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

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/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 for_each_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_put(&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         new->resync_max = MaxSector;
 279
 280         new->queue = blk_alloc_queue(GFP_KERNEL);
 281         if (!new->queue) {
 282                 kfree(new);
 283                 return NULL;
 284         }
 285         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
 286
 287         blk_queue_make_request(new->queue, md_fail_request);
 288
 289         goto retry;
 290 }
 291
 292 static inline int mddev_lock(mddev_t * mddev)
 293 {
 294         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 295 }
 296
 297 static inline int mddev_trylock(mddev_t * mddev)
 298 {
 299         return mutex_trylock(&mddev->reconfig_mutex);
 300 }
 301
 302 static inline void mddev_unlock(mddev_t * mddev)
 303 {
 304         mutex_unlock(&mddev->reconfig_mutex);
 305
 306         md_wakeup_thread(mddev->thread);
 307 }
 308
 309 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 310 {
 311         mdk_rdev_t * rdev;
 312         struct list_head *tmp;
 313
 314         rdev_for_each(rdev, tmp, mddev) {
 315                 if (rdev->desc_nr == nr)
 316                         return rdev;
 317         }
 318         return NULL;
 319 }
 320
 321 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 322 {
 323         struct list_head *tmp;
 324         mdk_rdev_t *rdev;
 325
 326         rdev_for_each(rdev, tmp, mddev) {
 327                 if (rdev->bdev->bd_dev == dev)
 328                         return rdev;
 329         }
 330         return NULL;
 331 }
 332
 333 static struct mdk_personality *find_pers(int level, char *clevel)
 334 {
 335         struct mdk_personality *pers;
 336         list_for_each_entry(pers, &pers_list, list) {
 337                 if (level != LEVEL_NONE && pers->level == level)
 338                         return pers;
 339                 if (strcmp(pers->name, clevel)==0)
 340                         return pers;
 341         }
 342         return NULL;
 343 }
 344
 345 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 346 {
 347         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 348         return MD_NEW_SIZE_BLOCKS(size);
 349 }
 350
 351 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 352 {
 353         sector_t size;
 354
 355         size = rdev->sb_offset;
 356
 357         if (chunk_size)
 358                 size &= ~((sector_t)chunk_size/1024 - 1);
 359         return size;
 360 }
 361
 362 static int alloc_disk_sb(mdk_rdev_t * rdev)
 363 {
 364         if (rdev->sb_page)
 365                 MD_BUG();
 366
 367         rdev->sb_page = alloc_page(GFP_KERNEL);
 368         if (!rdev->sb_page) {
 369                 printk(KERN_ALERT "md: out of memory.\n");
 370                 return -EINVAL;
 371         }
 372
 373         return 0;
 374 }
 375
 376 static void free_disk_sb(mdk_rdev_t * rdev)
 377 {
 378         if (rdev->sb_page) {
 379                 put_page(rdev->sb_page);
 380                 rdev->sb_loaded = 0;
 381                 rdev->sb_page = NULL;
 382                 rdev->sb_offset = 0;
 383                 rdev->size = 0;
 384         }
 385 }
 386
 387
 388 static void super_written(struct bio *bio, int error)
 389 {
 390         mdk_rdev_t *rdev = bio->bi_private;
 391         mddev_t *mddev = rdev->mddev;
 392
 393         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 394                 printk("md: super_written gets error=%d, uptodate=%d\n",
 395                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 396                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 397                 md_error(mddev, rdev);
 398         }
 399
 400         if (atomic_dec_and_test(&mddev->pending_writes))
 401                 wake_up(&mddev->sb_wait);
 402         bio_put(bio);
 403 }
 404
 405 static void super_written_barrier(struct bio *bio, int error)
 406 {
 407         struct bio *bio2 = bio->bi_private;
 408         mdk_rdev_t *rdev = bio2->bi_private;
 409         mddev_t *mddev = rdev->mddev;
 410
 411         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 412             error == -EOPNOTSUPP) {
 413                 unsigned long flags;
 414                 /* barriers don't appear to be supported :-( */
 415                 set_bit(BarriersNotsupp, &rdev->flags);
 416                 mddev->barriers_work = 0;
 417                 spin_lock_irqsave(&mddev->write_lock, flags);
 418                 bio2->bi_next = mddev->biolist;
 419                 mddev->biolist = bio2;
 420                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 421                 wake_up(&mddev->sb_wait);
 422                 bio_put(bio);
 423         } else {
 424                 bio_put(bio2);
 425                 bio->bi_private = rdev;
 426                 super_written(bio, error);
 427         }
 428 }
 429
 430 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 431                    sector_t sector, int size, struct page *page)
 432 {
 433         /* write first size bytes of page to sector of rdev
 434          * Increment mddev->pending_writes before returning
 435          * and decrement it on completion, waking up sb_wait
 436          * if zero is reached.
 437          * If an error occurred, call md_error
 438          *
 439          * As we might need to resubmit the request if BIO_RW_BARRIER
 440          * causes ENOTSUPP, we allocate a spare bio...
 441          */
 442         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 443         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 444
 445         bio->bi_bdev = rdev->bdev;
 446         bio->bi_sector = sector;
 447         bio_add_page(bio, page, size, 0);
 448         bio->bi_private = rdev;
 449         bio->bi_end_io = super_written;
 450         bio->bi_rw = rw;
 451
 452         atomic_inc(&mddev->pending_writes);
 453         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 454                 struct bio *rbio;
 455                 rw |= (1<<BIO_RW_BARRIER);
 456                 rbio = bio_clone(bio, GFP_NOIO);
 457                 rbio->bi_private = bio;
 458                 rbio->bi_end_io = super_written_barrier;
 459                 submit_bio(rw, rbio);
 460         } else
 461                 submit_bio(rw, bio);
 462 }
 463
 464 void md_super_wait(mddev_t *mddev)
 465 {
 466         /* wait for all superblock writes that were scheduled to complete.
 467          * if any had to be retried (due to BARRIER problems), retry them
 468          */
 469         DEFINE_WAIT(wq);
 470         for(;;) {
 471                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 472                 if (atomic_read(&mddev->pending_writes)==0)
 473                         break;
 474                 while (mddev->biolist) {
 475                         struct bio *bio;
 476                         spin_lock_irq(&mddev->write_lock);
 477                         bio = mddev->biolist;
 478                         mddev->biolist = bio->bi_next ;
 479                         bio->bi_next = NULL;
 480                         spin_unlock_irq(&mddev->write_lock);
 481                         submit_bio(bio->bi_rw, bio);
 482                 }
 483                 schedule();
 484         }
 485         finish_wait(&mddev->sb_wait, &wq);
 486 }
 487
 488 static void bi_complete(struct bio *bio, int error)
 489 {
 490         complete((struct completion*)bio->bi_private);
 491 }
 492
 493 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 494                    struct page *page, int rw)
 495 {
 496         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 497         struct completion event;
 498         int ret;
 499
 500         rw |= (1 << BIO_RW_SYNC);
 501
 502         bio->bi_bdev = bdev;
 503         bio->bi_sector = sector;
 504         bio_add_page(bio, page, size, 0);
 505         init_completion(&event);
 506         bio->bi_private = &event;
 507         bio->bi_end_io = bi_complete;
 508         submit_bio(rw, bio);
 509         wait_for_completion(&event);
 510
 511         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 512         bio_put(bio);
 513         return ret;
 514 }
 515 EXPORT_SYMBOL_GPL(sync_page_io);
 516
 517 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 518 {
 519         char b[BDEVNAME_SIZE];
 520         if (!rdev->sb_page) {
 521                 MD_BUG();
 522                 return -EINVAL;
 523         }
 524         if (rdev->sb_loaded)
 525                 return 0;
 526
 527
 528         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 529                 goto fail;
 530         rdev->sb_loaded = 1;
 531         return 0;
 532
 533 fail:
 534         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 535                 bdevname(rdev->bdev,b));
 536         return -EINVAL;
 537 }
 538
 539 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 540 {
 541         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 542                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 543                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 544                 (sb1->set_uuid3 == sb2->set_uuid3))
 545
 546                 return 1;
 547
 548         return 0;
 549 }
 550
 551
 552 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 553 {
 554         int ret;
 555         mdp_super_t *tmp1, *tmp2;
 556
 557         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 558         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 559
 560         if (!tmp1 || !tmp2) {
 561                 ret = 0;
 562                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 563                 goto abort;
 564         }
 565
 566         *tmp1 = *sb1;
 567         *tmp2 = *sb2;
 568
 569         /*
 570          * nr_disks is not constant
 571          */
 572         tmp1->nr_disks = 0;
 573         tmp2->nr_disks = 0;
 574
 575         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 576                 ret = 0;
 577         else
 578                 ret = 1;
 579
 580 abort:
 581         kfree(tmp1);
 582         kfree(tmp2);
 583         return ret;
 584 }
 585
 586
 587 static u32 md_csum_fold(u32 csum)
 588 {
 589         csum = (csum & 0xffff) + (csum >> 16);
 590         return (csum & 0xffff) + (csum >> 16);
 591 }
 592
 593 static unsigned int calc_sb_csum(mdp_super_t * sb)
 594 {
 595         u64 newcsum = 0;
 596         u32 *sb32 = (u32*)sb;
 597         int i;
 598         unsigned int disk_csum, csum;
 599
 600         disk_csum = sb->sb_csum;
 601         sb->sb_csum = 0;
 602
 603         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 604                 newcsum += sb32[i];
 605         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 606
 607
 608 #ifdef CONFIG_ALPHA
 609         /* This used to use csum_partial, which was wrong for several
 610          * reasons including that different results are returned on
 611          * different architectures.  It isn't critical that we get exactly
 612          * the same return value as before (we always csum_fold before
 613          * testing, and that removes any differences).  However as we
 614          * know that csum_partial always returned a 16bit value on
 615          * alphas, do a fold to maximise conformity to previous behaviour.
 616          */
 617         sb->sb_csum = md_csum_fold(disk_csum);
 618 #else
 619         sb->sb_csum = disk_csum;
 620 #endif
 621         return csum;
 622 }
 623
 624
 625 /*
 626  * Handle superblock details.
 627  * We want to be able to handle multiple superblock formats
 628  * so we have a common interface to them all, and an array of
 629  * different handlers.
 630  * We rely on user-space to write the initial superblock, and support
 631  * reading and updating of superblocks.
 632  * Interface methods are:
 633  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 634  *      loads and validates a superblock on dev.
 635  *      if refdev != NULL, compare superblocks on both devices
 636  *    Return:
 637  *      0 - dev has a superblock that is compatible with refdev
 638  *      1 - dev has a superblock that is compatible and newer than refdev
 639  *          so dev should be used as the refdev in future
 640  *     -EINVAL superblock incompatible or invalid
 641  *     -othererror e.g. -EIO
 642  *
 643  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 644  *      Verify that dev is acceptable into mddev.
 645  *       The first time, mddev->raid_disks will be 0, and data from
 646  *       dev should be merged in.  Subsequent calls check that dev
 647  *       is new enough.  Return 0 or -EINVAL
 648  *
 649  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 650  *     Update the superblock for rdev with data in mddev
 651  *     This does not write to disc.
 652  *
 653  */
 654
 655 struct super_type  {
 656         char            *name;
 657         struct module   *owner;
 658         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 659         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 660         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 661 };
 662
 663 /*
 664  * load_super for 0.90.0
 665  */
 666 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 667 {
 668         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 669         mdp_super_t *sb;
 670         int ret;
 671         sector_t sb_offset;
 672
 673         /*
 674          * Calculate the position of the superblock,
 675          * it's at the end of the disk.
 676          *
 677          * It also happens to be a multiple of 4Kb.
 678          */
 679         sb_offset = calc_dev_sboffset(rdev->bdev);
 680         rdev->sb_offset = sb_offset;
 681
 682         ret = read_disk_sb(rdev, MD_SB_BYTES);
 683         if (ret) return ret;
 684
 685         ret = -EINVAL;
 686
 687         bdevname(rdev->bdev, b);
 688         sb = (mdp_super_t*)page_address(rdev->sb_page);
 689
 690         if (sb->md_magic != MD_SB_MAGIC) {
 691                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 692                        b);
 693                 goto abort;
 694         }
 695
 696         if (sb->major_version != 0 ||
 697             sb->minor_version < 90 ||
 698             sb->minor_version > 91) {
 699                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 700                         sb->major_version, sb->minor_version,
 701                         b);
 702                 goto abort;
 703         }
 704
 705         if (sb->raid_disks <= 0)
 706                 goto abort;
 707
 708         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 709                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 710                         b);
 711                 goto abort;
 712         }
 713
 714         rdev->preferred_minor = sb->md_minor;
 715         rdev->data_offset = 0;
 716         rdev->sb_size = MD_SB_BYTES;
 717
 718         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 719                 if (sb->level != 1 && sb->level != 4
 720                     && sb->level != 5 && sb->level != 6
 721                     && sb->level != 10) {
 722                         /* FIXME use a better test */
 723                         printk(KERN_WARNING
 724                                "md: bitmaps not supported for this level.\n");
 725                         goto abort;
 726                 }
 727         }
 728
 729         if (sb->level == LEVEL_MULTIPATH)
 730                 rdev->desc_nr = -1;
 731         else
 732                 rdev->desc_nr = sb->this_disk.number;
 733
 734         if (refdev == 0)
 735                 ret = 1;
 736         else {
 737                 __u64 ev1, ev2;
 738                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 739                 if (!uuid_equal(refsb, sb)) {
 740                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 741                                 b, bdevname(refdev->bdev,b2));
 742                         goto abort;
 743                 }
 744                 if (!sb_equal(refsb, sb)) {
 745                         printk(KERN_WARNING "md: %s has same UUID"
 746                                " but different superblock to %s\n",
 747                                b, bdevname(refdev->bdev, b2));
 748                         goto abort;
 749                 }
 750                 ev1 = md_event(sb);
 751                 ev2 = md_event(refsb);
 752                 if (ev1 > ev2)
 753                         ret = 1;
 754                 else
 755                         ret = 0;
 756         }
 757         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 758
 759         if (rdev->size < sb->size && sb->level > 1)
 760                 /* "this cannot possibly happen" ... */
 761                 ret = -EINVAL;
 762
 763  abort:
 764         return ret;
 765 }
 766
 767 /*
 768  * validate_super for 0.90.0
 769  */
 770 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 771 {
 772         mdp_disk_t *desc;
 773         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 774         __u64 ev1 = md_event(sb);
 775
 776         rdev->raid_disk = -1;
 777         clear_bit(Faulty, &rdev->flags);
 778         clear_bit(In_sync, &rdev->flags);
 779         clear_bit(WriteMostly, &rdev->flags);
 780         clear_bit(BarriersNotsupp, &rdev->flags);
 781
 782         if (mddev->raid_disks == 0) {
 783                 mddev->major_version = 0;
 784                 mddev->minor_version = sb->minor_version;
 785                 mddev->patch_version = sb->patch_version;
 786                 mddev->external = 0;
 787                 mddev->chunk_size = sb->chunk_size;
 788                 mddev->ctime = sb->ctime;
 789                 mddev->utime = sb->utime;
 790                 mddev->level = sb->level;
 791                 mddev->clevel[0] = 0;
 792                 mddev->layout = sb->layout;
 793                 mddev->raid_disks = sb->raid_disks;
 794                 mddev->size = sb->size;
 795                 mddev->events = ev1;
 796                 mddev->bitmap_offset = 0;
 797                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 798
 799                 if (mddev->minor_version >= 91) {
 800                         mddev->reshape_position = sb->reshape_position;
 801                         mddev->delta_disks = sb->delta_disks;
 802                         mddev->new_level = sb->new_level;
 803                         mddev->new_layout = sb->new_layout;
 804                         mddev->new_chunk = sb->new_chunk;
 805                 } else {
 806                         mddev->reshape_position = MaxSector;
 807                         mddev->delta_disks = 0;
 808                         mddev->new_level = mddev->level;
 809                         mddev->new_layout = mddev->layout;
 810                         mddev->new_chunk = mddev->chunk_size;
 811                 }
 812
 813                 if (sb->state & (1<<MD_SB_CLEAN))
 814                         mddev->recovery_cp = MaxSector;
 815                 else {
 816                         if (sb->events_hi == sb->cp_events_hi &&
 817                                 sb->events_lo == sb->cp_events_lo) {
 818                                 mddev->recovery_cp = sb->recovery_cp;
 819                         } else
 820                                 mddev->recovery_cp = 0;
 821                 }
 822
 823                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 824                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 825                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 826                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 827
 828                 mddev->max_disks = MD_SB_DISKS;
 829
 830                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 831                     mddev->bitmap_file == NULL)
 832                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 833
 834         } else if (mddev->pers == NULL) {
 835                 /* Insist on good event counter while assembling */
 836                 ++ev1;
 837                 if (ev1 < mddev->events)
 838                         return -EINVAL;
 839         } else if (mddev->bitmap) {
 840                 /* if adding to array with a bitmap, then we can accept an
 841                  * older device ... but not too old.
 842                  */
 843                 if (ev1 < mddev->bitmap->events_cleared)
 844                         return 0;
 845         } else {
 846                 if (ev1 < mddev->events)
 847                         /* just a hot-add of a new device, leave raid_disk at -1 */
 848                         return 0;
 849         }
 850
 851         if (mddev->level != LEVEL_MULTIPATH) {
 852                 desc = sb->disks + rdev->desc_nr;
 853
 854                 if (desc->state & (1<<MD_DISK_FAULTY))
 855                         set_bit(Faulty, &rdev->flags);
 856                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
 857                             desc->raid_disk < mddev->raid_disks */) {
 858                         set_bit(In_sync, &rdev->flags);
 859                         rdev->raid_disk = desc->raid_disk;
 860                 }
 861                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 862                         set_bit(WriteMostly, &rdev->flags);
 863         } else /* MULTIPATH are always insync */
 864                 set_bit(In_sync, &rdev->flags);
 865         return 0;
 866 }
 867
 868 /*
 869  * sync_super for 0.90.0
 870  */
 871 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 872 {
 873         mdp_super_t *sb;
 874         struct list_head *tmp;
 875         mdk_rdev_t *rdev2;
 876         int next_spare = mddev->raid_disks;
 877
 878
 879         /* make rdev->sb match mddev data..
 880          *
 881          * 1/ zero out disks
 882          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 883          * 3/ any empty disks < next_spare become removed
 884          *
 885          * disks[0] gets initialised to REMOVED because
 886          * we cannot be sure from other fields if it has
 887          * been initialised or not.
 888          */
 889         int i;
 890         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 891
 892         rdev->sb_size = MD_SB_BYTES;
 893
 894         sb = (mdp_super_t*)page_address(rdev->sb_page);
 895
 896         memset(sb, 0, sizeof(*sb));
 897
 898         sb->md_magic = MD_SB_MAGIC;
 899         sb->major_version = mddev->major_version;
 900         sb->patch_version = mddev->patch_version;
 901         sb->gvalid_words  = 0; /* ignored */
 902         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 903         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 904         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 905         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 906
 907         sb->ctime = mddev->ctime;
 908         sb->level = mddev->level;
 909         sb->size  = mddev->size;
 910         sb->raid_disks = mddev->raid_disks;
 911         sb->md_minor = mddev->md_minor;
 912         sb->not_persistent = 0;
 913         sb->utime = mddev->utime;
 914         sb->state = 0;
 915         sb->events_hi = (mddev->events>>32);
 916         sb->events_lo = (u32)mddev->events;
 917
 918         if (mddev->reshape_position == MaxSector)
 919                 sb->minor_version = 90;
 920         else {
 921                 sb->minor_version = 91;
 922                 sb->reshape_position = mddev->reshape_position;
 923                 sb->new_level = mddev->new_level;
 924                 sb->delta_disks = mddev->delta_disks;
 925                 sb->new_layout = mddev->new_layout;
 926                 sb->new_chunk = mddev->new_chunk;
 927         }
 928         mddev->minor_version = sb->minor_version;
 929         if (mddev->in_sync)
 930         {
 931                 sb->recovery_cp = mddev->recovery_cp;
 932                 sb->cp_events_hi = (mddev->events>>32);
 933                 sb->cp_events_lo = (u32)mddev->events;
 934                 if (mddev->recovery_cp == MaxSector)
 935                         sb->state = (1<< MD_SB_CLEAN);
 936         } else
 937                 sb->recovery_cp = 0;
 938
 939         sb->layout = mddev->layout;
 940         sb->chunk_size = mddev->chunk_size;
 941
 942         if (mddev->bitmap && mddev->bitmap_file == NULL)
 943                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 944
 945         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 946         rdev_for_each(rdev2, tmp, mddev) {
 947                 mdp_disk_t *d;
 948                 int desc_nr;
 949                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 950                     && !test_bit(Faulty, &rdev2->flags))
 951                         desc_nr = rdev2->raid_disk;
 952                 else
 953                         desc_nr = next_spare++;
 954                 rdev2->desc_nr = desc_nr;
 955                 d = &sb->disks[rdev2->desc_nr];
 956                 nr_disks++;
 957                 d->number = rdev2->desc_nr;
 958                 d->major = MAJOR(rdev2->bdev->bd_dev);
 959                 d->minor = MINOR(rdev2->bdev->bd_dev);
 960                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 961                     && !test_bit(Faulty, &rdev2->flags))
 962                         d->raid_disk = rdev2->raid_disk;
 963                 else
 964                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 965                 if (test_bit(Faulty, &rdev2->flags))
 966                         d->state = (1<<MD_DISK_FAULTY);
 967                 else if (test_bit(In_sync, &rdev2->flags)) {
 968                         d->state = (1<<MD_DISK_ACTIVE);
 969                         d->state |= (1<<MD_DISK_SYNC);
 970                         active++;
 971                         working++;
 972                 } else {
 973                         d->state = 0;
 974                         spare++;
 975                         working++;
 976                 }
 977                 if (test_bit(WriteMostly, &rdev2->flags))
 978                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 979         }
 980         /* now set the "removed" and "faulty" bits on any missing devices */
 981         for (i=0 ; i < mddev->raid_disks ; i++) {
 982                 mdp_disk_t *d = &sb->disks[i];
 983                 if (d->state == 0 && d->number == 0) {
 984                         d->number = i;
 985                         d->raid_disk = i;
 986                         d->state = (1<<MD_DISK_REMOVED);
 987                         d->state |= (1<<MD_DISK_FAULTY);
 988                         failed++;
 989                 }
 990         }
 991         sb->nr_disks = nr_disks;
 992         sb->active_disks = active;
 993         sb->working_disks = working;
 994         sb->failed_disks = failed;
 995         sb->spare_disks = spare;
 996
 997         sb->this_disk = sb->disks[rdev->desc_nr];
 998         sb->sb_csum = calc_sb_csum(sb);
 999 }
1000
1001 /*
1002  * version 1 superblock
1003  */
1004
1005 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1006 {
1007         __le32 disk_csum;
1008         u32 csum;
1009         unsigned long long newcsum;
1010         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1011         __le32 *isuper = (__le32*)sb;
1012         int i;
1013
1014         disk_csum = sb->sb_csum;
1015         sb->sb_csum = 0;
1016         newcsum = 0;
1017         for (i=0; size>=4; size -= 4 )
1018                 newcsum += le32_to_cpu(*isuper++);
1019
1020         if (size == 2)
1021                 newcsum += le16_to_cpu(*(__le16*) isuper);
1022
1023         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1024         sb->sb_csum = disk_csum;
1025         return cpu_to_le32(csum);
1026 }
1027
1028 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1029 {
1030         struct mdp_superblock_1 *sb;
1031         int ret;
1032         sector_t sb_offset;
1033         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1034         int bmask;
1035
1036         /*
1037          * Calculate the position of the superblock.
1038          * It is always aligned to a 4K boundary and
1039          * depeding on minor_version, it can be:
1040          * 0: At least 8K, but less than 12K, from end of device
1041          * 1: At start of device
1042          * 2: 4K from start of device.
1043          */
1044         switch(minor_version) {
1045         case 0:
1046                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1047                 sb_offset -= 8*2;
1048                 sb_offset &= ~(sector_t)(4*2-1);
1049                 /* convert from sectors to K */
1050                 sb_offset /= 2;
1051                 break;
1052         case 1:
1053                 sb_offset = 0;
1054                 break;
1055         case 2:
1056                 sb_offset = 4;
1057                 break;
1058         default:
1059                 return -EINVAL;
1060         }
1061         rdev->sb_offset = sb_offset;
1062
1063         /* superblock is rarely larger than 1K, but it can be larger,
1064          * and it is safe to read 4k, so we do that
1065          */
1066         ret = read_disk_sb(rdev, 4096);
1067         if (ret) return ret;
1068
1069
1070         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1071
1072         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1073             sb->major_version != cpu_to_le32(1) ||
1074             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1075             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1076             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1077                 return -EINVAL;
1078
1079         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1080                 printk("md: invalid superblock checksum on %s\n",
1081                         bdevname(rdev->bdev,b));
1082                 return -EINVAL;
1083         }
1084         if (le64_to_cpu(sb->data_size) < 10) {
1085                 printk("md: data_size too small on %s\n",
1086                        bdevname(rdev->bdev,b));
1087                 return -EINVAL;
1088         }
1089         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1090                 if (sb->level != cpu_to_le32(1) &&
1091                     sb->level != cpu_to_le32(4) &&
1092                     sb->level != cpu_to_le32(5) &&
1093                     sb->level != cpu_to_le32(6) &&
1094                     sb->level != cpu_to_le32(10)) {
1095                         printk(KERN_WARNING
1096                                "md: bitmaps not supported for this level.\n");
1097                         return -EINVAL;
1098                 }
1099         }
1100
1101         rdev->preferred_minor = 0xffff;
1102         rdev->data_offset = le64_to_cpu(sb->data_offset);
1103         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1104
1105         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1106         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1107         if (rdev->sb_size & bmask)
1108                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1109
1110         if (minor_version
1111             && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1112                 return -EINVAL;
1113
1114         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1115                 rdev->desc_nr = -1;
1116         else
1117                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1118
1119         if (refdev == 0)
1120                 ret = 1;
1121         else {
1122                 __u64 ev1, ev2;
1123                 struct mdp_superblock_1 *refsb =
1124                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1125
1126                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1127                     sb->level != refsb->level ||
1128                     sb->layout != refsb->layout ||
1129                     sb->chunksize != refsb->chunksize) {
1130                         printk(KERN_WARNING "md: %s has strangely different"
1131                                 " superblock to %s\n",
1132                                 bdevname(rdev->bdev,b),
1133                                 bdevname(refdev->bdev,b2));
1134                         return -EINVAL;
1135                 }
1136                 ev1 = le64_to_cpu(sb->events);
1137                 ev2 = le64_to_cpu(refsb->events);
1138
1139                 if (ev1 > ev2)
1140                         ret = 1;
1141                 else
1142                         ret = 0;
1143         }
1144         if (minor_version)
1145                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1146         else
1147                 rdev->size = rdev->sb_offset;
1148         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1149                 return -EINVAL;
1150         rdev->size = le64_to_cpu(sb->data_size)/2;
1151         if (le32_to_cpu(sb->chunksize))
1152                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1153
1154         if (le64_to_cpu(sb->size) > rdev->size*2)
1155                 return -EINVAL;
1156         return ret;
1157 }
1158
1159 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1160 {
1161         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1162         __u64 ev1 = le64_to_cpu(sb->events);
1163
1164         rdev->raid_disk = -1;
1165         clear_bit(Faulty, &rdev->flags);
1166         clear_bit(In_sync, &rdev->flags);
1167         clear_bit(WriteMostly, &rdev->flags);
1168         clear_bit(BarriersNotsupp, &rdev->flags);
1169
1170         if (mddev->raid_disks == 0) {
1171                 mddev->major_version = 1;
1172                 mddev->patch_version = 0;
1173                 mddev->external = 0;
1174                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1175                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1176                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1177                 mddev->level = le32_to_cpu(sb->level);
1178                 mddev->clevel[0] = 0;
1179                 mddev->layout = le32_to_cpu(sb->layout);
1180                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1181                 mddev->size = le64_to_cpu(sb->size)/2;
1182                 mddev->events = ev1;
1183                 mddev->bitmap_offset = 0;
1184                 mddev->default_bitmap_offset = 1024 >> 9;
1185
1186                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1187                 memcpy(mddev->uuid, sb->set_uuid, 16);
1188
1189                 mddev->max_disks =  (4096-256)/2;
1190
1191                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1192                     mddev->bitmap_file == NULL )
1193                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1194
1195                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1196                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1197                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1198                         mddev->new_level = le32_to_cpu(sb->new_level);
1199                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1200                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1201                 } else {
1202                         mddev->reshape_position = MaxSector;
1203                         mddev->delta_disks = 0;
1204                         mddev->new_level = mddev->level;
1205                         mddev->new_layout = mddev->layout;
1206                         mddev->new_chunk = mddev->chunk_size;
1207                 }
1208
1209         } else if (mddev->pers == NULL) {
1210                 /* Insist of good event counter while assembling */
1211                 ++ev1;
1212                 if (ev1 < mddev->events)
1213                         return -EINVAL;
1214         } else if (mddev->bitmap) {
1215                 /* If adding to array with a bitmap, then we can accept an
1216                  * older device, but not too old.
1217                  */
1218                 if (ev1 < mddev->bitmap->events_cleared)
1219                         return 0;
1220         } else {
1221                 if (ev1 < mddev->events)
1222                         /* just a hot-add of a new device, leave raid_disk at -1 */
1223                         return 0;
1224         }
1225         if (mddev->level != LEVEL_MULTIPATH) {
1226                 int role;
1227                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1228                 switch(role) {
1229                 case 0xffff: /* spare */
1230                         break;
1231                 case 0xfffe: /* faulty */
1232                         set_bit(Faulty, &rdev->flags);
1233                         break;
1234                 default:
1235                         if ((le32_to_cpu(sb->feature_map) &
1236                              MD_FEATURE_RECOVERY_OFFSET))
1237                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1238                         else
1239                                 set_bit(In_sync, &rdev->flags);
1240                         rdev->raid_disk = role;
1241                         break;
1242                 }
1243                 if (sb->devflags & WriteMostly1)
1244                         set_bit(WriteMostly, &rdev->flags);
1245         } else /* MULTIPATH are always insync */
1246                 set_bit(In_sync, &rdev->flags);
1247
1248         return 0;
1249 }
1250
1251 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1252 {
1253         struct mdp_superblock_1 *sb;
1254         struct list_head *tmp;
1255         mdk_rdev_t *rdev2;
1256         int max_dev, i;
1257         /* make rdev->sb match mddev and rdev data. */
1258
1259         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1260
1261         sb->feature_map = 0;
1262         sb->pad0 = 0;
1263         sb->recovery_offset = cpu_to_le64(0);
1264         memset(sb->pad1, 0, sizeof(sb->pad1));
1265         memset(sb->pad2, 0, sizeof(sb->pad2));
1266         memset(sb->pad3, 0, sizeof(sb->pad3));
1267
1268         sb->utime = cpu_to_le64((__u64)mddev->utime);
1269         sb->events = cpu_to_le64(mddev->events);
1270         if (mddev->in_sync)
1271                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1272         else
1273                 sb->resync_offset = cpu_to_le64(0);
1274
1275         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1276
1277         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1278         sb->size = cpu_to_le64(mddev->size<<1);
1279
1280         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1281                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1282                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1283         }
1284
1285         if (rdev->raid_disk >= 0 &&
1286             !test_bit(In_sync, &rdev->flags) &&
1287             rdev->recovery_offset > 0) {
1288                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1289                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1290         }
1291
1292         if (mddev->reshape_position != MaxSector) {
1293                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1294                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1295                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1296                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1297                 sb->new_level = cpu_to_le32(mddev->new_level);
1298                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1299         }
1300
1301         max_dev = 0;
1302         rdev_for_each(rdev2, tmp, mddev)
1303                 if (rdev2->desc_nr+1 > max_dev)
1304                         max_dev = rdev2->desc_nr+1;
1305
1306         if (max_dev > le32_to_cpu(sb->max_dev))
1307                 sb->max_dev = cpu_to_le32(max_dev);
1308         for (i=0; i<max_dev;i++)
1309                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1310
1311         rdev_for_each(rdev2, tmp, mddev) {
1312                 i = rdev2->desc_nr;
1313                 if (test_bit(Faulty, &rdev2->flags))
1314                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1315                 else if (test_bit(In_sync, &rdev2->flags))
1316                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1317                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1318                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1319                 else
1320                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1321         }
1322
1323         sb->sb_csum = calc_sb_1_csum(sb);
1324 }
1325
1326
1327 static struct super_type super_types[] = {
1328         [0] = {
1329                 .name   = "0.90.0",
1330                 .owner  = THIS_MODULE,
1331                 .load_super     = super_90_load,
1332                 .validate_super = super_90_validate,
1333                 .sync_super     = super_90_sync,
1334         },
1335         [1] = {
1336                 .name   = "md-1",
1337                 .owner  = THIS_MODULE,
1338                 .load_super     = super_1_load,
1339                 .validate_super = super_1_validate,
1340                 .sync_super     = super_1_sync,
1341         },
1342 };
1343
1344 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1345 {
1346         struct list_head *tmp, *tmp2;
1347         mdk_rdev_t *rdev, *rdev2;
1348
1349         rdev_for_each(rdev, tmp, mddev1)
1350                 rdev_for_each(rdev2, tmp2, mddev2)
1351                         if (rdev->bdev->bd_contains ==
1352                             rdev2->bdev->bd_contains)
1353                                 return 1;
1354
1355         return 0;
1356 }
1357
1358 static LIST_HEAD(pending_raid_disks);
1359
1360 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1361 {
1362         char b[BDEVNAME_SIZE];
1363         struct kobject *ko;
1364         char *s;
1365         int err;
1366
1367         if (rdev->mddev) {
1368                 MD_BUG();
1369                 return -EINVAL;
1370         }
1371         /* make sure rdev->size exceeds mddev->size */
1372         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1373                 if (mddev->pers) {
1374                         /* Cannot change size, so fail
1375                          * If mddev->level <= 0, then we don't care
1376                          * about aligning sizes (e.g. linear)
1377                          */
1378                         if (mddev->level > 0)
1379                                 return -ENOSPC;
1380                 } else
1381                         mddev->size = rdev->size;
1382         }
1383
1384         /* Verify rdev->desc_nr is unique.
1385          * If it is -1, assign a free number, else
1386          * check number is not in use
1387          */
1388         if (rdev->desc_nr < 0) {
1389                 int choice = 0;
1390                 if (mddev->pers) choice = mddev->raid_disks;
1391                 while (find_rdev_nr(mddev, choice))
1392                         choice++;
1393                 rdev->desc_nr = choice;
1394         } else {
1395                 if (find_rdev_nr(mddev, rdev->desc_nr))
1396                         return -EBUSY;
1397         }
1398         bdevname(rdev->bdev,b);
1399         while ( (s=strchr(b, '/')) != NULL)
1400                 *s = '!';
1401
1402         rdev->mddev = mddev;
1403         printk(KERN_INFO "md: bind<%s>\n", b);
1404
1405         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1406                 goto fail;
1407
1408         if (rdev->bdev->bd_part)
1409                 ko = &rdev->bdev->bd_part->dev.kobj;
1410         else
1411                 ko = &rdev->bdev->bd_disk->dev.kobj;
1412         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1413                 kobject_del(&rdev->kobj);
1414                 goto fail;
1415         }
1416         list_add(&rdev->same_set, &mddev->disks);
1417         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1418         return 0;
1419
1420  fail:
1421         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1422                b, mdname(mddev));
1423         return err;
1424 }
1425
1426 static void md_delayed_delete(struct work_struct *ws)
1427 {
1428         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1429         kobject_del(&rdev->kobj);
1430         kobject_put(&rdev->kobj);
1431 }
1432
1433 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1434 {
1435         char b[BDEVNAME_SIZE];
1436         if (!rdev->mddev) {
1437                 MD_BUG();
1438                 return;
1439         }
1440         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1441         list_del_init(&rdev->same_set);
1442         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1443         rdev->mddev = NULL;
1444         sysfs_remove_link(&rdev->kobj, "block");
1445
1446         /* We need to delay this, otherwise we can deadlock when
1447          * writing to 'remove' to "dev/state"
1448          */
1449         INIT_WORK(&rdev->del_work, md_delayed_delete);
1450         kobject_get(&rdev->kobj);
1451         schedule_work(&rdev->del_work);
1452 }
1453
1454 /*
1455  * prevent the device from being mounted, repartitioned or
1456  * otherwise reused by a RAID array (or any other kernel
1457  * subsystem), by bd_claiming the device.
1458  */
1459 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1460 {
1461         int err = 0;
1462         struct block_device *bdev;
1463         char b[BDEVNAME_SIZE];
1464
1465         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1466         if (IS_ERR(bdev)) {
1467                 printk(KERN_ERR "md: could not open %s.\n",
1468                         __bdevname(dev, b));
1469                 return PTR_ERR(bdev);
1470         }
1471         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1472         if (err) {
1473                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1474                         bdevname(bdev, b));
1475                 blkdev_put(bdev);
1476                 return err;
1477         }
1478         if (!shared)
1479                 set_bit(AllReserved, &rdev->flags);
1480         rdev->bdev = bdev;
1481         return err;
1482 }
1483
1484 static void unlock_rdev(mdk_rdev_t *rdev)
1485 {
1486         struct block_device *bdev = rdev->bdev;
1487         rdev->bdev = NULL;
1488         if (!bdev)
1489                 MD_BUG();
1490         bd_release(bdev);
1491         blkdev_put(bdev);
1492 }
1493
1494 void md_autodetect_dev(dev_t dev);
1495
1496 static void export_rdev(mdk_rdev_t * rdev)
1497 {
1498         char b[BDEVNAME_SIZE];
1499         printk(KERN_INFO "md: export_rdev(%s)\n",
1500                 bdevname(rdev->bdev,b));
1501         if (rdev->mddev)
1502                 MD_BUG();
1503         free_disk_sb(rdev);
1504         list_del_init(&rdev->same_set);
1505 #ifndef MODULE
1506         md_autodetect_dev(rdev->bdev->bd_dev);
1507 #endif
1508         unlock_rdev(rdev);
1509         kobject_put(&rdev->kobj);
1510 }
1511
1512 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1513 {
1514         unbind_rdev_from_array(rdev);
1515         export_rdev(rdev);
1516 }
1517
1518 static void export_array(mddev_t *mddev)
1519 {
1520         struct list_head *tmp;
1521         mdk_rdev_t *rdev;
1522
1523         rdev_for_each(rdev, tmp, mddev) {
1524                 if (!rdev->mddev) {
1525                         MD_BUG();
1526                         continue;
1527                 }
1528                 kick_rdev_from_array(rdev);
1529         }
1530         if (!list_empty(&mddev->disks))
1531                 MD_BUG();
1532         mddev->raid_disks = 0;
1533         mddev->major_version = 0;
1534 }
1535
1536 static void print_desc(mdp_disk_t *desc)
1537 {
1538         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1539                 desc->major,desc->minor,desc->raid_disk,desc->state);
1540 }
1541
1542 static void print_sb(mdp_super_t *sb)
1543 {
1544         int i;
1545
1546         printk(KERN_INFO
1547                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1548                 sb->major_version, sb->minor_version, sb->patch_version,
1549                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1550                 sb->ctime);
1551         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1552                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1553                 sb->md_minor, sb->layout, sb->chunk_size);
1554         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1555                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1556                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1557                 sb->failed_disks, sb->spare_disks,
1558                 sb->sb_csum, (unsigned long)sb->events_lo);
1559
1560         printk(KERN_INFO);
1561         for (i = 0; i < MD_SB_DISKS; i++) {
1562                 mdp_disk_t *desc;
1563
1564                 desc = sb->disks + i;
1565                 if (desc->number || desc->major || desc->minor ||
1566                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1567                         printk("     D %2d: ", i);
1568                         print_desc(desc);
1569                 }
1570         }
1571         printk(KERN_INFO "md:     THIS: ");
1572         print_desc(&sb->this_disk);
1573
1574 }
1575
1576 static void print_rdev(mdk_rdev_t *rdev)
1577 {
1578         char b[BDEVNAME_SIZE];
1579         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1580                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1581                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1582                 rdev->desc_nr);
1583         if (rdev->sb_loaded) {
1584                 printk(KERN_INFO "md: rdev superblock:\n");
1585                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1586         } else
1587                 printk(KERN_INFO "md: no rdev superblock!\n");
1588 }
1589
1590 static void md_print_devices(void)
1591 {
1592         struct list_head *tmp, *tmp2;
1593         mdk_rdev_t *rdev;
1594         mddev_t *mddev;
1595         char b[BDEVNAME_SIZE];
1596
1597         printk("\n");
1598         printk("md:     **********************************\n");
1599         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1600         printk("md:     **********************************\n");
1601         for_each_mddev(mddev, tmp) {
1602
1603                 if (mddev->bitmap)
1604                         bitmap_print_sb(mddev->bitmap);
1605                 else
1606                         printk("%s: ", mdname(mddev));
1607                 rdev_for_each(rdev, tmp2, mddev)
1608                         printk("<%s>", bdevname(rdev->bdev,b));
1609                 printk("\n");
1610
1611                 rdev_for_each(rdev, tmp2, mddev)
1612                         print_rdev(rdev);
1613         }
1614         printk("md:     **********************************\n");
1615         printk("\n");
1616 }
1617
1618
1619 static void sync_sbs(mddev_t * mddev, int nospares)
1620 {
1621         /* Update each superblock (in-memory image), but
1622          * if we are allowed to, skip spares which already
1623          * have the right event counter, or have one earlier
1624          * (which would mean they aren't being marked as dirty
1625          * with the rest of the array)
1626          */
1627         mdk_rdev_t *rdev;
1628         struct list_head *tmp;
1629
1630         rdev_for_each(rdev, tmp, mddev) {
1631                 if (rdev->sb_events == mddev->events ||
1632                     (nospares &&
1633                      rdev->raid_disk < 0 &&
1634                      (rdev->sb_events&1)==0 &&
1635                      rdev->sb_events+1 == mddev->events)) {
1636                         /* Don't update this superblock */
1637                         rdev->sb_loaded = 2;
1638                 } else {
1639                         super_types[mddev->major_version].
1640                                 sync_super(mddev, rdev);
1641                         rdev->sb_loaded = 1;
1642                 }
1643         }
1644 }
1645
1646 static void md_update_sb(mddev_t * mddev, int force_change)
1647 {
1648         struct list_head *tmp;
1649         mdk_rdev_t *rdev;
1650         int sync_req;
1651         int nospares = 0;
1652
1653 repeat:
1654         spin_lock_irq(&mddev->write_lock);
1655
1656         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1657         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1658                 force_change = 1;
1659         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1660                 /* just a clean<-> dirty transition, possibly leave spares alone,
1661                  * though if events isn't the right even/odd, we will have to do
1662                  * spares after all
1663                  */
1664                 nospares = 1;
1665         if (force_change)
1666                 nospares = 0;
1667         if (mddev->degraded)
1668                 /* If the array is degraded, then skipping spares is both
1669                  * dangerous and fairly pointless.
1670                  * Dangerous because a device that was removed from the array
1671                  * might have a event_count that still looks up-to-date,
1672                  * so it can be re-added without a resync.
1673                  * Pointless because if there are any spares to skip,
1674                  * then a recovery will happen and soon that array won't
1675                  * be degraded any more and the spare can go back to sleep then.
1676                  */
1677                 nospares = 0;
1678
1679         sync_req = mddev->in_sync;
1680         mddev->utime = get_seconds();
1681
1682         /* If this is just a dirty<->clean transition, and the array is clean
1683          * and 'events' is odd, we can roll back to the previous clean state */
1684         if (nospares
1685             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1686             && (mddev->events & 1)
1687             && mddev->events != 1)
1688                 mddev->events--;
1689         else {
1690                 /* otherwise we have to go forward and ... */
1691                 mddev->events ++;
1692                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1693                         /* .. if the array isn't clean, insist on an odd 'events' */
1694                         if ((mddev->events&1)==0) {
1695                                 mddev->events++;
1696                                 nospares = 0;
1697                         }
1698                 } else {
1699                         /* otherwise insist on an even 'events' (for clean states) */
1700                         if ((mddev->events&1)) {
1701                                 mddev->events++;
1702                                 nospares = 0;
1703                         }
1704                 }
1705         }
1706
1707         if (!mddev->events) {
1708                 /*
1709                  * oops, this 64-bit counter should never wrap.
1710                  * Either we are in around ~1 trillion A.C., assuming
1711                  * 1 reboot per second, or we have a bug:
1712                  */
1713                 MD_BUG();
1714                 mddev->events --;
1715         }
1716
1717         /*
1718          * do not write anything to disk if using
1719          * nonpersistent superblocks
1720          */
1721         if (!mddev->persistent) {
1722                 if (!mddev->external)
1723                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1724
1725                 spin_unlock_irq(&mddev->write_lock);
1726                 wake_up(&mddev->sb_wait);
1727                 return;
1728         }
1729         sync_sbs(mddev, nospares);
1730         spin_unlock_irq(&mddev->write_lock);
1731
1732         dprintk(KERN_INFO
1733                 "md: updating %s RAID superblock on device (in sync %d)\n",
1734                 mdname(mddev),mddev->in_sync);
1735
1736         bitmap_update_sb(mddev->bitmap);
1737         rdev_for_each(rdev, tmp, mddev) {
1738                 char b[BDEVNAME_SIZE];
1739                 dprintk(KERN_INFO "md: ");
1740                 if (rdev->sb_loaded != 1)
1741                         continue; /* no noise on spare devices */
1742                 if (test_bit(Faulty, &rdev->flags))
1743                         dprintk("(skipping faulty ");
1744
1745                 dprintk("%s ", bdevname(rdev->bdev,b));
1746                 if (!test_bit(Faulty, &rdev->flags)) {
1747                         md_super_write(mddev,rdev,
1748                                        rdev->sb_offset<<1, rdev->sb_size,
1749                                        rdev->sb_page);
1750                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1751                                 bdevname(rdev->bdev,b),
1752                                 (unsigned long long)rdev->sb_offset);
1753                         rdev->sb_events = mddev->events;
1754
1755                 } else
1756                         dprintk(")\n");
1757                 if (mddev->level == LEVEL_MULTIPATH)
1758                         /* only need to write one superblock... */
1759                         break;
1760         }
1761         md_super_wait(mddev);
1762         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1763
1764         spin_lock_irq(&mddev->write_lock);
1765         if (mddev->in_sync != sync_req ||
1766             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1767                 /* have to write it out again */
1768                 spin_unlock_irq(&mddev->write_lock);
1769                 goto repeat;
1770         }
1771         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1772         spin_unlock_irq(&mddev->write_lock);
1773         wake_up(&mddev->sb_wait);
1774
1775 }
1776
1777 /* words written to sysfs files may, or my not, be \n terminated.
1778  * We want to accept with case. For this we use cmd_match.
1779  */
1780 static int cmd_match(const char *cmd, const char *str)
1781 {
1782         /* See if cmd, written into a sysfs file, matches
1783          * str.  They must either be the same, or cmd can
1784          * have a trailing newline
1785          */
1786         while (*cmd && *str && *cmd == *str) {
1787                 cmd++;
1788                 str++;
1789         }
1790         if (*cmd == '\n')
1791                 cmd++;
1792         if (*str || *cmd)
1793                 return 0;
1794         return 1;
1795 }
1796
1797 struct rdev_sysfs_entry {
1798         struct attribute attr;
1799         ssize_t (*show)(mdk_rdev_t *, char *);
1800         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1801 };
1802
1803 static ssize_t
1804 state_show(mdk_rdev_t *rdev, char *page)
1805 {
1806         char *sep = "";
1807         size_t len = 0;
1808
1809         if (test_bit(Faulty, &rdev->flags)) {
1810                 len+= sprintf(page+len, "%sfaulty",sep);
1811                 sep = ",";
1812         }
1813         if (test_bit(In_sync, &rdev->flags)) {
1814                 len += sprintf(page+len, "%sin_sync",sep);
1815                 sep = ",";
1816         }
1817         if (test_bit(WriteMostly, &rdev->flags)) {
1818                 len += sprintf(page+len, "%swrite_mostly",sep);
1819                 sep = ",";
1820         }
1821         if (!test_bit(Faulty, &rdev->flags) &&
1822             !test_bit(In_sync, &rdev->flags)) {
1823                 len += sprintf(page+len, "%sspare", sep);
1824                 sep = ",";
1825         }
1826         return len+sprintf(page+len, "\n");
1827 }
1828
1829 static ssize_t
1830 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1831 {
1832         /* can write
1833          *  faulty  - simulates and error
1834          *  remove  - disconnects the device
1835          *  writemostly - sets write_mostly
1836          *  -writemostly - clears write_mostly
1837          */
1838         int err = -EINVAL;
1839         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1840                 md_error(rdev->mddev, rdev);
1841                 err = 0;
1842         } else if (cmd_match(buf, "remove")) {
1843                 if (rdev->raid_disk >= 0)
1844                         err = -EBUSY;
1845                 else {
1846                         mddev_t *mddev = rdev->mddev;
1847                         kick_rdev_from_array(rdev);
1848                         if (mddev->pers)
1849                                 md_update_sb(mddev, 1);
1850                         md_new_event(mddev);
1851                         err = 0;
1852                 }
1853         } else if (cmd_match(buf, "writemostly")) {
1854                 set_bit(WriteMostly, &rdev->flags);
1855                 err = 0;
1856         } else if (cmd_match(buf, "-writemostly")) {
1857                 clear_bit(WriteMostly, &rdev->flags);
1858                 err = 0;
1859         }
1860         return err ? err : len;
1861 }
1862 static struct rdev_sysfs_entry rdev_state =
1863 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1864
1865 static ssize_t
1866 super_show(mdk_rdev_t *rdev, char *page)
1867 {
1868         if (rdev->sb_loaded && rdev->sb_size) {
1869                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1870                 return rdev->sb_size;
1871         } else
1872                 return 0;
1873 }
1874 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1875
1876 static ssize_t
1877 errors_show(mdk_rdev_t *rdev, char *page)
1878 {
1879         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1880 }
1881
1882 static ssize_t
1883 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1884 {
1885         char *e;
1886         unsigned long n = simple_strtoul(buf, &e, 10);
1887         if (*buf && (*e == 0 || *e == '\n')) {
1888                 atomic_set(&rdev->corrected_errors, n);
1889                 return len;
1890         }
1891         return -EINVAL;
1892 }
1893 static struct rdev_sysfs_entry rdev_errors =
1894 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1895
1896 static ssize_t
1897 slot_show(mdk_rdev_t *rdev, char *page)
1898 {
1899         if (rdev->raid_disk < 0)
1900                 return sprintf(page, "none\n");
1901         else
1902                 return sprintf(page, "%d\n", rdev->raid_disk);
1903 }
1904
1905 static ssize_t
1906 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1907 {
1908         char *e;
1909         int err;
1910         char nm[20];
1911         int slot = simple_strtoul(buf, &e, 10);
1912         if (strncmp(buf, "none", 4)==0)
1913                 slot = -1;
1914         else if (e==buf || (*e && *e!= '\n'))
1915                 return -EINVAL;
1916         if (rdev->mddev->pers) {
1917                 /* Setting 'slot' on an active array requires also
1918                  * updating the 'rd%d' link, and communicating
1919                  * with the personality with ->hot_*_disk.
1920                  * For now we only support removing
1921                  * failed/spare devices.  This normally happens automatically,
1922                  * but not when the metadata is externally managed.
1923                  */
1924                 if (slot != -1)
1925                         return -EBUSY;
1926                 if (rdev->raid_disk == -1)
1927                         return -EEXIST;
1928                 /* personality does all needed checks */
1929                 if (rdev->mddev->pers->hot_add_disk == NULL)
1930                         return -EINVAL;
1931                 err = rdev->mddev->pers->
1932                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
1933                 if (err)
1934                         return err;
1935                 sprintf(nm, "rd%d", rdev->raid_disk);
1936                 sysfs_remove_link(&rdev->mddev->kobj, nm);
1937                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1938                 md_wakeup_thread(rdev->mddev->thread);
1939         } else {
1940                 if (slot >= rdev->mddev->raid_disks)
1941                         return -ENOSPC;
1942                 rdev->raid_disk = slot;
1943                 /* assume it is working */
1944                 clear_bit(Faulty, &rdev->flags);
1945                 clear_bit(WriteMostly, &rdev->flags);
1946                 set_bit(In_sync, &rdev->flags);
1947         }
1948         return len;
1949 }
1950
1951
1952 static struct rdev_sysfs_entry rdev_slot =
1953 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1954
1955 static ssize_t
1956 offset_show(mdk_rdev_t *rdev, char *page)
1957 {
1958         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1959 }
1960
1961 static ssize_t
1962 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1963 {
1964         char *e;
1965         unsigned long long offset = simple_strtoull(buf, &e, 10);
1966         if (e==buf || (*e && *e != '\n'))
1967                 return -EINVAL;
1968         if (rdev->mddev->pers)
1969                 return -EBUSY;
1970         if (rdev->size && rdev->mddev->external)
1971                 /* Must set offset before size, so overlap checks
1972                  * can be sane */
1973                 return -EBUSY;
1974         rdev->data_offset = offset;
1975         return len;
1976 }
1977
1978 static struct rdev_sysfs_entry rdev_offset =
1979 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1980
1981 static ssize_t
1982 rdev_size_show(mdk_rdev_t *rdev, char *page)
1983 {
1984         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1985 }
1986
1987 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1988 {
1989         /* check if two start/length pairs overlap */
1990         if (s1+l1 <= s2)
1991                 return 0;
1992         if (s2+l2 <= s1)
1993                 return 0;
1994         return 1;
1995 }
1996
1997 static ssize_t
1998 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1999 {
2000         char *e;
2001         unsigned long long size = simple_strtoull(buf, &e, 10);
2002         unsigned long long oldsize = rdev->size;
2003         if (e==buf || (*e && *e != '\n'))
2004                 return -EINVAL;
2005         if (rdev->mddev->pers)
2006                 return -EBUSY;
2007         rdev->size = size;
2008         if (size > oldsize && rdev->mddev->external) {
2009                 /* need to check that all other rdevs with the same ->bdev
2010                  * do not overlap.  We need to unlock the mddev to avoid
2011                  * a deadlock.  We have already changed rdev->size, and if
2012                  * we have to change it back, we will have the lock again.
2013                  */
2014                 mddev_t *mddev;
2015                 int overlap = 0;
2016                 struct list_head *tmp, *tmp2;
2017
2018                 mddev_unlock(rdev->mddev);
2019                 for_each_mddev(mddev, tmp) {
2020                         mdk_rdev_t *rdev2;
2021
2022                         mddev_lock(mddev);
2023                         rdev_for_each(rdev2, tmp2, mddev)
2024                                 if (test_bit(AllReserved, &rdev2->flags) ||
2025                                     (rdev->bdev == rdev2->bdev &&
2026                                      rdev != rdev2 &&
2027                                      overlaps(rdev->data_offset, rdev->size,
2028                                             rdev2->data_offset, rdev2->size))) {
2029                                         overlap = 1;
2030                                         break;
2031                                 }
2032                         mddev_unlock(mddev);
2033                         if (overlap) {
2034                                 mddev_put(mddev);
2035                                 break;
2036                         }
2037                 }
2038                 mddev_lock(rdev->mddev);
2039                 if (overlap) {
2040                         /* Someone else could have slipped in a size
2041                          * change here, but doing so is just silly.
2042                          * We put oldsize back because we *know* it is
2043                          * safe, and trust userspace not to race with
2044                          * itself
2045                          */
2046                         rdev->size = oldsize;
2047                         return -EBUSY;
2048                 }
2049         }
2050         if (size < rdev->mddev->size || rdev->mddev->size == 0)
2051                 rdev->mddev->size = size;
2052         return len;
2053 }
2054
2055 static struct rdev_sysfs_entry rdev_size =
2056 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2057
2058 static struct attribute *rdev_default_attrs[] = {
2059         &rdev_state.attr,
2060         &rdev_super.attr,
2061         &rdev_errors.attr,
2062         &rdev_slot.attr,
2063         &rdev_offset.attr,
2064         &rdev_size.attr,
2065         NULL,
2066 };
2067 static ssize_t
2068 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2069 {
2070         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2071         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2072
2073         if (!entry->show)
2074                 return -EIO;
2075         return entry->show(rdev, page);
2076 }
2077
2078 static ssize_t
2079 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2080               const char *page, size_t length)
2081 {
2082         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2083         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2084         int rv;
2085
2086         if (!entry->store)
2087                 return -EIO;
2088         if (!capable(CAP_SYS_ADMIN))
2089                 return -EACCES;
2090         rv = mddev_lock(rdev->mddev);
2091         if (!rv) {
2092                 rv = entry->store(rdev, page, length);
2093                 mddev_unlock(rdev->mddev);
2094         }
2095         return rv;
2096 }
2097
2098 static void rdev_free(struct kobject *ko)
2099 {
2100         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2101         kfree(rdev);
2102 }
2103 static struct sysfs_ops rdev_sysfs_ops = {
2104         .show           = rdev_attr_show,
2105         .store          = rdev_attr_store,
2106 };
2107 static struct kobj_type rdev_ktype = {
2108         .release        = rdev_free,
2109         .sysfs_ops      = &rdev_sysfs_ops,
2110         .default_attrs  = rdev_default_attrs,
2111 };
2112
2113 /*
2114  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2115  *
2116  * mark the device faulty if:
2117  *
2118  *   - the device is nonexistent (zero size)
2119  *   - the device has no valid superblock
2120  *
2121  * a faulty rdev _never_ has rdev->sb set.
2122  */
2123 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2124 {
2125         char b[BDEVNAME_SIZE];
2126         int err;
2127         mdk_rdev_t *rdev;
2128         sector_t size;
2129
2130         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2131         if (!rdev) {
2132                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2133                 return ERR_PTR(-ENOMEM);
2134         }
2135
2136         if ((err = alloc_disk_sb(rdev)))
2137                 goto abort_free;
2138
2139         err = lock_rdev(rdev, newdev, super_format == -2);
2140         if (err)
2141                 goto abort_free;
2142
2143         kobject_init(&rdev->kobj, &rdev_ktype);
2144
2145         rdev->desc_nr = -1;
2146         rdev->saved_raid_disk = -1;
2147         rdev->raid_disk = -1;
2148         rdev->flags = 0;
2149         rdev->data_offset = 0;
2150         rdev->sb_events = 0;
2151         atomic_set(&rdev->nr_pending, 0);
2152         atomic_set(&rdev->read_errors, 0);
2153         atomic_set(&rdev->corrected_errors, 0);
2154
2155         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2156         if (!size) {
2157                 printk(KERN_WARNING
2158                         "md: %s has zero or unknown size, marking faulty!\n",
2159                         bdevname(rdev->bdev,b));
2160                 err = -EINVAL;
2161                 goto abort_free;
2162         }
2163
2164         if (super_format >= 0) {
2165                 err = super_types[super_format].
2166                         load_super(rdev, NULL, super_minor);
2167                 if (err == -EINVAL) {
2168                         printk(KERN_WARNING
2169                                 "md: %s does not have a valid v%d.%d "
2170                                "superblock, not importing!\n",
2171                                 bdevname(rdev->bdev,b),
2172                                super_format, super_minor);
2173                         goto abort_free;
2174                 }
2175                 if (err < 0) {
2176                         printk(KERN_WARNING
2177                                 "md: could not read %s's sb, not importing!\n",
2178                                 bdevname(rdev->bdev,b));
2179                         goto abort_free;
2180                 }
2181         }
2182         INIT_LIST_HEAD(&rdev->same_set);
2183
2184         return rdev;
2185
2186 abort_free:
2187         if (rdev->sb_page) {
2188                 if (rdev->bdev)
2189                         unlock_rdev(rdev);
2190                 free_disk_sb(rdev);
2191         }
2192         kfree(rdev);
2193         return ERR_PTR(err);
2194 }
2195
2196 /*
2197  * Check a full RAID array for plausibility
2198  */
2199
2200
2201 static void analyze_sbs(mddev_t * mddev)
2202 {
2203         int i;
2204         struct list_head *tmp;
2205         mdk_rdev_t *rdev, *freshest;
2206         char b[BDEVNAME_SIZE];
2207
2208         freshest = NULL;
2209         rdev_for_each(rdev, tmp, mddev)
2210                 switch (super_types[mddev->major_version].
2211                         load_super(rdev, freshest, mddev->minor_version)) {
2212                 case 1:
2213                         freshest = rdev;
2214                         break;
2215                 case 0:
2216                         break;
2217                 default:
2218                         printk( KERN_ERR \
2219                                 "md: fatal superblock inconsistency in %s"
2220                                 " -- removing from array\n",
2221                                 bdevname(rdev->bdev,b));
2222                         kick_rdev_from_array(rdev);
2223                 }
2224
2225
2226         super_types[mddev->major_version].
2227                 validate_super(mddev, freshest);
2228
2229         i = 0;
2230         rdev_for_each(rdev, tmp, mddev) {
2231                 if (rdev != freshest)
2232                         if (super_types[mddev->major_version].
2233                             validate_super(mddev, rdev)) {
2234                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2235                                         " from array!\n",
2236                                         bdevname(rdev->bdev,b));
2237                                 kick_rdev_from_array(rdev);
2238                                 continue;
2239                         }
2240                 if (mddev->level == LEVEL_MULTIPATH) {
2241                         rdev->desc_nr = i++;
2242                         rdev->raid_disk = rdev->desc_nr;
2243                         set_bit(In_sync, &rdev->flags);
2244                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2245                         rdev->raid_disk = -1;
2246                         clear_bit(In_sync, &rdev->flags);
2247                 }
2248         }
2249
2250
2251
2252         if (mddev->recovery_cp != MaxSector &&
2253             mddev->level >= 1)
2254                 printk(KERN_ERR "md: %s: raid array is not clean"
2255                        " -- starting background reconstruction\n",
2256                        mdname(mddev));
2257
2258 }
2259
2260 static ssize_t
2261 safe_delay_show(mddev_t *mddev, char *page)
2262 {
2263         int msec = (mddev->safemode_delay*1000)/HZ;
2264         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2265 }
2266 static ssize_t
2267 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2268 {
2269         int scale=1;
2270         int dot=0;
2271         int i;
2272         unsigned long msec;
2273         char buf[30];
2274         char *e;
2275         /* remove a period, and count digits after it */
2276         if (len >= sizeof(buf))
2277                 return -EINVAL;
2278         strlcpy(buf, cbuf, len);
2279         buf[len] = 0;
2280         for (i=0; i<len; i++) {
2281                 if (dot) {
2282                         if (isdigit(buf[i])) {
2283                                 buf[i-1] = buf[i];
2284                                 scale *= 10;
2285                         }
2286                         buf[i] = 0;
2287                 } else if (buf[i] == '.') {
2288                         dot=1;
2289                         buf[i] = 0;
2290                 }
2291         }
2292         msec = simple_strtoul(buf, &e, 10);
2293         if (e == buf || (*e && *e != '\n'))
2294                 return -EINVAL;
2295         msec = (msec * 1000) / scale;
2296         if (msec == 0)
2297                 mddev->safemode_delay = 0;
2298         else {
2299                 mddev->safemode_delay = (msec*HZ)/1000;
2300                 if (mddev->safemode_delay == 0)
2301                         mddev->safemode_delay = 1;
2302         }
2303         return len;
2304 }
2305 static struct md_sysfs_entry md_safe_delay =
2306 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2307
2308 static ssize_t
2309 level_show(mddev_t *mddev, char *page)
2310 {
2311         struct mdk_personality *p = mddev->pers;
2312         if (p)
2313                 return sprintf(page, "%s\n", p->name);
2314         else if (mddev->clevel[0])
2315                 return sprintf(page, "%s\n", mddev->clevel);
2316         else if (mddev->level != LEVEL_NONE)
2317                 return sprintf(page, "%d\n", mddev->level);
2318         else
2319                 return 0;
2320 }
2321
2322 static ssize_t
2323 level_store(mddev_t *mddev, const char *buf, size_t len)
2324 {
2325         ssize_t rv = len;
2326         if (mddev->pers)
2327                 return -EBUSY;
2328         if (len == 0)
2329                 return 0;
2330         if (len >= sizeof(mddev->clevel))
2331                 return -ENOSPC;
2332         strncpy(mddev->clevel, buf, len);
2333         if (mddev->clevel[len-1] == '\n')
2334                 len--;
2335         mddev->clevel[len] = 0;
2336         mddev->level = LEVEL_NONE;
2337         return rv;
2338 }
2339
2340 static struct md_sysfs_entry md_level =
2341 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2342
2343
2344 static ssize_t
2345 layout_show(mddev_t *mddev, char *page)
2346 {
2347         /* just a number, not meaningful for all levels */
2348         if (mddev->reshape_position != MaxSector &&
2349             mddev->layout != mddev->new_layout)
2350                 return sprintf(page, "%d (%d)\n",
2351                                mddev->new_layout, mddev->layout);
2352         return sprintf(page, "%d\n", mddev->layout);
2353 }
2354
2355 static ssize_t
2356 layout_store(mddev_t *mddev, const char *buf, size_t len)
2357 {
2358         char *e;
2359         unsigned long n = simple_strtoul(buf, &e, 10);
2360
2361         if (!*buf || (*e && *e != '\n'))
2362                 return -EINVAL;
2363
2364         if (mddev->pers)
2365                 return -EBUSY;
2366         if (mddev->reshape_position != MaxSector)
2367                 mddev->new_layout = n;
2368         else
2369                 mddev->layout = n;
2370         return len;
2371 }
2372 static struct md_sysfs_entry md_layout =
2373 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2374
2375
2376 static ssize_t
2377 raid_disks_show(mddev_t *mddev, char *page)
2378 {
2379         if (mddev->raid_disks == 0)
2380                 return 0;
2381         if (mddev->reshape_position != MaxSector &&
2382             mddev->delta_disks != 0)
2383                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2384                                mddev->raid_disks - mddev->delta_disks);
2385         return sprintf(page, "%d\n", mddev->raid_disks);
2386 }
2387
2388 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2389
2390 static ssize_t
2391 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2392 {
2393         char *e;
2394         int rv = 0;
2395         unsigned long n = simple_strtoul(buf, &e, 10);
2396
2397         if (!*buf || (*e && *e != '\n'))
2398                 return -EINVAL;
2399
2400         if (mddev->pers)
2401                 rv = update_raid_disks(mddev, n);
2402         else if (mddev->reshape_position != MaxSector) {
2403                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2404                 mddev->delta_disks = n - olddisks;
2405                 mddev->raid_disks = n;
2406         } else
2407                 mddev->raid_disks = n;
2408         return rv ? rv : len;
2409 }
2410 static struct md_sysfs_entry md_raid_disks =
2411 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2412
2413 static ssize_t
2414 chunk_size_show(mddev_t *mddev, char *page)
2415 {
2416         if (mddev->reshape_position != MaxSector &&
2417             mddev->chunk_size != mddev->new_chunk)
2418                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2419                                mddev->chunk_size);
2420         return sprintf(page, "%d\n", mddev->chunk_size);
2421 }
2422
2423 static ssize_t
2424 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2425 {
2426         /* can only set chunk_size if array is not yet active */
2427         char *e;
2428         unsigned long n = simple_strtoul(buf, &e, 10);
2429
2430         if (!*buf || (*e && *e != '\n'))
2431                 return -EINVAL;
2432
2433         if (mddev->pers)
2434                 return -EBUSY;
2435         else if (mddev->reshape_position != MaxSector)
2436                 mddev->new_chunk = n;
2437         else
2438                 mddev->chunk_size = n;
2439         return len;
2440 }
2441 static struct md_sysfs_entry md_chunk_size =
2442 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2443
2444 static ssize_t
2445 resync_start_show(mddev_t *mddev, char *page)
2446 {
2447         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2448 }
2449
2450 static ssize_t
2451 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2452 {
2453         /* can only set chunk_size if array is not yet active */
2454         char *e;
2455         unsigned long long n = simple_strtoull(buf, &e, 10);
2456
2457         if (mddev->pers)
2458                 return -EBUSY;
2459         if (!*buf || (*e && *e != '\n'))
2460                 return -EINVAL;
2461
2462         mddev->recovery_cp = n;
2463         return len;
2464 }
2465 static struct md_sysfs_entry md_resync_start =
2466 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2467
2468 /*
2469  * The array state can be:
2470  *
2471  * clear
2472  *     No devices, no size, no level
2473  *     Equivalent to STOP_ARRAY ioctl
2474  * inactive
2475  *     May have some settings, but array is not active
2476  *        all IO results in error
2477  *     When written, doesn't tear down array, but just stops it
2478  * suspended (not supported yet)
2479  *     All IO requests will block. The array can be reconfigured.
2480  *     Writing this, if accepted, will block until array is quiessent
2481  * readonly
2482  *     no resync can happen.  no superblocks get written.
2483  *     write requests fail
2484  * read-auto
2485  *     like readonly, but behaves like 'clean' on a write request.
2486  *
2487  * clean - no pending writes, but otherwise active.
2488  *     When written to inactive array, starts without resync
2489  *     If a write request arrives then
2490  *       if metadata is known, mark 'dirty' and switch to 'active'.
2491  *       if not known, block and switch to write-pending
2492  *     If written to an active array that has pending writes, then fails.
2493  * active
2494  *     fully active: IO and resync can be happening.
2495  *     When written to inactive array, starts with resync
2496  *
2497  * write-pending
2498  *     clean, but writes are blocked waiting for 'active' to be written.
2499  *
2500  * active-idle
2501  *     like active, but no writes have been seen for a while (100msec).
2502  *
2503  */
2504 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2505                    write_pending, active_idle, bad_word};
2506 static char *array_states[] = {
2507         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2508         "write-pending", "active-idle", NULL };
2509
2510 static int match_word(const char *word, char **list)
2511 {
2512         int n;
2513         for (n=0; list[n]; n++)
2514                 if (cmd_match(word, list[n]))
2515                         break;
2516         return n;
2517 }
2518
2519 static ssize_t
2520 array_state_show(mddev_t *mddev, char *page)
2521 {
2522         enum array_state st = inactive;
2523
2524         if (mddev->pers)
2525                 switch(mddev->ro) {
2526                 case 1:
2527                         st = readonly;
2528                         break;
2529                 case 2:
2530                         st = read_auto;
2531                         break;
2532                 case 0:
2533                         if (mddev->in_sync)
2534                                 st = clean;
2535                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2536                                 st = write_pending;
2537                         else if (mddev->safemode)
2538                                 st = active_idle;
2539                         else
2540                                 st = active;
2541                 }
2542         else {
2543                 if (list_empty(&mddev->disks) &&
2544                     mddev->raid_disks == 0 &&
2545                     mddev->size == 0)
2546                         st = clear;
2547                 else
2548                         st = inactive;
2549         }
2550         return sprintf(page, "%s\n", array_states[st]);
2551 }
2552
2553 static int do_md_stop(mddev_t * mddev, int ro);
2554 static int do_md_run(mddev_t * mddev);
2555 static int restart_array(mddev_t *mddev);
2556
2557 static ssize_t
2558 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2559 {
2560         int err = -EINVAL;
2561         enum array_state st = match_word(buf, array_states);
2562         switch(st) {
2563         case bad_word:
2564                 break;
2565         case clear:
2566                 /* stopping an active array */
2567                 if (atomic_read(&mddev->active) > 1)
2568                         return -EBUSY;
2569                 err = do_md_stop(mddev, 0);
2570                 break;
2571         case inactive:
2572                 /* stopping an active array */
2573                 if (mddev->pers) {
2574                         if (atomic_read(&mddev->active) > 1)
2575                                 return -EBUSY;
2576                         err = do_md_stop(mddev, 2);
2577                 } else
2578                         err = 0; /* already inactive */
2579                 break;
2580         case suspended:
2581                 break; /* not supported yet */
2582         case readonly:
2583                 if (mddev->pers)
2584                         err = do_md_stop(mddev, 1);
2585                 else {
2586                         mddev->ro = 1;
2587                         err = do_md_run(mddev);
2588                 }
2589                 break;
2590         case read_auto:
2591                 /* stopping an active array */
2592                 if (mddev->pers) {
2593                         err = do_md_stop(mddev, 1);
2594                         if (err == 0)
2595                                 mddev->ro = 2; /* FIXME mark devices writable */
2596                 } else {
2597                         mddev->ro = 2;
2598                         err = do_md_run(mddev);
2599                 }
2600                 break;
2601         case clean:
2602                 if (mddev->pers) {
2603                         restart_array(mddev);
2604                         spin_lock_irq(&mddev->write_lock);
2605                         if (atomic_read(&mddev->writes_pending) == 0) {
2606                                 if (mddev->in_sync == 0) {
2607                                         mddev->in_sync = 1;
2608                                         if (mddev->persistent)
2609                                                 set_bit(MD_CHANGE_CLEAN,
2610                                                         &mddev->flags);
2611                                 }
2612                                 err = 0;
2613                         } else
2614                                 err = -EBUSY;
2615                         spin_unlock_irq(&mddev->write_lock);
2616                 } else {
2617                         mddev->ro = 0;
2618                         mddev->recovery_cp = MaxSector;
2619                         err = do_md_run(mddev);
2620                 }
2621                 break;
2622         case active:
2623                 if (mddev->pers) {
2624                         restart_array(mddev);
2625                         if (mddev->external)
2626                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2627                         wake_up(&mddev->sb_wait);
2628                         err = 0;
2629                 } else {
2630                         mddev->ro = 0;
2631                         err = do_md_run(mddev);
2632                 }
2633                 break;
2634         case write_pending:
2635         case active_idle:
2636                 /* these cannot be set */
2637                 break;
2638         }
2639         if (err)
2640                 return err;
2641         else
2642                 return len;
2643 }
2644 static struct md_sysfs_entry md_array_state =
2645 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2646
2647 static ssize_t
2648 null_show(mddev_t *mddev, char *page)
2649 {
2650         return -EINVAL;
2651 }
2652
2653 static ssize_t
2654 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2655 {
2656         /* buf must be %d:%d\n? giving major and minor numbers */
2657         /* The new device is added to the array.
2658          * If the array has a persistent superblock, we read the
2659          * superblock to initialise info and check validity.
2660          * Otherwise, only checking done is that in bind_rdev_to_array,
2661          * which mainly checks size.
2662          */
2663         char *e;
2664         int major = simple_strtoul(buf, &e, 10);
2665         int minor;
2666         dev_t dev;
2667         mdk_rdev_t *rdev;
2668         int err;
2669
2670         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2671                 return -EINVAL;
2672         minor = simple_strtoul(e+1, &e, 10);
2673         if (*e && *e != '\n')
2674                 return -EINVAL;
2675         dev = MKDEV(major, minor);
2676         if (major != MAJOR(dev) ||
2677             minor != MINOR(dev))
2678                 return -EOVERFLOW;
2679
2680
2681         if (mddev->persistent) {
2682                 rdev = md_import_device(dev, mddev->major_version,
2683                                         mddev->minor_version);
2684                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2685                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2686                                                        mdk_rdev_t, same_set);
2687                         err = super_types[mddev->major_version]
2688                                 .load_super(rdev, rdev0, mddev->minor_version);
2689                         if (err < 0)
2690                                 goto out;
2691                 }
2692         } else if (mddev->external)
2693                 rdev = md_import_device(dev, -2, -1);
2694         else
2695                 rdev = md_import_device(dev, -1, -1);
2696
2697         if (IS_ERR(rdev))
2698                 return PTR_ERR(rdev);
2699         err = bind_rdev_to_array(rdev, mddev);
2700  out:
2701         if (err)
2702                 export_rdev(rdev);
2703         return err ? err : len;
2704 }
2705
2706 static struct md_sysfs_entry md_new_device =
2707 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2708
2709 static ssize_t
2710 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2711 {
2712         char *end;
2713         unsigned long chunk, end_chunk;
2714
2715         if (!mddev->bitmap)
2716                 goto out;
2717         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2718         while (*buf) {
2719                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2720                 if (buf == end) break;
2721                 if (*end == '-') { /* range */
2722                         buf = end + 1;
2723                         end_chunk = simple_strtoul(buf, &end, 0);
2724                         if (buf == end) break;
2725                 }
2726                 if (*end && !isspace(*end)) break;
2727                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2728                 buf = end;
2729                 while (isspace(*buf)) buf++;
2730         }
2731         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2732 out:
2733         return len;
2734 }
2735
2736 static struct md_sysfs_entry md_bitmap =
2737 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2738
2739 static ssize_t
2740 size_show(mddev_t *mddev, char *page)
2741 {
2742         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2743 }
2744
2745 static int update_size(mddev_t *mddev, unsigned long size);
2746
2747 static ssize_t
2748 size_store(mddev_t *mddev, const char *buf, size_t len)
2749 {
2750         /* If array is inactive, we can reduce the component size, but
2751          * not increase it (except from 0).
2752          * If array is active, we can try an on-line resize
2753          */
2754         char *e;
2755         int err = 0;
2756         unsigned long long size = simple_strtoull(buf, &e, 10);
2757         if (!*buf || *buf == '\n' ||
2758             (*e && *e != '\n'))
2759                 return -EINVAL;
2760
2761         if (mddev->pers) {
2762                 err = update_size(mddev, size);
2763                 md_update_sb(mddev, 1);
2764         } else {
2765                 if (mddev->size == 0 ||
2766                     mddev->size > size)
2767                         mddev->size = size;
2768                 else
2769                         err = -ENOSPC;
2770         }
2771         return err ? err : len;
2772 }
2773
2774 static struct md_sysfs_entry md_size =
2775 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2776
2777
2778 /* Metdata version.
2779  * This is one of
2780  *   'none' for arrays with no metadata (good luck...)
2781  *   'external' for arrays with externally managed metadata,
2782  * or N.M for internally known formats
2783  */
2784 static ssize_t
2785 metadata_show(mddev_t *mddev, char *page)
2786 {
2787         if (mddev->persistent)
2788                 return sprintf(page, "%d.%d\n",
2789                                mddev->major_version, mddev->minor_version);
2790         else if (mddev->external)
2791                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2792         else
2793                 return sprintf(page, "none\n");
2794 }
2795
2796 static ssize_t
2797 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2798 {
2799         int major, minor;
2800         char *e;
2801         if (!list_empty(&mddev->disks))
2802                 return -EBUSY;
2803
2804         if (cmd_match(buf, "none")) {
2805                 mddev->persistent = 0;
2806                 mddev->external = 0;
2807                 mddev->major_version = 0;
2808                 mddev->minor_version = 90;
2809                 return len;
2810         }
2811         if (strncmp(buf, "external:", 9) == 0) {
2812                 size_t namelen = len-9;
2813                 if (namelen >= sizeof(mddev->metadata_type))
2814                         namelen = sizeof(mddev->metadata_type)-1;
2815                 strncpy(mddev->metadata_type, buf+9, namelen);
2816                 mddev->metadata_type[namelen] = 0;
2817                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2818                         mddev->metadata_type[--namelen] = 0;
2819                 mddev->persistent = 0;
2820                 mddev->external = 1;
2821                 mddev->major_version = 0;
2822                 mddev->minor_version = 90;
2823                 return len;
2824         }
2825         major = simple_strtoul(buf, &e, 10);
2826         if (e==buf || *e != '.')
2827                 return -EINVAL;
2828         buf = e+1;
2829         minor = simple_strtoul(buf, &e, 10);
2830         if (e==buf || (*e && *e != '\n') )
2831                 return -EINVAL;
2832         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2833                 return -ENOENT;
2834         mddev->major_version = major;
2835         mddev->minor_version = minor;
2836         mddev->persistent = 1;
2837         mddev->external = 0;
2838         return len;
2839 }
2840
2841 static struct md_sysfs_entry md_metadata =
2842 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2843
2844 static ssize_t
2845 action_show(mddev_t *mddev, char *page)
2846 {
2847         char *type = "idle";
2848         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2849             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2850                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2851                         type = "reshape";
2852                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2853                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2854                                 type = "resync";
2855                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2856                                 type = "check";
2857                         else
2858                                 type = "repair";
2859                 } else
2860                         type = "recover";
2861         }
2862         return sprintf(page, "%s\n", type);
2863 }
2864
2865 static ssize_t
2866 action_store(mddev_t *mddev, const char *page, size_t len)
2867 {
2868         if (!mddev->pers || !mddev->pers->sync_request)
2869                 return -EINVAL;
2870
2871         if (cmd_match(page, "idle")) {
2872                 if (mddev->sync_thread) {
2873                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2874                         md_unregister_thread(mddev->sync_thread);
2875                         mddev->sync_thread = NULL;
2876                         mddev->recovery = 0;
2877                 }
2878         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2879                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2880                 return -EBUSY;
2881         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2882                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2883         else if (cmd_match(page, "reshape")) {
2884                 int err;
2885                 if (mddev->pers->start_reshape == NULL)
2886                         return -EINVAL;
2887                 err = mddev->pers->start_reshape(mddev);
2888                 if (err)
2889                         return err;
2890         } else {
2891                 if (cmd_match(page, "check"))
2892                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2893                 else if (!cmd_match(page, "repair"))
2894                         return -EINVAL;
2895                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2896                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2897         }
2898         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2899         md_wakeup_thread(mddev->thread);
2900         return len;
2901 }
2902
2903 static ssize_t
2904 mismatch_cnt_show(mddev_t *mddev, char *page)
2905 {
2906         return sprintf(page, "%llu\n",
2907                        (unsigned long long) mddev->resync_mismatches);
2908 }
2909
2910 static struct md_sysfs_entry md_scan_mode =
2911 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2912
2913
2914 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2915
2916 static ssize_t
2917 sync_min_show(mddev_t *mddev, char *page)
2918 {
2919         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2920                        mddev->sync_speed_min ? "local": "system");
2921 }
2922
2923 static ssize_t
2924 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2925 {
2926         int min;
2927         char *e;
2928         if (strncmp(buf, "system", 6)==0) {
2929                 mddev->sync_speed_min = 0;
2930                 return len;
2931         }
2932         min = simple_strtoul(buf, &e, 10);
2933         if (buf == e || (*e && *e != '\n') || min <= 0)
2934                 return -EINVAL;
2935         mddev->sync_speed_min = min;
2936         return len;
2937 }
2938
2939 static struct md_sysfs_entry md_sync_min =
2940 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2941
2942 static ssize_t
2943 sync_max_show(mddev_t *mddev, char *page)
2944 {
2945         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2946                        mddev->sync_speed_max ? "local": "system");
2947 }
2948
2949 static ssize_t
2950 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2951 {
2952         int max;
2953         char *e;
2954         if (strncmp(buf, "system", 6)==0) {
2955                 mddev->sync_speed_max = 0;
2956                 return len;
2957         }
2958         max = simple_strtoul(buf, &e, 10);
2959         if (buf == e || (*e && *e != '\n') || max <= 0)
2960                 return -EINVAL;
2961         mddev->sync_speed_max = max;
2962         return len;
2963 }
2964
2965 static struct md_sysfs_entry md_sync_max =
2966 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2967
2968 static ssize_t
2969 degraded_show(mddev_t *mddev, char *page)
2970 {
2971         return sprintf(page, "%d\n", mddev->degraded);
2972 }
2973 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2974
2975 static ssize_t
2976 sync_speed_show(mddev_t *mddev, char *page)
2977 {
2978         unsigned long resync, dt, db;
2979         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2980         dt = ((jiffies - mddev->resync_mark) / HZ);
2981         if (!dt) dt++;
2982         db = resync - (mddev->resync_mark_cnt);
2983         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2984 }
2985
2986 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2987
2988 static ssize_t
2989 sync_completed_show(mddev_t *mddev, char *page)
2990 {
2991         unsigned long max_blocks, resync;
2992
2993         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2994                 max_blocks = mddev->resync_max_sectors;
2995         else
2996                 max_blocks = mddev->size << 1;
2997
2998         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2999         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3000 }
3001
3002 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3003
3004 static ssize_t
3005 max_sync_show(mddev_t *mddev, char *page)
3006 {
3007         if (mddev->resync_max == MaxSector)
3008                 return sprintf(page, "max\n");
3009         else
3010                 return sprintf(page, "%llu\n",
3011                                (unsigned long long)mddev->resync_max);
3012 }
3013 static ssize_t
3014 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3015 {
3016         if (strncmp(buf, "max", 3) == 0)
3017                 mddev->resync_max = MaxSector;
3018         else {
3019                 char *ep;
3020                 unsigned long long max = simple_strtoull(buf, &ep, 10);
3021                 if (ep == buf || (*ep != 0 && *ep != '\n'))
3022                         return -EINVAL;
3023                 if (max < mddev->resync_max &&
3024                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3025                         return -EBUSY;
3026
3027                 /* Must be a multiple of chunk_size */
3028                 if (mddev->chunk_size) {
3029                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3030                                 return -EINVAL;
3031                 }
3032                 mddev->resync_max = max;
3033         }
3034         wake_up(&mddev->recovery_wait);
3035         return len;
3036 }
3037
3038 static struct md_sysfs_entry md_max_sync =
3039 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3040
3041 static ssize_t
3042 suspend_lo_show(mddev_t *mddev, char *page)
3043 {
3044         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3045 }
3046
3047 static ssize_t
3048 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3049 {
3050         char *e;
3051         unsigned long long new = simple_strtoull(buf, &e, 10);
3052
3053         if (mddev->pers->quiesce == NULL)
3054                 return -EINVAL;
3055         if (buf == e || (*e && *e != '\n'))
3056                 return -EINVAL;
3057         if (new >= mddev->suspend_hi ||
3058             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3059                 mddev->suspend_lo = new;
3060                 mddev->pers->quiesce(mddev, 2);
3061                 return len;
3062         } else
3063                 return -EINVAL;
3064 }
3065 static struct md_sysfs_entry md_suspend_lo =
3066 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3067
3068
3069 static ssize_t
3070 suspend_hi_show(mddev_t *mddev, char *page)
3071 {
3072         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3073 }
3074
3075 static ssize_t
3076 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3077 {
3078         char *e;
3079         unsigned long long new = simple_strtoull(buf, &e, 10);
3080
3081         if (mddev->pers->quiesce == NULL)
3082                 return -EINVAL;
3083         if (buf == e || (*e && *e != '\n'))
3084                 return -EINVAL;
3085         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3086             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3087                 mddev->suspend_hi = new;
3088                 mddev->pers->quiesce(mddev, 1);
3089                 mddev->pers->quiesce(mddev, 0);
3090                 return len;
3091         } else
3092                 return -EINVAL;
3093 }
3094 static struct md_sysfs_entry md_suspend_hi =
3095 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3096
3097 static ssize_t
3098 reshape_position_show(mddev_t *mddev, char *page)
3099 {
3100         if (mddev->reshape_position != MaxSector)
3101                 return sprintf(page, "%llu\n",
3102                                (unsigned long long)mddev->reshape_position);
3103         strcpy(page, "none\n");
3104         return 5;
3105 }
3106
3107 static ssize_t
3108 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3109 {
3110         char *e;
3111         unsigned long long new = simple_strtoull(buf, &e, 10);
3112         if (mddev->pers)
3113                 return -EBUSY;
3114         if (buf == e || (*e && *e != '\n'))
3115                 return -EINVAL;
3116         mddev->reshape_position = new;
3117         mddev->delta_disks = 0;
3118         mddev->new_level = mddev->level;
3119         mddev->new_layout = mddev->layout;
3120         mddev->new_chunk = mddev->chunk_size;
3121         return len;
3122 }
3123
3124 static struct md_sysfs_entry md_reshape_position =
3125 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3126        reshape_position_store);
3127
3128
3129 static struct attribute *md_default_attrs[] = {
3130         &md_level.attr,
3131         &md_layout.attr,
3132         &md_raid_disks.attr,
3133         &md_chunk_size.attr,
3134         &md_size.attr,
3135         &md_resync_start.attr,
3136         &md_metadata.attr,
3137         &md_new_device.attr,
3138         &md_safe_delay.attr,
3139         &md_array_state.attr,
3140         &md_reshape_position.attr,
3141         NULL,
3142 };
3143
3144 static struct attribute *md_redundancy_attrs[] = {
3145         &md_scan_mode.attr,
3146         &md_mismatches.attr,
3147         &md_sync_min.attr,
3148         &md_sync_max.attr,
3149         &md_sync_speed.attr,
3150         &md_sync_completed.attr,
3151         &md_max_sync.attr,
3152         &md_suspend_lo.attr,
3153         &md_suspend_hi.attr,
3154         &md_bitmap.attr,
3155         &md_degraded.attr,
3156         NULL,
3157 };
3158 static struct attribute_group md_redundancy_group = {
3159         .name = NULL,
3160         .attrs = md_redundancy_attrs,
3161 };
3162
3163
3164 static ssize_t
3165 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3166 {
3167         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3168         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3169         ssize_t rv;
3170
3171         if (!entry->show)
3172                 return -EIO;
3173         rv = mddev_lock(mddev);
3174         if (!rv) {
3175                 rv = entry->show(mddev, page);
3176                 mddev_unlock(mddev);
3177         }
3178         return rv;
3179 }
3180
3181 static ssize_t
3182 md_attr_store(struct kobject *kobj, struct attribute *attr,
3183               const char *page, size_t length)
3184 {
3185         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3186         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3187         ssize_t rv;
3188
3189         if (!entry->store)
3190                 return -EIO;
3191         if (!capable(CAP_SYS_ADMIN))
3192                 return -EACCES;
3193         rv = mddev_lock(mddev);
3194         if (!rv) {
3195                 rv = entry->store(mddev, page, length);
3196                 mddev_unlock(mddev);
3197         }
3198         return rv;
3199 }
3200
3201 static void md_free(struct kobject *ko)
3202 {
3203         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3204         kfree(mddev);
3205 }
3206
3207 static struct sysfs_ops md_sysfs_ops = {
3208         .show   = md_attr_show,
3209         .store  = md_attr_store,
3210 };
3211 static struct kobj_type md_ktype = {
3212         .release        = md_free,
3213         .sysfs_ops      = &md_sysfs_ops,
3214         .default_attrs  = md_default_attrs,
3215 };
3216
3217 int mdp_major = 0;
3218
3219 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3220 {
3221         static DEFINE_MUTEX(disks_mutex);
3222         mddev_t *mddev = mddev_find(dev);
3223         struct gendisk *disk;
3224         int partitioned = (MAJOR(dev) != MD_MAJOR);
3225         int shift = partitioned ? MdpMinorShift : 0;
3226         int unit = MINOR(dev) >> shift;
3227         int error;
3228
3229         if (!mddev)
3230                 return NULL;
3231
3232         mutex_lock(&disks_mutex);
3233         if (mddev->gendisk) {
3234                 mutex_unlock(&disks_mutex);
3235                 mddev_put(mddev);
3236                 return NULL;
3237         }
3238         disk = alloc_disk(1 << shift);
3239         if (!disk) {
3240                 mutex_unlock(&disks_mutex);
3241                 mddev_put(mddev);
3242                 return NULL;
3243         }
3244         disk->major = MAJOR(dev);
3245         disk->first_minor = unit << shift;
3246         if (partitioned)
3247                 sprintf(disk->disk_name, "md_d%d", unit);
3248         else
3249                 sprintf(disk->disk_name, "md%d", unit);
3250         disk->fops = &md_fops;
3251         disk->private_data = mddev;
3252         disk->queue = mddev->queue;
3253         add_disk(disk);
3254         mddev->gendisk = disk;
3255         mutex_unlock(&disks_mutex);
3256         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3257                                      "%s", "md");
3258         if (error)
3259                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3260                        disk->disk_name);
3261         else
3262                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3263         return NULL;
3264 }
3265
3266 static void md_safemode_timeout(unsigned long data)
3267 {
3268         mddev_t *mddev = (mddev_t *) data;
3269
3270         mddev->safemode = 1;
3271         md_wakeup_thread(mddev->thread);
3272 }
3273
3274 static int start_dirty_degraded;
3275
3276 static int do_md_run(mddev_t * mddev)
3277 {
3278         int err;
3279         int chunk_size;
3280         struct list_head *tmp;
3281         mdk_rdev_t *rdev;
3282         struct gendisk *disk;
3283         struct mdk_personality *pers;
3284         char b[BDEVNAME_SIZE];
3285
3286         if (list_empty(&mddev->disks))
3287                 /* cannot run an array with no devices.. */
3288                 return -EINVAL;
3289
3290         if (mddev->pers)
3291                 return -EBUSY;
3292
3293         /*
3294          * Analyze all RAID superblock(s)
3295          */
3296         if (!mddev->raid_disks) {
3297                 if (!mddev->persistent)
3298                         return -EINVAL;
3299                 analyze_sbs(mddev);
3300         }
3301
3302         chunk_size = mddev->chunk_size;
3303
3304         if (chunk_size) {
3305                 if (chunk_size > MAX_CHUNK_SIZE) {
3306                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3307                                 chunk_size, MAX_CHUNK_SIZE);
3308                         return -EINVAL;
3309                 }
3310                 /*
3311                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3312                  */
3313                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3314                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3315                         return -EINVAL;
3316                 }
3317                 if (chunk_size < PAGE_SIZE) {
3318                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3319                                 chunk_size, PAGE_SIZE);
3320                         return -EINVAL;
3321                 }
3322
3323                 /* devices must have minimum size of one chunk */
3324                 rdev_for_each(rdev, tmp, mddev) {
3325                         if (test_bit(Faulty, &rdev->flags))
3326                                 continue;
3327                         if (rdev->size < chunk_size / 1024) {
3328                                 printk(KERN_WARNING
3329                                         "md: Dev %s smaller than chunk_size:"
3330                                         " %lluk < %dk\n",
3331                                         bdevname(rdev->bdev,b),
3332                                         (unsigned long long)rdev->size,
3333                                         chunk_size / 1024);
3334                                 return -EINVAL;
3335                         }
3336                 }
3337         }
3338
3339 #ifdef CONFIG_KMOD
3340         if (mddev->level != LEVEL_NONE)
3341                 request_module("md-level-%d", mddev->level);
3342         else if (mddev->clevel[0])
3343                 request_module("md-%s", mddev->clevel);
3344 #endif
3345
3346         /*
3347          * Drop all container device buffers, from now on
3348          * the only valid external interface is through the md
3349          * device.
3350          */
3351         rdev_for_each(rdev, tmp, mddev) {
3352                 if (test_bit(Faulty, &rdev->flags))
3353                         continue;
3354                 sync_blockdev(rdev->bdev);
3355                 invalidate_bdev(rdev->bdev);
3356
3357                 /* perform some consistency tests on the device.
3358                  * We don't want the data to overlap the metadata,
3359                  * Internal Bitmap issues has handled elsewhere.
3360                  */
3361                 if (rdev->data_offset < rdev->sb_offset) {
3362                         if (mddev->size &&
3363                             rdev->data_offset + mddev->size*2
3364                             > rdev->sb_offset*2) {
3365                                 printk("md: %s: data overlaps metadata\n",
3366                                        mdname(mddev));
3367                                 return -EINVAL;
3368                         }
3369                 } else {
3370                         if (rdev->sb_offset*2 + rdev->sb_size/512
3371                             > rdev->data_offset) {
3372                                 printk("md: %s: metadata overlaps data\n",
3373                                        mdname(mddev));
3374                                 return -EINVAL;
3375                         }
3376                 }
3377         }
3378
3379         md_probe(mddev->unit, NULL, NULL);
3380         disk = mddev->gendisk;
3381         if (!disk)
3382                 return -ENOMEM;
3383
3384         spin_lock(&pers_lock);
3385         pers = find_pers(mddev->level, mddev->clevel);
3386         if (!pers || !try_module_get(pers->owner)) {
3387                 spin_unlock(&pers_lock);
3388                 if (mddev->level != LEVEL_NONE)
3389                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3390                                mddev->level);
3391                 else
3392                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3393                                mddev->clevel);
3394                 return -EINVAL;
3395         }
3396         mddev->pers = pers;
3397         spin_unlock(&pers_lock);
3398         mddev->level = pers->level;
3399         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3400
3401         if (mddev->reshape_position != MaxSector &&
3402             pers->start_reshape == NULL) {
3403                 /* This personality cannot handle reshaping... */
3404                 mddev->pers = NULL;
3405                 module_put(pers->owner);
3406                 return -EINVAL;
3407         }
3408
3409         if (pers->sync_request) {
3410                 /* Warn if this is a potentially silly
3411                  * configuration.
3412                  */
3413                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3414                 mdk_rdev_t *rdev2;
3415                 struct list_head *tmp2;
3416                 int warned = 0;
3417                 rdev_for_each(rdev, tmp, mddev) {
3418                         rdev_for_each(rdev2, tmp2, mddev) {
3419                                 if (rdev < rdev2 &&
3420                                     rdev->bdev->bd_contains ==
3421                                     rdev2->bdev->bd_contains) {
3422                                         printk(KERN_WARNING
3423                                                "%s: WARNING: %s appears to be"
3424                                                " on the same physical disk as"
3425                                                " %s.\n",
3426                                                mdname(mddev),
3427                                                bdevname(rdev->bdev,b),
3428                                                bdevname(rdev2->bdev,b2));
3429                                         warned = 1;
3430                                 }
3431                         }
3432                 }
3433                 if (warned)
3434                         printk(KERN_WARNING
3435                                "True protection against single-disk"
3436                                " failure might be compromised.\n");
3437         }
3438
3439         mddev->recovery = 0;
3440         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3441         mddev->barriers_work = 1;
3442         mddev->ok_start_degraded = start_dirty_degraded;
3443
3444         if (start_readonly)
3445                 mddev->ro = 2; /* read-only, but switch on first write */
3446
3447         err = mddev->pers->run(mddev);
3448         if (!err && mddev->pers->sync_request) {
3449                 err = bitmap_create(mddev);
3450                 if (err) {
3451                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3452                                mdname(mddev), err);
3453                         mddev->pers->stop(mddev);
3454                 }
3455         }
3456         if (err) {
3457                 printk(KERN_ERR "md: pers->run() failed ...\n");
3458                 module_put(mddev->pers->owner);
3459                 mddev->pers = NULL;
3460                 bitmap_destroy(mddev);
3461                 return err;
3462         }
3463         if (mddev->pers->sync_request) {
3464                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3465                         printk(KERN_WARNING
3466                                "md: cannot register extra attributes for %s\n",
3467                                mdname(mddev));
3468         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3469                 mddev->ro = 0;
3470
3471         atomic_set(&mddev->writes_pending,0);
3472         mddev->safemode = 0;
3473         mddev->safemode_timer.function = md_safemode_timeout;
3474         mddev->safemode_timer.data = (unsigned long) mddev;
3475         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3476         mddev->in_sync = 1;
3477
3478         rdev_for_each(rdev, tmp, mddev)
3479                 if (rdev->raid_disk >= 0) {
3480                         char nm[20];
3481                         sprintf(nm, "rd%d", rdev->raid_disk);
3482                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3483                                 printk("md: cannot register %s for %s\n",
3484                                        nm, mdname(mddev));
3485                 }
3486
3487         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3488
3489         if (mddev->flags)
3490                 md_update_sb(mddev, 0);
3491
3492         set_capacity(disk, mddev->array_size<<1);
3493
3494         /* If we call blk_queue_make_request here, it will
3495          * re-initialise max_sectors etc which may have been
3496          * refined inside -> run.  So just set the bits we need to set.
3497          * Most initialisation happended when we called
3498          * blk_queue_make_request(..., md_fail_request)
3499          * earlier.
3500          */
3501         mddev->queue->queuedata = mddev;
3502         mddev->queue->make_request_fn = mddev->pers->make_request;
3503
3504         /* If there is a partially-recovered drive we need to
3505          * start recovery here.  If we leave it to md_check_recovery,
3506          * it will remove the drives and not do the right thing
3507          */
3508         if (mddev->degraded && !mddev->sync_thread) {
3509                 struct list_head *rtmp;
3510                 int spares = 0;
3511                 rdev_for_each(rdev, rtmp, mddev)
3512                         if (rdev->raid_disk >= 0 &&
3513                             !test_bit(In_sync, &rdev->flags) &&
3514                             !test_bit(Faulty, &rdev->flags))
3515                                 /* complete an interrupted recovery */
3516                                 spares++;
3517                 if (spares && mddev->pers->sync_request) {
3518                         mddev->recovery = 0;
3519                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3520                         mddev->sync_thread = md_register_thread(md_do_sync,
3521                                                                 mddev,
3522                                                                 "%s_resync");
3523                         if (!mddev->sync_thread) {
3524                                 printk(KERN_ERR "%s: could not start resync"
3525                                        " thread...\n",
3526                                        mdname(mddev));
3527                                 /* leave the spares where they are, it shouldn't hurt */
3528                                 mddev->recovery = 0;
3529                         }
3530                 }
3531         }
3532         md_wakeup_thread(mddev->thread);
3533         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3534
3535         mddev->changed = 1;
3536         md_new_event(mddev);
3537         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3538         return 0;
3539 }
3540
3541 static int restart_array(mddev_t *mddev)
3542 {
3543         struct gendisk *disk = mddev->gendisk;
3544         int err;
3545
3546         /*
3547          * Complain if it has no devices
3548          */
3549         err = -ENXIO;
3550         if (list_empty(&mddev->disks))
3551                 goto out;
3552
3553         if (mddev->pers) {
3554                 err = -EBUSY;
3555                 if (!mddev->ro)
3556                         goto out;
3557
3558                 mddev->safemode = 0;
3559                 mddev->ro = 0;
3560                 set_disk_ro(disk, 0);
3561
3562                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3563                         mdname(mddev));
3564                 /*
3565                  * Kick recovery or resync if necessary
3566                  */
3567                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3568                 md_wakeup_thread(mddev->thread);
3569                 md_wakeup_thread(mddev->sync_thread);
3570                 err = 0;
3571         } else
3572                 err = -EINVAL;
3573
3574 out:
3575         return err;
3576 }
3577
3578 /* similar to deny_write_access, but accounts for our holding a reference
3579  * to the file ourselves */
3580 static int deny_bitmap_write_access(struct file * file)
3581 {
3582         struct inode *inode = file->f_mapping->host;
3583
3584         spin_lock(&inode->i_lock);
3585         if (atomic_read(&inode->i_writecount) > 1) {
3586                 spin_unlock(&inode->i_lock);
3587                 return -ETXTBSY;
3588         }
3589         atomic_set(&inode->i_writecount, -1);
3590         spin_unlock(&inode->i_lock);
3591
3592         return 0;
3593 }
3594
3595 static void restore_bitmap_write_access(struct file *file)
3596 {
3597         struct inode *inode = file->f_mapping->host;
3598
3599         spin_lock(&inode->i_lock);
3600         atomic_set(&inode->i_writecount, 1);
3601         spin_unlock(&inode->i_lock);
3602 }
3603
3604 /* mode:
3605  *   0 - completely stop and dis-assemble array
3606  *   1 - switch to readonly
3607  *   2 - stop but do not disassemble array
3608  */
3609 static int do_md_stop(mddev_t * mddev, int mode)
3610 {
3611         int err = 0;
3612         struct gendisk *disk = mddev->gendisk;
3613
3614         if (mddev->pers) {
3615                 if (atomic_read(&mddev->active)>2) {
3616                         printk("md: %s still in use.\n",mdname(mddev));
3617                         return -EBUSY;
3618                 }
3619
3620                 if (mddev->sync_thread) {
3621                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3622                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3623                         md_unregister_thread(mddev->sync_thread);
3624                         mddev->sync_thread = NULL;
3625                 }
3626
3627                 del_timer_sync(&mddev->safemode_timer);
3628
3629                 invalidate_partition(disk, 0);
3630
3631                 switch(mode) {
3632                 case 1: /* readonly */
3633                         err  = -ENXIO;
3634                         if (mddev->ro==1)
3635                                 goto out;
3636                         mddev->ro = 1;
3637                         break;
3638                 case 0: /* disassemble */
3639                 case 2: /* stop */
3640                         bitmap_flush(mddev);
3641                         md_super_wait(mddev);
3642                         if (mddev->ro)
3643                                 set_disk_ro(disk, 0);
3644                         blk_queue_make_request(mddev->queue, md_fail_request);
3645                         mddev->pers->stop(mddev);
3646                         mddev->queue->merge_bvec_fn = NULL;
3647                         mddev->queue->unplug_fn = NULL;
3648                         mddev->queue->backing_dev_info.congested_fn = NULL;
3649                         if (mddev->pers->sync_request)
3650                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3651
3652                         module_put(mddev->pers->owner);
3653                         mddev->pers = NULL;
3654
3655                         set_capacity(disk, 0);
3656                         mddev->changed = 1;
3657
3658                         if (mddev->ro)
3659                                 mddev->ro = 0;
3660                 }
3661                 if (!mddev->in_sync || mddev->flags) {
3662                         /* mark array as shutdown cleanly */
3663                         mddev->in_sync = 1;
3664                         md_update_sb(mddev, 1);
3665                 }
3666                 if (mode == 1)
3667                         set_disk_ro(disk, 1);
3668                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3669         }
3670
3671         /*
3672          * Free resources if final stop
3673          */
3674         if (mode == 0) {
3675                 mdk_rdev_t *rdev;
3676                 struct list_head *tmp;
3677
3678                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3679
3680                 bitmap_destroy(mddev);
3681                 if (mddev->bitmap_file) {
3682                         restore_bitmap_write_access(mddev->bitmap_file);
3683                         fput(mddev->bitmap_file);
3684                         mddev->bitmap_file = NULL;
3685                 }
3686                 mddev->bitmap_offset = 0;
3687
3688                 rdev_for_each(rdev, tmp, mddev)
3689                         if (rdev->raid_disk >= 0) {
3690                                 char nm[20];
3691                                 sprintf(nm, "rd%d", rdev->raid_disk);
3692                                 sysfs_remove_link(&mddev->kobj, nm);
3693                         }
3694
3695                 /* make sure all md_delayed_delete calls have finished */
3696                 flush_scheduled_work();
3697
3698                 export_array(mddev);
3699
3700                 mddev->array_size = 0;
3701                 mddev->size = 0;
3702                 mddev->raid_disks = 0;
3703                 mddev->recovery_cp = 0;
3704                 mddev->resync_max = MaxSector;
3705                 mddev->reshape_position = MaxSector;
3706                 mddev->external = 0;
3707                 mddev->persistent = 0;
3708
3709         } else if (mddev->pers)
3710                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3711                         mdname(mddev));
3712         err = 0;
3713         md_new_event(mddev);
3714 out:
3715         return err;
3716 }
3717
3718 #ifndef MODULE
3719 static void autorun_array(mddev_t *mddev)
3720 {
3721         mdk_rdev_t *rdev;
3722         struct list_head *tmp;
3723         int err;
3724
3725         if (list_empty(&mddev->disks))
3726                 return;
3727
3728         printk(KERN_INFO "md: running: ");
3729
3730         rdev_for_each(rdev, tmp, mddev) {
3731                 char b[BDEVNAME_SIZE];
3732                 printk("<%s>", bdevname(rdev->bdev,b));
3733         }
3734         printk("\n");
3735
3736         err = do_md_run (mddev);
3737         if (err) {
3738                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3739                 do_md_stop (mddev, 0);
3740         }
3741 }
3742
3743 /*
3744  * lets try to run arrays based on all disks that have arrived
3745  * until now. (those are in pending_raid_disks)
3746  *
3747  * the method: pick the first pending disk, collect all disks with
3748  * the same UUID, remove all from the pending list and put them into
3749  * the 'same_array' list. Then order this list based on superblock
3750  * update time (freshest comes first), kick out 'old' disks and
3751  * compare superblocks. If everything's fine then run it.
3752  *
3753  * If "unit" is allocated, then bump its reference count
3754  */
3755 static void autorun_devices(int part)
3756 {
3757         struct list_head *tmp;
3758         mdk_rdev_t *rdev0, *rdev;
3759         mddev_t *mddev;
3760         char b[BDEVNAME_SIZE];
3761
3762         printk(KERN_INFO "md: autorun ...\n");
3763         while (!list_empty(&pending_raid_disks)) {
3764                 int unit;
3765                 dev_t dev;
3766                 LIST_HEAD(candidates);
3767                 rdev0 = list_entry(pending_raid_disks.next,
3768                                          mdk_rdev_t, same_set);
3769
3770                 printk(KERN_INFO "md: considering %s ...\n",
3771                         bdevname(rdev0->bdev,b));
3772                 INIT_LIST_HEAD(&candidates);
3773                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3774                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3775                                 printk(KERN_INFO "md:  adding %s ...\n",
3776                                         bdevname(rdev->bdev,b));
3777                                 list_move(&rdev->same_set, &candidates);
3778                         }
3779                 /*
3780                  * now we have a set of devices, with all of them having
3781                  * mostly sane superblocks. It's time to allocate the
3782                  * mddev.
3783                  */
3784                 if (part) {
3785                         dev = MKDEV(mdp_major,
3786                                     rdev0->preferred_minor << MdpMinorShift);
3787                         unit = MINOR(dev) >> MdpMinorShift;
3788                 } else {
3789                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3790                         unit = MINOR(dev);
3791                 }
3792                 if (rdev0->preferred_minor != unit) {
3793                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3794                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3795                         break;
3796                 }
3797
3798                 md_probe(dev, NULL, NULL);
3799                 mddev = mddev_find(dev);
3800                 if (!mddev) {
3801                         printk(KERN_ERR
3802                                 "md: cannot allocate memory for md drive.\n");
3803                         break;
3804                 }
3805                 if (mddev_lock(mddev))
3806                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3807                                mdname(mddev));
3808                 else if (mddev->raid_disks || mddev->major_version
3809                          || !list_empty(&mddev->disks)) {
3810                         printk(KERN_WARNING
3811                                 "md: %s already running, cannot run %s\n",
3812                                 mdname(mddev), bdevname(rdev0->bdev,b));
3813                         mddev_unlock(mddev);
3814                 } else {
3815                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3816                         mddev->persistent = 1;
3817                         rdev_for_each_list(rdev, tmp, candidates) {
3818                                 list_del_init(&rdev->same_set);
3819                                 if (bind_rdev_to_array(rdev, mddev))
3820                                         export_rdev(rdev);
3821                         }
3822                         autorun_array(mddev);
3823                         mddev_unlock(mddev);
3824                 }
3825                 /* on success, candidates will be empty, on error
3826                  * it won't...
3827                  */
3828                 rdev_for_each_list(rdev, tmp, candidates)
3829                         export_rdev(rdev);
3830                 mddev_put(mddev);
3831         }
3832         printk(KERN_INFO "md: ... autorun DONE.\n");
3833 }
3834 #endif /* !MODULE */
3835
3836 static int get_version(void __user * arg)
3837 {
3838         mdu_version_t ver;
3839
3840         ver.major = MD_MAJOR_VERSION;
3841         ver.minor = MD_MINOR_VERSION;
3842         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3843
3844         if (copy_to_user(arg, &ver, sizeof(ver)))
3845                 return -EFAULT;
3846
3847         return 0;
3848 }
3849
3850 static int get_array_info(mddev_t * mddev, void __user * arg)
3851 {
3852         mdu_array_info_t info;
3853         int nr,working,active,failed,spare;
3854         mdk_rdev_t *rdev;
3855         struct list_head *tmp;
3856
3857         nr=working=active=failed=spare=0;
3858         rdev_for_each(rdev, tmp, mddev) {
3859                 nr++;
3860                 if (test_bit(Faulty, &rdev->flags))
3861                         failed++;
3862                 else {
3863                         working++;
3864                         if (test_bit(In_sync, &rdev->flags))
3865                                 active++;
3866                         else
3867                                 spare++;
3868                 }
3869         }
3870
3871         info.major_version = mddev->major_version;
3872         info.minor_version = mddev->minor_version;
3873         info.patch_version = MD_PATCHLEVEL_VERSION;
3874         info.ctime         = mddev->ctime;
3875         info.level         = mddev->level;
3876         info.size          = mddev->size;
3877         if (info.size != mddev->size) /* overflow */
3878                 info.size = -1;
3879         info.nr_disks      = nr;
3880         info.raid_disks    = mddev->raid_disks;
3881         info.md_minor      = mddev->md_minor;
3882         info.not_persistent= !mddev->persistent;
3883
3884         info.utime         = mddev->utime;
3885         info.state         = 0;
3886         if (mddev->in_sync)
3887                 info.state = (1<<MD_SB_CLEAN);
3888         if (mddev->bitmap && mddev->bitmap_offset)
3889                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3890         info.active_disks  = active;
3891         info.working_disks = working;
3892         info.failed_disks  = failed;
3893         info.spare_disks   = spare;
3894
3895         info.layout        = mddev->layout;
3896         info.chunk_size    = mddev->chunk_size;
3897
3898         if (copy_to_user(arg, &info, sizeof(info)))
3899                 return -EFAULT;
3900
3901         return 0;
3902 }
3903
3904 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3905 {
3906         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3907         char *ptr, *buf = NULL;
3908         int err = -ENOMEM;
3909
3910         md_allow_write(mddev);
3911
3912         file = kmalloc(sizeof(*file), GFP_KERNEL);
3913         if (!file)
3914                 goto out;
3915
3916         /* bitmap disabled, zero the first byte and copy out */
3917         if (!mddev->bitmap || !mddev->bitmap->file) {
3918                 file->pathname[0] = '\0';
3919                 goto copy_out;
3920         }
3921
3922         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3923         if (!buf)
3924                 goto out;
3925
3926         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3927         if (!ptr)
3928                 goto out;
3929
3930         strcpy(file->pathname, ptr);
3931
3932 copy_out:
3933         err = 0;
3934         if (copy_to_user(arg, file, sizeof(*file)))
3935                 err = -EFAULT;
3936 out:
3937         kfree(buf);
3938         kfree(file);
3939         return err;
3940 }
3941
3942 static int get_disk_info(mddev_t * mddev, void __user * arg)
3943 {
3944         mdu_disk_info_t info;
3945         unsigned int nr;
3946         mdk_rdev_t *rdev;
3947
3948         if (copy_from_user(&info, arg, sizeof(info)))
3949                 return -EFAULT;
3950
3951         nr = info.number;
3952
3953         rdev = find_rdev_nr(mddev, nr);
3954         if (rdev) {
3955                 info.major = MAJOR(rdev->bdev->bd_dev);
3956                 info.minor = MINOR(rdev->bdev->bd_dev);
3957                 info.raid_disk = rdev->raid_disk;
3958                 info.state = 0;
3959                 if (test_bit(Faulty, &rdev->flags))
3960                         info.state |= (1<<MD_DISK_FAULTY);
3961                 else if (test_bit(In_sync, &rdev->flags)) {
3962                         info.state |= (1<<MD_DISK_ACTIVE);
3963                         info.state |= (1<<MD_DISK_SYNC);
3964                 }
3965                 if (test_bit(WriteMostly, &rdev->flags))
3966                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3967         } else {
3968                 info.major = info.minor = 0;
3969                 info.raid_disk = -1;
3970                 info.state = (1<<MD_DISK_REMOVED);
3971         }
3972
3973         if (copy_to_user(arg, &info, sizeof(info)))
3974                 return -EFAULT;
3975
3976         return 0;
3977 }
3978
3979 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3980 {
3981         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3982         mdk_rdev_t *rdev;
3983         dev_t dev = MKDEV(info->major,info->minor);
3984
3985         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3986                 return -EOVERFLOW;
3987
3988         if (!mddev->raid_disks) {
3989                 int err;
3990                 /* expecting a device which has a superblock */
3991                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3992                 if (IS_ERR(rdev)) {
3993                         printk(KERN_WARNING
3994                                 "md: md_import_device returned %ld\n",
3995                                 PTR_ERR(rdev));
3996                         return PTR_ERR(rdev);
3997                 }
3998                 if (!list_empty(&mddev->disks)) {
3999                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4000                                                         mdk_rdev_t, same_set);
4001                         int err = super_types[mddev->major_version]
4002                                 .load_super(rdev, rdev0, mddev->minor_version);
4003                         if (err < 0) {
4004                                 printk(KERN_WARNING
4005                                         "md: %s has different UUID to %s\n",
4006                                         bdevname(rdev->bdev,b),
4007                                         bdevname(rdev0->bdev,b2));
4008                                 export_rdev(rdev);
4009                                 return -EINVAL;
4010                         }
4011                 }
4012                 err = bind_rdev_to_array(rdev, mddev);
4013                 if (err)
4014                         export_rdev(rdev);
4015                 return err;
4016         }
4017
4018         /*
4019          * add_new_disk can be used once the array is assembled
4020          * to add "hot spares".  They must already have a superblock
4021          * written
4022          */
4023         if (mddev->pers) {
4024                 int err;
4025                 if (!mddev->pers->hot_add_disk) {
4026                         printk(KERN_WARNING
4027                                 "%s: personality does not support diskops!\n",
4028                                mdname(mddev));
4029                         return -EINVAL;
4030                 }
4031                 if (mddev->persistent)
4032                         rdev = md_import_device(dev, mddev->major_version,
4033                                                 mddev->minor_version);
4034                 else
4035                         rdev = md_import_device(dev, -1, -1);
4036                 if (IS_ERR(rdev)) {
4037                         printk(KERN_WARNING
4038                                 "md: md_import_device returned %ld\n",
4039                                 PTR_ERR(rdev));
4040                         return PTR_ERR(rdev);
4041                 }
4042                 /* set save_raid_disk if appropriate */
4043                 if (!mddev->persistent) {
4044                         if (info->state & (1<<MD_DISK_SYNC)  &&
4045                             info->raid_disk < mddev->raid_disks)
4046                                 rdev->raid_disk = info->raid_disk;
4047                         else
4048                                 rdev->raid_disk = -1;
4049                 } else
4050                         super_types[mddev->major_version].
4051                                 validate_super(mddev, rdev);
4052                 rdev->saved_raid_disk = rdev->raid_disk;
4053
4054                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4055                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4056                         set_bit(WriteMostly, &rdev->flags);
4057
4058                 rdev->raid_disk = -1;
4059                 err = bind_rdev_to_array(rdev, mddev);
4060                 if (!err && !mddev->pers->hot_remove_disk) {
4061                         /* If there is hot_add_disk but no hot_remove_disk
4062                          * then added disks for geometry changes,
4063                          * and should be added immediately.
4064                          */
4065                         super_types[mddev->major_version].
4066                                 validate_super(mddev, rdev);
4067                         err = mddev->pers->hot_add_disk(mddev, rdev);
4068                         if (err)
4069                                 unbind_rdev_from_array(rdev);
4070                 }
4071                 if (err)
4072                         export_rdev(rdev);
4073
4074                 md_update_sb(mddev, 1);
4075                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4076                 md_wakeup_thread(mddev->thread);
4077                 return err;
4078         }
4079
4080         /* otherwise, add_new_disk is only allowed
4081          * for major_version==0 superblocks
4082          */
4083         if (mddev->major_version != 0) {
4084                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4085                        mdname(mddev));
4086                 return -EINVAL;
4087         }
4088
4089         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4090                 int err;
4091                 rdev = md_import_device (dev, -1, 0);
4092                 if (IS_ERR(rdev)) {
4093                         printk(KERN_WARNING
4094                                 "md: error, md_import_device() returned %ld\n",
4095                                 PTR_ERR(rdev));
4096                         return PTR_ERR(rdev);
4097                 }
4098                 rdev->desc_nr = info->number;
4099                 if (info->raid_disk < mddev->raid_disks)
4100                         rdev->raid_disk = info->raid_disk;
4101                 else
4102                         rdev->raid_disk = -1;
4103
4104                 if (rdev->raid_disk < mddev->raid_disks)
4105                         if (info->state & (1<<MD_DISK_SYNC))
4106                                 set_bit(In_sync, &rdev->flags);
4107
4108                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4109                         set_bit(WriteMostly, &rdev->flags);
4110
4111                 if (!mddev->persistent) {
4112                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4113                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4114                 } else
4115                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4116                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4117
4118                 err = bind_rdev_to_array(rdev, mddev);
4119                 if (err) {
4120                         export_rdev(rdev);
4121                         return err;
4122                 }
4123         }
4124
4125         return 0;
4126 }
4127
4128 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4129 {
4130         char b[BDEVNAME_SIZE];
4131         mdk_rdev_t *rdev;
4132
4133         if (!mddev->pers)
4134                 return -ENODEV;
4135
4136         rdev = find_rdev(mddev, dev);
4137         if (!rdev)
4138                 return -ENXIO;
4139
4140         if (rdev->raid_disk >= 0)
4141                 goto busy;
4142
4143         kick_rdev_from_array(rdev);
4144         md_update_sb(mddev, 1);
4145         md_new_event(mddev);
4146
4147         return 0;
4148 busy:
4149         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
4150                 bdevname(rdev->bdev,b), mdname(mddev));
4151         return -EBUSY;
4152 }
4153
4154 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4155 {
4156         char b[BDEVNAME_SIZE];
4157         int err;
4158         unsigned int size;
4159         mdk_rdev_t *rdev;
4160
4161         if (!mddev->pers)
4162                 return -ENODEV;
4163
4164         if (mddev->major_version != 0) {
4165                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4166                         " version-0 superblocks.\n",
4167                         mdname(mddev));
4168                 return -EINVAL;
4169         }
4170         if (!mddev->pers->hot_add_disk) {
4171                 printk(KERN_WARNING
4172                         "%s: personality does not support diskops!\n",
4173                         mdname(mddev));
4174                 return -EINVAL;
4175         }
4176
4177         rdev = md_import_device (dev, -1, 0);
4178         if (IS_ERR(rdev)) {
4179                 printk(KERN_WARNING
4180                         "md: error, md_import_device() returned %ld\n",
4181                         PTR_ERR(rdev));
4182                 return -EINVAL;
4183         }
4184
4185         if (mddev->persistent)
4186                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4187         else
4188                 rdev->sb_offset =
4189                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4190
4191         size = calc_dev_size(rdev, mddev->chunk_size);
4192         rdev->size = size;
4193
4194         if (test_bit(Faulty, &rdev->flags)) {
4195                 printk(KERN_WARNING
4196                         "md: can not hot-add faulty %s disk to %s!\n",
4197                         bdevname(rdev->bdev,b), mdname(mddev));
4198                 err = -EINVAL;
4199                 goto abort_export;
4200         }
4201         clear_bit(In_sync, &rdev->flags);
4202         rdev->desc_nr = -1;
4203         rdev->saved_raid_disk = -1;
4204         err = bind_rdev_to_array(rdev, mddev);
4205         if (err)
4206                 goto abort_export;
4207
4208         /*
4209          * The rest should better be atomic, we can have disk failures
4210          * noticed in interrupt contexts ...
4211          */
4212
4213         if (rdev->desc_nr == mddev->max_disks) {
4214                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4215                         mdname(mddev));
4216                 err = -EBUSY;
4217                 goto abort_unbind_export;
4218         }
4219
4220         rdev->raid_disk = -1;
4221
4222         md_update_sb(mddev, 1);
4223
4224         /*
4225          * Kick recovery, maybe this spare has to be added to the
4226          * array immediately.
4227          */
4228         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4229         md_wakeup_thread(mddev->thread);
4230         md_new_event(mddev);
4231         return 0;
4232
4233 abort_unbind_export:
4234         unbind_rdev_from_array(rdev);
4235
4236 abort_export:
4237         export_rdev(rdev);
4238         return err;
4239 }
4240
4241 static int set_bitmap_file(mddev_t *mddev, int fd)
4242 {
4243         int err;
4244
4245         if (mddev->pers) {
4246                 if (!mddev->pers->quiesce)
4247                         return -EBUSY;
4248                 if (mddev->recovery || mddev->sync_thread)
4249                         return -EBUSY;
4250                 /* we should be able to change the bitmap.. */
4251         }
4252
4253
4254         if (fd >= 0) {
4255                 if (mddev->bitmap)
4256                         return -EEXIST; /* cannot add when bitmap is present */
4257                 mddev->bitmap_file = fget(fd);
4258
4259                 if (mddev->bitmap_file == NULL) {
4260                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4261                                mdname(mddev));
4262                         return -EBADF;
4263                 }
4264
4265                 err = deny_bitmap_write_access(mddev->bitmap_file);
4266                 if (err) {
4267                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4268                                mdname(mddev));
4269                         fput(mddev->bitmap_file);
4270                         mddev->bitmap_file = NULL;
4271                         return err;
4272                 }
4273                 mddev->bitmap_offset = 0; /* file overrides offset */
4274         } else if (mddev->bitmap == NULL)
4275                 return -ENOENT; /* cannot remove what isn't there */
4276         err = 0;
4277         if (mddev->pers) {
4278                 mddev->pers->quiesce(mddev, 1);
4279                 if (fd >= 0)
4280                         err = bitmap_create(mddev);
4281                 if (fd < 0 || err) {
4282                         bitmap_destroy(mddev);
4283                         fd = -1; /* make sure to put the file */
4284                 }
4285                 mddev->pers->quiesce(mddev, 0);
4286         }
4287         if (fd < 0) {
4288                 if (mddev->bitmap_file) {
4289                         restore_bitmap_write_access(mddev->bitmap_file);
4290                         fput(mddev->bitmap_file);
4291                 }
4292                 mddev->bitmap_file = NULL;
4293         }
4294
4295         return err;
4296 }
4297
4298 /*
4299  * set_array_info is used two different ways
4300  * The original usage is when creating a new array.
4301  * In this usage, raid_disks is > 0 and it together with
4302  *  level, size, not_persistent,layout,chunksize determine the
4303  *  shape of the array.
4304  *  This will always create an array with a type-0.90.0 superblock.
4305  * The newer usage is when assembling an array.
4306  *  In this case raid_disks will be 0, and the major_version field is
4307  *  use to determine which style super-blocks are to be found on the devices.
4308  *  The minor and patch _version numbers are also kept incase the
4309  *  super_block handler wishes to interpret them.
4310  */
4311 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4312 {
4313
4314         if (info->raid_disks == 0) {
4315                 /* just setting version number for superblock loading */
4316                 if (info->major_version < 0 ||
4317                     info->major_version >= ARRAY_SIZE(super_types) ||
4318                     super_types[info->major_version].name == NULL) {
4319                         /* maybe try to auto-load a module? */
4320                         printk(KERN_INFO
4321                                 "md: superblock version %d not known\n",
4322                                 info->major_version);
4323                         return -EINVAL;
4324                 }
4325                 mddev->major_version = info->major_version;
4326                 mddev->minor_version = info->minor_version;
4327                 mddev->patch_version = info->patch_version;
4328                 mddev->persistent = !info->not_persistent;
4329                 return 0;
4330         }
4331         mddev->major_version = MD_MAJOR_VERSION;
4332         mddev->minor_version = MD_MINOR_VERSION;
4333         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4334         mddev->ctime         = get_seconds();
4335
4336         mddev->level         = info->level;
4337         mddev->clevel[0]     = 0;
4338         mddev->size          = info->size;
4339         mddev->raid_disks    = info->raid_disks;
4340         /* don't set md_minor, it is determined by which /dev/md* was
4341          * openned
4342          */
4343         if (info->state & (1<<MD_SB_CLEAN))
4344                 mddev->recovery_cp = MaxSector;
4345         else
4346                 mddev->recovery_cp = 0;
4347         mddev->persistent    = ! info->not_persistent;
4348         mddev->external      = 0;
4349
4350         mddev->layout        = info->layout;
4351         mddev->chunk_size    = info->chunk_size;
4352
4353         mddev->max_disks     = MD_SB_DISKS;
4354
4355         if (mddev->persistent)
4356                 mddev->flags         = 0;
4357         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4358
4359         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4360         mddev->bitmap_offset = 0;
4361
4362         mddev->reshape_position = MaxSector;
4363
4364         /*
4365          * Generate a 128 bit UUID
4366          */
4367         get_random_bytes(mddev->uuid, 16);
4368
4369         mddev->new_level = mddev->level;
4370         mddev->new_chunk = mddev->chunk_size;
4371         mddev->new_layout = mddev->layout;
4372         mddev->delta_disks = 0;
4373
4374         return 0;
4375 }
4376
4377 static int update_size(mddev_t *mddev, unsigned long size)
4378 {
4379         mdk_rdev_t * rdev;
4380         int rv;
4381         struct list_head *tmp;
4382         int fit = (size == 0);
4383
4384         if (mddev->pers->resize == NULL)
4385                 return -EINVAL;
4386         /* The "size" is the amount of each device that is used.
4387          * This can only make sense for arrays with redundancy.
4388          * linear and raid0 always use whatever space is available
4389          * We can only consider changing the size if no resync
4390          * or reconstruction is happening, and if the new size
4391          * is acceptable. It must fit before the sb_offset or,
4392          * if that is <data_offset, it must fit before the
4393          * size of each device.
4394          * If size is zero, we find the largest size that fits.
4395          */
4396         if (mddev->sync_thread)
4397                 return -EBUSY;
4398         rdev_for_each(rdev, tmp, mddev) {
4399                 sector_t avail;
4400                 avail = rdev->size * 2;
4401
4402                 if (fit && (size == 0 || size > avail/2))
4403                         size = avail/2;
4404                 if (avail < ((sector_t)size << 1))
4405                         return -ENOSPC;
4406         }
4407         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4408         if (!rv) {
4409                 struct block_device *bdev;
4410
4411                 bdev = bdget_disk(mddev->gendisk, 0);
4412                 if (bdev) {
4413                         mutex_lock(&bdev->bd_inode->i_mutex);
4414                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4415                         mutex_unlock(&bdev->bd_inode->i_mutex);
4416                         bdput(bdev);
4417                 }
4418         }
4419         return rv;
4420 }
4421
4422 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4423 {
4424         int rv;
4425         /* change the number of raid disks */
4426         if (mddev->pers->check_reshape == NULL)
4427                 return -EINVAL;
4428         if (raid_disks <= 0 ||
4429             raid_disks >= mddev->max_disks)
4430                 return -EINVAL;
4431         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4432                 return -EBUSY;
4433         mddev->delta_disks = raid_disks - mddev->raid_disks;
4434
4435         rv = mddev->pers->check_reshape(mddev);
4436         return rv;
4437 }
4438
4439
4440 /*
4441  * update_array_info is used to change the configuration of an
4442  * on-line array.
4443  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4444  * fields in the info are checked against the array.
4445  * Any differences that cannot be handled will cause an error.
4446  * Normally, only one change can be managed at a time.
4447  */
4448 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4449 {
4450         int rv = 0;
4451         int cnt = 0;
4452         int state = 0;
4453
4454         /* calculate expected state,ignoring low bits */
4455         if (mddev->bitmap && mddev->bitmap_offset)
4456                 state |= (1 << MD_SB_BITMAP_PRESENT);
4457
4458         if (mddev->major_version != info->major_version ||
4459             mddev->minor_version != info->minor_version ||
4460 /*          mddev->patch_version != info->patch_version || */
4461             mddev->ctime         != info->ctime         ||
4462             mddev->level         != info->level         ||
4463 /*          mddev->layout        != info->layout        || */
4464             !mddev->persistent   != info->not_persistent||
4465             mddev->chunk_size    != info->chunk_size    ||
4466             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4467             ((state^info->state) & 0xfffffe00)
4468                 )
4469                 return -EINVAL;
4470         /* Check there is only one change */
4471         if (info->size >= 0 && mddev->size != info->size) cnt++;
4472         if (mddev->raid_disks != info->raid_disks) cnt++;
4473         if (mddev->layout != info->layout) cnt++;
4474         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4475         if (cnt == 0) return 0;
4476         if (cnt > 1) return -EINVAL;
4477
4478         if (mddev->layout != info->layout) {
4479                 /* Change layout
4480                  * we don't need to do anything at the md level, the
4481                  * personality will take care of it all.
4482                  */
4483                 if (mddev->pers->reconfig == NULL)
4484                         return -EINVAL;
4485                 else
4486                         return mddev->pers->reconfig(mddev, info->layout, -1);
4487         }
4488         if (info->size >= 0 && mddev->size != info->size)
4489                 rv = update_size(mddev, info->size);
4490
4491         if (mddev->raid_disks    != info->raid_disks)
4492                 rv = update_raid_disks(mddev, info->raid_disks);
4493
4494         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4495                 if (mddev->pers->quiesce == NULL)
4496                         return -EINVAL;
4497                 if (mddev->recovery || mddev->sync_thread)
4498                         return -EBUSY;
4499                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4500                         /* add the bitmap */
4501                         if (mddev->bitmap)
4502                                 return -EEXIST;
4503                         if (mddev->default_bitmap_offset == 0)
4504                                 return -EINVAL;
4505                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4506                         mddev->pers->quiesce(mddev, 1);
4507                         rv = bitmap_create(mddev);
4508                         if (rv)
4509                                 bitmap_destroy(mddev);
4510                         mddev->pers->quiesce(mddev, 0);
4511                 } else {
4512                         /* remove the bitmap */
4513                         if (!mddev->bitmap)
4514                                 return -ENOENT;
4515                         if (mddev->bitmap->file)
4516                                 return -EINVAL;
4517                         mddev->pers->quiesce(mddev, 1);
4518                         bitmap_destroy(mddev);
4519                         mddev->pers->quiesce(mddev, 0);
4520                         mddev->bitmap_offset = 0;
4521                 }
4522         }
4523         md_update_sb(mddev, 1);
4524         return rv;
4525 }
4526
4527 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4528 {
4529         mdk_rdev_t *rdev;
4530
4531         if (mddev->pers == NULL)
4532                 return -ENODEV;
4533
4534         rdev = find_rdev(mddev, dev);
4535         if (!rdev)
4536                 return -ENODEV;
4537
4538         md_error(mddev, rdev);
4539         return 0;
4540 }
4541
4542 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4543 {
4544         mddev_t *mddev = bdev->bd_disk->private_data;
4545
4546         geo->heads = 2;
4547         geo->sectors = 4;
4548         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4549         return 0;
4550 }
4551
4552 static int md_ioctl(struct inode *inode, struct file *file,
4553                         unsigned int cmd, unsigned long arg)
4554 {
4555         int err = 0;
4556         void __user *argp = (void __user *)arg;
4557         mddev_t *mddev = NULL;
4558
4559         if (!capable(CAP_SYS_ADMIN))
4560                 return -EACCES;
4561
4562         /*
4563          * Commands dealing with the RAID driver but not any
4564          * particular array:
4565          */
4566         switch (cmd)
4567         {
4568                 case RAID_VERSION:
4569                         err = get_version(argp);
4570                         goto done;
4571
4572                 case PRINT_RAID_DEBUG:
4573                         err = 0;
4574                         md_print_devices();
4575                         goto done;
4576
4577 #ifndef MODULE
4578                 case RAID_AUTORUN:
4579                         err = 0;
4580                         autostart_arrays(arg);
4581                         goto done;
4582 #endif
4583                 default:;
4584         }
4585
4586         /*
4587          * Commands creating/starting a new array:
4588          */
4589
4590         mddev = inode->i_bdev->bd_disk->private_data;
4591
4592         if (!mddev) {
4593                 BUG();
4594                 goto abort;
4595         }
4596
4597         err = mddev_lock(mddev);
4598         if (err) {
4599                 printk(KERN_INFO
4600                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4601                         err, cmd);
4602                 goto abort;
4603         }
4604
4605         switch (cmd)
4606         {
4607                 case SET_ARRAY_INFO:
4608                         {
4609                                 mdu_array_info_t info;
4610                                 if (!arg)
4611                                         memset(&info, 0, sizeof(info));
4612                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4613                                         err = -EFAULT;
4614                                         goto abort_unlock;
4615                                 }
4616                                 if (mddev->pers) {
4617                                         err = update_array_info(mddev, &info);
4618                                         if (err) {
4619                                                 printk(KERN_WARNING "md: couldn't update"
4620                                                        " array info. %d\n", err);
4621                                                 goto abort_unlock;
4622                                         }
4623                                         goto done_unlock;
4624                                 }
4625                                 if (!list_empty(&mddev->disks)) {
4626                                         printk(KERN_WARNING
4627                                                "md: array %s already has disks!\n",
4628                                                mdname(mddev));
4629                                         err = -EBUSY;
4630                                         goto abort_unlock;
4631                                 }
4632                                 if (mddev->raid_disks) {
4633                                         printk(KERN_WARNING
4634                                                "md: array %s already initialised!\n",
4635                                                mdname(mddev));
4636                                         err = -EBUSY;
4637                                         goto abort_unlock;
4638                                 }
4639                                 err = set_array_info(mddev, &info);
4640                                 if (err) {
4641                                         printk(KERN_WARNING "md: couldn't set"
4642                                                " array info. %d\n", err);
4643                                         goto abort_unlock;
4644                                 }
4645                         }
4646                         goto done_unlock;
4647
4648                 default:;
4649         }
4650
4651         /*
4652          * Commands querying/configuring an existing array:
4653          */
4654         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4655          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4656         if ((!mddev->raid_disks && !mddev->external)
4657             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4658             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4659             && cmd != GET_BITMAP_FILE) {
4660                 err = -ENODEV;
4661                 goto abort_unlock;
4662         }
4663
4664         /*
4665          * Commands even a read-only array can execute:
4666          */
4667         switch (cmd)
4668         {
4669                 case GET_ARRAY_INFO:
4670                         err = get_array_info(mddev, argp);
4671                         goto done_unlock;
4672
4673                 case GET_BITMAP_FILE:
4674                         err = get_bitmap_file(mddev, argp);
4675                         goto done_unlock;
4676
4677                 case GET_DISK_INFO:
4678                         err = get_disk_info(mddev, argp);
4679                         goto done_unlock;
4680
4681                 case RESTART_ARRAY_RW:
4682                         err = restart_array(mddev);
4683                         goto done_unlock;
4684
4685                 case STOP_ARRAY:
4686                         err = do_md_stop (mddev, 0);
4687                         goto done_unlock;
4688
4689                 case STOP_ARRAY_RO:
4690                         err = do_md_stop (mddev, 1);
4691                         goto done_unlock;
4692
4693         /*
4694          * We have a problem here : there is no easy way to give a CHS
4695          * virtual geometry. We currently pretend that we have a 2 heads
4696          * 4 sectors (with a BIG number of cylinders...). This drives
4697          * dosfs just mad... ;-)
4698          */
4699         }
4700
4701         /*
4702          * The remaining ioctls are changing the state of the
4703          * superblock, so we do not allow them on read-only arrays.
4704          * However non-MD ioctls (e.g. get-size) will still come through
4705          * here and hit the 'default' below, so only disallow
4706          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4707          */
4708         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4709             mddev->ro && mddev->pers) {
4710                 if (mddev->ro == 2) {
4711                         mddev->ro = 0;
4712                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4713                 md_wakeup_thread(mddev->thread);
4714
4715                 } else {
4716                         err = -EROFS;
4717                         goto abort_unlock;
4718                 }
4719         }
4720
4721         switch (cmd)
4722         {
4723                 case ADD_NEW_DISK:
4724                 {
4725                         mdu_disk_info_t info;
4726                         if (copy_from_user(&info, argp, sizeof(info)))
4727                                 err = -EFAULT;
4728                         else
4729                                 err = add_new_disk(mddev, &info);
4730                         goto done_unlock;
4731                 }
4732
4733                 case HOT_REMOVE_DISK:
4734                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4735                         goto done_unlock;
4736
4737                 case HOT_ADD_DISK:
4738                         err = hot_add_disk(mddev, new_decode_dev(arg));
4739                         goto done_unlock;
4740
4741                 case SET_DISK_FAULTY:
4742                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4743                         goto done_unlock;
4744
4745                 case RUN_ARRAY:
4746                         err = do_md_run (mddev);
4747                         goto done_unlock;
4748
4749                 case SET_BITMAP_FILE:
4750                         err = set_bitmap_file(mddev, (int)arg);
4751                         goto done_unlock;
4752
4753                 default:
4754                         err = -EINVAL;
4755                         goto abort_unlock;
4756         }
4757
4758 done_unlock:
4759 abort_unlock:
4760         mddev_unlock(mddev);
4761
4762         return err;
4763 done:
4764         if (err)
4765                 MD_BUG();
4766 abort:
4767         return err;
4768 }
4769
4770 static int md_open(struct inode *inode, struct file *file)
4771 {
4772         /*
4773          * Succeed if we can lock the mddev, which confirms that
4774          * it isn't being stopped right now.
4775          */
4776         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4777         int err;
4778
4779         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4780                 goto out;
4781
4782         err = 0;
4783         mddev_get(mddev);
4784         mddev_unlock(mddev);
4785
4786         check_disk_change(inode->i_bdev);
4787  out:
4788         return err;
4789 }
4790
4791 static int md_release(struct inode *inode, struct file * file)
4792 {
4793         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4794
4795         BUG_ON(!mddev);
4796         mddev_put(mddev);
4797
4798         return 0;
4799 }
4800
4801 static int md_media_changed(struct gendisk *disk)
4802 {
4803         mddev_t *mddev = disk->private_data;
4804
4805         return mddev->changed;
4806 }
4807
4808 static int md_revalidate(struct gendisk *disk)
4809 {
4810         mddev_t *mddev = disk->private_data;
4811
4812         mddev->changed = 0;
4813         return 0;
4814 }
4815 static struct block_device_operations md_fops =
4816 {
4817         .owner          = THIS_MODULE,
4818         .open           = md_open,
4819         .release        = md_release,
4820         .ioctl          = md_ioctl,
4821         .getgeo         = md_getgeo,
4822         .media_changed  = md_media_changed,
4823         .revalidate_disk= md_revalidate,
4824 };
4825
4826 static int md_thread(void * arg)
4827 {
4828         mdk_thread_t *thread = arg;
4829
4830         /*
4831          * md_thread is a 'system-thread', it's priority should be very
4832          * high. We avoid resource deadlocks individually in each
4833          * raid personality. (RAID5 does preallocation) We also use RR and
4834          * the very same RT priority as kswapd, thus we will never get
4835          * into a priority inversion deadlock.
4836          *
4837          * we definitely have to have equal or higher priority than
4838          * bdflush, otherwise bdflush will deadlock if there are too
4839          * many dirty RAID5 blocks.
4840          */
4841
4842         allow_signal(SIGKILL);
4843         while (!kthread_should_stop()) {
4844
4845                 /* We need to wait INTERRUPTIBLE so that
4846                  * we don't add to the load-average.
4847                  * That means we need to be sure no signals are
4848                  * pending
4849                  */
4850                 if (signal_pending(current))
4851                         flush_signals(current);
4852
4853                 wait_event_interruptible_timeout
4854                         (thread->wqueue,
4855                          test_bit(THREAD_WAKEUP, &thread->flags)
4856                          || kthread_should_stop(),
4857                          thread->timeout);
4858
4859                 clear_bit(THREAD_WAKEUP, &thread->flags);
4860
4861                 thread->run(thread->mddev);
4862         }
4863
4864         return 0;
4865 }
4866
4867 void md_wakeup_thread(mdk_thread_t *thread)
4868 {
4869         if (thread) {
4870                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4871                 set_bit(THREAD_WAKEUP, &thread->flags);
4872                 wake_up(&thread->wqueue);
4873         }
4874 }
4875
4876 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4877                                  const char *name)
4878 {
4879         mdk_thread_t *thread;
4880
4881         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4882         if (!thread)
4883                 return NULL;
4884
4885         init_waitqueue_head(&thread->wqueue);
4886
4887         thread->run = run;
4888         thread->mddev = mddev;
4889         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4890         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4891         if (IS_ERR(thread->tsk)) {
4892                 kfree(thread);
4893                 return NULL;
4894         }
4895         return thread;
4896 }
4897
4898 void md_unregister_thread(mdk_thread_t *thread)
4899 {
4900         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4901
4902         kthread_stop(thread->tsk);
4903         kfree(thread);
4904 }
4905
4906 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4907 {
4908         if (!mddev) {
4909                 MD_BUG();
4910                 return;
4911         }
4912
4913         if (!rdev || test_bit(Faulty, &rdev->flags))
4914                 return;
4915 /*
4916         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4917                 mdname(mddev),
4918                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4919                 __builtin_return_address(0),__builtin_return_address(1),
4920                 __builtin_return_address(2),__builtin_return_address(3));
4921 */
4922         if (!mddev->pers)
4923                 return;
4924         if (!mddev->pers->error_handler)
4925                 return;
4926         mddev->pers->error_handler(mddev,rdev);
4927         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4928         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4929         md_wakeup_thread(mddev->thread);
4930         md_new_event_inintr(mddev);
4931 }
4932
4933 /* seq_file implementation /proc/mdstat */
4934
4935 static void status_unused(struct seq_file *seq)
4936 {
4937         int i = 0;
4938         mdk_rdev_t *rdev;
4939         struct list_head *tmp;
4940
4941         seq_printf(seq, "unused devices: ");
4942
4943         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
4944                 char b[BDEVNAME_SIZE];
4945                 i++;
4946                 seq_printf(seq, "%s ",
4947                               bdevname(rdev->bdev,b));
4948         }
4949         if (!i)
4950                 seq_printf(seq, "<none>");
4951
4952         seq_printf(seq, "\n");
4953 }
4954
4955
4956 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4957 {
4958         sector_t max_blocks, resync, res;
4959         unsigned long dt, db, rt;
4960         int scale;
4961         unsigned int per_milli;
4962
4963         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4964
4965         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4966                 max_blocks = mddev->resync_max_sectors >> 1;
4967         else
4968                 max_blocks = mddev->size;
4969
4970         /*
4971          * Should not happen.
4972          */
4973         if (!max_blocks) {
4974                 MD_BUG();
4975                 return;
4976         }
4977         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4978          * in a sector_t, and (max_blocks>>scale) will fit in a
4979          * u32, as those are the requirements for sector_div.
4980          * Thus 'scale' must be at least 10
4981          */
4982         scale = 10;
4983         if (sizeof(sector_t) > sizeof(unsigned long)) {
4984                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4985                         scale++;
4986         }
4987         res = (resync>>scale)*1000;
4988         sector_div(res, (u32)((max_blocks>>scale)+1));
4989
4990         per_milli = res;
4991         {
4992                 int i, x = per_milli/50, y = 20-x;
4993                 seq_printf(seq, "[");
4994                 for (i = 0; i < x; i++)
4995                         seq_printf(seq, "=");
4996                 seq_printf(seq, ">");
4997                 for (i = 0; i < y; i++)
4998                         seq_printf(seq, ".");
4999                 seq_printf(seq, "] ");
5000         }
5001         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5002                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5003                     "reshape" :
5004                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5005                      "check" :
5006                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5007                       "resync" : "recovery"))),
5008                    per_milli/10, per_milli % 10,
5009                    (unsigned long long) resync,
5010                    (unsigned long long) max_blocks);
5011
5012         /*
5013          * We do not want to overflow, so the order of operands and
5014          * the * 100 / 100 trick are important. We do a +1 to be
5015          * safe against division by zero. We only estimate anyway.
5016          *
5017          * dt: time from mark until now
5018          * db: blocks written from mark until now
5019          * rt: remaining time
5020          */
5021         dt = ((jiffies - mddev->resync_mark) / HZ);
5022         if (!dt) dt++;
5023         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5024                 - mddev->resync_mark_cnt;
5025         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5026
5027         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5028
5029         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5030 }
5031
5032 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5033 {
5034         struct list_head *tmp;
5035         loff_t l = *pos;
5036         mddev_t *mddev;
5037
5038         if (l >= 0x10000)
5039                 return NULL;
5040         if (!l--)
5041                 /* header */
5042                 return (void*)1;
5043
5044         spin_lock(&all_mddevs_lock);
5045         list_for_each(tmp,&all_mddevs)
5046                 if (!l--) {
5047                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5048                         mddev_get(mddev);
5049                         spin_unlock(&all_mddevs_lock);
5050                         return mddev;
5051                 }
5052         spin_unlock(&all_mddevs_lock);
5053         if (!l--)
5054                 return (void*)2;/* tail */
5055         return NULL;
5056 }
5057
5058 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5059 {
5060         struct list_head *tmp;
5061         mddev_t *next_mddev, *mddev = v;
5062
5063         ++*pos;
5064         if (v == (void*)2)
5065                 return NULL;
5066
5067         spin_lock(&all_mddevs_lock);
5068         if (v == (void*)1)
5069                 tmp = all_mddevs.next;
5070         else
5071                 tmp = mddev->all_mddevs.next;
5072         if (tmp != &all_mddevs)
5073                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5074         else {
5075                 next_mddev = (void*)2;
5076                 *pos = 0x10000;
5077         }
5078         spin_unlock(&all_mddevs_lock);
5079
5080         if (v != (void*)1)
5081                 mddev_put(mddev);
5082         return next_mddev;
5083
5084 }
5085
5086 static void md_seq_stop(struct seq_file *seq, void *v)
5087 {
5088         mddev_t *mddev = v;
5089
5090         if (mddev && v != (void*)1 && v != (void*)2)
5091                 mddev_put(mddev);
5092 }
5093
5094 struct mdstat_info {
5095         int event;
5096 };
5097
5098 static int md_seq_show(struct seq_file *seq, void *v)
5099 {
5100         mddev_t *mddev = v;
5101         sector_t size;
5102         struct list_head *tmp2;
5103         mdk_rdev_t *rdev;
5104         struct mdstat_info *mi = seq->private;
5105         struct bitmap *bitmap;
5106
5107         if (v == (void*)1) {
5108                 struct mdk_personality *pers;
5109                 seq_printf(seq, "Personalities : ");
5110                 spin_lock(&pers_lock);
5111                 list_for_each_entry(pers, &pers_list, list)
5112                         seq_printf(seq, "[%s] ", pers->name);
5113
5114                 spin_unlock(&pers_lock);
5115                 seq_printf(seq, "\n");
5116                 mi->event = atomic_read(&md_event_count);
5117                 return 0;
5118         }
5119         if (v == (void*)2) {
5120                 status_unused(seq);
5121                 return 0;
5122         }
5123
5124         if (mddev_lock(mddev) < 0)
5125                 return -EINTR;
5126
5127         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5128                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5129                                                 mddev->pers ? "" : "in");
5130                 if (mddev->pers) {
5131                         if (mddev->ro==1)
5132                                 seq_printf(seq, " (read-only)");
5133                         if (mddev->ro==2)
5134                                 seq_printf(seq, "(auto-read-only)");
5135                         seq_printf(seq, " %s", mddev->pers->name);
5136                 }
5137
5138                 size = 0;
5139                 rdev_for_each(rdev, tmp2, mddev) {
5140                         char b[BDEVNAME_SIZE];
5141                         seq_printf(seq, " %s[%d]",
5142                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5143                         if (test_bit(WriteMostly, &rdev->flags))
5144                                 seq_printf(seq, "(W)");
5145                         if (test_bit(Faulty, &rdev->flags)) {
5146                                 seq_printf(seq, "(F)");
5147                                 continue;
5148                         } else if (rdev->raid_disk < 0)
5149                                 seq_printf(seq, "(S)"); /* spare */
5150                         size += rdev->size;
5151                 }
5152
5153                 if (!list_empty(&mddev->disks)) {
5154                         if (mddev->pers)
5155                                 seq_printf(seq, "\n      %llu blocks",
5156                                         (unsigned long long)mddev->array_size);
5157                         else
5158                                 seq_printf(seq, "\n      %llu blocks",
5159                                         (unsigned long long)size);
5160                 }
5161                 if (mddev->persistent) {
5162                         if (mddev->major_version != 0 ||
5163                             mddev->minor_version != 90) {
5164                                 seq_printf(seq," super %d.%d",
5165                                            mddev->major_version,
5166                                            mddev->minor_version);
5167                         }
5168                 } else if (mddev->external)
5169                         seq_printf(seq, " super external:%s",
5170                                    mddev->metadata_type);
5171                 else
5172                         seq_printf(seq, " super non-persistent");
5173
5174                 if (mddev->pers) {
5175                         mddev->pers->status (seq, mddev);
5176                         seq_printf(seq, "\n      ");
5177                         if (mddev->pers->sync_request) {
5178                                 if (mddev->curr_resync > 2) {
5179                                         status_resync (seq, mddev);
5180                                         seq_printf(seq, "\n      ");
5181                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5182                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5183                                 else if (mddev->recovery_cp < MaxSector)
5184                                         seq_printf(seq, "\tresync=PENDING\n      ");
5185                         }
5186                 } else
5187                         seq_printf(seq, "\n       ");
5188
5189                 if ((bitmap = mddev->bitmap)) {
5190                         unsigned long chunk_kb;
5191                         unsigned long flags;
5192                         spin_lock_irqsave(&bitmap->lock, flags);
5193                         chunk_kb = bitmap->chunksize >> 10;
5194                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5195                                 "%lu%s chunk",
5196                                 bitmap->pages - bitmap->missing_pages,
5197                                 bitmap->pages,
5198                                 (bitmap->pages - bitmap->missing_pages)
5199                                         << (PAGE_SHIFT - 10),
5200                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5201                                 chunk_kb ? "KB" : "B");
5202                         if (bitmap->file) {
5203                                 seq_printf(seq, ", file: ");
5204                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5205                         }
5206
5207                         seq_printf(seq, "\n");
5208                         spin_unlock_irqrestore(&bitmap->lock, flags);
5209                 }
5210
5211                 seq_printf(seq, "\n");
5212         }
5213         mddev_unlock(mddev);
5214
5215         return 0;
5216 }
5217
5218 static struct seq_operations md_seq_ops = {
5219         .start  = md_seq_start,
5220         .next   = md_seq_next,
5221         .stop   = md_seq_stop,
5222         .show   = md_seq_show,
5223 };
5224
5225 static int md_seq_open(struct inode *inode, struct file *file)
5226 {
5227         int error;
5228         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5229         if (mi == NULL)
5230                 return -ENOMEM;
5231
5232         error = seq_open(file, &md_seq_ops);
5233         if (error)
5234                 kfree(mi);
5235         else {
5236                 struct seq_file *p = file->private_data;
5237                 p->private = mi;
5238                 mi->event = atomic_read(&md_event_count);
5239         }
5240         return error;
5241 }
5242
5243 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5244 {
5245         struct seq_file *m = filp->private_data;
5246         struct mdstat_info *mi = m->private;
5247         int mask;
5248
5249         poll_wait(filp, &md_event_waiters, wait);
5250
5251         /* always allow read */
5252         mask = POLLIN | POLLRDNORM;
5253
5254         if (mi->event != atomic_read(&md_event_count))
5255                 mask |= POLLERR | POLLPRI;
5256         return mask;
5257 }
5258
5259 static const struct file_operations md_seq_fops = {
5260         .owner          = THIS_MODULE,
5261         .open           = md_seq_open,
5262         .read           = seq_read,
5263         .llseek         = seq_lseek,
5264         .release        = seq_release_private,
5265         .poll           = mdstat_poll,
5266 };
5267
5268 int register_md_personality(struct mdk_personality *p)
5269 {
5270         spin_lock(&pers_lock);
5271         list_add_tail(&p->list, &pers_list);
5272         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5273         spin_unlock(&pers_lock);
5274         return 0;
5275 }
5276
5277 int unregister_md_personality(struct mdk_personality *p)
5278 {
5279         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5280         spin_lock(&pers_lock);
5281         list_del_init(&p->list);
5282         spin_unlock(&pers_lock);
5283         return 0;
5284 }
5285
5286 static int is_mddev_idle(mddev_t *mddev)
5287 {
5288         mdk_rdev_t * rdev;
5289         struct list_head *tmp;
5290         int idle;
5291         long curr_events;
5292
5293         idle = 1;
5294         rdev_for_each(rdev, tmp, mddev) {
5295                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5296                 curr_events = disk_stat_read(disk, sectors[0]) +
5297                                 disk_stat_read(disk, sectors[1]) -
5298                                 atomic_read(&disk->sync_io);
5299                 /* sync IO will cause sync_io to increase before the disk_stats
5300                  * as sync_io is counted when a request starts, and
5301                  * disk_stats is counted when it completes.
5302                  * So resync activity will cause curr_events to be smaller than
5303                  * when there was no such activity.
5304                  * non-sync IO will cause disk_stat to increase without
5305                  * increasing sync_io so curr_events will (eventually)
5306                  * be larger than it was before.  Once it becomes
5307                  * substantially larger, the test below will cause
5308                  * the array to appear non-idle, and resync will slow
5309                  * down.
5310                  * If there is a lot of outstanding resync activity when
5311                  * we set last_event to curr_events, then all that activity
5312                  * completing might cause the array to appear non-idle
5313                  * and resync will be slowed down even though there might
5314                  * not have been non-resync activity.  This will only
5315                  * happen once though.  'last_events' will soon reflect
5316                  * the state where there is little or no outstanding
5317                  * resync requests, and further resync activity will
5318                  * always make curr_events less than last_events.
5319                  *
5320                  */
5321                 if (curr_events - rdev->last_events > 4096) {
5322                         rdev->last_events = curr_events;
5323                         idle = 0;
5324                 }
5325         }
5326         return idle;
5327 }
5328
5329 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5330 {
5331         /* another "blocks" (512byte) blocks have been synced */
5332         atomic_sub(blocks, &mddev->recovery_active);
5333         wake_up(&mddev->recovery_wait);
5334         if (!ok) {
5335                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5336                 md_wakeup_thread(mddev->thread);
5337                 // stop recovery, signal do_sync ....
5338         }
5339 }
5340
5341
5342 /* md_write_start(mddev, bi)
5343  * If we need to update some array metadata (e.g. 'active' flag
5344  * in superblock) before writing, schedule a superblock update
5345  * and wait for it to complete.
5346  */
5347 void md_write_start(mddev_t *mddev, struct bio *bi)
5348 {
5349         if (bio_data_dir(bi) != WRITE)
5350                 return;
5351
5352         BUG_ON(mddev->ro == 1);
5353         if (mddev->ro == 2) {
5354                 /* need to switch to read/write */
5355                 mddev->ro = 0;
5356                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5357                 md_wakeup_thread(mddev->thread);
5358         }
5359         atomic_inc(&mddev->writes_pending);
5360         if (mddev->in_sync) {
5361                 spin_lock_irq(&mddev->write_lock);
5362                 if (mddev->in_sync) {
5363                         mddev->in_sync = 0;
5364                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5365                         md_wakeup_thread(mddev->thread);
5366                 }
5367                 spin_unlock_irq(&mddev->write_lock);
5368         }
5369         wait_event(mddev->sb_wait, mddev->flags==0);
5370 }
5371
5372 void md_write_end(mddev_t *mddev)
5373 {
5374         if (atomic_dec_and_test(&mddev->writes_pending)) {
5375                 if (mddev->safemode == 2)
5376                         md_wakeup_thread(mddev->thread);
5377                 else if (mddev->safemode_delay)
5378                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5379         }
5380 }
5381
5382 /* md_allow_write(mddev)
5383  * Calling this ensures that the array is marked 'active' so that writes
5384  * may proceed without blocking.  It is important to call this before
5385  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5386  * Must be called with mddev_lock held.
5387  */
5388 void md_allow_write(mddev_t *mddev)
5389 {
5390         if (!mddev->pers)
5391                 return;
5392         if (mddev->ro)
5393                 return;
5394
5395         spin_lock_irq(&mddev->write_lock);
5396         if (mddev->in_sync) {
5397                 mddev->in_sync = 0;
5398                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5399                 if (mddev->safemode_delay &&
5400                     mddev->safemode == 0)
5401                         mddev->safemode = 1;
5402                 spin_unlock_irq(&mddev->write_lock);
5403                 md_update_sb(mddev, 0);
5404         } else
5405                 spin_unlock_irq(&mddev->write_lock);
5406 }
5407 EXPORT_SYMBOL_GPL(md_allow_write);
5408
5409 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5410
5411 #define SYNC_MARKS      10
5412 #define SYNC_MARK_STEP  (3*HZ)
5413 void md_do_sync(mddev_t *mddev)
5414 {
5415         mddev_t *mddev2;
5416         unsigned int currspeed = 0,
5417                  window;
5418         sector_t max_sectors,j, io_sectors;
5419         unsigned long mark[SYNC_MARKS];
5420         sector_t mark_cnt[SYNC_MARKS];
5421         int last_mark,m;
5422         struct list_head *tmp;
5423         sector_t last_check;
5424         int skipped = 0;
5425         struct list_head *rtmp;
5426         mdk_rdev_t *rdev;
5427         char *desc;
5428
5429         /* just incase thread restarts... */
5430         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5431                 return;
5432         if (mddev->ro) /* never try to sync a read-only array */
5433                 return;
5434
5435         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5436                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5437                         desc = "data-check";
5438                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5439                         desc = "requested-resync";
5440                 else
5441                         desc = "resync";
5442         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5443                 desc = "reshape";
5444         else
5445                 desc = "recovery";
5446
5447         /* we overload curr_resync somewhat here.
5448          * 0 == not engaged in resync at all
5449          * 2 == checking that there is no conflict with another sync
5450          * 1 == like 2, but have yielded to allow conflicting resync to
5451          *              commense
5452          * other == active in resync - this many blocks
5453          *
5454          * Before starting a resync we must have set curr_resync to
5455          * 2, and then checked that every "conflicting" array has curr_resync
5456          * less than ours.  When we find one that is the same or higher
5457          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5458          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5459          * This will mean we have to start checking from the beginning again.
5460          *
5461          */
5462
5463         do {
5464                 mddev->curr_resync = 2;
5465
5466         try_again:
5467                 if (kthread_should_stop()) {
5468                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5469                         goto skip;
5470                 }
5471                 for_each_mddev(mddev2, tmp) {
5472                         if (mddev2 == mddev)
5473                                 continue;
5474                         if (mddev2->curr_resync &&
5475                             match_mddev_units(mddev,mddev2)) {
5476                                 DEFINE_WAIT(wq);
5477                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5478                                         /* arbitrarily yield */
5479                                         mddev->curr_resync = 1;
5480                                         wake_up(&resync_wait);
5481                                 }
5482                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5483                                         /* no need to wait here, we can wait the next
5484                                          * time 'round when curr_resync == 2
5485                                          */
5486                                         continue;
5487                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5488                                 if (!kthread_should_stop() &&
5489                                     mddev2->curr_resync >= mddev->curr_resync) {
5490                                         printk(KERN_INFO "md: delaying %s of %s"
5491                                                " until %s has finished (they"
5492                                                " share one or more physical units)\n",
5493                                                desc, mdname(mddev), mdname(mddev2));
5494                                         mddev_put(mddev2);
5495                                         schedule();
5496                                         finish_wait(&resync_wait, &wq);
5497                                         goto try_again;
5498                                 }
5499                                 finish_wait(&resync_wait, &wq);
5500                         }
5501                 }
5502         } while (mddev->curr_resync < 2);
5503
5504         j = 0;
5505         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5506                 /* resync follows the size requested by the personality,
5507                  * which defaults to physical size, but can be virtual size
5508                  */
5509                 max_sectors = mddev->resync_max_sectors;
5510                 mddev->resync_mismatches = 0;
5511                 /* we don't use the checkpoint if there's a bitmap */
5512                 if (!mddev->bitmap &&
5513                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5514                         j = mddev->recovery_cp;
5515         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5516                 max_sectors = mddev->size << 1;
5517         else {
5518                 /* recovery follows the physical size of devices */
5519                 max_sectors = mddev->size << 1;
5520                 j = MaxSector;
5521                 rdev_for_each(rdev, rtmp, mddev)
5522                         if (rdev->raid_disk >= 0 &&
5523                             !test_bit(Faulty, &rdev->flags) &&
5524                             !test_bit(In_sync, &rdev->flags) &&
5525                             rdev->recovery_offset < j)
5526                                 j = rdev->recovery_offset;
5527         }
5528
5529         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5530         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5531                 " %d KB/sec/disk.\n", speed_min(mddev));
5532         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5533                "(but not more than %d KB/sec) for %s.\n",
5534                speed_max(mddev), desc);
5535
5536         is_mddev_idle(mddev); /* this also initializes IO event counters */
5537
5538         io_sectors = 0;
5539         for (m = 0; m < SYNC_MARKS; m++) {
5540                 mark[m] = jiffies;
5541                 mark_cnt[m] = io_sectors;
5542         }
5543         last_mark = 0;
5544         mddev->resync_mark = mark[last_mark];
5545         mddev->resync_mark_cnt = mark_cnt[last_mark];
5546
5547         /*
5548          * Tune reconstruction:
5549          */
5550         window = 32*(PAGE_SIZE/512);
5551         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5552                 window/2,(unsigned long long) max_sectors/2);
5553
5554         atomic_set(&mddev->recovery_active, 0);
5555         init_waitqueue_head(&mddev->recovery_wait);
5556         last_check = 0;
5557
5558         if (j>2) {
5559                 printk(KERN_INFO
5560                        "md: resuming %s of %s from checkpoint.\n",
5561                        desc, mdname(mddev));
5562                 mddev->curr_resync = j;
5563         }
5564
5565         while (j < max_sectors) {
5566                 sector_t sectors;
5567
5568                 skipped = 0;
5569                 if (j >= mddev->resync_max) {
5570                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5571                         wait_event(mddev->recovery_wait,
5572                                    mddev->resync_max > j
5573                                    || kthread_should_stop());
5574                 }
5575                 if (kthread_should_stop())
5576                         goto interrupted;
5577                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5578                                                   currspeed < speed_min(mddev));
5579                 if (sectors == 0) {
5580                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5581                         goto out;
5582                 }
5583
5584                 if (!skipped) { /* actual IO requested */
5585                         io_sectors += sectors;
5586                         atomic_add(sectors, &mddev->recovery_active);
5587                 }
5588
5589                 j += sectors;
5590                 if (j>1) mddev->curr_resync = j;
5591                 mddev->curr_mark_cnt = io_sectors;
5592                 if (last_check == 0)
5593                         /* this is the earliers that rebuilt will be
5594                          * visible in /proc/mdstat
5595                          */
5596                         md_new_event(mddev);
5597
5598                 if (last_check + window > io_sectors || j == max_sectors)
5599                         continue;
5600
5601                 last_check = io_sectors;
5602
5603                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5604                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5605                         break;
5606
5607         repeat:
5608                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5609                         /* step marks */
5610                         int next = (last_mark+1) % SYNC_MARKS;
5611
5612                         mddev->resync_mark = mark[next];
5613                         mddev->resync_mark_cnt = mark_cnt[next];
5614                         mark[next] = jiffies;
5615                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5616                         last_mark = next;
5617                 }
5618
5619
5620                 if (kthread_should_stop())
5621                         goto interrupted;
5622
5623
5624                 /*
5625                  * this loop exits only if either when we are slower than
5626                  * the 'hard' speed limit, or the system was IO-idle for
5627                  * a jiffy.
5628                  * the system might be non-idle CPU-wise, but we only care
5629                  * about not overloading the IO subsystem. (things like an
5630                  * e2fsck being done on the RAID array should execute fast)
5631                  */
5632                 blk_unplug(mddev->queue);
5633                 cond_resched();
5634
5635                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5636                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5637
5638                 if (currspeed > speed_min(mddev)) {
5639                         if ((currspeed > speed_max(mddev)) ||
5640                                         !is_mddev_idle(mddev)) {
5641                                 msleep(500);
5642                                 goto repeat;
5643                         }
5644                 }
5645         }
5646         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5647         /*
5648          * this also signals 'finished resyncing' to md_stop
5649          */
5650  out:
5651         blk_unplug(mddev->queue);
5652
5653         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5654
5655         /* tell personality that we are finished */
5656         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5657
5658         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5659             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5660             mddev->curr_resync > 2) {
5661                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5662                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5663                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5664                                         printk(KERN_INFO
5665                                                "md: checkpointing %s of %s.\n",
5666                                                desc, mdname(mddev));
5667                                         mddev->recovery_cp = mddev->curr_resync;
5668                                 }
5669                         } else
5670                                 mddev->recovery_cp = MaxSector;
5671                 } else {
5672                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5673                                 mddev->curr_resync = MaxSector;
5674                         rdev_for_each(rdev, rtmp, mddev)
5675                                 if (rdev->raid_disk >= 0 &&
5676                                     !test_bit(Faulty, &rdev->flags) &&
5677                                     !test_bit(In_sync, &rdev->flags) &&
5678                                     rdev->recovery_offset < mddev->curr_resync)
5679                                         rdev->recovery_offset = mddev->curr_resync;
5680                 }
5681         }
5682         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5683
5684  skip:
5685         mddev->curr_resync = 0;
5686         mddev->resync_max = MaxSector;
5687         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5688         wake_up(&resync_wait);
5689         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5690         md_wakeup_thread(mddev->thread);
5691         return;
5692
5693  interrupted:
5694         /*
5695          * got a signal, exit.
5696          */
5697         printk(KERN_INFO
5698                "md: md_do_sync() got signal ... exiting\n");
5699         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5700         goto out;
5701
5702 }
5703 EXPORT_SYMBOL_GPL(md_do_sync);
5704
5705
5706 static int remove_and_add_spares(mddev_t *mddev)
5707 {
5708         mdk_rdev_t *rdev;
5709         struct list_head *rtmp;
5710         int spares = 0;
5711
5712         rdev_for_each(rdev, rtmp, mddev)
5713                 if (rdev->raid_disk >= 0 &&
5714                     !mddev->external &&
5715                     (test_bit(Faulty, &rdev->flags) ||
5716                      ! test_bit(In_sync, &rdev->flags)) &&
5717                     atomic_read(&rdev->nr_pending)==0) {
5718                         if (mddev->pers->hot_remove_disk(
5719                                     mddev, rdev->raid_disk)==0) {
5720                                 char nm[20];
5721                                 sprintf(nm,"rd%d", rdev->raid_disk);
5722                                 sysfs_remove_link(&mddev->kobj, nm);
5723                                 rdev->raid_disk = -1;
5724                         }
5725                 }
5726
5727         if (mddev->degraded) {
5728                 rdev_for_each(rdev, rtmp, mddev)
5729                         if (rdev->raid_disk < 0
5730                             && !test_bit(Faulty, &rdev->flags)) {
5731                                 rdev->recovery_offset = 0;
5732                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5733                                         char nm[20];
5734                                         sprintf(nm, "rd%d", rdev->raid_disk);
5735                                         if (sysfs_create_link(&mddev->kobj,
5736                                                               &rdev->kobj, nm))
5737                                                 printk(KERN_WARNING
5738                                                        "md: cannot register "
5739                                                        "%s for %s\n",
5740                                                        nm, mdname(mddev));
5741                                         spares++;
5742                                         md_new_event(mddev);
5743                                 } else
5744                                         break;
5745                         }
5746         }
5747         return spares;
5748 }
5749 /*
5750  * This routine is regularly called by all per-raid-array threads to
5751  * deal with generic issues like resync and super-block update.
5752  * Raid personalities that don't have a thread (linear/raid0) do not
5753  * need this as they never do any recovery or update the superblock.
5754  *
5755  * It does not do any resync itself, but rather "forks" off other threads
5756  * to do that as needed.
5757  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5758  * "->recovery" and create a thread at ->sync_thread.
5759  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5760  * and wakeups up this thread which will reap the thread and finish up.
5761  * This thread also removes any faulty devices (with nr_pending == 0).
5762  *
5763  * The overall approach is:
5764  *  1/ if the superblock needs updating, update it.
5765  *  2/ If a recovery thread is running, don't do anything else.
5766  *  3/ If recovery has finished, clean up, possibly marking spares active.
5767  *  4/ If there are any faulty devices, remove them.
5768  *  5/ If array is degraded, try to add spares devices
5769  *  6/ If array has spares or is not in-sync, start a resync thread.
5770  */
5771 void md_check_recovery(mddev_t *mddev)
5772 {
5773         mdk_rdev_t *rdev;
5774         struct list_head *rtmp;
5775
5776
5777         if (mddev->bitmap)
5778                 bitmap_daemon_work(mddev->bitmap);
5779
5780         if (mddev->ro)
5781                 return;
5782
5783         if (signal_pending(current)) {
5784                 if (mddev->pers->sync_request) {
5785                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5786                                mdname(mddev));
5787                         mddev->safemode = 2;
5788                 }
5789                 flush_signals(current);
5790         }
5791
5792         if ( ! (
5793                 (mddev->flags && !mddev->external) ||
5794                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5795                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5796                 (mddev->safemode == 1) ||
5797                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5798                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5799                 ))
5800                 return;
5801
5802         if (mddev_trylock(mddev)) {
5803                 int spares = 0;
5804
5805                 spin_lock_irq(&mddev->write_lock);
5806                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5807                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5808                         mddev->in_sync = 1;
5809                         if (mddev->persistent)
5810                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5811                 }
5812                 if (mddev->safemode == 1)
5813                         mddev->safemode = 0;
5814                 spin_unlock_irq(&mddev->write_lock);
5815
5816                 if (mddev->flags)
5817                         md_update_sb(mddev, 0);
5818
5819
5820                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5821                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5822                         /* resync/recovery still happening */
5823                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5824                         goto unlock;
5825                 }
5826                 if (mddev->sync_thread) {
5827                         /* resync has finished, collect result */
5828                         md_unregister_thread(mddev->sync_thread);
5829                         mddev->sync_thread = NULL;
5830                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5831                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5832                                 /* success...*/
5833                                 /* activate any spares */
5834                                 mddev->pers->spare_active(mddev);
5835                         }
5836                         md_update_sb(mddev, 1);
5837
5838                         /* if array is no-longer degraded, then any saved_raid_disk
5839                          * information must be scrapped
5840                          */
5841                         if (!mddev->degraded)
5842                                 rdev_for_each(rdev, rtmp, mddev)
5843                                         rdev->saved_raid_disk = -1;
5844
5845                         mddev->recovery = 0;
5846                         /* flag recovery needed just to double check */
5847                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5848                         md_new_event(mddev);
5849                         goto unlock;
5850                 }
5851                 /* Clear some bits that don't mean anything, but
5852                  * might be left set
5853                  */
5854                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5855                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5856                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5857                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5858
5859                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5860                         goto unlock;
5861                 /* no recovery is running.
5862                  * remove any failed drives, then
5863                  * add spares if possible.
5864                  * Spare are also removed and re-added, to allow
5865                  * the personality to fail the re-add.
5866                  */
5867
5868                 if (mddev->reshape_position != MaxSector) {
5869                         if (mddev->pers->check_reshape(mddev) != 0)
5870                                 /* Cannot proceed */
5871                                 goto unlock;
5872                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5873                 } else if ((spares = remove_and_add_spares(mddev))) {
5874                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5875                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5876                 } else if (mddev->recovery_cp < MaxSector) {
5877                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5878                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5879                         /* nothing to be done ... */
5880                         goto unlock;
5881
5882                 if (mddev->pers->sync_request) {
5883                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5884                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5885                                 /* We are adding a device or devices to an array
5886                                  * which has the bitmap stored on all devices.
5887                                  * So make sure all bitmap pages get written
5888                                  */
5889                                 bitmap_write_all(mddev->bitmap);
5890                         }
5891                         mddev->sync_thread = md_register_thread(md_do_sync,
5892                                                                 mddev,
5893                                                                 "%s_resync");
5894                         if (!mddev->sync_thread) {
5895                                 printk(KERN_ERR "%s: could not start resync"
5896                                         " thread...\n",
5897                                         mdname(mddev));
5898                                 /* leave the spares where they are, it shouldn't hurt */
5899                                 mddev->recovery = 0;
5900                         } else
5901                                 md_wakeup_thread(mddev->sync_thread);
5902                         md_new_event(mddev);
5903                 }
5904         unlock:
5905                 mddev_unlock(mddev);
5906         }
5907 }
5908
5909 static int md_notify_reboot(struct notifier_block *this,
5910                             unsigned long code, void *x)
5911 {
5912         struct list_head *tmp;
5913         mddev_t *mddev;
5914
5915         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5916
5917                 printk(KERN_INFO "md: stopping all md devices.\n");
5918
5919                 for_each_mddev(mddev, tmp)
5920                         if (mddev_trylock(mddev)) {
5921                                 do_md_stop (mddev, 1);
5922                                 mddev_unlock(mddev);
5923                         }
5924                 /*
5925                  * certain more exotic SCSI devices are known to be
5926                  * volatile wrt too early system reboots. While the
5927                  * right place to handle this issue is the given
5928                  * driver, we do want to have a safe RAID driver ...
5929                  */
5930                 mdelay(1000*1);
5931         }
5932         return NOTIFY_DONE;
5933 }
5934
5935 static struct notifier_block md_notifier = {
5936         .notifier_call  = md_notify_reboot,
5937         .next           = NULL,
5938         .priority       = INT_MAX, /* before any real devices */
5939 };
5940
5941 static void md_geninit(void)
5942 {
5943         struct proc_dir_entry *p;
5944
5945         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5946
5947         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5948         if (p)
5949                 p->proc_fops = &md_seq_fops;
5950 }
5951
5952 static int __init md_init(void)
5953 {
5954         if (register_blkdev(MAJOR_NR, "md"))
5955                 return -1;
5956         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5957                 unregister_blkdev(MAJOR_NR, "md");
5958                 return -1;
5959         }
5960         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5961                             md_probe, NULL, NULL);
5962         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5963                             md_probe, NULL, NULL);
5964
5965         register_reboot_notifier(&md_notifier);
5966         raid_table_header = register_sysctl_table(raid_root_table);
5967
5968         md_geninit();
5969         return (0);
5970 }
5971
5972
5973 #ifndef MODULE
5974
5975 /*
5976  * Searches all registered partitions for autorun RAID arrays
5977  * at boot time.
5978  */
5979
5980 static LIST_HEAD(all_detected_devices);
5981 struct detected_devices_node {
5982         struct list_head list;
5983         dev_t dev;
5984 };
5985
5986 void md_autodetect_dev(dev_t dev)
5987 {
5988         struct detected_devices_node *node_detected_dev;
5989
5990         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5991         if (node_detected_dev) {
5992                 node_detected_dev->dev = dev;
5993                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5994         } else {
5995                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5996                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
5997         }
5998 }
5999
6000
6001 static void autostart_arrays(int part)
6002 {
6003         mdk_rdev_t *rdev;
6004         struct detected_devices_node *node_detected_dev;
6005         dev_t dev;
6006         int i_scanned, i_passed;
6007
6008         i_scanned = 0;
6009         i_passed = 0;
6010
6011         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6012
6013         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6014                 i_scanned++;
6015                 node_detected_dev = list_entry(all_detected_devices.next,
6016                                         struct detected_devices_node, list);
6017                 list_del(&node_detected_dev->list);
6018                 dev = node_detected_dev->dev;
6019                 kfree(node_detected_dev);
6020                 rdev = md_import_device(dev,0, 90);
6021                 if (IS_ERR(rdev))
6022                         continue;
6023
6024                 if (test_bit(Faulty, &rdev->flags)) {
6025                         MD_BUG();
6026                         continue;
6027                 }
6028                 list_add(&rdev->same_set, &pending_raid_disks);
6029                 i_passed++;
6030         }
6031
6032         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6033                                                 i_scanned, i_passed);
6034
6035         autorun_devices(part);
6036 }
6037
6038 #endif /* !MODULE */
6039
6040 static __exit void md_exit(void)
6041 {
6042         mddev_t *mddev;
6043         struct list_head *tmp;
6044
6045         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6046         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6047
6048         unregister_blkdev(MAJOR_NR,"md");
6049         unregister_blkdev(mdp_major, "mdp");
6050         unregister_reboot_notifier(&md_notifier);
6051         unregister_sysctl_table(raid_table_header);
6052         remove_proc_entry("mdstat", NULL);
6053         for_each_mddev(mddev, tmp) {
6054                 struct gendisk *disk = mddev->gendisk;
6055                 if (!disk)
6056                         continue;
6057                 export_array(mddev);
6058                 del_gendisk(disk);
6059                 put_disk(disk);
6060                 mddev->gendisk = NULL;
6061                 mddev_put(mddev);
6062         }
6063 }
6064
6065 subsys_initcall(md_init);
6066 module_exit(md_exit)
6067
6068 static int get_ro(char *buffer, struct kernel_param *kp)
6069 {
6070         return sprintf(buffer, "%d", start_readonly);
6071 }
6072 static int set_ro(const char *val, struct kernel_param *kp)
6073 {
6074         char *e;
6075         int num = simple_strtoul(val, &e, 10);
6076         if (*val && (*e == '\0' || *e == '\n')) {
6077                 start_readonly = num;
6078                 return 0;
6079         }
6080         return -EINVAL;
6081 }
6082
6083 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6084 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6085
6086
6087 EXPORT_SYMBOL(register_md_personality);
6088 EXPORT_SYMBOL(unregister_md_personality);
6089 EXPORT_SYMBOL(md_error);
6090 EXPORT_SYMBOL(md_done_sync);
6091 EXPORT_SYMBOL(md_write_start);
6092 EXPORT_SYMBOL(md_write_end);
6093 EXPORT_SYMBOL(md_register_thread);
6094 EXPORT_SYMBOL(md_unregister_thread);
6095 EXPORT_SYMBOL(md_wakeup_thread);
6096 EXPORT_SYMBOL(md_check_recovery);
6097 MODULE_LICENSE("GPL");
6098 MODULE_ALIAS("md");
6099 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);