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/kernel.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/buffer_head.h> /* for invalidate_bdev */
  43 #include <linux/poll.h>
  44 #include <linux/mutex.h>
  45 #include <linux/ctype.h>
  46 #include <linux/freezer.h>
  47
  48 #include <linux/init.h>
  49
  50 #include <linux/file.h>
  51
  52 #ifdef CONFIG_KMOD
  53 #include <linux/kmod.h>
  54 #endif
  55
  56 #include <asm/unaligned.h>
  57
  58 #define MAJOR_NR MD_MAJOR
  59 #define MD_DRIVER
  60
  61 /* 63 partitions with the alternate major number (mdp) */
  62 #define MdpMinorShift 6
  63
  64 #define DEBUG 0
  65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  66
  67
  68 #ifndef MODULE
  69 static void autostart_arrays (int part);
  70 #endif
  71
  72 static LIST_HEAD(pers_list);
  73 static DEFINE_SPINLOCK(pers_lock);
  74
  75 static void md_print_devices(void);
  76
  77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  78
  79 /*
  80  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  81  * is 1000 KB/sec, so the extra system load does not show up that much.
  82  * Increase it if you want to have more _guaranteed_ speed. Note that
  83  * the RAID driver will use the maximum available bandwidth if the IO
  84  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  85  * speed limit - in case reconstruction slows down your system despite
  86  * idle IO detection.
  87  *
  88  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  89  * or /sys/block/mdX/md/sync_speed_{min,max}
  90  */
  91
  92 static int sysctl_speed_limit_min = 1000;
  93 static int sysctl_speed_limit_max = 200000;
  94 static inline int speed_min(mddev_t *mddev)
  95 {
  96         return mddev->sync_speed_min ?
  97                 mddev->sync_speed_min : sysctl_speed_limit_min;
  98 }
  99
 100 static inline int speed_max(mddev_t *mddev)
 101 {
 102         return mddev->sync_speed_max ?
 103                 mddev->sync_speed_max : sysctl_speed_limit_max;
 104 }
 105
 106 static struct ctl_table_header *raid_table_header;
 107
 108 static ctl_table raid_table[] = {
 109         {
 110                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 111                 .procname       = "speed_limit_min",
 112                 .data           = &sysctl_speed_limit_min,
 113                 .maxlen         = sizeof(int),
 114                 .mode           = S_IRUGO|S_IWUSR,
 115                 .proc_handler   = &proc_dointvec,
 116         },
 117         {
 118                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 119                 .procname       = "speed_limit_max",
 120                 .data           = &sysctl_speed_limit_max,
 121                 .maxlen         = sizeof(int),
 122                 .mode           = S_IRUGO|S_IWUSR,
 123                 .proc_handler   = &proc_dointvec,
 124         },
 125         { .ctl_name = 0 }
 126 };
 127
 128 static ctl_table raid_dir_table[] = {
 129         {
 130                 .ctl_name       = DEV_RAID,
 131                 .procname       = "raid",
 132                 .maxlen         = 0,
 133                 .mode           = S_IRUGO|S_IXUGO,
 134                 .child          = raid_table,
 135         },
 136         { .ctl_name = 0 }
 137 };
 138
 139 static ctl_table raid_root_table[] = {
 140         {
 141                 .ctl_name       = CTL_DEV,
 142                 .procname       = "dev",
 143                 .maxlen         = 0,
 144                 .mode           = 0555,
 145                 .child          = raid_dir_table,
 146         },
 147         { .ctl_name = 0 }
 148 };
 149
 150 static struct block_device_operations md_fops;
 151
 152 static int start_readonly;
 153
 154 /*
 155  * We have a system wide 'event count' that is incremented
 156  * on any 'interesting' event, and readers of /proc/mdstat
 157  * can use 'poll' or 'select' to find out when the event
 158  * count increases.
 159  *
 160  * Events are:
 161  *  start array, stop array, error, add device, remove device,
 162  *  start build, activate spare
 163  */
 164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 165 static atomic_t md_event_count;
 166 void md_new_event(mddev_t *mddev)
 167 {
 168         atomic_inc(&md_event_count);
 169         wake_up(&md_event_waiters);
 170         sysfs_notify(&mddev->kobj, NULL, "sync_action");
 171 }
 172 EXPORT_SYMBOL_GPL(md_new_event);
 173
 174 /* Alternate version that can be called from interrupts
 175  * when calling sysfs_notify isn't needed.
 176  */
 177 static void md_new_event_inintr(mddev_t *mddev)
 178 {
 179         atomic_inc(&md_event_count);
 180         wake_up(&md_event_waiters);
 181 }
 182
 183 /*
 184  * Enables to iterate over all existing md arrays
 185  * all_mddevs_lock protects this list.
 186  */
 187 static LIST_HEAD(all_mddevs);
 188 static DEFINE_SPINLOCK(all_mddevs_lock);
 189
 190
 191 /*
 192  * iterates through all used mddevs in the system.
 193  * We take care to grab the all_mddevs_lock whenever navigating
 194  * the list, and to always hold a refcount when unlocked.
 195  * Any code which breaks out of this loop while own
 196  * a reference to the current mddev and must mddev_put it.
 197  */
 198 #define ITERATE_MDDEV(mddev,tmp)                                        \
 199                                                                         \
 200         for (({ spin_lock(&all_mddevs_lock);                            \
 201                 tmp = all_mddevs.next;                                  \
 202                 mddev = NULL;});                                        \
 203              ({ if (tmp != &all_mddevs)                                 \
 204                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 205                 spin_unlock(&all_mddevs_lock);                          \
 206                 if (mddev) mddev_put(mddev);                            \
 207                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 208                 tmp != &all_mddevs;});                                  \
 209              ({ spin_lock(&all_mddevs_lock);                            \
 210                 tmp = tmp->next;})                                      \
 211                 )
 212
 213
 214 static int md_fail_request (struct request_queue *q, struct bio *bio)
 215 {
 216         bio_io_error(bio);
 217         return 0;
 218 }
 219
 220 static inline mddev_t *mddev_get(mddev_t *mddev)
 221 {
 222         atomic_inc(&mddev->active);
 223         return mddev;
 224 }
 225
 226 static void mddev_put(mddev_t *mddev)
 227 {
 228         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 229                 return;
 230         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 231                 list_del(&mddev->all_mddevs);
 232                 spin_unlock(&all_mddevs_lock);
 233                 blk_cleanup_queue(mddev->queue);
 234                 kobject_unregister(&mddev->kobj);
 235         } else
 236                 spin_unlock(&all_mddevs_lock);
 237 }
 238
 239 static mddev_t * mddev_find(dev_t unit)
 240 {
 241         mddev_t *mddev, *new = NULL;
 242
 243  retry:
 244         spin_lock(&all_mddevs_lock);
 245         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 246                 if (mddev->unit == unit) {
 247                         mddev_get(mddev);
 248                         spin_unlock(&all_mddevs_lock);
 249                         kfree(new);
 250                         return mddev;
 251                 }
 252
 253         if (new) {
 254                 list_add(&new->all_mddevs, &all_mddevs);
 255                 spin_unlock(&all_mddevs_lock);
 256                 return new;
 257         }
 258         spin_unlock(&all_mddevs_lock);
 259
 260         new = kzalloc(sizeof(*new), GFP_KERNEL);
 261         if (!new)
 262                 return NULL;
 263
 264         new->unit = unit;
 265         if (MAJOR(unit) == MD_MAJOR)
 266                 new->md_minor = MINOR(unit);
 267         else
 268                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 269
 270         mutex_init(&new->reconfig_mutex);
 271         INIT_LIST_HEAD(&new->disks);
 272         INIT_LIST_HEAD(&new->all_mddevs);
 273         init_timer(&new->safemode_timer);
 274         atomic_set(&new->active, 1);
 275         spin_lock_init(&new->write_lock);
 276         init_waitqueue_head(&new->sb_wait);
 277         new->reshape_position = MaxSector;
 278
 279         new->queue = blk_alloc_queue(GFP_KERNEL);
 280         if (!new->queue) {
 281                 kfree(new);
 282                 return NULL;
 283         }
 284         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
 285
 286         blk_queue_make_request(new->queue, md_fail_request);
 287
 288         goto retry;
 289 }
 290
 291 static inline int mddev_lock(mddev_t * mddev)
 292 {
 293         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 294 }
 295
 296 static inline int mddev_trylock(mddev_t * mddev)
 297 {
 298         return mutex_trylock(&mddev->reconfig_mutex);
 299 }
 300
 301 static inline void mddev_unlock(mddev_t * mddev)
 302 {
 303         mutex_unlock(&mddev->reconfig_mutex);
 304
 305         md_wakeup_thread(mddev->thread);
 306 }
 307
 308 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 309 {
 310         mdk_rdev_t * rdev;
 311         struct list_head *tmp;
 312
 313         ITERATE_RDEV(mddev,rdev,tmp) {
 314                 if (rdev->desc_nr == nr)
 315                         return rdev;
 316         }
 317         return NULL;
 318 }
 319
 320 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 321 {
 322         struct list_head *tmp;
 323         mdk_rdev_t *rdev;
 324
 325         ITERATE_RDEV(mddev,rdev,tmp) {
 326                 if (rdev->bdev->bd_dev == dev)
 327                         return rdev;
 328         }
 329         return NULL;
 330 }
 331
 332 static struct mdk_personality *find_pers(int level, char *clevel)
 333 {
 334         struct mdk_personality *pers;
 335         list_for_each_entry(pers, &pers_list, list) {
 336                 if (level != LEVEL_NONE && pers->level == level)
 337                         return pers;
 338                 if (strcmp(pers->name, clevel)==0)
 339                         return pers;
 340         }
 341         return NULL;
 342 }
 343
 344 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 345 {
 346         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 347         return MD_NEW_SIZE_BLOCKS(size);
 348 }
 349
 350 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 351 {
 352         sector_t size;
 353
 354         size = rdev->sb_offset;
 355
 356         if (chunk_size)
 357                 size &= ~((sector_t)chunk_size/1024 - 1);
 358         return size;
 359 }
 360
 361 static int alloc_disk_sb(mdk_rdev_t * rdev)
 362 {
 363         if (rdev->sb_page)
 364                 MD_BUG();
 365
 366         rdev->sb_page = alloc_page(GFP_KERNEL);
 367         if (!rdev->sb_page) {
 368                 printk(KERN_ALERT "md: out of memory.\n");
 369                 return -EINVAL;
 370         }
 371
 372         return 0;
 373 }
 374
 375 static void free_disk_sb(mdk_rdev_t * rdev)
 376 {
 377         if (rdev->sb_page) {
 378                 put_page(rdev->sb_page);
 379                 rdev->sb_loaded = 0;
 380                 rdev->sb_page = NULL;
 381                 rdev->sb_offset = 0;
 382                 rdev->size = 0;
 383         }
 384 }
 385
 386
 387 static void super_written(struct bio *bio, int error)
 388 {
 389         mdk_rdev_t *rdev = bio->bi_private;
 390         mddev_t *mddev = rdev->mddev;
 391
 392         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 393                 printk("md: super_written gets error=%d, uptodate=%d\n",
 394                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 395                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 396                 md_error(mddev, rdev);
 397         }
 398
 399         if (atomic_dec_and_test(&mddev->pending_writes))
 400                 wake_up(&mddev->sb_wait);
 401         bio_put(bio);
 402 }
 403
 404 static void super_written_barrier(struct bio *bio, int error)
 405 {
 406         struct bio *bio2 = bio->bi_private;
 407         mdk_rdev_t *rdev = bio2->bi_private;
 408         mddev_t *mddev = rdev->mddev;
 409
 410         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 411             error == -EOPNOTSUPP) {
 412                 unsigned long flags;
 413                 /* barriers don't appear to be supported :-( */
 414                 set_bit(BarriersNotsupp, &rdev->flags);
 415                 mddev->barriers_work = 0;
 416                 spin_lock_irqsave(&mddev->write_lock, flags);
 417                 bio2->bi_next = mddev->biolist;
 418                 mddev->biolist = bio2;
 419                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 420                 wake_up(&mddev->sb_wait);
 421                 bio_put(bio);
 422         } else {
 423                 bio_put(bio2);
 424                 bio->bi_private = rdev;
 425                 super_written(bio, error);
 426         }
 427 }
 428
 429 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 430                    sector_t sector, int size, struct page *page)
 431 {
 432         /* write first size bytes of page to sector of rdev
 433          * Increment mddev->pending_writes before returning
 434          * and decrement it on completion, waking up sb_wait
 435          * if zero is reached.
 436          * If an error occurred, call md_error
 437          *
 438          * As we might need to resubmit the request if BIO_RW_BARRIER
 439          * causes ENOTSUPP, we allocate a spare bio...
 440          */
 441         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 442         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 443
 444         bio->bi_bdev = rdev->bdev;
 445         bio->bi_sector = sector;
 446         bio_add_page(bio, page, size, 0);
 447         bio->bi_private = rdev;
 448         bio->bi_end_io = super_written;
 449         bio->bi_rw = rw;
 450
 451         atomic_inc(&mddev->pending_writes);
 452         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 453                 struct bio *rbio;
 454                 rw |= (1<<BIO_RW_BARRIER);
 455                 rbio = bio_clone(bio, GFP_NOIO);
 456                 rbio->bi_private = bio;
 457                 rbio->bi_end_io = super_written_barrier;
 458                 submit_bio(rw, rbio);
 459         } else
 460                 submit_bio(rw, bio);
 461 }
 462
 463 void md_super_wait(mddev_t *mddev)
 464 {
 465         /* wait for all superblock writes that were scheduled to complete.
 466          * if any had to be retried (due to BARRIER problems), retry them
 467          */
 468         DEFINE_WAIT(wq);
 469         for(;;) {
 470                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 471                 if (atomic_read(&mddev->pending_writes)==0)
 472                         break;
 473                 while (mddev->biolist) {
 474                         struct bio *bio;
 475                         spin_lock_irq(&mddev->write_lock);
 476                         bio = mddev->biolist;
 477                         mddev->biolist = bio->bi_next ;
 478                         bio->bi_next = NULL;
 479                         spin_unlock_irq(&mddev->write_lock);
 480                         submit_bio(bio->bi_rw, bio);
 481                 }
 482                 schedule();
 483         }
 484         finish_wait(&mddev->sb_wait, &wq);
 485 }
 486
 487 static void bi_complete(struct bio *bio, int error)
 488 {
 489         complete((struct completion*)bio->bi_private);
 490 }
 491
 492 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 493                    struct page *page, int rw)
 494 {
 495         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 496         struct completion event;
 497         int ret;
 498
 499         rw |= (1 << BIO_RW_SYNC);
 500
 501         bio->bi_bdev = bdev;
 502         bio->bi_sector = sector;
 503         bio_add_page(bio, page, size, 0);
 504         init_completion(&event);
 505         bio->bi_private = &event;
 506         bio->bi_end_io = bi_complete;
 507         submit_bio(rw, bio);
 508         wait_for_completion(&event);
 509
 510         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 511         bio_put(bio);
 512         return ret;
 513 }
 514 EXPORT_SYMBOL_GPL(sync_page_io);
 515
 516 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 517 {
 518         char b[BDEVNAME_SIZE];
 519         if (!rdev->sb_page) {
 520                 MD_BUG();
 521                 return -EINVAL;
 522         }
 523         if (rdev->sb_loaded)
 524                 return 0;
 525
 526
 527         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 528                 goto fail;
 529         rdev->sb_loaded = 1;
 530         return 0;
 531
 532 fail:
 533         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 534                 bdevname(rdev->bdev,b));
 535         return -EINVAL;
 536 }
 537
 538 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 539 {
 540         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 541                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 542                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 543                 (sb1->set_uuid3 == sb2->set_uuid3))
 544
 545                 return 1;
 546
 547         return 0;
 548 }
 549
 550
 551 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 552 {
 553         int ret;
 554         mdp_super_t *tmp1, *tmp2;
 555
 556         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 557         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 558
 559         if (!tmp1 || !tmp2) {
 560                 ret = 0;
 561                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 562                 goto abort;
 563         }
 564
 565         *tmp1 = *sb1;
 566         *tmp2 = *sb2;
 567
 568         /*
 569          * nr_disks is not constant
 570          */
 571         tmp1->nr_disks = 0;
 572         tmp2->nr_disks = 0;
 573
 574         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 575                 ret = 0;
 576         else
 577                 ret = 1;
 578
 579 abort:
 580         kfree(tmp1);
 581         kfree(tmp2);
 582         return ret;
 583 }
 584
 585
 586 static u32 md_csum_fold(u32 csum)
 587 {
 588         csum = (csum & 0xffff) + (csum >> 16);
 589         return (csum & 0xffff) + (csum >> 16);
 590 }
 591
 592 static unsigned int calc_sb_csum(mdp_super_t * sb)
 593 {
 594         u64 newcsum = 0;
 595         u32 *sb32 = (u32*)sb;
 596         int i;
 597         unsigned int disk_csum, csum;
 598
 599         disk_csum = sb->sb_csum;
 600         sb->sb_csum = 0;
 601
 602         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 603                 newcsum += sb32[i];
 604         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 605
 606
 607 #ifdef CONFIG_ALPHA
 608         /* This used to use csum_partial, which was wrong for several
 609          * reasons including that different results are returned on
 610          * different architectures.  It isn't critical that we get exactly
 611          * the same return value as before (we always csum_fold before
 612          * testing, and that removes any differences).  However as we
 613          * know that csum_partial always returned a 16bit value on
 614          * alphas, do a fold to maximise conformity to previous behaviour.
 615          */
 616         sb->sb_csum = md_csum_fold(disk_csum);
 617 #else
 618         sb->sb_csum = disk_csum;
 619 #endif
 620         return csum;
 621 }
 622
 623
 624 /*
 625  * Handle superblock details.
 626  * We want to be able to handle multiple superblock formats
 627  * so we have a common interface to them all, and an array of
 628  * different handlers.
 629  * We rely on user-space to write the initial superblock, and support
 630  * reading and updating of superblocks.
 631  * Interface methods are:
 632  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 633  *      loads and validates a superblock on dev.
 634  *      if refdev != NULL, compare superblocks on both devices
 635  *    Return:
 636  *      0 - dev has a superblock that is compatible with refdev
 637  *      1 - dev has a superblock that is compatible and newer than refdev
 638  *          so dev should be used as the refdev in future
 639  *     -EINVAL superblock incompatible or invalid
 640  *     -othererror e.g. -EIO
 641  *
 642  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 643  *      Verify that dev is acceptable into mddev.
 644  *       The first time, mddev->raid_disks will be 0, and data from
 645  *       dev should be merged in.  Subsequent calls check that dev
 646  *       is new enough.  Return 0 or -EINVAL
 647  *
 648  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 649  *     Update the superblock for rdev with data in mddev
 650  *     This does not write to disc.
 651  *
 652  */
 653
 654 struct super_type  {
 655         char            *name;
 656         struct module   *owner;
 657         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 658         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 659         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 660 };
 661
 662 /*
 663  * load_super for 0.90.0
 664  */
 665 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 666 {
 667         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 668         mdp_super_t *sb;
 669         int ret;
 670         sector_t sb_offset;
 671
 672         /*
 673          * Calculate the position of the superblock,
 674          * it's at the end of the disk.
 675          *
 676          * It also happens to be a multiple of 4Kb.
 677          */
 678         sb_offset = calc_dev_sboffset(rdev->bdev);
 679         rdev->sb_offset = sb_offset;
 680
 681         ret = read_disk_sb(rdev, MD_SB_BYTES);
 682         if (ret) return ret;
 683
 684         ret = -EINVAL;
 685
 686         bdevname(rdev->bdev, b);
 687         sb = (mdp_super_t*)page_address(rdev->sb_page);
 688
 689         if (sb->md_magic != MD_SB_MAGIC) {
 690                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 691                        b);
 692                 goto abort;
 693         }
 694
 695         if (sb->major_version != 0 ||
 696             sb->minor_version < 90 ||
 697             sb->minor_version > 91) {
 698                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 699                         sb->major_version, sb->minor_version,
 700                         b);
 701                 goto abort;
 702         }
 703
 704         if (sb->raid_disks <= 0)
 705                 goto abort;
 706
 707         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 708                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 709                         b);
 710                 goto abort;
 711         }
 712
 713         rdev->preferred_minor = sb->md_minor;
 714         rdev->data_offset = 0;
 715         rdev->sb_size = MD_SB_BYTES;
 716
 717         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 718                 if (sb->level != 1 && sb->level != 4
 719                     && sb->level != 5 && sb->level != 6
 720                     && sb->level != 10) {
 721                         /* FIXME use a better test */
 722                         printk(KERN_WARNING
 723                                "md: bitmaps not supported for this level.\n");
 724                         goto abort;
 725                 }
 726         }
 727
 728         if (sb->level == LEVEL_MULTIPATH)
 729                 rdev->desc_nr = -1;
 730         else
 731                 rdev->desc_nr = sb->this_disk.number;
 732
 733         if (refdev == 0)
 734                 ret = 1;
 735         else {
 736                 __u64 ev1, ev2;
 737                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 738                 if (!uuid_equal(refsb, sb)) {
 739                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 740                                 b, bdevname(refdev->bdev,b2));
 741                         goto abort;
 742                 }
 743                 if (!sb_equal(refsb, sb)) {
 744                         printk(KERN_WARNING "md: %s has same UUID"
 745                                " but different superblock to %s\n",
 746                                b, bdevname(refdev->bdev, b2));
 747                         goto abort;
 748                 }
 749                 ev1 = md_event(sb);
 750                 ev2 = md_event(refsb);
 751                 if (ev1 > ev2)
 752                         ret = 1;
 753                 else
 754                         ret = 0;
 755         }
 756         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 757
 758         if (rdev->size < sb->size && sb->level > 1)
 759                 /* "this cannot possibly happen" ... */
 760                 ret = -EINVAL;
 761
 762  abort:
 763         return ret;
 764 }
 765
 766 /*
 767  * validate_super for 0.90.0
 768  */
 769 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 770 {
 771         mdp_disk_t *desc;
 772         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 773         __u64 ev1 = md_event(sb);
 774
 775         rdev->raid_disk = -1;
 776         rdev->flags = 0;
 777         if (mddev->raid_disks == 0) {
 778                 mddev->major_version = 0;
 779                 mddev->minor_version = sb->minor_version;
 780                 mddev->patch_version = sb->patch_version;
 781                 mddev->persistent = ! sb->not_persistent;
 782                 mddev->chunk_size = sb->chunk_size;
 783                 mddev->ctime = sb->ctime;
 784                 mddev->utime = sb->utime;
 785                 mddev->level = sb->level;
 786                 mddev->clevel[0] = 0;
 787                 mddev->layout = sb->layout;
 788                 mddev->raid_disks = sb->raid_disks;
 789                 mddev->size = sb->size;
 790                 mddev->events = ev1;
 791                 mddev->bitmap_offset = 0;
 792                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 793
 794                 if (mddev->minor_version >= 91) {
 795                         mddev->reshape_position = sb->reshape_position;
 796                         mddev->delta_disks = sb->delta_disks;
 797                         mddev->new_level = sb->new_level;
 798                         mddev->new_layout = sb->new_layout;
 799                         mddev->new_chunk = sb->new_chunk;
 800                 } else {
 801                         mddev->reshape_position = MaxSector;
 802                         mddev->delta_disks = 0;
 803                         mddev->new_level = mddev->level;
 804                         mddev->new_layout = mddev->layout;
 805                         mddev->new_chunk = mddev->chunk_size;
 806                 }
 807
 808                 if (sb->state & (1<<MD_SB_CLEAN))
 809                         mddev->recovery_cp = MaxSector;
 810                 else {
 811                         if (sb->events_hi == sb->cp_events_hi &&
 812                                 sb->events_lo == sb->cp_events_lo) {
 813                                 mddev->recovery_cp = sb->recovery_cp;
 814                         } else
 815                                 mddev->recovery_cp = 0;
 816                 }
 817
 818                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 819                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 820                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 821                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 822
 823                 mddev->max_disks = MD_SB_DISKS;
 824
 825                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 826                     mddev->bitmap_file == NULL)
 827                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 828
 829         } else if (mddev->pers == NULL) {
 830                 /* Insist on good event counter while assembling */
 831                 ++ev1;
 832                 if (ev1 < mddev->events)
 833                         return -EINVAL;
 834         } else if (mddev->bitmap) {
 835                 /* if adding to array with a bitmap, then we can accept an
 836                  * older device ... but not too old.
 837                  */
 838                 if (ev1 < mddev->bitmap->events_cleared)
 839                         return 0;
 840         } else {
 841                 if (ev1 < mddev->events)
 842                         /* just a hot-add of a new device, leave raid_disk at -1 */
 843                         return 0;
 844         }
 845
 846         if (mddev->level != LEVEL_MULTIPATH) {
 847                 desc = sb->disks + rdev->desc_nr;
 848
 849                 if (desc->state & (1<<MD_DISK_FAULTY))
 850                         set_bit(Faulty, &rdev->flags);
 851                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
 852                             desc->raid_disk < mddev->raid_disks */) {
 853                         set_bit(In_sync, &rdev->flags);
 854                         rdev->raid_disk = desc->raid_disk;
 855                 }
 856                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 857                         set_bit(WriteMostly, &rdev->flags);
 858         } else /* MULTIPATH are always insync */
 859                 set_bit(In_sync, &rdev->flags);
 860         return 0;
 861 }
 862
 863 /*
 864  * sync_super for 0.90.0
 865  */
 866 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 867 {
 868         mdp_super_t *sb;
 869         struct list_head *tmp;
 870         mdk_rdev_t *rdev2;
 871         int next_spare = mddev->raid_disks;
 872
 873
 874         /* make rdev->sb match mddev data..
 875          *
 876          * 1/ zero out disks
 877          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 878          * 3/ any empty disks < next_spare become removed
 879          *
 880          * disks[0] gets initialised to REMOVED because
 881          * we cannot be sure from other fields if it has
 882          * been initialised or not.
 883          */
 884         int i;
 885         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 886
 887         rdev->sb_size = MD_SB_BYTES;
 888
 889         sb = (mdp_super_t*)page_address(rdev->sb_page);
 890
 891         memset(sb, 0, sizeof(*sb));
 892
 893         sb->md_magic = MD_SB_MAGIC;
 894         sb->major_version = mddev->major_version;
 895         sb->patch_version = mddev->patch_version;
 896         sb->gvalid_words  = 0; /* ignored */
 897         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 898         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 899         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 900         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 901
 902         sb->ctime = mddev->ctime;
 903         sb->level = mddev->level;
 904         sb->size  = mddev->size;
 905         sb->raid_disks = mddev->raid_disks;
 906         sb->md_minor = mddev->md_minor;
 907         sb->not_persistent = !mddev->persistent;
 908         sb->utime = mddev->utime;
 909         sb->state = 0;
 910         sb->events_hi = (mddev->events>>32);
 911         sb->events_lo = (u32)mddev->events;
 912
 913         if (mddev->reshape_position == MaxSector)
 914                 sb->minor_version = 90;
 915         else {
 916                 sb->minor_version = 91;
 917                 sb->reshape_position = mddev->reshape_position;
 918                 sb->new_level = mddev->new_level;
 919                 sb->delta_disks = mddev->delta_disks;
 920                 sb->new_layout = mddev->new_layout;
 921                 sb->new_chunk = mddev->new_chunk;
 922         }
 923         mddev->minor_version = sb->minor_version;
 924         if (mddev->in_sync)
 925         {
 926                 sb->recovery_cp = mddev->recovery_cp;
 927                 sb->cp_events_hi = (mddev->events>>32);
 928                 sb->cp_events_lo = (u32)mddev->events;
 929                 if (mddev->recovery_cp == MaxSector)
 930                         sb->state = (1<< MD_SB_CLEAN);
 931         } else
 932                 sb->recovery_cp = 0;
 933
 934         sb->layout = mddev->layout;
 935         sb->chunk_size = mddev->chunk_size;
 936
 937         if (mddev->bitmap && mddev->bitmap_file == NULL)
 938                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 939
 940         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 941         ITERATE_RDEV(mddev,rdev2,tmp) {
 942                 mdp_disk_t *d;
 943                 int desc_nr;
 944                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 945                     && !test_bit(Faulty, &rdev2->flags))
 946                         desc_nr = rdev2->raid_disk;
 947                 else
 948                         desc_nr = next_spare++;
 949                 rdev2->desc_nr = desc_nr;
 950                 d = &sb->disks[rdev2->desc_nr];
 951                 nr_disks++;
 952                 d->number = rdev2->desc_nr;
 953                 d->major = MAJOR(rdev2->bdev->bd_dev);
 954                 d->minor = MINOR(rdev2->bdev->bd_dev);
 955                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 956                     && !test_bit(Faulty, &rdev2->flags))
 957                         d->raid_disk = rdev2->raid_disk;
 958                 else
 959                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 960                 if (test_bit(Faulty, &rdev2->flags))
 961                         d->state = (1<<MD_DISK_FAULTY);
 962                 else if (test_bit(In_sync, &rdev2->flags)) {
 963                         d->state = (1<<MD_DISK_ACTIVE);
 964                         d->state |= (1<<MD_DISK_SYNC);
 965                         active++;
 966                         working++;
 967                 } else {
 968                         d->state = 0;
 969                         spare++;
 970                         working++;
 971                 }
 972                 if (test_bit(WriteMostly, &rdev2->flags))
 973                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 974         }
 975         /* now set the "removed" and "faulty" bits on any missing devices */
 976         for (i=0 ; i < mddev->raid_disks ; i++) {
 977                 mdp_disk_t *d = &sb->disks[i];
 978                 if (d->state == 0 && d->number == 0) {
 979                         d->number = i;
 980                         d->raid_disk = i;
 981                         d->state = (1<<MD_DISK_REMOVED);
 982                         d->state |= (1<<MD_DISK_FAULTY);
 983                         failed++;
 984                 }
 985         }
 986         sb->nr_disks = nr_disks;
 987         sb->active_disks = active;
 988         sb->working_disks = working;
 989         sb->failed_disks = failed;
 990         sb->spare_disks = spare;
 991
 992         sb->this_disk = sb->disks[rdev->desc_nr];
 993         sb->sb_csum = calc_sb_csum(sb);
 994 }
 995
 996 /*
 997  * version 1 superblock
 998  */
 999
1000 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1001 {
1002         __le32 disk_csum;
1003         u32 csum;
1004         unsigned long long newcsum;
1005         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1006         __le32 *isuper = (__le32*)sb;
1007         int i;
1008
1009         disk_csum = sb->sb_csum;
1010         sb->sb_csum = 0;
1011         newcsum = 0;
1012         for (i=0; size>=4; size -= 4 )
1013                 newcsum += le32_to_cpu(*isuper++);
1014
1015         if (size == 2)
1016                 newcsum += le16_to_cpu(*(__le16*) isuper);
1017
1018         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1019         sb->sb_csum = disk_csum;
1020         return cpu_to_le32(csum);
1021 }
1022
1023 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1024 {
1025         struct mdp_superblock_1 *sb;
1026         int ret;
1027         sector_t sb_offset;
1028         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1029         int bmask;
1030
1031         /*
1032          * Calculate the position of the superblock.
1033          * It is always aligned to a 4K boundary and
1034          * depeding on minor_version, it can be:
1035          * 0: At least 8K, but less than 12K, from end of device
1036          * 1: At start of device
1037          * 2: 4K from start of device.
1038          */
1039         switch(minor_version) {
1040         case 0:
1041                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1042                 sb_offset -= 8*2;
1043                 sb_offset &= ~(sector_t)(4*2-1);
1044                 /* convert from sectors to K */
1045                 sb_offset /= 2;
1046                 break;
1047         case 1:
1048                 sb_offset = 0;
1049                 break;
1050         case 2:
1051                 sb_offset = 4;
1052                 break;
1053         default:
1054                 return -EINVAL;
1055         }
1056         rdev->sb_offset = sb_offset;
1057
1058         /* superblock is rarely larger than 1K, but it can be larger,
1059          * and it is safe to read 4k, so we do that
1060          */
1061         ret = read_disk_sb(rdev, 4096);
1062         if (ret) return ret;
1063
1064
1065         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1066
1067         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1068             sb->major_version != cpu_to_le32(1) ||
1069             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1070             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1071             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1072                 return -EINVAL;
1073
1074         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1075                 printk("md: invalid superblock checksum on %s\n",
1076                         bdevname(rdev->bdev,b));
1077                 return -EINVAL;
1078         }
1079         if (le64_to_cpu(sb->data_size) < 10) {
1080                 printk("md: data_size too small on %s\n",
1081                        bdevname(rdev->bdev,b));
1082                 return -EINVAL;
1083         }
1084         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1085                 if (sb->level != cpu_to_le32(1) &&
1086                     sb->level != cpu_to_le32(4) &&
1087                     sb->level != cpu_to_le32(5) &&
1088                     sb->level != cpu_to_le32(6) &&
1089                     sb->level != cpu_to_le32(10)) {
1090                         printk(KERN_WARNING
1091                                "md: bitmaps not supported for this level.\n");
1092                         return -EINVAL;
1093                 }
1094         }
1095
1096         rdev->preferred_minor = 0xffff;
1097         rdev->data_offset = le64_to_cpu(sb->data_offset);
1098         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1099
1100         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1101         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1102         if (rdev->sb_size & bmask)
1103                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1104
1105         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1106                 rdev->desc_nr = -1;
1107         else
1108                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1109
1110         if (refdev == 0)
1111                 ret = 1;
1112         else {
1113                 __u64 ev1, ev2;
1114                 struct mdp_superblock_1 *refsb =
1115                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1116
1117                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1118                     sb->level != refsb->level ||
1119                     sb->layout != refsb->layout ||
1120                     sb->chunksize != refsb->chunksize) {
1121                         printk(KERN_WARNING "md: %s has strangely different"
1122                                 " superblock to %s\n",
1123                                 bdevname(rdev->bdev,b),
1124                                 bdevname(refdev->bdev,b2));
1125                         return -EINVAL;
1126                 }
1127                 ev1 = le64_to_cpu(sb->events);
1128                 ev2 = le64_to_cpu(refsb->events);
1129
1130                 if (ev1 > ev2)
1131                         ret = 1;
1132                 else
1133                         ret = 0;
1134         }
1135         if (minor_version)
1136                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1137         else
1138                 rdev->size = rdev->sb_offset;
1139         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1140                 return -EINVAL;
1141         rdev->size = le64_to_cpu(sb->data_size)/2;
1142         if (le32_to_cpu(sb->chunksize))
1143                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1144
1145         if (le64_to_cpu(sb->size) > rdev->size*2)
1146                 return -EINVAL;
1147         return ret;
1148 }
1149
1150 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1151 {
1152         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1153         __u64 ev1 = le64_to_cpu(sb->events);
1154
1155         rdev->raid_disk = -1;
1156         rdev->flags = 0;
1157         if (mddev->raid_disks == 0) {
1158                 mddev->major_version = 1;
1159                 mddev->patch_version = 0;
1160                 mddev->persistent = 1;
1161                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1162                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1163                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1164                 mddev->level = le32_to_cpu(sb->level);
1165                 mddev->clevel[0] = 0;
1166                 mddev->layout = le32_to_cpu(sb->layout);
1167                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1168                 mddev->size = le64_to_cpu(sb->size)/2;
1169                 mddev->events = ev1;
1170                 mddev->bitmap_offset = 0;
1171                 mddev->default_bitmap_offset = 1024 >> 9;
1172
1173                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1174                 memcpy(mddev->uuid, sb->set_uuid, 16);
1175
1176                 mddev->max_disks =  (4096-256)/2;
1177
1178                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1179                     mddev->bitmap_file == NULL )
1180                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1181
1182                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1183                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1184                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1185                         mddev->new_level = le32_to_cpu(sb->new_level);
1186                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1187                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1188                 } else {
1189                         mddev->reshape_position = MaxSector;
1190                         mddev->delta_disks = 0;
1191                         mddev->new_level = mddev->level;
1192                         mddev->new_layout = mddev->layout;
1193                         mddev->new_chunk = mddev->chunk_size;
1194                 }
1195
1196         } else if (mddev->pers == NULL) {
1197                 /* Insist of good event counter while assembling */
1198                 ++ev1;
1199                 if (ev1 < mddev->events)
1200                         return -EINVAL;
1201         } else if (mddev->bitmap) {
1202                 /* If adding to array with a bitmap, then we can accept an
1203                  * older device, but not too old.
1204                  */
1205                 if (ev1 < mddev->bitmap->events_cleared)
1206                         return 0;
1207         } else {
1208                 if (ev1 < mddev->events)
1209                         /* just a hot-add of a new device, leave raid_disk at -1 */
1210                         return 0;
1211         }
1212         if (mddev->level != LEVEL_MULTIPATH) {
1213                 int role;
1214                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1215                 switch(role) {
1216                 case 0xffff: /* spare */
1217                         break;
1218                 case 0xfffe: /* faulty */
1219                         set_bit(Faulty, &rdev->flags);
1220                         break;
1221                 default:
1222                         if ((le32_to_cpu(sb->feature_map) &
1223                              MD_FEATURE_RECOVERY_OFFSET))
1224                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1225                         else
1226                                 set_bit(In_sync, &rdev->flags);
1227                         rdev->raid_disk = role;
1228                         break;
1229                 }
1230                 if (sb->devflags & WriteMostly1)
1231                         set_bit(WriteMostly, &rdev->flags);
1232         } else /* MULTIPATH are always insync */
1233                 set_bit(In_sync, &rdev->flags);
1234
1235         return 0;
1236 }
1237
1238 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1239 {
1240         struct mdp_superblock_1 *sb;
1241         struct list_head *tmp;
1242         mdk_rdev_t *rdev2;
1243         int max_dev, i;
1244         /* make rdev->sb match mddev and rdev data. */
1245
1246         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1247
1248         sb->feature_map = 0;
1249         sb->pad0 = 0;
1250         sb->recovery_offset = cpu_to_le64(0);
1251         memset(sb->pad1, 0, sizeof(sb->pad1));
1252         memset(sb->pad2, 0, sizeof(sb->pad2));
1253         memset(sb->pad3, 0, sizeof(sb->pad3));
1254
1255         sb->utime = cpu_to_le64((__u64)mddev->utime);
1256         sb->events = cpu_to_le64(mddev->events);
1257         if (mddev->in_sync)
1258                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1259         else
1260                 sb->resync_offset = cpu_to_le64(0);
1261
1262         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1263
1264         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1265         sb->size = cpu_to_le64(mddev->size<<1);
1266
1267         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1268                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1269                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1270         }
1271
1272         if (rdev->raid_disk >= 0 &&
1273             !test_bit(In_sync, &rdev->flags) &&
1274             rdev->recovery_offset > 0) {
1275                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1276                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1277         }
1278
1279         if (mddev->reshape_position != MaxSector) {
1280                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1281                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1282                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1283                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1284                 sb->new_level = cpu_to_le32(mddev->new_level);
1285                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1286         }
1287
1288         max_dev = 0;
1289         ITERATE_RDEV(mddev,rdev2,tmp)
1290                 if (rdev2->desc_nr+1 > max_dev)
1291                         max_dev = rdev2->desc_nr+1;
1292
1293         if (max_dev > le32_to_cpu(sb->max_dev))
1294                 sb->max_dev = cpu_to_le32(max_dev);
1295         for (i=0; i<max_dev;i++)
1296                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1297
1298         ITERATE_RDEV(mddev,rdev2,tmp) {
1299                 i = rdev2->desc_nr;
1300                 if (test_bit(Faulty, &rdev2->flags))
1301                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1302                 else if (test_bit(In_sync, &rdev2->flags))
1303                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1304                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1305                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1306                 else
1307                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1308         }
1309
1310         sb->sb_csum = calc_sb_1_csum(sb);
1311 }
1312
1313
1314 static struct super_type super_types[] = {
1315         [0] = {
1316                 .name   = "0.90.0",
1317                 .owner  = THIS_MODULE,
1318                 .load_super     = super_90_load,
1319                 .validate_super = super_90_validate,
1320                 .sync_super     = super_90_sync,
1321         },
1322         [1] = {
1323                 .name   = "md-1",
1324                 .owner  = THIS_MODULE,
1325                 .load_super     = super_1_load,
1326                 .validate_super = super_1_validate,
1327                 .sync_super     = super_1_sync,
1328         },
1329 };
1330
1331 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1332 {
1333         struct list_head *tmp, *tmp2;
1334         mdk_rdev_t *rdev, *rdev2;
1335
1336         ITERATE_RDEV(mddev1,rdev,tmp)
1337                 ITERATE_RDEV(mddev2, rdev2, tmp2)
1338                         if (rdev->bdev->bd_contains ==
1339                             rdev2->bdev->bd_contains)
1340                                 return 1;
1341
1342         return 0;
1343 }
1344
1345 static LIST_HEAD(pending_raid_disks);
1346
1347 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1348 {
1349         char b[BDEVNAME_SIZE];
1350         struct kobject *ko;
1351         char *s;
1352         int err;
1353
1354         if (rdev->mddev) {
1355                 MD_BUG();
1356                 return -EINVAL;
1357         }
1358         /* make sure rdev->size exceeds mddev->size */
1359         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1360                 if (mddev->pers) {
1361                         /* Cannot change size, so fail
1362                          * If mddev->level <= 0, then we don't care
1363                          * about aligning sizes (e.g. linear)
1364                          */
1365                         if (mddev->level > 0)
1366                                 return -ENOSPC;
1367                 } else
1368                         mddev->size = rdev->size;
1369         }
1370
1371         /* Verify rdev->desc_nr is unique.
1372          * If it is -1, assign a free number, else
1373          * check number is not in use
1374          */
1375         if (rdev->desc_nr < 0) {
1376                 int choice = 0;
1377                 if (mddev->pers) choice = mddev->raid_disks;
1378                 while (find_rdev_nr(mddev, choice))
1379                         choice++;
1380                 rdev->desc_nr = choice;
1381         } else {
1382                 if (find_rdev_nr(mddev, rdev->desc_nr))
1383                         return -EBUSY;
1384         }
1385         bdevname(rdev->bdev,b);
1386         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1387                 return -ENOMEM;
1388         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1389                 *s = '!';
1390
1391         rdev->mddev = mddev;
1392         printk(KERN_INFO "md: bind<%s>\n", b);
1393
1394         rdev->kobj.parent = &mddev->kobj;
1395         if ((err = kobject_add(&rdev->kobj)))
1396                 goto fail;
1397
1398         if (rdev->bdev->bd_part)
1399                 ko = &rdev->bdev->bd_part->kobj;
1400         else
1401                 ko = &rdev->bdev->bd_disk->kobj;
1402         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1403                 kobject_del(&rdev->kobj);
1404                 goto fail;
1405         }
1406         list_add(&rdev->same_set, &mddev->disks);
1407         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1408         return 0;
1409
1410  fail:
1411         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1412                b, mdname(mddev));
1413         return err;
1414 }
1415
1416 static void delayed_delete(struct work_struct *ws)
1417 {
1418         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1419         kobject_del(&rdev->kobj);
1420 }
1421
1422 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1423 {
1424         char b[BDEVNAME_SIZE];
1425         if (!rdev->mddev) {
1426                 MD_BUG();
1427                 return;
1428         }
1429         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1430         list_del_init(&rdev->same_set);
1431         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1432         rdev->mddev = NULL;
1433         sysfs_remove_link(&rdev->kobj, "block");
1434
1435         /* We need to delay this, otherwise we can deadlock when
1436          * writing to 'remove' to "dev/state"
1437          */
1438         INIT_WORK(&rdev->del_work, delayed_delete);
1439         schedule_work(&rdev->del_work);
1440 }
1441
1442 /*
1443  * prevent the device from being mounted, repartitioned or
1444  * otherwise reused by a RAID array (or any other kernel
1445  * subsystem), by bd_claiming the device.
1446  */
1447 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1448 {
1449         int err = 0;
1450         struct block_device *bdev;
1451         char b[BDEVNAME_SIZE];
1452
1453         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1454         if (IS_ERR(bdev)) {
1455                 printk(KERN_ERR "md: could not open %s.\n",
1456                         __bdevname(dev, b));
1457                 return PTR_ERR(bdev);
1458         }
1459         err = bd_claim(bdev, rdev);
1460         if (err) {
1461                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1462                         bdevname(bdev, b));
1463                 blkdev_put(bdev);
1464                 return err;
1465         }
1466         rdev->bdev = bdev;
1467         return err;
1468 }
1469
1470 static void unlock_rdev(mdk_rdev_t *rdev)
1471 {
1472         struct block_device *bdev = rdev->bdev;
1473         rdev->bdev = NULL;
1474         if (!bdev)
1475                 MD_BUG();
1476         bd_release(bdev);
1477         blkdev_put(bdev);
1478 }
1479
1480 void md_autodetect_dev(dev_t dev);
1481
1482 static void export_rdev(mdk_rdev_t * rdev)
1483 {
1484         char b[BDEVNAME_SIZE];
1485         printk(KERN_INFO "md: export_rdev(%s)\n",
1486                 bdevname(rdev->bdev,b));
1487         if (rdev->mddev)
1488                 MD_BUG();
1489         free_disk_sb(rdev);
1490         list_del_init(&rdev->same_set);
1491 #ifndef MODULE
1492         md_autodetect_dev(rdev->bdev->bd_dev);
1493 #endif
1494         unlock_rdev(rdev);
1495         kobject_put(&rdev->kobj);
1496 }
1497
1498 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1499 {
1500         unbind_rdev_from_array(rdev);
1501         export_rdev(rdev);
1502 }
1503
1504 static void export_array(mddev_t *mddev)
1505 {
1506         struct list_head *tmp;
1507         mdk_rdev_t *rdev;
1508
1509         ITERATE_RDEV(mddev,rdev,tmp) {
1510                 if (!rdev->mddev) {
1511                         MD_BUG();
1512                         continue;
1513                 }
1514                 kick_rdev_from_array(rdev);
1515         }
1516         if (!list_empty(&mddev->disks))
1517                 MD_BUG();
1518         mddev->raid_disks = 0;
1519         mddev->major_version = 0;
1520 }
1521
1522 static void print_desc(mdp_disk_t *desc)
1523 {
1524         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1525                 desc->major,desc->minor,desc->raid_disk,desc->state);
1526 }
1527
1528 static void print_sb(mdp_super_t *sb)
1529 {
1530         int i;
1531
1532         printk(KERN_INFO
1533                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1534                 sb->major_version, sb->minor_version, sb->patch_version,
1535                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1536                 sb->ctime);
1537         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1538                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1539                 sb->md_minor, sb->layout, sb->chunk_size);
1540         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1541                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1542                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1543                 sb->failed_disks, sb->spare_disks,
1544                 sb->sb_csum, (unsigned long)sb->events_lo);
1545
1546         printk(KERN_INFO);
1547         for (i = 0; i < MD_SB_DISKS; i++) {
1548                 mdp_disk_t *desc;
1549
1550                 desc = sb->disks + i;
1551                 if (desc->number || desc->major || desc->minor ||
1552                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1553                         printk("     D %2d: ", i);
1554                         print_desc(desc);
1555                 }
1556         }
1557         printk(KERN_INFO "md:     THIS: ");
1558         print_desc(&sb->this_disk);
1559
1560 }
1561
1562 static void print_rdev(mdk_rdev_t *rdev)
1563 {
1564         char b[BDEVNAME_SIZE];
1565         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1566                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1567                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1568                 rdev->desc_nr);
1569         if (rdev->sb_loaded) {
1570                 printk(KERN_INFO "md: rdev superblock:\n");
1571                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1572         } else
1573                 printk(KERN_INFO "md: no rdev superblock!\n");
1574 }
1575
1576 static void md_print_devices(void)
1577 {
1578         struct list_head *tmp, *tmp2;
1579         mdk_rdev_t *rdev;
1580         mddev_t *mddev;
1581         char b[BDEVNAME_SIZE];
1582
1583         printk("\n");
1584         printk("md:     **********************************\n");
1585         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1586         printk("md:     **********************************\n");
1587         ITERATE_MDDEV(mddev,tmp) {
1588
1589                 if (mddev->bitmap)
1590                         bitmap_print_sb(mddev->bitmap);
1591                 else
1592                         printk("%s: ", mdname(mddev));
1593                 ITERATE_RDEV(mddev,rdev,tmp2)
1594                         printk("<%s>", bdevname(rdev->bdev,b));
1595                 printk("\n");
1596
1597                 ITERATE_RDEV(mddev,rdev,tmp2)
1598                         print_rdev(rdev);
1599         }
1600         printk("md:     **********************************\n");
1601         printk("\n");
1602 }
1603
1604
1605 static void sync_sbs(mddev_t * mddev, int nospares)
1606 {
1607         /* Update each superblock (in-memory image), but
1608          * if we are allowed to, skip spares which already
1609          * have the right event counter, or have one earlier
1610          * (which would mean they aren't being marked as dirty
1611          * with the rest of the array)
1612          */
1613         mdk_rdev_t *rdev;
1614         struct list_head *tmp;
1615
1616         ITERATE_RDEV(mddev,rdev,tmp) {
1617                 if (rdev->sb_events == mddev->events ||
1618                     (nospares &&
1619                      rdev->raid_disk < 0 &&
1620                      (rdev->sb_events&1)==0 &&
1621                      rdev->sb_events+1 == mddev->events)) {
1622                         /* Don't update this superblock */
1623                         rdev->sb_loaded = 2;
1624                 } else {
1625                         super_types[mddev->major_version].
1626                                 sync_super(mddev, rdev);
1627                         rdev->sb_loaded = 1;
1628                 }
1629         }
1630 }
1631
1632 static void md_update_sb(mddev_t * mddev, int force_change)
1633 {
1634         struct list_head *tmp;
1635         mdk_rdev_t *rdev;
1636         int sync_req;
1637         int nospares = 0;
1638
1639 repeat:
1640         spin_lock_irq(&mddev->write_lock);
1641
1642         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1643         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1644                 force_change = 1;
1645         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1646                 /* just a clean<-> dirty transition, possibly leave spares alone,
1647                  * though if events isn't the right even/odd, we will have to do
1648                  * spares after all
1649                  */
1650                 nospares = 1;
1651         if (force_change)
1652                 nospares = 0;
1653         if (mddev->degraded)
1654                 /* If the array is degraded, then skipping spares is both
1655                  * dangerous and fairly pointless.
1656                  * Dangerous because a device that was removed from the array
1657                  * might have a event_count that still looks up-to-date,
1658                  * so it can be re-added without a resync.
1659                  * Pointless because if there are any spares to skip,
1660                  * then a recovery will happen and soon that array won't
1661                  * be degraded any more and the spare can go back to sleep then.
1662                  */
1663                 nospares = 0;
1664
1665         sync_req = mddev->in_sync;
1666         mddev->utime = get_seconds();
1667
1668         /* If this is just a dirty<->clean transition, and the array is clean
1669          * and 'events' is odd, we can roll back to the previous clean state */
1670         if (nospares
1671             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1672             && (mddev->events & 1)
1673             && mddev->events != 1)
1674                 mddev->events--;
1675         else {
1676                 /* otherwise we have to go forward and ... */
1677                 mddev->events ++;
1678                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1679                         /* .. if the array isn't clean, insist on an odd 'events' */
1680                         if ((mddev->events&1)==0) {
1681                                 mddev->events++;
1682                                 nospares = 0;
1683                         }
1684                 } else {
1685                         /* otherwise insist on an even 'events' (for clean states) */
1686                         if ((mddev->events&1)) {
1687                                 mddev->events++;
1688                                 nospares = 0;
1689                         }
1690                 }
1691         }
1692
1693         if (!mddev->events) {
1694                 /*
1695                  * oops, this 64-bit counter should never wrap.
1696                  * Either we are in around ~1 trillion A.C., assuming
1697                  * 1 reboot per second, or we have a bug:
1698                  */
1699                 MD_BUG();
1700                 mddev->events --;
1701         }
1702         sync_sbs(mddev, nospares);
1703
1704         /*
1705          * do not write anything to disk if using
1706          * nonpersistent superblocks
1707          */
1708         if (!mddev->persistent) {
1709                 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1710                 spin_unlock_irq(&mddev->write_lock);
1711                 wake_up(&mddev->sb_wait);
1712                 return;
1713         }
1714         spin_unlock_irq(&mddev->write_lock);
1715
1716         dprintk(KERN_INFO
1717                 "md: updating %s RAID superblock on device (in sync %d)\n",
1718                 mdname(mddev),mddev->in_sync);
1719
1720         bitmap_update_sb(mddev->bitmap);
1721         ITERATE_RDEV(mddev,rdev,tmp) {
1722                 char b[BDEVNAME_SIZE];
1723                 dprintk(KERN_INFO "md: ");
1724                 if (rdev->sb_loaded != 1)
1725                         continue; /* no noise on spare devices */
1726                 if (test_bit(Faulty, &rdev->flags))
1727                         dprintk("(skipping faulty ");
1728
1729                 dprintk("%s ", bdevname(rdev->bdev,b));
1730                 if (!test_bit(Faulty, &rdev->flags)) {
1731                         md_super_write(mddev,rdev,
1732                                        rdev->sb_offset<<1, rdev->sb_size,
1733                                        rdev->sb_page);
1734                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1735                                 bdevname(rdev->bdev,b),
1736                                 (unsigned long long)rdev->sb_offset);
1737                         rdev->sb_events = mddev->events;
1738
1739                 } else
1740                         dprintk(")\n");
1741                 if (mddev->level == LEVEL_MULTIPATH)
1742                         /* only need to write one superblock... */
1743                         break;
1744         }
1745         md_super_wait(mddev);
1746         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1747
1748         spin_lock_irq(&mddev->write_lock);
1749         if (mddev->in_sync != sync_req ||
1750             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1751                 /* have to write it out again */
1752                 spin_unlock_irq(&mddev->write_lock);
1753                 goto repeat;
1754         }
1755         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1756         spin_unlock_irq(&mddev->write_lock);
1757         wake_up(&mddev->sb_wait);
1758
1759 }
1760
1761 /* words written to sysfs files may, or my not, be \n terminated.
1762  * We want to accept with case. For this we use cmd_match.
1763  */
1764 static int cmd_match(const char *cmd, const char *str)
1765 {
1766         /* See if cmd, written into a sysfs file, matches
1767          * str.  They must either be the same, or cmd can
1768          * have a trailing newline
1769          */
1770         while (*cmd && *str && *cmd == *str) {
1771                 cmd++;
1772                 str++;
1773         }
1774         if (*cmd == '\n')
1775                 cmd++;
1776         if (*str || *cmd)
1777                 return 0;
1778         return 1;
1779 }
1780
1781 struct rdev_sysfs_entry {
1782         struct attribute attr;
1783         ssize_t (*show)(mdk_rdev_t *, char *);
1784         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1785 };
1786
1787 static ssize_t
1788 state_show(mdk_rdev_t *rdev, char *page)
1789 {
1790         char *sep = "";
1791         int len=0;
1792
1793         if (test_bit(Faulty, &rdev->flags)) {
1794                 len+= sprintf(page+len, "%sfaulty",sep);
1795                 sep = ",";
1796         }
1797         if (test_bit(In_sync, &rdev->flags)) {
1798                 len += sprintf(page+len, "%sin_sync",sep);
1799                 sep = ",";
1800         }
1801         if (test_bit(WriteMostly, &rdev->flags)) {
1802                 len += sprintf(page+len, "%swrite_mostly",sep);
1803                 sep = ",";
1804         }
1805         if (!test_bit(Faulty, &rdev->flags) &&
1806             !test_bit(In_sync, &rdev->flags)) {
1807                 len += sprintf(page+len, "%sspare", sep);
1808                 sep = ",";
1809         }
1810         return len+sprintf(page+len, "\n");
1811 }
1812
1813 static ssize_t
1814 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1815 {
1816         /* can write
1817          *  faulty  - simulates and error
1818          *  remove  - disconnects the device
1819          *  writemostly - sets write_mostly
1820          *  -writemostly - clears write_mostly
1821          */
1822         int err = -EINVAL;
1823         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1824                 md_error(rdev->mddev, rdev);
1825                 err = 0;
1826         } else if (cmd_match(buf, "remove")) {
1827                 if (rdev->raid_disk >= 0)
1828                         err = -EBUSY;
1829                 else {
1830                         mddev_t *mddev = rdev->mddev;
1831                         kick_rdev_from_array(rdev);
1832                         if (mddev->pers)
1833                                 md_update_sb(mddev, 1);
1834                         md_new_event(mddev);
1835                         err = 0;
1836                 }
1837         } else if (cmd_match(buf, "writemostly")) {
1838                 set_bit(WriteMostly, &rdev->flags);
1839                 err = 0;
1840         } else if (cmd_match(buf, "-writemostly")) {
1841                 clear_bit(WriteMostly, &rdev->flags);
1842                 err = 0;
1843         }
1844         return err ? err : len;
1845 }
1846 static struct rdev_sysfs_entry rdev_state =
1847 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1848
1849 static ssize_t
1850 super_show(mdk_rdev_t *rdev, char *page)
1851 {
1852         if (rdev->sb_loaded && rdev->sb_size) {
1853                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1854                 return rdev->sb_size;
1855         } else
1856                 return 0;
1857 }
1858 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1859
1860 static ssize_t
1861 errors_show(mdk_rdev_t *rdev, char *page)
1862 {
1863         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1864 }
1865
1866 static ssize_t
1867 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1868 {
1869         char *e;
1870         unsigned long n = simple_strtoul(buf, &e, 10);
1871         if (*buf && (*e == 0 || *e == '\n')) {
1872                 atomic_set(&rdev->corrected_errors, n);
1873                 return len;
1874         }
1875         return -EINVAL;
1876 }
1877 static struct rdev_sysfs_entry rdev_errors =
1878 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1879
1880 static ssize_t
1881 slot_show(mdk_rdev_t *rdev, char *page)
1882 {
1883         if (rdev->raid_disk < 0)
1884                 return sprintf(page, "none\n");
1885         else
1886                 return sprintf(page, "%d\n", rdev->raid_disk);
1887 }
1888
1889 static ssize_t
1890 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1891 {
1892         char *e;
1893         int slot = simple_strtoul(buf, &e, 10);
1894         if (strncmp(buf, "none", 4)==0)
1895                 slot = -1;
1896         else if (e==buf || (*e && *e!= '\n'))
1897                 return -EINVAL;
1898         if (rdev->mddev->pers)
1899                 /* Cannot set slot in active array (yet) */
1900                 return -EBUSY;
1901         if (slot >= rdev->mddev->raid_disks)
1902                 return -ENOSPC;
1903         rdev->raid_disk = slot;
1904         /* assume it is working */
1905         rdev->flags = 0;
1906         set_bit(In_sync, &rdev->flags);
1907         return len;
1908 }
1909
1910
1911 static struct rdev_sysfs_entry rdev_slot =
1912 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1913
1914 static ssize_t
1915 offset_show(mdk_rdev_t *rdev, char *page)
1916 {
1917         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1918 }
1919
1920 static ssize_t
1921 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1922 {
1923         char *e;
1924         unsigned long long offset = simple_strtoull(buf, &e, 10);
1925         if (e==buf || (*e && *e != '\n'))
1926                 return -EINVAL;
1927         if (rdev->mddev->pers)
1928                 return -EBUSY;
1929         rdev->data_offset = offset;
1930         return len;
1931 }
1932
1933 static struct rdev_sysfs_entry rdev_offset =
1934 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1935
1936 static ssize_t
1937 rdev_size_show(mdk_rdev_t *rdev, char *page)
1938 {
1939         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1940 }
1941
1942 static ssize_t
1943 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1944 {
1945         char *e;
1946         unsigned long long size = simple_strtoull(buf, &e, 10);
1947         if (e==buf || (*e && *e != '\n'))
1948                 return -EINVAL;
1949         if (rdev->mddev->pers)
1950                 return -EBUSY;
1951         rdev->size = size;
1952         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1953                 rdev->mddev->size = size;
1954         return len;
1955 }
1956
1957 static struct rdev_sysfs_entry rdev_size =
1958 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1959
1960 static struct attribute *rdev_default_attrs[] = {
1961         &rdev_state.attr,
1962         &rdev_super.attr,
1963         &rdev_errors.attr,
1964         &rdev_slot.attr,
1965         &rdev_offset.attr,
1966         &rdev_size.attr,
1967         NULL,
1968 };
1969 static ssize_t
1970 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1971 {
1972         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1973         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1974
1975         if (!entry->show)
1976                 return -EIO;
1977         return entry->show(rdev, page);
1978 }
1979
1980 static ssize_t
1981 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1982               const char *page, size_t length)
1983 {
1984         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1985         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1986
1987         if (!entry->store)
1988                 return -EIO;
1989         if (!capable(CAP_SYS_ADMIN))
1990                 return -EACCES;
1991         return entry->store(rdev, page, length);
1992 }
1993
1994 static void rdev_free(struct kobject *ko)
1995 {
1996         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1997         kfree(rdev);
1998 }
1999 static struct sysfs_ops rdev_sysfs_ops = {
2000         .show           = rdev_attr_show,
2001         .store          = rdev_attr_store,
2002 };
2003 static struct kobj_type rdev_ktype = {
2004         .release        = rdev_free,
2005         .sysfs_ops      = &rdev_sysfs_ops,
2006         .default_attrs  = rdev_default_attrs,
2007 };
2008
2009 /*
2010  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2011  *
2012  * mark the device faulty if:
2013  *
2014  *   - the device is nonexistent (zero size)
2015  *   - the device has no valid superblock
2016  *
2017  * a faulty rdev _never_ has rdev->sb set.
2018  */
2019 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2020 {
2021         char b[BDEVNAME_SIZE];
2022         int err;
2023         mdk_rdev_t *rdev;
2024         sector_t size;
2025
2026         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2027         if (!rdev) {
2028                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2029                 return ERR_PTR(-ENOMEM);
2030         }
2031
2032         if ((err = alloc_disk_sb(rdev)))
2033                 goto abort_free;
2034
2035         err = lock_rdev(rdev, newdev);
2036         if (err)
2037                 goto abort_free;
2038
2039         rdev->kobj.parent = NULL;
2040         rdev->kobj.ktype = &rdev_ktype;
2041         kobject_init(&rdev->kobj);
2042
2043         rdev->desc_nr = -1;
2044         rdev->saved_raid_disk = -1;
2045         rdev->raid_disk = -1;
2046         rdev->flags = 0;
2047         rdev->data_offset = 0;
2048         rdev->sb_events = 0;
2049         atomic_set(&rdev->nr_pending, 0);
2050         atomic_set(&rdev->read_errors, 0);
2051         atomic_set(&rdev->corrected_errors, 0);
2052
2053         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2054         if (!size) {
2055                 printk(KERN_WARNING
2056                         "md: %s has zero or unknown size, marking faulty!\n",
2057                         bdevname(rdev->bdev,b));
2058                 err = -EINVAL;
2059                 goto abort_free;
2060         }
2061
2062         if (super_format >= 0) {
2063                 err = super_types[super_format].
2064                         load_super(rdev, NULL, super_minor);
2065                 if (err == -EINVAL) {
2066                         printk(KERN_WARNING
2067                                 "md: %s does not have a valid v%d.%d "
2068                                "superblock, not importing!\n",
2069                                 bdevname(rdev->bdev,b),
2070                                super_format, super_minor);
2071                         goto abort_free;
2072                 }
2073                 if (err < 0) {
2074                         printk(KERN_WARNING
2075                                 "md: could not read %s's sb, not importing!\n",
2076                                 bdevname(rdev->bdev,b));
2077                         goto abort_free;
2078                 }
2079         }
2080         INIT_LIST_HEAD(&rdev->same_set);
2081
2082         return rdev;
2083
2084 abort_free:
2085         if (rdev->sb_page) {
2086                 if (rdev->bdev)
2087                         unlock_rdev(rdev);
2088                 free_disk_sb(rdev);
2089         }
2090         kfree(rdev);
2091         return ERR_PTR(err);
2092 }
2093
2094 /*
2095  * Check a full RAID array for plausibility
2096  */
2097
2098
2099 static void analyze_sbs(mddev_t * mddev)
2100 {
2101         int i;
2102         struct list_head *tmp;
2103         mdk_rdev_t *rdev, *freshest;
2104         char b[BDEVNAME_SIZE];
2105
2106         freshest = NULL;
2107         ITERATE_RDEV(mddev,rdev,tmp)
2108                 switch (super_types[mddev->major_version].
2109                         load_super(rdev, freshest, mddev->minor_version)) {
2110                 case 1:
2111                         freshest = rdev;
2112                         break;
2113                 case 0:
2114                         break;
2115                 default:
2116                         printk( KERN_ERR \
2117                                 "md: fatal superblock inconsistency in %s"
2118                                 " -- removing from array\n",
2119                                 bdevname(rdev->bdev,b));
2120                         kick_rdev_from_array(rdev);
2121                 }
2122
2123
2124         super_types[mddev->major_version].
2125                 validate_super(mddev, freshest);
2126
2127         i = 0;
2128         ITERATE_RDEV(mddev,rdev,tmp) {
2129                 if (rdev != freshest)
2130                         if (super_types[mddev->major_version].
2131                             validate_super(mddev, rdev)) {
2132                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2133                                         " from array!\n",
2134                                         bdevname(rdev->bdev,b));
2135                                 kick_rdev_from_array(rdev);
2136                                 continue;
2137                         }
2138                 if (mddev->level == LEVEL_MULTIPATH) {
2139                         rdev->desc_nr = i++;
2140                         rdev->raid_disk = rdev->desc_nr;
2141                         set_bit(In_sync, &rdev->flags);
2142                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2143                         rdev->raid_disk = -1;
2144                         clear_bit(In_sync, &rdev->flags);
2145                 }
2146         }
2147
2148
2149
2150         if (mddev->recovery_cp != MaxSector &&
2151             mddev->level >= 1)
2152                 printk(KERN_ERR "md: %s: raid array is not clean"
2153                        " -- starting background reconstruction\n",
2154                        mdname(mddev));
2155
2156 }
2157
2158 static ssize_t
2159 safe_delay_show(mddev_t *mddev, char *page)
2160 {
2161         int msec = (mddev->safemode_delay*1000)/HZ;
2162         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2163 }
2164 static ssize_t
2165 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2166 {
2167         int scale=1;
2168         int dot=0;
2169         int i;
2170         unsigned long msec;
2171         char buf[30];
2172         char *e;
2173         /* remove a period, and count digits after it */
2174         if (len >= sizeof(buf))
2175                 return -EINVAL;
2176         strlcpy(buf, cbuf, len);
2177         buf[len] = 0;
2178         for (i=0; i<len; i++) {
2179                 if (dot) {
2180                         if (isdigit(buf[i])) {
2181                                 buf[i-1] = buf[i];
2182                                 scale *= 10;
2183                         }
2184                         buf[i] = 0;
2185                 } else if (buf[i] == '.') {
2186                         dot=1;
2187                         buf[i] = 0;
2188                 }
2189         }
2190         msec = simple_strtoul(buf, &e, 10);
2191         if (e == buf || (*e && *e != '\n'))
2192                 return -EINVAL;
2193         msec = (msec * 1000) / scale;
2194         if (msec == 0)
2195                 mddev->safemode_delay = 0;
2196         else {
2197                 mddev->safemode_delay = (msec*HZ)/1000;
2198                 if (mddev->safemode_delay == 0)
2199                         mddev->safemode_delay = 1;
2200         }
2201         return len;
2202 }
2203 static struct md_sysfs_entry md_safe_delay =
2204 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2205
2206 static ssize_t
2207 level_show(mddev_t *mddev, char *page)
2208 {
2209         struct mdk_personality *p = mddev->pers;
2210         if (p)
2211                 return sprintf(page, "%s\n", p->name);
2212         else if (mddev->clevel[0])
2213                 return sprintf(page, "%s\n", mddev->clevel);
2214         else if (mddev->level != LEVEL_NONE)
2215                 return sprintf(page, "%d\n", mddev->level);
2216         else
2217                 return 0;
2218 }
2219
2220 static ssize_t
2221 level_store(mddev_t *mddev, const char *buf, size_t len)
2222 {
2223         int rv = len;
2224         if (mddev->pers)
2225                 return -EBUSY;
2226         if (len == 0)
2227                 return 0;
2228         if (len >= sizeof(mddev->clevel))
2229                 return -ENOSPC;
2230         strncpy(mddev->clevel, buf, len);
2231         if (mddev->clevel[len-1] == '\n')
2232                 len--;
2233         mddev->clevel[len] = 0;
2234         mddev->level = LEVEL_NONE;
2235         return rv;
2236 }
2237
2238 static struct md_sysfs_entry md_level =
2239 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2240
2241
2242 static ssize_t
2243 layout_show(mddev_t *mddev, char *page)
2244 {
2245         /* just a number, not meaningful for all levels */
2246         if (mddev->reshape_position != MaxSector &&
2247             mddev->layout != mddev->new_layout)
2248                 return sprintf(page, "%d (%d)\n",
2249                                mddev->new_layout, mddev->layout);
2250         return sprintf(page, "%d\n", mddev->layout);
2251 }
2252
2253 static ssize_t
2254 layout_store(mddev_t *mddev, const char *buf, size_t len)
2255 {
2256         char *e;
2257         unsigned long n = simple_strtoul(buf, &e, 10);
2258
2259         if (!*buf || (*e && *e != '\n'))
2260                 return -EINVAL;
2261
2262         if (mddev->pers)
2263                 return -EBUSY;
2264         if (mddev->reshape_position != MaxSector)
2265                 mddev->new_layout = n;
2266         else
2267                 mddev->layout = n;
2268         return len;
2269 }
2270 static struct md_sysfs_entry md_layout =
2271 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2272
2273
2274 static ssize_t
2275 raid_disks_show(mddev_t *mddev, char *page)
2276 {
2277         if (mddev->raid_disks == 0)
2278                 return 0;
2279         if (mddev->reshape_position != MaxSector &&
2280             mddev->delta_disks != 0)
2281                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2282                                mddev->raid_disks - mddev->delta_disks);
2283         return sprintf(page, "%d\n", mddev->raid_disks);
2284 }
2285
2286 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2287
2288 static ssize_t
2289 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2290 {
2291         char *e;
2292         int rv = 0;
2293         unsigned long n = simple_strtoul(buf, &e, 10);
2294
2295         if (!*buf || (*e && *e != '\n'))
2296                 return -EINVAL;
2297
2298         if (mddev->pers)
2299                 rv = update_raid_disks(mddev, n);
2300         else if (mddev->reshape_position != MaxSector) {
2301                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2302                 mddev->delta_disks = n - olddisks;
2303                 mddev->raid_disks = n;
2304         } else
2305                 mddev->raid_disks = n;
2306         return rv ? rv : len;
2307 }
2308 static struct md_sysfs_entry md_raid_disks =
2309 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2310
2311 static ssize_t
2312 chunk_size_show(mddev_t *mddev, char *page)
2313 {
2314         if (mddev->reshape_position != MaxSector &&
2315             mddev->chunk_size != mddev->new_chunk)
2316                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2317                                mddev->chunk_size);
2318         return sprintf(page, "%d\n", mddev->chunk_size);
2319 }
2320
2321 static ssize_t
2322 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2323 {
2324         /* can only set chunk_size if array is not yet active */
2325         char *e;
2326         unsigned long n = simple_strtoul(buf, &e, 10);
2327
2328         if (!*buf || (*e && *e != '\n'))
2329                 return -EINVAL;
2330
2331         if (mddev->pers)
2332                 return -EBUSY;
2333         else if (mddev->reshape_position != MaxSector)
2334                 mddev->new_chunk = n;
2335         else
2336                 mddev->chunk_size = n;
2337         return len;
2338 }
2339 static struct md_sysfs_entry md_chunk_size =
2340 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2341
2342 static ssize_t
2343 resync_start_show(mddev_t *mddev, char *page)
2344 {
2345         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2346 }
2347
2348 static ssize_t
2349 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2350 {
2351         /* can only set chunk_size if array is not yet active */
2352         char *e;
2353         unsigned long long n = simple_strtoull(buf, &e, 10);
2354
2355         if (mddev->pers)
2356                 return -EBUSY;
2357         if (!*buf || (*e && *e != '\n'))
2358                 return -EINVAL;
2359
2360         mddev->recovery_cp = n;
2361         return len;
2362 }
2363 static struct md_sysfs_entry md_resync_start =
2364 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2365
2366 /*
2367  * The array state can be:
2368  *
2369  * clear
2370  *     No devices, no size, no level
2371  *     Equivalent to STOP_ARRAY ioctl
2372  * inactive
2373  *     May have some settings, but array is not active
2374  *        all IO results in error
2375  *     When written, doesn't tear down array, but just stops it
2376  * suspended (not supported yet)
2377  *     All IO requests will block. The array can be reconfigured.
2378  *     Writing this, if accepted, will block until array is quiessent
2379  * readonly
2380  *     no resync can happen.  no superblocks get written.
2381  *     write requests fail
2382  * read-auto
2383  *     like readonly, but behaves like 'clean' on a write request.
2384  *
2385  * clean - no pending writes, but otherwise active.
2386  *     When written to inactive array, starts without resync
2387  *     If a write request arrives then
2388  *       if metadata is known, mark 'dirty' and switch to 'active'.
2389  *       if not known, block and switch to write-pending
2390  *     If written to an active array that has pending writes, then fails.
2391  * active
2392  *     fully active: IO and resync can be happening.
2393  *     When written to inactive array, starts with resync
2394  *
2395  * write-pending
2396  *     clean, but writes are blocked waiting for 'active' to be written.
2397  *
2398  * active-idle
2399  *     like active, but no writes have been seen for a while (100msec).
2400  *
2401  */
2402 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2403                    write_pending, active_idle, bad_word};
2404 static char *array_states[] = {
2405         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2406         "write-pending", "active-idle", NULL };
2407
2408 static int match_word(const char *word, char **list)
2409 {
2410         int n;
2411         for (n=0; list[n]; n++)
2412                 if (cmd_match(word, list[n]))
2413                         break;
2414         return n;
2415 }
2416
2417 static ssize_t
2418 array_state_show(mddev_t *mddev, char *page)
2419 {
2420         enum array_state st = inactive;
2421
2422         if (mddev->pers)
2423                 switch(mddev->ro) {
2424                 case 1:
2425                         st = readonly;
2426                         break;
2427                 case 2:
2428                         st = read_auto;
2429                         break;
2430                 case 0:
2431                         if (mddev->in_sync)
2432                                 st = clean;
2433                         else if (mddev->safemode)
2434                                 st = active_idle;
2435                         else
2436                                 st = active;
2437                 }
2438         else {
2439                 if (list_empty(&mddev->disks) &&
2440                     mddev->raid_disks == 0 &&
2441                     mddev->size == 0)
2442                         st = clear;
2443                 else
2444                         st = inactive;
2445         }
2446         return sprintf(page, "%s\n", array_states[st]);
2447 }
2448
2449 static int do_md_stop(mddev_t * mddev, int ro);
2450 static int do_md_run(mddev_t * mddev);
2451 static int restart_array(mddev_t *mddev);
2452
2453 static ssize_t
2454 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2455 {
2456         int err = -EINVAL;
2457         enum array_state st = match_word(buf, array_states);
2458         switch(st) {
2459         case bad_word:
2460                 break;
2461         case clear:
2462                 /* stopping an active array */
2463                 if (mddev->pers) {
2464                         if (atomic_read(&mddev->active) > 1)
2465                                 return -EBUSY;
2466                         err = do_md_stop(mddev, 0);
2467                 }
2468                 break;
2469         case inactive:
2470                 /* stopping an active array */
2471                 if (mddev->pers) {
2472                         if (atomic_read(&mddev->active) > 1)
2473                                 return -EBUSY;
2474                         err = do_md_stop(mddev, 2);
2475                 }
2476                 break;
2477         case suspended:
2478                 break; /* not supported yet */
2479         case readonly:
2480                 if (mddev->pers)
2481                         err = do_md_stop(mddev, 1);
2482                 else {
2483                         mddev->ro = 1;
2484                         err = do_md_run(mddev);
2485                 }
2486                 break;
2487         case read_auto:
2488                 /* stopping an active array */
2489                 if (mddev->pers) {
2490                         err = do_md_stop(mddev, 1);
2491                         if (err == 0)
2492                                 mddev->ro = 2; /* FIXME mark devices writable */
2493                 } else {
2494                         mddev->ro = 2;
2495                         err = do_md_run(mddev);
2496                 }
2497                 break;
2498         case clean:
2499                 if (mddev->pers) {
2500                         restart_array(mddev);
2501                         spin_lock_irq(&mddev->write_lock);
2502                         if (atomic_read(&mddev->writes_pending) == 0) {
2503                                 mddev->in_sync = 1;
2504                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2505                         }
2506                         spin_unlock_irq(&mddev->write_lock);
2507                 } else {
2508                         mddev->ro = 0;
2509                         mddev->recovery_cp = MaxSector;
2510                         err = do_md_run(mddev);
2511                 }
2512                 break;
2513         case active:
2514                 if (mddev->pers) {
2515                         restart_array(mddev);
2516                         clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2517                         wake_up(&mddev->sb_wait);
2518                         err = 0;
2519                 } else {
2520                         mddev->ro = 0;
2521                         err = do_md_run(mddev);
2522                 }
2523                 break;
2524         case write_pending:
2525         case active_idle:
2526                 /* these cannot be set */
2527                 break;
2528         }
2529         if (err)
2530                 return err;
2531         else
2532                 return len;
2533 }
2534 static struct md_sysfs_entry md_array_state =
2535 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2536
2537 static ssize_t
2538 null_show(mddev_t *mddev, char *page)
2539 {
2540         return -EINVAL;
2541 }
2542
2543 static ssize_t
2544 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2545 {
2546         /* buf must be %d:%d\n? giving major and minor numbers */
2547         /* The new device is added to the array.
2548          * If the array has a persistent superblock, we read the
2549          * superblock to initialise info and check validity.
2550          * Otherwise, only checking done is that in bind_rdev_to_array,
2551          * which mainly checks size.
2552          */
2553         char *e;
2554         int major = simple_strtoul(buf, &e, 10);
2555         int minor;
2556         dev_t dev;
2557         mdk_rdev_t *rdev;
2558         int err;
2559
2560         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2561                 return -EINVAL;
2562         minor = simple_strtoul(e+1, &e, 10);
2563         if (*e && *e != '\n')
2564                 return -EINVAL;
2565         dev = MKDEV(major, minor);
2566         if (major != MAJOR(dev) ||
2567             minor != MINOR(dev))
2568                 return -EOVERFLOW;
2569
2570
2571         if (mddev->persistent) {
2572                 rdev = md_import_device(dev, mddev->major_version,
2573                                         mddev->minor_version);
2574                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2575                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2576                                                        mdk_rdev_t, same_set);
2577                         err = super_types[mddev->major_version]
2578                                 .load_super(rdev, rdev0, mddev->minor_version);
2579                         if (err < 0)
2580                                 goto out;
2581                 }
2582         } else
2583                 rdev = md_import_device(dev, -1, -1);
2584
2585         if (IS_ERR(rdev))
2586                 return PTR_ERR(rdev);
2587         err = bind_rdev_to_array(rdev, mddev);
2588  out:
2589         if (err)
2590                 export_rdev(rdev);
2591         return err ? err : len;
2592 }
2593
2594 static struct md_sysfs_entry md_new_device =
2595 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2596
2597 static ssize_t
2598 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2599 {
2600         char *end;
2601         unsigned long chunk, end_chunk;
2602
2603         if (!mddev->bitmap)
2604                 goto out;
2605         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2606         while (*buf) {
2607                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2608                 if (buf == end) break;
2609                 if (*end == '-') { /* range */
2610                         buf = end + 1;
2611                         end_chunk = simple_strtoul(buf, &end, 0);
2612                         if (buf == end) break;
2613                 }
2614                 if (*end && !isspace(*end)) break;
2615                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2616                 buf = end;
2617                 while (isspace(*buf)) buf++;
2618         }
2619         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2620 out:
2621         return len;
2622 }
2623
2624 static struct md_sysfs_entry md_bitmap =
2625 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2626
2627 static ssize_t
2628 size_show(mddev_t *mddev, char *page)
2629 {
2630         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2631 }
2632
2633 static int update_size(mddev_t *mddev, unsigned long size);
2634
2635 static ssize_t
2636 size_store(mddev_t *mddev, const char *buf, size_t len)
2637 {
2638         /* If array is inactive, we can reduce the component size, but
2639          * not increase it (except from 0).
2640          * If array is active, we can try an on-line resize
2641          */
2642         char *e;
2643         int err = 0;
2644         unsigned long long size = simple_strtoull(buf, &e, 10);
2645         if (!*buf || *buf == '\n' ||
2646             (*e && *e != '\n'))
2647                 return -EINVAL;
2648
2649         if (mddev->pers) {
2650                 err = update_size(mddev, size);
2651                 md_update_sb(mddev, 1);
2652         } else {
2653                 if (mddev->size == 0 ||
2654                     mddev->size > size)
2655                         mddev->size = size;
2656                 else
2657                         err = -ENOSPC;
2658         }
2659         return err ? err : len;
2660 }
2661
2662 static struct md_sysfs_entry md_size =
2663 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2664
2665
2666 /* Metdata version.
2667  * This is either 'none' for arrays with externally managed metadata,
2668  * or N.M for internally known formats
2669  */
2670 static ssize_t
2671 metadata_show(mddev_t *mddev, char *page)
2672 {
2673         if (mddev->persistent)
2674                 return sprintf(page, "%d.%d\n",
2675                                mddev->major_version, mddev->minor_version);
2676         else
2677                 return sprintf(page, "none\n");
2678 }
2679
2680 static ssize_t
2681 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2682 {
2683         int major, minor;
2684         char *e;
2685         if (!list_empty(&mddev->disks))
2686                 return -EBUSY;
2687
2688         if (cmd_match(buf, "none")) {
2689                 mddev->persistent = 0;
2690                 mddev->major_version = 0;
2691                 mddev->minor_version = 90;
2692                 return len;
2693         }
2694         major = simple_strtoul(buf, &e, 10);
2695         if (e==buf || *e != '.')
2696                 return -EINVAL;
2697         buf = e+1;
2698         minor = simple_strtoul(buf, &e, 10);
2699         if (e==buf || (*e && *e != '\n') )
2700                 return -EINVAL;
2701         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2702                 return -ENOENT;
2703         mddev->major_version = major;
2704         mddev->minor_version = minor;
2705         mddev->persistent = 1;
2706         return len;
2707 }
2708
2709 static struct md_sysfs_entry md_metadata =
2710 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2711
2712 static ssize_t
2713 action_show(mddev_t *mddev, char *page)
2714 {
2715         char *type = "idle";
2716         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2717             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2718                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2719                         type = "reshape";
2720                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2721                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2722                                 type = "resync";
2723                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2724                                 type = "check";
2725                         else
2726                                 type = "repair";
2727                 } else
2728                         type = "recover";
2729         }
2730         return sprintf(page, "%s\n", type);
2731 }
2732
2733 static ssize_t
2734 action_store(mddev_t *mddev, const char *page, size_t len)
2735 {
2736         if (!mddev->pers || !mddev->pers->sync_request)
2737                 return -EINVAL;
2738
2739         if (cmd_match(page, "idle")) {
2740                 if (mddev->sync_thread) {
2741                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2742                         md_unregister_thread(mddev->sync_thread);
2743                         mddev->sync_thread = NULL;
2744                         mddev->recovery = 0;
2745                 }
2746         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2747                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2748                 return -EBUSY;
2749         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2750                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2751         else if (cmd_match(page, "reshape")) {
2752                 int err;
2753                 if (mddev->pers->start_reshape == NULL)
2754                         return -EINVAL;
2755                 err = mddev->pers->start_reshape(mddev);
2756                 if (err)
2757                         return err;
2758         } else {
2759                 if (cmd_match(page, "check"))
2760                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2761                 else if (!cmd_match(page, "repair"))
2762                         return -EINVAL;
2763                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2764                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2765         }
2766         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2767         md_wakeup_thread(mddev->thread);
2768         return len;
2769 }
2770
2771 static ssize_t
2772 mismatch_cnt_show(mddev_t *mddev, char *page)
2773 {
2774         return sprintf(page, "%llu\n",
2775                        (unsigned long long) mddev->resync_mismatches);
2776 }
2777
2778 static struct md_sysfs_entry md_scan_mode =
2779 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2780
2781
2782 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2783
2784 static ssize_t
2785 sync_min_show(mddev_t *mddev, char *page)
2786 {
2787         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2788                        mddev->sync_speed_min ? "local": "system");
2789 }
2790
2791 static ssize_t
2792 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2793 {
2794         int min;
2795         char *e;
2796         if (strncmp(buf, "system", 6)==0) {
2797                 mddev->sync_speed_min = 0;
2798                 return len;
2799         }
2800         min = simple_strtoul(buf, &e, 10);
2801         if (buf == e || (*e && *e != '\n') || min <= 0)
2802                 return -EINVAL;
2803         mddev->sync_speed_min = min;
2804         return len;
2805 }
2806
2807 static struct md_sysfs_entry md_sync_min =
2808 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2809
2810 static ssize_t
2811 sync_max_show(mddev_t *mddev, char *page)
2812 {
2813         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2814                        mddev->sync_speed_max ? "local": "system");
2815 }
2816
2817 static ssize_t
2818 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2819 {
2820         int max;
2821         char *e;
2822         if (strncmp(buf, "system", 6)==0) {
2823                 mddev->sync_speed_max = 0;
2824                 return len;
2825         }
2826         max = simple_strtoul(buf, &e, 10);
2827         if (buf == e || (*e && *e != '\n') || max <= 0)
2828                 return -EINVAL;
2829         mddev->sync_speed_max = max;
2830         return len;
2831 }
2832
2833 static struct md_sysfs_entry md_sync_max =
2834 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2835
2836 static ssize_t
2837 degraded_show(mddev_t *mddev, char *page)
2838 {
2839         return sprintf(page, "%d\n", mddev->degraded);
2840 }
2841 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2842
2843 static ssize_t
2844 sync_speed_show(mddev_t *mddev, char *page)
2845 {
2846         unsigned long resync, dt, db;
2847         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2848         dt = ((jiffies - mddev->resync_mark) / HZ);
2849         if (!dt) dt++;
2850         db = resync - (mddev->resync_mark_cnt);
2851         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2852 }
2853
2854 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2855
2856 static ssize_t
2857 sync_completed_show(mddev_t *mddev, char *page)
2858 {
2859         unsigned long max_blocks, resync;
2860
2861         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2862                 max_blocks = mddev->resync_max_sectors;
2863         else
2864                 max_blocks = mddev->size << 1;
2865
2866         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2867         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2868 }
2869
2870 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2871
2872 static ssize_t
2873 suspend_lo_show(mddev_t *mddev, char *page)
2874 {
2875         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2876 }
2877
2878 static ssize_t
2879 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2880 {
2881         char *e;
2882         unsigned long long new = simple_strtoull(buf, &e, 10);
2883
2884         if (mddev->pers->quiesce == NULL)
2885                 return -EINVAL;
2886         if (buf == e || (*e && *e != '\n'))
2887                 return -EINVAL;
2888         if (new >= mddev->suspend_hi ||
2889             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2890                 mddev->suspend_lo = new;
2891                 mddev->pers->quiesce(mddev, 2);
2892                 return len;
2893         } else
2894                 return -EINVAL;
2895 }
2896 static struct md_sysfs_entry md_suspend_lo =
2897 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2898
2899
2900 static ssize_t
2901 suspend_hi_show(mddev_t *mddev, char *page)
2902 {
2903         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2904 }
2905
2906 static ssize_t
2907 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2908 {
2909         char *e;
2910         unsigned long long new = simple_strtoull(buf, &e, 10);
2911
2912         if (mddev->pers->quiesce == NULL)
2913                 return -EINVAL;
2914         if (buf == e || (*e && *e != '\n'))
2915                 return -EINVAL;
2916         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2917             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2918                 mddev->suspend_hi = new;
2919                 mddev->pers->quiesce(mddev, 1);
2920                 mddev->pers->quiesce(mddev, 0);
2921                 return len;
2922         } else
2923                 return -EINVAL;
2924 }
2925 static struct md_sysfs_entry md_suspend_hi =
2926 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2927
2928 static ssize_t
2929 reshape_position_show(mddev_t *mddev, char *page)
2930 {
2931         if (mddev->reshape_position != MaxSector)
2932                 return sprintf(page, "%llu\n",
2933                                (unsigned long long)mddev->reshape_position);
2934         strcpy(page, "none\n");
2935         return 5;
2936 }
2937
2938 static ssize_t
2939 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
2940 {
2941         char *e;
2942         unsigned long long new = simple_strtoull(buf, &e, 10);
2943         if (mddev->pers)
2944                 return -EBUSY;
2945         if (buf == e || (*e && *e != '\n'))
2946                 return -EINVAL;
2947         mddev->reshape_position = new;
2948         mddev->delta_disks = 0;
2949         mddev->new_level = mddev->level;
2950         mddev->new_layout = mddev->layout;
2951         mddev->new_chunk = mddev->chunk_size;
2952         return len;
2953 }
2954
2955 static struct md_sysfs_entry md_reshape_position =
2956 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
2957        reshape_position_store);
2958
2959
2960 static struct attribute *md_default_attrs[] = {
2961         &md_level.attr,
2962         &md_layout.attr,
2963         &md_raid_disks.attr,
2964         &md_chunk_size.attr,
2965         &md_size.attr,
2966         &md_resync_start.attr,
2967         &md_metadata.attr,
2968         &md_new_device.attr,
2969         &md_safe_delay.attr,
2970         &md_array_state.attr,
2971         &md_reshape_position.attr,
2972         NULL,
2973 };
2974
2975 static struct attribute *md_redundancy_attrs[] = {
2976         &md_scan_mode.attr,
2977         &md_mismatches.attr,
2978         &md_sync_min.attr,
2979         &md_sync_max.attr,
2980         &md_sync_speed.attr,
2981         &md_sync_completed.attr,
2982         &md_suspend_lo.attr,
2983         &md_suspend_hi.attr,
2984         &md_bitmap.attr,
2985         &md_degraded.attr,
2986         NULL,
2987 };
2988 static struct attribute_group md_redundancy_group = {
2989         .name = NULL,
2990         .attrs = md_redundancy_attrs,
2991 };
2992
2993
2994 static ssize_t
2995 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2996 {
2997         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2998         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2999         ssize_t rv;
3000
3001         if (!entry->show)
3002                 return -EIO;
3003         rv = mddev_lock(mddev);
3004         if (!rv) {
3005                 rv = entry->show(mddev, page);
3006                 mddev_unlock(mddev);
3007         }
3008         return rv;
3009 }
3010
3011 static ssize_t
3012 md_attr_store(struct kobject *kobj, struct attribute *attr,
3013               const char *page, size_t length)
3014 {
3015         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3016         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3017         ssize_t rv;
3018
3019         if (!entry->store)
3020                 return -EIO;
3021         if (!capable(CAP_SYS_ADMIN))
3022                 return -EACCES;
3023         rv = mddev_lock(mddev);
3024         if (!rv) {
3025                 rv = entry->store(mddev, page, length);
3026                 mddev_unlock(mddev);
3027         }
3028         return rv;
3029 }
3030
3031 static void md_free(struct kobject *ko)
3032 {
3033         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3034         kfree(mddev);
3035 }
3036
3037 static struct sysfs_ops md_sysfs_ops = {
3038         .show   = md_attr_show,
3039         .store  = md_attr_store,
3040 };
3041 static struct kobj_type md_ktype = {
3042         .release        = md_free,
3043         .sysfs_ops      = &md_sysfs_ops,
3044         .default_attrs  = md_default_attrs,
3045 };
3046
3047 int mdp_major = 0;
3048
3049 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3050 {
3051         static DEFINE_MUTEX(disks_mutex);
3052         mddev_t *mddev = mddev_find(dev);
3053         struct gendisk *disk;
3054         int partitioned = (MAJOR(dev) != MD_MAJOR);
3055         int shift = partitioned ? MdpMinorShift : 0;
3056         int unit = MINOR(dev) >> shift;
3057
3058         if (!mddev)
3059                 return NULL;
3060
3061         mutex_lock(&disks_mutex);
3062         if (mddev->gendisk) {
3063                 mutex_unlock(&disks_mutex);
3064                 mddev_put(mddev);
3065                 return NULL;
3066         }
3067         disk = alloc_disk(1 << shift);
3068         if (!disk) {
3069                 mutex_unlock(&disks_mutex);
3070                 mddev_put(mddev);
3071                 return NULL;
3072         }
3073         disk->major = MAJOR(dev);
3074         disk->first_minor = unit << shift;
3075         if (partitioned)
3076                 sprintf(disk->disk_name, "md_d%d", unit);
3077         else
3078                 sprintf(disk->disk_name, "md%d", unit);
3079         disk->fops = &md_fops;
3080         disk->private_data = mddev;
3081         disk->queue = mddev->queue;
3082         add_disk(disk);
3083         mddev->gendisk = disk;
3084         mutex_unlock(&disks_mutex);
3085         mddev->kobj.parent = &disk->kobj;
3086         kobject_set_name(&mddev->kobj, "%s", "md");
3087         mddev->kobj.ktype = &md_ktype;
3088         if (kobject_register(&mddev->kobj))
3089                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3090                        disk->disk_name);
3091         return NULL;
3092 }
3093
3094 static void md_safemode_timeout(unsigned long data)
3095 {
3096         mddev_t *mddev = (mddev_t *) data;
3097
3098         mddev->safemode = 1;
3099         md_wakeup_thread(mddev->thread);
3100 }
3101
3102 static int start_dirty_degraded;
3103
3104 static int do_md_run(mddev_t * mddev)
3105 {
3106         int err;
3107         int chunk_size;
3108         struct list_head *tmp;
3109         mdk_rdev_t *rdev;
3110         struct gendisk *disk;
3111         struct mdk_personality *pers;
3112         char b[BDEVNAME_SIZE];
3113
3114         if (list_empty(&mddev->disks))
3115                 /* cannot run an array with no devices.. */
3116                 return -EINVAL;
3117
3118         if (mddev->pers)
3119                 return -EBUSY;
3120
3121         /*
3122          * Analyze all RAID superblock(s)
3123          */
3124         if (!mddev->raid_disks)
3125                 analyze_sbs(mddev);
3126
3127         chunk_size = mddev->chunk_size;
3128
3129         if (chunk_size) {
3130                 if (chunk_size > MAX_CHUNK_SIZE) {
3131                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3132                                 chunk_size, MAX_CHUNK_SIZE);
3133                         return -EINVAL;
3134                 }
3135                 /*
3136                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3137                  */
3138                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3139                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3140                         return -EINVAL;
3141                 }
3142                 if (chunk_size < PAGE_SIZE) {
3143                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3144                                 chunk_size, PAGE_SIZE);
3145                         return -EINVAL;
3146                 }
3147
3148                 /* devices must have minimum size of one chunk */
3149                 ITERATE_RDEV(mddev,rdev,tmp) {
3150                         if (test_bit(Faulty, &rdev->flags))
3151                                 continue;
3152                         if (rdev->size < chunk_size / 1024) {
3153                                 printk(KERN_WARNING
3154                                         "md: Dev %s smaller than chunk_size:"
3155                                         " %lluk < %dk\n",
3156                                         bdevname(rdev->bdev,b),
3157                                         (unsigned long long)rdev->size,
3158                                         chunk_size / 1024);
3159                                 return -EINVAL;
3160                         }
3161                 }
3162         }
3163
3164 #ifdef CONFIG_KMOD
3165         if (mddev->level != LEVEL_NONE)
3166                 request_module("md-level-%d", mddev->level);
3167         else if (mddev->clevel[0])
3168                 request_module("md-%s", mddev->clevel);
3169 #endif
3170
3171         /*
3172          * Drop all container device buffers, from now on
3173          * the only valid external interface is through the md
3174          * device.
3175          */
3176         ITERATE_RDEV(mddev,rdev,tmp) {
3177                 if (test_bit(Faulty, &rdev->flags))
3178                         continue;
3179                 sync_blockdev(rdev->bdev);
3180                 invalidate_bdev(rdev->bdev);
3181
3182                 /* perform some consistency tests on the device.
3183                  * We don't want the data to overlap the metadata,
3184                  * Internal Bitmap issues has handled elsewhere.
3185                  */
3186                 if (rdev->data_offset < rdev->sb_offset) {
3187                         if (mddev->size &&
3188                             rdev->data_offset + mddev->size*2
3189                             > rdev->sb_offset*2) {
3190                                 printk("md: %s: data overlaps metadata\n",
3191                                        mdname(mddev));
3192                                 return -EINVAL;
3193                         }
3194                 } else {
3195                         if (rdev->sb_offset*2 + rdev->sb_size/512
3196                             > rdev->data_offset) {
3197                                 printk("md: %s: metadata overlaps data\n",
3198                                        mdname(mddev));
3199                                 return -EINVAL;
3200                         }
3201                 }
3202         }
3203
3204         md_probe(mddev->unit, NULL, NULL);
3205         disk = mddev->gendisk;
3206         if (!disk)
3207                 return -ENOMEM;
3208
3209         spin_lock(&pers_lock);
3210         pers = find_pers(mddev->level, mddev->clevel);
3211         if (!pers || !try_module_get(pers->owner)) {
3212                 spin_unlock(&pers_lock);
3213                 if (mddev->level != LEVEL_NONE)
3214                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3215                                mddev->level);
3216                 else
3217                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3218                                mddev->clevel);
3219                 return -EINVAL;
3220         }
3221         mddev->pers = pers;
3222         spin_unlock(&pers_lock);
3223         mddev->level = pers->level;
3224         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3225
3226         if (mddev->reshape_position != MaxSector &&
3227             pers->start_reshape == NULL) {
3228                 /* This personality cannot handle reshaping... */
3229                 mddev->pers = NULL;
3230                 module_put(pers->owner);
3231                 return -EINVAL;
3232         }
3233
3234         if (pers->sync_request) {
3235                 /* Warn if this is a potentially silly
3236                  * configuration.
3237                  */
3238                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3239                 mdk_rdev_t *rdev2;
3240                 struct list_head *tmp2;
3241                 int warned = 0;
3242                 ITERATE_RDEV(mddev, rdev, tmp) {
3243                         ITERATE_RDEV(mddev, rdev2, tmp2) {
3244                                 if (rdev < rdev2 &&
3245                                     rdev->bdev->bd_contains ==
3246                                     rdev2->bdev->bd_contains) {
3247                                         printk(KERN_WARNING
3248                                                "%s: WARNING: %s appears to be"
3249                                                " on the same physical disk as"
3250                                                " %s.\n",
3251                                                mdname(mddev),
3252                                                bdevname(rdev->bdev,b),
3253                                                bdevname(rdev2->bdev,b2));
3254                                         warned = 1;
3255                                 }
3256                         }
3257                 }
3258                 if (warned)
3259                         printk(KERN_WARNING
3260                                "True protection against single-disk"
3261                                " failure might be compromised.\n");
3262         }
3263
3264         mddev->recovery = 0;
3265         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3266         mddev->barriers_work = 1;
3267         mddev->ok_start_degraded = start_dirty_degraded;
3268
3269         if (start_readonly)
3270                 mddev->ro = 2; /* read-only, but switch on first write */
3271
3272         err = mddev->pers->run(mddev);
3273         if (!err && mddev->pers->sync_request) {
3274                 err = bitmap_create(mddev);
3275                 if (err) {
3276                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3277                                mdname(mddev), err);
3278                         mddev->pers->stop(mddev);
3279                 }
3280         }
3281         if (err) {
3282                 printk(KERN_ERR "md: pers->run() failed ...\n");
3283                 module_put(mddev->pers->owner);
3284                 mddev->pers = NULL;
3285                 bitmap_destroy(mddev);
3286                 return err;
3287         }
3288         if (mddev->pers->sync_request) {
3289                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3290                         printk(KERN_WARNING
3291                                "md: cannot register extra attributes for %s\n",
3292                                mdname(mddev));
3293         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3294                 mddev->ro = 0;
3295
3296         atomic_set(&mddev->writes_pending,0);
3297         mddev->safemode = 0;
3298         mddev->safemode_timer.function = md_safemode_timeout;
3299         mddev->safemode_timer.data = (unsigned long) mddev;
3300         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3301         mddev->in_sync = 1;
3302
3303         ITERATE_RDEV(mddev,rdev,tmp)
3304                 if (rdev->raid_disk >= 0) {
3305                         char nm[20];
3306                         sprintf(nm, "rd%d", rdev->raid_disk);
3307                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3308                                 printk("md: cannot register %s for %s\n",
3309                                        nm, mdname(mddev));
3310                 }
3311
3312         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3313
3314         if (mddev->flags)
3315                 md_update_sb(mddev, 0);
3316
3317         set_capacity(disk, mddev->array_size<<1);
3318
3319         /* If we call blk_queue_make_request here, it will
3320          * re-initialise max_sectors etc which may have been
3321          * refined inside -> run.  So just set the bits we need to set.
3322          * Most initialisation happended when we called
3323          * blk_queue_make_request(..., md_fail_request)
3324          * earlier.
3325          */
3326         mddev->queue->queuedata = mddev;
3327         mddev->queue->make_request_fn = mddev->pers->make_request;
3328
3329         /* If there is a partially-recovered drive we need to
3330          * start recovery here.  If we leave it to md_check_recovery,
3331          * it will remove the drives and not do the right thing
3332          */
3333         if (mddev->degraded && !mddev->sync_thread) {
3334                 struct list_head *rtmp;
3335                 int spares = 0;
3336                 ITERATE_RDEV(mddev,rdev,rtmp)
3337                         if (rdev->raid_disk >= 0 &&
3338                             !test_bit(In_sync, &rdev->flags) &&
3339                             !test_bit(Faulty, &rdev->flags))
3340                                 /* complete an interrupted recovery */
3341                                 spares++;
3342                 if (spares && mddev->pers->sync_request) {
3343                         mddev->recovery = 0;
3344                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3345                         mddev->sync_thread = md_register_thread(md_do_sync,
3346                                                                 mddev,
3347                                                                 "%s_resync");
3348                         if (!mddev->sync_thread) {
3349                                 printk(KERN_ERR "%s: could not start resync"
3350                                        " thread...\n",
3351                                        mdname(mddev));
3352                                 /* leave the spares where they are, it shouldn't hurt */
3353                                 mddev->recovery = 0;
3354                         }
3355                 }
3356         }
3357         md_wakeup_thread(mddev->thread);
3358         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3359
3360         mddev->changed = 1;
3361         md_new_event(mddev);
3362         kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3363         return 0;
3364 }
3365
3366 static int restart_array(mddev_t *mddev)
3367 {
3368         struct gendisk *disk = mddev->gendisk;
3369         int err;
3370
3371         /*
3372          * Complain if it has no devices
3373          */
3374         err = -ENXIO;
3375         if (list_empty(&mddev->disks))
3376                 goto out;
3377
3378         if (mddev->pers) {
3379                 err = -EBUSY;
3380                 if (!mddev->ro)
3381                         goto out;
3382
3383                 mddev->safemode = 0;
3384                 mddev->ro = 0;
3385                 set_disk_ro(disk, 0);
3386
3387                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3388                         mdname(mddev));
3389                 /*
3390                  * Kick recovery or resync if necessary
3391                  */
3392                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3393                 md_wakeup_thread(mddev->thread);
3394                 md_wakeup_thread(mddev->sync_thread);
3395                 err = 0;
3396         } else
3397                 err = -EINVAL;
3398
3399 out:
3400         return err;
3401 }
3402
3403 /* similar to deny_write_access, but accounts for our holding a reference
3404  * to the file ourselves */
3405 static int deny_bitmap_write_access(struct file * file)
3406 {
3407         struct inode *inode = file->f_mapping->host;
3408
3409         spin_lock(&inode->i_lock);
3410         if (atomic_read(&inode->i_writecount) > 1) {
3411                 spin_unlock(&inode->i_lock);
3412                 return -ETXTBSY;
3413         }
3414         atomic_set(&inode->i_writecount, -1);
3415         spin_unlock(&inode->i_lock);
3416
3417         return 0;
3418 }
3419
3420 static void restore_bitmap_write_access(struct file *file)
3421 {
3422         struct inode *inode = file->f_mapping->host;
3423
3424         spin_lock(&inode->i_lock);
3425         atomic_set(&inode->i_writecount, 1);
3426         spin_unlock(&inode->i_lock);
3427 }
3428
3429 /* mode:
3430  *   0 - completely stop and dis-assemble array
3431  *   1 - switch to readonly
3432  *   2 - stop but do not disassemble array
3433  */
3434 static int do_md_stop(mddev_t * mddev, int mode)
3435 {
3436         int err = 0;
3437         struct gendisk *disk = mddev->gendisk;
3438
3439         if (mddev->pers) {
3440                 if (atomic_read(&mddev->active)>2) {
3441                         printk("md: %s still in use.\n",mdname(mddev));
3442                         return -EBUSY;
3443                 }
3444
3445                 if (mddev->sync_thread) {
3446                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3447                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3448                         md_unregister_thread(mddev->sync_thread);
3449                         mddev->sync_thread = NULL;
3450                 }
3451
3452                 del_timer_sync(&mddev->safemode_timer);
3453
3454                 invalidate_partition(disk, 0);
3455
3456                 switch(mode) {
3457                 case 1: /* readonly */
3458                         err  = -ENXIO;
3459                         if (mddev->ro==1)
3460                                 goto out;
3461                         mddev->ro = 1;
3462                         break;
3463                 case 0: /* disassemble */
3464                 case 2: /* stop */
3465                         bitmap_flush(mddev);
3466                         md_super_wait(mddev);
3467                         if (mddev->ro)
3468                                 set_disk_ro(disk, 0);
3469                         blk_queue_make_request(mddev->queue, md_fail_request);
3470                         mddev->pers->stop(mddev);
3471                         mddev->queue->merge_bvec_fn = NULL;
3472                         mddev->queue->unplug_fn = NULL;
3473                         mddev->queue->backing_dev_info.congested_fn = NULL;
3474                         if (mddev->pers->sync_request)
3475                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3476
3477                         module_put(mddev->pers->owner);
3478                         mddev->pers = NULL;
3479
3480                         set_capacity(disk, 0);
3481                         mddev->changed = 1;
3482
3483                         if (mddev->ro)
3484                                 mddev->ro = 0;
3485                 }
3486                 if (!mddev->in_sync || mddev->flags) {
3487                         /* mark array as shutdown cleanly */
3488                         mddev->in_sync = 1;
3489                         md_update_sb(mddev, 1);
3490                 }
3491                 if (mode == 1)
3492                         set_disk_ro(disk, 1);
3493                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3494         }
3495
3496         /*
3497          * Free resources if final stop
3498          */
3499         if (mode == 0) {
3500                 mdk_rdev_t *rdev;
3501                 struct list_head *tmp;
3502
3503                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3504
3505                 bitmap_destroy(mddev);
3506                 if (mddev->bitmap_file) {
3507                         restore_bitmap_write_access(mddev->bitmap_file);
3508                         fput(mddev->bitmap_file);
3509                         mddev->bitmap_file = NULL;
3510                 }
3511                 mddev->bitmap_offset = 0;
3512
3513                 ITERATE_RDEV(mddev,rdev,tmp)
3514                         if (rdev->raid_disk >= 0) {
3515                                 char nm[20];
3516                                 sprintf(nm, "rd%d", rdev->raid_disk);
3517                                 sysfs_remove_link(&mddev->kobj, nm);
3518                         }
3519
3520                 /* make sure all delayed_delete calls have finished */
3521                 flush_scheduled_work();
3522
3523                 export_array(mddev);
3524
3525                 mddev->array_size = 0;
3526                 mddev->size = 0;
3527                 mddev->raid_disks = 0;
3528                 mddev->recovery_cp = 0;
3529                 mddev->reshape_position = MaxSector;
3530
3531         } else if (mddev->pers)
3532                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3533                         mdname(mddev));
3534         err = 0;
3535         md_new_event(mddev);
3536 out:
3537         return err;
3538 }
3539
3540 #ifndef MODULE
3541 static void autorun_array(mddev_t *mddev)
3542 {
3543         mdk_rdev_t *rdev;
3544         struct list_head *tmp;
3545         int err;
3546
3547         if (list_empty(&mddev->disks))
3548                 return;
3549
3550         printk(KERN_INFO "md: running: ");
3551
3552         ITERATE_RDEV(mddev,rdev,tmp) {
3553                 char b[BDEVNAME_SIZE];
3554                 printk("<%s>", bdevname(rdev->bdev,b));
3555         }
3556         printk("\n");
3557
3558         err = do_md_run (mddev);
3559         if (err) {
3560                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3561                 do_md_stop (mddev, 0);
3562         }
3563 }
3564
3565 /*
3566  * lets try to run arrays based on all disks that have arrived
3567  * until now. (those are in pending_raid_disks)
3568  *
3569  * the method: pick the first pending disk, collect all disks with
3570  * the same UUID, remove all from the pending list and put them into
3571  * the 'same_array' list. Then order this list based on superblock
3572  * update time (freshest comes first), kick out 'old' disks and
3573  * compare superblocks. If everything's fine then run it.
3574  *
3575  * If "unit" is allocated, then bump its reference count
3576  */
3577 static void autorun_devices(int part)
3578 {
3579         struct list_head *tmp;
3580         mdk_rdev_t *rdev0, *rdev;
3581         mddev_t *mddev;
3582         char b[BDEVNAME_SIZE];
3583
3584         printk(KERN_INFO "md: autorun ...\n");
3585         while (!list_empty(&pending_raid_disks)) {
3586                 int unit;
3587                 dev_t dev;
3588                 LIST_HEAD(candidates);
3589                 rdev0 = list_entry(pending_raid_disks.next,
3590                                          mdk_rdev_t, same_set);
3591
3592                 printk(KERN_INFO "md: considering %s ...\n",
3593                         bdevname(rdev0->bdev,b));
3594                 INIT_LIST_HEAD(&candidates);
3595                 ITERATE_RDEV_PENDING(rdev,tmp)
3596                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3597                                 printk(KERN_INFO "md:  adding %s ...\n",
3598                                         bdevname(rdev->bdev,b));
3599                                 list_move(&rdev->same_set, &candidates);
3600                         }
3601                 /*
3602                  * now we have a set of devices, with all of them having
3603                  * mostly sane superblocks. It's time to allocate the
3604                  * mddev.
3605                  */
3606                 if (part) {
3607                         dev = MKDEV(mdp_major,
3608                                     rdev0->preferred_minor << MdpMinorShift);
3609                         unit = MINOR(dev) >> MdpMinorShift;
3610                 } else {
3611                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3612                         unit = MINOR(dev);
3613                 }
3614                 if (rdev0->preferred_minor != unit) {
3615                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3616                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3617                         break;
3618                 }
3619
3620                 md_probe(dev, NULL, NULL);
3621                 mddev = mddev_find(dev);
3622                 if (!mddev) {
3623                         printk(KERN_ERR
3624                                 "md: cannot allocate memory for md drive.\n");
3625                         break;
3626                 }
3627                 if (mddev_lock(mddev))
3628                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3629                                mdname(mddev));
3630                 else if (mddev->raid_disks || mddev->major_version
3631                          || !list_empty(&mddev->disks)) {
3632                         printk(KERN_WARNING
3633                                 "md: %s already running, cannot run %s\n",
3634                                 mdname(mddev), bdevname(rdev0->bdev,b));
3635                         mddev_unlock(mddev);
3636                 } else {
3637                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3638                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3639                                 list_del_init(&rdev->same_set);
3640                                 if (bind_rdev_to_array(rdev, mddev))
3641                                         export_rdev(rdev);
3642                         }
3643                         autorun_array(mddev);
3644                         mddev_unlock(mddev);
3645                 }
3646                 /* on success, candidates will be empty, on error
3647                  * it won't...
3648                  */
3649                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3650                         export_rdev(rdev);
3651                 mddev_put(mddev);
3652         }
3653         printk(KERN_INFO "md: ... autorun DONE.\n");
3654 }
3655 #endif /* !MODULE */
3656
3657 static int get_version(void __user * arg)
3658 {
3659         mdu_version_t ver;
3660
3661         ver.major = MD_MAJOR_VERSION;
3662         ver.minor = MD_MINOR_VERSION;
3663         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3664
3665         if (copy_to_user(arg, &ver, sizeof(ver)))
3666                 return -EFAULT;
3667
3668         return 0;
3669 }
3670
3671 static int get_array_info(mddev_t * mddev, void __user * arg)
3672 {
3673         mdu_array_info_t info;
3674         int nr,working,active,failed,spare;
3675         mdk_rdev_t *rdev;
3676         struct list_head *tmp;
3677
3678         nr=working=active=failed=spare=0;
3679         ITERATE_RDEV(mddev,rdev,tmp) {
3680                 nr++;
3681                 if (test_bit(Faulty, &rdev->flags))
3682                         failed++;
3683                 else {
3684                         working++;
3685                         if (test_bit(In_sync, &rdev->flags))
3686                                 active++;
3687                         else
3688                                 spare++;
3689                 }
3690         }
3691
3692         info.major_version = mddev->major_version;
3693         info.minor_version = mddev->minor_version;
3694         info.patch_version = MD_PATCHLEVEL_VERSION;
3695         info.ctime         = mddev->ctime;
3696         info.level         = mddev->level;
3697         info.size          = mddev->size;
3698         if (info.size != mddev->size) /* overflow */
3699                 info.size = -1;
3700         info.nr_disks      = nr;
3701         info.raid_disks    = mddev->raid_disks;
3702         info.md_minor      = mddev->md_minor;
3703         info.not_persistent= !mddev->persistent;
3704
3705         info.utime         = mddev->utime;
3706         info.state         = 0;
3707         if (mddev->in_sync)
3708                 info.state = (1<<MD_SB_CLEAN);
3709         if (mddev->bitmap && mddev->bitmap_offset)
3710                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3711         info.active_disks  = active;
3712         info.working_disks = working;
3713         info.failed_disks  = failed;
3714         info.spare_disks   = spare;
3715
3716         info.layout        = mddev->layout;
3717         info.chunk_size    = mddev->chunk_size;
3718
3719         if (copy_to_user(arg, &info, sizeof(info)))
3720                 return -EFAULT;
3721
3722         return 0;
3723 }
3724
3725 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3726 {
3727         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3728         char *ptr, *buf = NULL;
3729         int err = -ENOMEM;
3730
3731         md_allow_write(mddev);
3732
3733         file = kmalloc(sizeof(*file), GFP_KERNEL);
3734         if (!file)
3735                 goto out;
3736
3737         /* bitmap disabled, zero the first byte and copy out */
3738         if (!mddev->bitmap || !mddev->bitmap->file) {
3739                 file->pathname[0] = '\0';
3740                 goto copy_out;
3741         }
3742
3743         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3744         if (!buf)
3745                 goto out;
3746
3747         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3748         if (!ptr)
3749                 goto out;
3750
3751         strcpy(file->pathname, ptr);
3752
3753 copy_out:
3754         err = 0;
3755         if (copy_to_user(arg, file, sizeof(*file)))
3756                 err = -EFAULT;
3757 out:
3758         kfree(buf);
3759         kfree(file);
3760         return err;
3761 }
3762
3763 static int get_disk_info(mddev_t * mddev, void __user * arg)
3764 {
3765         mdu_disk_info_t info;
3766         unsigned int nr;
3767         mdk_rdev_t *rdev;
3768
3769         if (copy_from_user(&info, arg, sizeof(info)))
3770                 return -EFAULT;
3771
3772         nr = info.number;
3773
3774         rdev = find_rdev_nr(mddev, nr);
3775         if (rdev) {
3776                 info.major = MAJOR(rdev->bdev->bd_dev);
3777                 info.minor = MINOR(rdev->bdev->bd_dev);
3778                 info.raid_disk = rdev->raid_disk;
3779                 info.state = 0;
3780                 if (test_bit(Faulty, &rdev->flags))
3781                         info.state |= (1<<MD_DISK_FAULTY);
3782                 else if (test_bit(In_sync, &rdev->flags)) {
3783                         info.state |= (1<<MD_DISK_ACTIVE);
3784                         info.state |= (1<<MD_DISK_SYNC);
3785                 }
3786                 if (test_bit(WriteMostly, &rdev->flags))
3787                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3788         } else {
3789                 info.major = info.minor = 0;
3790                 info.raid_disk = -1;
3791                 info.state = (1<<MD_DISK_REMOVED);
3792         }
3793
3794         if (copy_to_user(arg, &info, sizeof(info)))
3795                 return -EFAULT;
3796
3797         return 0;
3798 }
3799
3800 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3801 {
3802         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3803         mdk_rdev_t *rdev;
3804         dev_t dev = MKDEV(info->major,info->minor);
3805
3806         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3807                 return -EOVERFLOW;
3808
3809         if (!mddev->raid_disks) {
3810                 int err;
3811                 /* expecting a device which has a superblock */
3812                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3813                 if (IS_ERR(rdev)) {
3814                         printk(KERN_WARNING
3815                                 "md: md_import_device returned %ld\n",
3816                                 PTR_ERR(rdev));
3817                         return PTR_ERR(rdev);
3818                 }
3819                 if (!list_empty(&mddev->disks)) {
3820                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3821                                                         mdk_rdev_t, same_set);
3822                         int err = super_types[mddev->major_version]
3823                                 .load_super(rdev, rdev0, mddev->minor_version);
3824                         if (err < 0) {
3825                                 printk(KERN_WARNING
3826                                         "md: %s has different UUID to %s\n",
3827                                         bdevname(rdev->bdev,b),
3828                                         bdevname(rdev0->bdev,b2));
3829                                 export_rdev(rdev);
3830                                 return -EINVAL;
3831                         }
3832                 }
3833                 err = bind_rdev_to_array(rdev, mddev);
3834                 if (err)
3835                         export_rdev(rdev);
3836                 return err;
3837         }
3838
3839         /*
3840          * add_new_disk can be used once the array is assembled
3841          * to add "hot spares".  They must already have a superblock
3842          * written
3843          */
3844         if (mddev->pers) {
3845                 int err;
3846                 if (!mddev->pers->hot_add_disk) {
3847                         printk(KERN_WARNING
3848                                 "%s: personality does not support diskops!\n",
3849                                mdname(mddev));
3850                         return -EINVAL;
3851                 }
3852                 if (mddev->persistent)
3853                         rdev = md_import_device(dev, mddev->major_version,
3854                                                 mddev->minor_version);
3855                 else
3856                         rdev = md_import_device(dev, -1, -1);
3857                 if (IS_ERR(rdev)) {
3858                         printk(KERN_WARNING
3859                                 "md: md_import_device returned %ld\n",
3860                                 PTR_ERR(rdev));
3861                         return PTR_ERR(rdev);
3862                 }
3863                 /* set save_raid_disk if appropriate */
3864                 if (!mddev->persistent) {
3865                         if (info->state & (1<<MD_DISK_SYNC)  &&
3866                             info->raid_disk < mddev->raid_disks)
3867                                 rdev->raid_disk = info->raid_disk;
3868                         else
3869                                 rdev->raid_disk = -1;
3870                 } else
3871                         super_types[mddev->major_version].
3872                                 validate_super(mddev, rdev);
3873                 rdev->saved_raid_disk = rdev->raid_disk;
3874
3875                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3876                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3877                         set_bit(WriteMostly, &rdev->flags);
3878
3879                 rdev->raid_disk = -1;
3880                 err = bind_rdev_to_array(rdev, mddev);
3881                 if (!err && !mddev->pers->hot_remove_disk) {
3882                         /* If there is hot_add_disk but no hot_remove_disk
3883                          * then added disks for geometry changes,
3884                          * and should be added immediately.
3885                          */
3886                         super_types[mddev->major_version].
3887                                 validate_super(mddev, rdev);
3888                         err = mddev->pers->hot_add_disk(mddev, rdev);
3889                         if (err)
3890                                 unbind_rdev_from_array(rdev);
3891                 }
3892                 if (err)
3893                         export_rdev(rdev);
3894
3895                 md_update_sb(mddev, 1);
3896                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3897                 md_wakeup_thread(mddev->thread);
3898                 return err;
3899         }
3900
3901         /* otherwise, add_new_disk is only allowed
3902          * for major_version==0 superblocks
3903          */
3904         if (mddev->major_version != 0) {
3905                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3906                        mdname(mddev));
3907                 return -EINVAL;
3908         }
3909
3910         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3911                 int err;
3912                 rdev = md_import_device (dev, -1, 0);
3913                 if (IS_ERR(rdev)) {
3914                         printk(KERN_WARNING
3915                                 "md: error, md_import_device() returned %ld\n",
3916                                 PTR_ERR(rdev));
3917                         return PTR_ERR(rdev);
3918                 }
3919                 rdev->desc_nr = info->number;
3920                 if (info->raid_disk < mddev->raid_disks)
3921                         rdev->raid_disk = info->raid_disk;
3922                 else
3923                         rdev->raid_disk = -1;
3924
3925                 rdev->flags = 0;
3926
3927                 if (rdev->raid_disk < mddev->raid_disks)
3928                         if (info->state & (1<<MD_DISK_SYNC))
3929                                 set_bit(In_sync, &rdev->flags);
3930
3931                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3932                         set_bit(WriteMostly, &rdev->flags);
3933
3934                 if (!mddev->persistent) {
3935                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3936                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3937                 } else
3938                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3939                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3940
3941                 err = bind_rdev_to_array(rdev, mddev);
3942                 if (err) {
3943                         export_rdev(rdev);
3944                         return err;
3945                 }
3946         }
3947
3948         return 0;
3949 }
3950
3951 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3952 {
3953         char b[BDEVNAME_SIZE];
3954         mdk_rdev_t *rdev;
3955
3956         if (!mddev->pers)
3957                 return -ENODEV;
3958
3959         rdev = find_rdev(mddev, dev);
3960         if (!rdev)
3961                 return -ENXIO;
3962
3963         if (rdev->raid_disk >= 0)
3964                 goto busy;
3965
3966         kick_rdev_from_array(rdev);
3967         md_update_sb(mddev, 1);
3968         md_new_event(mddev);
3969
3970         return 0;
3971 busy:
3972         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3973                 bdevname(rdev->bdev,b), mdname(mddev));
3974         return -EBUSY;
3975 }
3976
3977 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3978 {
3979         char b[BDEVNAME_SIZE];
3980         int err;
3981         unsigned int size;
3982         mdk_rdev_t *rdev;
3983
3984         if (!mddev->pers)
3985                 return -ENODEV;
3986
3987         if (mddev->major_version != 0) {
3988                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3989                         " version-0 superblocks.\n",
3990                         mdname(mddev));
3991                 return -EINVAL;
3992         }
3993         if (!mddev->pers->hot_add_disk) {
3994                 printk(KERN_WARNING
3995                         "%s: personality does not support diskops!\n",
3996                         mdname(mddev));
3997                 return -EINVAL;
3998         }
3999
4000         rdev = md_import_device (dev, -1, 0);
4001         if (IS_ERR(rdev)) {
4002                 printk(KERN_WARNING
4003                         "md: error, md_import_device() returned %ld\n",
4004                         PTR_ERR(rdev));
4005                 return -EINVAL;
4006         }
4007
4008         if (mddev->persistent)
4009                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4010         else
4011                 rdev->sb_offset =
4012                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4013
4014         size = calc_dev_size(rdev, mddev->chunk_size);
4015         rdev->size = size;
4016
4017         if (test_bit(Faulty, &rdev->flags)) {
4018                 printk(KERN_WARNING
4019                         "md: can not hot-add faulty %s disk to %s!\n",
4020                         bdevname(rdev->bdev,b), mdname(mddev));
4021                 err = -EINVAL;
4022                 goto abort_export;
4023         }
4024         clear_bit(In_sync, &rdev->flags);
4025         rdev->desc_nr = -1;
4026         rdev->saved_raid_disk = -1;
4027         err = bind_rdev_to_array(rdev, mddev);
4028         if (err)
4029                 goto abort_export;
4030
4031         /*
4032          * The rest should better be atomic, we can have disk failures
4033          * noticed in interrupt contexts ...
4034          */
4035
4036         if (rdev->desc_nr == mddev->max_disks) {
4037                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4038                         mdname(mddev));
4039                 err = -EBUSY;
4040                 goto abort_unbind_export;
4041         }
4042
4043         rdev->raid_disk = -1;
4044
4045         md_update_sb(mddev, 1);
4046
4047         /*
4048          * Kick recovery, maybe this spare has to be added to the
4049          * array immediately.
4050          */
4051         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4052         md_wakeup_thread(mddev->thread);
4053         md_new_event(mddev);
4054         return 0;
4055
4056 abort_unbind_export:
4057         unbind_rdev_from_array(rdev);
4058
4059 abort_export:
4060         export_rdev(rdev);
4061         return err;
4062 }
4063
4064 static int set_bitmap_file(mddev_t *mddev, int fd)
4065 {
4066         int err;
4067
4068         if (mddev->pers) {
4069                 if (!mddev->pers->quiesce)
4070                         return -EBUSY;
4071                 if (mddev->recovery || mddev->sync_thread)
4072                         return -EBUSY;
4073                 /* we should be able to change the bitmap.. */
4074         }
4075
4076
4077         if (fd >= 0) {
4078                 if (mddev->bitmap)
4079                         return -EEXIST; /* cannot add when bitmap is present */
4080                 mddev->bitmap_file = fget(fd);
4081
4082                 if (mddev->bitmap_file == NULL) {
4083                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4084                                mdname(mddev));
4085                         return -EBADF;
4086                 }
4087
4088                 err = deny_bitmap_write_access(mddev->bitmap_file);
4089                 if (err) {
4090                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4091                                mdname(mddev));
4092                         fput(mddev->bitmap_file);
4093                         mddev->bitmap_file = NULL;
4094                         return err;
4095                 }
4096                 mddev->bitmap_offset = 0; /* file overrides offset */
4097         } else if (mddev->bitmap == NULL)
4098                 return -ENOENT; /* cannot remove what isn't there */
4099         err = 0;
4100         if (mddev->pers) {
4101                 mddev->pers->quiesce(mddev, 1);
4102                 if (fd >= 0)
4103                         err = bitmap_create(mddev);
4104                 if (fd < 0 || err) {
4105                         bitmap_destroy(mddev);
4106                         fd = -1; /* make sure to put the file */
4107                 }
4108                 mddev->pers->quiesce(mddev, 0);
4109         }
4110         if (fd < 0) {
4111                 if (mddev->bitmap_file) {
4112                         restore_bitmap_write_access(mddev->bitmap_file);
4113                         fput(mddev->bitmap_file);
4114                 }
4115                 mddev->bitmap_file = NULL;
4116         }
4117
4118         return err;
4119 }
4120
4121 /*
4122  * set_array_info is used two different ways
4123  * The original usage is when creating a new array.
4124  * In this usage, raid_disks is > 0 and it together with
4125  *  level, size, not_persistent,layout,chunksize determine the
4126  *  shape of the array.
4127  *  This will always create an array with a type-0.90.0 superblock.
4128  * The newer usage is when assembling an array.
4129  *  In this case raid_disks will be 0, and the major_version field is
4130  *  use to determine which style super-blocks are to be found on the devices.
4131  *  The minor and patch _version numbers are also kept incase the
4132  *  super_block handler wishes to interpret them.
4133  */
4134 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4135 {
4136
4137         if (info->raid_disks == 0) {
4138                 /* just setting version number for superblock loading */
4139                 if (info->major_version < 0 ||
4140                     info->major_version >= ARRAY_SIZE(super_types) ||
4141                     super_types[info->major_version].name == NULL) {
4142                         /* maybe try to auto-load a module? */
4143                         printk(KERN_INFO
4144                                 "md: superblock version %d not known\n",
4145                                 info->major_version);
4146                         return -EINVAL;
4147                 }
4148                 mddev->major_version = info->major_version;
4149                 mddev->minor_version = info->minor_version;
4150                 mddev->patch_version = info->patch_version;
4151                 mddev->persistent = !info->not_persistent;
4152                 return 0;
4153         }
4154         mddev->major_version = MD_MAJOR_VERSION;
4155         mddev->minor_version = MD_MINOR_VERSION;
4156         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4157         mddev->ctime         = get_seconds();
4158
4159         mddev->level         = info->level;
4160         mddev->clevel[0]     = 0;
4161         mddev->size          = info->size;
4162         mddev->raid_disks    = info->raid_disks;
4163         /* don't set md_minor, it is determined by which /dev/md* was
4164          * openned
4165          */
4166         if (info->state & (1<<MD_SB_CLEAN))
4167                 mddev->recovery_cp = MaxSector;
4168         else
4169                 mddev->recovery_cp = 0;
4170         mddev->persistent    = ! info->not_persistent;
4171
4172         mddev->layout        = info->layout;
4173         mddev->chunk_size    = info->chunk_size;
4174
4175         mddev->max_disks     = MD_SB_DISKS;
4176
4177         mddev->flags         = 0;
4178         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4179
4180         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4181         mddev->bitmap_offset = 0;
4182
4183         mddev->reshape_position = MaxSector;
4184
4185         /*
4186          * Generate a 128 bit UUID
4187          */
4188         get_random_bytes(mddev->uuid, 16);
4189
4190         mddev->new_level = mddev->level;
4191         mddev->new_chunk = mddev->chunk_size;
4192         mddev->new_layout = mddev->layout;
4193         mddev->delta_disks = 0;
4194
4195         return 0;
4196 }
4197
4198 static int update_size(mddev_t *mddev, unsigned long size)
4199 {
4200         mdk_rdev_t * rdev;
4201         int rv;
4202         struct list_head *tmp;
4203         int fit = (size == 0);
4204
4205         if (mddev->pers->resize == NULL)
4206                 return -EINVAL;
4207         /* The "size" is the amount of each device that is used.
4208          * This can only make sense for arrays with redundancy.
4209          * linear and raid0 always use whatever space is available
4210          * We can only consider changing the size if no resync
4211          * or reconstruction is happening, and if the new size
4212          * is acceptable. It must fit before the sb_offset or,
4213          * if that is <data_offset, it must fit before the
4214          * size of each device.
4215          * If size is zero, we find the largest size that fits.
4216          */
4217         if (mddev->sync_thread)
4218                 return -EBUSY;
4219         ITERATE_RDEV(mddev,rdev,tmp) {
4220                 sector_t avail;
4221                 avail = rdev->size * 2;
4222
4223                 if (fit && (size == 0 || size > avail/2))
4224                         size = avail/2;
4225                 if (avail < ((sector_t)size << 1))
4226                         return -ENOSPC;
4227         }
4228         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4229         if (!rv) {
4230                 struct block_device *bdev;
4231
4232                 bdev = bdget_disk(mddev->gendisk, 0);
4233                 if (bdev) {
4234                         mutex_lock(&bdev->bd_inode->i_mutex);
4235                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4236                         mutex_unlock(&bdev->bd_inode->i_mutex);
4237                         bdput(bdev);
4238                 }
4239         }
4240         return rv;
4241 }
4242
4243 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4244 {
4245         int rv;
4246         /* change the number of raid disks */
4247         if (mddev->pers->check_reshape == NULL)
4248                 return -EINVAL;
4249         if (raid_disks <= 0 ||
4250             raid_disks >= mddev->max_disks)
4251                 return -EINVAL;
4252         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4253                 return -EBUSY;
4254         mddev->delta_disks = raid_disks - mddev->raid_disks;
4255
4256         rv = mddev->pers->check_reshape(mddev);
4257         return rv;
4258 }
4259
4260
4261 /*
4262  * update_array_info is used to change the configuration of an
4263  * on-line array.
4264  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4265  * fields in the info are checked against the array.
4266  * Any differences that cannot be handled will cause an error.
4267  * Normally, only one change can be managed at a time.
4268  */
4269 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4270 {
4271         int rv = 0;
4272         int cnt = 0;
4273         int state = 0;
4274
4275         /* calculate expected state,ignoring low bits */
4276         if (mddev->bitmap && mddev->bitmap_offset)
4277                 state |= (1 << MD_SB_BITMAP_PRESENT);
4278
4279         if (mddev->major_version != info->major_version ||
4280             mddev->minor_version != info->minor_version ||
4281 /*          mddev->patch_version != info->patch_version || */
4282             mddev->ctime         != info->ctime         ||
4283             mddev->level         != info->level         ||
4284 /*          mddev->layout        != info->layout        || */
4285             !mddev->persistent   != info->not_persistent||
4286             mddev->chunk_size    != info->chunk_size    ||
4287             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4288             ((state^info->state) & 0xfffffe00)
4289                 )
4290                 return -EINVAL;
4291         /* Check there is only one change */
4292         if (info->size >= 0 && mddev->size != info->size) cnt++;
4293         if (mddev->raid_disks != info->raid_disks) cnt++;
4294         if (mddev->layout != info->layout) cnt++;
4295         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4296         if (cnt == 0) return 0;
4297         if (cnt > 1) return -EINVAL;
4298
4299         if (mddev->layout != info->layout) {
4300                 /* Change layout
4301                  * we don't need to do anything at the md level, the
4302                  * personality will take care of it all.
4303                  */
4304                 if (mddev->pers->reconfig == NULL)
4305                         return -EINVAL;
4306                 else
4307                         return mddev->pers->reconfig(mddev, info->layout, -1);
4308         }
4309         if (info->size >= 0 && mddev->size != info->size)
4310                 rv = update_size(mddev, info->size);
4311
4312         if (mddev->raid_disks    != info->raid_disks)
4313                 rv = update_raid_disks(mddev, info->raid_disks);
4314
4315         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4316                 if (mddev->pers->quiesce == NULL)
4317                         return -EINVAL;
4318                 if (mddev->recovery || mddev->sync_thread)
4319                         return -EBUSY;
4320                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4321                         /* add the bitmap */
4322                         if (mddev->bitmap)
4323                                 return -EEXIST;
4324                         if (mddev->default_bitmap_offset == 0)
4325                                 return -EINVAL;
4326                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4327                         mddev->pers->quiesce(mddev, 1);
4328                         rv = bitmap_create(mddev);
4329                         if (rv)
4330                                 bitmap_destroy(mddev);
4331                         mddev->pers->quiesce(mddev, 0);
4332                 } else {
4333                         /* remove the bitmap */
4334                         if (!mddev->bitmap)
4335                                 return -ENOENT;
4336                         if (mddev->bitmap->file)
4337                                 return -EINVAL;
4338                         mddev->pers->quiesce(mddev, 1);
4339                         bitmap_destroy(mddev);
4340                         mddev->pers->quiesce(mddev, 0);
4341                         mddev->bitmap_offset = 0;
4342                 }
4343         }
4344         md_update_sb(mddev, 1);
4345         return rv;
4346 }
4347
4348 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4349 {
4350         mdk_rdev_t *rdev;
4351
4352         if (mddev->pers == NULL)
4353                 return -ENODEV;
4354
4355         rdev = find_rdev(mddev, dev);
4356         if (!rdev)
4357                 return -ENODEV;
4358
4359         md_error(mddev, rdev);
4360         return 0;
4361 }
4362
4363 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4364 {
4365         mddev_t *mddev = bdev->bd_disk->private_data;
4366
4367         geo->heads = 2;
4368         geo->sectors = 4;
4369         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4370         return 0;
4371 }
4372
4373 static int md_ioctl(struct inode *inode, struct file *file,
4374                         unsigned int cmd, unsigned long arg)
4375 {
4376         int err = 0;
4377         void __user *argp = (void __user *)arg;
4378         mddev_t *mddev = NULL;
4379
4380         if (!capable(CAP_SYS_ADMIN))
4381                 return -EACCES;
4382
4383         /*
4384          * Commands dealing with the RAID driver but not any
4385          * particular array:
4386          */
4387         switch (cmd)
4388         {
4389                 case RAID_VERSION:
4390                         err = get_version(argp);
4391                         goto done;
4392
4393                 case PRINT_RAID_DEBUG:
4394                         err = 0;
4395                         md_print_devices();
4396                         goto done;
4397
4398 #ifndef MODULE
4399                 case RAID_AUTORUN:
4400                         err = 0;
4401                         autostart_arrays(arg);
4402                         goto done;
4403 #endif
4404                 default:;
4405         }
4406
4407         /*
4408          * Commands creating/starting a new array:
4409          */
4410
4411         mddev = inode->i_bdev->bd_disk->private_data;
4412
4413         if (!mddev) {
4414                 BUG();
4415                 goto abort;
4416         }
4417
4418         err = mddev_lock(mddev);
4419         if (err) {
4420                 printk(KERN_INFO
4421                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4422                         err, cmd);
4423                 goto abort;
4424         }
4425
4426         switch (cmd)
4427         {
4428                 case SET_ARRAY_INFO:
4429                         {
4430                                 mdu_array_info_t info;
4431                                 if (!arg)
4432                                         memset(&info, 0, sizeof(info));
4433                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4434                                         err = -EFAULT;
4435                                         goto abort_unlock;
4436                                 }
4437                                 if (mddev->pers) {
4438                                         err = update_array_info(mddev, &info);
4439                                         if (err) {
4440                                                 printk(KERN_WARNING "md: couldn't update"
4441                                                        " array info. %d\n", err);
4442                                                 goto abort_unlock;
4443                                         }
4444                                         goto done_unlock;
4445                                 }
4446                                 if (!list_empty(&mddev->disks)) {
4447                                         printk(KERN_WARNING
4448                                                "md: array %s already has disks!\n",
4449                                                mdname(mddev));
4450                                         err = -EBUSY;
4451                                         goto abort_unlock;
4452                                 }
4453                                 if (mddev->raid_disks) {
4454                                         printk(KERN_WARNING
4455                                                "md: array %s already initialised!\n",
4456                                                mdname(mddev));
4457                                         err = -EBUSY;
4458                                         goto abort_unlock;
4459                                 }
4460                                 err = set_array_info(mddev, &info);
4461                                 if (err) {
4462                                         printk(KERN_WARNING "md: couldn't set"
4463                                                " array info. %d\n", err);
4464                                         goto abort_unlock;
4465                                 }
4466                         }
4467                         goto done_unlock;
4468
4469                 default:;
4470         }
4471
4472         /*
4473          * Commands querying/configuring an existing array:
4474          */
4475         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4476          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4477         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4478                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4479                         && cmd != GET_BITMAP_FILE) {
4480                 err = -ENODEV;
4481                 goto abort_unlock;
4482         }
4483
4484         /*
4485          * Commands even a read-only array can execute:
4486          */
4487         switch (cmd)
4488         {
4489                 case GET_ARRAY_INFO:
4490                         err = get_array_info(mddev, argp);
4491                         goto done_unlock;
4492
4493                 case GET_BITMAP_FILE:
4494                         err = get_bitmap_file(mddev, argp);
4495                         goto done_unlock;
4496
4497                 case GET_DISK_INFO:
4498                         err = get_disk_info(mddev, argp);
4499                         goto done_unlock;
4500
4501                 case RESTART_ARRAY_RW:
4502                         err = restart_array(mddev);
4503                         goto done_unlock;
4504
4505                 case STOP_ARRAY:
4506                         err = do_md_stop (mddev, 0);
4507                         goto done_unlock;
4508
4509                 case STOP_ARRAY_RO:
4510                         err = do_md_stop (mddev, 1);
4511                         goto done_unlock;
4512
4513         /*
4514          * We have a problem here : there is no easy way to give a CHS
4515          * virtual geometry. We currently pretend that we have a 2 heads
4516          * 4 sectors (with a BIG number of cylinders...). This drives
4517          * dosfs just mad... ;-)
4518          */
4519         }
4520
4521         /*
4522          * The remaining ioctls are changing the state of the
4523          * superblock, so we do not allow them on read-only arrays.
4524          * However non-MD ioctls (e.g. get-size) will still come through
4525          * here and hit the 'default' below, so only disallow
4526          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4527          */
4528         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4529             mddev->ro && mddev->pers) {
4530                 if (mddev->ro == 2) {
4531                         mddev->ro = 0;
4532                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4533                 md_wakeup_thread(mddev->thread);
4534
4535                 } else {
4536                         err = -EROFS;
4537                         goto abort_unlock;
4538                 }
4539         }
4540
4541         switch (cmd)
4542         {
4543                 case ADD_NEW_DISK:
4544                 {
4545                         mdu_disk_info_t info;
4546                         if (copy_from_user(&info, argp, sizeof(info)))
4547                                 err = -EFAULT;
4548                         else
4549                                 err = add_new_disk(mddev, &info);
4550                         goto done_unlock;
4551                 }
4552
4553                 case HOT_REMOVE_DISK:
4554                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4555                         goto done_unlock;
4556
4557                 case HOT_ADD_DISK:
4558                         err = hot_add_disk(mddev, new_decode_dev(arg));
4559                         goto done_unlock;
4560
4561                 case SET_DISK_FAULTY:
4562                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4563                         goto done_unlock;
4564
4565                 case RUN_ARRAY:
4566                         err = do_md_run (mddev);
4567                         goto done_unlock;
4568
4569                 case SET_BITMAP_FILE:
4570                         err = set_bitmap_file(mddev, (int)arg);
4571                         goto done_unlock;
4572
4573                 default:
4574                         err = -EINVAL;
4575                         goto abort_unlock;
4576         }
4577
4578 done_unlock:
4579 abort_unlock:
4580         mddev_unlock(mddev);
4581
4582         return err;
4583 done:
4584         if (err)
4585                 MD_BUG();
4586 abort:
4587         return err;
4588 }
4589
4590 static int md_open(struct inode *inode, struct file *file)
4591 {
4592         /*
4593          * Succeed if we can lock the mddev, which confirms that
4594          * it isn't being stopped right now.
4595          */
4596         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4597         int err;
4598
4599         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4600                 goto out;
4601
4602         err = 0;
4603         mddev_get(mddev);
4604         mddev_unlock(mddev);
4605
4606         check_disk_change(inode->i_bdev);
4607  out:
4608         return err;
4609 }
4610
4611 static int md_release(struct inode *inode, struct file * file)
4612 {
4613         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4614
4615         BUG_ON(!mddev);
4616         mddev_put(mddev);
4617
4618         return 0;
4619 }
4620
4621 static int md_media_changed(struct gendisk *disk)
4622 {
4623         mddev_t *mddev = disk->private_data;
4624
4625         return mddev->changed;
4626 }
4627
4628 static int md_revalidate(struct gendisk *disk)
4629 {
4630         mddev_t *mddev = disk->private_data;
4631
4632         mddev->changed = 0;
4633         return 0;
4634 }
4635 static struct block_device_operations md_fops =
4636 {
4637         .owner          = THIS_MODULE,
4638         .open           = md_open,
4639         .release        = md_release,
4640         .ioctl          = md_ioctl,
4641         .getgeo         = md_getgeo,
4642         .media_changed  = md_media_changed,
4643         .revalidate_disk= md_revalidate,
4644 };
4645
4646 static int md_thread(void * arg)
4647 {
4648         mdk_thread_t *thread = arg;
4649
4650         /*
4651          * md_thread is a 'system-thread', it's priority should be very
4652          * high. We avoid resource deadlocks individually in each
4653          * raid personality. (RAID5 does preallocation) We also use RR and
4654          * the very same RT priority as kswapd, thus we will never get
4655          * into a priority inversion deadlock.
4656          *
4657          * we definitely have to have equal or higher priority than
4658          * bdflush, otherwise bdflush will deadlock if there are too
4659          * many dirty RAID5 blocks.
4660          */
4661
4662         allow_signal(SIGKILL);
4663         while (!kthread_should_stop()) {
4664
4665                 /* We need to wait INTERRUPTIBLE so that
4666                  * we don't add to the load-average.
4667                  * That means we need to be sure no signals are
4668                  * pending
4669                  */
4670                 if (signal_pending(current))
4671                         flush_signals(current);
4672
4673                 wait_event_interruptible_timeout
4674                         (thread->wqueue,
4675                          test_bit(THREAD_WAKEUP, &thread->flags)
4676                          || kthread_should_stop(),
4677                          thread->timeout);
4678
4679                 clear_bit(THREAD_WAKEUP, &thread->flags);
4680
4681                 thread->run(thread->mddev);
4682         }
4683
4684         return 0;
4685 }
4686
4687 void md_wakeup_thread(mdk_thread_t *thread)
4688 {
4689         if (thread) {
4690                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4691                 set_bit(THREAD_WAKEUP, &thread->flags);
4692                 wake_up(&thread->wqueue);
4693         }
4694 }
4695
4696 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4697                                  const char *name)
4698 {
4699         mdk_thread_t *thread;
4700
4701         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4702         if (!thread)
4703                 return NULL;
4704
4705         init_waitqueue_head(&thread->wqueue);
4706
4707         thread->run = run;
4708         thread->mddev = mddev;
4709         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4710         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4711         if (IS_ERR(thread->tsk)) {
4712                 kfree(thread);
4713                 return NULL;
4714         }
4715         return thread;
4716 }
4717
4718 void md_unregister_thread(mdk_thread_t *thread)
4719 {
4720         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4721
4722         kthread_stop(thread->tsk);
4723         kfree(thread);
4724 }
4725
4726 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4727 {
4728         if (!mddev) {
4729                 MD_BUG();
4730                 return;
4731         }
4732
4733         if (!rdev || test_bit(Faulty, &rdev->flags))
4734                 return;
4735 /*
4736         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4737                 mdname(mddev),
4738                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4739                 __builtin_return_address(0),__builtin_return_address(1),
4740                 __builtin_return_address(2),__builtin_return_address(3));
4741 */
4742         if (!mddev->pers)
4743                 return;
4744         if (!mddev->pers->error_handler)
4745                 return;
4746         mddev->pers->error_handler(mddev,rdev);
4747         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4748         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4749         md_wakeup_thread(mddev->thread);
4750         md_new_event_inintr(mddev);
4751 }
4752
4753 /* seq_file implementation /proc/mdstat */
4754
4755 static void status_unused(struct seq_file *seq)
4756 {
4757         int i = 0;
4758         mdk_rdev_t *rdev;
4759         struct list_head *tmp;
4760
4761         seq_printf(seq, "unused devices: ");
4762
4763         ITERATE_RDEV_PENDING(rdev,tmp) {
4764                 char b[BDEVNAME_SIZE];
4765                 i++;
4766                 seq_printf(seq, "%s ",
4767                               bdevname(rdev->bdev,b));
4768         }
4769         if (!i)
4770                 seq_printf(seq, "<none>");
4771
4772         seq_printf(seq, "\n");
4773 }
4774
4775
4776 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4777 {
4778         sector_t max_blocks, resync, res;
4779         unsigned long dt, db, rt;
4780         int scale;
4781         unsigned int per_milli;
4782
4783         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4784
4785         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4786                 max_blocks = mddev->resync_max_sectors >> 1;
4787         else
4788                 max_blocks = mddev->size;
4789
4790         /*
4791          * Should not happen.
4792          */
4793         if (!max_blocks) {
4794                 MD_BUG();
4795                 return;
4796         }
4797         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4798          * in a sector_t, and (max_blocks>>scale) will fit in a
4799          * u32, as those are the requirements for sector_div.
4800          * Thus 'scale' must be at least 10
4801          */
4802         scale = 10;
4803         if (sizeof(sector_t) > sizeof(unsigned long)) {
4804                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4805                         scale++;
4806         }
4807         res = (resync>>scale)*1000;
4808         sector_div(res, (u32)((max_blocks>>scale)+1));
4809
4810         per_milli = res;
4811         {
4812                 int i, x = per_milli/50, y = 20-x;
4813                 seq_printf(seq, "[");
4814                 for (i = 0; i < x; i++)
4815                         seq_printf(seq, "=");
4816                 seq_printf(seq, ">");
4817                 for (i = 0; i < y; i++)
4818                         seq_printf(seq, ".");
4819                 seq_printf(seq, "] ");
4820         }
4821         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4822                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4823                     "reshape" :
4824                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4825                      "check" :
4826                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4827                       "resync" : "recovery"))),
4828                    per_milli/10, per_milli % 10,
4829                    (unsigned long long) resync,
4830                    (unsigned long long) max_blocks);
4831
4832         /*
4833          * We do not want to overflow, so the order of operands and
4834          * the * 100 / 100 trick are important. We do a +1 to be
4835          * safe against division by zero. We only estimate anyway.
4836          *
4837          * dt: time from mark until now
4838          * db: blocks written from mark until now
4839          * rt: remaining time
4840          */
4841         dt = ((jiffies - mddev->resync_mark) / HZ);
4842         if (!dt) dt++;
4843         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4844                 - mddev->resync_mark_cnt;
4845         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4846
4847         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4848
4849         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4850 }
4851
4852 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4853 {
4854         struct list_head *tmp;
4855         loff_t l = *pos;
4856         mddev_t *mddev;
4857
4858         if (l >= 0x10000)
4859                 return NULL;
4860         if (!l--)
4861                 /* header */
4862                 return (void*)1;
4863
4864         spin_lock(&all_mddevs_lock);
4865         list_for_each(tmp,&all_mddevs)
4866                 if (!l--) {
4867                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4868                         mddev_get(mddev);
4869                         spin_unlock(&all_mddevs_lock);
4870                         return mddev;
4871                 }
4872         spin_unlock(&all_mddevs_lock);
4873         if (!l--)
4874                 return (void*)2;/* tail */
4875         return NULL;
4876 }
4877
4878 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4879 {
4880         struct list_head *tmp;
4881         mddev_t *next_mddev, *mddev = v;
4882
4883         ++*pos;
4884         if (v == (void*)2)
4885                 return NULL;
4886
4887         spin_lock(&all_mddevs_lock);
4888         if (v == (void*)1)
4889                 tmp = all_mddevs.next;
4890         else
4891                 tmp = mddev->all_mddevs.next;
4892         if (tmp != &all_mddevs)
4893                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4894         else {
4895                 next_mddev = (void*)2;
4896                 *pos = 0x10000;
4897         }
4898         spin_unlock(&all_mddevs_lock);
4899
4900         if (v != (void*)1)
4901                 mddev_put(mddev);
4902         return next_mddev;
4903
4904 }
4905
4906 static void md_seq_stop(struct seq_file *seq, void *v)
4907 {
4908         mddev_t *mddev = v;
4909
4910         if (mddev && v != (void*)1 && v != (void*)2)
4911                 mddev_put(mddev);
4912 }
4913
4914 struct mdstat_info {
4915         int event;
4916 };
4917
4918 static int md_seq_show(struct seq_file *seq, void *v)
4919 {
4920         mddev_t *mddev = v;
4921         sector_t size;
4922         struct list_head *tmp2;
4923         mdk_rdev_t *rdev;
4924         struct mdstat_info *mi = seq->private;
4925         struct bitmap *bitmap;
4926
4927         if (v == (void*)1) {
4928                 struct mdk_personality *pers;
4929                 seq_printf(seq, "Personalities : ");
4930                 spin_lock(&pers_lock);
4931                 list_for_each_entry(pers, &pers_list, list)
4932                         seq_printf(seq, "[%s] ", pers->name);
4933
4934                 spin_unlock(&pers_lock);
4935                 seq_printf(seq, "\n");
4936                 mi->event = atomic_read(&md_event_count);
4937                 return 0;
4938         }
4939         if (v == (void*)2) {
4940                 status_unused(seq);
4941                 return 0;
4942         }
4943
4944         if (mddev_lock(mddev) < 0)
4945                 return -EINTR;
4946
4947         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4948                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4949                                                 mddev->pers ? "" : "in");
4950                 if (mddev->pers) {
4951                         if (mddev->ro==1)
4952                                 seq_printf(seq, " (read-only)");
4953                         if (mddev->ro==2)
4954                                 seq_printf(seq, "(auto-read-only)");
4955                         seq_printf(seq, " %s", mddev->pers->name);
4956                 }
4957
4958                 size = 0;
4959                 ITERATE_RDEV(mddev,rdev,tmp2) {
4960                         char b[BDEVNAME_SIZE];
4961                         seq_printf(seq, " %s[%d]",
4962                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4963                         if (test_bit(WriteMostly, &rdev->flags))
4964                                 seq_printf(seq, "(W)");
4965                         if (test_bit(Faulty, &rdev->flags)) {
4966                                 seq_printf(seq, "(F)");
4967                                 continue;
4968                         } else if (rdev->raid_disk < 0)
4969                                 seq_printf(seq, "(S)"); /* spare */
4970                         size += rdev->size;
4971                 }
4972
4973                 if (!list_empty(&mddev->disks)) {
4974                         if (mddev->pers)
4975                                 seq_printf(seq, "\n      %llu blocks",
4976                                         (unsigned long long)mddev->array_size);
4977                         else
4978                                 seq_printf(seq, "\n      %llu blocks",
4979                                         (unsigned long long)size);
4980                 }
4981                 if (mddev->persistent) {
4982                         if (mddev->major_version != 0 ||
4983                             mddev->minor_version != 90) {
4984                                 seq_printf(seq," super %d.%d",
4985                                            mddev->major_version,
4986                                            mddev->minor_version);
4987                         }
4988                 } else
4989                         seq_printf(seq, " super non-persistent");
4990
4991                 if (mddev->pers) {
4992                         mddev->pers->status (seq, mddev);
4993                         seq_printf(seq, "\n      ");
4994                         if (mddev->pers->sync_request) {
4995                                 if (mddev->curr_resync > 2) {
4996                                         status_resync (seq, mddev);
4997                                         seq_printf(seq, "\n      ");
4998                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4999                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5000                                 else if (mddev->recovery_cp < MaxSector)
5001                                         seq_printf(seq, "\tresync=PENDING\n      ");
5002                         }
5003                 } else
5004                         seq_printf(seq, "\n       ");
5005
5006                 if ((bitmap = mddev->bitmap)) {
5007                         unsigned long chunk_kb;
5008                         unsigned long flags;
5009                         spin_lock_irqsave(&bitmap->lock, flags);
5010                         chunk_kb = bitmap->chunksize >> 10;
5011                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5012                                 "%lu%s chunk",
5013                                 bitmap->pages - bitmap->missing_pages,
5014                                 bitmap->pages,
5015                                 (bitmap->pages - bitmap->missing_pages)
5016                                         << (PAGE_SHIFT - 10),
5017                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5018                                 chunk_kb ? "KB" : "B");
5019                         if (bitmap->file) {
5020                                 seq_printf(seq, ", file: ");
5021                                 seq_path(seq, bitmap->file->f_path.mnt,
5022                                          bitmap->file->f_path.dentry," \t\n");
5023                         }
5024
5025                         seq_printf(seq, "\n");
5026                         spin_unlock_irqrestore(&bitmap->lock, flags);
5027                 }
5028
5029                 seq_printf(seq, "\n");
5030         }
5031         mddev_unlock(mddev);
5032
5033         return 0;
5034 }
5035
5036 static struct seq_operations md_seq_ops = {
5037         .start  = md_seq_start,
5038         .next   = md_seq_next,
5039         .stop   = md_seq_stop,
5040         .show   = md_seq_show,
5041 };
5042
5043 static int md_seq_open(struct inode *inode, struct file *file)
5044 {
5045         int error;
5046         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5047         if (mi == NULL)
5048                 return -ENOMEM;
5049
5050         error = seq_open(file, &md_seq_ops);
5051         if (error)
5052                 kfree(mi);
5053         else {
5054                 struct seq_file *p = file->private_data;
5055                 p->private = mi;
5056                 mi->event = atomic_read(&md_event_count);
5057         }
5058         return error;
5059 }
5060
5061 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5062 {
5063         struct seq_file *m = filp->private_data;
5064         struct mdstat_info *mi = m->private;
5065         int mask;
5066
5067         poll_wait(filp, &md_event_waiters, wait);
5068
5069         /* always allow read */
5070         mask = POLLIN | POLLRDNORM;
5071
5072         if (mi->event != atomic_read(&md_event_count))
5073                 mask |= POLLERR | POLLPRI;
5074         return mask;
5075 }
5076
5077 static const struct file_operations md_seq_fops = {
5078         .owner          = THIS_MODULE,
5079         .open           = md_seq_open,
5080         .read           = seq_read,
5081         .llseek         = seq_lseek,
5082         .release        = seq_release_private,
5083         .poll           = mdstat_poll,
5084 };
5085
5086 int register_md_personality(struct mdk_personality *p)
5087 {
5088         spin_lock(&pers_lock);
5089         list_add_tail(&p->list, &pers_list);
5090         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5091         spin_unlock(&pers_lock);
5092         return 0;
5093 }
5094
5095 int unregister_md_personality(struct mdk_personality *p)
5096 {
5097         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5098         spin_lock(&pers_lock);
5099         list_del_init(&p->list);
5100         spin_unlock(&pers_lock);
5101         return 0;
5102 }
5103
5104 static int is_mddev_idle(mddev_t *mddev)
5105 {
5106         mdk_rdev_t * rdev;
5107         struct list_head *tmp;
5108         int idle;
5109         long curr_events;
5110
5111         idle = 1;
5112         ITERATE_RDEV(mddev,rdev,tmp) {
5113                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5114                 curr_events = disk_stat_read(disk, sectors[0]) +
5115                                 disk_stat_read(disk, sectors[1]) -
5116                                 atomic_read(&disk->sync_io);
5117                 /* sync IO will cause sync_io to increase before the disk_stats
5118                  * as sync_io is counted when a request starts, and
5119                  * disk_stats is counted when it completes.
5120                  * So resync activity will cause curr_events to be smaller than
5121                  * when there was no such activity.
5122                  * non-sync IO will cause disk_stat to increase without
5123                  * increasing sync_io so curr_events will (eventually)
5124                  * be larger than it was before.  Once it becomes
5125                  * substantially larger, the test below will cause
5126                  * the array to appear non-idle, and resync will slow
5127                  * down.
5128                  * If there is a lot of outstanding resync activity when
5129                  * we set last_event to curr_events, then all that activity
5130                  * completing might cause the array to appear non-idle
5131                  * and resync will be slowed down even though there might
5132                  * not have been non-resync activity.  This will only
5133                  * happen once though.  'last_events' will soon reflect
5134                  * the state where there is little or no outstanding
5135                  * resync requests, and further resync activity will
5136                  * always make curr_events less than last_events.
5137                  *
5138                  */
5139                 if (curr_events - rdev->last_events > 4096) {
5140                         rdev->last_events = curr_events;
5141                         idle = 0;
5142                 }
5143         }
5144         return idle;
5145 }
5146
5147 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5148 {
5149         /* another "blocks" (512byte) blocks have been synced */
5150         atomic_sub(blocks, &mddev->recovery_active);
5151         wake_up(&mddev->recovery_wait);
5152         if (!ok) {
5153                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5154                 md_wakeup_thread(mddev->thread);
5155                 // stop recovery, signal do_sync ....
5156         }
5157 }
5158
5159
5160 /* md_write_start(mddev, bi)
5161  * If we need to update some array metadata (e.g. 'active' flag
5162  * in superblock) before writing, schedule a superblock update
5163  * and wait for it to complete.
5164  */
5165 void md_write_start(mddev_t *mddev, struct bio *bi)
5166 {
5167         if (bio_data_dir(bi) != WRITE)
5168                 return;
5169
5170         BUG_ON(mddev->ro == 1);
5171         if (mddev->ro == 2) {
5172                 /* need to switch to read/write */
5173                 mddev->ro = 0;
5174                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5175                 md_wakeup_thread(mddev->thread);
5176         }
5177         atomic_inc(&mddev->writes_pending);
5178         if (mddev->in_sync) {
5179                 spin_lock_irq(&mddev->write_lock);
5180                 if (mddev->in_sync) {
5181                         mddev->in_sync = 0;
5182                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5183                         md_wakeup_thread(mddev->thread);
5184                 }
5185                 spin_unlock_irq(&mddev->write_lock);
5186         }
5187         wait_event(mddev->sb_wait, mddev->flags==0);
5188 }
5189
5190 void md_write_end(mddev_t *mddev)
5191 {
5192         if (atomic_dec_and_test(&mddev->writes_pending)) {
5193                 if (mddev->safemode == 2)
5194                         md_wakeup_thread(mddev->thread);
5195                 else if (mddev->safemode_delay)
5196                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5197         }
5198 }
5199
5200 /* md_allow_write(mddev)
5201  * Calling this ensures that the array is marked 'active' so that writes
5202  * may proceed without blocking.  It is important to call this before
5203  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5204  * Must be called with mddev_lock held.
5205  */
5206 void md_allow_write(mddev_t *mddev)
5207 {
5208         if (!mddev->pers)
5209                 return;
5210         if (mddev->ro)
5211                 return;
5212
5213         spin_lock_irq(&mddev->write_lock);
5214         if (mddev->in_sync) {
5215                 mddev->in_sync = 0;
5216                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5217                 if (mddev->safemode_delay &&
5218                     mddev->safemode == 0)
5219                         mddev->safemode = 1;
5220                 spin_unlock_irq(&mddev->write_lock);
5221                 md_update_sb(mddev, 0);
5222         } else
5223                 spin_unlock_irq(&mddev->write_lock);
5224 }
5225 EXPORT_SYMBOL_GPL(md_allow_write);
5226
5227 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5228
5229 #define SYNC_MARKS      10
5230 #define SYNC_MARK_STEP  (3*HZ)
5231 void md_do_sync(mddev_t *mddev)
5232 {
5233         mddev_t *mddev2;
5234         unsigned int currspeed = 0,
5235                  window;
5236         sector_t max_sectors,j, io_sectors;
5237         unsigned long mark[SYNC_MARKS];
5238         sector_t mark_cnt[SYNC_MARKS];
5239         int last_mark,m;
5240         struct list_head *tmp;
5241         sector_t last_check;
5242         int skipped = 0;
5243         struct list_head *rtmp;
5244         mdk_rdev_t *rdev;
5245         char *desc;
5246
5247         /* just incase thread restarts... */
5248         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5249                 return;
5250         if (mddev->ro) /* never try to sync a read-only array */
5251                 return;
5252
5253         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5254                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5255                         desc = "data-check";
5256                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5257                         desc = "requested-resync";
5258                 else
5259                         desc = "resync";
5260         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5261                 desc = "reshape";
5262         else
5263                 desc = "recovery";
5264
5265         /* we overload curr_resync somewhat here.
5266          * 0 == not engaged in resync at all
5267          * 2 == checking that there is no conflict with another sync
5268          * 1 == like 2, but have yielded to allow conflicting resync to
5269          *              commense
5270          * other == active in resync - this many blocks
5271          *
5272          * Before starting a resync we must have set curr_resync to
5273          * 2, and then checked that every "conflicting" array has curr_resync
5274          * less than ours.  When we find one that is the same or higher
5275          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5276          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5277          * This will mean we have to start checking from the beginning again.
5278          *
5279          */
5280
5281         do {
5282                 mddev->curr_resync = 2;
5283
5284         try_again:
5285                 if (kthread_should_stop()) {
5286                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5287                         goto skip;
5288                 }
5289                 ITERATE_MDDEV(mddev2,tmp) {
5290                         if (mddev2 == mddev)
5291                                 continue;
5292                         if (mddev2->curr_resync &&
5293                             match_mddev_units(mddev,mddev2)) {
5294                                 DEFINE_WAIT(wq);
5295                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5296                                         /* arbitrarily yield */
5297                                         mddev->curr_resync = 1;
5298                                         wake_up(&resync_wait);
5299                                 }
5300                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5301                                         /* no need to wait here, we can wait the next
5302                                          * time 'round when curr_resync == 2
5303                                          */
5304                                         continue;
5305                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5306                                 if (!kthread_should_stop() &&
5307                                     mddev2->curr_resync >= mddev->curr_resync) {
5308                                         printk(KERN_INFO "md: delaying %s of %s"
5309                                                " until %s has finished (they"
5310                                                " share one or more physical units)\n",
5311                                                desc, mdname(mddev), mdname(mddev2));
5312                                         mddev_put(mddev2);
5313                                         schedule();
5314                                         finish_wait(&resync_wait, &wq);
5315                                         goto try_again;
5316                                 }
5317                                 finish_wait(&resync_wait, &wq);
5318                         }
5319                 }
5320         } while (mddev->curr_resync < 2);
5321
5322         j = 0;
5323         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5324                 /* resync follows the size requested by the personality,
5325                  * which defaults to physical size, but can be virtual size
5326                  */
5327                 max_sectors = mddev->resync_max_sectors;
5328                 mddev->resync_mismatches = 0;
5329                 /* we don't use the checkpoint if there's a bitmap */
5330                 if (!mddev->bitmap &&
5331                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5332                         j = mddev->recovery_cp;
5333         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5334                 max_sectors = mddev->size << 1;
5335         else {
5336                 /* recovery follows the physical size of devices */
5337                 max_sectors = mddev->size << 1;
5338                 j = MaxSector;
5339                 ITERATE_RDEV(mddev,rdev,rtmp)
5340                         if (rdev->raid_disk >= 0 &&
5341                             !test_bit(Faulty, &rdev->flags) &&
5342                             !test_bit(In_sync, &rdev->flags) &&
5343                             rdev->recovery_offset < j)
5344                                 j = rdev->recovery_offset;
5345         }
5346
5347         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5348         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5349                 " %d KB/sec/disk.\n", speed_min(mddev));
5350         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5351                "(but not more than %d KB/sec) for %s.\n",
5352                speed_max(mddev), desc);
5353
5354         is_mddev_idle(mddev); /* this also initializes IO event counters */
5355
5356         io_sectors = 0;
5357         for (m = 0; m < SYNC_MARKS; m++) {
5358                 mark[m] = jiffies;
5359                 mark_cnt[m] = io_sectors;
5360         }
5361         last_mark = 0;
5362         mddev->resync_mark = mark[last_mark];
5363         mddev->resync_mark_cnt = mark_cnt[last_mark];
5364
5365         /*
5366          * Tune reconstruction:
5367          */
5368         window = 32*(PAGE_SIZE/512);
5369         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5370                 window/2,(unsigned long long) max_sectors/2);
5371
5372         atomic_set(&mddev->recovery_active, 0);
5373         init_waitqueue_head(&mddev->recovery_wait);
5374         last_check = 0;
5375
5376         if (j>2) {
5377                 printk(KERN_INFO
5378                        "md: resuming %s of %s from checkpoint.\n",
5379                        desc, mdname(mddev));
5380                 mddev->curr_resync = j;
5381         }
5382
5383         while (j < max_sectors) {
5384                 sector_t sectors;
5385
5386                 skipped = 0;
5387                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5388                                             currspeed < speed_min(mddev));
5389                 if (sectors == 0) {
5390                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5391                         goto out;
5392                 }
5393
5394                 if (!skipped) { /* actual IO requested */
5395                         io_sectors += sectors;
5396                         atomic_add(sectors, &mddev->recovery_active);
5397                 }
5398
5399                 j += sectors;
5400                 if (j>1) mddev->curr_resync = j;
5401                 mddev->curr_mark_cnt = io_sectors;
5402                 if (last_check == 0)
5403                         /* this is the earliers that rebuilt will be
5404                          * visible in /proc/mdstat
5405                          */
5406                         md_new_event(mddev);
5407
5408                 if (last_check + window > io_sectors || j == max_sectors)
5409                         continue;
5410
5411                 last_check = io_sectors;
5412
5413                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5414                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5415                         break;
5416
5417         repeat:
5418                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5419                         /* step marks */
5420                         int next = (last_mark+1) % SYNC_MARKS;
5421
5422                         mddev->resync_mark = mark[next];
5423                         mddev->resync_mark_cnt = mark_cnt[next];
5424                         mark[next] = jiffies;
5425                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5426                         last_mark = next;
5427                 }
5428
5429
5430                 if (kthread_should_stop()) {
5431                         /*
5432                          * got a signal, exit.
5433                          */
5434                         printk(KERN_INFO
5435                                 "md: md_do_sync() got signal ... exiting\n");
5436                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5437                         goto out;
5438                 }
5439
5440                 /*
5441                  * this loop exits only if either when we are slower than
5442                  * the 'hard' speed limit, or the system was IO-idle for
5443                  * a jiffy.
5444                  * the system might be non-idle CPU-wise, but we only care
5445                  * about not overloading the IO subsystem. (things like an
5446                  * e2fsck being done on the RAID array should execute fast)
5447                  */
5448                 blk_unplug(mddev->queue);
5449                 cond_resched();
5450
5451                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5452                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5453
5454                 if (currspeed > speed_min(mddev)) {
5455                         if ((currspeed > speed_max(mddev)) ||
5456                                         !is_mddev_idle(mddev)) {
5457                                 msleep(500);
5458                                 goto repeat;
5459                         }
5460                 }
5461         }
5462         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5463         /*
5464          * this also signals 'finished resyncing' to md_stop
5465          */
5466  out:
5467         blk_unplug(mddev->queue);
5468
5469         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5470
5471         /* tell personality that we are finished */
5472         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5473
5474         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5475             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5476             mddev->curr_resync > 2) {
5477                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5478                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5479                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5480                                         printk(KERN_INFO
5481                                                "md: checkpointing %s of %s.\n",
5482                                                desc, mdname(mddev));
5483                                         mddev->recovery_cp = mddev->curr_resync;
5484                                 }
5485                         } else
5486                                 mddev->recovery_cp = MaxSector;
5487                 } else {
5488                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5489                                 mddev->curr_resync = MaxSector;
5490                         ITERATE_RDEV(mddev,rdev,rtmp)
5491                                 if (rdev->raid_disk >= 0 &&
5492                                     !test_bit(Faulty, &rdev->flags) &&
5493                                     !test_bit(In_sync, &rdev->flags) &&
5494                                     rdev->recovery_offset < mddev->curr_resync)
5495                                         rdev->recovery_offset = mddev->curr_resync;
5496                 }
5497         }
5498         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5499
5500  skip:
5501         mddev->curr_resync = 0;
5502         wake_up(&resync_wait);
5503         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5504         md_wakeup_thread(mddev->thread);
5505 }
5506 EXPORT_SYMBOL_GPL(md_do_sync);
5507
5508
5509 static int remove_and_add_spares(mddev_t *mddev)
5510 {
5511         mdk_rdev_t *rdev;
5512         struct list_head *rtmp;
5513         int spares = 0;
5514
5515         ITERATE_RDEV(mddev,rdev,rtmp)
5516                 if (rdev->raid_disk >= 0 &&
5517                     (test_bit(Faulty, &rdev->flags) ||
5518                      ! test_bit(In_sync, &rdev->flags)) &&
5519                     atomic_read(&rdev->nr_pending)==0) {
5520                         if (mddev->pers->hot_remove_disk(
5521                                     mddev, rdev->raid_disk)==0) {
5522                                 char nm[20];
5523                                 sprintf(nm,"rd%d", rdev->raid_disk);
5524                                 sysfs_remove_link(&mddev->kobj, nm);
5525                                 rdev->raid_disk = -1;
5526                         }
5527                 }
5528
5529         if (mddev->degraded) {
5530                 ITERATE_RDEV(mddev,rdev,rtmp)
5531                         if (rdev->raid_disk < 0
5532                             && !test_bit(Faulty, &rdev->flags)) {
5533                                 rdev->recovery_offset = 0;
5534                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5535                                         char nm[20];
5536                                         sprintf(nm, "rd%d", rdev->raid_disk);
5537                                         if (sysfs_create_link(&mddev->kobj,
5538                                                               &rdev->kobj, nm))
5539                                                 printk(KERN_WARNING
5540                                                        "md: cannot register "
5541                                                        "%s for %s\n",
5542                                                        nm, mdname(mddev));
5543                                         spares++;
5544                                         md_new_event(mddev);
5545                                 } else
5546                                         break;
5547                         }
5548         }
5549         return spares;
5550 }
5551 /*
5552  * This routine is regularly called by all per-raid-array threads to
5553  * deal with generic issues like resync and super-block update.
5554  * Raid personalities that don't have a thread (linear/raid0) do not
5555  * need this as they never do any recovery or update the superblock.
5556  *
5557  * It does not do any resync itself, but rather "forks" off other threads
5558  * to do that as needed.
5559  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5560  * "->recovery" and create a thread at ->sync_thread.
5561  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5562  * and wakeups up this thread which will reap the thread and finish up.
5563  * This thread also removes any faulty devices (with nr_pending == 0).
5564  *
5565  * The overall approach is:
5566  *  1/ if the superblock needs updating, update it.
5567  *  2/ If a recovery thread is running, don't do anything else.
5568  *  3/ If recovery has finished, clean up, possibly marking spares active.
5569  *  4/ If there are any faulty devices, remove them.
5570  *  5/ If array is degraded, try to add spares devices
5571  *  6/ If array has spares or is not in-sync, start a resync thread.
5572  */
5573 void md_check_recovery(mddev_t *mddev)
5574 {
5575         mdk_rdev_t *rdev;
5576         struct list_head *rtmp;
5577
5578
5579         if (mddev->bitmap)
5580                 bitmap_daemon_work(mddev->bitmap);
5581
5582         if (mddev->ro)
5583                 return;
5584
5585         if (signal_pending(current)) {
5586                 if (mddev->pers->sync_request) {
5587                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5588                                mdname(mddev));
5589                         mddev->safemode = 2;
5590                 }
5591                 flush_signals(current);
5592         }
5593
5594         if ( ! (
5595                 mddev->flags ||
5596                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5597                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5598                 (mddev->safemode == 1) ||
5599                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5600                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5601                 ))
5602                 return;
5603
5604         if (mddev_trylock(mddev)) {
5605                 int spares = 0;
5606
5607                 spin_lock_irq(&mddev->write_lock);
5608                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5609                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5610                         mddev->in_sync = 1;
5611                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5612                 }
5613                 if (mddev->safemode == 1)
5614                         mddev->safemode = 0;
5615                 spin_unlock_irq(&mddev->write_lock);
5616
5617                 if (mddev->flags)
5618                         md_update_sb(mddev, 0);
5619
5620
5621                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5622                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5623                         /* resync/recovery still happening */
5624                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5625                         goto unlock;
5626                 }
5627                 if (mddev->sync_thread) {
5628                         /* resync has finished, collect result */
5629                         md_unregister_thread(mddev->sync_thread);
5630                         mddev->sync_thread = NULL;
5631                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5632                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5633                                 /* success...*/
5634                                 /* activate any spares */
5635                                 mddev->pers->spare_active(mddev);
5636                         }
5637                         md_update_sb(mddev, 1);
5638
5639                         /* if array is no-longer degraded, then any saved_raid_disk
5640                          * information must be scrapped
5641                          */
5642                         if (!mddev->degraded)
5643                                 ITERATE_RDEV(mddev,rdev,rtmp)
5644                                         rdev->saved_raid_disk = -1;
5645
5646                         mddev->recovery = 0;
5647                         /* flag recovery needed just to double check */
5648                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5649                         md_new_event(mddev);
5650                         goto unlock;
5651                 }
5652                 /* Clear some bits that don't mean anything, but
5653                  * might be left set
5654                  */
5655                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5656                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5657                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5658                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5659
5660                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5661                         goto unlock;
5662                 /* no recovery is running.
5663                  * remove any failed drives, then
5664                  * add spares if possible.
5665                  * Spare are also removed and re-added, to allow
5666                  * the personality to fail the re-add.
5667                  */
5668
5669                 if (mddev->reshape_position != MaxSector) {
5670                         if (mddev->pers->check_reshape(mddev) != 0)
5671                                 /* Cannot proceed */
5672                                 goto unlock;
5673                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5674                 } else if ((spares = remove_and_add_spares(mddev))) {
5675                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5676                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5677                 } else if (mddev->recovery_cp < MaxSector) {
5678                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5679                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5680                         /* nothing to be done ... */
5681                         goto unlock;
5682
5683                 if (mddev->pers->sync_request) {
5684                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5685                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5686                                 /* We are adding a device or devices to an array
5687                                  * which has the bitmap stored on all devices.
5688                                  * So make sure all bitmap pages get written
5689                                  */
5690                                 bitmap_write_all(mddev->bitmap);
5691                         }
5692                         mddev->sync_thread = md_register_thread(md_do_sync,
5693                                                                 mddev,
5694                                                                 "%s_resync");
5695                         if (!mddev->sync_thread) {
5696                                 printk(KERN_ERR "%s: could not start resync"
5697                                         " thread...\n",
5698                                         mdname(mddev));
5699                                 /* leave the spares where they are, it shouldn't hurt */
5700                                 mddev->recovery = 0;
5701                         } else
5702                                 md_wakeup_thread(mddev->sync_thread);
5703                         md_new_event(mddev);
5704                 }
5705         unlock:
5706                 mddev_unlock(mddev);
5707         }
5708 }
5709
5710 static int md_notify_reboot(struct notifier_block *this,
5711                             unsigned long code, void *x)
5712 {
5713         struct list_head *tmp;
5714         mddev_t *mddev;
5715
5716         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5717
5718                 printk(KERN_INFO "md: stopping all md devices.\n");
5719
5720                 ITERATE_MDDEV(mddev,tmp)
5721                         if (mddev_trylock(mddev)) {
5722                                 do_md_stop (mddev, 1);
5723                                 mddev_unlock(mddev);
5724                         }
5725                 /*
5726                  * certain more exotic SCSI devices are known to be
5727                  * volatile wrt too early system reboots. While the
5728                  * right place to handle this issue is the given
5729                  * driver, we do want to have a safe RAID driver ...
5730                  */
5731                 mdelay(1000*1);
5732         }
5733         return NOTIFY_DONE;
5734 }
5735
5736 static struct notifier_block md_notifier = {
5737         .notifier_call  = md_notify_reboot,
5738         .next           = NULL,
5739         .priority       = INT_MAX, /* before any real devices */
5740 };
5741
5742 static void md_geninit(void)
5743 {
5744         struct proc_dir_entry *p;
5745
5746         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5747
5748         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5749         if (p)
5750                 p->proc_fops = &md_seq_fops;
5751 }
5752
5753 static int __init md_init(void)
5754 {
5755         if (register_blkdev(MAJOR_NR, "md"))
5756                 return -1;
5757         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5758                 unregister_blkdev(MAJOR_NR, "md");
5759                 return -1;
5760         }
5761         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5762                             md_probe, NULL, NULL);
5763         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5764                             md_probe, NULL, NULL);
5765
5766         register_reboot_notifier(&md_notifier);
5767         raid_table_header = register_sysctl_table(raid_root_table);
5768
5769         md_geninit();
5770         return (0);
5771 }
5772
5773
5774 #ifndef MODULE
5775
5776 /*
5777  * Searches all registered partitions for autorun RAID arrays
5778  * at boot time.
5779  */
5780
5781 static LIST_HEAD(all_detected_devices);
5782 struct detected_devices_node {
5783         struct list_head list;
5784         dev_t dev;
5785 };
5786
5787 void md_autodetect_dev(dev_t dev)
5788 {
5789         struct detected_devices_node *node_detected_dev;
5790
5791         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5792         if (node_detected_dev) {
5793                 node_detected_dev->dev = dev;
5794                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5795         } else {
5796                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5797                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
5798         }
5799 }
5800
5801
5802 static void autostart_arrays(int part)
5803 {
5804         mdk_rdev_t *rdev;
5805         struct detected_devices_node *node_detected_dev;
5806         dev_t dev;
5807         int i_scanned, i_passed;
5808
5809         i_scanned = 0;
5810         i_passed = 0;
5811
5812         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5813
5814         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
5815                 i_scanned++;
5816                 node_detected_dev = list_entry(all_detected_devices.next,
5817                                         struct detected_devices_node, list);
5818                 list_del(&node_detected_dev->list);
5819                 dev = node_detected_dev->dev;
5820                 kfree(node_detected_dev);
5821                 rdev = md_import_device(dev,0, 90);
5822                 if (IS_ERR(rdev))
5823                         continue;
5824
5825                 if (test_bit(Faulty, &rdev->flags)) {
5826                         MD_BUG();
5827                         continue;
5828                 }
5829                 list_add(&rdev->same_set, &pending_raid_disks);
5830                 i_passed++;
5831         }
5832
5833         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
5834                                                 i_scanned, i_passed);
5835
5836         autorun_devices(part);
5837 }
5838
5839 #endif /* !MODULE */
5840
5841 static __exit void md_exit(void)
5842 {
5843         mddev_t *mddev;
5844         struct list_head *tmp;
5845
5846         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5847         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5848
5849         unregister_blkdev(MAJOR_NR,"md");
5850         unregister_blkdev(mdp_major, "mdp");
5851         unregister_reboot_notifier(&md_notifier);
5852         unregister_sysctl_table(raid_table_header);
5853         remove_proc_entry("mdstat", NULL);
5854         ITERATE_MDDEV(mddev,tmp) {
5855                 struct gendisk *disk = mddev->gendisk;
5856                 if (!disk)
5857                         continue;
5858                 export_array(mddev);
5859                 del_gendisk(disk);
5860                 put_disk(disk);
5861                 mddev->gendisk = NULL;
5862                 mddev_put(mddev);
5863         }
5864 }
5865
5866 subsys_initcall(md_init);
5867 module_exit(md_exit)
5868
5869 static int get_ro(char *buffer, struct kernel_param *kp)
5870 {
5871         return sprintf(buffer, "%d", start_readonly);
5872 }
5873 static int set_ro(const char *val, struct kernel_param *kp)
5874 {
5875         char *e;
5876         int num = simple_strtoul(val, &e, 10);
5877         if (*val && (*e == '\0' || *e == '\n')) {
5878                 start_readonly = num;
5879                 return 0;
5880         }
5881         return -EINVAL;
5882 }
5883
5884 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5885 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5886
5887
5888 EXPORT_SYMBOL(register_md_personality);
5889 EXPORT_SYMBOL(unregister_md_personality);
5890 EXPORT_SYMBOL(md_error);
5891 EXPORT_SYMBOL(md_done_sync);
5892 EXPORT_SYMBOL(md_write_start);
5893 EXPORT_SYMBOL(md_write_end);
5894 EXPORT_SYMBOL(md_register_thread);
5895 EXPORT_SYMBOL(md_unregister_thread);
5896 EXPORT_SYMBOL(md_wakeup_thread);
5897 EXPORT_SYMBOL(md_check_recovery);
5898 MODULE_LICENSE("GPL");
5899 MODULE_ALIAS("md");
5900 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);