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 <[email protected]>
  11    - boot support for linear and striped mode by Harald Hoyer <[email protected]>
  12    - kerneld support by Boris Tobotras <[email protected]>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <[email protected]>
  15    - Devfs support by Richard Gooch <[email protected]>
  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 <[email protected]>.
  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         if (rdev->size < sb->size && sb->level > 1)
 699                 /* "this cannot possibly happen" ... */
 700                 ret = -EINVAL;
 701
 702  abort:
 703         return ret;
 704 }
 705
 706 /*
 707  * validate_super for 0.90.0
 708  */
 709 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 710 {
 711         mdp_disk_t *desc;
 712         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 713
 714         rdev->raid_disk = -1;
 715         rdev->flags = 0;
 716         if (mddev->raid_disks == 0) {
 717                 mddev->major_version = 0;
 718                 mddev->minor_version = sb->minor_version;
 719                 mddev->patch_version = sb->patch_version;
 720                 mddev->persistent = ! sb->not_persistent;
 721                 mddev->chunk_size = sb->chunk_size;
 722                 mddev->ctime = sb->ctime;
 723                 mddev->utime = sb->utime;
 724                 mddev->level = sb->level;
 725                 mddev->clevel[0] = 0;
 726                 mddev->layout = sb->layout;
 727                 mddev->raid_disks = sb->raid_disks;
 728                 mddev->size = sb->size;
 729                 mddev->events = md_event(sb);
 730                 mddev->bitmap_offset = 0;
 731                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 732
 733                 if (sb->state & (1<<MD_SB_CLEAN))
 734                         mddev->recovery_cp = MaxSector;
 735                 else {
 736                         if (sb->events_hi == sb->cp_events_hi &&
 737                                 sb->events_lo == sb->cp_events_lo) {
 738                                 mddev->recovery_cp = sb->recovery_cp;
 739                         } else
 740                                 mddev->recovery_cp = 0;
 741                 }
 742
 743                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 744                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 745                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 746                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 747
 748                 mddev->max_disks = MD_SB_DISKS;
 749
 750                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 751                     mddev->bitmap_file == NULL) {
 752                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
 753                             && mddev->level != 10) {
 754                                 /* FIXME use a better test */
 755                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
 756                                 return -EINVAL;
 757                         }
 758                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 759                 }
 760
 761         } else if (mddev->pers == NULL) {
 762                 /* Insist on good event counter while assembling */
 763                 __u64 ev1 = md_event(sb);
 764                 ++ev1;
 765                 if (ev1 < mddev->events)
 766                         return -EINVAL;
 767         } else if (mddev->bitmap) {
 768                 /* if adding to array with a bitmap, then we can accept an
 769                  * older device ... but not too old.
 770                  */
 771                 __u64 ev1 = md_event(sb);
 772                 if (ev1 < mddev->bitmap->events_cleared)
 773                         return 0;
 774         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 775                 return 0;
 776
 777         if (mddev->level != LEVEL_MULTIPATH) {
 778                 desc = sb->disks + rdev->desc_nr;
 779
 780                 if (desc->state & (1<<MD_DISK_FAULTY))
 781                         set_bit(Faulty, &rdev->flags);
 782                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 783                          desc->raid_disk < mddev->raid_disks) {
 784                         set_bit(In_sync, &rdev->flags);
 785                         rdev->raid_disk = desc->raid_disk;
 786                 }
 787                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 788                         set_bit(WriteMostly, &rdev->flags);
 789         } else /* MULTIPATH are always insync */
 790                 set_bit(In_sync, &rdev->flags);
 791         return 0;
 792 }
 793
 794 /*
 795  * sync_super for 0.90.0
 796  */
 797 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 798 {
 799         mdp_super_t *sb;
 800         struct list_head *tmp;
 801         mdk_rdev_t *rdev2;
 802         int next_spare = mddev->raid_disks;
 803
 804
 805         /* make rdev->sb match mddev data..
 806          *
 807          * 1/ zero out disks
 808          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 809          * 3/ any empty disks < next_spare become removed
 810          *
 811          * disks[0] gets initialised to REMOVED because
 812          * we cannot be sure from other fields if it has
 813          * been initialised or not.
 814          */
 815         int i;
 816         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 817
 818         rdev->sb_size = MD_SB_BYTES;
 819
 820         sb = (mdp_super_t*)page_address(rdev->sb_page);
 821
 822         memset(sb, 0, sizeof(*sb));
 823
 824         sb->md_magic = MD_SB_MAGIC;
 825         sb->major_version = mddev->major_version;
 826         sb->minor_version = mddev->minor_version;
 827         sb->patch_version = mddev->patch_version;
 828         sb->gvalid_words  = 0; /* ignored */
 829         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 830         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 831         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 832         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 833
 834         sb->ctime = mddev->ctime;
 835         sb->level = mddev->level;
 836         sb->size  = mddev->size;
 837         sb->raid_disks = mddev->raid_disks;
 838         sb->md_minor = mddev->md_minor;
 839         sb->not_persistent = !mddev->persistent;
 840         sb->utime = mddev->utime;
 841         sb->state = 0;
 842         sb->events_hi = (mddev->events>>32);
 843         sb->events_lo = (u32)mddev->events;
 844
 845         if (mddev->in_sync)
 846         {
 847                 sb->recovery_cp = mddev->recovery_cp;
 848                 sb->cp_events_hi = (mddev->events>>32);
 849                 sb->cp_events_lo = (u32)mddev->events;
 850                 if (mddev->recovery_cp == MaxSector)
 851                         sb->state = (1<< MD_SB_CLEAN);
 852         } else
 853                 sb->recovery_cp = 0;
 854
 855         sb->layout = mddev->layout;
 856         sb->chunk_size = mddev->chunk_size;
 857
 858         if (mddev->bitmap && mddev->bitmap_file == NULL)
 859                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 860
 861         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 862         ITERATE_RDEV(mddev,rdev2,tmp) {
 863                 mdp_disk_t *d;
 864                 int desc_nr;
 865                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 866                     && !test_bit(Faulty, &rdev2->flags))
 867                         desc_nr = rdev2->raid_disk;
 868                 else
 869                         desc_nr = next_spare++;
 870                 rdev2->desc_nr = desc_nr;
 871                 d = &sb->disks[rdev2->desc_nr];
 872                 nr_disks++;
 873                 d->number = rdev2->desc_nr;
 874                 d->major = MAJOR(rdev2->bdev->bd_dev);
 875                 d->minor = MINOR(rdev2->bdev->bd_dev);
 876                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 877                     && !test_bit(Faulty, &rdev2->flags))
 878                         d->raid_disk = rdev2->raid_disk;
 879                 else
 880                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 881                 if (test_bit(Faulty, &rdev2->flags)) {
 882                         d->state = (1<<MD_DISK_FAULTY);
 883                         failed++;
 884                 } else if (test_bit(In_sync, &rdev2->flags)) {
 885                         d->state = (1<<MD_DISK_ACTIVE);
 886                         d->state |= (1<<MD_DISK_SYNC);
 887                         active++;
 888                         working++;
 889                 } else {
 890                         d->state = 0;
 891                         spare++;
 892                         working++;
 893                 }
 894                 if (test_bit(WriteMostly, &rdev2->flags))
 895                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 896         }
 897         /* now set the "removed" and "faulty" bits on any missing devices */
 898         for (i=0 ; i < mddev->raid_disks ; i++) {
 899                 mdp_disk_t *d = &sb->disks[i];
 900                 if (d->state == 0 && d->number == 0) {
 901                         d->number = i;
 902                         d->raid_disk = i;
 903                         d->state = (1<<MD_DISK_REMOVED);
 904                         d->state |= (1<<MD_DISK_FAULTY);
 905                         failed++;
 906                 }
 907         }
 908         sb->nr_disks = nr_disks;
 909         sb->active_disks = active;
 910         sb->working_disks = working;
 911         sb->failed_disks = failed;
 912         sb->spare_disks = spare;
 913
 914         sb->this_disk = sb->disks[rdev->desc_nr];
 915         sb->sb_csum = calc_sb_csum(sb);
 916 }
 917
 918 /*
 919  * version 1 superblock
 920  */
 921
 922 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 923 {
 924         unsigned int disk_csum, csum;
 925         unsigned long long newcsum;
 926         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 927         unsigned int *isuper = (unsigned int*)sb;
 928         int i;
 929
 930         disk_csum = sb->sb_csum;
 931         sb->sb_csum = 0;
 932         newcsum = 0;
 933         for (i=0; size>=4; size -= 4 )
 934                 newcsum += le32_to_cpu(*isuper++);
 935
 936         if (size == 2)
 937                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 938
 939         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 940         sb->sb_csum = disk_csum;
 941         return cpu_to_le32(csum);
 942 }
 943
 944 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 945 {
 946         struct mdp_superblock_1 *sb;
 947         int ret;
 948         sector_t sb_offset;
 949         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 950         int bmask;
 951
 952         /*
 953          * Calculate the position of the superblock.
 954          * It is always aligned to a 4K boundary and
 955          * depeding on minor_version, it can be:
 956          * 0: At least 8K, but less than 12K, from end of device
 957          * 1: At start of device
 958          * 2: 4K from start of device.
 959          */
 960         switch(minor_version) {
 961         case 0:
 962                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 963                 sb_offset -= 8*2;
 964                 sb_offset &= ~(sector_t)(4*2-1);
 965                 /* convert from sectors to K */
 966                 sb_offset /= 2;
 967                 break;
 968         case 1:
 969                 sb_offset = 0;
 970                 break;
 971         case 2:
 972                 sb_offset = 4;
 973                 break;
 974         default:
 975                 return -EINVAL;
 976         }
 977         rdev->sb_offset = sb_offset;
 978
 979         /* superblock is rarely larger than 1K, but it can be larger,
 980          * and it is safe to read 4k, so we do that
 981          */
 982         ret = read_disk_sb(rdev, 4096);
 983         if (ret) return ret;
 984
 985
 986         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 987
 988         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 989             sb->major_version != cpu_to_le32(1) ||
 990             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 991             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 992             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 993                 return -EINVAL;
 994
 995         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 996                 printk("md: invalid superblock checksum on %s\n",
 997                         bdevname(rdev->bdev,b));
 998                 return -EINVAL;
 999         }
1000         if (le64_to_cpu(sb->data_size) < 10) {
1001                 printk("md: data_size too small on %s\n",
1002                        bdevname(rdev->bdev,b));
1003                 return -EINVAL;
1004         }
1005         rdev->preferred_minor = 0xffff;
1006         rdev->data_offset = le64_to_cpu(sb->data_offset);
1007         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1008
1009         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1010         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1011         if (rdev->sb_size & bmask)
1012                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1013
1014         if (refdev == 0)
1015                 return 1;
1016         else {
1017                 __u64 ev1, ev2;
1018                 struct mdp_superblock_1 *refsb =
1019                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1020
1021                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1022                     sb->level != refsb->level ||
1023                     sb->layout != refsb->layout ||
1024                     sb->chunksize != refsb->chunksize) {
1025                         printk(KERN_WARNING "md: %s has strangely different"
1026                                 " superblock to %s\n",
1027                                 bdevname(rdev->bdev,b),
1028                                 bdevname(refdev->bdev,b2));
1029                         return -EINVAL;
1030                 }
1031                 ev1 = le64_to_cpu(sb->events);
1032                 ev2 = le64_to_cpu(refsb->events);
1033
1034                 if (ev1 > ev2)
1035                         return 1;
1036         }
1037         if (minor_version)
1038                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1039         else
1040                 rdev->size = rdev->sb_offset;
1041         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1042                 return -EINVAL;
1043         rdev->size = le64_to_cpu(sb->data_size)/2;
1044         if (le32_to_cpu(sb->chunksize))
1045                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1046
1047         if (le32_to_cpu(sb->size) > rdev->size*2)
1048                 return -EINVAL;
1049         return 0;
1050 }
1051
1052 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1053 {
1054         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1055
1056         rdev->raid_disk = -1;
1057         rdev->flags = 0;
1058         if (mddev->raid_disks == 0) {
1059                 mddev->major_version = 1;
1060                 mddev->patch_version = 0;
1061                 mddev->persistent = 1;
1062                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1063                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1064                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1065                 mddev->level = le32_to_cpu(sb->level);
1066                 mddev->clevel[0] = 0;
1067                 mddev->layout = le32_to_cpu(sb->layout);
1068                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1069                 mddev->size = le64_to_cpu(sb->size)/2;
1070                 mddev->events = le64_to_cpu(sb->events);
1071                 mddev->bitmap_offset = 0;
1072                 mddev->default_bitmap_offset = 1024;
1073
1074                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1075                 memcpy(mddev->uuid, sb->set_uuid, 16);
1076
1077                 mddev->max_disks =  (4096-256)/2;
1078
1079                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1080                     mddev->bitmap_file == NULL ) {
1081                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1082                             && mddev->level != 10) {
1083                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1084                                 return -EINVAL;
1085                         }
1086                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1087                 }
1088         } else if (mddev->pers == NULL) {
1089                 /* Insist of good event counter while assembling */
1090                 __u64 ev1 = le64_to_cpu(sb->events);
1091                 ++ev1;
1092                 if (ev1 < mddev->events)
1093                         return -EINVAL;
1094         } else if (mddev->bitmap) {
1095                 /* If adding to array with a bitmap, then we can accept an
1096                  * older device, but not too old.
1097                  */
1098                 __u64 ev1 = le64_to_cpu(sb->events);
1099                 if (ev1 < mddev->bitmap->events_cleared)
1100                         return 0;
1101         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1102                 return 0;
1103
1104         if (mddev->level != LEVEL_MULTIPATH) {
1105                 int role;
1106                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1107                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1108                 switch(role) {
1109                 case 0xffff: /* spare */
1110                         break;
1111                 case 0xfffe: /* faulty */
1112                         set_bit(Faulty, &rdev->flags);
1113                         break;
1114                 default:
1115                         set_bit(In_sync, &rdev->flags);
1116                         rdev->raid_disk = role;
1117                         break;
1118                 }
1119                 if (sb->devflags & WriteMostly1)
1120                         set_bit(WriteMostly, &rdev->flags);
1121         } else /* MULTIPATH are always insync */
1122                 set_bit(In_sync, &rdev->flags);
1123
1124         return 0;
1125 }
1126
1127 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1128 {
1129         struct mdp_superblock_1 *sb;
1130         struct list_head *tmp;
1131         mdk_rdev_t *rdev2;
1132         int max_dev, i;
1133         /* make rdev->sb match mddev and rdev data. */
1134
1135         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1136
1137         sb->feature_map = 0;
1138         sb->pad0 = 0;
1139         memset(sb->pad1, 0, sizeof(sb->pad1));
1140         memset(sb->pad2, 0, sizeof(sb->pad2));
1141         memset(sb->pad3, 0, sizeof(sb->pad3));
1142
1143         sb->utime = cpu_to_le64((__u64)mddev->utime);
1144         sb->events = cpu_to_le64(mddev->events);
1145         if (mddev->in_sync)
1146                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1147         else
1148                 sb->resync_offset = cpu_to_le64(0);
1149
1150         sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1151
1152         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1153                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1154                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1155         }
1156
1157         max_dev = 0;
1158         ITERATE_RDEV(mddev,rdev2,tmp)
1159                 if (rdev2->desc_nr+1 > max_dev)
1160                         max_dev = rdev2->desc_nr+1;
1161
1162         sb->max_dev = cpu_to_le32(max_dev);
1163         for (i=0; i<max_dev;i++)
1164                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1165
1166         ITERATE_RDEV(mddev,rdev2,tmp) {
1167                 i = rdev2->desc_nr;
1168                 if (test_bit(Faulty, &rdev2->flags))
1169                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1170                 else if (test_bit(In_sync, &rdev2->flags))
1171                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1172                 else
1173                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1174         }
1175
1176         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1177         sb->sb_csum = calc_sb_1_csum(sb);
1178 }
1179
1180
1181 static struct super_type super_types[] = {
1182         [0] = {
1183                 .name   = "0.90.0",
1184                 .owner  = THIS_MODULE,
1185                 .load_super     = super_90_load,
1186                 .validate_super = super_90_validate,
1187                 .sync_super     = super_90_sync,
1188         },
1189         [1] = {
1190                 .name   = "md-1",
1191                 .owner  = THIS_MODULE,
1192                 .load_super     = super_1_load,
1193                 .validate_super = super_1_validate,
1194                 .sync_super     = super_1_sync,
1195         },
1196 };
1197
1198 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1199 {
1200         struct list_head *tmp;
1201         mdk_rdev_t *rdev;
1202
1203         ITERATE_RDEV(mddev,rdev,tmp)
1204                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1205                         return rdev;
1206
1207         return NULL;
1208 }
1209
1210 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1211 {
1212         struct list_head *tmp;
1213         mdk_rdev_t *rdev;
1214
1215         ITERATE_RDEV(mddev1,rdev,tmp)
1216                 if (match_dev_unit(mddev2, rdev))
1217                         return 1;
1218
1219         return 0;
1220 }
1221
1222 static LIST_HEAD(pending_raid_disks);
1223
1224 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1225 {
1226         mdk_rdev_t *same_pdev;
1227         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1228         struct kobject *ko;
1229
1230         if (rdev->mddev) {
1231                 MD_BUG();
1232                 return -EINVAL;
1233         }
1234         /* make sure rdev->size exceeds mddev->size */
1235         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1236                 if (mddev->pers)
1237                         /* Cannot change size, so fail */
1238                         return -ENOSPC;
1239                 else
1240                         mddev->size = rdev->size;
1241         }
1242         same_pdev = match_dev_unit(mddev, rdev);
1243         if (same_pdev)
1244                 printk(KERN_WARNING
1245                         "%s: WARNING: %s appears to be on the same physical"
1246                         " disk as %s. True\n     protection against single-disk"
1247                         " failure might be compromised.\n",
1248                         mdname(mddev), bdevname(rdev->bdev,b),
1249                         bdevname(same_pdev->bdev,b2));
1250
1251         /* Verify rdev->desc_nr is unique.
1252          * If it is -1, assign a free number, else
1253          * check number is not in use
1254          */
1255         if (rdev->desc_nr < 0) {
1256                 int choice = 0;
1257                 if (mddev->pers) choice = mddev->raid_disks;
1258                 while (find_rdev_nr(mddev, choice))
1259                         choice++;
1260                 rdev->desc_nr = choice;
1261         } else {
1262                 if (find_rdev_nr(mddev, rdev->desc_nr))
1263                         return -EBUSY;
1264         }
1265         bdevname(rdev->bdev,b);
1266         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1267                 return -ENOMEM;
1268
1269         list_add(&rdev->same_set, &mddev->disks);
1270         rdev->mddev = mddev;
1271         printk(KERN_INFO "md: bind<%s>\n", b);
1272
1273         rdev->kobj.parent = &mddev->kobj;
1274         kobject_add(&rdev->kobj);
1275
1276         if (rdev->bdev->bd_part)
1277                 ko = &rdev->bdev->bd_part->kobj;
1278         else
1279                 ko = &rdev->bdev->bd_disk->kobj;
1280         sysfs_create_link(&rdev->kobj, ko, "block");
1281         return 0;
1282 }
1283
1284 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1285 {
1286         char b[BDEVNAME_SIZE];
1287         if (!rdev->mddev) {
1288                 MD_BUG();
1289                 return;
1290         }
1291         list_del_init(&rdev->same_set);
1292         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1293         rdev->mddev = NULL;
1294         sysfs_remove_link(&rdev->kobj, "block");
1295         kobject_del(&rdev->kobj);
1296 }
1297
1298 /*
1299  * prevent the device from being mounted, repartitioned or
1300  * otherwise reused by a RAID array (or any other kernel
1301  * subsystem), by bd_claiming the device.
1302  */
1303 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1304 {
1305         int err = 0;
1306         struct block_device *bdev;
1307         char b[BDEVNAME_SIZE];
1308
1309         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1310         if (IS_ERR(bdev)) {
1311                 printk(KERN_ERR "md: could not open %s.\n",
1312                         __bdevname(dev, b));
1313                 return PTR_ERR(bdev);
1314         }
1315         err = bd_claim(bdev, rdev);
1316         if (err) {
1317                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1318                         bdevname(bdev, b));
1319                 blkdev_put(bdev);
1320                 return err;
1321         }
1322         rdev->bdev = bdev;
1323         return err;
1324 }
1325
1326 static void unlock_rdev(mdk_rdev_t *rdev)
1327 {
1328         struct block_device *bdev = rdev->bdev;
1329         rdev->bdev = NULL;
1330         if (!bdev)
1331                 MD_BUG();
1332         bd_release(bdev);
1333         blkdev_put(bdev);
1334 }
1335
1336 void md_autodetect_dev(dev_t dev);
1337
1338 static void export_rdev(mdk_rdev_t * rdev)
1339 {
1340         char b[BDEVNAME_SIZE];
1341         printk(KERN_INFO "md: export_rdev(%s)\n",
1342                 bdevname(rdev->bdev,b));
1343         if (rdev->mddev)
1344                 MD_BUG();
1345         free_disk_sb(rdev);
1346         list_del_init(&rdev->same_set);
1347 #ifndef MODULE
1348         md_autodetect_dev(rdev->bdev->bd_dev);
1349 #endif
1350         unlock_rdev(rdev);
1351         kobject_put(&rdev->kobj);
1352 }
1353
1354 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1355 {
1356         unbind_rdev_from_array(rdev);
1357         export_rdev(rdev);
1358 }
1359
1360 static void export_array(mddev_t *mddev)
1361 {
1362         struct list_head *tmp;
1363         mdk_rdev_t *rdev;
1364
1365         ITERATE_RDEV(mddev,rdev,tmp) {
1366                 if (!rdev->mddev) {
1367                         MD_BUG();
1368                         continue;
1369                 }
1370                 kick_rdev_from_array(rdev);
1371         }
1372         if (!list_empty(&mddev->disks))
1373                 MD_BUG();
1374         mddev->raid_disks = 0;
1375         mddev->major_version = 0;
1376 }
1377
1378 static void print_desc(mdp_disk_t *desc)
1379 {
1380         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1381                 desc->major,desc->minor,desc->raid_disk,desc->state);
1382 }
1383
1384 static void print_sb(mdp_super_t *sb)
1385 {
1386         int i;
1387
1388         printk(KERN_INFO
1389                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1390                 sb->major_version, sb->minor_version, sb->patch_version,
1391                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1392                 sb->ctime);
1393         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1394                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1395                 sb->md_minor, sb->layout, sb->chunk_size);
1396         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1397                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1398                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1399                 sb->failed_disks, sb->spare_disks,
1400                 sb->sb_csum, (unsigned long)sb->events_lo);
1401
1402         printk(KERN_INFO);
1403         for (i = 0; i < MD_SB_DISKS; i++) {
1404                 mdp_disk_t *desc;
1405
1406                 desc = sb->disks + i;
1407                 if (desc->number || desc->major || desc->minor ||
1408                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1409                         printk("     D %2d: ", i);
1410                         print_desc(desc);
1411                 }
1412         }
1413         printk(KERN_INFO "md:     THIS: ");
1414         print_desc(&sb->this_disk);
1415
1416 }
1417
1418 static void print_rdev(mdk_rdev_t *rdev)
1419 {
1420         char b[BDEVNAME_SIZE];
1421         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1422                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1423                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1424                 rdev->desc_nr);
1425         if (rdev->sb_loaded) {
1426                 printk(KERN_INFO "md: rdev superblock:\n");
1427                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1428         } else
1429                 printk(KERN_INFO "md: no rdev superblock!\n");
1430 }
1431
1432 void md_print_devices(void)
1433 {
1434         struct list_head *tmp, *tmp2;
1435         mdk_rdev_t *rdev;
1436         mddev_t *mddev;
1437         char b[BDEVNAME_SIZE];
1438
1439         printk("\n");
1440         printk("md:     **********************************\n");
1441         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1442         printk("md:     **********************************\n");
1443         ITERATE_MDDEV(mddev,tmp) {
1444
1445                 if (mddev->bitmap)
1446                         bitmap_print_sb(mddev->bitmap);
1447                 else
1448                         printk("%s: ", mdname(mddev));
1449                 ITERATE_RDEV(mddev,rdev,tmp2)
1450                         printk("<%s>", bdevname(rdev->bdev,b));
1451                 printk("\n");
1452
1453                 ITERATE_RDEV(mddev,rdev,tmp2)
1454                         print_rdev(rdev);
1455         }
1456         printk("md:     **********************************\n");
1457         printk("\n");
1458 }
1459
1460
1461 static void sync_sbs(mddev_t * mddev)
1462 {
1463         mdk_rdev_t *rdev;
1464         struct list_head *tmp;
1465
1466         ITERATE_RDEV(mddev,rdev,tmp) {
1467                 super_types[mddev->major_version].
1468                         sync_super(mddev, rdev);
1469                 rdev->sb_loaded = 1;
1470         }
1471 }
1472
1473 static void md_update_sb(mddev_t * mddev)
1474 {
1475         int err;
1476         struct list_head *tmp;
1477         mdk_rdev_t *rdev;
1478         int sync_req;
1479
1480 repeat:
1481         spin_lock_irq(&mddev->write_lock);
1482         sync_req = mddev->in_sync;
1483         mddev->utime = get_seconds();
1484         mddev->events ++;
1485
1486         if (!mddev->events) {
1487                 /*
1488                  * oops, this 64-bit counter should never wrap.
1489                  * Either we are in around ~1 trillion A.C., assuming
1490                  * 1 reboot per second, or we have a bug:
1491                  */
1492                 MD_BUG();
1493                 mddev->events --;
1494         }
1495         mddev->sb_dirty = 2;
1496         sync_sbs(mddev);
1497
1498         /*
1499          * do not write anything to disk if using
1500          * nonpersistent superblocks
1501          */
1502         if (!mddev->persistent) {
1503                 mddev->sb_dirty = 0;
1504                 spin_unlock_irq(&mddev->write_lock);
1505                 wake_up(&mddev->sb_wait);
1506                 return;
1507         }
1508         spin_unlock_irq(&mddev->write_lock);
1509
1510         dprintk(KERN_INFO
1511                 "md: updating %s RAID superblock on device (in sync %d)\n",
1512                 mdname(mddev),mddev->in_sync);
1513
1514         err = bitmap_update_sb(mddev->bitmap);
1515         ITERATE_RDEV(mddev,rdev,tmp) {
1516                 char b[BDEVNAME_SIZE];
1517                 dprintk(KERN_INFO "md: ");
1518                 if (test_bit(Faulty, &rdev->flags))
1519                         dprintk("(skipping faulty ");
1520
1521                 dprintk("%s ", bdevname(rdev->bdev,b));
1522                 if (!test_bit(Faulty, &rdev->flags)) {
1523                         md_super_write(mddev,rdev,
1524                                        rdev->sb_offset<<1, rdev->sb_size,
1525                                        rdev->sb_page);
1526                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1527                                 bdevname(rdev->bdev,b),
1528                                 (unsigned long long)rdev->sb_offset);
1529
1530                 } else
1531                         dprintk(")\n");
1532                 if (mddev->level == LEVEL_MULTIPATH)
1533                         /* only need to write one superblock... */
1534                         break;
1535         }
1536         md_super_wait(mddev);
1537         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1538
1539         spin_lock_irq(&mddev->write_lock);
1540         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1541                 /* have to write it out again */
1542                 spin_unlock_irq(&mddev->write_lock);
1543                 goto repeat;
1544         }
1545         mddev->sb_dirty = 0;
1546         spin_unlock_irq(&mddev->write_lock);
1547         wake_up(&mddev->sb_wait);
1548
1549 }
1550
1551 /* words written to sysfs files may, or my not, be \n terminated.
1552  * We want to accept with case. For this we use cmd_match.
1553  */
1554 static int cmd_match(const char *cmd, const char *str)
1555 {
1556         /* See if cmd, written into a sysfs file, matches
1557          * str.  They must either be the same, or cmd can
1558          * have a trailing newline
1559          */
1560         while (*cmd && *str && *cmd == *str) {
1561                 cmd++;
1562                 str++;
1563         }
1564         if (*cmd == '\n')
1565                 cmd++;
1566         if (*str || *cmd)
1567                 return 0;
1568         return 1;
1569 }
1570
1571 struct rdev_sysfs_entry {
1572         struct attribute attr;
1573         ssize_t (*show)(mdk_rdev_t *, char *);
1574         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1575 };
1576
1577 static ssize_t
1578 state_show(mdk_rdev_t *rdev, char *page)
1579 {
1580         char *sep = "";
1581         int len=0;
1582
1583         if (test_bit(Faulty, &rdev->flags)) {
1584                 len+= sprintf(page+len, "%sfaulty",sep);
1585                 sep = ",";
1586         }
1587         if (test_bit(In_sync, &rdev->flags)) {
1588                 len += sprintf(page+len, "%sin_sync",sep);
1589                 sep = ",";
1590         }
1591         if (!test_bit(Faulty, &rdev->flags) &&
1592             !test_bit(In_sync, &rdev->flags)) {
1593                 len += sprintf(page+len, "%sspare", sep);
1594                 sep = ",";
1595         }
1596         return len+sprintf(page+len, "\n");
1597 }
1598
1599 static struct rdev_sysfs_entry
1600 rdev_state = __ATTR_RO(state);
1601
1602 static ssize_t
1603 super_show(mdk_rdev_t *rdev, char *page)
1604 {
1605         if (rdev->sb_loaded && rdev->sb_size) {
1606                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1607                 return rdev->sb_size;
1608         } else
1609                 return 0;
1610 }
1611 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1612
1613 static ssize_t
1614 errors_show(mdk_rdev_t *rdev, char *page)
1615 {
1616         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1617 }
1618
1619 static ssize_t
1620 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1621 {
1622         char *e;
1623         unsigned long n = simple_strtoul(buf, &e, 10);
1624         if (*buf && (*e == 0 || *e == '\n')) {
1625                 atomic_set(&rdev->corrected_errors, n);
1626                 return len;
1627         }
1628         return -EINVAL;
1629 }
1630 static struct rdev_sysfs_entry rdev_errors =
1631 __ATTR(errors, 0644, errors_show, errors_store);
1632
1633 static ssize_t
1634 slot_show(mdk_rdev_t *rdev, char *page)
1635 {
1636         if (rdev->raid_disk < 0)
1637                 return sprintf(page, "none\n");
1638         else
1639                 return sprintf(page, "%d\n", rdev->raid_disk);
1640 }
1641
1642 static ssize_t
1643 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1644 {
1645         char *e;
1646         int slot = simple_strtoul(buf, &e, 10);
1647         if (strncmp(buf, "none", 4)==0)
1648                 slot = -1;
1649         else if (e==buf || (*e && *e!= '\n'))
1650                 return -EINVAL;
1651         if (rdev->mddev->pers)
1652                 /* Cannot set slot in active array (yet) */
1653                 return -EBUSY;
1654         if (slot >= rdev->mddev->raid_disks)
1655                 return -ENOSPC;
1656         rdev->raid_disk = slot;
1657         /* assume it is working */
1658         rdev->flags = 0;
1659         set_bit(In_sync, &rdev->flags);
1660         return len;
1661 }
1662
1663
1664 static struct rdev_sysfs_entry rdev_slot =
1665 __ATTR(slot, 0644, slot_show, slot_store);
1666
1667 static ssize_t
1668 offset_show(mdk_rdev_t *rdev, char *page)
1669 {
1670         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1671 }
1672
1673 static ssize_t
1674 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1675 {
1676         char *e;
1677         unsigned long long offset = simple_strtoull(buf, &e, 10);
1678         if (e==buf || (*e && *e != '\n'))
1679                 return -EINVAL;
1680         if (rdev->mddev->pers)
1681                 return -EBUSY;
1682         rdev->data_offset = offset;
1683         return len;
1684 }
1685
1686 static struct rdev_sysfs_entry rdev_offset =
1687 __ATTR(offset, 0644, offset_show, offset_store);
1688
1689 static ssize_t
1690 rdev_size_show(mdk_rdev_t *rdev, char *page)
1691 {
1692         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1693 }
1694
1695 static ssize_t
1696 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1697 {
1698         char *e;
1699         unsigned long long size = simple_strtoull(buf, &e, 10);
1700         if (e==buf || (*e && *e != '\n'))
1701                 return -EINVAL;
1702         if (rdev->mddev->pers)
1703                 return -EBUSY;
1704         rdev->size = size;
1705         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1706                 rdev->mddev->size = size;
1707         return len;
1708 }
1709
1710 static struct rdev_sysfs_entry rdev_size =
1711 __ATTR(size, 0644, rdev_size_show, rdev_size_store);
1712
1713 static struct attribute *rdev_default_attrs[] = {
1714         &rdev_state.attr,
1715         &rdev_super.attr,
1716         &rdev_errors.attr,
1717         &rdev_slot.attr,
1718         &rdev_offset.attr,
1719         &rdev_size.attr,
1720         NULL,
1721 };
1722 static ssize_t
1723 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1724 {
1725         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1726         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1727
1728         if (!entry->show)
1729                 return -EIO;
1730         return entry->show(rdev, page);
1731 }
1732
1733 static ssize_t
1734 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1735               const char *page, size_t length)
1736 {
1737         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1738         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1739
1740         if (!entry->store)
1741                 return -EIO;
1742         return entry->store(rdev, page, length);
1743 }
1744
1745 static void rdev_free(struct kobject *ko)
1746 {
1747         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1748         kfree(rdev);
1749 }
1750 static struct sysfs_ops rdev_sysfs_ops = {
1751         .show           = rdev_attr_show,
1752         .store          = rdev_attr_store,
1753 };
1754 static struct kobj_type rdev_ktype = {
1755         .release        = rdev_free,
1756         .sysfs_ops      = &rdev_sysfs_ops,
1757         .default_attrs  = rdev_default_attrs,
1758 };
1759
1760 /*
1761  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1762  *
1763  * mark the device faulty if:
1764  *
1765  *   - the device is nonexistent (zero size)
1766  *   - the device has no valid superblock
1767  *
1768  * a faulty rdev _never_ has rdev->sb set.
1769  */
1770 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1771 {
1772         char b[BDEVNAME_SIZE];
1773         int err;
1774         mdk_rdev_t *rdev;
1775         sector_t size;
1776
1777         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1778         if (!rdev) {
1779                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1780                 return ERR_PTR(-ENOMEM);
1781         }
1782
1783         if ((err = alloc_disk_sb(rdev)))
1784                 goto abort_free;
1785
1786         err = lock_rdev(rdev, newdev);
1787         if (err)
1788                 goto abort_free;
1789
1790         rdev->kobj.parent = NULL;
1791         rdev->kobj.ktype = &rdev_ktype;
1792         kobject_init(&rdev->kobj);
1793
1794         rdev->desc_nr = -1;
1795         rdev->flags = 0;
1796         rdev->data_offset = 0;
1797         atomic_set(&rdev->nr_pending, 0);
1798         atomic_set(&rdev->read_errors, 0);
1799         atomic_set(&rdev->corrected_errors, 0);
1800
1801         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1802         if (!size) {
1803                 printk(KERN_WARNING
1804                         "md: %s has zero or unknown size, marking faulty!\n",
1805                         bdevname(rdev->bdev,b));
1806                 err = -EINVAL;
1807                 goto abort_free;
1808         }
1809
1810         if (super_format >= 0) {
1811                 err = super_types[super_format].
1812                         load_super(rdev, NULL, super_minor);
1813                 if (err == -EINVAL) {
1814                         printk(KERN_WARNING
1815                                 "md: %s has invalid sb, not importing!\n",
1816                                 bdevname(rdev->bdev,b));
1817                         goto abort_free;
1818                 }
1819                 if (err < 0) {
1820                         printk(KERN_WARNING
1821                                 "md: could not read %s's sb, not importing!\n",
1822                                 bdevname(rdev->bdev,b));
1823                         goto abort_free;
1824                 }
1825         }
1826         INIT_LIST_HEAD(&rdev->same_set);
1827
1828         return rdev;
1829
1830 abort_free:
1831         if (rdev->sb_page) {
1832                 if (rdev->bdev)
1833                         unlock_rdev(rdev);
1834                 free_disk_sb(rdev);
1835         }
1836         kfree(rdev);
1837         return ERR_PTR(err);
1838 }
1839
1840 /*
1841  * Check a full RAID array for plausibility
1842  */
1843
1844
1845 static void analyze_sbs(mddev_t * mddev)
1846 {
1847         int i;
1848         struct list_head *tmp;
1849         mdk_rdev_t *rdev, *freshest;
1850         char b[BDEVNAME_SIZE];
1851
1852         freshest = NULL;
1853         ITERATE_RDEV(mddev,rdev,tmp)
1854                 switch (super_types[mddev->major_version].
1855                         load_super(rdev, freshest, mddev->minor_version)) {
1856                 case 1:
1857                         freshest = rdev;
1858                         break;
1859                 case 0:
1860                         break;
1861                 default:
1862                         printk( KERN_ERR \
1863                                 "md: fatal superblock inconsistency in %s"
1864                                 " -- removing from array\n",
1865                                 bdevname(rdev->bdev,b));
1866                         kick_rdev_from_array(rdev);
1867                 }
1868
1869
1870         super_types[mddev->major_version].
1871                 validate_super(mddev, freshest);
1872
1873         i = 0;
1874         ITERATE_RDEV(mddev,rdev,tmp) {
1875                 if (rdev != freshest)
1876                         if (super_types[mddev->major_version].
1877                             validate_super(mddev, rdev)) {
1878                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1879                                         " from array!\n",
1880                                         bdevname(rdev->bdev,b));
1881                                 kick_rdev_from_array(rdev);
1882                                 continue;
1883                         }
1884                 if (mddev->level == LEVEL_MULTIPATH) {
1885                         rdev->desc_nr = i++;
1886                         rdev->raid_disk = rdev->desc_nr;
1887                         set_bit(In_sync, &rdev->flags);
1888                 }
1889         }
1890
1891
1892
1893         if (mddev->recovery_cp != MaxSector &&
1894             mddev->level >= 1)
1895                 printk(KERN_ERR "md: %s: raid array is not clean"
1896                        " -- starting background reconstruction\n",
1897                        mdname(mddev));
1898
1899 }
1900
1901 static ssize_t
1902 level_show(mddev_t *mddev, char *page)
1903 {
1904         struct mdk_personality *p = mddev->pers;
1905         if (p)
1906                 return sprintf(page, "%s\n", p->name);
1907         else if (mddev->clevel[0])
1908                 return sprintf(page, "%s\n", mddev->clevel);
1909         else if (mddev->level != LEVEL_NONE)
1910                 return sprintf(page, "%d\n", mddev->level);
1911         else
1912                 return 0;
1913 }
1914
1915 static ssize_t
1916 level_store(mddev_t *mddev, const char *buf, size_t len)
1917 {
1918         int rv = len;
1919         if (mddev->pers)
1920                 return -EBUSY;
1921         if (len == 0)
1922                 return 0;
1923         if (len >= sizeof(mddev->clevel))
1924                 return -ENOSPC;
1925         strncpy(mddev->clevel, buf, len);
1926         if (mddev->clevel[len-1] == '\n')
1927                 len--;
1928         mddev->clevel[len] = 0;
1929         mddev->level = LEVEL_NONE;
1930         return rv;
1931 }
1932
1933 static struct md_sysfs_entry md_level =
1934 __ATTR(level, 0644, level_show, level_store);
1935
1936 static ssize_t
1937 raid_disks_show(mddev_t *mddev, char *page)
1938 {
1939         if (mddev->raid_disks == 0)
1940                 return 0;
1941         return sprintf(page, "%d\n", mddev->raid_disks);
1942 }
1943
1944 static int update_raid_disks(mddev_t *mddev, int raid_disks);
1945
1946 static ssize_t
1947 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
1948 {
1949         /* can only set raid_disks if array is not yet active */
1950         char *e;
1951         int rv = 0;
1952         unsigned long n = simple_strtoul(buf, &e, 10);
1953
1954         if (!*buf || (*e && *e != '\n'))
1955                 return -EINVAL;
1956
1957         if (mddev->pers)
1958                 rv = update_raid_disks(mddev, n);
1959         else
1960                 mddev->raid_disks = n;
1961         return rv ? rv : len;
1962 }
1963 static struct md_sysfs_entry md_raid_disks =
1964 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
1965
1966 static ssize_t
1967 chunk_size_show(mddev_t *mddev, char *page)
1968 {
1969         return sprintf(page, "%d\n", mddev->chunk_size);
1970 }
1971
1972 static ssize_t
1973 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1974 {
1975         /* can only set chunk_size if array is not yet active */
1976         char *e;
1977         unsigned long n = simple_strtoul(buf, &e, 10);
1978
1979         if (mddev->pers)
1980                 return -EBUSY;
1981         if (!*buf || (*e && *e != '\n'))
1982                 return -EINVAL;
1983
1984         mddev->chunk_size = n;
1985         return len;
1986 }
1987 static struct md_sysfs_entry md_chunk_size =
1988 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
1989
1990 static ssize_t
1991 null_show(mddev_t *mddev, char *page)
1992 {
1993         return -EINVAL;
1994 }
1995
1996 static ssize_t
1997 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
1998 {
1999         /* buf must be %d:%d\n? giving major and minor numbers */
2000         /* The new device is added to the array.
2001          * If the array has a persistent superblock, we read the
2002          * superblock to initialise info and check validity.
2003          * Otherwise, only checking done is that in bind_rdev_to_array,
2004          * which mainly checks size.
2005          */
2006         char *e;
2007         int major = simple_strtoul(buf, &e, 10);
2008         int minor;
2009         dev_t dev;
2010         mdk_rdev_t *rdev;
2011         int err;
2012
2013         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2014                 return -EINVAL;
2015         minor = simple_strtoul(e+1, &e, 10);
2016         if (*e && *e != '\n')
2017                 return -EINVAL;
2018         dev = MKDEV(major, minor);
2019         if (major != MAJOR(dev) ||
2020             minor != MINOR(dev))
2021                 return -EOVERFLOW;
2022
2023
2024         if (mddev->persistent) {
2025                 rdev = md_import_device(dev, mddev->major_version,
2026                                         mddev->minor_version);
2027                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2028                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2029                                                        mdk_rdev_t, same_set);
2030                         err = super_types[mddev->major_version]
2031                                 .load_super(rdev, rdev0, mddev->minor_version);
2032                         if (err < 0)
2033                                 goto out;
2034                 }
2035         } else
2036                 rdev = md_import_device(dev, -1, -1);
2037
2038         if (IS_ERR(rdev))
2039                 return PTR_ERR(rdev);
2040         err = bind_rdev_to_array(rdev, mddev);
2041  out:
2042         if (err)
2043                 export_rdev(rdev);
2044         return err ? err : len;
2045 }
2046
2047 static struct md_sysfs_entry md_new_device =
2048 __ATTR(new_dev, 0200, null_show, new_dev_store);
2049
2050 static ssize_t
2051 size_show(mddev_t *mddev, char *page)
2052 {
2053         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2054 }
2055
2056 static int update_size(mddev_t *mddev, unsigned long size);
2057
2058 static ssize_t
2059 size_store(mddev_t *mddev, const char *buf, size_t len)
2060 {
2061         /* If array is inactive, we can reduce the component size, but
2062          * not increase it (except from 0).
2063          * If array is active, we can try an on-line resize
2064          */
2065         char *e;
2066         int err = 0;
2067         unsigned long long size = simple_strtoull(buf, &e, 10);
2068         if (!*buf || *buf == '\n' ||
2069             (*e && *e != '\n'))
2070                 return -EINVAL;
2071
2072         if (mddev->pers) {
2073                 err = update_size(mddev, size);
2074                 md_update_sb(mddev);
2075         } else {
2076                 if (mddev->size == 0 ||
2077                     mddev->size > size)
2078                         mddev->size = size;
2079                 else
2080                         err = -ENOSPC;
2081         }
2082         return err ? err : len;
2083 }
2084
2085 static struct md_sysfs_entry md_size =
2086 __ATTR(component_size, 0644, size_show, size_store);
2087
2088
2089 /* Metdata version.
2090  * This is either 'none' for arrays with externally managed metadata,
2091  * or N.M for internally known formats
2092  */
2093 static ssize_t
2094 metadata_show(mddev_t *mddev, char *page)
2095 {
2096         if (mddev->persistent)
2097                 return sprintf(page, "%d.%d\n",
2098                                mddev->major_version, mddev->minor_version);
2099         else
2100                 return sprintf(page, "none\n");
2101 }
2102
2103 static ssize_t
2104 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2105 {
2106         int major, minor;
2107         char *e;
2108         if (!list_empty(&mddev->disks))
2109                 return -EBUSY;
2110
2111         if (cmd_match(buf, "none")) {
2112                 mddev->persistent = 0;
2113                 mddev->major_version = 0;
2114                 mddev->minor_version = 90;
2115                 return len;
2116         }
2117         major = simple_strtoul(buf, &e, 10);
2118         if (e==buf || *e != '.')
2119                 return -EINVAL;
2120         buf = e+1;
2121         minor = simple_strtoul(buf, &e, 10);
2122         if (e==buf || *e != '\n')
2123                 return -EINVAL;
2124         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2125             super_types[major].name == NULL)
2126                 return -ENOENT;
2127         mddev->major_version = major;
2128         mddev->minor_version = minor;
2129         mddev->persistent = 1;
2130         return len;
2131 }
2132
2133 static struct md_sysfs_entry md_metadata =
2134 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
2135
2136 static ssize_t
2137 action_show(mddev_t *mddev, char *page)
2138 {
2139         char *type = "idle";
2140         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2141             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2142                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2143                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2144                                 type = "resync";
2145                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2146                                 type = "check";
2147                         else
2148                                 type = "repair";
2149                 } else
2150                         type = "recover";
2151         }
2152         return sprintf(page, "%s\n", type);
2153 }
2154
2155 static ssize_t
2156 action_store(mddev_t *mddev, const char *page, size_t len)
2157 {
2158         if (!mddev->pers || !mddev->pers->sync_request)
2159                 return -EINVAL;
2160
2161         if (cmd_match(page, "idle")) {
2162                 if (mddev->sync_thread) {
2163                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2164                         md_unregister_thread(mddev->sync_thread);
2165                         mddev->sync_thread = NULL;
2166                         mddev->recovery = 0;
2167                 }
2168         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2169                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2170                 return -EBUSY;
2171         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2172                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2173         else {
2174                 if (cmd_match(page, "check"))
2175                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2176                 else if (cmd_match(page, "repair"))
2177                         return -EINVAL;
2178                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2179                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2180         }
2181         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2182         md_wakeup_thread(mddev->thread);
2183         return len;
2184 }
2185
2186 static ssize_t
2187 mismatch_cnt_show(mddev_t *mddev, char *page)
2188 {
2189         return sprintf(page, "%llu\n",
2190                        (unsigned long long) mddev->resync_mismatches);
2191 }
2192
2193 static struct md_sysfs_entry
2194 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2195
2196
2197 static struct md_sysfs_entry
2198 md_mismatches = __ATTR_RO(mismatch_cnt);
2199
2200 static struct attribute *md_default_attrs[] = {
2201         &md_level.attr,
2202         &md_raid_disks.attr,
2203         &md_chunk_size.attr,
2204         &md_size.attr,
2205         &md_metadata.attr,
2206         &md_new_device.attr,
2207         NULL,
2208 };
2209
2210 static struct attribute *md_redundancy_attrs[] = {
2211         &md_scan_mode.attr,
2212         &md_mismatches.attr,
2213         NULL,
2214 };
2215 static struct attribute_group md_redundancy_group = {
2216         .name = NULL,
2217         .attrs = md_redundancy_attrs,
2218 };
2219
2220
2221 static ssize_t
2222 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2223 {
2224         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2225         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2226         ssize_t rv;
2227
2228         if (!entry->show)
2229                 return -EIO;
2230         mddev_lock(mddev);
2231         rv = entry->show(mddev, page);
2232         mddev_unlock(mddev);
2233         return rv;
2234 }
2235
2236 static ssize_t
2237 md_attr_store(struct kobject *kobj, struct attribute *attr,
2238               const char *page, size_t length)
2239 {
2240         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2241         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2242         ssize_t rv;
2243
2244         if (!entry->store)
2245                 return -EIO;
2246         mddev_lock(mddev);
2247         rv = entry->store(mddev, page, length);
2248         mddev_unlock(mddev);
2249         return rv;
2250 }
2251
2252 static void md_free(struct kobject *ko)
2253 {
2254         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2255         kfree(mddev);
2256 }
2257
2258 static struct sysfs_ops md_sysfs_ops = {
2259         .show   = md_attr_show,
2260         .store  = md_attr_store,
2261 };
2262 static struct kobj_type md_ktype = {
2263         .release        = md_free,
2264         .sysfs_ops      = &md_sysfs_ops,
2265         .default_attrs  = md_default_attrs,
2266 };
2267
2268 int mdp_major = 0;
2269
2270 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2271 {
2272         static DECLARE_MUTEX(disks_sem);
2273         mddev_t *mddev = mddev_find(dev);
2274         struct gendisk *disk;
2275         int partitioned = (MAJOR(dev) != MD_MAJOR);
2276         int shift = partitioned ? MdpMinorShift : 0;
2277         int unit = MINOR(dev) >> shift;
2278
2279         if (!mddev)
2280                 return NULL;
2281
2282         down(&disks_sem);
2283         if (mddev->gendisk) {
2284                 up(&disks_sem);
2285                 mddev_put(mddev);
2286                 return NULL;
2287         }
2288         disk = alloc_disk(1 << shift);
2289         if (!disk) {
2290                 up(&disks_sem);
2291                 mddev_put(mddev);
2292                 return NULL;
2293         }
2294         disk->major = MAJOR(dev);
2295         disk->first_minor = unit << shift;
2296         if (partitioned) {
2297                 sprintf(disk->disk_name, "md_d%d", unit);
2298                 sprintf(disk->devfs_name, "md/d%d", unit);
2299         } else {
2300                 sprintf(disk->disk_name, "md%d", unit);
2301                 sprintf(disk->devfs_name, "md/%d", unit);
2302         }
2303         disk->fops = &md_fops;
2304         disk->private_data = mddev;
2305         disk->queue = mddev->queue;
2306         add_disk(disk);
2307         mddev->gendisk = disk;
2308         up(&disks_sem);
2309         mddev->kobj.parent = &disk->kobj;
2310         mddev->kobj.k_name = NULL;
2311         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2312         mddev->kobj.ktype = &md_ktype;
2313         kobject_register(&mddev->kobj);
2314         return NULL;
2315 }
2316
2317 void md_wakeup_thread(mdk_thread_t *thread);
2318
2319 static void md_safemode_timeout(unsigned long data)
2320 {
2321         mddev_t *mddev = (mddev_t *) data;
2322
2323         mddev->safemode = 1;
2324         md_wakeup_thread(mddev->thread);
2325 }
2326
2327 static int start_dirty_degraded;
2328
2329 static int do_md_run(mddev_t * mddev)
2330 {
2331         int err;
2332         int chunk_size;
2333         struct list_head *tmp;
2334         mdk_rdev_t *rdev;
2335         struct gendisk *disk;
2336         struct mdk_personality *pers;
2337         char b[BDEVNAME_SIZE];
2338
2339         if (list_empty(&mddev->disks))
2340                 /* cannot run an array with no devices.. */
2341                 return -EINVAL;
2342
2343         if (mddev->pers)
2344                 return -EBUSY;
2345
2346         /*
2347          * Analyze all RAID superblock(s)
2348          */
2349         if (!mddev->raid_disks)
2350                 analyze_sbs(mddev);
2351
2352         chunk_size = mddev->chunk_size;
2353
2354         if (chunk_size) {
2355                 if (chunk_size > MAX_CHUNK_SIZE) {
2356                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2357                                 chunk_size, MAX_CHUNK_SIZE);
2358                         return -EINVAL;
2359                 }
2360                 /*
2361                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2362                  */
2363                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2364                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2365                         return -EINVAL;
2366                 }
2367                 if (chunk_size < PAGE_SIZE) {
2368                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2369                                 chunk_size, PAGE_SIZE);
2370                         return -EINVAL;
2371                 }
2372
2373                 /* devices must have minimum size of one chunk */
2374                 ITERATE_RDEV(mddev,rdev,tmp) {
2375                         if (test_bit(Faulty, &rdev->flags))
2376                                 continue;
2377                         if (rdev->size < chunk_size / 1024) {
2378                                 printk(KERN_WARNING
2379                                         "md: Dev %s smaller than chunk_size:"
2380                                         " %lluk < %dk\n",
2381                                         bdevname(rdev->bdev,b),
2382                                         (unsigned long long)rdev->size,
2383                                         chunk_size / 1024);
2384                                 return -EINVAL;
2385                         }
2386                 }
2387         }
2388
2389 #ifdef CONFIG_KMOD
2390         if (mddev->level != LEVEL_NONE)
2391                 request_module("md-level-%d", mddev->level);
2392         else if (mddev->clevel[0])
2393                 request_module("md-%s", mddev->clevel);
2394 #endif
2395
2396         /*
2397          * Drop all container device buffers, from now on
2398          * the only valid external interface is through the md
2399          * device.
2400          * Also find largest hardsector size
2401          */
2402         ITERATE_RDEV(mddev,rdev,tmp) {
2403                 if (test_bit(Faulty, &rdev->flags))
2404                         continue;
2405                 sync_blockdev(rdev->bdev);
2406                 invalidate_bdev(rdev->bdev, 0);
2407         }
2408
2409         md_probe(mddev->unit, NULL, NULL);
2410         disk = mddev->gendisk;
2411         if (!disk)
2412                 return -ENOMEM;
2413
2414         spin_lock(&pers_lock);
2415         pers = find_pers(mddev->level, mddev->clevel);
2416         if (!pers || !try_module_get(pers->owner)) {
2417                 spin_unlock(&pers_lock);
2418                 if (mddev->level != LEVEL_NONE)
2419                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2420                                mddev->level);
2421                 else
2422                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2423                                mddev->clevel);
2424                 return -EINVAL;
2425         }
2426         mddev->pers = pers;
2427         spin_unlock(&pers_lock);
2428         mddev->level = pers->level;
2429         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2430
2431         mddev->recovery = 0;
2432         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2433         mddev->barriers_work = 1;
2434         mddev->ok_start_degraded = start_dirty_degraded;
2435
2436         if (start_readonly)
2437                 mddev->ro = 2; /* read-only, but switch on first write */
2438
2439         err = mddev->pers->run(mddev);
2440         if (!err && mddev->pers->sync_request) {
2441                 err = bitmap_create(mddev);
2442                 if (err) {
2443                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2444                                mdname(mddev), err);
2445                         mddev->pers->stop(mddev);
2446                 }
2447         }
2448         if (err) {
2449                 printk(KERN_ERR "md: pers->run() failed ...\n");
2450                 module_put(mddev->pers->owner);
2451                 mddev->pers = NULL;
2452                 bitmap_destroy(mddev);
2453                 return err;
2454         }
2455         if (mddev->pers->sync_request)
2456                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2457         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2458                 mddev->ro = 0;
2459
2460         atomic_set(&mddev->writes_pending,0);
2461         mddev->safemode = 0;
2462         mddev->safemode_timer.function = md_safemode_timeout;
2463         mddev->safemode_timer.data = (unsigned long) mddev;
2464         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2465         mddev->in_sync = 1;
2466
2467         ITERATE_RDEV(mddev,rdev,tmp)
2468                 if (rdev->raid_disk >= 0) {
2469                         char nm[20];
2470                         sprintf(nm, "rd%d", rdev->raid_disk);
2471                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2472                 }
2473
2474         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2475         md_wakeup_thread(mddev->thread);
2476
2477         if (mddev->sb_dirty)
2478                 md_update_sb(mddev);
2479
2480         set_capacity(disk, mddev->array_size<<1);
2481
2482         /* If we call blk_queue_make_request here, it will
2483          * re-initialise max_sectors etc which may have been
2484          * refined inside -> run.  So just set the bits we need to set.
2485          * Most initialisation happended when we called
2486          * blk_queue_make_request(..., md_fail_request)
2487          * earlier.
2488          */
2489         mddev->queue->queuedata = mddev;
2490         mddev->queue->make_request_fn = mddev->pers->make_request;
2491
2492         mddev->changed = 1;
2493         md_new_event(mddev);
2494         return 0;
2495 }
2496
2497 static int restart_array(mddev_t *mddev)
2498 {
2499         struct gendisk *disk = mddev->gendisk;
2500         int err;
2501
2502         /*
2503          * Complain if it has no devices
2504          */
2505         err = -ENXIO;
2506         if (list_empty(&mddev->disks))
2507                 goto out;
2508
2509         if (mddev->pers) {
2510                 err = -EBUSY;
2511                 if (!mddev->ro)
2512                         goto out;
2513
2514                 mddev->safemode = 0;
2515                 mddev->ro = 0;
2516                 set_disk_ro(disk, 0);
2517
2518                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2519                         mdname(mddev));
2520                 /*
2521                  * Kick recovery or resync if necessary
2522                  */
2523                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2524                 md_wakeup_thread(mddev->thread);
2525                 err = 0;
2526         } else {
2527                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2528                         mdname(mddev));
2529                 err = -EINVAL;
2530         }
2531
2532 out:
2533         return err;
2534 }
2535
2536 static int do_md_stop(mddev_t * mddev, int ro)
2537 {
2538         int err = 0;
2539         struct gendisk *disk = mddev->gendisk;
2540
2541         if (mddev->pers) {
2542                 if (atomic_read(&mddev->active)>2) {
2543                         printk("md: %s still in use.\n",mdname(mddev));
2544                         return -EBUSY;
2545                 }
2546
2547                 if (mddev->sync_thread) {
2548                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2549                         md_unregister_thread(mddev->sync_thread);
2550                         mddev->sync_thread = NULL;
2551                 }
2552
2553                 del_timer_sync(&mddev->safemode_timer);
2554
2555                 invalidate_partition(disk, 0);
2556
2557                 if (ro) {
2558                         err  = -ENXIO;
2559                         if (mddev->ro==1)
2560                                 goto out;
2561                         mddev->ro = 1;
2562                 } else {
2563                         bitmap_flush(mddev);
2564                         md_super_wait(mddev);
2565                         if (mddev->ro)
2566                                 set_disk_ro(disk, 0);
2567                         blk_queue_make_request(mddev->queue, md_fail_request);
2568                         mddev->pers->stop(mddev);
2569                         if (mddev->pers->sync_request)
2570                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2571
2572                         module_put(mddev->pers->owner);
2573                         mddev->pers = NULL;
2574                         if (mddev->ro)
2575                                 mddev->ro = 0;
2576                 }
2577                 if (!mddev->in_sync) {
2578                         /* mark array as shutdown cleanly */
2579                         mddev->in_sync = 1;
2580                         md_update_sb(mddev);
2581                 }
2582                 if (ro)
2583                         set_disk_ro(disk, 1);
2584         }
2585
2586         bitmap_destroy(mddev);
2587         if (mddev->bitmap_file) {
2588                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2589                 fput(mddev->bitmap_file);
2590                 mddev->bitmap_file = NULL;
2591         }
2592         mddev->bitmap_offset = 0;
2593
2594         /*
2595          * Free resources if final stop
2596          */
2597         if (!ro) {
2598                 mdk_rdev_t *rdev;
2599                 struct list_head *tmp;
2600                 struct gendisk *disk;
2601                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2602
2603                 ITERATE_RDEV(mddev,rdev,tmp)
2604                         if (rdev->raid_disk >= 0) {
2605                                 char nm[20];
2606                                 sprintf(nm, "rd%d", rdev->raid_disk);
2607                                 sysfs_remove_link(&mddev->kobj, nm);
2608                         }
2609
2610                 export_array(mddev);
2611
2612                 mddev->array_size = 0;
2613                 disk = mddev->gendisk;
2614                 if (disk)
2615                         set_capacity(disk, 0);
2616                 mddev->changed = 1;
2617         } else
2618                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2619                         mdname(mddev));
2620         err = 0;
2621         md_new_event(mddev);
2622 out:
2623         return err;
2624 }
2625
2626 static void autorun_array(mddev_t *mddev)
2627 {
2628         mdk_rdev_t *rdev;
2629         struct list_head *tmp;
2630         int err;
2631
2632         if (list_empty(&mddev->disks))
2633                 return;
2634
2635         printk(KERN_INFO "md: running: ");
2636
2637         ITERATE_RDEV(mddev,rdev,tmp) {
2638                 char b[BDEVNAME_SIZE];
2639                 printk("<%s>", bdevname(rdev->bdev,b));
2640         }
2641         printk("\n");
2642
2643         err = do_md_run (mddev);
2644         if (err) {
2645                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2646                 do_md_stop (mddev, 0);
2647         }
2648 }
2649
2650 /*
2651  * lets try to run arrays based on all disks that have arrived
2652  * until now. (those are in pending_raid_disks)
2653  *
2654  * the method: pick the first pending disk, collect all disks with
2655  * the same UUID, remove all from the pending list and put them into
2656  * the 'same_array' list. Then order this list based on superblock
2657  * update time (freshest comes first), kick out 'old' disks and
2658  * compare superblocks. If everything's fine then run it.
2659  *
2660  * If "unit" is allocated, then bump its reference count
2661  */
2662 static void autorun_devices(int part)
2663 {
2664         struct list_head candidates;
2665         struct list_head *tmp;
2666         mdk_rdev_t *rdev0, *rdev;
2667         mddev_t *mddev;
2668         char b[BDEVNAME_SIZE];
2669
2670         printk(KERN_INFO "md: autorun ...\n");
2671         while (!list_empty(&pending_raid_disks)) {
2672                 dev_t dev;
2673                 rdev0 = list_entry(pending_raid_disks.next,
2674                                          mdk_rdev_t, same_set);
2675
2676                 printk(KERN_INFO "md: considering %s ...\n",
2677                         bdevname(rdev0->bdev,b));
2678                 INIT_LIST_HEAD(&candidates);
2679                 ITERATE_RDEV_PENDING(rdev,tmp)
2680                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2681                                 printk(KERN_INFO "md:  adding %s ...\n",
2682                                         bdevname(rdev->bdev,b));
2683                                 list_move(&rdev->same_set, &candidates);
2684                         }
2685                 /*
2686                  * now we have a set of devices, with all of them having
2687                  * mostly sane superblocks. It's time to allocate the
2688                  * mddev.
2689                  */
2690                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2691                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2692                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2693                         break;
2694                 }
2695                 if (part)
2696                         dev = MKDEV(mdp_major,
2697                                     rdev0->preferred_minor << MdpMinorShift);
2698                 else
2699                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2700
2701                 md_probe(dev, NULL, NULL);
2702                 mddev = mddev_find(dev);
2703                 if (!mddev) {
2704                         printk(KERN_ERR
2705                                 "md: cannot allocate memory for md drive.\n");
2706                         break;
2707                 }
2708                 if (mddev_lock(mddev))
2709                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2710                                mdname(mddev));
2711                 else if (mddev->raid_disks || mddev->major_version
2712                          || !list_empty(&mddev->disks)) {
2713                         printk(KERN_WARNING
2714                                 "md: %s already running, cannot run %s\n",
2715                                 mdname(mddev), bdevname(rdev0->bdev,b));
2716                         mddev_unlock(mddev);
2717                 } else {
2718                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2719                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2720                                 list_del_init(&rdev->same_set);
2721                                 if (bind_rdev_to_array(rdev, mddev))
2722                                         export_rdev(rdev);
2723                         }
2724                         autorun_array(mddev);
2725                         mddev_unlock(mddev);
2726                 }
2727                 /* on success, candidates will be empty, on error
2728                  * it won't...
2729                  */
2730                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2731                         export_rdev(rdev);
2732                 mddev_put(mddev);
2733         }
2734         printk(KERN_INFO "md: ... autorun DONE.\n");
2735 }
2736
2737 /*
2738  * import RAID devices based on one partition
2739  * if possible, the array gets run as well.
2740  */
2741
2742 static int autostart_array(dev_t startdev)
2743 {
2744         char b[BDEVNAME_SIZE];
2745         int err = -EINVAL, i;
2746         mdp_super_t *sb = NULL;
2747         mdk_rdev_t *start_rdev = NULL, *rdev;
2748
2749         start_rdev = md_import_device(startdev, 0, 0);
2750         if (IS_ERR(start_rdev))
2751                 return err;
2752
2753
2754         /* NOTE: this can only work for 0.90.0 superblocks */
2755         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2756         if (sb->major_version != 0 ||
2757             sb->minor_version != 90 ) {
2758                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2759                 export_rdev(start_rdev);
2760                 return err;
2761         }
2762
2763         if (test_bit(Faulty, &start_rdev->flags)) {
2764                 printk(KERN_WARNING
2765                         "md: can not autostart based on faulty %s!\n",
2766                         bdevname(start_rdev->bdev,b));
2767                 export_rdev(start_rdev);
2768                 return err;
2769         }
2770         list_add(&start_rdev->same_set, &pending_raid_disks);
2771
2772         for (i = 0; i < MD_SB_DISKS; i++) {
2773                 mdp_disk_t *desc = sb->disks + i;
2774                 dev_t dev = MKDEV(desc->major, desc->minor);
2775
2776                 if (!dev)
2777                         continue;
2778                 if (dev == startdev)
2779                         continue;
2780                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2781                         continue;
2782                 rdev = md_import_device(dev, 0, 0);
2783                 if (IS_ERR(rdev))
2784                         continue;
2785
2786                 list_add(&rdev->same_set, &pending_raid_disks);
2787         }
2788
2789         /*
2790          * possibly return codes
2791          */
2792         autorun_devices(0);
2793         return 0;
2794
2795 }
2796
2797
2798 static int get_version(void __user * arg)
2799 {
2800         mdu_version_t ver;
2801
2802         ver.major = MD_MAJOR_VERSION;
2803         ver.minor = MD_MINOR_VERSION;
2804         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2805
2806         if (copy_to_user(arg, &ver, sizeof(ver)))
2807                 return -EFAULT;
2808
2809         return 0;
2810 }
2811
2812 static int get_array_info(mddev_t * mddev, void __user * arg)
2813 {
2814         mdu_array_info_t info;
2815         int nr,working,active,failed,spare;
2816         mdk_rdev_t *rdev;
2817         struct list_head *tmp;
2818
2819         nr=working=active=failed=spare=0;
2820         ITERATE_RDEV(mddev,rdev,tmp) {
2821                 nr++;
2822                 if (test_bit(Faulty, &rdev->flags))
2823                         failed++;
2824                 else {
2825                         working++;
2826                         if (test_bit(In_sync, &rdev->flags))
2827                                 active++;
2828                         else
2829                                 spare++;
2830                 }
2831         }
2832
2833         info.major_version = mddev->major_version;
2834         info.minor_version = mddev->minor_version;
2835         info.patch_version = MD_PATCHLEVEL_VERSION;
2836         info.ctime         = mddev->ctime;
2837         info.level         = mddev->level;
2838         info.size          = mddev->size;
2839         info.nr_disks      = nr;
2840         info.raid_disks    = mddev->raid_disks;
2841         info.md_minor      = mddev->md_minor;
2842         info.not_persistent= !mddev->persistent;
2843
2844         info.utime         = mddev->utime;
2845         info.state         = 0;
2846         if (mddev->in_sync)
2847                 info.state = (1<<MD_SB_CLEAN);
2848         if (mddev->bitmap && mddev->bitmap_offset)
2849                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2850         info.active_disks  = active;
2851         info.working_disks = working;
2852         info.failed_disks  = failed;
2853         info.spare_disks   = spare;
2854
2855         info.layout        = mddev->layout;
2856         info.chunk_size    = mddev->chunk_size;
2857
2858         if (copy_to_user(arg, &info, sizeof(info)))
2859                 return -EFAULT;
2860
2861         return 0;
2862 }
2863
2864 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2865 {
2866         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2867         char *ptr, *buf = NULL;
2868         int err = -ENOMEM;
2869
2870         file = kmalloc(sizeof(*file), GFP_KERNEL);
2871         if (!file)
2872                 goto out;
2873
2874         /* bitmap disabled, zero the first byte and copy out */
2875         if (!mddev->bitmap || !mddev->bitmap->file) {
2876                 file->pathname[0] = '\0';
2877                 goto copy_out;
2878         }
2879
2880         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2881         if (!buf)
2882                 goto out;
2883
2884         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2885         if (!ptr)
2886                 goto out;
2887
2888         strcpy(file->pathname, ptr);
2889
2890 copy_out:
2891         err = 0;
2892         if (copy_to_user(arg, file, sizeof(*file)))
2893                 err = -EFAULT;
2894 out:
2895         kfree(buf);
2896         kfree(file);
2897         return err;
2898 }
2899
2900 static int get_disk_info(mddev_t * mddev, void __user * arg)
2901 {
2902         mdu_disk_info_t info;
2903         unsigned int nr;
2904         mdk_rdev_t *rdev;
2905
2906         if (copy_from_user(&info, arg, sizeof(info)))
2907                 return -EFAULT;
2908
2909         nr = info.number;
2910
2911         rdev = find_rdev_nr(mddev, nr);
2912         if (rdev) {
2913                 info.major = MAJOR(rdev->bdev->bd_dev);
2914                 info.minor = MINOR(rdev->bdev->bd_dev);
2915                 info.raid_disk = rdev->raid_disk;
2916                 info.state = 0;
2917                 if (test_bit(Faulty, &rdev->flags))
2918                         info.state |= (1<<MD_DISK_FAULTY);
2919                 else if (test_bit(In_sync, &rdev->flags)) {
2920                         info.state |= (1<<MD_DISK_ACTIVE);
2921                         info.state |= (1<<MD_DISK_SYNC);
2922                 }
2923                 if (test_bit(WriteMostly, &rdev->flags))
2924                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2925         } else {
2926                 info.major = info.minor = 0;
2927                 info.raid_disk = -1;
2928                 info.state = (1<<MD_DISK_REMOVED);
2929         }
2930
2931         if (copy_to_user(arg, &info, sizeof(info)))
2932                 return -EFAULT;
2933
2934         return 0;
2935 }
2936
2937 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2938 {
2939         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2940         mdk_rdev_t *rdev;
2941         dev_t dev = MKDEV(info->major,info->minor);
2942
2943         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2944                 return -EOVERFLOW;
2945
2946         if (!mddev->raid_disks) {
2947                 int err;
2948                 /* expecting a device which has a superblock */
2949                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2950                 if (IS_ERR(rdev)) {
2951                         printk(KERN_WARNING
2952                                 "md: md_import_device returned %ld\n",
2953                                 PTR_ERR(rdev));
2954                         return PTR_ERR(rdev);
2955                 }
2956                 if (!list_empty(&mddev->disks)) {
2957                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2958                                                         mdk_rdev_t, same_set);
2959                         int err = super_types[mddev->major_version]
2960                                 .load_super(rdev, rdev0, mddev->minor_version);
2961                         if (err < 0) {
2962                                 printk(KERN_WARNING
2963                                         "md: %s has different UUID to %s\n",
2964                                         bdevname(rdev->bdev,b),
2965                                         bdevname(rdev0->bdev,b2));
2966                                 export_rdev(rdev);
2967                                 return -EINVAL;
2968                         }
2969                 }
2970                 err = bind_rdev_to_array(rdev, mddev);
2971                 if (err)
2972                         export_rdev(rdev);
2973                 return err;
2974         }
2975
2976         /*
2977          * add_new_disk can be used once the array is assembled
2978          * to add "hot spares".  They must already have a superblock
2979          * written
2980          */
2981         if (mddev->pers) {
2982                 int err;
2983                 if (!mddev->pers->hot_add_disk) {
2984                         printk(KERN_WARNING
2985                                 "%s: personality does not support diskops!\n",
2986                                mdname(mddev));
2987                         return -EINVAL;
2988                 }
2989                 if (mddev->persistent)
2990                         rdev = md_import_device(dev, mddev->major_version,
2991                                                 mddev->minor_version);
2992                 else
2993                         rdev = md_import_device(dev, -1, -1);
2994                 if (IS_ERR(rdev)) {
2995                         printk(KERN_WARNING
2996                                 "md: md_import_device returned %ld\n",
2997                                 PTR_ERR(rdev));
2998                         return PTR_ERR(rdev);
2999                 }
3000                 /* set save_raid_disk if appropriate */
3001                 if (!mddev->persistent) {
3002                         if (info->state & (1<<MD_DISK_SYNC)  &&
3003                             info->raid_disk < mddev->raid_disks)
3004                                 rdev->raid_disk = info->raid_disk;
3005                         else
3006                                 rdev->raid_disk = -1;
3007                 } else
3008                         super_types[mddev->major_version].
3009                                 validate_super(mddev, rdev);
3010                 rdev->saved_raid_disk = rdev->raid_disk;
3011
3012                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3013                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3014                         set_bit(WriteMostly, &rdev->flags);
3015
3016                 rdev->raid_disk = -1;
3017                 err = bind_rdev_to_array(rdev, mddev);
3018                 if (err)
3019                         export_rdev(rdev);
3020
3021                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3022                 md_wakeup_thread(mddev->thread);
3023                 return err;
3024         }
3025
3026         /* otherwise, add_new_disk is only allowed
3027          * for major_version==0 superblocks
3028          */
3029         if (mddev->major_version != 0) {
3030                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3031                        mdname(mddev));
3032                 return -EINVAL;
3033         }
3034
3035         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3036                 int err;
3037                 rdev = md_import_device (dev, -1, 0);
3038                 if (IS_ERR(rdev)) {
3039                         printk(KERN_WARNING
3040                                 "md: error, md_import_device() returned %ld\n",
3041                                 PTR_ERR(rdev));
3042                         return PTR_ERR(rdev);
3043                 }
3044                 rdev->desc_nr = info->number;
3045                 if (info->raid_disk < mddev->raid_disks)
3046                         rdev->raid_disk = info->raid_disk;
3047                 else
3048                         rdev->raid_disk = -1;
3049
3050                 rdev->flags = 0;
3051
3052                 if (rdev->raid_disk < mddev->raid_disks)
3053                         if (info->state & (1<<MD_DISK_SYNC))
3054                                 set_bit(In_sync, &rdev->flags);
3055
3056                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3057                         set_bit(WriteMostly, &rdev->flags);
3058
3059                 if (!mddev->persistent) {
3060                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3061                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3062                 } else
3063                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3064                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3065
3066                 err = bind_rdev_to_array(rdev, mddev);
3067                 if (err) {
3068                         export_rdev(rdev);
3069                         return err;
3070                 }
3071         }
3072
3073         return 0;
3074 }
3075
3076 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3077 {
3078         char b[BDEVNAME_SIZE];
3079         mdk_rdev_t *rdev;
3080
3081         if (!mddev->pers)
3082                 return -ENODEV;
3083
3084         rdev = find_rdev(mddev, dev);
3085         if (!rdev)
3086                 return -ENXIO;
3087
3088         if (rdev->raid_disk >= 0)
3089                 goto busy;
3090
3091         kick_rdev_from_array(rdev);
3092         md_update_sb(mddev);
3093         md_new_event(mddev);
3094
3095         return 0;
3096 busy:
3097         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3098                 bdevname(rdev->bdev,b), mdname(mddev));
3099         return -EBUSY;
3100 }
3101
3102 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3103 {
3104         char b[BDEVNAME_SIZE];
3105         int err;
3106         unsigned int size;
3107         mdk_rdev_t *rdev;
3108
3109         if (!mddev->pers)
3110                 return -ENODEV;
3111
3112         if (mddev->major_version != 0) {
3113                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3114                         " version-0 superblocks.\n",
3115                         mdname(mddev));
3116                 return -EINVAL;
3117         }
3118         if (!mddev->pers->hot_add_disk) {
3119                 printk(KERN_WARNING
3120                         "%s: personality does not support diskops!\n",
3121                         mdname(mddev));
3122                 return -EINVAL;
3123         }
3124
3125         rdev = md_import_device (dev, -1, 0);
3126         if (IS_ERR(rdev)) {
3127                 printk(KERN_WARNING
3128                         "md: error, md_import_device() returned %ld\n",
3129                         PTR_ERR(rdev));
3130                 return -EINVAL;
3131         }
3132
3133         if (mddev->persistent)
3134                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3135         else
3136                 rdev->sb_offset =
3137                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3138
3139         size = calc_dev_size(rdev, mddev->chunk_size);
3140         rdev->size = size;
3141
3142         if (test_bit(Faulty, &rdev->flags)) {
3143                 printk(KERN_WARNING
3144                         "md: can not hot-add faulty %s disk to %s!\n",
3145                         bdevname(rdev->bdev,b), mdname(mddev));
3146                 err = -EINVAL;
3147                 goto abort_export;
3148         }
3149         clear_bit(In_sync, &rdev->flags);
3150         rdev->desc_nr = -1;
3151         err = bind_rdev_to_array(rdev, mddev);
3152         if (err)
3153                 goto abort_export;
3154
3155         /*
3156          * The rest should better be atomic, we can have disk failures
3157          * noticed in interrupt contexts ...
3158          */
3159
3160         if (rdev->desc_nr == mddev->max_disks) {
3161                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3162                         mdname(mddev));
3163                 err = -EBUSY;
3164                 goto abort_unbind_export;
3165         }
3166
3167         rdev->raid_disk = -1;
3168
3169         md_update_sb(mddev);
3170
3171         /*
3172          * Kick recovery, maybe this spare has to be added to the
3173          * array immediately.
3174          */
3175         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3176         md_wakeup_thread(mddev->thread);
3177         md_new_event(mddev);
3178         return 0;
3179
3180 abort_unbind_export:
3181         unbind_rdev_from_array(rdev);
3182
3183 abort_export:
3184         export_rdev(rdev);
3185         return err;
3186 }
3187
3188 /* similar to deny_write_access, but accounts for our holding a reference
3189  * to the file ourselves */
3190 static int deny_bitmap_write_access(struct file * file)
3191 {
3192         struct inode *inode = file->f_mapping->host;
3193
3194         spin_lock(&inode->i_lock);
3195         if (atomic_read(&inode->i_writecount) > 1) {
3196                 spin_unlock(&inode->i_lock);
3197                 return -ETXTBSY;
3198         }
3199         atomic_set(&inode->i_writecount, -1);
3200         spin_unlock(&inode->i_lock);
3201
3202         return 0;
3203 }
3204
3205 static int set_bitmap_file(mddev_t *mddev, int fd)
3206 {
3207         int err;
3208
3209         if (mddev->pers) {
3210                 if (!mddev->pers->quiesce)
3211                         return -EBUSY;
3212                 if (mddev->recovery || mddev->sync_thread)
3213                         return -EBUSY;
3214                 /* we should be able to change the bitmap.. */
3215         }
3216
3217
3218         if (fd >= 0) {
3219                 if (mddev->bitmap)
3220                         return -EEXIST; /* cannot add when bitmap is present */
3221                 mddev->bitmap_file = fget(fd);
3222
3223                 if (mddev->bitmap_file == NULL) {
3224                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3225                                mdname(mddev));
3226                         return -EBADF;
3227                 }
3228
3229                 err = deny_bitmap_write_access(mddev->bitmap_file);
3230                 if (err) {
3231                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3232                                mdname(mddev));
3233                         fput(mddev->bitmap_file);
3234                         mddev->bitmap_file = NULL;
3235                         return err;
3236                 }
3237                 mddev->bitmap_offset = 0; /* file overrides offset */
3238         } else if (mddev->bitmap == NULL)
3239                 return -ENOENT; /* cannot remove what isn't there */
3240         err = 0;
3241         if (mddev->pers) {
3242                 mddev->pers->quiesce(mddev, 1);
3243                 if (fd >= 0)
3244                         err = bitmap_create(mddev);
3245                 if (fd < 0 || err)
3246                         bitmap_destroy(mddev);
3247                 mddev->pers->quiesce(mddev, 0);
3248         } else if (fd < 0) {
3249                 if (mddev->bitmap_file)
3250                         fput(mddev->bitmap_file);
3251                 mddev->bitmap_file = NULL;
3252         }
3253
3254         return err;
3255 }
3256
3257 /*
3258  * set_array_info is used two different ways
3259  * The original usage is when creating a new array.
3260  * In this usage, raid_disks is > 0 and it together with
3261  *  level, size, not_persistent,layout,chunksize determine the
3262  *  shape of the array.
3263  *  This will always create an array with a type-0.90.0 superblock.
3264  * The newer usage is when assembling an array.
3265  *  In this case raid_disks will be 0, and the major_version field is
3266  *  use to determine which style super-blocks are to be found on the devices.
3267  *  The minor and patch _version numbers are also kept incase the
3268  *  super_block handler wishes to interpret them.
3269  */
3270 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3271 {
3272
3273         if (info->raid_disks == 0) {
3274                 /* just setting version number for superblock loading */
3275                 if (info->major_version < 0 ||
3276                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3277                     super_types[info->major_version].name == NULL) {
3278                         /* maybe try to auto-load a module? */
3279                         printk(KERN_INFO
3280                                 "md: superblock version %d not known\n",
3281                                 info->major_version);
3282                         return -EINVAL;
3283                 }
3284                 mddev->major_version = info->major_version;
3285                 mddev->minor_version = info->minor_version;
3286                 mddev->patch_version = info->patch_version;
3287                 return 0;
3288         }
3289         mddev->major_version = MD_MAJOR_VERSION;
3290         mddev->minor_version = MD_MINOR_VERSION;
3291         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3292         mddev->ctime         = get_seconds();
3293
3294         mddev->level         = info->level;
3295         mddev->size          = info->size;
3296         mddev->raid_disks    = info->raid_disks;
3297         /* don't set md_minor, it is determined by which /dev/md* was
3298          * openned
3299          */
3300         if (info->state & (1<<MD_SB_CLEAN))
3301                 mddev->recovery_cp = MaxSector;
3302         else
3303                 mddev->recovery_cp = 0;
3304         mddev->persistent    = ! info->not_persistent;
3305
3306         mddev->layout        = info->layout;
3307         mddev->chunk_size    = info->chunk_size;
3308
3309         mddev->max_disks     = MD_SB_DISKS;
3310
3311         mddev->sb_dirty      = 1;
3312
3313         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3314         mddev->bitmap_offset = 0;
3315
3316         /*
3317          * Generate a 128 bit UUID
3318          */
3319         get_random_bytes(mddev->uuid, 16);
3320
3321         return 0;
3322 }
3323
3324 static int update_size(mddev_t *mddev, unsigned long size)
3325 {
3326         mdk_rdev_t * rdev;
3327         int rv;
3328         struct list_head *tmp;
3329
3330         if (mddev->pers->resize == NULL)
3331                 return -EINVAL;
3332         /* The "size" is the amount of each device that is used.
3333          * This can only make sense for arrays with redundancy.
3334          * linear and raid0 always use whatever space is available
3335          * We can only consider changing the size if no resync
3336          * or reconstruction is happening, and if the new size
3337          * is acceptable. It must fit before the sb_offset or,
3338          * if that is <data_offset, it must fit before the
3339          * size of each device.
3340          * If size is zero, we find the largest size that fits.
3341          */
3342         if (mddev->sync_thread)
3343                 return -EBUSY;
3344         ITERATE_RDEV(mddev,rdev,tmp) {
3345                 sector_t avail;
3346                 int fit = (size == 0);
3347                 if (rdev->sb_offset > rdev->data_offset)
3348                         avail = (rdev->sb_offset*2) - rdev->data_offset;
3349                 else
3350                         avail = get_capacity(rdev->bdev->bd_disk)
3351                                 - rdev->data_offset;
3352                 if (fit && (size == 0 || size > avail/2))
3353                         size = avail/2;
3354                 if (avail < ((sector_t)size << 1))
3355                         return -ENOSPC;
3356         }
3357         rv = mddev->pers->resize(mddev, (sector_t)size *2);
3358         if (!rv) {
3359                 struct block_device *bdev;
3360
3361                 bdev = bdget_disk(mddev->gendisk, 0);
3362                 if (bdev) {
3363                         down(&bdev->bd_inode->i_sem);
3364                         i_size_write(bdev->bd_inode, mddev->array_size << 10);
3365                         up(&bdev->bd_inode->i_sem);
3366                         bdput(bdev);
3367                 }
3368         }
3369         return rv;
3370 }
3371
3372 static int update_raid_disks(mddev_t *mddev, int raid_disks)
3373 {
3374         int rv;
3375         /* change the number of raid disks */
3376         if (mddev->pers->reshape == NULL)
3377                 return -EINVAL;
3378         if (raid_disks <= 0 ||
3379             raid_disks >= mddev->max_disks)
3380                 return -EINVAL;
3381         if (mddev->sync_thread)
3382                 return -EBUSY;
3383         rv = mddev->pers->reshape(mddev, raid_disks);
3384         if (!rv) {
3385                 struct block_device *bdev;
3386
3387                 bdev = bdget_disk(mddev->gendisk, 0);
3388                 if (bdev) {
3389                         down(&bdev->bd_inode->i_sem);
3390                         i_size_write(bdev->bd_inode, mddev->array_size << 10);
3391                         up(&bdev->bd_inode->i_sem);
3392                         bdput(bdev);
3393                 }
3394         }
3395         return rv;
3396 }
3397
3398
3399 /*
3400  * update_array_info is used to change the configuration of an
3401  * on-line array.
3402  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3403  * fields in the info are checked against the array.
3404  * Any differences that cannot be handled will cause an error.
3405  * Normally, only one change can be managed at a time.
3406  */
3407 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3408 {
3409         int rv = 0;
3410         int cnt = 0;
3411         int state = 0;
3412
3413         /* calculate expected state,ignoring low bits */
3414         if (mddev->bitmap && mddev->bitmap_offset)
3415                 state |= (1 << MD_SB_BITMAP_PRESENT);
3416
3417         if (mddev->major_version != info->major_version ||
3418             mddev->minor_version != info->minor_version ||
3419 /*          mddev->patch_version != info->patch_version || */
3420             mddev->ctime         != info->ctime         ||
3421             mddev->level         != info->level         ||
3422 /*          mddev->layout        != info->layout        || */
3423             !mddev->persistent   != info->not_persistent||
3424             mddev->chunk_size    != info->chunk_size    ||
3425             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3426             ((state^info->state) & 0xfffffe00)
3427                 )
3428                 return -EINVAL;
3429         /* Check there is only one change */
3430         if (mddev->size != info->size) cnt++;
3431         if (mddev->raid_disks != info->raid_disks) cnt++;
3432         if (mddev->layout != info->layout) cnt++;
3433         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3434         if (cnt == 0) return 0;
3435         if (cnt > 1) return -EINVAL;
3436
3437         if (mddev->layout != info->layout) {
3438                 /* Change layout
3439                  * we don't need to do anything at the md level, the
3440                  * personality will take care of it all.
3441                  */
3442                 if (mddev->pers->reconfig == NULL)
3443                         return -EINVAL;
3444                 else
3445                         return mddev->pers->reconfig(mddev, info->layout, -1);
3446         }
3447         if (mddev->size != info->size)
3448                 rv = update_size(mddev, info->size);
3449
3450         if (mddev->raid_disks    != info->raid_disks)
3451                 rv = update_raid_disks(mddev, info->raid_disks);
3452
3453         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3454                 if (mddev->pers->quiesce == NULL)
3455                         return -EINVAL;
3456                 if (mddev->recovery || mddev->sync_thread)
3457                         return -EBUSY;
3458                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3459                         /* add the bitmap */
3460                         if (mddev->bitmap)
3461                                 return -EEXIST;
3462                         if (mddev->default_bitmap_offset == 0)
3463                                 return -EINVAL;
3464                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3465                         mddev->pers->quiesce(mddev, 1);
3466                         rv = bitmap_create(mddev);
3467                         if (rv)
3468                                 bitmap_destroy(mddev);
3469                         mddev->pers->quiesce(mddev, 0);
3470                 } else {
3471                         /* remove the bitmap */
3472                         if (!mddev->bitmap)
3473                                 return -ENOENT;
3474                         if (mddev->bitmap->file)
3475                                 return -EINVAL;
3476                         mddev->pers->quiesce(mddev, 1);
3477                         bitmap_destroy(mddev);
3478                         mddev->pers->quiesce(mddev, 0);
3479                         mddev->bitmap_offset = 0;
3480                 }
3481         }
3482         md_update_sb(mddev);
3483         return rv;
3484 }
3485
3486 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3487 {
3488         mdk_rdev_t *rdev;
3489
3490         if (mddev->pers == NULL)
3491                 return -ENODEV;
3492
3493         rdev = find_rdev(mddev, dev);
3494         if (!rdev)
3495                 return -ENODEV;
3496
3497         md_error(mddev, rdev);
3498         return 0;
3499 }
3500
3501 static int md_ioctl(struct inode *inode, struct file *file,
3502                         unsigned int cmd, unsigned long arg)
3503 {
3504         int err = 0;
3505         void __user *argp = (void __user *)arg;
3506         struct hd_geometry __user *loc = argp;
3507         mddev_t *mddev = NULL;
3508
3509         if (!capable(CAP_SYS_ADMIN))
3510                 return -EACCES;
3511
3512         /*
3513          * Commands dealing with the RAID driver but not any
3514          * particular array:
3515          */
3516         switch (cmd)
3517         {
3518                 case RAID_VERSION:
3519                         err = get_version(argp);
3520                         goto done;
3521
3522                 case PRINT_RAID_DEBUG:
3523                         err = 0;
3524                         md_print_devices();
3525                         goto done;
3526
3527 #ifndef MODULE
3528                 case RAID_AUTORUN:
3529                         err = 0;
3530                         autostart_arrays(arg);
3531                         goto done;
3532 #endif
3533                 default:;
3534         }
3535
3536         /*
3537          * Commands creating/starting a new array:
3538          */
3539
3540         mddev = inode->i_bdev->bd_disk->private_data;
3541
3542         if (!mddev) {
3543                 BUG();
3544                 goto abort;
3545         }
3546
3547
3548         if (cmd == START_ARRAY) {
3549                 /* START_ARRAY doesn't need to lock the array as autostart_array
3550                  * does the locking, and it could even be a different array
3551                  */
3552                 static int cnt = 3;
3553                 if (cnt > 0 ) {
3554                         printk(KERN_WARNING
3555                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3556                                "This will not be supported beyond July 2006\n",
3557                                current->comm, current->pid);
3558                         cnt--;
3559                 }
3560                 err = autostart_array(new_decode_dev(arg));
3561                 if (err) {
3562                         printk(KERN_WARNING "md: autostart failed!\n");
3563                         goto abort;
3564                 }
3565                 goto done;
3566         }
3567
3568         err = mddev_lock(mddev);
3569         if (err) {
3570                 printk(KERN_INFO
3571                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3572                         err, cmd);
3573                 goto abort;
3574         }
3575
3576         switch (cmd)
3577         {
3578                 case SET_ARRAY_INFO:
3579                         {
3580                                 mdu_array_info_t info;
3581                                 if (!arg)
3582                                         memset(&info, 0, sizeof(info));
3583                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3584                                         err = -EFAULT;
3585                                         goto abort_unlock;
3586                                 }
3587                                 if (mddev->pers) {
3588                                         err = update_array_info(mddev, &info);
3589                                         if (err) {
3590                                                 printk(KERN_WARNING "md: couldn't update"
3591                                                        " array info. %d\n", err);
3592                                                 goto abort_unlock;
3593                                         }
3594                                         goto done_unlock;
3595                                 }
3596                                 if (!list_empty(&mddev->disks)) {
3597                                         printk(KERN_WARNING
3598                                                "md: array %s already has disks!\n",
3599                                                mdname(mddev));
3600                                         err = -EBUSY;
3601                                         goto abort_unlock;
3602                                 }
3603                                 if (mddev->raid_disks) {
3604                                         printk(KERN_WARNING
3605                                                "md: array %s already initialised!\n",
3606                                                mdname(mddev));
3607                                         err = -EBUSY;
3608                                         goto abort_unlock;
3609                                 }
3610                                 err = set_array_info(mddev, &info);
3611                                 if (err) {
3612                                         printk(KERN_WARNING "md: couldn't set"
3613                                                " array info. %d\n", err);
3614                                         goto abort_unlock;
3615                                 }
3616                         }
3617                         goto done_unlock;
3618
3619                 default:;
3620         }
3621
3622         /*
3623          * Commands querying/configuring an existing array:
3624          */
3625         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3626          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3627         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3628                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3629                 err = -ENODEV;
3630                 goto abort_unlock;
3631         }
3632
3633         /*
3634          * Commands even a read-only array can execute:
3635          */
3636         switch (cmd)
3637         {
3638                 case GET_ARRAY_INFO:
3639                         err = get_array_info(mddev, argp);
3640                         goto done_unlock;
3641
3642                 case GET_BITMAP_FILE:
3643                         err = get_bitmap_file(mddev, argp);
3644                         goto done_unlock;
3645
3646                 case GET_DISK_INFO:
3647                         err = get_disk_info(mddev, argp);
3648                         goto done_unlock;
3649
3650                 case RESTART_ARRAY_RW:
3651                         err = restart_array(mddev);
3652                         goto done_unlock;
3653
3654                 case STOP_ARRAY:
3655                         err = do_md_stop (mddev, 0);
3656                         goto done_unlock;
3657
3658                 case STOP_ARRAY_RO:
3659                         err = do_md_stop (mddev, 1);
3660                         goto done_unlock;
3661
3662         /*
3663          * We have a problem here : there is no easy way to give a CHS
3664          * virtual geometry. We currently pretend that we have a 2 heads
3665          * 4 sectors (with a BIG number of cylinders...). This drives
3666          * dosfs just mad... ;-)
3667          */
3668                 case HDIO_GETGEO:
3669                         if (!loc) {
3670                                 err = -EINVAL;
3671                                 goto abort_unlock;
3672                         }
3673                         err = put_user (2, (char __user *) &loc->heads);
3674                         if (err)
3675                                 goto abort_unlock;
3676                         err = put_user (4, (char __user *) &loc->sectors);
3677                         if (err)
3678                                 goto abort_unlock;
3679                         err = put_user(get_capacity(mddev->gendisk)/8,
3680                                         (short __user *) &loc->cylinders);
3681                         if (err)
3682                                 goto abort_unlock;
3683                         err = put_user (get_start_sect(inode->i_bdev),
3684                                                 (long __user *) &loc->start);
3685                         goto done_unlock;
3686         }
3687
3688         /*
3689          * The remaining ioctls are changing the state of the
3690          * superblock, so we do not allow them on read-only arrays.
3691          * However non-MD ioctls (e.g. get-size) will still come through
3692          * here and hit the 'default' below, so only disallow
3693          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3694          */
3695         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3696             mddev->ro && mddev->pers) {
3697                 if (mddev->ro == 2) {
3698                         mddev->ro = 0;
3699                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3700                 md_wakeup_thread(mddev->thread);
3701
3702                 } else {
3703                         err = -EROFS;
3704                         goto abort_unlock;
3705                 }
3706         }
3707
3708         switch (cmd)
3709         {
3710                 case ADD_NEW_DISK:
3711                 {
3712                         mdu_disk_info_t info;
3713                         if (copy_from_user(&info, argp, sizeof(info)))
3714                                 err = -EFAULT;
3715                         else
3716                                 err = add_new_disk(mddev, &info);
3717                         goto done_unlock;
3718                 }
3719
3720                 case HOT_REMOVE_DISK:
3721                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3722                         goto done_unlock;
3723
3724                 case HOT_ADD_DISK:
3725                         err = hot_add_disk(mddev, new_decode_dev(arg));
3726                         goto done_unlock;
3727
3728                 case SET_DISK_FAULTY:
3729                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3730                         goto done_unlock;
3731
3732                 case RUN_ARRAY:
3733                         err = do_md_run (mddev);
3734                         goto done_unlock;
3735
3736                 case SET_BITMAP_FILE:
3737                         err = set_bitmap_file(mddev, (int)arg);
3738                         goto done_unlock;
3739
3740                 default:
3741                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3742                                 printk(KERN_WARNING "md: %s(pid %d) used"
3743                                         " obsolete MD ioctl, upgrade your"
3744                                         " software to use new ictls.\n",
3745                                         current->comm, current->pid);
3746                         err = -EINVAL;
3747                         goto abort_unlock;
3748         }
3749
3750 done_unlock:
3751 abort_unlock:
3752         mddev_unlock(mddev);
3753
3754         return err;
3755 done:
3756         if (err)
3757                 MD_BUG();
3758 abort:
3759         return err;
3760 }
3761
3762 static int md_open(struct inode *inode, struct file *file)
3763 {
3764         /*
3765          * Succeed if we can lock the mddev, which confirms that
3766          * it isn't being stopped right now.
3767          */
3768         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3769         int err;
3770
3771         if ((err = mddev_lock(mddev)))
3772                 goto out;
3773
3774         err = 0;
3775         mddev_get(mddev);
3776         mddev_unlock(mddev);
3777
3778         check_disk_change(inode->i_bdev);
3779  out:
3780         return err;
3781 }
3782
3783 static int md_release(struct inode *inode, struct file * file)
3784 {
3785         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3786
3787         if (!mddev)
3788                 BUG();
3789         mddev_put(mddev);
3790
3791         return 0;
3792 }
3793
3794 static int md_media_changed(struct gendisk *disk)
3795 {
3796         mddev_t *mddev = disk->private_data;
3797
3798         return mddev->changed;
3799 }
3800
3801 static int md_revalidate(struct gendisk *disk)
3802 {
3803         mddev_t *mddev = disk->private_data;
3804
3805         mddev->changed = 0;
3806         return 0;
3807 }
3808 static struct block_device_operations md_fops =
3809 {
3810         .owner          = THIS_MODULE,
3811         .open           = md_open,
3812         .release        = md_release,
3813         .ioctl          = md_ioctl,
3814         .media_changed  = md_media_changed,
3815         .revalidate_disk= md_revalidate,
3816 };
3817
3818 static int md_thread(void * arg)
3819 {
3820         mdk_thread_t *thread = arg;
3821
3822         /*
3823          * md_thread is a 'system-thread', it's priority should be very
3824          * high. We avoid resource deadlocks individually in each
3825          * raid personality. (RAID5 does preallocation) We also use RR and
3826          * the very same RT priority as kswapd, thus we will never get
3827          * into a priority inversion deadlock.
3828          *
3829          * we definitely have to have equal or higher priority than
3830          * bdflush, otherwise bdflush will deadlock if there are too
3831          * many dirty RAID5 blocks.
3832          */
3833
3834         allow_signal(SIGKILL);
3835         while (!kthread_should_stop()) {
3836
3837                 /* We need to wait INTERRUPTIBLE so that
3838                  * we don't add to the load-average.
3839                  * That means we need to be sure no signals are
3840                  * pending
3841                  */
3842                 if (signal_pending(current))
3843                         flush_signals(current);
3844
3845                 wait_event_interruptible_timeout
3846                         (thread->wqueue,
3847                          test_bit(THREAD_WAKEUP, &thread->flags)
3848                          || kthread_should_stop(),
3849                          thread->timeout);
3850                 try_to_freeze();
3851
3852                 clear_bit(THREAD_WAKEUP, &thread->flags);
3853
3854                 thread->run(thread->mddev);
3855         }
3856
3857         return 0;
3858 }
3859
3860 void md_wakeup_thread(mdk_thread_t *thread)
3861 {
3862         if (thread) {
3863                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3864                 set_bit(THREAD_WAKEUP, &thread->flags);
3865                 wake_up(&thread->wqueue);
3866         }
3867 }
3868
3869 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3870                                  const char *name)
3871 {
3872         mdk_thread_t *thread;
3873
3874         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3875         if (!thread)
3876                 return NULL;
3877
3878         init_waitqueue_head(&thread->wqueue);
3879
3880         thread->run = run;
3881         thread->mddev = mddev;
3882         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3883         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3884         if (IS_ERR(thread->tsk)) {
3885                 kfree(thread);
3886                 return NULL;
3887         }
3888         return thread;
3889 }
3890
3891 void md_unregister_thread(mdk_thread_t *thread)
3892 {
3893         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3894
3895         kthread_stop(thread->tsk);
3896         kfree(thread);
3897 }
3898
3899 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3900 {
3901         if (!mddev) {
3902                 MD_BUG();
3903                 return;
3904         }
3905
3906         if (!rdev || test_bit(Faulty, &rdev->flags))
3907                 return;
3908 /*
3909         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3910                 mdname(mddev),
3911                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3912                 __builtin_return_address(0),__builtin_return_address(1),
3913                 __builtin_return_address(2),__builtin_return_address(3));
3914 */
3915         if (!mddev->pers->error_handler)
3916                 return;
3917         mddev->pers->error_handler(mddev,rdev);
3918         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3919         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3920         md_wakeup_thread(mddev->thread);
3921         md_new_event(mddev);
3922 }
3923
3924 /* seq_file implementation /proc/mdstat */
3925
3926 static void status_unused(struct seq_file *seq)
3927 {
3928         int i = 0;
3929         mdk_rdev_t *rdev;
3930         struct list_head *tmp;
3931
3932         seq_printf(seq, "unused devices: ");
3933
3934         ITERATE_RDEV_PENDING(rdev,tmp) {
3935                 char b[BDEVNAME_SIZE];
3936                 i++;
3937                 seq_printf(seq, "%s ",
3938                               bdevname(rdev->bdev,b));
3939         }
3940         if (!i)
3941                 seq_printf(seq, "<none>");
3942
3943         seq_printf(seq, "\n");
3944 }
3945
3946
3947 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3948 {
3949         unsigned long max_blocks, resync, res, dt, db, rt;
3950
3951         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3952
3953         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3954                 max_blocks = mddev->resync_max_sectors >> 1;
3955         else
3956                 max_blocks = mddev->size;
3957
3958         /*
3959          * Should not happen.
3960          */
3961         if (!max_blocks) {
3962                 MD_BUG();
3963                 return;
3964         }
3965         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3966         {
3967                 int i, x = res/50, y = 20-x;
3968                 seq_printf(seq, "[");
3969                 for (i = 0; i < x; i++)
3970                         seq_printf(seq, "=");
3971                 seq_printf(seq, ">");
3972                 for (i = 0; i < y; i++)
3973                         seq_printf(seq, ".");
3974                 seq_printf(seq, "] ");
3975         }
3976         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3977                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3978                        "resync" : "recovery"),
3979                       res/10, res % 10, resync, max_blocks);
3980
3981         /*
3982          * We do not want to overflow, so the order of operands and
3983          * the * 100 / 100 trick are important. We do a +1 to be
3984          * safe against division by zero. We only estimate anyway.
3985          *
3986          * dt: time from mark until now
3987          * db: blocks written from mark until now
3988          * rt: remaining time
3989          */
3990         dt = ((jiffies - mddev->resync_mark) / HZ);
3991         if (!dt) dt++;
3992         db = resync - (mddev->resync_mark_cnt/2);
3993         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3994
3995         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3996
3997         seq_printf(seq, " speed=%ldK/sec", db/dt);
3998 }
3999
4000 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4001 {
4002         struct list_head *tmp;
4003         loff_t l = *pos;
4004         mddev_t *mddev;
4005
4006         if (l >= 0x10000)
4007                 return NULL;
4008         if (!l--)
4009                 /* header */
4010                 return (void*)1;
4011
4012         spin_lock(&all_mddevs_lock);
4013         list_for_each(tmp,&all_mddevs)
4014                 if (!l--) {
4015                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4016                         mddev_get(mddev);
4017                         spin_unlock(&all_mddevs_lock);
4018                         return mddev;
4019                 }
4020         spin_unlock(&all_mddevs_lock);
4021         if (!l--)
4022                 return (void*)2;/* tail */
4023         return NULL;
4024 }
4025
4026 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4027 {
4028         struct list_head *tmp;
4029         mddev_t *next_mddev, *mddev = v;
4030
4031         ++*pos;
4032         if (v == (void*)2)
4033                 return NULL;
4034
4035         spin_lock(&all_mddevs_lock);
4036         if (v == (void*)1)
4037                 tmp = all_mddevs.next;
4038         else
4039                 tmp = mddev->all_mddevs.next;
4040         if (tmp != &all_mddevs)
4041                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4042         else {
4043                 next_mddev = (void*)2;
4044                 *pos = 0x10000;
4045         }
4046         spin_unlock(&all_mddevs_lock);
4047
4048         if (v != (void*)1)
4049                 mddev_put(mddev);
4050         return next_mddev;
4051
4052 }
4053
4054 static void md_seq_stop(struct seq_file *seq, void *v)
4055 {
4056         mddev_t *mddev = v;
4057
4058         if (mddev && v != (void*)1 && v != (void*)2)
4059                 mddev_put(mddev);
4060 }
4061
4062 struct mdstat_info {
4063         int event;
4064 };
4065
4066 static int md_seq_show(struct seq_file *seq, void *v)
4067 {
4068         mddev_t *mddev = v;
4069         sector_t size;
4070         struct list_head *tmp2;
4071         mdk_rdev_t *rdev;
4072         struct mdstat_info *mi = seq->private;
4073         struct bitmap *bitmap;
4074
4075         if (v == (void*)1) {
4076                 struct mdk_personality *pers;
4077                 seq_printf(seq, "Personalities : ");
4078                 spin_lock(&pers_lock);
4079                 list_for_each_entry(pers, &pers_list, list)
4080                         seq_printf(seq, "[%s] ", pers->name);
4081
4082                 spin_unlock(&pers_lock);
4083                 seq_printf(seq, "\n");
4084                 mi->event = atomic_read(&md_event_count);
4085                 return 0;
4086         }
4087         if (v == (void*)2) {
4088                 status_unused(seq);
4089                 return 0;
4090         }
4091
4092         if (mddev_lock(mddev)!=0)
4093                 return -EINTR;
4094         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4095                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4096                                                 mddev->pers ? "" : "in");
4097                 if (mddev->pers) {
4098                         if (mddev->ro==1)
4099                                 seq_printf(seq, " (read-only)");
4100                         if (mddev->ro==2)
4101                                 seq_printf(seq, "(auto-read-only)");
4102                         seq_printf(seq, " %s", mddev->pers->name);
4103                 }
4104
4105                 size = 0;
4106                 ITERATE_RDEV(mddev,rdev,tmp2) {
4107                         char b[BDEVNAME_SIZE];
4108                         seq_printf(seq, " %s[%d]",
4109                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4110                         if (test_bit(WriteMostly, &rdev->flags))
4111                                 seq_printf(seq, "(W)");
4112                         if (test_bit(Faulty, &rdev->flags)) {
4113                                 seq_printf(seq, "(F)");
4114                                 continue;
4115                         } else if (rdev->raid_disk < 0)
4116                                 seq_printf(seq, "(S)"); /* spare */
4117                         size += rdev->size;
4118                 }
4119
4120                 if (!list_empty(&mddev->disks)) {
4121                         if (mddev->pers)
4122                                 seq_printf(seq, "\n      %llu blocks",
4123                                         (unsigned long long)mddev->array_size);
4124                         else
4125                                 seq_printf(seq, "\n      %llu blocks",
4126                                         (unsigned long long)size);
4127                 }
4128                 if (mddev->persistent) {
4129                         if (mddev->major_version != 0 ||
4130                             mddev->minor_version != 90) {
4131                                 seq_printf(seq," super %d.%d",
4132                                            mddev->major_version,
4133                                            mddev->minor_version);
4134                         }
4135                 } else
4136                         seq_printf(seq, " super non-persistent");
4137
4138                 if (mddev->pers) {
4139                         mddev->pers->status (seq, mddev);
4140                         seq_printf(seq, "\n      ");
4141                         if (mddev->pers->sync_request) {
4142                                 if (mddev->curr_resync > 2) {
4143                                         status_resync (seq, mddev);
4144                                         seq_printf(seq, "\n      ");
4145                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4146                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4147                                 else if (mddev->recovery_cp < MaxSector)
4148                                         seq_printf(seq, "\tresync=PENDING\n      ");
4149                         }
4150                 } else
4151                         seq_printf(seq, "\n       ");
4152
4153                 if ((bitmap = mddev->bitmap)) {
4154                         unsigned long chunk_kb;
4155                         unsigned long flags;
4156                         spin_lock_irqsave(&bitmap->lock, flags);
4157                         chunk_kb = bitmap->chunksize >> 10;
4158                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4159                                 "%lu%s chunk",
4160                                 bitmap->pages - bitmap->missing_pages,
4161                                 bitmap->pages,
4162                                 (bitmap->pages - bitmap->missing_pages)
4163                                         << (PAGE_SHIFT - 10),
4164                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4165                                 chunk_kb ? "KB" : "B");
4166                         if (bitmap->file) {
4167                                 seq_printf(seq, ", file: ");
4168                                 seq_path(seq, bitmap->file->f_vfsmnt,
4169                                          bitmap->file->f_dentry," \t\n");
4170                         }
4171
4172                         seq_printf(seq, "\n");
4173                         spin_unlock_irqrestore(&bitmap->lock, flags);
4174                 }
4175
4176                 seq_printf(seq, "\n");
4177         }
4178         mddev_unlock(mddev);
4179
4180         return 0;
4181 }
4182
4183 static struct seq_operations md_seq_ops = {
4184         .start  = md_seq_start,
4185         .next   = md_seq_next,
4186         .stop   = md_seq_stop,
4187         .show   = md_seq_show,
4188 };
4189
4190 static int md_seq_open(struct inode *inode, struct file *file)
4191 {
4192         int error;
4193         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4194         if (mi == NULL)
4195                 return -ENOMEM;
4196
4197         error = seq_open(file, &md_seq_ops);
4198         if (error)
4199                 kfree(mi);
4200         else {
4201                 struct seq_file *p = file->private_data;
4202                 p->private = mi;
4203                 mi->event = atomic_read(&md_event_count);
4204         }
4205         return error;
4206 }
4207
4208 static int md_seq_release(struct inode *inode, struct file *file)
4209 {
4210         struct seq_file *m = file->private_data;
4211         struct mdstat_info *mi = m->private;
4212         m->private = NULL;
4213         kfree(mi);
4214         return seq_release(inode, file);
4215 }
4216
4217 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4218 {
4219         struct seq_file *m = filp->private_data;
4220         struct mdstat_info *mi = m->private;
4221         int mask;
4222
4223         poll_wait(filp, &md_event_waiters, wait);
4224
4225         /* always allow read */
4226         mask = POLLIN | POLLRDNORM;
4227
4228         if (mi->event != atomic_read(&md_event_count))
4229                 mask |= POLLERR | POLLPRI;
4230         return mask;
4231 }
4232
4233 static struct file_operations md_seq_fops = {
4234         .open           = md_seq_open,
4235         .read           = seq_read,
4236         .llseek         = seq_lseek,
4237         .release        = md_seq_release,
4238         .poll           = mdstat_poll,
4239 };
4240
4241 int register_md_personality(struct mdk_personality *p)
4242 {
4243         spin_lock(&pers_lock);
4244         list_add_tail(&p->list, &pers_list);
4245         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4246         spin_unlock(&pers_lock);
4247         return 0;
4248 }
4249
4250 int unregister_md_personality(struct mdk_personality *p)
4251 {
4252         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4253         spin_lock(&pers_lock);
4254         list_del_init(&p->list);
4255         spin_unlock(&pers_lock);
4256         return 0;
4257 }
4258
4259 static int is_mddev_idle(mddev_t *mddev)
4260 {
4261         mdk_rdev_t * rdev;
4262         struct list_head *tmp;
4263         int idle;
4264         unsigned long curr_events;
4265
4266         idle = 1;
4267         ITERATE_RDEV(mddev,rdev,tmp) {
4268                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4269                 curr_events = disk_stat_read(disk, sectors[0]) +
4270                                 disk_stat_read(disk, sectors[1]) -
4271                                 atomic_read(&disk->sync_io);
4272                 /* The difference between curr_events and last_events
4273                  * will be affected by any new non-sync IO (making
4274                  * curr_events bigger) and any difference in the amount of
4275                  * in-flight syncio (making current_events bigger or smaller)
4276                  * The amount in-flight is currently limited to
4277                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4278                  * which is at most 4096 sectors.
4279                  * These numbers are fairly fragile and should be made
4280                  * more robust, probably by enforcing the
4281                  * 'window size' that md_do_sync sort-of uses.
4282                  *
4283                  * Note: the following is an unsigned comparison.
4284                  */
4285                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4286                         rdev->last_events = curr_events;
4287                         idle = 0;
4288                 }
4289         }
4290         return idle;
4291 }
4292
4293 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4294 {
4295         /* another "blocks" (512byte) blocks have been synced */
4296         atomic_sub(blocks, &mddev->recovery_active);
4297         wake_up(&mddev->recovery_wait);
4298         if (!ok) {
4299                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4300                 md_wakeup_thread(mddev->thread);
4301                 // stop recovery, signal do_sync ....
4302         }
4303 }
4304
4305
4306 /* md_write_start(mddev, bi)
4307  * If we need to update some array metadata (e.g. 'active' flag
4308  * in superblock) before writing, schedule a superblock update
4309  * and wait for it to complete.
4310  */
4311 void md_write_start(mddev_t *mddev, struct bio *bi)
4312 {
4313         if (bio_data_dir(bi) != WRITE)
4314                 return;
4315
4316         BUG_ON(mddev->ro == 1);
4317         if (mddev->ro == 2) {
4318                 /* need to switch to read/write */
4319                 mddev->ro = 0;
4320                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4321                 md_wakeup_thread(mddev->thread);
4322         }
4323         atomic_inc(&mddev->writes_pending);
4324         if (mddev->in_sync) {
4325                 spin_lock_irq(&mddev->write_lock);
4326                 if (mddev->in_sync) {
4327                         mddev->in_sync = 0;
4328                         mddev->sb_dirty = 1;
4329                         md_wakeup_thread(mddev->thread);
4330                 }
4331                 spin_unlock_irq(&mddev->write_lock);
4332         }
4333         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4334 }
4335
4336 void md_write_end(mddev_t *mddev)
4337 {
4338         if (atomic_dec_and_test(&mddev->writes_pending)) {
4339                 if (mddev->safemode == 2)
4340                         md_wakeup_thread(mddev->thread);
4341                 else
4342                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4343         }
4344 }
4345
4346 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4347
4348 #define SYNC_MARKS      10
4349 #define SYNC_MARK_STEP  (3*HZ)
4350 static void md_do_sync(mddev_t *mddev)
4351 {
4352         mddev_t *mddev2;
4353         unsigned int currspeed = 0,
4354                  window;
4355         sector_t max_sectors,j, io_sectors;
4356         unsigned long mark[SYNC_MARKS];
4357         sector_t mark_cnt[SYNC_MARKS];
4358         int last_mark,m;
4359         struct list_head *tmp;
4360         sector_t last_check;
4361         int skipped = 0;
4362
4363         /* just incase thread restarts... */
4364         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4365                 return;
4366
4367         /* we overload curr_resync somewhat here.
4368          * 0 == not engaged in resync at all
4369          * 2 == checking that there is no conflict with another sync
4370          * 1 == like 2, but have yielded to allow conflicting resync to
4371          *              commense
4372          * other == active in resync - this many blocks
4373          *
4374          * Before starting a resync we must have set curr_resync to
4375          * 2, and then checked that every "conflicting" array has curr_resync
4376          * less than ours.  When we find one that is the same or higher
4377          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4378          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4379          * This will mean we have to start checking from the beginning again.
4380          *
4381          */
4382
4383         do {
4384                 mddev->curr_resync = 2;
4385
4386         try_again:
4387                 if (kthread_should_stop()) {
4388                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4389                         goto skip;
4390                 }
4391                 ITERATE_MDDEV(mddev2,tmp) {
4392                         if (mddev2 == mddev)
4393                                 continue;
4394                         if (mddev2->curr_resync &&
4395                             match_mddev_units(mddev,mddev2)) {
4396                                 DEFINE_WAIT(wq);
4397                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4398                                         /* arbitrarily yield */
4399                                         mddev->curr_resync = 1;
4400                                         wake_up(&resync_wait);
4401                                 }
4402                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4403                                         /* no need to wait here, we can wait the next
4404                                          * time 'round when curr_resync == 2
4405                                          */
4406                                         continue;
4407                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4408                                 if (!kthread_should_stop() &&
4409                                     mddev2->curr_resync >= mddev->curr_resync) {
4410                                         printk(KERN_INFO "md: delaying resync of %s"
4411                                                " until %s has finished resync (they"
4412                                                " share one or more physical units)\n",
4413                                                mdname(mddev), mdname(mddev2));
4414                                         mddev_put(mddev2);
4415                                         schedule();
4416                                         finish_wait(&resync_wait, &wq);
4417                                         goto try_again;
4418                                 }
4419                                 finish_wait(&resync_wait, &wq);
4420                         }
4421                 }
4422         } while (mddev->curr_resync < 2);
4423
4424         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4425                 /* resync follows the size requested by the personality,
4426                  * which defaults to physical size, but can be virtual size
4427                  */
4428                 max_sectors = mddev->resync_max_sectors;
4429                 mddev->resync_mismatches = 0;
4430         } else
4431                 /* recovery follows the physical size of devices */
4432                 max_sectors = mddev->size << 1;
4433
4434         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4435         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4436                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4437         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4438                "(but not more than %d KB/sec) for reconstruction.\n",
4439                sysctl_speed_limit_max);
4440
4441         is_mddev_idle(mddev); /* this also initializes IO event counters */
4442         /* we don't use the checkpoint if there's a bitmap */
4443         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4444             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4445                 j = mddev->recovery_cp;
4446         else
4447                 j = 0;
4448         io_sectors = 0;
4449         for (m = 0; m < SYNC_MARKS; m++) {
4450                 mark[m] = jiffies;
4451                 mark_cnt[m] = io_sectors;
4452         }
4453         last_mark = 0;
4454         mddev->resync_mark = mark[last_mark];
4455         mddev->resync_mark_cnt = mark_cnt[last_mark];
4456
4457         /*
4458          * Tune reconstruction:
4459          */
4460         window = 32*(PAGE_SIZE/512);
4461         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4462                 window/2,(unsigned long long) max_sectors/2);
4463
4464         atomic_set(&mddev->recovery_active, 0);
4465         init_waitqueue_head(&mddev->recovery_wait);
4466         last_check = 0;
4467
4468         if (j>2) {
4469                 printk(KERN_INFO
4470                         "md: resuming recovery of %s from checkpoint.\n",
4471                         mdname(mddev));
4472                 mddev->curr_resync = j;
4473         }
4474
4475         while (j < max_sectors) {
4476                 sector_t sectors;
4477
4478                 skipped = 0;
4479                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4480                                             currspeed < sysctl_speed_limit_min);
4481                 if (sectors == 0) {
4482                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4483                         goto out;
4484                 }
4485
4486                 if (!skipped) { /* actual IO requested */
4487                         io_sectors += sectors;
4488                         atomic_add(sectors, &mddev->recovery_active);
4489                 }
4490
4491                 j += sectors;
4492                 if (j>1) mddev->curr_resync = j;
4493                 if (last_check == 0)
4494                         /* this is the earliers that rebuilt will be
4495                          * visible in /proc/mdstat
4496                          */
4497                         md_new_event(mddev);
4498
4499                 if (last_check + window > io_sectors || j == max_sectors)
4500                         continue;
4501
4502                 last_check = io_sectors;
4503
4504                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4505                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4506                         break;
4507
4508         repeat:
4509                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4510                         /* step marks */
4511                         int next = (last_mark+1) % SYNC_MARKS;
4512
4513                         mddev->resync_mark = mark[next];
4514                         mddev->resync_mark_cnt = mark_cnt[next];
4515                         mark[next] = jiffies;
4516                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4517                         last_mark = next;
4518                 }
4519
4520
4521                 if (kthread_should_stop()) {
4522                         /*
4523                          * got a signal, exit.
4524                          */
4525                         printk(KERN_INFO
4526                                 "md: md_do_sync() got signal ... exiting\n");
4527                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4528                         goto out;
4529                 }
4530
4531                 /*
4532                  * this loop exits only if either when we are slower than
4533                  * the 'hard' speed limit, or the system was IO-idle for
4534                  * a jiffy.
4535                  * the system might be non-idle CPU-wise, but we only care
4536                  * about not overloading the IO subsystem. (things like an
4537                  * e2fsck being done on the RAID array should execute fast)
4538                  */
4539                 mddev->queue->unplug_fn(mddev->queue);
4540                 cond_resched();
4541
4542                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4543                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4544
4545                 if (currspeed > sysctl_speed_limit_min) {
4546                         if ((currspeed > sysctl_speed_limit_max) ||
4547                                         !is_mddev_idle(mddev)) {
4548                                 msleep(500);
4549                                 goto repeat;
4550                         }
4551                 }
4552         }
4553         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4554         /*
4555          * this also signals 'finished resyncing' to md_stop
4556          */
4557  out:
4558         mddev->queue->unplug_fn(mddev->queue);
4559
4560         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4561
4562         /* tell personality that we are finished */
4563         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4564
4565         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4566             mddev->curr_resync > 2 &&
4567             mddev->curr_resync >= mddev->recovery_cp) {
4568                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4569                         printk(KERN_INFO
4570                                 "md: checkpointing recovery of %s.\n",
4571                                 mdname(mddev));
4572                         mddev->recovery_cp = mddev->curr_resync;
4573                 } else
4574                         mddev->recovery_cp = MaxSector;
4575         }
4576
4577  skip:
4578         mddev->curr_resync = 0;
4579         wake_up(&resync_wait);
4580         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4581         md_wakeup_thread(mddev->thread);
4582 }
4583
4584
4585 /*
4586  * This routine is regularly called by all per-raid-array threads to
4587  * deal with generic issues like resync and super-block update.
4588  * Raid personalities that don't have a thread (linear/raid0) do not
4589  * need this as they never do any recovery or update the superblock.
4590  *
4591  * It does not do any resync itself, but rather "forks" off other threads
4592  * to do that as needed.
4593  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4594  * "->recovery" and create a thread at ->sync_thread.
4595  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4596  * and wakeups up this thread which will reap the thread and finish up.
4597  * This thread also removes any faulty devices (with nr_pending == 0).
4598  *
4599  * The overall approach is:
4600  *  1/ if the superblock needs updating, update it.
4601  *  2/ If a recovery thread is running, don't do anything else.
4602  *  3/ If recovery has finished, clean up, possibly marking spares active.
4603  *  4/ If there are any faulty devices, remove them.
4604  *  5/ If array is degraded, try to add spares devices
4605  *  6/ If array has spares or is not in-sync, start a resync thread.
4606  */
4607 void md_check_recovery(mddev_t *mddev)
4608 {
4609         mdk_rdev_t *rdev;
4610         struct list_head *rtmp;
4611
4612
4613         if (mddev->bitmap)
4614                 bitmap_daemon_work(mddev->bitmap);
4615
4616         if (mddev->ro)
4617                 return;
4618
4619         if (signal_pending(current)) {
4620                 if (mddev->pers->sync_request) {
4621                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4622                                mdname(mddev));
4623                         mddev->safemode = 2;
4624                 }
4625                 flush_signals(current);
4626         }
4627
4628         if ( ! (
4629                 mddev->sb_dirty ||
4630                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4631                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4632                 (mddev->safemode == 1) ||
4633                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4634                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4635                 ))
4636                 return;
4637
4638         if (mddev_trylock(mddev)==0) {
4639                 int spares =0;
4640
4641                 spin_lock_irq(&mddev->write_lock);
4642                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4643                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4644                         mddev->in_sync = 1;
4645                         mddev->sb_dirty = 1;
4646                 }
4647                 if (mddev->safemode == 1)
4648                         mddev->safemode = 0;
4649                 spin_unlock_irq(&mddev->write_lock);
4650
4651                 if (mddev->sb_dirty)
4652                         md_update_sb(mddev);
4653
4654
4655                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4656                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4657                         /* resync/recovery still happening */
4658                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4659                         goto unlock;
4660                 }
4661                 if (mddev->sync_thread) {
4662                         /* resync has finished, collect result */
4663                         md_unregister_thread(mddev->sync_thread);
4664                         mddev->sync_thread = NULL;
4665                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4666                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4667                                 /* success...*/
4668                                 /* activate any spares */
4669                                 mddev->pers->spare_active(mddev);
4670                         }
4671                         md_update_sb(mddev);
4672
4673                         /* if array is no-longer degraded, then any saved_raid_disk
4674                          * information must be scrapped
4675                          */
4676                         if (!mddev->degraded)
4677                                 ITERATE_RDEV(mddev,rdev,rtmp)
4678                                         rdev->saved_raid_disk = -1;
4679
4680                         mddev->recovery = 0;
4681                         /* flag recovery needed just to double check */
4682                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4683                         md_new_event(mddev);
4684                         goto unlock;
4685                 }
4686                 /* Clear some bits that don't mean anything, but
4687                  * might be left set
4688                  */
4689                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4690                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4691                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4692                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4693
4694                 /* no recovery is running.
4695                  * remove any failed drives, then
4696                  * add spares if possible.
4697                  * Spare are also removed and re-added, to allow
4698                  * the personality to fail the re-add.
4699                  */
4700                 ITERATE_RDEV(mddev,rdev,rtmp)
4701                         if (rdev->raid_disk >= 0 &&
4702                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4703                             atomic_read(&rdev->nr_pending)==0) {
4704                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4705                                         char nm[20];
4706                                         sprintf(nm,"rd%d", rdev->raid_disk);
4707                                         sysfs_remove_link(&mddev->kobj, nm);
4708                                         rdev->raid_disk = -1;
4709                                 }
4710                         }
4711
4712                 if (mddev->degraded) {
4713                         ITERATE_RDEV(mddev,rdev,rtmp)
4714                                 if (rdev->raid_disk < 0
4715                                     && !test_bit(Faulty, &rdev->flags)) {
4716                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4717                                                 char nm[20];
4718                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4719                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4720                                                 spares++;
4721                                                 md_new_event(mddev);
4722                                         } else
4723                                                 break;
4724                                 }
4725                 }
4726
4727                 if (spares) {
4728                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4729                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4730                 } else if (mddev->recovery_cp < MaxSector) {
4731                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4732                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4733                         /* nothing to be done ... */
4734                         goto unlock;
4735
4736                 if (mddev->pers->sync_request) {
4737                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4738                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4739                                 /* We are adding a device or devices to an array
4740                                  * which has the bitmap stored on all devices.
4741                                  * So make sure all bitmap pages get written
4742                                  */
4743                                 bitmap_write_all(mddev->bitmap);
4744                         }
4745                         mddev->sync_thread = md_register_thread(md_do_sync,
4746                                                                 mddev,
4747                                                                 "%s_resync");
4748                         if (!mddev->sync_thread) {
4749                                 printk(KERN_ERR "%s: could not start resync"
4750                                         " thread...\n",
4751                                         mdname(mddev));
4752                                 /* leave the spares where they are, it shouldn't hurt */
4753                                 mddev->recovery = 0;
4754                         } else
4755                                 md_wakeup_thread(mddev->sync_thread);
4756                         md_new_event(mddev);
4757                 }
4758         unlock:
4759                 mddev_unlock(mddev);
4760         }
4761 }
4762
4763 static int md_notify_reboot(struct notifier_block *this,
4764                             unsigned long code, void *x)
4765 {
4766         struct list_head *tmp;
4767         mddev_t *mddev;
4768
4769         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4770
4771                 printk(KERN_INFO "md: stopping all md devices.\n");
4772
4773                 ITERATE_MDDEV(mddev,tmp)
4774                         if (mddev_trylock(mddev)==0)
4775                                 do_md_stop (mddev, 1);
4776                 /*
4777                  * certain more exotic SCSI devices are known to be
4778                  * volatile wrt too early system reboots. While the
4779                  * right place to handle this issue is the given
4780                  * driver, we do want to have a safe RAID driver ...
4781                  */
4782                 mdelay(1000*1);
4783         }
4784         return NOTIFY_DONE;
4785 }
4786
4787 static struct notifier_block md_notifier = {
4788         .notifier_call  = md_notify_reboot,
4789         .next           = NULL,
4790         .priority       = INT_MAX, /* before any real devices */
4791 };
4792
4793 static void md_geninit(void)
4794 {
4795         struct proc_dir_entry *p;
4796
4797         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4798
4799         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4800         if (p)
4801                 p->proc_fops = &md_seq_fops;
4802 }
4803
4804 static int __init md_init(void)
4805 {
4806         int minor;
4807
4808         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4809                         " MD_SB_DISKS=%d\n",
4810                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4811                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4812         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4813                         BITMAP_MINOR);
4814
4815         if (register_blkdev(MAJOR_NR, "md"))
4816                 return -1;
4817         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4818                 unregister_blkdev(MAJOR_NR, "md");
4819                 return -1;
4820         }
4821         devfs_mk_dir("md");
4822         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4823                                 md_probe, NULL, NULL);
4824         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4825                             md_probe, NULL, NULL);
4826
4827         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4828                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4829                                 S_IFBLK|S_IRUSR|S_IWUSR,
4830                                 "md/%d", minor);
4831
4832         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4833                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4834                               S_IFBLK|S_IRUSR|S_IWUSR,
4835                               "md/mdp%d", minor);
4836
4837
4838         register_reboot_notifier(&md_notifier);
4839         raid_table_header = register_sysctl_table(raid_root_table, 1);
4840
4841         md_geninit();
4842         return (0);
4843 }
4844
4845
4846 #ifndef MODULE
4847
4848 /*
4849  * Searches all registered partitions for autorun RAID arrays
4850  * at boot time.
4851  */
4852 static dev_t detected_devices[128];
4853 static int dev_cnt;
4854
4855 void md_autodetect_dev(dev_t dev)
4856 {
4857         if (dev_cnt >= 0 && dev_cnt < 127)
4858                 detected_devices[dev_cnt++] = dev;
4859 }
4860
4861
4862 static void autostart_arrays(int part)
4863 {
4864         mdk_rdev_t *rdev;
4865         int i;
4866
4867         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4868
4869         for (i = 0; i < dev_cnt; i++) {
4870                 dev_t dev = detected_devices[i];
4871
4872                 rdev = md_import_device(dev,0, 0);
4873                 if (IS_ERR(rdev))
4874                         continue;
4875
4876                 if (test_bit(Faulty, &rdev->flags)) {
4877                         MD_BUG();
4878                         continue;
4879                 }
4880                 list_add(&rdev->same_set, &pending_raid_disks);
4881         }
4882         dev_cnt = 0;
4883
4884         autorun_devices(part);
4885 }
4886
4887 #endif
4888
4889 static __exit void md_exit(void)
4890 {
4891         mddev_t *mddev;
4892         struct list_head *tmp;
4893         int i;
4894         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4895         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4896         for (i=0; i < MAX_MD_DEVS; i++)
4897                 devfs_remove("md/%d", i);
4898         for (i=0; i < MAX_MD_DEVS; i++)
4899                 devfs_remove("md/d%d", i);
4900
4901         devfs_remove("md");
4902
4903         unregister_blkdev(MAJOR_NR,"md");
4904         unregister_blkdev(mdp_major, "mdp");
4905         unregister_reboot_notifier(&md_notifier);
4906         unregister_sysctl_table(raid_table_header);
4907         remove_proc_entry("mdstat", NULL);
4908         ITERATE_MDDEV(mddev,tmp) {
4909                 struct gendisk *disk = mddev->gendisk;
4910                 if (!disk)
4911                         continue;
4912                 export_array(mddev);
4913                 del_gendisk(disk);
4914                 put_disk(disk);
4915                 mddev->gendisk = NULL;
4916                 mddev_put(mddev);
4917         }
4918 }
4919
4920 module_init(md_init)
4921 module_exit(md_exit)
4922
4923 static int get_ro(char *buffer, struct kernel_param *kp)
4924 {
4925         return sprintf(buffer, "%d", start_readonly);
4926 }
4927 static int set_ro(const char *val, struct kernel_param *kp)
4928 {
4929         char *e;
4930         int num = simple_strtoul(val, &e, 10);
4931         if (*val && (*e == '\0' || *e == '\n')) {
4932                 start_readonly = num;
4933                 return 0;
4934         }
4935         return -EINVAL;
4936 }
4937
4938 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4939 module_param(start_dirty_degraded, int, 0644);
4940
4941
4942 EXPORT_SYMBOL(register_md_personality);
4943 EXPORT_SYMBOL(unregister_md_personality);
4944 EXPORT_SYMBOL(md_error);
4945 EXPORT_SYMBOL(md_done_sync);
4946 EXPORT_SYMBOL(md_write_start);
4947 EXPORT_SYMBOL(md_write_end);
4948 EXPORT_SYMBOL(md_register_thread);
4949 EXPORT_SYMBOL(md_unregister_thread);
4950 EXPORT_SYMBOL(md_wakeup_thread);
4951 EXPORT_SYMBOL(md_print_devices);
4952 EXPORT_SYMBOL(md_check_recovery);
4953 MODULE_LICENSE("GPL");
4954 MODULE_ALIAS("md");
4955 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);