drivers/md/md.c

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