop all container device buffers, from now on
2240          * the only valid external interface is through the md
2241          * device.
2242          * Also find largest hardsector size
2243          */
2244         ITERATE_RDEV(mddev,rdev,tmp) {
2245                 if (test_bit(Faulty, &rdev->flags))
2246                         continue;
2247                 sync_blockdev(rdev->bdev);
2248                 invalidate_bdev(rdev->bdev, 0);
2249         }
2250
2251         md_probe(mddev->unit, NULL, NULL);
2252         disk = mddev->gendisk;
2253         if (!disk)
2254                 return -ENOMEM;
2255
2256         spin_lock(&pers_lock);
2257         pers = find_pers(mddev->level, mddev->clevel);
2258         if (!pers || !try_module_get(pers->owner)) {
2259                 spin_unlock(&pers_lock);
2260                 if (mddev->level != LEVEL_NONE)
2261                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2262                                mddev->level);
2263                 else
2264                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2265                                mddev->clevel);
2266                 return -EINVAL;
2267         }
2268         mddev->pers = pers;
2269         spin_unlock(&pers_lock);
2270         mddev->level = pers->level;
2271         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2272
2273         mddev->recovery = 0;
2274         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2275         mddev->barriers_work = 1;
2276         mddev->ok_start_degraded = start_dirty_degraded;
2277
2278         if (start_readonly)
2279                 mddev->ro = 2; /* read-only, but switch on first write */
2280
2281         err = mddev->pers->run(mddev);
2282         if (!err && mddev->pers->sync_request) {
2283                 err = bitmap_create(mddev);
2284                 if (err) {
2285                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2286                                mdname(mddev), err);
2287                         mddev->pers->stop(mddev);
2288                 }
2289         }
2290         if (err) {
2291                 printk(KERN_ERR "md: pers->run() failed ...\n");
2292                 module_put(mddev->pers->owner);
2293                 mddev->pers = NULL;
2294                 bitmap_destroy(mddev);
2295                 return err;
2296         }
2297         if (mddev->pers->sync_request)
2298                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2299         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2300                 mddev->ro = 0;
2301
2302         atomic_set(&mddev->writes_pending,0);
2303         mddev->safemode = 0;
2304         mddev->safemode_timer.function = md_safemode_timeout;
2305         mddev->safemode_timer.data = (unsigned long) mddev;
2306         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2307         mddev->in_sync = 1;
2308
2309         ITERATE_RDEV(mddev,rdev,tmp)
2310                 if (rdev->raid_disk >= 0) {
2311                         char nm[20];
2312                         sprintf(nm, "rd%d", rdev->raid_disk);
2313                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2314                 }
2315
2316         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2317         md_wakeup_thread(mddev->thread);
2318
2319         if (mddev->sb_dirty)
2320                 md_update_sb(mddev);
2321
2322         set_capacity(disk, mddev->array_size<<1);
2323
2324         /* If we call blk_queue_make_request here, it will
2325          * re-initialise max_sectors etc which may have been
2326          * refined inside -> run.  So just set the bits we need to set.
2327          * Most initialisation happended when we called
2328          * blk_queue_make_request(..., md_fail_request)
2329          * earlier.
2330          */
2331         mddev->queue->queuedata = mddev;
2332         mddev->queue->make_request_fn = mddev->pers->make_request;
2333
2334         mddev->changed = 1;
2335         md_new_event(mddev);
2336         return 0;
2337 }
2338
2339 static int restart_array(mddev_t *mddev)
2340 {
2341         struct gendisk *disk = mddev->gendisk;
2342         int err;
2343
2344         /*
2345          * Complain if it has no devices
2346          */
2347         err = -ENXIO;
2348         if (list_empty(&mddev->disks))
2349                 goto out;
2350
2351         if (mddev->pers) {
2352                 err = -EBUSY;
2353                 if (!mddev->ro)
2354                         goto out;
2355
2356                 mddev->safemode = 0;
2357                 mddev->ro = 0;
2358                 set_disk_ro(disk, 0);
2359
2360                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2361                         mdname(mddev));
2362                 /*
2363                  * Kick recovery or resync if necessary
2364                  */
2365                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2366                 md_wakeup_thread(mddev->thread);
2367                 err = 0;
2368         } else {
2369                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2370                         mdname(mddev));
2371                 err = -EINVAL;
2372         }
2373
2374 out:
2375         return err;
2376 }
2377
2378 static int do_md_stop(mddev_t * mddev, int ro)
2379 {
2380         int err = 0;
2381         struct gendisk *disk = mddev->gendisk;
2382
2383         if (mddev->pers) {
2384                 if (atomic_read(&mddev->active)>2) {
2385                         printk("md: %s still in use.\n",mdname(mddev));
2386                         return -EBUSY;
2387                 }
2388
2389                 if (mddev->sync_thread) {
2390                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2391                         md_unregister_thread(mddev->sync_thread);
2392                         mddev->sync_thread = NULL;
2393                 }
2394
2395                 del_timer_sync(&mddev->safemode_timer);
2396
2397                 invalidate_partition(disk, 0);
2398
2399                 if (ro) {
2400                         err  = -ENXIO;
2401                         if (mddev->ro==1)
2402                                 goto out;
2403                         mddev->ro = 1;
2404                 } else {
2405                         bitmap_flush(mddev);
2406                         md_super_wait(mddev);
2407                         if (mddev->ro)
2408                                 set_disk_ro(disk, 0);
2409                         blk_queue_make_request(mddev->queue, md_fail_request);
2410                         mddev->pers->stop(mddev);
2411                         if (mddev->pers->sync_request)
2412                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2413
2414                         module_put(mddev->pers->owner);
2415                         mddev->pers = NULL;
2416                         if (mddev->ro)
2417                                 mddev->ro = 0;
2418                 }
2419                 if (!mddev->in_sync) {
2420                         /* mark array as shutdown cleanly */
2421                         mddev->in_sync = 1;
2422                         md_update_sb(mddev);
2423                 }
2424                 if (ro)
2425                         set_disk_ro(disk, 1);
2426         }
2427
2428         bitmap_destroy(mddev);
2429         if (mddev->bitmap_file) {
2430                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2431                 fput(mddev->bitmap_file);
2432                 mddev->bitmap_file = NULL;
2433         }
2434         mddev->bitmap_offset = 0;
2435
2436         /*
2437          * Free resources if final stop
2438          */
2439         if (!ro) {
2440                 mdk_rdev_t *rdev;
2441                 struct list_head *tmp;
2442                 struct gendisk *disk;
2443                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2444
2445                 ITERATE_RDEV(mddev,rdev,tmp)
2446                         if (rdev->raid_disk >= 0) {
2447                                 char nm[20];
2448                                 sprintf(nm, "rd%d", rdev->raid_disk);
2449                                 sysfs_remove_link(&mddev->kobj, nm);
2450                         }
2451
2452                 export_array(mddev);
2453
2454                 mddev->array_size = 0;
2455                 disk = mddev->gendisk;
2456                 if (disk)
2457                         set_capacity(disk, 0);
2458                 mddev->changed = 1;
2459         } else
2460                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2461                         mdname(mddev));
2462         err = 0;
2463         md_new_event(mddev);
2464 out:
2465         return err;
2466 }
2467
2468 static void autorun_array(mddev_t *mddev)
2469 {
2470         mdk_rdev_t *rdev;
2471         struct list_head *tmp;
2472         int err;
2473
2474         if (list_empty(&mddev->disks))
2475                 return;
2476
2477         printk(KERN_INFO "md: running: ");
2478
2479         ITERATE_RDEV(mddev,rdev,tmp) {
2480                 char b[BDEVNAME_SIZE];
2481                 printk("<%s>", bdevname(rdev->bdev,b));
2482         }
2483         printk("\n");
2484
2485         err = do_md_run (mddev);
2486         if (err) {
2487                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2488                 do_md_stop (mddev, 0);
2489         }
2490 }
2491
2492 /*
2493  * lets try to run arrays based on all disks that have arrived
2494  * until now. (those are in pending_raid_disks)
2495  *
2496  * the method: pick the first pending disk, collect all disks with
2497  * the same UUID, remove all from the pending list and put them into
2498  * the 'same_array' list. Then order this list based on superblock
2499  * update time (freshest comes first), kick out 'old' disks and
2500  * compare superblocks. If everything's fine then run it.
2501  *
2502  * If "unit" is allocated, then bump its reference count
2503  */
2504 static void autorun_devices(int part)
2505 {
2506         struct list_head candidates;
2507         struct list_head *tmp;
2508         mdk_rdev_t *rdev0, *rdev;
2509         mddev_t *mddev;
2510         char b[BDEVNAME_SIZE];
2511
2512         printk(KERN_INFO "md: autorun ...\n");
2513         while (!list_empty(&pending_raid_disks)) {
2514                 dev_t dev;
2515                 rdev0 = list_entry(pending_raid_disks.next,
2516                                          mdk_rdev_t, same_set);
2517
2518                 printk(KERN_INFO "md: considering %s ...\n",
2519                         bdevname(rdev0->bdev,b));
2520                 INIT_LIST_HEAD(&candidates);
2521                 ITERATE_RDEV_PENDING(rdev,tmp)
2522                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2523                                 printk(KERN_INFO "md:  adding %s ...\n",
2524                                         bdevname(rdev->bdev,b));
2525                                 list_move(&rdev->same_set, &candidates);
2526                         }
2527                 /*
2528                  * now we have a set of devices, with all of them having
2529                  * mostly sane superblocks. It's time to allocate the
2530                  * mddev.
2531                  */
2532                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2533                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2534                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2535                         break;
2536                 }
2537                 if (part)
2538                         dev = MKDEV(mdp_major,
2539                                     rdev0->preferred_minor << MdpMinorShift);
2540                 else
2541                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2542
2543                 md_probe(dev, NULL, NULL);
2544                 mddev = mddev_find(dev);
2545                 if (!mddev) {
2546                         printk(KERN_ERR
2547                                 "md: cannot allocate memory for md drive.\n");
2548                         break;
2549                 }
2550                 if (mddev_lock(mddev))
2551                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2552                                mdname(mddev));
2553                 else if (mddev->raid_disks || mddev->major_version
2554                          || !list_empty(&mddev->disks)) {
2555                         printk(KERN_WARNING
2556                                 "md: %s already running, cannot run %s\n",
2557                                 mdname(mddev), bdevname(rdev0->bdev,b));
2558                         mddev_unlock(mddev);
2559                 } else {
2560                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2561                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2562                                 list_del_init(&rdev->same_set);
2563                                 if (bind_rdev_to_array(rdev, mddev))
2564                                         export_rdev(rdev);
2565                         }
2566                         autorun_array(mddev);
2567                         mddev_unlock(mddev);
2568                 }
2569                 /* on success, candidates will be empty, on error
2570                  * it won't...
2571                  */
2572                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2573                         export_rdev(rdev);
2574                 mddev_put(mddev);
2575         }
2576         printk(KERN_INFO "md: ... autorun DONE.\n");
2577 }
2578
2579 /*
2580  * import RAID devices based on one partition
2581  * if possible, the array gets run as well.
2582  */
2583
2584 static int autostart_array(dev_t startdev)
2585 {
2586         char b[BDEVNAME_SIZE];
2587         int err = -EINVAL, i;
2588         mdp_super_t *sb = NULL;
2589         mdk_rdev_t *start_rdev = NULL, *rdev;
2590
2591         start_rdev = md_import_device(startdev, 0, 0);
2592         if (IS_ERR(start_rdev))
2593                 return err;
2594
2595
2596         /* NOTE: this can only work for 0.90.0 superblocks */
2597         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2598         if (sb->major_version != 0 ||
2599             sb->minor_version != 90 ) {
2600                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2601                 export_rdev(start_rdev);
2602                 return err;
2603         }
2604
2605         if (test_bit(Faulty, &start_rdev->flags)) {
2606                 printk(KERN_WARNING
2607                         "md: can not autostart based on faulty %s!\n",
2608                         bdevname(start_rdev->bdev,b));
2609                 export_rdev(start_rdev);
2610                 return err;
2611         }
2612         list_add(&start_rdev->same_set, &pending_raid_disks);
2613
2614         for (i = 0; i < MD_SB_DISKS; i++) {
2615                 mdp_disk_t *desc = sb->disks + i;
2616                 dev_t dev = MKDEV(desc->major, desc->minor);
2617
2618                 if (!dev)
2619                         continue;
2620                 if (dev == startdev)
2621                         continue;
2622                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2623                         continue;
2624                 rdev = md_import_device(dev, 0, 0);
2625                 if (IS_ERR(rdev))
2626                         continue;
2627
2628                 list_add(&rdev->same_set, &pending_raid_disks);
2629         }
2630
2631         /*
2632          * possibly return codes
2633          */
2634         autorun_devices(0);
2635         return 0;
2636
2637 }
2638
2639
2640 static int get_version(void __user * arg)
2641 {
2642         mdu_version_t ver;
2643
2644         ver.major = MD_MAJOR_VERSION;
2645         ver.minor = MD_MINOR_VERSION;
2646         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2647
2648         if (copy_to_user(arg, &ver, sizeof(ver)))
2649                 return -EFAULT;
2650
2651         return 0;
2652 }
2653
2654 static int get_array_info(mddev_t * mddev, void __user * arg)
2655 {
2656         mdu_array_info_t info;
2657         int nr,working,active,failed,spare;
2658         mdk_rdev_t *rdev;
2659         struct list_head *tmp;
2660
2661         nr=working=active=failed=spare=0;
2662         ITERATE_RDEV(mddev,rdev,tmp) {
2663                 nr++;
2664                 if (test_bit(Faulty, &rdev->flags))
2665                         failed++;
2666                 else {
2667                         working++;
2668                         if (test_bit(In_sync, &rdev->flags))
2669                                 active++;
2670                         else
2671                                 spare++;
2672                 }
2673         }
2674
2675         info.major_version = mddev->major_version;
2676         info.minor_version = mddev->minor_version;
2677         info.patch_version = MD_PATCHLEVEL_VERSION;
2678         info.ctime         = mddev->ctime;
2679         info.level         = mddev->level;
2680         info.size          = mddev->size;
2681         info.nr_disks      = nr;
2682         info.raid_disks    = mddev->raid_disks;
2683         info.md_minor      = mddev->md_minor;
2684         info.not_persistent= !mddev->persistent;
2685
2686         info.utime         = mddev->utime;
2687         info.state         = 0;
2688         if (mddev->in_sync)
2689                 info.state = (1<<MD_SB_CLEAN);
2690         if (mddev->bitmap && mddev->bitmap_offset)
2691                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2692         info.active_disks  = active;
2693         info.working_disks = working;
2694         info.failed_disks  = failed;
2695         info.spare_disks   = spare;
2696
2697         info.layout        = mddev->layout;
2698         info.chunk_size    = mddev->chunk_size;
2699
2700         if (copy_to_user(arg, &info, sizeof(info)))
2701                 return -EFAULT;
2702
2703         return 0;
2704 }
2705
2706 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2707 {
2708         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2709         char *ptr, *buf = NULL;
2710         int err = -ENOMEM;
2711
2712         file = kmalloc(sizeof(*file), GFP_KERNEL);
2713         if (!file)
2714                 goto out;
2715
2716         /* bitmap disabled, zero the first byte and copy out */
2717         if (!mddev->bitmap || !mddev->bitmap->file) {
2718                 file->pathname[0] = '\0';
2719                 goto copy_out;
2720         }
2721
2722         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2723         if (!buf)
2724                 goto out;
2725
2726         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2727         if (!ptr)
2728                 goto out;
2729
2730         strcpy(file->pathname, ptr);
2731
2732 copy_out:
2733         err = 0;
2734         if (copy_to_user(arg, file, sizeof(*file)))
2735                 err = -EFAULT;
2736 out:
2737         kfree(buf);
2738         kfree(file);
2739         return err;
2740 }
2741
2742 static int get_disk_info(mddev_t * mddev, void __user * arg)
2743 {
2744         mdu_disk_info_t info;
2745         unsigned int nr;
2746         mdk_rdev_t *rdev;
2747
2748         if (copy_from_user(&info, arg, sizeof(info)))
2749                 return -EFAULT;
2750
2751         nr = info.number;
2752
2753         rdev = find_rdev_nr(mddev, nr);
2754         if (rdev) {
2755                 info.major = MAJOR(rdev->bdev->bd_dev);
2756                 info.minor = MINOR(rdev->bdev->bd_dev);
2757                 info.raid_disk = rdev->raid_disk;
2758                 info.state = 0;
2759                 if (test_bit(Faulty, &rdev->flags))
2760                         info.state |= (1<<MD_DISK_FAULTY);
2761                 else if (test_bit(In_sync, &rdev->flags)) {
2762                         info.state |= (1<<MD_DISK_ACTIVE);
2763                         info.state |= (1<<MD_DISK_SYNC);
2764                 }
2765                 if (test_bit(WriteMostly, &rdev->flags))
2766                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2767         } else {
2768                 info.major = info.minor = 0;
2769                 info.raid_disk = -1;
2770                 info.state = (1<<MD_DISK_REMOVED);
2771         }
2772
2773         if (copy_to_user(arg, &info, sizeof(info)))
2774                 return -EFAULT;
2775
2776         return 0;
2777 }
2778
2779 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2780 {
2781         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2782         mdk_rdev_t *rdev;
2783         dev_t dev = MKDEV(info->major,info->minor);
2784
2785         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2786                 return -EOVERFLOW;
2787
2788         if (!mddev->raid_disks) {
2789                 int err;
2790                 /* expecting a device which has a superblock */
2791                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2792                 if (IS_ERR(rdev)) {
2793                         printk(KERN_WARNING
2794                                 "md: md_import_device returned %ld\n",
2795                                 PTR_ERR(rdev));
2796                         return PTR_ERR(rdev);
2797                 }
2798                 if (!list_empty(&mddev->disks)) {
2799                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2800                                                         mdk_rdev_t, same_set);
2801                         int err = super_types[mddev->major_version]
2802                                 .load_super(rdev, rdev0, mddev->minor_version);
2803                         if (err < 0) {
2804                                 printk(KERN_WARNING
2805                                         "md: %s has different UUID to %s\n",
2806                                         bdevname(rdev->bdev,b),
2807                                         bdevname(rdev0->bdev,b2));
2808                                 export_rdev(rdev);
2809                                 return -EINVAL;
2810                         }
2811                 }
2812                 err = bind_rdev_to_array(rdev, mddev);
2813                 if (err)
2814                         export_rdev(rdev);
2815                 return err;
2816         }
2817
2818         /*
2819          * add_new_disk can be used once the array is assembled
2820          * to add "hot spares".  They must already have a superblock
2821          * written
2822          */
2823         if (mddev->pers) {
2824                 int err;
2825                 if (!mddev->pers->hot_add_disk) {
2826                         printk(KERN_WARNING
2827                                 "%s: personality does not support diskops!\n",
2828                                mdname(mddev));
2829                         return -EINVAL;
2830                 }
2831                 if (mddev->persistent)
2832                         rdev = md_import_device(dev, mddev->major_version,
2833                                                 mddev->minor_version);
2834                 else
2835                         rdev = md_import_device(dev, -1, -1);
2836                 if (IS_ERR(rdev)) {
2837                         printk(KERN_WARNING
2838                                 "md: md_import_device returned %ld\n",
2839                                 PTR_ERR(rdev));
2840                         return PTR_ERR(rdev);
2841                 }
2842                 /* set save_raid_disk if appropriate */
2843                 if (!mddev->persistent) {
2844                         if (info->state & (1<<MD_DISK_SYNC)  &&
2845                             info->raid_disk < mddev->raid_disks)
2846                                 rdev->raid_disk = info->raid_disk;
2847                         else
2848                                 rdev->raid_disk = -1;
2849                 } else
2850                         super_types[mddev->major_version].
2851                                 validate_super(mddev, rdev);
2852                 rdev->saved_raid_disk = rdev->raid_disk;
2853
2854                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2855                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2856                         set_bit(WriteMostly, &rdev->flags);
2857
2858                 rdev->raid_disk = -1;
2859                 err = bind_rdev_to_array(rdev, mddev);
2860                 if (err)
2861                         export_rdev(rdev);
2862
2863                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2864                 md_wakeup_thread(mddev->thread);
2865                 return err;
2866         }
2867
2868         /* otherwise, add_new_disk is only allowed
2869          * for major_version==0 superblocks
2870          */
2871         if (mddev->major_version != 0) {
2872                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2873                        mdname(mddev));
2874                 return -EINVAL;
2875         }
2876
2877         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2878                 int err;
2879                 rdev = md_import_device (dev, -1, 0);
2880                 if (IS_ERR(rdev)) {
2881                         printk(KERN_WARNING
2882                                 "md: error, md_import_device() returned %ld\n",
2883                                 PTR_ERR(rdev));
2884                         return PTR_ERR(rdev);
2885                 }
2886                 rdev->desc_nr = info->number;
2887                 if (info->raid_disk < mddev->raid_disks)
2888                         rdev->raid_disk = info->raid_disk;
2889                 else
2890                         rdev->raid_disk = -1;
2891
2892                 rdev->flags = 0;
2893
2894                 if (rdev->raid_disk < mddev->raid_disks)
2895                         if (info->state & (1<<MD_DISK_SYNC))
2896                                 set_bit(In_sync, &rdev->flags);
2897
2898                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2899                         set_bit(WriteMostly, &rdev->flags);
2900
2901                 err = bind_rdev_to_array(rdev, mddev);
2902                 if (err) {
2903                         export_rdev(rdev);
2904                         return err;
2905                 }
2906
2907                 if (!mddev->persistent) {
2908                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2909                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2910                 } else
2911                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2912                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2913
2914                 if (!mddev->size || (mddev->size > rdev->size))
2915                         mddev->size = rdev->size;
2916         }
2917
2918         return 0;
2919 }
2920
2921 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2922 {
2923         char b[BDEVNAME_SIZE];
2924         mdk_rdev_t *rdev;
2925
2926         if (!mddev->pers)
2927                 return -ENODEV;
2928
2929         rdev = find_rdev(mddev, dev);
2930         if (!rdev)
2931                 return -ENXIO;
2932
2933         if (rdev->raid_disk >= 0)
2934                 goto busy;
2935
2936         kick_rdev_from_array(rdev);
2937         md_update_sb(mddev);
2938         md_new_event(mddev);
2939
2940         return 0;
2941 busy:
2942         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2943                 bdevname(rdev->bdev,b), mdname(mddev));
2944         return -EBUSY;
2945 }
2946
2947 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2948 {
2949         char b[BDEVNAME_SIZE];
2950         int err;
2951         unsigned int size;
2952         mdk_rdev_t *rdev;
2953
2954         if (!mddev->pers)
2955                 return -ENODEV;
2956
2957         if (mddev->major_version != 0) {
2958                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2959                         " version-0 superblocks.\n",
2960                         mdname(mddev));
2961                 return -EINVAL;
2962         }
2963         if (!mddev->pers->hot_add_disk) {
2964                 printk(KERN_WARNING
2965                         "%s: personality does not support diskops!\n",
2966                         mdname(mddev));
2967                 return -EINVAL;
2968         }
2969
2970         rdev = md_import_device (dev, -1, 0);
2971         if (IS_ERR(rdev)) {
2972                 printk(KERN_WARNING
2973                         "md: error, md_import_device() returned %ld\n",
2974                         PTR_ERR(rdev));
2975                 return -EINVAL;
2976         }
2977
2978         if (mddev->persistent)
2979                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2980         else
2981                 rdev->sb_offset =
2982                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2983
2984         size = calc_dev_size(rdev, mddev->chunk_size);
2985         rdev->size = size;
2986
2987         if (size < mddev->size) {
2988                 printk(KERN_WARNING
2989                         "%s: disk size %llu blocks < array size %llu\n",
2990                         mdname(mddev), (unsigned long long)size,
2991                         (unsigned long long)mddev->size);
2992                 err = -ENOSPC;
2993                 goto abort_export;
2994         }
2995
2996         if (test_bit(Faulty, &rdev->flags)) {
2997                 printk(KERN_WARNING
2998                         "md: can not hot-add faulty %s disk to %s!\n",
2999                         bdevname(rdev->bdev,b), mdname(mddev));
3000                 err = -EINVAL;
3001                 goto abort_export;
3002         }
3003         clear_bit(In_sync, &rdev->flags);
3004         rdev->desc_nr = -1;
3005         bind_rdev_to_array(rdev, mddev);
3006
3007         /*
3008          * The rest should better be atomic, we can have disk failures
3009          * noticed in interrupt contexts ...
3010          */
3011
3012         if (rdev->desc_nr == mddev->max_disks) {
3013                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3014                         mdname(mddev));
3015                 err = -EBUSY;
3016                 goto abort_unbind_export;
3017         }
3018
3019         rdev->raid_disk = -1;
3020
3021         md_update_sb(mddev);
3022
3023         /*
3024          * Kick recovery, maybe this spare has to be added to the
3025          * array immediately.
3026          */
3027         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3028         md_wakeup_thread(mddev->thread);
3029         md_new_event(mddev);
3030         return 0;
3031
3032 abort_unbind_export:
3033         unbind_rdev_from_array(rdev);
3034
3035 abort_export:
3036         export_rdev(rdev);
3037         return err;
3038 }
3039
3040 /* similar to deny_write_access, but accounts for our holding a reference
3041  * to the file ourselves */
3042 static int deny_bitmap_write_access(struct file * file)
3043 {
3044         struct inode *inode = file->f_mapping->host;
3045
3046         spin_lock(&inode->i_lock);
3047         if (atomic_read(&inode->i_writecount) > 1) {
3048                 spin_unlock(&inode->i_lock);
3049                 return -ETXTBSY;
3050         }
3051         atomic_set(&inode->i_writecount, -1);
3052         spin_unlock(&inode->i_lock);
3053
3054         return 0;
3055 }
3056
3057 static int set_bitmap_file(mddev_t *mddev, int fd)
3058 {
3059         int err;
3060
3061         if (mddev->pers) {
3062                 if (!mddev->pers->quiesce)
3063                         return -EBUSY;
3064                 if (mddev->recovery || mddev->sync_thread)
3065                         return -EBUSY;
3066                 /* we should be able to change the bitmap.. */
3067         }
3068
3069
3070         if (fd >= 0) {
3071                 if (mddev->bitmap)
3072                         return -EEXIST; /* cannot add when bitmap is present */
3073                 mddev->bitmap_file = fget(fd);
3074
3075                 if (mddev->bitmap_file == NULL) {
3076                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3077                                mdname(mddev));
3078                         return -EBADF;
3079                 }
3080
3081                 err = deny_bitmap_write_access(mddev->bitmap_file);
3082                 if (err) {
3083                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3084                                mdname(mddev));
3085                         fput(mddev->bitmap_file);
3086                         mddev->bitmap_file = NULL;
3087                         return err;
3088                 }
3089                 mddev->bitmap_offset = 0; /* file overrides offset */
3090         } else if (mddev->bitmap == NULL)
3091                 return -ENOENT; /* cannot remove what isn't there */
3092         err = 0;
3093         if (mddev->pers) {
3094                 mddev->pers->quiesce(mddev, 1);
3095                 if (fd >= 0)
3096                         err = bitmap_create(mddev);
3097                 if (fd < 0 || err)
3098                         bitmap_destroy(mddev);
3099                 mddev->pers->quiesce(mddev, 0);
3100         } else if (fd < 0) {
3101                 if (mddev->bitmap_file)
3102                         fput(mddev->bitmap_file);
3103                 mddev->bitmap_file = NULL;
3104         }
3105
3106         return err;
3107 }
3108
3109 /*
3110  * set_array_info is used two different ways
3111  * The original usage is when creating a new array.
3112  * In this usage, raid_disks is > 0 and it together with
3113  *  level, size, not_persistent,layout,chunksize determine the
3114  *  shape of the array.
3115  *  This will always create an array with a type-0.90.0 superblock.
3116  * The newer usage is when assembling an array.
3117  *  In this case raid_disks will be 0, and the major_version field is
3118  *  use to determine which style super-blocks are to be found on the devices.
3119  *  The minor and patch _version numbers are also kept incase the
3120  *  super_block handler wishes to interpret them.
3121  */
3122 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3123 {
3124
3125         if (info->raid_disks == 0) {
3126                 /* just setting version number for superblock loading */
3127                 if (info->major_version < 0 ||
3128                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3129                     super_types[info->major_version].name == NULL) {
3130                         /* maybe try to auto-load a module? */
3131                         printk(KERN_INFO
3132                                 "md: superblock version %d not known\n",
3133                                 info->major_version);
3134                         return -EINVAL;
3135                 }
3136                 mddev->major_version = info->major_version;
3137                 mddev->minor_version = info->minor_version;
3138                 mddev->patch_version = info->patch_version;
3139                 return 0;
3140         }
3141         mddev->major_version = MD_MAJOR_VERSION;
3142         mddev->minor_version = MD_MINOR_VERSION;
3143         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3144         mddev->ctime         = get_seconds();
3145
3146         mddev->level         = info->level;
3147         mddev->size          = info->size;
3148         mddev->raid_disks    = info->raid_disks;
3149         /* don't set md_minor, it is determined by which /dev/md* was
3150          * openned
3151          */
3152         if (info->state & (1<<MD_SB_CLEAN))
3153                 mddev->recovery_cp = MaxSector;
3154         else
3155                 mddev->recovery_cp = 0;
3156         mddev->persistent    = ! info->not_persistent;
3157
3158         mddev->layout        = info->layout;
3159         mddev->chunk_size    = info->chunk_size;
3160
3161         mddev->max_disks     = MD_SB_DISKS;
3162
3163         mddev->sb_dirty      = 1;
3164
3165         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3166         mddev->bitmap_offset = 0;
3167
3168         /*
3169          * Generate a 128 bit UUID
3170          */
3171         get_random_bytes(mddev->uuid, 16);
3172
3173         return 0;
3174 }
3175
3176 static int update_size(mddev_t *mddev, unsigned long size)
3177 {
3178         mdk_rdev_t * rdev;
3179         int rv;
3180         struct list_head *tmp;
3181
3182         if (mddev->pers->resize == NULL)
3183                 return -EINVAL;
3184         /* The "size" is the amount of each device that is used.
3185          * This can only make sense for arrays with redundancy.
3186          * linear and raid0 always use whatever space is available
3187          * We can only consider changing the size if no resync
3188          * or reconstruction is happening, and if the new size
3189          * is acceptable. It must fit before the sb_offset or,
3190          * if that is <data_offset, it must fit before the
3191          * size of each device.
3192          * If size is zero, we find the largest size that fits.
3193          */
3194         if (mddev->sync_thread)
3195                 return -EBUSY;
3196         ITERATE_RDEV(mddev,rdev,tmp) {
3197                 sector_t avail;
3198                 int fit = (size == 0);
3199                 if (rdev->sb_offset > rdev->data_offset)
3200                         avail = (rdev->sb_offset*2) - rdev->data_offset;
3201                 else
3202                         avail = get_capacity(rdev->bdev->bd_disk)
3203                                 - rdev->data_offset;
3204                 if (fit && (size == 0 || size > avail/2))
3205                         size = avail/2;
3206                 if (avail < ((sector_t)size << 1))
3207                         return -ENOSPC;
3208         }
3209         rv = mddev->pers->resize(mddev, (sector_t)size *2);
3210         if (!rv) {
3211                 struct block_device *bdev;
3212
3213                 bdev = bdget_disk(mddev->gendisk, 0);
3214                 if (bdev) {
3215                         down(&bdev->bd_inode->i_sem);
3216                         i_size_write(bdev->bd_inode, mddev->array_size << 10);
3217                         up(&bdev->bd_inode->i_sem);
3218                         bdput(bdev);
3219                 }
3220         }
3221         return rv;
3222 }
3223
3224 static int update_raid_disks(mddev_t *mddev, int raid_disks)
3225 {
3226         int rv;
3227         /* change the number of raid disks */
3228         if (mddev->pers->reshape == NULL)
3229                 return -EINVAL;
3230         if (raid_disks <= 0 ||
3231             raid_disks >= mddev->max_disks)
3232                 return -EINVAL;
3233         if (mddev->sync_thread)
3234                 return -EBUSY;
3235         rv = mddev->pers->reshape(mddev, raid_disks);
3236         if (!rv) {
3237                 struct block_device *bdev;
3238
3239                 bdev = bdget_disk(mddev->gendisk, 0);
3240                 if (bdev) {
3241                         down(&bdev->bd_inode->i_sem);
3242                         i_size_write(bdev->bd_inode, mddev->array_size << 10);
3243                         up(&bdev->bd_inode->i_sem);
3244                         bdput(bdev);
3245                 }
3246         }
3247         return rv;
3248 }
3249
3250
3251 /*
3252  * update_array_info is used to change the configuration of an
3253  * on-line array.
3254  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3255  * fields in the info are checked against the array.
3256  * Any differences that cannot be handled will cause an error.
3257  * Normally, only one change can be managed at a time.
3258  */
3259 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3260 {
3261         int rv = 0;
3262         int cnt = 0;
3263         int state = 0;
3264
3265         /* calculate expected state,ignoring low bits */
3266         if (mddev->bitmap && mddev->bitmap_offset)
3267                 state |= (1 << MD_SB_BITMAP_PRESENT);
3268
3269         if (mddev->major_version != info->major_version ||
3270             mddev->minor_version != info->minor_version ||
3271 /*          mddev->patch_version != info->patch_version || */
3272             mddev->ctime         != info->ctime         ||
3273             mddev->level         != info->level         ||
3274 /*          mddev->layout        != info->layout        || */
3275             !mddev->persistent   != info->not_persistent||
3276             mddev->chunk_size    != info->chunk_size    ||
3277             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3278             ((state^info->state) & 0xfffffe00)
3279                 )
3280                 return -EINVAL;
3281         /* Check there is only one change */
3282         if (mddev->size != info->size) cnt++;
3283         if (mddev->raid_disks != info->raid_disks) cnt++;
3284         if (mddev->layout != info->layout) cnt++;
3285         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3286         if (cnt == 0) return 0;
3287         if (cnt > 1) return -EINVAL;
3288
3289         if (mddev->layout != info->layout) {
3290                 /* Change layout
3291                  * we don't need to do anything at the md level, the
3292                  * personality will take care of it all.
3293                  */
3294                 if (mddev->pers->reconfig == NULL)
3295                         return -EINVAL;
3296                 else
3297                         return mddev->pers->reconfig(mddev, info->layout, -1);
3298         }
3299         if (mddev->size != info->size)
3300                 rv = update_size(mddev, info->size);
3301
3302         if (mddev->raid_disks    != info->raid_disks)
3303                 rv = update_raid_disks(mddev, info->raid_disks);
3304
3305         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3306                 if (mddev->pers->quiesce == NULL)
3307                         return -EINVAL;
3308                 if (mddev->recovery || mddev->sync_thread)
3309                         return -EBUSY;
3310                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3311                         /* add the bitmap */
3312                         if (mddev->bitmap)
3313                                 return -EEXIST;
3314                         if (mddev->default_bitmap_offset == 0)
3315                                 return -EINVAL;
3316                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3317                         mddev->pers->quiesce(mddev, 1);
3318                         rv = bitmap_create(mddev);
3319                         if (rv)
3320                                 bitmap_destroy(mddev);
3321                         mddev->pers->quiesce(mddev, 0);
3322                 } else {
3323                         /* remove the bitmap */
3324                         if (!mddev->bitmap)
3325                                 return -ENOENT;
3326                         if (mddev->bitmap->file)
3327                                 return -EINVAL;
3328                         mddev->pers->quiesce(mddev, 1);
3329                         bitmap_destroy(mddev);
3330                         mddev->pers->quiesce(mddev, 0);
3331                         mddev->bitmap_offset = 0;
3332                 }
3333         }
3334         md_update_sb(mddev);
3335         return rv;
3336 }
3337
3338 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3339 {
3340         mdk_rdev_t *rdev;
3341
3342         if (mddev->pers == NULL)
3343                 return -ENODEV;
3344
3345         rdev = find_rdev(mddev, dev);
3346         if (!rdev)
3347                 return -ENODEV;
3348
3349         md_error(mddev, rdev);
3350         return 0;
3351 }
3352
3353 static int md_ioctl(struct inode *inode, struct file *file,
3354                         unsigned int cmd, unsigned long arg)
3355 {
3356         int err = 0;
3357         void __user *argp = (void __user *)arg;
3358         struct hd_geometry __user *loc = argp;
3359         mddev_t *mddev = NULL;
3360
3361         if (!capable(CAP_SYS_ADMIN))
3362                 return -EACCES;
3363
3364         /*
3365          * Commands dealing with the RAID driver but not any
3366          * particular array:
3367          */
3368         switch (cmd)
3369         {
3370                 case RAID_VERSION:
3371                         err = get_version(argp);
3372                         goto done;
3373
3374                 case PRINT_RAID_DEBUG:
3375                         err = 0;
3376                         md_print_devices();
3377                         goto done;
3378
3379 #ifndef MODULE
3380                 case RAID_AUTORUN:
3381                         err = 0;
3382                         autostart_arrays(arg);
3383                         goto done;
3384 #endif
3385                 default:;
3386         }
3387
3388         /*
3389          * Commands creating/starting a new array:
3390          */
3391
3392         mddev = inode->i_bdev->bd_disk->private_data;
3393
3394         if (!mddev) {
3395                 BUG();
3396                 goto abort;
3397         }
3398
3399
3400         if (cmd == START_ARRAY) {
3401                 /* START_ARRAY doesn't need to lock the array as autostart_array
3402                  * does the locking, and it could even be a different array
3403                  */
3404                 static int cnt = 3;
3405                 if (cnt > 0 ) {
3406                         printk(KERN_WARNING
3407                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3408                                "This will not be supported beyond July 2006\n",
3409                                current->comm, current->pid);
3410                         cnt--;
3411                 }
3412                 err = autostart_array(new_decode_dev(arg));
3413                 if (err) {
3414                         printk(KERN_WARNING "md: autostart failed!\n");
3415                         goto abort;
3416                 }
3417                 goto done;
3418         }
3419
3420         err = mddev_lock(mddev);
3421         if (err) {
3422                 printk(KERN_INFO
3423                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3424                         err, cmd);
3425                 goto abort;
3426         }
3427
3428         switch (cmd)
3429         {
3430                 case SET_ARRAY_INFO:
3431                         {
3432                                 mdu_array_info_t info;
3433                                 if (!arg)
3434                                         memset(&info, 0, sizeof(info));
3435                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3436                                         err = -EFAULT;
3437                                         goto abort_unlock;
3438                                 }
3439                                 if (mddev->pers) {
3440                                         err = update_array_info(mddev, &info);
3441                                         if (err) {
3442                                                 printk(KERN_WARNING "md: couldn't update"
3443                                                        " array info. %d\n", err);
3444                                                 goto abort_unlock;
3445                                         }
3446                                         goto done_unlock;
3447                                 }
3448                                 if (!list_empty(&mddev->disks)) {
3449                                         printk(KERN_WARNING
3450                                                "md: array %s already has disks!\n",
3451                                                mdname(mddev));
3452                                         err = -EBUSY;
3453                                         goto abort_unlock;
3454                                 }
3455                                 if (mddev->raid_disks) {
3456                                         printk(KERN_WARNING
3457                                                "md: array %s already initialised!\n",
3458                                                mdname(mddev));
3459                                         err = -EBUSY;
3460                                         goto abort_unlock;
3461                                 }
3462                                 err = set_array_info(mddev, &info);
3463                                 if (err) {
3464                                         printk(KERN_WARNING "md: couldn't set"
3465                                                " array info. %d\n", err);
3466                                         goto abort_unlock;
3467                                 }
3468                         }
3469                         goto done_unlock;
3470
3471                 default:;
3472         }
3473
3474         /*
3475          * Commands querying/configuring an existing array:
3476          */
3477         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3478          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3479         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3480                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3481                 err = -ENODEV;
3482                 goto abort_unlock;
3483         }
3484
3485         /*
3486          * Commands even a read-only array can execute:
3487          */
3488         switch (cmd)
3489         {
3490                 case GET_ARRAY_INFO:
3491                         err = get_array_info(mddev, argp);
3492                         goto done_unlock;
3493
3494                 case GET_BITMAP_FILE:
3495                         err = get_bitmap_file(mddev, argp);
3496                         goto done_unlock;
3497
3498                 case GET_DISK_INFO:
3499                         err = get_disk_info(mddev, argp);
3500                         goto done_unlock;
3501
3502                 case RESTART_ARRAY_RW:
3503                         err = restart_array(mddev);
3504                         goto done_unlock;
3505
3506                 case STOP_ARRAY:
3507                         err = do_md_stop (mddev, 0);
3508                         goto done_unlock;
3509
3510                 case STOP_ARRAY_RO:
3511                         err = do_md_stop (mddev, 1);
3512                         goto done_unlock;
3513
3514         /*
3515          * We have a problem here : there is no easy way to give a CHS
3516          * virtual geometry. We currently pretend that we have a 2 heads
3517          * 4 sectors (with a BIG number of cylinders...). This drives
3518          * dosfs just mad... ;-)
3519          */
3520                 case HDIO_GETGEO:
3521                         if (!loc) {
3522                                 err = -EINVAL;
3523                                 goto abort_unlock;
3524                         }
3525                         err = put_user (2, (char __user *) &loc->heads);
3526                         if (err)
3527                                 goto abort_unlock;
3528                         err = put_user (4, (char __user *) &loc->sectors);
3529                         if (err)
3530                                 goto abort_unlock;
3531                         err = put_user(get_capacity(mddev->gendisk)/8,
3532                                         (short __user *) &loc->cylinders);
3533                         if (err)
3534                                 goto abort_unlock;
3535                         err = put_user (get_start_sect(inode->i_bdev),
3536                                                 (long __user *) &loc->start);
3537                         goto done_unlock;
3538         }
3539
3540         /*
3541          * The remaining ioctls are changing the state of the
3542          * superblock, so we do not allow them on read-only arrays.
3543          * However non-MD ioctls (e.g. get-size) will still come through
3544          * here and hit the 'default' below, so only disallow
3545          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3546          */
3547         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3548             mddev->ro && mddev->pers) {
3549                 if (mddev->ro == 2) {
3550                         mddev->ro = 0;
3551                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3552                 md_wakeup_thread(mddev->thread);
3553
3554                 } else {
3555                         err = -EROFS;
3556                         goto abort_unlock;
3557                 }
3558         }
3559
3560         switch (cmd)
3561         {
3562                 case ADD_NEW_DISK:
3563                 {
3564                         mdu_disk_info_t info;
3565                         if (copy_from_user(&info, argp, sizeof(info)))
3566                                 err = -EFAULT;
3567                         else
3568                                 err = add_new_disk(mddev, &info);
3569                         goto done_unlock;
3570                 }
3571
3572                 case HOT_REMOVE_DISK:
3573                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3574                         goto done_unlock;
3575
3576                 case HOT_ADD_DISK:
3577                         err = hot_add_disk(mddev, new_decode_dev(arg));
3578                         goto done_unlock;
3579
3580                 case SET_DISK_FAULTY:
3581                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3582                         goto done_unlock;
3583
3584                 case RUN_ARRAY:
3585                         err = do_md_run (mddev);
3586                         goto done_unlock;
3587
3588                 case SET_BITMAP_FILE:
3589                         err = set_bitmap_file(mddev, (int)arg);
3590                         goto done_unlock;
3591
3592                 default:
3593                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3594                                 printk(KERN_WARNING "md: %s(pid %d) used"
3595                                         " obsolete MD ioctl, upgrade your"
3596                                         " software to use new ictls.\n",
3597                                         current->comm, current->pid);
3598                         err = -EINVAL;
3599                         goto abort_unlock;
3600         }
3601
3602 done_unlock:
3603 abort_unlock:
3604         mddev_unlock(mddev);
3605
3606         return err;
3607 done:
3608         if (err)
3609                 MD_BUG();
3610 abort:
3611         return err;
3612 }
3613
3614 static int md_open(struct inode *inode, struct file *file)
3615 {
3616         /*
3617          * Succeed if we can lock the mddev, which confirms that
3618          * it isn't being stopped right now.
3619          */
3620         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3621         int err;
3622
3623         if ((err = mddev_lock(mddev)))
3624                 goto out;
3625
3626         err = 0;
3627         mddev_get(mddev);
3628         mddev_unlock(mddev);
3629
3630         check_disk_change(inode->i_bdev);
3631  out:
3632         return err;
3633 }
3634
3635 static int md_release(struct inode *inode, struct file * file)
3636 {
3637         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3638
3639         if (!mddev)
3640                 BUG();
3641         mddev_put(mddev);
3642
3643         return 0;
3644 }
3645
3646 static int md_media_changed(struct gendisk *disk)
3647 {
3648         mddev_t *mddev = disk->private_data;
3649
3650         return mddev->changed;
3651 }
3652
3653 static int md_revalidate(struct gendisk *disk)
3654 {
3655         mddev_t *mddev = disk->private_data;
3656
3657         mddev->changed = 0;
3658         return 0;
3659 }
3660 static struct block_device_operations md_fops =
3661 {
3662         .owner          = THIS_MODULE,
3663         .open           = md_open,
3664         .release        = md_release,
3665         .ioctl          = md_ioctl,
3666         .media_changed  = md_media_changed,
3667         .revalidate_disk= md_revalidate,
3668 };
3669
3670 static int md_thread(void * arg)
3671 {
3672         mdk_thread_t *thread = arg;
3673
3674         /*
3675          * md_thread is a 'system-thread', it's priority should be very
3676          * high. We avoid resource deadlocks individually in each
3677          * raid personality. (RAID5 does preallocation) We also use RR and
3678          * the very same RT priority as kswapd, thus we will never get
3679          * into a priority inversion deadlock.
3680          *
3681          * we definitely have to have equal or higher priority than
3682          * bdflush, otherwise bdflush will deadlock if there are too
3683          * many dirty RAID5 blocks.
3684          */
3685
3686         allow_signal(SIGKILL);
3687         while (!kthread_should_stop()) {
3688
3689                 /* We need to wait INTERRUPTIBLE so that
3690                  * we don't add to the load-average.
3691                  * That means we need to be sure no signals are
3692                  * pending
3693                  */
3694                 if (signal_pending(current))
3695                         flush_signals(current);
3696
3697                 wait_event_interruptible_timeout
3698                         (thread->wqueue,
3699                          test_bit(THREAD_WAKEUP, &thread->flags)
3700                          || kthread_should_stop(),
3701                          thread->timeout);
3702                 try_to_freeze();
3703
3704                 clear_bit(THREAD_WAKEUP, &thread->flags);
3705
3706                 thread->run(thread->mddev);
3707         }
3708
3709         return 0;
3710 }
3711
3712 void md_wakeup_thread(mdk_thread_t *thread)
3713 {
3714         if (thread) {
3715                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3716                 set_bit(THREAD_WAKEUP, &thread->flags);
3717                 wake_up(&thread->wqueue);
3718         }
3719 }
3720
3721 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3722                                  const char *name)
3723 {
3724         mdk_thread_t *thread;
3725
3726         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3727         if (!thread)
3728                 return NULL;
3729
3730         init_waitqueue_head(&thread->wqueue);
3731
3732         thread->run = run;
3733         thread->mddev = mddev;
3734         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3735         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3736         if (IS_ERR(thread->tsk)) {
3737                 kfree(thread);
3738                 return NULL;
3739         }
3740         return thread;
3741 }
3742
3743 void md_unregister_thread(mdk_thread_t *thread)
3744 {
3745         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3746
3747         kthread_stop(thread->tsk);
3748         kfree(thread);
3749 }
3750
3751 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3752 {
3753         if (!mddev) {
3754                 MD_BUG();
3755                 return;
3756         }
3757
3758         if (!rdev || test_bit(Faulty, &rdev->flags))
3759                 return;
3760 /*
3761         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3762                 mdname(mddev),
3763                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3764                 __builtin_return_address(0),__builtin_return_address(1),
3765                 __builtin_return_address(2),__builtin_return_address(3));
3766 */
3767         if (!mddev->pers->error_handler)
3768                 return;
3769         mddev->pers->error_handler(mddev,rdev);
3770         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3771         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3772         md_wakeup_thread(mddev->thread);
3773         md_new_event(mddev);
3774 }
3775
3776 /* seq_file implementation /proc/mdstat */
3777
3778 static void status_unused(struct seq_file *seq)
3779 {
3780         int i = 0;
3781         mdk_rdev_t *rdev;
3782         struct list_head *tmp;
3783
3784         seq_printf(seq, "unused devices: ");
3785
3786         ITERATE_RDEV_PENDING(rdev,tmp) {
3787                 char b[BDEVNAME_SIZE];
3788                 i++;
3789                 seq_printf(seq, "%s ",
3790                               bdevname(rdev->bdev,b));
3791         }
3792         if (!i)
3793                 seq_printf(seq, "<none>");
3794
3795         seq_printf(seq, "\n");
3796 }
3797
3798
3799 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3800 {
3801         unsigned long max_blocks, resync, res, dt, db, rt;
3802
3803         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3804
3805         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3806                 max_blocks = mddev->resync_max_sectors >> 1;
3807         else
3808                 max_blocks = mddev->size;
3809
3810         /*
3811          * Should not happen.
3812          */
3813         if (!max_blocks) {
3814                 MD_BUG();
3815                 return;
3816         }
3817         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3818         {
3819                 int i, x = res/50, y = 20-x;
3820                 seq_printf(seq, "[");
3821                 for (i = 0; i < x; i++)
3822                         seq_printf(seq, "=");
3823                 seq_printf(seq, ">");
3824                 for (i = 0; i < y; i++)
3825                         seq_printf(seq, ".");
3826                 seq_printf(seq, "] ");
3827         }
3828         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3829                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3830                        "resync" : "recovery"),
3831                       res/10, res % 10, resync, max_blocks);
3832
3833         /*
3834          * We do not want to overflow, so the order of operands and
3835          * the * 100 / 100 trick are important. We do a +1 to be
3836          * safe against division by zero. We only estimate anyway.
3837          *
3838          * dt: time from mark until now
3839          * db: blocks written from mark until now
3840          * rt: remaining time
3841          */
3842         dt = ((jiffies - mddev->resync_mark) / HZ);
3843         if (!dt) dt++;
3844         db = resync - (mddev->resync_mark_cnt/2);
3845         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3846
3847         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3848
3849         seq_printf(seq, " speed=%ldK/sec", db/dt);
3850 }
3851
3852 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3853 {
3854         struct list_head *tmp;
3855         loff_t l = *pos;
3856         mddev_t *mddev;
3857
3858         if (l >= 0x10000)
3859                 return NULL;
3860         if (!l--)
3861                 /* header */
3862                 return (void*)1;
3863
3864         spin_lock(&all_mddevs_lock);
3865         list_for_each(tmp,&all_mddevs)
3866                 if (!l--) {
3867                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3868                         mddev_get(mddev);
3869                         spin_unlock(&all_mddevs_lock);
3870                         return mddev;
3871                 }
3872         spin_unlock(&all_mddevs_lock);
3873         if (!l--)
3874                 return (void*)2;/* tail */
3875         return NULL;
3876 }
3877
3878 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3879 {
3880         struct list_head *tmp;
3881         mddev_t *next_mddev, *mddev = v;
3882
3883         ++*pos;
3884         if (v == (void*)2)
3885                 return NULL;
3886
3887         spin_lock(&all_mddevs_lock);
3888         if (v == (void*)1)
3889                 tmp = all_mddevs.next;
3890         else
3891                 tmp = mddev->all_mddevs.next;
3892         if (tmp != &all_mddevs)
3893                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3894         else {
3895                 next_mddev = (void*)2;
3896                 *pos = 0x10000;
3897         }
3898         spin_unlock(&all_mddevs_lock);
3899
3900         if (v != (void*)1)
3901                 mddev_put(mddev);
3902         return next_mddev;
3903
3904 }
3905
3906 static void md_seq_stop(struct seq_file *seq, void *v)
3907 {
3908         mddev_t *mddev = v;
3909
3910         if (mddev && v != (void*)1 && v != (void*)2)
3911                 mddev_put(mddev);
3912 }
3913
3914 struct mdstat_info {
3915         int event;
3916 };
3917
3918 static int md_seq_show(struct seq_file *seq, void *v)
3919 {
3920         mddev_t *mddev = v;
3921         sector_t size;
3922         struct list_head *tmp2;
3923         mdk_rdev_t *rdev;
3924         struct mdstat_info *mi = seq->private;
3925         struct bitmap *bitmap;
3926
3927         if (v == (void*)1) {
3928                 struct mdk_personality *pers;
3929                 seq_printf(seq, "Personalities : ");
3930                 spin_lock(&pers_lock);
3931                 list_for_each_entry(pers, &pers_list, list)
3932                         seq_printf(seq, "[%s] ", pers->name);
3933
3934                 spin_unlock(&pers_lock);
3935                 seq_printf(seq, "\n");
3936                 mi->event = atomic_read(&md_event_count);
3937                 return 0;
3938         }
3939         if (v == (void*)2) {
3940                 status_unused(seq);
3941                 return 0;
3942         }
3943
3944         if (mddev_lock(mddev)!=0)
3945                 return -EINTR;
3946         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3947                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3948                                                 mddev->pers ? "" : "in");
3949                 if (mddev->pers) {
3950                         if (mddev->ro==1)
3951                                 seq_printf(seq, " (read-only)");
3952                         if (mddev->ro==2)
3953                                 seq_printf(seq, "(auto-read-only)");
3954                         seq_printf(seq, " %s", mddev->pers->name);
3955                 }
3956
3957                 size = 0;
3958                 ITERATE_RDEV(mddev,rdev,tmp2) {
3959                         char b[BDEVNAME_SIZE];
3960                         seq_printf(seq, " %s[%d]",
3961                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3962                         if (test_bit(WriteMostly, &rdev->flags))
3963                                 seq_printf(seq, "(W)");
3964                         if (test_bit(Faulty, &rdev->flags)) {
3965                                 seq_printf(seq, "(F)");
3966                                 continue;
3967                         } else if (rdev->raid_disk < 0)
3968                                 seq_printf(seq, "(S)"); /* spare */
3969                         size += rdev->size;
3970                 }
3971
3972                 if (!list_empty(&mddev->disks)) {
3973                         if (mddev->pers)
3974                                 seq_printf(seq, "\n      %llu blocks",
3975                                         (unsigned long long)mddev->array_size);
3976                         else
3977                                 seq_printf(seq, "\n      %llu blocks",
3978                                         (unsigned long long)size);
3979                 }
3980                 if (mddev->persistent) {
3981                         if (mddev->major_version != 0 ||
3982                             mddev->minor_version != 90) {
3983                                 seq_printf(seq," super %d.%d",
3984                                            mddev->major_version,
3985                                            mddev->minor_version);
3986                         }
3987                 } else
3988                         seq_printf(seq, " super non-persistent");
3989
3990                 if (mddev->pers) {
3991                         mddev->pers->status (seq, mddev);
3992                         seq_printf(seq, "\n      ");
3993                         if (mddev->pers->sync_request) {
3994                                 if (mddev->curr_resync > 2) {
3995                                         status_resync (seq, mddev);
3996                                         seq_printf(seq, "\n      ");
3997                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3998                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3999                                 else if (mddev->recovery_cp < MaxSector)
4000                                         seq_printf(seq, "\tresync=PENDING\n      ");
4001                         }
4002                 } else
4003                         seq_printf(seq, "\n       ");
4004
4005                 if ((bitmap = mddev->bitmap)) {
4006                         unsigned long chunk_kb;
4007                         unsigned long flags;
4008                         spin_lock_irqsave(&bitmap->lock, flags);
4009                         chunk_kb = bitmap->chunksize >> 10;
4010                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4011                                 "%lu%s chunk",
4012                                 bitmap->pages - bitmap->missing_pages,
4013                                 bitmap->pages,
4014                                 (bitmap->pages - bitmap->missing_pages)
4015                                         << (PAGE_SHIFT - 10),
4016                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4017                                 chunk_kb ? "KB" : "B");
4018                         if (bitmap->file) {
4019                                 seq_printf(seq, ", file: ");
4020                                 seq_path(seq, bitmap->file->f_vfsmnt,
4021                                          bitmap->file->f_dentry," \t\n");
4022                         }
4023
4024                         seq_printf(seq, "\n");
4025                         spin_unlock_irqrestore(&bitmap->lock, flags);
4026                 }
4027
4028                 seq_printf(seq, "\n");
4029         }
4030         mddev_unlock(mddev);
4031
4032         return 0;
4033 }
4034
4035 static struct seq_operations md_seq_ops = {
4036         .start  = md_seq_start,
4037         .next   = md_seq_next,
4038         .stop   = md_seq_stop,
4039         .show   = md_seq_show,
4040 };
4041
4042 static int md_seq_open(struct inode *inode, struct file *file)
4043 {
4044         int error;
4045         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4046         if (mi == NULL)
4047                 return -ENOMEM;
4048
4049         error = seq_open(file, &md_seq_ops);
4050         if (error)
4051                 kfree(mi);
4052         else {
4053                 struct seq_file *p = file->private_data;
4054                 p->private = mi;
4055                 mi->event = atomic_read(&md_event_count);
4056         }
4057         return error;
4058 }
4059
4060 static int md_seq_release(struct inode *inode, struct file *file)
4061 {
4062         struct seq_file *m = file->private_data;
4063         struct mdstat_info *mi = m->private;
4064         m->private = NULL;
4065         kfree(mi);
4066         return seq_release(inode, file);
4067 }
4068
4069 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4070 {
4071         struct seq_file *m = filp->private_data;
4072         struct mdstat_info *mi = m->private;
4073         int mask;
4074
4075         poll_wait(filp, &md_event_waiters, wait);
4076
4077         /* always allow read */
4078         mask = POLLIN | POLLRDNORM;
4079
4080         if (mi->event != atomic_read(&md_event_count))
4081                 mask |= POLLERR | POLLPRI;
4082         return mask;
4083 }
4084
4085 static struct file_operations md_seq_fops = {
4086         .open           = md_seq_open,
4087         .read           = seq_read,
4088         .llseek         = seq_lseek,
4089         .release        = md_seq_release,
4090         .poll           = mdstat_poll,
4091 };
4092
4093 int register_md_personality(struct mdk_personality *p)
4094 {
4095         spin_lock(&pers_lock);
4096         list_add_tail(&p->list, &pers_list);
4097         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4098         spin_unlock(&pers_lock);
4099         return 0;
4100 }
4101
4102 int unregister_md_personality(struct mdk_personality *p)
4103 {
4104         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4105         spin_lock(&pers_lock);
4106         list_del_init(&p->list);
4107         spin_unlock(&pers_lock);
4108         return 0;
4109 }
4110
4111 static int is_mddev_idle(mddev_t *mddev)
4112 {
4113         mdk_rdev_t * rdev;
4114         struct list_head *tmp;
4115         int idle;
4116         unsigned long curr_events;
4117
4118         idle = 1;
4119         ITERATE_RDEV(mddev,rdev,tmp) {
4120                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4121                 curr_events = disk_stat_read(disk, sectors[0]) +
4122                                 disk_stat_read(disk, sectors[1]) -
4123                                 atomic_read(&disk->sync_io);
4124                 /* The difference between curr_events and last_events
4125                  * will be affected by any new non-sync IO (making
4126                  * curr_events bigger) and any difference in the amount of
4127                  * in-flight syncio (making current_events bigger or smaller)
4128                  * The amount in-flight is currently limited to
4129                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4130                  * which is at most 4096 sectors.
4131                  * These numbers are fairly fragile and should be made
4132                  * more robust, probably by enforcing the
4133                  * 'window size' that md_do_sync sort-of uses.
4134                  *
4135                  * Note: the following is an unsigned comparison.
4136                  */
4137                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4138                         rdev->last_events = curr_events;
4139                         idle = 0;
4140                 }
4141         }
4142         return idle;
4143 }
4144
4145 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4146 {
4147         /* another "blocks" (512byte) blocks have been synced */
4148         atomic_sub(blocks, &mddev->recovery_active);
4149         wake_up(&mddev->recovery_wait);
4150         if (!ok) {
4151                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4152                 md_wakeup_thread(mddev->thread);
4153                 // stop recovery, signal do_sync ....
4154         }
4155 }
4156
4157
4158 /* md_write_start(mddev, bi)
4159  * If we need to update some array metadata (e.g. 'active' flag
4160  * in superblock) before writing, schedule a superblock update
4161  * and wait for it to complete.
4162  */
4163 void md_write_start(mddev_t *mddev, struct bio *bi)
4164 {
4165         if (bio_data_dir(bi) != WRITE)
4166                 return;
4167
4168         BUG_ON(mddev->ro == 1);
4169         if (mddev->ro == 2) {
4170                 /* need to switch to read/write */
4171                 mddev->ro = 0;
4172                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4173                 md_wakeup_thread(mddev->thread);
4174         }
4175         atomic_inc(&mddev->writes_pending);
4176         if (mddev->in_sync) {
4177                 spin_lock_irq(&mddev->write_lock);
4178                 if (mddev->in_sync) {
4179                         mddev->in_sync = 0;
4180                         mddev->sb_dirty = 1;
4181                         md_wakeup_thread(mddev->thread);
4182                 }
4183                 spin_unlock_irq(&mddev->write_lock);
4184         }
4185         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4186 }
4187
4188 void md_write_end(mddev_t *mddev)
4189 {
4190         if (atomic_dec_and_test(&mddev->writes_pending)) {
4191                 if (mddev->safemode == 2)
4192                         md_wakeup_thread(mddev->thread);
4193                 else
4194                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4195         }
4196 }
4197
4198 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4199
4200 #define SYNC_MARKS      10
4201 #define SYNC_MARK_STEP  (3*HZ)
4202 static void md_do_sync(mddev_t *mddev)
4203 {
4204         mddev_t *mddev2;
4205         unsigned int currspeed = 0,
4206                  window;
4207         sector_t max_sectors,j, io_sectors;
4208         unsigned long mark[SYNC_MARKS];
4209         sector_t mark_cnt[SYNC_MARKS];
4210         int last_mark,m;
4211         struct list_head *tmp;
4212         sector_t last_check;
4213         int skipped = 0;
4214
4215         /* just incase thread restarts... */
4216         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4217                 return;
4218
4219         /* we overload curr_resync somewhat here.
4220          * 0 == not engaged in resync at all
4221          * 2 == checking that there is no conflict with another sync
4222          * 1 == like 2, but have yielded to allow conflicting resync to
4223          *              commense
4224          * other == active in resync - this many blocks
4225          *
4226          * Before starting a resync we must have set curr_resync to
4227          * 2, and then checked that every "conflicting" array has curr_resync
4228          * less than ours.  When we find one that is the same or higher
4229          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4230          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4231          * This will mean we have to start checking from the beginning again.
4232          *
4233          */
4234
4235         do {
4236                 mddev->curr_resync = 2;
4237
4238         try_again:
4239                 if (kthread_should_stop()) {
4240                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4241                         goto skip;
4242                 }
4243                 ITERATE_MDDEV(mddev2,tmp) {
4244                         if (mddev2 == mddev)
4245                                 continue;
4246                         if (mddev2->curr_resync &&
4247                             match_mddev_units(mddev,mddev2)) {
4248                                 DEFINE_WAIT(wq);
4249                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4250                                         /* arbitrarily yield */
4251                                         mddev->curr_resync = 1;
4252                                         wake_up(&resync_wait);
4253                                 }
4254                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4255                                         /* no need to wait here, we can wait the next
4256                                          * time 'round when curr_resync == 2
4257                                          */
4258                                         continue;
4259                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4260                                 if (!kthread_should_stop() &&
4261                                     mddev2->curr_resync >= mddev->curr_resync) {
4262                                         printk(KERN_INFO "md: delaying resync of %s"
4263                                                " until %s has finished resync (they"
4264                                                " share one or more physical units)\n",
4265                                                mdname(mddev), mdname(mddev2));
4266                                         mddev_put(mddev2);
4267                                         schedule();
4268                                         finish_wait(&resync_wait, &wq);
4269                                         goto try_again;
4270                                 }
4271                                 finish_wait(&resync_wait, &wq);
4272                         }
4273                 }
4274         } while (mddev->curr_resync < 2);
4275
4276         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4277                 /* resync follows the size requested by the personality,
4278                  * which defaults to physical size, but can be virtual size
4279                  */
4280                 max_sectors = mddev->resync_max_sectors;
4281                 mddev->resync_mismatches = 0;
4282         } else
4283                 /* recovery follows the physical size of devices */
4284                 max_sectors = mddev->size << 1;
4285
4286         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4287         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4288                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4289         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4290                "(but not more than %d KB/sec) for reconstruction.\n",
4291                sysctl_speed_limit_max);
4292
4293         is_mddev_idle(mddev); /* this also initializes IO event counters */
4294         /* we don't use the checkpoint if there's a bitmap */
4295         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4296             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4297                 j = mddev->recovery_cp;
4298         else
4299                 j = 0;
4300         io_sectors = 0;
4301         for (m = 0; m < SYNC_MARKS; m++) {
4302                 mark[m] = jiffies;
4303                 mark_cnt[m] = io_sectors;
4304         }
4305         last_mark = 0;
4306         mddev->resync_mark = mark[last_mark];
4307         mddev->resync_mark_cnt = mark_cnt[last_mark];
4308
4309         /*
4310          * Tune reconstruction:
4311          */
4312         window = 32*(PAGE_SIZE/512);
4313         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4314                 window/2,(unsigned long long) max_sectors/2);
4315
4316         atomic_set(&mddev->recovery_active, 0);
4317         init_waitqueue_head(&mddev->recovery_wait);
4318         last_check = 0;
4319
4320         if (j>2) {
4321                 printk(KERN_INFO
4322                         "md: resuming recovery of %s from checkpoint.\n",
4323                         mdname(mddev));
4324                 mddev->curr_resync = j;
4325         }
4326
4327         while (j < max_sectors) {
4328                 sector_t sectors;
4329
4330                 skipped = 0;
4331                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4332                                             currspeed < sysctl_speed_limit_min);
4333                 if (sectors == 0) {
4334                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4335                         goto out;
4336                 }
4337
4338                 if (!skipped) { /* actual IO requested */
4339                         io_sectors += sectors;
4340                         atomic_add(sectors, &mddev->recovery_active);
4341                 }
4342
4343                 j += sectors;
4344                 if (j>1) mddev->curr_resync = j;
4345                 if (last_check == 0)
4346                         /* this is the earliers that rebuilt will be
4347                          * visible in /proc/mdstat
4348                          */
4349                         md_new_event(mddev);
4350
4351                 if (last_check + window > io_sectors || j == max_sectors)
4352                         continue;
4353
4354                 last_check = io_sectors;
4355
4356                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4357                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4358                         break;
4359
4360         repeat:
4361                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4362                         /* step marks */
4363                         int next = (last_mark+1) % SYNC_MARKS;
4364
4365                         mddev->resync_mark = mark[next];
4366                         mddev->resync_mark_cnt = mark_cnt[next];
4367                         mark[next] = jiffies;
4368                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4369                         last_mark = next;
4370                 }
4371
4372
4373                 if (kthread_should_stop()) {
4374                         /*
4375                          * got a signal, exit.
4376                          */
4377                         printk(KERN_INFO
4378                                 "md: md_do_sync() got signal ... exiting\n");
4379                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4380                         goto out;
4381                 }
4382
4383                 /*
4384                  * this loop exits only if either when we are slower than
4385                  * the 'hard' speed limit, or the system was IO-idle for
4386                  * a jiffy.
4387                  * the system might be non-idle CPU-wise, but we only care
4388                  * about not overloading the IO subsystem. (things like an
4389                  * e2fsck being done on the RAID array should execute fast)
4390                  */
4391                 mddev->queue->unplug_fn(mddev->queue);
4392                 cond_resched();
4393
4394                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4395                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4396
4397                 if (currspeed > sysctl_speed_limit_min) {
4398                         if ((currspeed > sysctl_speed_limit_max) ||
4399                                         !is_mddev_idle(mddev)) {
4400                                 msleep(500);
4401                                 goto repeat;
4402                         }
4403                 }
4404         }
4405         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4406         /*
4407          * this also signals 'finished resyncing' to md_stop
4408          */
4409  out:
4410         mddev->queue->unplug_fn(mddev->queue);
4411
4412         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4413
4414         /* tell personality that we are finished */
4415         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4416
4417         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4418             mddev->curr_resync > 2 &&
4419             mddev->curr_resync >= mddev->recovery_cp) {
4420                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4421                         printk(KERN_INFO
4422                                 "md: checkpointing recovery of %s.\n",
4423                                 mdname(mddev));
4424                         mddev->recovery_cp = mddev->curr_resync;
4425                 } else
4426                         mddev->recovery_cp = MaxSector;
4427         }
4428
4429  skip:
4430         mddev->curr_resync = 0;
4431         wake_up(&resync_wait);
4432         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4433         md_wakeup_thread(mddev->thread);
4434 }
4435
4436
4437 /*
4438  * This routine is regularly called by all per-raid-array threads to
4439  * deal with generic issues like resync and super-block update.
4440  * Raid personalities that don't have a thread (linear/raid0) do not
4441  * need this as they never do any recovery or update the superblock.
4442  *
4443  * It does not do any resync itself, but rather "forks" off other threads
4444  * to do that as needed.
4445  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4446  * "->recovery" and create a thread at ->sync_thread.
4447  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4448  * and wakeups up this thread which will reap the thread and finish up.
4449  * This thread also removes any faulty devices (with nr_pending == 0).
4450  *
4451  * The overall approach is:
4452  *  1/ if the superblock needs updating, update it.
4453  *  2/ If a recovery thread is running, don't do anything else.
4454  *  3/ If recovery has finished, clean up, possibly marking spares active.
4455  *  4/ If there are any faulty devices, remove them.
4456  *  5/ If array is degraded, try to add spares devices
4457  *  6/ If array has spares or is not in-sync, start a resync thread.
4458  */
4459 void md_check_recovery(mddev_t *mddev)
4460 {
4461         mdk_rdev_t *rdev;
4462         struct list_head *rtmp;
4463
4464
4465         if (mddev->bitmap)
4466                 bitmap_daemon_work(mddev->bitmap);
4467
4468         if (mddev->ro)
4469                 return;
4470
4471         if (signal_pending(current)) {
4472                 if (mddev->pers->sync_request) {
4473                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4474                                mdname(mddev));
4475                         mddev->safemode = 2;
4476                 }
4477                 flush_signals(current);
4478         }
4479
4480         if ( ! (
4481                 mddev->sb_dirty ||
4482                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4483                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4484                 (mddev->safemode == 1) ||
4485                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4486                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4487                 ))
4488                 return;
4489
4490         if (mddev_trylock(mddev)==0) {
4491                 int spares =0;
4492
4493                 spin_lock_irq(&mddev->write_lock);
4494                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4495                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4496                         mddev->in_sync = 1;
4497                         mddev->sb_dirty = 1;
4498                 }
4499                 if (mddev->safemode == 1)
4500                         mddev->safemode = 0;
4501                 spin_unlock_irq(&mddev->write_lock);
4502
4503                 if (mddev->sb_dirty)
4504                         md_update_sb(mddev);
4505
4506
4507                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4508                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4509                         /* resync/recovery still happening */
4510                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4511                         goto unlock;
4512                 }
4513                 if (mddev->sync_thread) {
4514                         /* resync has finished, collect result */
4515                         md_unregister_thread(mddev->sync_thread);
4516                         mddev->sync_thread = NULL;
4517                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4518                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4519                                 /* success...*/
4520                                 /* activate any spares */
4521                                 mddev->pers->spare_active(mddev);
4522                         }
4523                         md_update_sb(mddev);
4524
4525                         /* if array is no-longer degraded, then any saved_raid_disk
4526                          * information must be scrapped
4527                          */
4528                         if (!mddev->degraded)
4529                                 ITERATE_RDEV(mddev,rdev,rtmp)
4530                                         rdev->saved_raid_disk = -1;
4531
4532                         mddev->recovery = 0;
4533                         /* flag recovery needed just to double check */
4534                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4535                         md_new_event(mddev);
4536                         goto unlock;
4537                 }
4538                 /* Clear some bits that don't mean anything, but
4539                  * might be left set
4540                  */
4541                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4542                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4543                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4544                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4545
4546                 /* no recovery is running.
4547                  * remove any failed drives, then
4548                  * add spares if possible.
4549                  * Spare are also removed and re-added, to allow
4550                  * the personality to fail the re-add.
4551                  */
4552                 ITERATE_RDEV(mddev,rdev,rtmp)
4553                         if (rdev->raid_disk >= 0 &&
4554                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4555                             atomic_read(&rdev->nr_pending)==0) {
4556                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4557                                         char nm[20];
4558                                         sprintf(nm,"rd%d", rdev->raid_disk);
4559                                         sysfs_remove_link(&mddev->kobj, nm);
4560                                         rdev->raid_disk = -1;
4561                                 }
4562                         }
4563
4564                 if (mddev->degraded) {
4565                         ITERATE_RDEV(mddev,rdev,rtmp)
4566                                 if (rdev->raid_disk < 0
4567                                     && !test_bit(Faulty, &rdev->flags)) {
4568                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4569                                                 char nm[20];
4570                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4571                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4572                                                 spares++;
4573                                                 md_new_event(mddev);
4574                                         } else
4575                                                 break;
4576                                 }
4577                 }
4578
4579                 if (spares) {
4580                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4581                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4582                 } else if (mddev->recovery_cp < MaxSector) {
4583                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4584                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4585                         /* nothing to be done ... */
4586                         goto unlock;
4587
4588                 if (mddev->pers->sync_request) {
4589                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4590                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4591                                 /* We are adding a device or devices to an array
4592                                  * which has the bitmap stored on all devices.
4593                                  * So make sure all bitmap pages get written
4594                                  */
4595                                 bitmap_write_all(mddev->bitmap);
4596                         }
4597                         mddev->sync_thread = md_register_thread(md_do_sync,
4598                                                                 mddev,
4599                                                                 "%s_resync");
4600                         if (!mddev->sync_thread) {
4601                                 printk(KERN_ERR "%s: could not start resync"
4602                                         " thread...\n",
4603                                         mdname(mddev));
4604                                 /* leave the spares where they are, it shouldn't hurt */
4605                                 mddev->recovery = 0;
4606                         } else
4607                                 md_wakeup_thread(mddev->sync_thread);
4608                         md_new_event(mddev);
4609                 }
4610         unlock:
4611                 mddev_unlock(mddev);
4612         }
4613 }
4614
4615 static int md_notify_reboot(struct notifier_block *this,
4616                             unsigned long code, void *x)
4617 {
4618         struct list_head *tmp;
4619         mddev_t *mddev;
4620
4621         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4622
4623                 printk(KERN_INFO "md: stopping all md devices.\n");
4624
4625                 ITERATE_MDDEV(mddev,tmp)
4626                         if (mddev_trylock(mddev)==0)
4627                                 do_md_stop (mddev, 1);
4628                 /*
4629                  * certain more exotic SCSI devices are known to be
4630                  * volatile wrt too early system reboots. While the
4631                  * right place to handle this issue is the given
4632                  * driver, we do want to have a safe RAID driver ...
4633                  */
4634                 mdelay(1000*1);
4635         }
4636         return NOTIFY_DONE;
4637 }
4638
4639 static struct notifier_block md_notifier = {
4640         .notifier_call  = md_notify_reboot,
4641         .next           = NULL,
4642         .priority       = INT_MAX, /* before any real devices */
4643 };
4644
4645 static void md_geninit(void)
4646 {
4647         struct proc_dir_entry *p;
4648
4649         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4650
4651         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4652         if (p)
4653                 p->proc_fops = &md_seq_fops;
4654 }
4655
4656 static int __init md_init(void)
4657 {
4658         int minor;
4659
4660         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4661                         " MD_SB_DISKS=%d\n",
4662                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4663                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4664         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4665                         BITMAP_MINOR);
4666
4667         if (register_blkdev(MAJOR_NR, "md"))
4668                 return -1;
4669         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4670                 unregister_blkdev(MAJOR_NR, "md");
4671                 return -1;
4672         }
4673         devfs_mk_dir("md");
4674         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4675                                 md_probe, NULL, NULL);
4676         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4677                             md_probe, NULL, NULL);
4678
4679         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4680                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4681                                 S_IFBLK|S_IRUSR|S_IWUSR,
4682                                 "md/%d", minor);
4683
4684         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4685                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4686                               S_IFBLK|S_IRUSR|S_IWUSR,
4687                               "md/mdp%d", minor);
4688
4689
4690         register_reboot_notifier(&md_notifier);
4691         raid_table_header = register_sysctl_table(raid_root_table, 1);
4692
4693         md_geninit();
4694         return (0);
4695 }
4696
4697
4698 #ifndef MODULE
4699
4700 /*
4701  * Searches all registered partitions for autorun RAID arrays
4702  * at boot time.
4703  */
4704 static dev_t detected_devices[128];
4705 static int dev_cnt;
4706
4707 void md_autodetect_dev(dev_t dev)
4708 {
4709         if (dev_cnt >= 0 && dev_cnt < 127)
4710                 detected_devices[dev_cnt++] = dev;
4711 }
4712
4713
4714 static void autostart_arrays(int part)
4715 {
4716         mdk_rdev_t *rdev;
4717         int i;
4718
4719         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4720
4721         for (i = 0; i < dev_cnt; i++) {
4722                 dev_t dev = detected_devices[i];
4723
4724                 rdev = md_import_device(dev,0, 0);
4725                 if (IS_ERR(rdev))
4726                         continue;
4727
4728                 if (test_bit(Faulty, &rdev->flags)) {
4729                         MD_BUG();
4730                         continue;
4731                 }
4732                 list_add(&rdev->same_set, &pending_raid_disks);
4733         }
4734         dev_cnt = 0;
4735
4736         autorun_devices(part);
4737 }
4738
4739 #endif
4740
4741 static __exit void md_exit(void)
4742 {
4743         mddev_t *mddev;
4744         struct list_head *tmp;
4745         int i;
4746         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4747         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4748         for (i=0; i < MAX_MD_DEVS; i++)
4749                 devfs_remove("md/%d", i);
4750         for (i=0; i < MAX_MD_DEVS; i++)
4751                 devfs_remove("md/d%d", i);
4752
4753         devfs_remove("md");
4754
4755         unregister_blkdev(MAJOR_NR,"md");
4756         unregister_blkdev(mdp_major, "mdp");
4757         unregister_reboot_notifier(&md_notifier);
4758         unregister_sysctl_table(raid_table_header);
4759         remove_proc_entry("mdstat", NULL);
4760         ITERATE_MDDEV(mddev,tmp) {
4761                 struct gendisk *disk = mddev->gendisk;
4762                 if (!disk)
4763                         continue;
4764                 export_array(mddev);
4765                 del_gendisk(disk);
4766                 put_disk(disk);
4767                 mddev->gendisk = NULL;
4768                 mddev_put(mddev);
4769         }
4770 }
4771
4772 module_init(md_init)
4773 module_exit(md_exit)
4774
4775 static int get_ro(char *buffer, struct kernel_param *kp)
4776 {
4777         return sprintf(buffer, "%d", start_readonly);
4778 }
4779 static int set_ro(const char *val, struct kernel_param *kp)
4780 {
4781         char *e;
4782         int num = simple_strtoul(val, &e, 10);
4783         if (*val && (*e == '\0' || *e == '\n')) {
4784                 start_readonly = num;
4785                 return 0;
4786         }
4787         return -EINVAL;
4788 }
4789
4790 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4791 module_param(start_dirty_degraded, int, 0644);
4792
4793
4794 EXPORT_SYMBOL(register_md_personality);
4795 EXPORT_SYMBOL(unregister_md_personality);
4796 EXPORT_SYMBOL(md_error);
4797 EXPORT_SYMBOL(md_done_sync);
4798 EXPORT_SYMBOL(md_write_start);
4799 EXPORT_SYMBOL(md_write_end);
4800 EXPORT_SYMBOL(md_register_thread);
4801 EXPORT_SYMBOL(md_unregister_thread);
4802 EXPORT_SYMBOL(md_wakeup_thread);
4803 EXPORT_SYMBOL(md_print_devices);
4804 EXPORT_SYMBOL(md_check_recovery);
4805 MODULE_LICENSE("GPL");
4806 MODULE_ALIAS("md");
4807 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);