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         if (test_bit(AutoDetected, &rdev->flags))
1507                 md_autodetect_dev(rdev->bdev->bd_dev);
1508 #endif
1509         unlock_rdev(rdev);
1510         kobject_put(&rdev->kobj);
1511 }
1512
1513 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1514 {
1515         unbind_rdev_from_array(rdev);
1516         export_rdev(rdev);
1517 }
1518
1519 static void export_array(mddev_t *mddev)
1520 {
1521         struct list_head *tmp;
1522         mdk_rdev_t *rdev;
1523
1524         rdev_for_each(rdev, tmp, mddev) {
1525                 if (!rdev->mddev) {
1526                         MD_BUG();
1527                         continue;
1528                 }
1529                 kick_rdev_from_array(rdev);
1530         }
1531         if (!list_empty(&mddev->disks))
1532                 MD_BUG();
1533         mddev->raid_disks = 0;
1534         mddev->major_version = 0;
1535 }
1536
1537 static void print_desc(mdp_disk_t *desc)
1538 {
1539         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1540                 desc->major,desc->minor,desc->raid_disk,desc->state);
1541 }
1542
1543 static void print_sb(mdp_super_t *sb)
1544 {
1545         int i;
1546
1547         printk(KERN_INFO
1548                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1549                 sb->major_version, sb->minor_version, sb->patch_version,
1550                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1551                 sb->ctime);
1552         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1553                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1554                 sb->md_minor, sb->layout, sb->chunk_size);
1555         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1556                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1557                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1558                 sb->failed_disks, sb->spare_disks,
1559                 sb->sb_csum, (unsigned long)sb->events_lo);
1560
1561         printk(KERN_INFO);
1562         for (i = 0; i < MD_SB_DISKS; i++) {
1563                 mdp_disk_t *desc;
1564
1565                 desc = sb->disks + i;
1566                 if (desc->number || desc->major || desc->minor ||
1567                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1568                         printk("     D %2d: ", i);
1569                         print_desc(desc);
1570                 }
1571         }
1572         printk(KERN_INFO "md:     THIS: ");
1573         print_desc(&sb->this_disk);
1574
1575 }
1576
1577 static void print_rdev(mdk_rdev_t *rdev)
1578 {
1579         char b[BDEVNAME_SIZE];
1580         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1581                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1582                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1583                 rdev->desc_nr);
1584         if (rdev->sb_loaded) {
1585                 printk(KERN_INFO "md: rdev superblock:\n");
1586                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1587         } else
1588                 printk(KERN_INFO "md: no rdev superblock!\n");
1589 }
1590
1591 static void md_print_devices(void)
1592 {
1593         struct list_head *tmp, *tmp2;
1594         mdk_rdev_t *rdev;
1595         mddev_t *mddev;
1596         char b[BDEVNAME_SIZE];
1597
1598         printk("\n");
1599         printk("md:     **********************************\n");
1600         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1601         printk("md:     **********************************\n");
1602         for_each_mddev(mddev, tmp) {
1603
1604                 if (mddev->bitmap)
1605                         bitmap_print_sb(mddev->bitmap);
1606                 else
1607                         printk("%s: ", mdname(mddev));
1608                 rdev_for_each(rdev, tmp2, mddev)
1609                         printk("<%s>", bdevname(rdev->bdev,b));
1610                 printk("\n");
1611
1612                 rdev_for_each(rdev, tmp2, mddev)
1613                         print_rdev(rdev);
1614         }
1615         printk("md:     **********************************\n");
1616         printk("\n");
1617 }
1618
1619
1620 static void sync_sbs(mddev_t * mddev, int nospares)
1621 {
1622         /* Update each superblock (in-memory image), but
1623          * if we are allowed to, skip spares which already
1624          * have the right event counter, or have one earlier
1625          * (which would mean they aren't being marked as dirty
1626          * with the rest of the array)
1627          */
1628         mdk_rdev_t *rdev;
1629         struct list_head *tmp;
1630
1631         rdev_for_each(rdev, tmp, mddev) {
1632                 if (rdev->sb_events == mddev->events ||
1633                     (nospares &&
1634                      rdev->raid_disk < 0 &&
1635                      (rdev->sb_events&1)==0 &&
1636                      rdev->sb_events+1 == mddev->events)) {
1637                         /* Don't update this superblock */
1638                         rdev->sb_loaded = 2;
1639                 } else {
1640                         super_types[mddev->major_version].
1641                                 sync_super(mddev, rdev);
1642                         rdev->sb_loaded = 1;
1643                 }
1644         }
1645 }
1646
1647 static void md_update_sb(mddev_t * mddev, int force_change)
1648 {
1649         struct list_head *tmp;
1650         mdk_rdev_t *rdev;
1651         int sync_req;
1652         int nospares = 0;
1653
1654 repeat:
1655         spin_lock_irq(&mddev->write_lock);
1656
1657         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1658         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1659                 force_change = 1;
1660         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1661                 /* just a clean<-> dirty transition, possibly leave spares alone,
1662                  * though if events isn't the right even/odd, we will have to do
1663                  * spares after all
1664                  */
1665                 nospares = 1;
1666         if (force_change)
1667                 nospares = 0;
1668         if (mddev->degraded)
1669                 /* If the array is degraded, then skipping spares is both
1670                  * dangerous and fairly pointless.
1671                  * Dangerous because a device that was removed from the array
1672                  * might have a event_count that still looks up-to-date,
1673                  * so it can be re-added without a resync.
1674                  * Pointless because if there are any spares to skip,
1675                  * then a recovery will happen and soon that array won't
1676                  * be degraded any more and the spare can go back to sleep then.
1677                  */
1678                 nospares = 0;
1679
1680         sync_req = mddev->in_sync;
1681         mddev->utime = get_seconds();
1682
1683         /* If this is just a dirty<->clean transition, and the array is clean
1684          * and 'events' is odd, we can roll back to the previous clean state */
1685         if (nospares
1686             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1687             && (mddev->events & 1)
1688             && mddev->events != 1)
1689                 mddev->events--;
1690         else {
1691                 /* otherwise we have to go forward and ... */
1692                 mddev->events ++;
1693                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1694                         /* .. if the array isn't clean, insist on an odd 'events' */
1695                         if ((mddev->events&1)==0) {
1696                                 mddev->events++;
1697                                 nospares = 0;
1698                         }
1699                 } else {
1700                         /* otherwise insist on an even 'events' (for clean states) */
1701                         if ((mddev->events&1)) {
1702                                 mddev->events++;
1703                                 nospares = 0;
1704                         }
1705                 }
1706         }
1707
1708         if (!mddev->events) {
1709                 /*
1710                  * oops, this 64-bit counter should never wrap.
1711                  * Either we are in around ~1 trillion A.C., assuming
1712                  * 1 reboot per second, or we have a bug:
1713                  */
1714                 MD_BUG();
1715                 mddev->events --;
1716         }
1717
1718         /*
1719          * do not write anything to disk if using
1720          * nonpersistent superblocks
1721          */
1722         if (!mddev->persistent) {
1723                 if (!mddev->external)
1724                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1725
1726                 spin_unlock_irq(&mddev->write_lock);
1727                 wake_up(&mddev->sb_wait);
1728                 return;
1729         }
1730         sync_sbs(mddev, nospares);
1731         spin_unlock_irq(&mddev->write_lock);
1732
1733         dprintk(KERN_INFO
1734                 "md: updating %s RAID superblock on device (in sync %d)\n",
1735                 mdname(mddev),mddev->in_sync);
1736
1737         bitmap_update_sb(mddev->bitmap);
1738         rdev_for_each(rdev, tmp, mddev) {
1739                 char b[BDEVNAME_SIZE];
1740                 dprintk(KERN_INFO "md: ");
1741                 if (rdev->sb_loaded != 1)
1742                         continue; /* no noise on spare devices */
1743                 if (test_bit(Faulty, &rdev->flags))
1744                         dprintk("(skipping faulty ");
1745
1746                 dprintk("%s ", bdevname(rdev->bdev,b));
1747                 if (!test_bit(Faulty, &rdev->flags)) {
1748                         md_super_write(mddev,rdev,
1749                                        rdev->sb_offset<<1, rdev->sb_size,
1750                                        rdev->sb_page);
1751                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1752                                 bdevname(rdev->bdev,b),
1753                                 (unsigned long long)rdev->sb_offset);
1754                         rdev->sb_events = mddev->events;
1755
1756                 } else
1757                         dprintk(")\n");
1758                 if (mddev->level == LEVEL_MULTIPATH)
1759                         /* only need to write one superblock... */
1760                         break;
1761         }
1762         md_super_wait(mddev);
1763         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1764
1765         spin_lock_irq(&mddev->write_lock);
1766         if (mddev->in_sync != sync_req ||
1767             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1768                 /* have to write it out again */
1769                 spin_unlock_irq(&mddev->write_lock);
1770                 goto repeat;
1771         }
1772         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1773         spin_unlock_irq(&mddev->write_lock);
1774         wake_up(&mddev->sb_wait);
1775
1776 }
1777
1778 /* words written to sysfs files may, or my not, be \n terminated.
1779  * We want to accept with case. For this we use cmd_match.
1780  */
1781 static int cmd_match(const char *cmd, const char *str)
1782 {
1783         /* See if cmd, written into a sysfs file, matches
1784          * str.  They must either be the same, or cmd can
1785          * have a trailing newline
1786          */
1787         while (*cmd && *str && *cmd == *str) {
1788                 cmd++;
1789                 str++;
1790         }
1791         if (*cmd == '\n')
1792                 cmd++;
1793         if (*str || *cmd)
1794                 return 0;
1795         return 1;
1796 }
1797
1798 struct rdev_sysfs_entry {
1799         struct attribute attr;
1800         ssize_t (*show)(mdk_rdev_t *, char *);
1801         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1802 };
1803
1804 static ssize_t
1805 state_show(mdk_rdev_t *rdev, char *page)
1806 {
1807         char *sep = "";
1808         size_t len = 0;
1809
1810         if (test_bit(Faulty, &rdev->flags)) {
1811                 len+= sprintf(page+len, "%sfaulty",sep);
1812                 sep = ",";
1813         }
1814         if (test_bit(In_sync, &rdev->flags)) {
1815                 len += sprintf(page+len, "%sin_sync",sep);
1816                 sep = ",";
1817         }
1818         if (test_bit(WriteMostly, &rdev->flags)) {
1819                 len += sprintf(page+len, "%swrite_mostly",sep);
1820                 sep = ",";
1821         }
1822         if (!test_bit(Faulty, &rdev->flags) &&
1823             !test_bit(In_sync, &rdev->flags)) {
1824                 len += sprintf(page+len, "%sspare", sep);
1825                 sep = ",";
1826         }
1827         return len+sprintf(page+len, "\n");
1828 }
1829
1830 static ssize_t
1831 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1832 {
1833         /* can write
1834          *  faulty  - simulates and error
1835          *  remove  - disconnects the device
1836          *  writemostly - sets write_mostly
1837          *  -writemostly - clears write_mostly
1838          */
1839         int err = -EINVAL;
1840         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1841                 md_error(rdev->mddev, rdev);
1842                 err = 0;
1843         } else if (cmd_match(buf, "remove")) {
1844                 if (rdev->raid_disk >= 0)
1845                         err = -EBUSY;
1846                 else {
1847                         mddev_t *mddev = rdev->mddev;
1848                         kick_rdev_from_array(rdev);
1849                         if (mddev->pers)
1850                                 md_update_sb(mddev, 1);
1851                         md_new_event(mddev);
1852                         err = 0;
1853                 }
1854         } else if (cmd_match(buf, "writemostly")) {
1855                 set_bit(WriteMostly, &rdev->flags);
1856                 err = 0;
1857         } else if (cmd_match(buf, "-writemostly")) {
1858                 clear_bit(WriteMostly, &rdev->flags);
1859                 err = 0;
1860         }
1861         return err ? err : len;
1862 }
1863 static struct rdev_sysfs_entry rdev_state =
1864 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1865
1866 static ssize_t
1867 super_show(mdk_rdev_t *rdev, char *page)
1868 {
1869         if (rdev->sb_loaded && rdev->sb_size) {
1870                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1871                 return rdev->sb_size;
1872         } else
1873                 return 0;
1874 }
1875 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1876
1877 static ssize_t
1878 errors_show(mdk_rdev_t *rdev, char *page)
1879 {
1880         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1881 }
1882
1883 static ssize_t
1884 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1885 {
1886         char *e;
1887         unsigned long n = simple_strtoul(buf, &e, 10);
1888         if (*buf && (*e == 0 || *e == '\n')) {
1889                 atomic_set(&rdev->corrected_errors, n);
1890                 return len;
1891         }
1892         return -EINVAL;
1893 }
1894 static struct rdev_sysfs_entry rdev_errors =
1895 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1896
1897 static ssize_t
1898 slot_show(mdk_rdev_t *rdev, char *page)
1899 {
1900         if (rdev->raid_disk < 0)
1901                 return sprintf(page, "none\n");
1902         else
1903                 return sprintf(page, "%d\n", rdev->raid_disk);
1904 }
1905
1906 static ssize_t
1907 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1908 {
1909         char *e;
1910         int err;
1911         char nm[20];
1912         int slot = simple_strtoul(buf, &e, 10);
1913         if (strncmp(buf, "none", 4)==0)
1914                 slot = -1;
1915         else if (e==buf || (*e && *e!= '\n'))
1916                 return -EINVAL;
1917         if (rdev->mddev->pers) {
1918                 /* Setting 'slot' on an active array requires also
1919                  * updating the 'rd%d' link, and communicating
1920                  * with the personality with ->hot_*_disk.
1921                  * For now we only support removing
1922                  * failed/spare devices.  This normally happens automatically,
1923                  * but not when the metadata is externally managed.
1924                  */
1925                 if (slot != -1)
1926                         return -EBUSY;
1927                 if (rdev->raid_disk == -1)
1928                         return -EEXIST;
1929                 /* personality does all needed checks */
1930                 if (rdev->mddev->pers->hot_add_disk == NULL)
1931                         return -EINVAL;
1932                 err = rdev->mddev->pers->
1933                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
1934                 if (err)
1935                         return err;
1936                 sprintf(nm, "rd%d", rdev->raid_disk);
1937                 sysfs_remove_link(&rdev->mddev->kobj, nm);
1938                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1939                 md_wakeup_thread(rdev->mddev->thread);
1940         } else {
1941                 if (slot >= rdev->mddev->raid_disks)
1942                         return -ENOSPC;
1943                 rdev->raid_disk = slot;
1944                 /* assume it is working */
1945                 clear_bit(Faulty, &rdev->flags);
1946                 clear_bit(WriteMostly, &rdev->flags);
1947                 set_bit(In_sync, &rdev->flags);
1948         }
1949         return len;
1950 }
1951
1952
1953 static struct rdev_sysfs_entry rdev_slot =
1954 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1955
1956 static ssize_t
1957 offset_show(mdk_rdev_t *rdev, char *page)
1958 {
1959         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1960 }
1961
1962 static ssize_t
1963 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1964 {
1965         char *e;
1966         unsigned long long offset = simple_strtoull(buf, &e, 10);
1967         if (e==buf || (*e && *e != '\n'))
1968                 return -EINVAL;
1969         if (rdev->mddev->pers)
1970                 return -EBUSY;
1971         if (rdev->size && rdev->mddev->external)
1972                 /* Must set offset before size, so overlap checks
1973                  * can be sane */
1974                 return -EBUSY;
1975         rdev->data_offset = offset;
1976         return len;
1977 }
1978
1979 static struct rdev_sysfs_entry rdev_offset =
1980 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1981
1982 static ssize_t
1983 rdev_size_show(mdk_rdev_t *rdev, char *page)
1984 {
1985         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1986 }
1987
1988 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1989 {
1990         /* check if two start/length pairs overlap */
1991         if (s1+l1 <= s2)
1992                 return 0;
1993         if (s2+l2 <= s1)
1994                 return 0;
1995         return 1;
1996 }
1997
1998 static ssize_t
1999 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2000 {
2001         char *e;
2002         unsigned long long size = simple_strtoull(buf, &e, 10);
2003         unsigned long long oldsize = rdev->size;
2004         if (e==buf || (*e && *e != '\n'))
2005                 return -EINVAL;
2006         if (rdev->mddev->pers)
2007                 return -EBUSY;
2008         rdev->size = size;
2009         if (size > oldsize && rdev->mddev->external) {
2010                 /* need to check that all other rdevs with the same ->bdev
2011                  * do not overlap.  We need to unlock the mddev to avoid
2012                  * a deadlock.  We have already changed rdev->size, and if
2013                  * we have to change it back, we will have the lock again.
2014                  */
2015                 mddev_t *mddev;
2016                 int overlap = 0;
2017                 struct list_head *tmp, *tmp2;
2018
2019                 mddev_unlock(rdev->mddev);
2020                 for_each_mddev(mddev, tmp) {
2021                         mdk_rdev_t *rdev2;
2022
2023                         mddev_lock(mddev);
2024                         rdev_for_each(rdev2, tmp2, mddev)
2025                                 if (test_bit(AllReserved, &rdev2->flags) ||
2026                                     (rdev->bdev == rdev2->bdev &&
2027                                      rdev != rdev2 &&
2028                                      overlaps(rdev->data_offset, rdev->size,
2029                                             rdev2->data_offset, rdev2->size))) {
2030                                         overlap = 1;
2031                                         break;
2032                                 }
2033                         mddev_unlock(mddev);
2034                         if (overlap) {
2035                                 mddev_put(mddev);
2036                                 break;
2037                         }
2038                 }
2039                 mddev_lock(rdev->mddev);
2040                 if (overlap) {
2041                         /* Someone else could have slipped in a size
2042                          * change here, but doing so is just silly.
2043                          * We put oldsize back because we *know* it is
2044                          * safe, and trust userspace not to race with
2045                          * itself
2046                          */
2047                         rdev->size = oldsize;
2048                         return -EBUSY;
2049                 }
2050         }
2051         if (size < rdev->mddev->size || rdev->mddev->size == 0)
2052                 rdev->mddev->size = size;
2053         return len;
2054 }
2055
2056 static struct rdev_sysfs_entry rdev_size =
2057 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2058
2059 static struct attribute *rdev_default_attrs[] = {
2060         &rdev_state.attr,
2061         &rdev_super.attr,
2062         &rdev_errors.attr,
2063         &rdev_slot.attr,
2064         &rdev_offset.attr,
2065         &rdev_size.attr,
2066         NULL,
2067 };
2068 static ssize_t
2069 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2070 {
2071         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2072         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2073
2074         if (!entry->show)
2075                 return -EIO;
2076         return entry->show(rdev, page);
2077 }
2078
2079 static ssize_t
2080 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2081               const char *page, size_t length)
2082 {
2083         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2084         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2085         int rv;
2086
2087         if (!entry->store)
2088                 return -EIO;
2089         if (!capable(CAP_SYS_ADMIN))
2090                 return -EACCES;
2091         rv = mddev_lock(rdev->mddev);
2092         if (!rv) {
2093                 rv = entry->store(rdev, page, length);
2094                 mddev_unlock(rdev->mddev);
2095         }
2096         return rv;
2097 }
2098
2099 static void rdev_free(struct kobject *ko)
2100 {
2101         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2102         kfree(rdev);
2103 }
2104 static struct sysfs_ops rdev_sysfs_ops = {
2105         .show           = rdev_attr_show,
2106         .store          = rdev_attr_store,
2107 };
2108 static struct kobj_type rdev_ktype = {
2109         .release        = rdev_free,
2110         .sysfs_ops      = &rdev_sysfs_ops,
2111         .default_attrs  = rdev_default_attrs,
2112 };
2113
2114 /*
2115  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2116  *
2117  * mark the device faulty if:
2118  *
2119  *   - the device is nonexistent (zero size)
2120  *   - the device has no valid superblock
2121  *
2122  * a faulty rdev _never_ has rdev->sb set.
2123  */
2124 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2125 {
2126         char b[BDEVNAME_SIZE];
2127         int err;
2128         mdk_rdev_t *rdev;
2129         sector_t size;
2130
2131         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2132         if (!rdev) {
2133                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2134                 return ERR_PTR(-ENOMEM);
2135         }
2136
2137         if ((err = alloc_disk_sb(rdev)))
2138                 goto abort_free;
2139
2140         err = lock_rdev(rdev, newdev, super_format == -2);
2141         if (err)
2142                 goto abort_free;
2143
2144         kobject_init(&rdev->kobj, &rdev_ktype);
2145
2146         rdev->desc_nr = -1;
2147         rdev->saved_raid_disk = -1;
2148         rdev->raid_disk = -1;
2149         rdev->flags = 0;
2150         rdev->data_offset = 0;
2151         rdev->sb_events = 0;
2152         atomic_set(&rdev->nr_pending, 0);
2153         atomic_set(&rdev->read_errors, 0);
2154         atomic_set(&rdev->corrected_errors, 0);
2155
2156         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2157         if (!size) {
2158                 printk(KERN_WARNING
2159                         "md: %s has zero or unknown size, marking faulty!\n",
2160                         bdevname(rdev->bdev,b));
2161                 err = -EINVAL;
2162                 goto abort_free;
2163         }
2164
2165         if (super_format >= 0) {
2166                 err = super_types[super_format].
2167                         load_super(rdev, NULL, super_minor);
2168                 if (err == -EINVAL) {
2169                         printk(KERN_WARNING
2170                                 "md: %s does not have a valid v%d.%d "
2171                                "superblock, not importing!\n",
2172                                 bdevname(rdev->bdev,b),
2173                                super_format, super_minor);
2174                         goto abort_free;
2175                 }
2176                 if (err < 0) {
2177                         printk(KERN_WARNING
2178                                 "md: could not read %s's sb, not importing!\n",
2179                                 bdevname(rdev->bdev,b));
2180                         goto abort_free;
2181                 }
2182         }
2183         INIT_LIST_HEAD(&rdev->same_set);
2184
2185         return rdev;
2186
2187 abort_free:
2188         if (rdev->sb_page) {
2189                 if (rdev->bdev)
2190                         unlock_rdev(rdev);
2191                 free_disk_sb(rdev);
2192         }
2193         kfree(rdev);
2194         return ERR_PTR(err);
2195 }
2196
2197 /*
2198  * Check a full RAID array for plausibility
2199  */
2200
2201
2202 static void analyze_sbs(mddev_t * mddev)
2203 {
2204         int i;
2205         struct list_head *tmp;
2206         mdk_rdev_t *rdev, *freshest;
2207         char b[BDEVNAME_SIZE];
2208
2209         freshest = NULL;
2210         rdev_for_each(rdev, tmp, mddev)
2211                 switch (super_types[mddev->major_version].
2212                         load_super(rdev, freshest, mddev->minor_version)) {
2213                 case 1:
2214                         freshest = rdev;
2215                         break;
2216                 case 0:
2217                         break;
2218                 default:
2219                         printk( KERN_ERR \
2220                                 "md: fatal superblock inconsistency in %s"
2221                                 " -- removing from array\n",
2222                                 bdevname(rdev->bdev,b));
2223                         kick_rdev_from_array(rdev);
2224                 }
2225
2226
2227         super_types[mddev->major_version].
2228                 validate_super(mddev, freshest);
2229
2230         i = 0;
2231         rdev_for_each(rdev, tmp, mddev) {
2232                 if (rdev != freshest)
2233                         if (super_types[mddev->major_version].
2234                             validate_super(mddev, rdev)) {
2235                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2236                                         " from array!\n",
2237                                         bdevname(rdev->bdev,b));
2238                                 kick_rdev_from_array(rdev);
2239                                 continue;
2240                         }
2241                 if (mddev->level == LEVEL_MULTIPATH) {
2242                         rdev->desc_nr = i++;
2243                         rdev->raid_disk = rdev->desc_nr;
2244                         set_bit(In_sync, &rdev->flags);
2245                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2246                         rdev->raid_disk = -1;
2247                         clear_bit(In_sync, &rdev->flags);
2248                 }
2249         }
2250
2251
2252
2253         if (mddev->recovery_cp != MaxSector &&
2254             mddev->level >= 1)
2255                 printk(KERN_ERR "md: %s: raid array is not clean"
2256                        " -- starting background reconstruction\n",
2257                        mdname(mddev));
2258
2259 }
2260
2261 static ssize_t
2262 safe_delay_show(mddev_t *mddev, char *page)
2263 {
2264         int msec = (mddev->safemode_delay*1000)/HZ;
2265         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2266 }
2267 static ssize_t
2268 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2269 {
2270         int scale=1;
2271         int dot=0;
2272         int i;
2273         unsigned long msec;
2274         char buf[30];
2275         char *e;
2276         /* remove a period, and count digits after it */
2277         if (len >= sizeof(buf))
2278                 return -EINVAL;
2279         strlcpy(buf, cbuf, len);
2280         buf[len] = 0;
2281         for (i=0; i<len; i++) {
2282                 if (dot) {
2283                         if (isdigit(buf[i])) {
2284                                 buf[i-1] = buf[i];
2285                                 scale *= 10;
2286                         }
2287                         buf[i] = 0;
2288                 } else if (buf[i] == '.') {
2289                         dot=1;
2290                         buf[i] = 0;
2291                 }
2292         }
2293         msec = simple_strtoul(buf, &e, 10);
2294         if (e == buf || (*e && *e != '\n'))
2295                 return -EINVAL;
2296         msec = (msec * 1000) / scale;
2297         if (msec == 0)
2298                 mddev->safemode_delay = 0;
2299         else {
2300                 mddev->safemode_delay = (msec*HZ)/1000;
2301                 if (mddev->safemode_delay == 0)
2302                         mddev->safemode_delay = 1;
2303         }
2304         return len;
2305 }
2306 static struct md_sysfs_entry md_safe_delay =
2307 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2308
2309 static ssize_t
2310 level_show(mddev_t *mddev, char *page)
2311 {
2312         struct mdk_personality *p = mddev->pers;
2313         if (p)
2314                 return sprintf(page, "%s\n", p->name);
2315         else if (mddev->clevel[0])
2316                 return sprintf(page, "%s\n", mddev->clevel);
2317         else if (mddev->level != LEVEL_NONE)
2318                 return sprintf(page, "%d\n", mddev->level);
2319         else
2320                 return 0;
2321 }
2322
2323 static ssize_t
2324 level_store(mddev_t *mddev, const char *buf, size_t len)
2325 {
2326         ssize_t rv = len;
2327         if (mddev->pers)
2328                 return -EBUSY;
2329         if (len == 0)
2330                 return 0;
2331         if (len >= sizeof(mddev->clevel))
2332                 return -ENOSPC;
2333         strncpy(mddev->clevel, buf, len);
2334         if (mddev->clevel[len-1] == '\n')
2335                 len--;
2336         mddev->clevel[len] = 0;
2337         mddev->level = LEVEL_NONE;
2338         return rv;
2339 }
2340
2341 static struct md_sysfs_entry md_level =
2342 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2343
2344
2345 static ssize_t
2346 layout_show(mddev_t *mddev, char *page)
2347 {
2348         /* just a number, not meaningful for all levels */
2349         if (mddev->reshape_position != MaxSector &&
2350             mddev->layout != mddev->new_layout)
2351                 return sprintf(page, "%d (%d)\n",
2352                                mddev->new_layout, mddev->layout);
2353         return sprintf(page, "%d\n", mddev->layout);
2354 }
2355
2356 static ssize_t
2357 layout_store(mddev_t *mddev, const char *buf, size_t len)
2358 {
2359         char *e;
2360         unsigned long n = simple_strtoul(buf, &e, 10);
2361
2362         if (!*buf || (*e && *e != '\n'))
2363                 return -EINVAL;
2364
2365         if (mddev->pers)
2366                 return -EBUSY;
2367         if (mddev->reshape_position != MaxSector)
2368                 mddev->new_layout = n;
2369         else
2370                 mddev->layout = n;
2371         return len;
2372 }
2373 static struct md_sysfs_entry md_layout =
2374 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2375
2376
2377 static ssize_t
2378 raid_disks_show(mddev_t *mddev, char *page)
2379 {
2380         if (mddev->raid_disks == 0)
2381                 return 0;
2382         if (mddev->reshape_position != MaxSector &&
2383             mddev->delta_disks != 0)
2384                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2385                                mddev->raid_disks - mddev->delta_disks);
2386         return sprintf(page, "%d\n", mddev->raid_disks);
2387 }
2388
2389 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2390
2391 static ssize_t
2392 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2393 {
2394         char *e;
2395         int rv = 0;
2396         unsigned long n = simple_strtoul(buf, &e, 10);
2397
2398         if (!*buf || (*e && *e != '\n'))
2399                 return -EINVAL;
2400
2401         if (mddev->pers)
2402                 rv = update_raid_disks(mddev, n);
2403         else if (mddev->reshape_position != MaxSector) {
2404                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2405                 mddev->delta_disks = n - olddisks;
2406                 mddev->raid_disks = n;
2407         } else
2408                 mddev->raid_disks = n;
2409         return rv ? rv : len;
2410 }
2411 static struct md_sysfs_entry md_raid_disks =
2412 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2413
2414 static ssize_t
2415 chunk_size_show(mddev_t *mddev, char *page)
2416 {
2417         if (mddev->reshape_position != MaxSector &&
2418             mddev->chunk_size != mddev->new_chunk)
2419                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2420                                mddev->chunk_size);
2421         return sprintf(page, "%d\n", mddev->chunk_size);
2422 }
2423
2424 static ssize_t
2425 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2426 {
2427         /* can only set chunk_size if array is not yet active */
2428         char *e;
2429         unsigned long n = simple_strtoul(buf, &e, 10);
2430
2431         if (!*buf || (*e && *e != '\n'))
2432                 return -EINVAL;
2433
2434         if (mddev->pers)
2435                 return -EBUSY;
2436         else if (mddev->reshape_position != MaxSector)
2437                 mddev->new_chunk = n;
2438         else
2439                 mddev->chunk_size = n;
2440         return len;
2441 }
2442 static struct md_sysfs_entry md_chunk_size =
2443 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2444
2445 static ssize_t
2446 resync_start_show(mddev_t *mddev, char *page)
2447 {
2448         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2449 }
2450
2451 static ssize_t
2452 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2453 {
2454         /* can only set chunk_size if array is not yet active */
2455         char *e;
2456         unsigned long long n = simple_strtoull(buf, &e, 10);
2457
2458         if (mddev->pers)
2459                 return -EBUSY;
2460         if (!*buf || (*e && *e != '\n'))
2461                 return -EINVAL;
2462
2463         mddev->recovery_cp = n;
2464         return len;
2465 }
2466 static struct md_sysfs_entry md_resync_start =
2467 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2468
2469 /*
2470  * The array state can be:
2471  *
2472  * clear
2473  *     No devices, no size, no level
2474  *     Equivalent to STOP_ARRAY ioctl
2475  * inactive
2476  *     May have some settings, but array is not active
2477  *        all IO results in error
2478  *     When written, doesn't tear down array, but just stops it
2479  * suspended (not supported yet)
2480  *     All IO requests will block. The array can be reconfigured.
2481  *     Writing this, if accepted, will block until array is quiessent
2482  * readonly
2483  *     no resync can happen.  no superblocks get written.
2484  *     write requests fail
2485  * read-auto
2486  *     like readonly, but behaves like 'clean' on a write request.
2487  *
2488  * clean - no pending writes, but otherwise active.
2489  *     When written to inactive array, starts without resync
2490  *     If a write request arrives then
2491  *       if metadata is known, mark 'dirty' and switch to 'active'.
2492  *       if not known, block and switch to write-pending
2493  *     If written to an active array that has pending writes, then fails.
2494  * active
2495  *     fully active: IO and resync can be happening.
2496  *     When written to inactive array, starts with resync
2497  *
2498  * write-pending
2499  *     clean, but writes are blocked waiting for 'active' to be written.
2500  *
2501  * active-idle
2502  *     like active, but no writes have been seen for a while (100msec).
2503  *
2504  */
2505 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2506                    write_pending, active_idle, bad_word};
2507 static char *array_states[] = {
2508         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2509         "write-pending", "active-idle", NULL };
2510
2511 static int match_word(const char *word, char **list)
2512 {
2513         int n;
2514         for (n=0; list[n]; n++)
2515                 if (cmd_match(word, list[n]))
2516                         break;
2517         return n;
2518 }
2519
2520 static ssize_t
2521 array_state_show(mddev_t *mddev, char *page)
2522 {
2523         enum array_state st = inactive;
2524
2525         if (mddev->pers)
2526                 switch(mddev->ro) {
2527                 case 1:
2528                         st = readonly;
2529                         break;
2530                 case 2:
2531                         st = read_auto;
2532                         break;
2533                 case 0:
2534                         if (mddev->in_sync)
2535                                 st = clean;
2536                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2537                                 st = write_pending;
2538                         else if (mddev->safemode)
2539                                 st = active_idle;
2540                         else
2541                                 st = active;
2542                 }
2543         else {
2544                 if (list_empty(&mddev->disks) &&
2545                     mddev->raid_disks == 0 &&
2546                     mddev->size == 0)
2547                         st = clear;
2548                 else
2549                         st = inactive;
2550         }
2551         return sprintf(page, "%s\n", array_states[st]);
2552 }
2553
2554 static int do_md_stop(mddev_t * mddev, int ro);
2555 static int do_md_run(mddev_t * mddev);
2556 static int restart_array(mddev_t *mddev);
2557
2558 static ssize_t
2559 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2560 {
2561         int err = -EINVAL;
2562         enum array_state st = match_word(buf, array_states);
2563         switch(st) {
2564         case bad_word:
2565                 break;
2566         case clear:
2567                 /* stopping an active array */
2568                 if (atomic_read(&mddev->active) > 1)
2569                         return -EBUSY;
2570                 err = do_md_stop(mddev, 0);
2571                 break;
2572         case inactive:
2573                 /* stopping an active array */
2574                 if (mddev->pers) {
2575                         if (atomic_read(&mddev->active) > 1)
2576                                 return -EBUSY;
2577                         err = do_md_stop(mddev, 2);
2578                 } else
2579                         err = 0; /* already inactive */
2580                 break;
2581         case suspended:
2582                 break; /* not supported yet */
2583         case readonly:
2584                 if (mddev->pers)
2585                         err = do_md_stop(mddev, 1);
2586                 else {
2587                         mddev->ro = 1;
2588                         err = do_md_run(mddev);
2589                 }
2590                 break;
2591         case read_auto:
2592                 /* stopping an active array */
2593                 if (mddev->pers) {
2594                         err = do_md_stop(mddev, 1);
2595                         if (err == 0)
2596                                 mddev->ro = 2; /* FIXME mark devices writable */
2597                 } else {
2598                         mddev->ro = 2;
2599                         err = do_md_run(mddev);
2600                 }
2601                 break;
2602         case clean:
2603                 if (mddev->pers) {
2604                         restart_array(mddev);
2605                         spin_lock_irq(&mddev->write_lock);
2606                         if (atomic_read(&mddev->writes_pending) == 0) {
2607                                 if (mddev->in_sync == 0) {
2608                                         mddev->in_sync = 1;
2609                                         if (mddev->persistent)
2610                                                 set_bit(MD_CHANGE_CLEAN,
2611                                                         &mddev->flags);
2612                                 }
2613                                 err = 0;
2614                         } else
2615                                 err = -EBUSY;
2616                         spin_unlock_irq(&mddev->write_lock);
2617                 } else {
2618                         mddev->ro = 0;
2619                         mddev->recovery_cp = MaxSector;
2620                         err = do_md_run(mddev);
2621                 }
2622                 break;
2623         case active:
2624                 if (mddev->pers) {
2625                         restart_array(mddev);
2626                         if (mddev->external)
2627                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2628                         wake_up(&mddev->sb_wait);
2629                         err = 0;
2630                 } else {
2631                         mddev->ro = 0;
2632                         err = do_md_run(mddev);
2633                 }
2634                 break;
2635         case write_pending:
2636         case active_idle:
2637                 /* these cannot be set */
2638                 break;
2639         }
2640         if (err)
2641                 return err;
2642         else
2643                 return len;
2644 }
2645 static struct md_sysfs_entry md_array_state =
2646 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2647
2648 static ssize_t
2649 null_show(mddev_t *mddev, char *page)
2650 {
2651         return -EINVAL;
2652 }
2653
2654 static ssize_t
2655 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2656 {
2657         /* buf must be %d:%d\n? giving major and minor numbers */
2658         /* The new device is added to the array.
2659          * If the array has a persistent superblock, we read the
2660          * superblock to initialise info and check validity.
2661          * Otherwise, only checking done is that in bind_rdev_to_array,
2662          * which mainly checks size.
2663          */
2664         char *e;
2665         int major = simple_strtoul(buf, &e, 10);
2666         int minor;
2667         dev_t dev;
2668         mdk_rdev_t *rdev;
2669         int err;
2670
2671         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2672                 return -EINVAL;
2673         minor = simple_strtoul(e+1, &e, 10);
2674         if (*e && *e != '\n')
2675                 return -EINVAL;
2676         dev = MKDEV(major, minor);
2677         if (major != MAJOR(dev) ||
2678             minor != MINOR(dev))
2679                 return -EOVERFLOW;
2680
2681
2682         if (mddev->persistent) {
2683                 rdev = md_import_device(dev, mddev->major_version,
2684                                         mddev->minor_version);
2685                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2686                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2687                                                        mdk_rdev_t, same_set);
2688                         err = super_types[mddev->major_version]
2689                                 .load_super(rdev, rdev0, mddev->minor_version);
2690                         if (err < 0)
2691                                 goto out;
2692                 }
2693         } else if (mddev->external)
2694                 rdev = md_import_device(dev, -2, -1);
2695         else
2696                 rdev = md_import_device(dev, -1, -1);
2697
2698         if (IS_ERR(rdev))
2699                 return PTR_ERR(rdev);
2700         err = bind_rdev_to_array(rdev, mddev);
2701  out:
2702         if (err)
2703                 export_rdev(rdev);
2704         return err ? err : len;
2705 }
2706
2707 static struct md_sysfs_entry md_new_device =
2708 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2709
2710 static ssize_t
2711 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2712 {
2713         char *end;
2714         unsigned long chunk, end_chunk;
2715
2716         if (!mddev->bitmap)
2717                 goto out;
2718         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2719         while (*buf) {
2720                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2721                 if (buf == end) break;
2722                 if (*end == '-') { /* range */
2723                         buf = end + 1;
2724                         end_chunk = simple_strtoul(buf, &end, 0);
2725                         if (buf == end) break;
2726                 }
2727                 if (*end && !isspace(*end)) break;
2728                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2729                 buf = end;
2730                 while (isspace(*buf)) buf++;
2731         }
2732         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2733 out:
2734         return len;
2735 }
2736
2737 static struct md_sysfs_entry md_bitmap =
2738 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2739
2740 static ssize_t
2741 size_show(mddev_t *mddev, char *page)
2742 {
2743         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2744 }
2745
2746 static int update_size(mddev_t *mddev, unsigned long size);
2747
2748 static ssize_t
2749 size_store(mddev_t *mddev, const char *buf, size_t len)
2750 {
2751         /* If array is inactive, we can reduce the component size, but
2752          * not increase it (except from 0).
2753          * If array is active, we can try an on-line resize
2754          */
2755         char *e;
2756         int err = 0;
2757         unsigned long long size = simple_strtoull(buf, &e, 10);
2758         if (!*buf || *buf == '\n' ||
2759             (*e && *e != '\n'))
2760                 return -EINVAL;
2761
2762         if (mddev->pers) {
2763                 err = update_size(mddev, size);
2764                 md_update_sb(mddev, 1);
2765         } else {
2766                 if (mddev->size == 0 ||
2767                     mddev->size > size)
2768                         mddev->size = size;
2769                 else
2770                         err = -ENOSPC;
2771         }
2772         return err ? err : len;
2773 }
2774
2775 static struct md_sysfs_entry md_size =
2776 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2777
2778
2779 /* Metdata version.
2780  * This is one of
2781  *   'none' for arrays with no metadata (good luck...)
2782  *   'external' for arrays with externally managed metadata,
2783  * or N.M for internally known formats
2784  */
2785 static ssize_t
2786 metadata_show(mddev_t *mddev, char *page)
2787 {
2788         if (mddev->persistent)
2789                 return sprintf(page, "%d.%d\n",
2790                                mddev->major_version, mddev->minor_version);
2791         else if (mddev->external)
2792                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2793         else
2794                 return sprintf(page, "none\n");
2795 }
2796
2797 static ssize_t
2798 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2799 {
2800         int major, minor;
2801         char *e;
2802         if (!list_empty(&mddev->disks))
2803                 return -EBUSY;
2804
2805         if (cmd_match(buf, "none")) {
2806                 mddev->persistent = 0;
2807                 mddev->external = 0;
2808                 mddev->major_version = 0;
2809                 mddev->minor_version = 90;
2810                 return len;
2811         }
2812         if (strncmp(buf, "external:", 9) == 0) {
2813                 size_t namelen = len-9;
2814                 if (namelen >= sizeof(mddev->metadata_type))
2815                         namelen = sizeof(mddev->metadata_type)-1;
2816                 strncpy(mddev->metadata_type, buf+9, namelen);
2817                 mddev->metadata_type[namelen] = 0;
2818                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2819                         mddev->metadata_type[--namelen] = 0;
2820                 mddev->persistent = 0;
2821                 mddev->external = 1;
2822                 mddev->major_version = 0;
2823                 mddev->minor_version = 90;
2824                 return len;
2825         }
2826         major = simple_strtoul(buf, &e, 10);
2827         if (e==buf || *e != '.')
2828                 return -EINVAL;
2829         buf = e+1;
2830         minor = simple_strtoul(buf, &e, 10);
2831         if (e==buf || (*e && *e != '\n') )
2832                 return -EINVAL;
2833         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2834                 return -ENOENT;
2835         mddev->major_version = major;
2836         mddev->minor_version = minor;
2837         mddev->persistent = 1;
2838         mddev->external = 0;
2839         return len;
2840 }
2841
2842 static struct md_sysfs_entry md_metadata =
2843 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2844
2845 static ssize_t
2846 action_show(mddev_t *mddev, char *page)
2847 {
2848         char *type = "idle";
2849         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2850             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2851                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2852                         type = "reshape";
2853                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2854                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2855                                 type = "resync";
2856                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2857                                 type = "check";
2858                         else
2859                                 type = "repair";
2860                 } else
2861                         type = "recover";
2862         }
2863         return sprintf(page, "%s\n", type);
2864 }
2865
2866 static ssize_t
2867 action_store(mddev_t *mddev, const char *page, size_t len)
2868 {
2869         if (!mddev->pers || !mddev->pers->sync_request)
2870                 return -EINVAL;
2871
2872         if (cmd_match(page, "idle")) {
2873                 if (mddev->sync_thread) {
2874                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2875                         md_unregister_thread(mddev->sync_thread);
2876                         mddev->sync_thread = NULL;
2877                         mddev->recovery = 0;
2878                 }
2879         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2880                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2881                 return -EBUSY;
2882         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2883                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2884         else if (cmd_match(page, "reshape")) {
2885                 int err;
2886                 if (mddev->pers->start_reshape == NULL)
2887                         return -EINVAL;
2888                 err = mddev->pers->start_reshape(mddev);
2889                 if (err)
2890                         return err;
2891         } else {
2892                 if (cmd_match(page, "check"))
2893                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2894                 else if (!cmd_match(page, "repair"))
2895                         return -EINVAL;
2896                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2897                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2898         }
2899         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2900         md_wakeup_thread(mddev->thread);
2901         return len;
2902 }
2903
2904 static ssize_t
2905 mismatch_cnt_show(mddev_t *mddev, char *page)
2906 {
2907         return sprintf(page, "%llu\n",
2908                        (unsigned long long) mddev->resync_mismatches);
2909 }
2910
2911 static struct md_sysfs_entry md_scan_mode =
2912 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2913
2914
2915 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2916
2917 static ssize_t
2918 sync_min_show(mddev_t *mddev, char *page)
2919 {
2920         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2921                        mddev->sync_speed_min ? "local": "system");
2922 }
2923
2924 static ssize_t
2925 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2926 {
2927         int min;
2928         char *e;
2929         if (strncmp(buf, "system", 6)==0) {
2930                 mddev->sync_speed_min = 0;
2931                 return len;
2932         }
2933         min = simple_strtoul(buf, &e, 10);
2934         if (buf == e || (*e && *e != '\n') || min <= 0)
2935                 return -EINVAL;
2936         mddev->sync_speed_min = min;
2937         return len;
2938 }
2939
2940 static struct md_sysfs_entry md_sync_min =
2941 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2942
2943 static ssize_t
2944 sync_max_show(mddev_t *mddev, char *page)
2945 {
2946         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2947                        mddev->sync_speed_max ? "local": "system");
2948 }
2949
2950 static ssize_t
2951 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2952 {
2953         int max;
2954         char *e;
2955         if (strncmp(buf, "system", 6)==0) {
2956                 mddev->sync_speed_max = 0;
2957                 return len;
2958         }
2959         max = simple_strtoul(buf, &e, 10);
2960         if (buf == e || (*e && *e != '\n') || max <= 0)
2961                 return -EINVAL;
2962         mddev->sync_speed_max = max;
2963         return len;
2964 }
2965
2966 static struct md_sysfs_entry md_sync_max =
2967 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2968
2969 static ssize_t
2970 degraded_show(mddev_t *mddev, char *page)
2971 {
2972         return sprintf(page, "%d\n", mddev->degraded);
2973 }
2974 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2975
2976 static ssize_t
2977 sync_speed_show(mddev_t *mddev, char *page)
2978 {
2979         unsigned long resync, dt, db;
2980         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2981         dt = ((jiffies - mddev->resync_mark) / HZ);
2982         if (!dt) dt++;
2983         db = resync - (mddev->resync_mark_cnt);
2984         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2985 }
2986
2987 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2988
2989 static ssize_t
2990 sync_completed_show(mddev_t *mddev, char *page)
2991 {
2992         unsigned long max_blocks, resync;
2993
2994         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2995                 max_blocks = mddev->resync_max_sectors;
2996         else
2997                 max_blocks = mddev->size << 1;
2998
2999         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3000         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3001 }
3002
3003 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3004
3005 static ssize_t
3006 max_sync_show(mddev_t *mddev, char *page)
3007 {
3008         if (mddev->resync_max == MaxSector)
3009                 return sprintf(page, "max\n");
3010         else
3011                 return sprintf(page, "%llu\n",
3012                                (unsigned long long)mddev->resync_max);
3013 }
3014 static ssize_t
3015 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3016 {
3017         if (strncmp(buf, "max", 3) == 0)
3018                 mddev->resync_max = MaxSector;
3019         else {
3020                 char *ep;
3021                 unsigned long long max = simple_strtoull(buf, &ep, 10);
3022                 if (ep == buf || (*ep != 0 && *ep != '\n'))
3023                         return -EINVAL;
3024                 if (max < mddev->resync_max &&
3025                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3026                         return -EBUSY;
3027
3028                 /* Must be a multiple of chunk_size */
3029                 if (mddev->chunk_size) {
3030                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3031                                 return -EINVAL;
3032                 }
3033                 mddev->resync_max = max;
3034         }
3035         wake_up(&mddev->recovery_wait);
3036         return len;
3037 }
3038
3039 static struct md_sysfs_entry md_max_sync =
3040 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3041
3042 static ssize_t
3043 suspend_lo_show(mddev_t *mddev, char *page)
3044 {
3045         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3046 }
3047
3048 static ssize_t
3049 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3050 {
3051         char *e;
3052         unsigned long long new = simple_strtoull(buf, &e, 10);
3053
3054         if (mddev->pers->quiesce == NULL)
3055                 return -EINVAL;
3056         if (buf == e || (*e && *e != '\n'))
3057                 return -EINVAL;
3058         if (new >= mddev->suspend_hi ||
3059             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3060                 mddev->suspend_lo = new;
3061                 mddev->pers->quiesce(mddev, 2);
3062                 return len;
3063         } else
3064                 return -EINVAL;
3065 }
3066 static struct md_sysfs_entry md_suspend_lo =
3067 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3068
3069
3070 static ssize_t
3071 suspend_hi_show(mddev_t *mddev, char *page)
3072 {
3073         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3074 }
3075
3076 static ssize_t
3077 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3078 {
3079         char *e;
3080         unsigned long long new = simple_strtoull(buf, &e, 10);
3081
3082         if (mddev->pers->quiesce == NULL)
3083                 return -EINVAL;
3084         if (buf == e || (*e && *e != '\n'))
3085                 return -EINVAL;
3086         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3087             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3088                 mddev->suspend_hi = new;
3089                 mddev->pers->quiesce(mddev, 1);
3090                 mddev->pers->quiesce(mddev, 0);
3091                 return len;
3092         } else
3093                 return -EINVAL;
3094 }
3095 static struct md_sysfs_entry md_suspend_hi =
3096 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3097
3098 static ssize_t
3099 reshape_position_show(mddev_t *mddev, char *page)
3100 {
3101         if (mddev->reshape_position != MaxSector)
3102                 return sprintf(page, "%llu\n",
3103                                (unsigned long long)mddev->reshape_position);
3104         strcpy(page, "none\n");
3105         return 5;
3106 }
3107
3108 static ssize_t
3109 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3110 {
3111         char *e;
3112         unsigned long long new = simple_strtoull(buf, &e, 10);
3113         if (mddev->pers)
3114                 return -EBUSY;
3115         if (buf == e || (*e && *e != '\n'))
3116                 return -EINVAL;
3117         mddev->reshape_position = new;
3118         mddev->delta_disks = 0;
3119         mddev->new_level = mddev->level;
3120         mddev->new_layout = mddev->layout;
3121         mddev->new_chunk = mddev->chunk_size;
3122         return len;
3123 }
3124
3125 static struct md_sysfs_entry md_reshape_position =
3126 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3127        reshape_position_store);
3128
3129
3130 static struct attribute *md_default_attrs[] = {
3131         &md_level.attr,
3132         &md_layout.attr,
3133         &md_raid_disks.attr,
3134         &md_chunk_size.attr,
3135         &md_size.attr,
3136         &md_resync_start.attr,
3137         &md_metadata.attr,
3138         &md_new_device.attr,
3139         &md_safe_delay.attr,
3140         &md_array_state.attr,
3141         &md_reshape_position.attr,
3142         NULL,
3143 };
3144
3145 static struct attribute *md_redundancy_attrs[] = {
3146         &md_scan_mode.attr,
3147         &md_mismatches.attr,
3148         &md_sync_min.attr,
3149         &md_sync_max.attr,
3150         &md_sync_speed.attr,
3151         &md_sync_completed.attr,
3152         &md_max_sync.attr,
3153         &md_suspend_lo.attr,
3154         &md_suspend_hi.attr,
3155         &md_bitmap.attr,
3156         &md_degraded.attr,
3157         NULL,
3158 };
3159 static struct attribute_group md_redundancy_group = {
3160         .name = NULL,
3161         .attrs = md_redundancy_attrs,
3162 };
3163
3164
3165 static ssize_t
3166 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3167 {
3168         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3169         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3170         ssize_t rv;
3171
3172         if (!entry->show)
3173                 return -EIO;
3174         rv = mddev_lock(mddev);
3175         if (!rv) {
3176                 rv = entry->show(mddev, page);
3177                 mddev_unlock(mddev);
3178         }
3179         return rv;
3180 }
3181
3182 static ssize_t
3183 md_attr_store(struct kobject *kobj, struct attribute *attr,
3184               const char *page, size_t length)
3185 {
3186         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3187         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3188         ssize_t rv;
3189
3190         if (!entry->store)
3191                 return -EIO;
3192         if (!capable(CAP_SYS_ADMIN))
3193                 return -EACCES;
3194         rv = mddev_lock(mddev);
3195         if (!rv) {
3196                 rv = entry->store(mddev, page, length);
3197                 mddev_unlock(mddev);
3198         }
3199         return rv;
3200 }
3201
3202 static void md_free(struct kobject *ko)
3203 {
3204         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3205         kfree(mddev);
3206 }
3207
3208 static struct sysfs_ops md_sysfs_ops = {
3209         .show   = md_attr_show,
3210         .store  = md_attr_store,
3211 };
3212 static struct kobj_type md_ktype = {
3213         .release        = md_free,
3214         .sysfs_ops      = &md_sysfs_ops,
3215         .default_attrs  = md_default_attrs,
3216 };
3217
3218 int mdp_major = 0;
3219
3220 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3221 {
3222         static DEFINE_MUTEX(disks_mutex);
3223         mddev_t *mddev = mddev_find(dev);
3224         struct gendisk *disk;
3225         int partitioned = (MAJOR(dev) != MD_MAJOR);
3226         int shift = partitioned ? MdpMinorShift : 0;
3227         int unit = MINOR(dev) >> shift;
3228         int error;
3229
3230         if (!mddev)
3231                 return NULL;
3232
3233         mutex_lock(&disks_mutex);
3234         if (mddev->gendisk) {
3235                 mutex_unlock(&disks_mutex);
3236                 mddev_put(mddev);
3237                 return NULL;
3238         }
3239         disk = alloc_disk(1 << shift);
3240         if (!disk) {
3241                 mutex_unlock(&disks_mutex);
3242                 mddev_put(mddev);
3243                 return NULL;
3244         }
3245         disk->major = MAJOR(dev);
3246         disk->first_minor = unit << shift;
3247         if (partitioned)
3248                 sprintf(disk->disk_name, "md_d%d", unit);
3249         else
3250                 sprintf(disk->disk_name, "md%d", unit);
3251         disk->fops = &md_fops;
3252         disk->private_data = mddev;
3253         disk->queue = mddev->queue;
3254         add_disk(disk);
3255         mddev->gendisk = disk;
3256         mutex_unlock(&disks_mutex);
3257         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3258                                      "%s", "md");
3259         if (error)
3260                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3261                        disk->disk_name);
3262         else
3263                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3264         return NULL;
3265 }
3266
3267 static void md_safemode_timeout(unsigned long data)
3268 {
3269         mddev_t *mddev = (mddev_t *) data;
3270
3271         mddev->safemode = 1;
3272         md_wakeup_thread(mddev->thread);
3273 }
3274
3275 static int start_dirty_degraded;
3276
3277 static int do_md_run(mddev_t * mddev)
3278 {
3279         int err;
3280         int chunk_size;
3281         struct list_head *tmp;
3282         mdk_rdev_t *rdev;
3283         struct gendisk *disk;
3284         struct mdk_personality *pers;
3285         char b[BDEVNAME_SIZE];
3286
3287         if (list_empty(&mddev->disks))
3288                 /* cannot run an array with no devices.. */
3289                 return -EINVAL;
3290
3291         if (mddev->pers)
3292                 return -EBUSY;
3293
3294         /*
3295          * Analyze all RAID superblock(s)
3296          */
3297         if (!mddev->raid_disks) {
3298                 if (!mddev->persistent)
3299                         return -EINVAL;
3300                 analyze_sbs(mddev);
3301         }
3302
3303         chunk_size = mddev->chunk_size;
3304
3305         if (chunk_size) {
3306                 if (chunk_size > MAX_CHUNK_SIZE) {
3307                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3308                                 chunk_size, MAX_CHUNK_SIZE);
3309                         return -EINVAL;
3310                 }
3311                 /*
3312                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3313                  */
3314                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3315                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3316                         return -EINVAL;
3317                 }
3318                 if (chunk_size < PAGE_SIZE) {
3319                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3320                                 chunk_size, PAGE_SIZE);
3321                         return -EINVAL;
3322                 }
3323
3324                 /* devices must have minimum size of one chunk */
3325                 rdev_for_each(rdev, tmp, mddev) {
3326                         if (test_bit(Faulty, &rdev->flags))
3327                                 continue;
3328                         if (rdev->size < chunk_size / 1024) {
3329                                 printk(KERN_WARNING
3330                                         "md: Dev %s smaller than chunk_size:"
3331                                         " %lluk < %dk\n",
3332                                         bdevname(rdev->bdev,b),
3333                                         (unsigned long long)rdev->size,
3334                                         chunk_size / 1024);
3335                                 return -EINVAL;
3336                         }
3337                 }
3338         }
3339
3340 #ifdef CONFIG_KMOD
3341         if (mddev->level != LEVEL_NONE)
3342                 request_module("md-level-%d", mddev->level);
3343         else if (mddev->clevel[0])
3344                 request_module("md-%s", mddev->clevel);
3345 #endif
3346
3347         /*
3348          * Drop all container device buffers, from now on
3349          * the only valid external interface is through the md
3350          * device.
3351          */
3352         rdev_for_each(rdev, tmp, mddev) {
3353                 if (test_bit(Faulty, &rdev->flags))
3354                         continue;
3355                 sync_blockdev(rdev->bdev);
3356                 invalidate_bdev(rdev->bdev);
3357
3358                 /* perform some consistency tests on the device.
3359                  * We don't want the data to overlap the metadata,
3360                  * Internal Bitmap issues has handled elsewhere.
3361                  */
3362                 if (rdev->data_offset < rdev->sb_offset) {
3363                         if (mddev->size &&
3364                             rdev->data_offset + mddev->size*2
3365                             > rdev->sb_offset*2) {
3366                                 printk("md: %s: data overlaps metadata\n",
3367                                        mdname(mddev));
3368                                 return -EINVAL;
3369                         }
3370                 } else {
3371                         if (rdev->sb_offset*2 + rdev->sb_size/512
3372                             > rdev->data_offset) {
3373                                 printk("md: %s: metadata overlaps data\n",
3374                                        mdname(mddev));
3375                                 return -EINVAL;
3376                         }
3377                 }
3378         }
3379
3380         md_probe(mddev->unit, NULL, NULL);
3381         disk = mddev->gendisk;
3382         if (!disk)
3383                 return -ENOMEM;
3384
3385         spin_lock(&pers_lock);
3386         pers = find_pers(mddev->level, mddev->clevel);
3387         if (!pers || !try_module_get(pers->owner)) {
3388                 spin_unlock(&pers_lock);
3389                 if (mddev->level != LEVEL_NONE)
3390                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3391                                mddev->level);
3392                 else
3393                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3394                                mddev->clevel);
3395                 return -EINVAL;
3396         }
3397         mddev->pers = pers;
3398         spin_unlock(&pers_lock);
3399         mddev->level = pers->level;
3400         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3401
3402         if (mddev->reshape_position != MaxSector &&
3403             pers->start_reshape == NULL) {
3404                 /* This personality cannot handle reshaping... */
3405                 mddev->pers = NULL;
3406                 module_put(pers->owner);
3407                 return -EINVAL;
3408         }
3409
3410         if (pers->sync_request) {
3411                 /* Warn if this is a potentially silly
3412                  * configuration.
3413                  */
3414                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3415                 mdk_rdev_t *rdev2;
3416                 struct list_head *tmp2;
3417                 int warned = 0;
3418                 rdev_for_each(rdev, tmp, mddev) {
3419                         rdev_for_each(rdev2, tmp2, mddev) {
3420                                 if (rdev < rdev2 &&
3421                                     rdev->bdev->bd_contains ==
3422                                     rdev2->bdev->bd_contains) {
3423                                         printk(KERN_WARNING
3424                                                "%s: WARNING: %s appears to be"
3425                                                " on the same physical disk as"
3426                                                " %s.\n",
3427                                                mdname(mddev),
3428                                                bdevname(rdev->bdev,b),
3429                                                bdevname(rdev2->bdev,b2));
3430                                         warned = 1;
3431                                 }
3432                         }
3433                 }
3434                 if (warned)
3435                         printk(KERN_WARNING
3436                                "True protection against single-disk"
3437                                " failure might be compromised.\n");
3438         }
3439
3440         mddev->recovery = 0;
3441         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3442         mddev->barriers_work = 1;
3443         mddev->ok_start_degraded = start_dirty_degraded;
3444
3445         if (start_readonly)
3446                 mddev->ro = 2; /* read-only, but switch on first write */
3447
3448         err = mddev->pers->run(mddev);
3449         if (!err && mddev->pers->sync_request) {
3450                 err = bitmap_create(mddev);
3451                 if (err) {
3452                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3453                                mdname(mddev), err);
3454                         mddev->pers->stop(mddev);
3455                 }
3456         }
3457         if (err) {
3458                 printk(KERN_ERR "md: pers->run() failed ...\n");
3459                 module_put(mddev->pers->owner);
3460                 mddev->pers = NULL;
3461                 bitmap_destroy(mddev);
3462                 return err;
3463         }
3464         if (mddev->pers->sync_request) {
3465                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3466                         printk(KERN_WARNING
3467                                "md: cannot register extra attributes for %s\n",
3468                                mdname(mddev));
3469         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3470                 mddev->ro = 0;
3471
3472         atomic_set(&mddev->writes_pending,0);
3473         mddev->safemode = 0;
3474         mddev->safemode_timer.function = md_safemode_timeout;
3475         mddev->safemode_timer.data = (unsigned long) mddev;
3476         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3477         mddev->in_sync = 1;
3478
3479         rdev_for_each(rdev, tmp, mddev)
3480                 if (rdev->raid_disk >= 0) {
3481                         char nm[20];
3482                         sprintf(nm, "rd%d", rdev->raid_disk);
3483                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3484                                 printk("md: cannot register %s for %s\n",
3485                                        nm, mdname(mddev));
3486                 }
3487
3488         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3489
3490         if (mddev->flags)
3491                 md_update_sb(mddev, 0);
3492
3493         set_capacity(disk, mddev->array_size<<1);
3494
3495         /* If we call blk_queue_make_request here, it will
3496          * re-initialise max_sectors etc which may have been
3497          * refined inside -> run.  So just set the bits we need to set.
3498          * Most initialisation happended when we called
3499          * blk_queue_make_request(..., md_fail_request)
3500          * earlier.
3501          */
3502         mddev->queue->queuedata = mddev;
3503         mddev->queue->make_request_fn = mddev->pers->make_request;
3504
3505         /* If there is a partially-recovered drive we need to
3506          * start recovery here.  If we leave it to md_check_recovery,
3507          * it will remove the drives and not do the right thing
3508          */
3509         if (mddev->degraded && !mddev->sync_thread) {
3510                 struct list_head *rtmp;
3511                 int spares = 0;
3512                 rdev_for_each(rdev, rtmp, mddev)
3513                         if (rdev->raid_disk >= 0 &&
3514                             !test_bit(In_sync, &rdev->flags) &&
3515                             !test_bit(Faulty, &rdev->flags))
3516                                 /* complete an interrupted recovery */
3517                                 spares++;
3518                 if (spares && mddev->pers->sync_request) {
3519                         mddev->recovery = 0;
3520                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3521                         mddev->sync_thread = md_register_thread(md_do_sync,
3522                                                                 mddev,
3523                                                                 "%s_resync");
3524                         if (!mddev->sync_thread) {
3525                                 printk(KERN_ERR "%s: could not start resync"
3526                                        " thread...\n",
3527                                        mdname(mddev));
3528                                 /* leave the spares where they are, it shouldn't hurt */
3529                                 mddev->recovery = 0;
3530                         }
3531                 }
3532         }
3533         md_wakeup_thread(mddev->thread);
3534         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3535
3536         mddev->changed = 1;
3537         md_new_event(mddev);
3538         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3539         return 0;
3540 }
3541
3542 static int restart_array(mddev_t *mddev)
3543 {
3544         struct gendisk *disk = mddev->gendisk;
3545         int err;
3546
3547         /*
3548          * Complain if it has no devices
3549          */
3550         err = -ENXIO;
3551         if (list_empty(&mddev->disks))
3552                 goto out;
3553
3554         if (mddev->pers) {
3555                 err = -EBUSY;
3556                 if (!mddev->ro)
3557                         goto out;
3558
3559                 mddev->safemode = 0;
3560                 mddev->ro = 0;
3561                 set_disk_ro(disk, 0);
3562
3563                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3564                         mdname(mddev));
3565                 /*
3566                  * Kick recovery or resync if necessary
3567                  */
3568                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3569                 md_wakeup_thread(mddev->thread);
3570                 md_wakeup_thread(mddev->sync_thread);
3571                 err = 0;
3572         } else
3573                 err = -EINVAL;
3574
3575 out:
3576         return err;
3577 }
3578
3579 /* similar to deny_write_access, but accounts for our holding a reference
3580  * to the file ourselves */
3581 static int deny_bitmap_write_access(struct file * file)
3582 {
3583         struct inode *inode = file->f_mapping->host;
3584
3585         spin_lock(&inode->i_lock);
3586         if (atomic_read(&inode->i_writecount) > 1) {
3587                 spin_unlock(&inode->i_lock);
3588                 return -ETXTBSY;
3589         }
3590         atomic_set(&inode->i_writecount, -1);
3591         spin_unlock(&inode->i_lock);
3592
3593         return 0;
3594 }
3595
3596 static void restore_bitmap_write_access(struct file *file)
3597 {
3598         struct inode *inode = file->f_mapping->host;
3599
3600         spin_lock(&inode->i_lock);
3601         atomic_set(&inode->i_writecount, 1);
3602         spin_unlock(&inode->i_lock);
3603 }
3604
3605 /* mode:
3606  *   0 - completely stop and dis-assemble array
3607  *   1 - switch to readonly
3608  *   2 - stop but do not disassemble array
3609  */
3610 static int do_md_stop(mddev_t * mddev, int mode)
3611 {
3612         int err = 0;
3613         struct gendisk *disk = mddev->gendisk;
3614
3615         if (mddev->pers) {
3616                 if (atomic_read(&mddev->active)>2) {
3617                         printk("md: %s still in use.\n",mdname(mddev));
3618                         return -EBUSY;
3619                 }
3620
3621                 if (mddev->sync_thread) {
3622                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3623                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3624                         md_unregister_thread(mddev->sync_thread);
3625                         mddev->sync_thread = NULL;
3626                 }
3627
3628                 del_timer_sync(&mddev->safemode_timer);
3629
3630                 invalidate_partition(disk, 0);
3631
3632                 switch(mode) {
3633                 case 1: /* readonly */
3634                         err  = -ENXIO;
3635                         if (mddev->ro==1)
3636                                 goto out;
3637                         mddev->ro = 1;
3638                         break;
3639                 case 0: /* disassemble */
3640                 case 2: /* stop */
3641                         bitmap_flush(mddev);
3642                         md_super_wait(mddev);
3643                         if (mddev->ro)
3644                                 set_disk_ro(disk, 0);
3645                         blk_queue_make_request(mddev->queue, md_fail_request);
3646                         mddev->pers->stop(mddev);
3647                         mddev->queue->merge_bvec_fn = NULL;
3648                         mddev->queue->unplug_fn = NULL;
3649                         mddev->queue->backing_dev_info.congested_fn = NULL;
3650                         if (mddev->pers->sync_request)
3651                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3652
3653                         module_put(mddev->pers->owner);
3654                         mddev->pers = NULL;
3655
3656                         set_capacity(disk, 0);
3657                         mddev->changed = 1;
3658
3659                         if (mddev->ro)
3660                                 mddev->ro = 0;
3661                 }
3662                 if (!mddev->in_sync || mddev->flags) {
3663                         /* mark array as shutdown cleanly */
3664                         mddev->in_sync = 1;
3665                         md_update_sb(mddev, 1);
3666                 }
3667                 if (mode == 1)
3668                         set_disk_ro(disk, 1);
3669                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3670         }
3671
3672         /*
3673          * Free resources if final stop
3674          */
3675         if (mode == 0) {
3676                 mdk_rdev_t *rdev;
3677                 struct list_head *tmp;
3678
3679                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3680
3681                 bitmap_destroy(mddev);
3682                 if (mddev->bitmap_file) {
3683                         restore_bitmap_write_access(mddev->bitmap_file);
3684                         fput(mddev->bitmap_file);
3685                         mddev->bitmap_file = NULL;
3686                 }
3687                 mddev->bitmap_offset = 0;
3688
3689                 rdev_for_each(rdev, tmp, mddev)
3690                         if (rdev->raid_disk >= 0) {
3691                                 char nm[20];
3692                                 sprintf(nm, "rd%d", rdev->raid_disk);
3693                                 sysfs_remove_link(&mddev->kobj, nm);
3694                         }
3695
3696                 /* make sure all md_delayed_delete calls have finished */
3697                 flush_scheduled_work();
3698
3699                 export_array(mddev);
3700
3701                 mddev->array_size = 0;
3702                 mddev->size = 0;
3703                 mddev->raid_disks = 0;
3704                 mddev->recovery_cp = 0;
3705                 mddev->resync_max = MaxSector;
3706                 mddev->reshape_position = MaxSector;
3707                 mddev->external = 0;
3708                 mddev->persistent = 0;
3709
3710         } else if (mddev->pers)
3711                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3712                         mdname(mddev));
3713         err = 0;
3714         md_new_event(mddev);
3715 out:
3716         return err;
3717 }
3718
3719 #ifndef MODULE
3720 static void autorun_array(mddev_t *mddev)
3721 {
3722         mdk_rdev_t *rdev;
3723         struct list_head *tmp;
3724         int err;
3725
3726         if (list_empty(&mddev->disks))
3727                 return;
3728
3729         printk(KERN_INFO "md: running: ");
3730
3731         rdev_for_each(rdev, tmp, mddev) {
3732                 char b[BDEVNAME_SIZE];
3733                 printk("<%s>", bdevname(rdev->bdev,b));
3734         }
3735         printk("\n");
3736
3737         err = do_md_run (mddev);
3738         if (err) {
3739                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3740                 do_md_stop (mddev, 0);
3741         }
3742 }
3743
3744 /*
3745  * lets try to run arrays based on all disks that have arrived
3746  * until now. (those are in pending_raid_disks)
3747  *
3748  * the method: pick the first pending disk, collect all disks with
3749  * the same UUID, remove all from the pending list and put them into
3750  * the 'same_array' list. Then order this list based on superblock
3751  * update time (freshest comes first), kick out 'old' disks and
3752  * compare superblocks. If everything's fine then run it.
3753  *
3754  * If "unit" is allocated, then bump its reference count
3755  */
3756 static void autorun_devices(int part)
3757 {
3758         struct list_head *tmp;
3759         mdk_rdev_t *rdev0, *rdev;
3760         mddev_t *mddev;
3761         char b[BDEVNAME_SIZE];
3762
3763         printk(KERN_INFO "md: autorun ...\n");
3764         while (!list_empty(&pending_raid_disks)) {
3765                 int unit;
3766                 dev_t dev;
3767                 LIST_HEAD(candidates);
3768                 rdev0 = list_entry(pending_raid_disks.next,
3769                                          mdk_rdev_t, same_set);
3770
3771                 printk(KERN_INFO "md: considering %s ...\n",
3772                         bdevname(rdev0->bdev,b));
3773                 INIT_LIST_HEAD(&candidates);
3774                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3775                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3776                                 printk(KERN_INFO "md:  adding %s ...\n",
3777                                         bdevname(rdev->bdev,b));
3778                                 list_move(&rdev->same_set, &candidates);
3779                         }
3780                 /*
3781                  * now we have a set of devices, with all of them having
3782                  * mostly sane superblocks. It's time to allocate the
3783                  * mddev.
3784                  */
3785                 if (part) {
3786                         dev = MKDEV(mdp_major,
3787                                     rdev0->preferred_minor << MdpMinorShift);
3788                         unit = MINOR(dev) >> MdpMinorShift;
3789                 } else {
3790                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3791                         unit = MINOR(dev);
3792                 }
3793                 if (rdev0->preferred_minor != unit) {
3794                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3795                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3796                         break;
3797                 }
3798
3799                 md_probe(dev, NULL, NULL);
3800                 mddev = mddev_find(dev);
3801                 if (!mddev) {
3802                         printk(KERN_ERR
3803                                 "md: cannot allocate memory for md drive.\n");
3804                         break;
3805                 }
3806                 if (mddev_lock(mddev))
3807                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3808                                mdname(mddev));
3809                 else if (mddev->raid_disks || mddev->major_version
3810                          || !list_empty(&mddev->disks)) {
3811                         printk(KERN_WARNING
3812                                 "md: %s already running, cannot run %s\n",
3813                                 mdname(mddev), bdevname(rdev0->bdev,b));
3814                         mddev_unlock(mddev);
3815                 } else {
3816                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3817                         mddev->persistent = 1;
3818                         rdev_for_each_list(rdev, tmp, candidates) {
3819                                 list_del_init(&rdev->same_set);
3820                                 if (bind_rdev_to_array(rdev, mddev))
3821                                         export_rdev(rdev);
3822                         }
3823                         autorun_array(mddev);
3824                         mddev_unlock(mddev);
3825                 }
3826                 /* on success, candidates will be empty, on error
3827                  * it won't...
3828                  */
3829                 rdev_for_each_list(rdev, tmp, candidates)
3830                         export_rdev(rdev);
3831                 mddev_put(mddev);
3832         }
3833         printk(KERN_INFO "md: ... autorun DONE.\n");
3834 }
3835 #endif /* !MODULE */
3836
3837 static int get_version(void __user * arg)
3838 {
3839         mdu_version_t ver;
3840
3841         ver.major = MD_MAJOR_VERSION;
3842         ver.minor = MD_MINOR_VERSION;
3843         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3844
3845         if (copy_to_user(arg, &ver, sizeof(ver)))
3846                 return -EFAULT;
3847
3848         return 0;
3849 }
3850
3851 static int get_array_info(mddev_t * mddev, void __user * arg)
3852 {
3853         mdu_array_info_t info;
3854         int nr,working,active,failed,spare;
3855         mdk_rdev_t *rdev;
3856         struct list_head *tmp;
3857
3858         nr=working=active=failed=spare=0;
3859         rdev_for_each(rdev, tmp, mddev) {
3860                 nr++;
3861                 if (test_bit(Faulty, &rdev->flags))
3862                         failed++;
3863                 else {
3864                         working++;
3865                         if (test_bit(In_sync, &rdev->flags))
3866                                 active++;
3867                         else
3868                                 spare++;
3869                 }
3870         }
3871
3872         info.major_version = mddev->major_version;
3873         info.minor_version = mddev->minor_version;
3874         info.patch_version = MD_PATCHLEVEL_VERSION;
3875         info.ctime         = mddev->ctime;
3876         info.level         = mddev->level;
3877         info.size          = mddev->size;
3878         if (info.size != mddev->size) /* overflow */
3879                 info.size = -1;
3880         info.nr_disks      = nr;
3881         info.raid_disks    = mddev->raid_disks;
3882         info.md_minor      = mddev->md_minor;
3883         info.not_persistent= !mddev->persistent;
3884
3885         info.utime         = mddev->utime;
3886         info.state         = 0;
3887         if (mddev->in_sync)
3888                 info.state = (1<<MD_SB_CLEAN);
3889         if (mddev->bitmap && mddev->bitmap_offset)
3890                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3891         info.active_disks  = active;
3892         info.working_disks = working;
3893         info.failed_disks  = failed;
3894         info.spare_disks   = spare;
3895
3896         info.layout        = mddev->layout;
3897         info.chunk_size    = mddev->chunk_size;
3898
3899         if (copy_to_user(arg, &info, sizeof(info)))
3900                 return -EFAULT;
3901
3902         return 0;
3903 }
3904
3905 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3906 {
3907         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3908         char *ptr, *buf = NULL;
3909         int err = -ENOMEM;
3910
3911         md_allow_write(mddev);
3912
3913         file = kmalloc(sizeof(*file), GFP_KERNEL);
3914         if (!file)
3915                 goto out;
3916
3917         /* bitmap disabled, zero the first byte and copy out */
3918         if (!mddev->bitmap || !mddev->bitmap->file) {
3919                 file->pathname[0] = '\0';
3920                 goto copy_out;
3921         }
3922
3923         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3924         if (!buf)
3925                 goto out;
3926
3927         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3928         if (!ptr)
3929                 goto out;
3930
3931         strcpy(file->pathname, ptr);
3932
3933 copy_out:
3934         err = 0;
3935         if (copy_to_user(arg, file, sizeof(*file)))
3936                 err = -EFAULT;
3937 out:
3938         kfree(buf);
3939         kfree(file);
3940         return err;
3941 }
3942
3943 static int get_disk_info(mddev_t * mddev, void __user * arg)
3944 {
3945         mdu_disk_info_t info;
3946         unsigned int nr;
3947         mdk_rdev_t *rdev;
3948
3949         if (copy_from_user(&info, arg, sizeof(info)))
3950                 return -EFAULT;
3951
3952         nr = info.number;
3953
3954         rdev = find_rdev_nr(mddev, nr);
3955         if (rdev) {
3956                 info.major = MAJOR(rdev->bdev->bd_dev);
3957                 info.minor = MINOR(rdev->bdev->bd_dev);
3958                 info.raid_disk = rdev->raid_disk;
3959                 info.state = 0;
3960                 if (test_bit(Faulty, &rdev->flags))
3961                         info.state |= (1<<MD_DISK_FAULTY);
3962                 else if (test_bit(In_sync, &rdev->flags)) {
3963                         info.state |= (1<<MD_DISK_ACTIVE);
3964                         info.state |= (1<<MD_DISK_SYNC);
3965                 }
3966                 if (test_bit(WriteMostly, &rdev->flags))
3967                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3968         } else {
3969                 info.major = info.minor = 0;
3970                 info.raid_disk = -1;
3971                 info.state = (1<<MD_DISK_REMOVED);
3972         }
3973
3974         if (copy_to_user(arg, &info, sizeof(info)))
3975                 return -EFAULT;
3976
3977         return 0;
3978 }
3979
3980 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3981 {
3982         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3983         mdk_rdev_t *rdev;
3984         dev_t dev = MKDEV(info->major,info->minor);
3985
3986         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3987                 return -EOVERFLOW;
3988
3989         if (!mddev->raid_disks) {
3990                 int err;
3991                 /* expecting a device which has a superblock */
3992                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3993                 if (IS_ERR(rdev)) {
3994                         printk(KERN_WARNING
3995                                 "md: md_import_device returned %ld\n",
3996                                 PTR_ERR(rdev));
3997                         return PTR_ERR(rdev);
3998                 }
3999                 if (!list_empty(&mddev->disks)) {
4000                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4001                                                         mdk_rdev_t, same_set);
4002                         int err = super_types[mddev->major_version]
4003                                 .load_super(rdev, rdev0, mddev->minor_version);
4004                         if (err < 0) {
4005                                 printk(KERN_WARNING
4006                                         "md: %s has different UUID to %s\n",
4007                                         bdevname(rdev->bdev,b),
4008                                         bdevname(rdev0->bdev,b2));
4009                                 export_rdev(rdev);
4010                                 return -EINVAL;
4011                         }
4012                 }
4013                 err = bind_rdev_to_array(rdev, mddev);
4014                 if (err)
4015                         export_rdev(rdev);
4016                 return err;
4017         }
4018
4019         /*
4020          * add_new_disk can be used once the array is assembled
4021          * to add "hot spares".  They must already have a superblock
4022          * written
4023          */
4024         if (mddev->pers) {
4025                 int err;
4026                 if (!mddev->pers->hot_add_disk) {
4027                         printk(KERN_WARNING
4028                                 "%s: personality does not support diskops!\n",
4029                                mdname(mddev));
4030                         return -EINVAL;
4031                 }
4032                 if (mddev->persistent)
4033                         rdev = md_import_device(dev, mddev->major_version,
4034                                                 mddev->minor_version);
4035                 else
4036                         rdev = md_import_device(dev, -1, -1);
4037                 if (IS_ERR(rdev)) {
4038                         printk(KERN_WARNING
4039                                 "md: md_import_device returned %ld\n",
4040                                 PTR_ERR(rdev));
4041                         return PTR_ERR(rdev);
4042                 }
4043                 /* set save_raid_disk if appropriate */
4044                 if (!mddev->persistent) {
4045                         if (info->state & (1<<MD_DISK_SYNC)  &&
4046                             info->raid_disk < mddev->raid_disks)
4047                                 rdev->raid_disk = info->raid_disk;
4048                         else
4049                                 rdev->raid_disk = -1;
4050                 } else
4051                         super_types[mddev->major_version].
4052                                 validate_super(mddev, rdev);
4053                 rdev->saved_raid_disk = rdev->raid_disk;
4054
4055                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4056                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4057                         set_bit(WriteMostly, &rdev->flags);
4058
4059                 rdev->raid_disk = -1;
4060                 err = bind_rdev_to_array(rdev, mddev);
4061                 if (!err && !mddev->pers->hot_remove_disk) {
4062                         /* If there is hot_add_disk but no hot_remove_disk
4063                          * then added disks for geometry changes,
4064                          * and should be added immediately.
4065                          */
4066                         super_types[mddev->major_version].
4067                                 validate_super(mddev, rdev);
4068                         err = mddev->pers->hot_add_disk(mddev, rdev);
4069                         if (err)
4070                                 unbind_rdev_from_array(rdev);
4071                 }
4072                 if (err)
4073                         export_rdev(rdev);
4074
4075                 md_update_sb(mddev, 1);
4076                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4077                 md_wakeup_thread(mddev->thread);
4078                 return err;
4079         }
4080
4081         /* otherwise, add_new_disk is only allowed
4082          * for major_version==0 superblocks
4083          */
4084         if (mddev->major_version != 0) {
4085                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4086                        mdname(mddev));
4087                 return -EINVAL;
4088         }
4089
4090         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4091                 int err;
4092                 rdev = md_import_device (dev, -1, 0);
4093                 if (IS_ERR(rdev)) {
4094                         printk(KERN_WARNING
4095                                 "md: error, md_import_device() returned %ld\n",
4096                                 PTR_ERR(rdev));
4097                         return PTR_ERR(rdev);
4098                 }
4099                 rdev->desc_nr = info->number;
4100                 if (info->raid_disk < mddev->raid_disks)
4101                         rdev->raid_disk = info->raid_disk;
4102                 else
4103                         rdev->raid_disk = -1;
4104
4105                 if (rdev->raid_disk < mddev->raid_disks)
4106                         if (info->state & (1<<MD_DISK_SYNC))
4107                                 set_bit(In_sync, &rdev->flags);
4108
4109                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4110                         set_bit(WriteMostly, &rdev->flags);
4111
4112                 if (!mddev->persistent) {
4113                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4114                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4115                 } else
4116                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4117                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4118
4119                 err = bind_rdev_to_array(rdev, mddev);
4120                 if (err) {
4121                         export_rdev(rdev);
4122                         return err;
4123                 }
4124         }
4125
4126         return 0;
4127 }
4128
4129 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4130 {
4131         char b[BDEVNAME_SIZE];
4132         mdk_rdev_t *rdev;
4133
4134         if (!mddev->pers)
4135                 return -ENODEV;
4136
4137         rdev = find_rdev(mddev, dev);
4138         if (!rdev)
4139                 return -ENXIO;
4140
4141         if (rdev->raid_disk >= 0)
4142                 goto busy;
4143
4144         kick_rdev_from_array(rdev);
4145         md_update_sb(mddev, 1);
4146         md_new_event(mddev);
4147
4148         return 0;
4149 busy:
4150         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
4151                 bdevname(rdev->bdev,b), mdname(mddev));
4152         return -EBUSY;
4153 }
4154
4155 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4156 {
4157         char b[BDEVNAME_SIZE];
4158         int err;
4159         unsigned int size;
4160         mdk_rdev_t *rdev;
4161
4162         if (!mddev->pers)
4163                 return -ENODEV;
4164
4165         if (mddev->major_version != 0) {
4166                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4167                         " version-0 superblocks.\n",
4168                         mdname(mddev));
4169                 return -EINVAL;
4170         }
4171         if (!mddev->pers->hot_add_disk) {
4172                 printk(KERN_WARNING
4173                         "%s: personality does not support diskops!\n",
4174                         mdname(mddev));
4175                 return -EINVAL;
4176         }
4177
4178         rdev = md_import_device (dev, -1, 0);
4179         if (IS_ERR(rdev)) {
4180                 printk(KERN_WARNING
4181                         "md: error, md_import_device() returned %ld\n",
4182                         PTR_ERR(rdev));
4183                 return -EINVAL;
4184         }
4185
4186         if (mddev->persistent)
4187                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4188         else
4189                 rdev->sb_offset =
4190                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4191
4192         size = calc_dev_size(rdev, mddev->chunk_size);
4193         rdev->size = size;
4194
4195         if (test_bit(Faulty, &rdev->flags)) {
4196                 printk(KERN_WARNING
4197                         "md: can not hot-add faulty %s disk to %s!\n",
4198                         bdevname(rdev->bdev,b), mdname(mddev));
4199                 err = -EINVAL;
4200                 goto abort_export;
4201         }
4202         clear_bit(In_sync, &rdev->flags);
4203         rdev->desc_nr = -1;
4204         rdev->saved_raid_disk = -1;
4205         err = bind_rdev_to_array(rdev, mddev);
4206         if (err)
4207                 goto abort_export;
4208
4209         /*
4210          * The rest should better be atomic, we can have disk failures
4211          * noticed in interrupt contexts ...
4212          */
4213
4214         if (rdev->desc_nr == mddev->max_disks) {
4215                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4216                         mdname(mddev));
4217                 err = -EBUSY;
4218                 goto abort_unbind_export;
4219         }
4220
4221         rdev->raid_disk = -1;
4222
4223         md_update_sb(mddev, 1);
4224
4225         /*
4226          * Kick recovery, maybe this spare has to be added to the
4227          * array immediately.
4228          */
4229         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4230         md_wakeup_thread(mddev->thread);
4231         md_new_event(mddev);
4232         return 0;
4233
4234 abort_unbind_export:
4235         unbind_rdev_from_array(rdev);
4236
4237 abort_export:
4238         export_rdev(rdev);
4239         return err;
4240 }
4241
4242 static int set_bitmap_file(mddev_t *mddev, int fd)
4243 {
4244         int err;
4245
4246         if (mddev->pers) {
4247                 if (!mddev->pers->quiesce)
4248                         return -EBUSY;
4249                 if (mddev->recovery || mddev->sync_thread)
4250                         return -EBUSY;
4251                 /* we should be able to change the bitmap.. */
4252         }
4253
4254
4255         if (fd >= 0) {
4256                 if (mddev->bitmap)
4257                         return -EEXIST; /* cannot add when bitmap is present */
4258                 mddev->bitmap_file = fget(fd);
4259
4260                 if (mddev->bitmap_file == NULL) {
4261                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4262                                mdname(mddev));
4263                         return -EBADF;
4264                 }
4265
4266                 err = deny_bitmap_write_access(mddev->bitmap_file);
4267                 if (err) {
4268                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4269                                mdname(mddev));
4270                         fput(mddev->bitmap_file);
4271                         mddev->bitmap_file = NULL;
4272                         return err;
4273                 }
4274                 mddev->bitmap_offset = 0; /* file overrides offset */
4275         } else if (mddev->bitmap == NULL)
4276                 return -ENOENT; /* cannot remove what isn't there */
4277         err = 0;
4278         if (mddev->pers) {
4279                 mddev->pers->quiesce(mddev, 1);
4280                 if (fd >= 0)
4281                         err = bitmap_create(mddev);
4282                 if (fd < 0 || err) {
4283                         bitmap_destroy(mddev);
4284                         fd = -1; /* make sure to put the file */
4285                 }
4286                 mddev->pers->quiesce(mddev, 0);
4287         }
4288         if (fd < 0) {
4289                 if (mddev->bitmap_file) {
4290                         restore_bitmap_write_access(mddev->bitmap_file);
4291                         fput(mddev->bitmap_file);
4292                 }
4293                 mddev->bitmap_file = NULL;
4294         }
4295
4296         return err;
4297 }
4298
4299 /*
4300  * set_array_info is used two different ways
4301  * The original usage is when creating a new array.
4302  * In this usage, raid_disks is > 0 and it together with
4303  *  level, size, not_persistent,layout,chunksize determine the
4304  *  shape of the array.
4305  *  This will always create an array with a type-0.90.0 superblock.
4306  * The newer usage is when assembling an array.
4307  *  In this case raid_disks will be 0, and the major_version field is
4308  *  use to determine which style super-blocks are to be found on the devices.
4309  *  The minor and patch _version numbers are also kept incase the
4310  *  super_block handler wishes to interpret them.
4311  */
4312 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4313 {
4314
4315         if (info->raid_disks == 0) {
4316                 /* just setting version number for superblock loading */
4317                 if (info->major_version < 0 ||
4318                     info->major_version >= ARRAY_SIZE(super_types) ||
4319                     super_types[info->major_version].name == NULL) {
4320                         /* maybe try to auto-load a module? */
4321                         printk(KERN_INFO
4322                                 "md: superblock version %d not known\n",
4323                                 info->major_version);
4324                         return -EINVAL;
4325                 }
4326                 mddev->major_version = info->major_version;
4327                 mddev->minor_version = info->minor_version;
4328                 mddev->patch_version = info->patch_version;
4329                 mddev->persistent = !info->not_persistent;
4330                 return 0;
4331         }
4332         mddev->major_version = MD_MAJOR_VERSION;
4333         mddev->minor_version = MD_MINOR_VERSION;
4334         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4335         mddev->ctime         = get_seconds();
4336
4337         mddev->level         = info->level;
4338         mddev->clevel[0]     = 0;
4339         mddev->size          = info->size;
4340         mddev->raid_disks    = info->raid_disks;
4341         /* don't set md_minor, it is determined by which /dev/md* was
4342          * openned
4343          */
4344         if (info->state & (1<<MD_SB_CLEAN))
4345                 mddev->recovery_cp = MaxSector;
4346         else
4347                 mddev->recovery_cp = 0;
4348         mddev->persistent    = ! info->not_persistent;
4349         mddev->external      = 0;
4350
4351         mddev->layout        = info->layout;
4352         mddev->chunk_size    = info->chunk_size;
4353
4354         mddev->max_disks     = MD_SB_DISKS;
4355
4356         if (mddev->persistent)
4357                 mddev->flags         = 0;
4358         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4359
4360         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4361         mddev->bitmap_offset = 0;
4362
4363         mddev->reshape_position = MaxSector;
4364
4365         /*
4366          * Generate a 128 bit UUID
4367          */
4368         get_random_bytes(mddev->uuid, 16);
4369
4370         mddev->new_level = mddev->level;
4371         mddev->new_chunk = mddev->chunk_size;
4372         mddev->new_layout = mddev->layout;
4373         mddev->delta_disks = 0;
4374
4375         return 0;
4376 }
4377
4378 static int update_size(mddev_t *mddev, unsigned long size)
4379 {
4380         mdk_rdev_t * rdev;
4381         int rv;
4382         struct list_head *tmp;
4383         int fit = (size == 0);
4384
4385         if (mddev->pers->resize == NULL)
4386                 return -EINVAL;
4387         /* The "size" is the amount of each device that is used.
4388          * This can only make sense for arrays with redundancy.
4389          * linear and raid0 always use whatever space is available
4390          * We can only consider changing the size if no resync
4391          * or reconstruction is happening, and if the new size
4392          * is acceptable. It must fit before the sb_offset or,
4393          * if that is <data_offset, it must fit before the
4394          * size of each device.
4395          * If size is zero, we find the largest size that fits.
4396          */
4397         if (mddev->sync_thread)
4398                 return -EBUSY;
4399         rdev_for_each(rdev, tmp, mddev) {
4400                 sector_t avail;
4401                 avail = rdev->size * 2;
4402
4403                 if (fit && (size == 0 || size > avail/2))
4404                         size = avail/2;
4405                 if (avail < ((sector_t)size << 1))
4406                         return -ENOSPC;
4407         }
4408         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4409         if (!rv) {
4410                 struct block_device *bdev;
4411
4412                 bdev = bdget_disk(mddev->gendisk, 0);
4413                 if (bdev) {
4414                         mutex_lock(&bdev->bd_inode->i_mutex);
4415                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4416                         mutex_unlock(&bdev->bd_inode->i_mutex);
4417                         bdput(bdev);
4418                 }
4419         }
4420         return rv;
4421 }
4422
4423 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4424 {
4425         int rv;
4426         /* change the number of raid disks */
4427         if (mddev->pers->check_reshape == NULL)
4428                 return -EINVAL;
4429         if (raid_disks <= 0 ||
4430             raid_disks >= mddev->max_disks)
4431                 return -EINVAL;
4432         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4433                 return -EBUSY;
4434         mddev->delta_disks = raid_disks - mddev->raid_disks;
4435
4436         rv = mddev->pers->check_reshape(mddev);
4437         return rv;
4438 }
4439
4440
4441 /*
4442  * update_array_info is used to change the configuration of an
4443  * on-line array.
4444  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4445  * fields in the info are checked against the array.
4446  * Any differences that cannot be handled will cause an error.
4447  * Normally, only one change can be managed at a time.
4448  */
4449 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4450 {
4451         int rv = 0;
4452         int cnt = 0;
4453         int state = 0;
4454
4455         /* calculate expected state,ignoring low bits */
4456         if (mddev->bitmap && mddev->bitmap_offset)
4457                 state |= (1 << MD_SB_BITMAP_PRESENT);
4458
4459         if (mddev->major_version != info->major_version ||
4460             mddev->minor_version != info->minor_version ||
4461 /*          mddev->patch_version != info->patch_version || */
4462             mddev->ctime         != info->ctime         ||
4463             mddev->level         != info->level         ||
4464 /*          mddev->layout        != info->layout        || */
4465             !mddev->persistent   != info->not_persistent||
4466             mddev->chunk_size    != info->chunk_size    ||
4467             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4468             ((state^info->state) & 0xfffffe00)
4469                 )
4470                 return -EINVAL;
4471         /* Check there is only one change */
4472         if (info->size >= 0 && mddev->size != info->size) cnt++;
4473         if (mddev->raid_disks != info->raid_disks) cnt++;
4474         if (mddev->layout != info->layout) cnt++;
4475         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4476         if (cnt == 0) return 0;
4477         if (cnt > 1) return -EINVAL;
4478
4479         if (mddev->layout != info->layout) {
4480                 /* Change layout
4481                  * we don't need to do anything at the md level, the
4482                  * personality will take care of it all.
4483                  */
4484                 if (mddev->pers->reconfig == NULL)
4485                         return -EINVAL;
4486                 else
4487                         return mddev->pers->reconfig(mddev, info->layout, -1);
4488         }
4489         if (info->size >= 0 && mddev->size != info->size)
4490                 rv = update_size(mddev, info->size);
4491
4492         if (mddev->raid_disks    != info->raid_disks)
4493                 rv = update_raid_disks(mddev, info->raid_disks);
4494
4495         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4496                 if (mddev->pers->quiesce == NULL)
4497                         return -EINVAL;
4498                 if (mddev->recovery || mddev->sync_thread)
4499                         return -EBUSY;
4500                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4501                         /* add the bitmap */
4502                         if (mddev->bitmap)
4503                                 return -EEXIST;
4504                         if (mddev->default_bitmap_offset == 0)
4505                                 return -EINVAL;
4506                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4507                         mddev->pers->quiesce(mddev, 1);
4508                         rv = bitmap_create(mddev);
4509                         if (rv)
4510                                 bitmap_destroy(mddev);
4511                         mddev->pers->quiesce(mddev, 0);
4512                 } else {
4513                         /* remove the bitmap */
4514                         if (!mddev->bitmap)
4515                                 return -ENOENT;
4516                         if (mddev->bitmap->file)
4517                                 return -EINVAL;
4518                         mddev->pers->quiesce(mddev, 1);
4519                         bitmap_destroy(mddev);
4520                         mddev->pers->quiesce(mddev, 0);
4521                         mddev->bitmap_offset = 0;
4522                 }
4523         }
4524         md_update_sb(mddev, 1);
4525         return rv;
4526 }
4527
4528 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4529 {
4530         mdk_rdev_t *rdev;
4531
4532         if (mddev->pers == NULL)
4533                 return -ENODEV;
4534
4535         rdev = find_rdev(mddev, dev);
4536         if (!rdev)
4537                 return -ENODEV;
4538
4539         md_error(mddev, rdev);
4540         return 0;
4541 }
4542
4543 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4544 {
4545         mddev_t *mddev = bdev->bd_disk->private_data;
4546
4547         geo->heads = 2;
4548         geo->sectors = 4;
4549         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4550         return 0;
4551 }
4552
4553 static int md_ioctl(struct inode *inode, struct file *file,
4554                         unsigned int cmd, unsigned long arg)
4555 {
4556         int err = 0;
4557         void __user *argp = (void __user *)arg;
4558         mddev_t *mddev = NULL;
4559
4560         if (!capable(CAP_SYS_ADMIN))
4561                 return -EACCES;
4562
4563         /*
4564          * Commands dealing with the RAID driver but not any
4565          * particular array:
4566          */
4567         switch (cmd)
4568         {
4569                 case RAID_VERSION:
4570                         err = get_version(argp);
4571                         goto done;
4572
4573                 case PRINT_RAID_DEBUG:
4574                         err = 0;
4575                         md_print_devices();
4576                         goto done;
4577
4578 #ifndef MODULE
4579                 case RAID_AUTORUN:
4580                         err = 0;
4581                         autostart_arrays(arg);
4582                         goto done;
4583 #endif
4584                 default:;
4585         }
4586
4587         /*
4588          * Commands creating/starting a new array:
4589          */
4590
4591         mddev = inode->i_bdev->bd_disk->private_data;
4592
4593         if (!mddev) {
4594                 BUG();
4595                 goto abort;
4596         }
4597
4598         err = mddev_lock(mddev);
4599         if (err) {
4600                 printk(KERN_INFO
4601                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4602                         err, cmd);
4603                 goto abort;
4604         }
4605
4606         switch (cmd)
4607         {
4608                 case SET_ARRAY_INFO:
4609                         {
4610                                 mdu_array_info_t info;
4611                                 if (!arg)
4612                                         memset(&info, 0, sizeof(info));
4613                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4614                                         err = -EFAULT;
4615                                         goto abort_unlock;
4616                                 }
4617                                 if (mddev->pers) {
4618                                         err = update_array_info(mddev, &info);
4619                                         if (err) {
4620                                                 printk(KERN_WARNING "md: couldn't update"
4621                                                        " array info. %d\n", err);
4622                                                 goto abort_unlock;
4623                                         }
4624                                         goto done_unlock;
4625                                 }
4626                                 if (!list_empty(&mddev->disks)) {
4627                                         printk(KERN_WARNING
4628                                                "md: array %s already has disks!\n",
4629                                                mdname(mddev));
4630                                         err = -EBUSY;
4631                                         goto abort_unlock;
4632                                 }
4633                                 if (mddev->raid_disks) {
4634                                         printk(KERN_WARNING
4635                                                "md: array %s already initialised!\n",
4636                                                mdname(mddev));
4637                                         err = -EBUSY;
4638                                         goto abort_unlock;
4639                                 }
4640                                 err = set_array_info(mddev, &info);
4641                                 if (err) {
4642                                         printk(KERN_WARNING "md: couldn't set"
4643                                                " array info. %d\n", err);
4644                                         goto abort_unlock;
4645                                 }
4646                         }
4647                         goto done_unlock;
4648
4649                 default:;
4650         }
4651
4652         /*
4653          * Commands querying/configuring an existing array:
4654          */
4655         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4656          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4657         if ((!mddev->raid_disks && !mddev->external)
4658             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4659             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4660             && cmd != GET_BITMAP_FILE) {
4661                 err = -ENODEV;
4662                 goto abort_unlock;
4663         }
4664
4665         /*
4666          * Commands even a read-only array can execute:
4667          */
4668         switch (cmd)
4669         {
4670                 case GET_ARRAY_INFO:
4671                         err = get_array_info(mddev, argp);
4672                         goto done_unlock;
4673
4674                 case GET_BITMAP_FILE:
4675                         err = get_bitmap_file(mddev, argp);
4676                         goto done_unlock;
4677
4678                 case GET_DISK_INFO:
4679                         err = get_disk_info(mddev, argp);
4680                         goto done_unlock;
4681
4682                 case RESTART_ARRAY_RW:
4683                         err = restart_array(mddev);
4684                         goto done_unlock;
4685
4686                 case STOP_ARRAY:
4687                         err = do_md_stop (mddev, 0);
4688                         goto done_unlock;
4689
4690                 case STOP_ARRAY_RO:
4691                         err = do_md_stop (mddev, 1);
4692                         goto done_unlock;
4693
4694         /*
4695          * We have a problem here : there is no easy way to give a CHS
4696          * virtual geometry. We currently pretend that we have a 2 heads
4697          * 4 sectors (with a BIG number of cylinders...). This drives
4698          * dosfs just mad... ;-)
4699          */
4700         }
4701
4702         /*
4703          * The remaining ioctls are changing the state of the
4704          * superblock, so we do not allow them on read-only arrays.
4705          * However non-MD ioctls (e.g. get-size) will still come through
4706          * here and hit the 'default' below, so only disallow
4707          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4708          */
4709         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4710             mddev->ro && mddev->pers) {
4711                 if (mddev->ro == 2) {
4712                         mddev->ro = 0;
4713                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4714                 md_wakeup_thread(mddev->thread);
4715
4716                 } else {
4717                         err = -EROFS;
4718                         goto abort_unlock;
4719                 }
4720         }
4721
4722         switch (cmd)
4723         {
4724                 case ADD_NEW_DISK:
4725                 {
4726                         mdu_disk_info_t info;
4727                         if (copy_from_user(&info, argp, sizeof(info)))
4728                                 err = -EFAULT;
4729                         else
4730                                 err = add_new_disk(mddev, &info);
4731                         goto done_unlock;
4732                 }
4733
4734                 case HOT_REMOVE_DISK:
4735                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4736                         goto done_unlock;
4737
4738                 case HOT_ADD_DISK:
4739                         err = hot_add_disk(mddev, new_decode_dev(arg));
4740                         goto done_unlock;
4741
4742                 case SET_DISK_FAULTY:
4743                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4744                         goto done_unlock;
4745
4746                 case RUN_ARRAY:
4747                         err = do_md_run (mddev);
4748                         goto done_unlock;
4749
4750                 case SET_BITMAP_FILE:
4751                         err = set_bitmap_file(mddev, (int)arg);
4752                         goto done_unlock;
4753
4754                 default:
4755                         err = -EINVAL;
4756                         goto abort_unlock;
4757         }
4758
4759 done_unlock:
4760 abort_unlock:
4761         mddev_unlock(mddev);
4762
4763         return err;
4764 done:
4765         if (err)
4766                 MD_BUG();
4767 abort:
4768         return err;
4769 }
4770
4771 static int md_open(struct inode *inode, struct file *file)
4772 {
4773         /*
4774          * Succeed if we can lock the mddev, which confirms that
4775          * it isn't being stopped right now.
4776          */
4777         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4778         int err;
4779
4780         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4781                 goto out;
4782
4783         err = 0;
4784         mddev_get(mddev);
4785         mddev_unlock(mddev);
4786
4787         check_disk_change(inode->i_bdev);
4788  out:
4789         return err;
4790 }
4791
4792 static int md_release(struct inode *inode, struct file * file)
4793 {
4794         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4795
4796         BUG_ON(!mddev);
4797         mddev_put(mddev);
4798
4799         return 0;
4800 }
4801
4802 static int md_media_changed(struct gendisk *disk)
4803 {
4804         mddev_t *mddev = disk->private_data;
4805
4806         return mddev->changed;
4807 }
4808
4809 static int md_revalidate(struct gendisk *disk)
4810 {
4811         mddev_t *mddev = disk->private_data;
4812
4813         mddev->changed = 0;
4814         return 0;
4815 }
4816 static struct block_device_operations md_fops =
4817 {
4818         .owner          = THIS_MODULE,
4819         .open           = md_open,
4820         .release        = md_release,
4821         .ioctl          = md_ioctl,
4822         .getgeo         = md_getgeo,
4823         .media_changed  = md_media_changed,
4824         .revalidate_disk= md_revalidate,
4825 };
4826
4827 static int md_thread(void * arg)
4828 {
4829         mdk_thread_t *thread = arg;
4830
4831         /*
4832          * md_thread is a 'system-thread', it's priority should be very
4833          * high. We avoid resource deadlocks individually in each
4834          * raid personality. (RAID5 does preallocation) We also use RR and
4835          * the very same RT priority as kswapd, thus we will never get
4836          * into a priority inversion deadlock.
4837          *
4838          * we definitely have to have equal or higher priority than
4839          * bdflush, otherwise bdflush will deadlock if there are too
4840          * many dirty RAID5 blocks.
4841          */
4842
4843         allow_signal(SIGKILL);
4844         while (!kthread_should_stop()) {
4845
4846                 /* We need to wait INTERRUPTIBLE so that
4847                  * we don't add to the load-average.
4848                  * That means we need to be sure no signals are
4849                  * pending
4850                  */
4851                 if (signal_pending(current))
4852                         flush_signals(current);
4853
4854                 wait_event_interruptible_timeout
4855                         (thread->wqueue,
4856                          test_bit(THREAD_WAKEUP, &thread->flags)
4857                          || kthread_should_stop(),
4858                          thread->timeout);
4859
4860                 clear_bit(THREAD_WAKEUP, &thread->flags);
4861
4862                 thread->run(thread->mddev);
4863         }
4864
4865         return 0;
4866 }
4867
4868 void md_wakeup_thread(mdk_thread_t *thread)
4869 {
4870         if (thread) {
4871                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4872                 set_bit(THREAD_WAKEUP, &thread->flags);
4873                 wake_up(&thread->wqueue);
4874         }
4875 }
4876
4877 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4878                                  const char *name)
4879 {
4880         mdk_thread_t *thread;
4881
4882         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4883         if (!thread)
4884                 return NULL;
4885
4886         init_waitqueue_head(&thread->wqueue);
4887
4888         thread->run = run;
4889         thread->mddev = mddev;
4890         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4891         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4892         if (IS_ERR(thread->tsk)) {
4893                 kfree(thread);
4894                 return NULL;
4895         }
4896         return thread;
4897 }
4898
4899 void md_unregister_thread(mdk_thread_t *thread)
4900 {
4901         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4902
4903         kthread_stop(thread->tsk);
4904         kfree(thread);
4905 }
4906
4907 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4908 {
4909         if (!mddev) {
4910                 MD_BUG();
4911                 return;
4912         }
4913
4914         if (!rdev || test_bit(Faulty, &rdev->flags))
4915                 return;
4916 /*
4917         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4918                 mdname(mddev),
4919                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4920                 __builtin_return_address(0),__builtin_return_address(1),
4921                 __builtin_return_address(2),__builtin_return_address(3));
4922 */
4923         if (!mddev->pers)
4924                 return;
4925         if (!mddev->pers->error_handler)
4926                 return;
4927         mddev->pers->error_handler(mddev,rdev);
4928         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4929         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4930         md_wakeup_thread(mddev->thread);
4931         md_new_event_inintr(mddev);
4932 }
4933
4934 /* seq_file implementation /proc/mdstat */
4935
4936 static void status_unused(struct seq_file *seq)
4937 {
4938         int i = 0;
4939         mdk_rdev_t *rdev;
4940         struct list_head *tmp;
4941
4942         seq_printf(seq, "unused devices: ");
4943
4944         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
4945                 char b[BDEVNAME_SIZE];
4946                 i++;
4947                 seq_printf(seq, "%s ",
4948                               bdevname(rdev->bdev,b));
4949         }
4950         if (!i)
4951                 seq_printf(seq, "<none>");
4952
4953         seq_printf(seq, "\n");
4954 }
4955
4956
4957 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4958 {
4959         sector_t max_blocks, resync, res;
4960         unsigned long dt, db, rt;
4961         int scale;
4962         unsigned int per_milli;
4963
4964         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4965
4966         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4967                 max_blocks = mddev->resync_max_sectors >> 1;
4968         else
4969                 max_blocks = mddev->size;
4970
4971         /*
4972          * Should not happen.
4973          */
4974         if (!max_blocks) {
4975                 MD_BUG();
4976                 return;
4977         }
4978         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4979          * in a sector_t, and (max_blocks>>scale) will fit in a
4980          * u32, as those are the requirements for sector_div.
4981          * Thus 'scale' must be at least 10
4982          */
4983         scale = 10;
4984         if (sizeof(sector_t) > sizeof(unsigned long)) {
4985                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4986                         scale++;
4987         }
4988         res = (resync>>scale)*1000;
4989         sector_div(res, (u32)((max_blocks>>scale)+1));
4990
4991         per_milli = res;
4992         {
4993                 int i, x = per_milli/50, y = 20-x;
4994                 seq_printf(seq, "[");
4995                 for (i = 0; i < x; i++)
4996                         seq_printf(seq, "=");
4997                 seq_printf(seq, ">");
4998                 for (i = 0; i < y; i++)
4999                         seq_printf(seq, ".");
5000                 seq_printf(seq, "] ");
5001         }
5002         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5003                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5004                     "reshape" :
5005                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5006                      "check" :
5007                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5008                       "resync" : "recovery"))),
5009                    per_milli/10, per_milli % 10,
5010                    (unsigned long long) resync,
5011                    (unsigned long long) max_blocks);
5012
5013         /*
5014          * We do not want to overflow, so the order of operands and
5015          * the * 100 / 100 trick are important. We do a +1 to be
5016          * safe against division by zero. We only estimate anyway.
5017          *
5018          * dt: time from mark until now
5019          * db: blocks written from mark until now
5020          * rt: remaining time
5021          */
5022         dt = ((jiffies - mddev->resync_mark) / HZ);
5023         if (!dt) dt++;
5024         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5025                 - mddev->resync_mark_cnt;
5026         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5027
5028         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5029
5030         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5031 }
5032
5033 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5034 {
5035         struct list_head *tmp;
5036         loff_t l = *pos;
5037         mddev_t *mddev;
5038
5039         if (l >= 0x10000)
5040                 return NULL;
5041         if (!l--)
5042                 /* header */
5043                 return (void*)1;
5044
5045         spin_lock(&all_mddevs_lock);
5046         list_for_each(tmp,&all_mddevs)
5047                 if (!l--) {
5048                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5049                         mddev_get(mddev);
5050                         spin_unlock(&all_mddevs_lock);
5051                         return mddev;
5052                 }
5053         spin_unlock(&all_mddevs_lock);
5054         if (!l--)
5055                 return (void*)2;/* tail */
5056         return NULL;
5057 }
5058
5059 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5060 {
5061         struct list_head *tmp;
5062         mddev_t *next_mddev, *mddev = v;
5063
5064         ++*pos;
5065         if (v == (void*)2)
5066                 return NULL;
5067
5068         spin_lock(&all_mddevs_lock);
5069         if (v == (void*)1)
5070                 tmp = all_mddevs.next;
5071         else
5072                 tmp = mddev->all_mddevs.next;
5073         if (tmp != &all_mddevs)
5074                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5075         else {
5076                 next_mddev = (void*)2;
5077                 *pos = 0x10000;
5078         }
5079         spin_unlock(&all_mddevs_lock);
5080
5081         if (v != (void*)1)
5082                 mddev_put(mddev);
5083         return next_mddev;
5084
5085 }
5086
5087 static void md_seq_stop(struct seq_file *seq, void *v)
5088 {
5089         mddev_t *mddev = v;
5090
5091         if (mddev && v != (void*)1 && v != (void*)2)
5092                 mddev_put(mddev);
5093 }
5094
5095 struct mdstat_info {
5096         int event;
5097 };
5098
5099 static int md_seq_show(struct seq_file *seq, void *v)
5100 {
5101         mddev_t *mddev = v;
5102         sector_t size;
5103         struct list_head *tmp2;
5104         mdk_rdev_t *rdev;
5105         struct mdstat_info *mi = seq->private;
5106         struct bitmap *bitmap;
5107
5108         if (v == (void*)1) {
5109                 struct mdk_personality *pers;
5110                 seq_printf(seq, "Personalities : ");
5111                 spin_lock(&pers_lock);
5112                 list_for_each_entry(pers, &pers_list, list)
5113                         seq_printf(seq, "[%s] ", pers->name);
5114
5115                 spin_unlock(&pers_lock);
5116                 seq_printf(seq, "\n");
5117                 mi->event = atomic_read(&md_event_count);
5118                 return 0;
5119         }
5120         if (v == (void*)2) {
5121                 status_unused(seq);
5122                 return 0;
5123         }
5124
5125         if (mddev_lock(mddev) < 0)
5126                 return -EINTR;
5127
5128         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5129                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5130                                                 mddev->pers ? "" : "in");
5131                 if (mddev->pers) {
5132                         if (mddev->ro==1)
5133                                 seq_printf(seq, " (read-only)");
5134                         if (mddev->ro==2)
5135                                 seq_printf(seq, "(auto-read-only)");
5136                         seq_printf(seq, " %s", mddev->pers->name);
5137                 }
5138
5139                 size = 0;
5140                 rdev_for_each(rdev, tmp2, mddev) {
5141                         char b[BDEVNAME_SIZE];
5142                         seq_printf(seq, " %s[%d]",
5143                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5144                         if (test_bit(WriteMostly, &rdev->flags))
5145                                 seq_printf(seq, "(W)");
5146                         if (test_bit(Faulty, &rdev->flags)) {
5147                                 seq_printf(seq, "(F)");
5148                                 continue;
5149                         } else if (rdev->raid_disk < 0)
5150                                 seq_printf(seq, "(S)"); /* spare */
5151                         size += rdev->size;
5152                 }
5153
5154                 if (!list_empty(&mddev->disks)) {
5155                         if (mddev->pers)
5156                                 seq_printf(seq, "\n      %llu blocks",
5157                                         (unsigned long long)mddev->array_size);
5158                         else
5159                                 seq_printf(seq, "\n      %llu blocks",
5160                                         (unsigned long long)size);
5161                 }
5162                 if (mddev->persistent) {
5163                         if (mddev->major_version != 0 ||
5164                             mddev->minor_version != 90) {
5165                                 seq_printf(seq," super %d.%d",
5166                                            mddev->major_version,
5167                                            mddev->minor_version);
5168                         }
5169                 } else if (mddev->external)
5170                         seq_printf(seq, " super external:%s",
5171                                    mddev->metadata_type);
5172                 else
5173                         seq_printf(seq, " super non-persistent");
5174
5175                 if (mddev->pers) {
5176                         mddev->pers->status (seq, mddev);
5177                         seq_printf(seq, "\n      ");
5178                         if (mddev->pers->sync_request) {
5179                                 if (mddev->curr_resync > 2) {
5180                                         status_resync (seq, mddev);
5181                                         seq_printf(seq, "\n      ");
5182                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5183                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5184                                 else if (mddev->recovery_cp < MaxSector)
5185                                         seq_printf(seq, "\tresync=PENDING\n      ");
5186                         }
5187                 } else
5188                         seq_printf(seq, "\n       ");
5189
5190                 if ((bitmap = mddev->bitmap)) {
5191                         unsigned long chunk_kb;
5192                         unsigned long flags;
5193                         spin_lock_irqsave(&bitmap->lock, flags);
5194                         chunk_kb = bitmap->chunksize >> 10;
5195                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5196                                 "%lu%s chunk",
5197                                 bitmap->pages - bitmap->missing_pages,
5198                                 bitmap->pages,
5199                                 (bitmap->pages - bitmap->missing_pages)
5200                                         << (PAGE_SHIFT - 10),
5201                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5202                                 chunk_kb ? "KB" : "B");
5203                         if (bitmap->file) {
5204                                 seq_printf(seq, ", file: ");
5205                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5206                         }
5207
5208                         seq_printf(seq, "\n");
5209                         spin_unlock_irqrestore(&bitmap->lock, flags);
5210                 }
5211
5212                 seq_printf(seq, "\n");
5213         }
5214         mddev_unlock(mddev);
5215
5216         return 0;
5217 }
5218
5219 static struct seq_operations md_seq_ops = {
5220         .start  = md_seq_start,
5221         .next   = md_seq_next,
5222         .stop   = md_seq_stop,
5223         .show   = md_seq_show,
5224 };
5225
5226 static int md_seq_open(struct inode *inode, struct file *file)
5227 {
5228         int error;
5229         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5230         if (mi == NULL)
5231                 return -ENOMEM;
5232
5233         error = seq_open(file, &md_seq_ops);
5234         if (error)
5235                 kfree(mi);
5236         else {
5237                 struct seq_file *p = file->private_data;
5238                 p->private = mi;
5239                 mi->event = atomic_read(&md_event_count);
5240         }
5241         return error;
5242 }
5243
5244 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5245 {
5246         struct seq_file *m = filp->private_data;
5247         struct mdstat_info *mi = m->private;
5248         int mask;
5249
5250         poll_wait(filp, &md_event_waiters, wait);
5251
5252         /* always allow read */
5253         mask = POLLIN | POLLRDNORM;
5254
5255         if (mi->event != atomic_read(&md_event_count))
5256                 mask |= POLLERR | POLLPRI;
5257         return mask;
5258 }
5259
5260 static const struct file_operations md_seq_fops = {
5261         .owner          = THIS_MODULE,
5262         .open           = md_seq_open,
5263         .read           = seq_read,
5264         .llseek         = seq_lseek,
5265         .release        = seq_release_private,
5266         .poll           = mdstat_poll,
5267 };
5268
5269 int register_md_personality(struct mdk_personality *p)
5270 {
5271         spin_lock(&pers_lock);
5272         list_add_tail(&p->list, &pers_list);
5273         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5274         spin_unlock(&pers_lock);
5275         return 0;
5276 }
5277
5278 int unregister_md_personality(struct mdk_personality *p)
5279 {
5280         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5281         spin_lock(&pers_lock);
5282         list_del_init(&p->list);
5283         spin_unlock(&pers_lock);
5284         return 0;
5285 }
5286
5287 static int is_mddev_idle(mddev_t *mddev)
5288 {
5289         mdk_rdev_t * rdev;
5290         struct list_head *tmp;
5291         int idle;
5292         long curr_events;
5293
5294         idle = 1;
5295         rdev_for_each(rdev, tmp, mddev) {
5296                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5297                 curr_events = disk_stat_read(disk, sectors[0]) +
5298                                 disk_stat_read(disk, sectors[1]) -
5299                                 atomic_read(&disk->sync_io);
5300                 /* sync IO will cause sync_io to increase before the disk_stats
5301                  * as sync_io is counted when a request starts, and
5302                  * disk_stats is counted when it completes.
5303                  * So resync activity will cause curr_events to be smaller than
5304                  * when there was no such activity.
5305                  * non-sync IO will cause disk_stat to increase without
5306                  * increasing sync_io so curr_events will (eventually)
5307                  * be larger than it was before.  Once it becomes
5308                  * substantially larger, the test below will cause
5309                  * the array to appear non-idle, and resync will slow
5310                  * down.
5311                  * If there is a lot of outstanding resync activity when
5312                  * we set last_event to curr_events, then all that activity
5313                  * completing might cause the array to appear non-idle
5314                  * and resync will be slowed down even though there might
5315                  * not have been non-resync activity.  This will only
5316                  * happen once though.  'last_events' will soon reflect
5317                  * the state where there is little or no outstanding
5318                  * resync requests, and further resync activity will
5319                  * always make curr_events less than last_events.
5320                  *
5321                  */
5322                 if (curr_events - rdev->last_events > 4096) {
5323                         rdev->last_events = curr_events;
5324                         idle = 0;
5325                 }
5326         }
5327         return idle;
5328 }
5329
5330 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5331 {
5332         /* another "blocks" (512byte) blocks have been synced */
5333         atomic_sub(blocks, &mddev->recovery_active);
5334         wake_up(&mddev->recovery_wait);
5335         if (!ok) {
5336                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5337                 md_wakeup_thread(mddev->thread);
5338                 // stop recovery, signal do_sync ....
5339         }
5340 }
5341
5342
5343 /* md_write_start(mddev, bi)
5344  * If we need to update some array metadata (e.g. 'active' flag
5345  * in superblock) before writing, schedule a superblock update
5346  * and wait for it to complete.
5347  */
5348 void md_write_start(mddev_t *mddev, struct bio *bi)
5349 {
5350         if (bio_data_dir(bi) != WRITE)
5351                 return;
5352
5353         BUG_ON(mddev->ro == 1);
5354         if (mddev->ro == 2) {
5355                 /* need to switch to read/write */
5356                 mddev->ro = 0;
5357                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5358                 md_wakeup_thread(mddev->thread);
5359         }
5360         atomic_inc(&mddev->writes_pending);
5361         if (mddev->in_sync) {
5362                 spin_lock_irq(&mddev->write_lock);
5363                 if (mddev->in_sync) {
5364                         mddev->in_sync = 0;
5365                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5366                         md_wakeup_thread(mddev->thread);
5367                 }
5368                 spin_unlock_irq(&mddev->write_lock);
5369         }
5370         wait_event(mddev->sb_wait, mddev->flags==0);
5371 }
5372
5373 void md_write_end(mddev_t *mddev)
5374 {
5375         if (atomic_dec_and_test(&mddev->writes_pending)) {
5376                 if (mddev->safemode == 2)
5377                         md_wakeup_thread(mddev->thread);
5378                 else if (mddev->safemode_delay)
5379                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5380         }
5381 }
5382
5383 /* md_allow_write(mddev)
5384  * Calling this ensures that the array is marked 'active' so that writes
5385  * may proceed without blocking.  It is important to call this before
5386  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5387  * Must be called with mddev_lock held.
5388  */
5389 void md_allow_write(mddev_t *mddev)
5390 {
5391         if (!mddev->pers)
5392                 return;
5393         if (mddev->ro)
5394                 return;
5395
5396         spin_lock_irq(&mddev->write_lock);
5397         if (mddev->in_sync) {
5398                 mddev->in_sync = 0;
5399                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5400                 if (mddev->safemode_delay &&
5401                     mddev->safemode == 0)
5402                         mddev->safemode = 1;
5403                 spin_unlock_irq(&mddev->write_lock);
5404                 md_update_sb(mddev, 0);
5405         } else
5406                 spin_unlock_irq(&mddev->write_lock);
5407 }
5408 EXPORT_SYMBOL_GPL(md_allow_write);
5409
5410 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5411
5412 #define SYNC_MARKS      10
5413 #define SYNC_MARK_STEP  (3*HZ)
5414 void md_do_sync(mddev_t *mddev)
5415 {
5416         mddev_t *mddev2;
5417         unsigned int currspeed = 0,
5418                  window;
5419         sector_t max_sectors,j, io_sectors;
5420         unsigned long mark[SYNC_MARKS];
5421         sector_t mark_cnt[SYNC_MARKS];
5422         int last_mark,m;
5423         struct list_head *tmp;
5424         sector_t last_check;
5425         int skipped = 0;
5426         struct list_head *rtmp;
5427         mdk_rdev_t *rdev;
5428         char *desc;
5429
5430         /* just incase thread restarts... */
5431         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5432                 return;
5433         if (mddev->ro) /* never try to sync a read-only array */
5434                 return;
5435
5436         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5437                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5438                         desc = "data-check";
5439                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5440                         desc = "requested-resync";
5441                 else
5442                         desc = "resync";
5443         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5444                 desc = "reshape";
5445         else
5446                 desc = "recovery";
5447
5448         /* we overload curr_resync somewhat here.
5449          * 0 == not engaged in resync at all
5450          * 2 == checking that there is no conflict with another sync
5451          * 1 == like 2, but have yielded to allow conflicting resync to
5452          *              commense
5453          * other == active in resync - this many blocks
5454          *
5455          * Before starting a resync we must have set curr_resync to
5456          * 2, and then checked that every "conflicting" array has curr_resync
5457          * less than ours.  When we find one that is the same or higher
5458          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5459          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5460          * This will mean we have to start checking from the beginning again.
5461          *
5462          */
5463
5464         do {
5465                 mddev->curr_resync = 2;
5466
5467         try_again:
5468                 if (kthread_should_stop()) {
5469                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5470                         goto skip;
5471                 }
5472                 for_each_mddev(mddev2, tmp) {
5473                         if (mddev2 == mddev)
5474                                 continue;
5475                         if (mddev2->curr_resync &&
5476                             match_mddev_units(mddev,mddev2)) {
5477                                 DEFINE_WAIT(wq);
5478                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5479                                         /* arbitrarily yield */
5480                                         mddev->curr_resync = 1;
5481                                         wake_up(&resync_wait);
5482                                 }
5483                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5484                                         /* no need to wait here, we can wait the next
5485                                          * time 'round when curr_resync == 2
5486                                          */
5487                                         continue;
5488                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5489                                 if (!kthread_should_stop() &&
5490                                     mddev2->curr_resync >= mddev->curr_resync) {
5491                                         printk(KERN_INFO "md: delaying %s of %s"
5492                                                " until %s has finished (they"
5493                                                " share one or more physical units)\n",
5494                                                desc, mdname(mddev), mdname(mddev2));
5495                                         mddev_put(mddev2);
5496                                         schedule();
5497                                         finish_wait(&resync_wait, &wq);
5498                                         goto try_again;
5499                                 }
5500                                 finish_wait(&resync_wait, &wq);
5501                         }
5502                 }
5503         } while (mddev->curr_resync < 2);
5504
5505         j = 0;
5506         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5507                 /* resync follows the size requested by the personality,
5508                  * which defaults to physical size, but can be virtual size
5509                  */
5510                 max_sectors = mddev->resync_max_sectors;
5511                 mddev->resync_mismatches = 0;
5512                 /* we don't use the checkpoint if there's a bitmap */
5513                 if (!mddev->bitmap &&
5514                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5515                         j = mddev->recovery_cp;
5516         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5517                 max_sectors = mddev->size << 1;
5518         else {
5519                 /* recovery follows the physical size of devices */
5520                 max_sectors = mddev->size << 1;
5521                 j = MaxSector;
5522                 rdev_for_each(rdev, rtmp, mddev)
5523                         if (rdev->raid_disk >= 0 &&
5524                             !test_bit(Faulty, &rdev->flags) &&
5525                             !test_bit(In_sync, &rdev->flags) &&
5526                             rdev->recovery_offset < j)
5527                                 j = rdev->recovery_offset;
5528         }
5529
5530         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5531         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5532                 " %d KB/sec/disk.\n", speed_min(mddev));
5533         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5534                "(but not more than %d KB/sec) for %s.\n",
5535                speed_max(mddev), desc);
5536
5537         is_mddev_idle(mddev); /* this also initializes IO event counters */
5538
5539         io_sectors = 0;
5540         for (m = 0; m < SYNC_MARKS; m++) {
5541                 mark[m] = jiffies;
5542                 mark_cnt[m] = io_sectors;
5543         }
5544         last_mark = 0;
5545         mddev->resync_mark = mark[last_mark];
5546         mddev->resync_mark_cnt = mark_cnt[last_mark];
5547
5548         /*
5549          * Tune reconstruction:
5550          */
5551         window = 32*(PAGE_SIZE/512);
5552         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5553                 window/2,(unsigned long long) max_sectors/2);
5554
5555         atomic_set(&mddev->recovery_active, 0);
5556         init_waitqueue_head(&mddev->recovery_wait);
5557         last_check = 0;
5558
5559         if (j>2) {
5560                 printk(KERN_INFO
5561                        "md: resuming %s of %s from checkpoint.\n",
5562                        desc, mdname(mddev));
5563                 mddev->curr_resync = j;
5564         }
5565
5566         while (j < max_sectors) {
5567                 sector_t sectors;
5568
5569                 skipped = 0;
5570                 if (j >= mddev->resync_max) {
5571                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5572                         wait_event(mddev->recovery_wait,
5573                                    mddev->resync_max > j
5574                                    || kthread_should_stop());
5575                 }
5576                 if (kthread_should_stop())
5577                         goto interrupted;
5578                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5579                                                   currspeed < speed_min(mddev));
5580                 if (sectors == 0) {
5581                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5582                         goto out;
5583                 }
5584
5585                 if (!skipped) { /* actual IO requested */
5586                         io_sectors += sectors;
5587                         atomic_add(sectors, &mddev->recovery_active);
5588                 }
5589
5590                 j += sectors;
5591                 if (j>1) mddev->curr_resync = j;
5592                 mddev->curr_mark_cnt = io_sectors;
5593                 if (last_check == 0)
5594                         /* this is the earliers that rebuilt will be
5595                          * visible in /proc/mdstat
5596                          */
5597                         md_new_event(mddev);
5598
5599                 if (last_check + window > io_sectors || j == max_sectors)
5600                         continue;
5601
5602                 last_check = io_sectors;
5603
5604                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5605                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5606                         break;
5607
5608         repeat:
5609                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5610                         /* step marks */
5611                         int next = (last_mark+1) % SYNC_MARKS;
5612
5613                         mddev->resync_mark = mark[next];
5614                         mddev->resync_mark_cnt = mark_cnt[next];
5615                         mark[next] = jiffies;
5616                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5617                         last_mark = next;
5618                 }
5619
5620
5621                 if (kthread_should_stop())
5622                         goto interrupted;
5623
5624
5625                 /*
5626                  * this loop exits only if either when we are slower than
5627                  * the 'hard' speed limit, or the system was IO-idle for
5628                  * a jiffy.
5629                  * the system might be non-idle CPU-wise, but we only care
5630                  * about not overloading the IO subsystem. (things like an
5631                  * e2fsck being done on the RAID array should execute fast)
5632                  */
5633                 blk_unplug(mddev->queue);
5634                 cond_resched();
5635
5636                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5637                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5638
5639                 if (currspeed > speed_min(mddev)) {
5640                         if ((currspeed > speed_max(mddev)) ||
5641                                         !is_mddev_idle(mddev)) {
5642                                 msleep(500);
5643                                 goto repeat;
5644                         }
5645                 }
5646         }
5647         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5648         /*
5649          * this also signals 'finished resyncing' to md_stop
5650          */
5651  out:
5652         blk_unplug(mddev->queue);
5653
5654         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5655
5656         /* tell personality that we are finished */
5657         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5658
5659         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5660             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5661             mddev->curr_resync > 2) {
5662                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5663                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5664                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5665                                         printk(KERN_INFO
5666                                                "md: checkpointing %s of %s.\n",
5667                                                desc, mdname(mddev));
5668                                         mddev->recovery_cp = mddev->curr_resync;
5669                                 }
5670                         } else
5671                                 mddev->recovery_cp = MaxSector;
5672                 } else {
5673                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5674                                 mddev->curr_resync = MaxSector;
5675                         rdev_for_each(rdev, rtmp, mddev)
5676                                 if (rdev->raid_disk >= 0 &&
5677                                     !test_bit(Faulty, &rdev->flags) &&
5678                                     !test_bit(In_sync, &rdev->flags) &&
5679                                     rdev->recovery_offset < mddev->curr_resync)
5680                                         rdev->recovery_offset = mddev->curr_resync;
5681                 }
5682         }
5683         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5684
5685  skip:
5686         mddev->curr_resync = 0;
5687         mddev->resync_max = MaxSector;
5688         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5689         wake_up(&resync_wait);
5690         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5691         md_wakeup_thread(mddev->thread);
5692         return;
5693
5694  interrupted:
5695         /*
5696          * got a signal, exit.
5697          */
5698         printk(KERN_INFO
5699                "md: md_do_sync() got signal ... exiting\n");
5700         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5701         goto out;
5702
5703 }
5704 EXPORT_SYMBOL_GPL(md_do_sync);
5705
5706
5707 static int remove_and_add_spares(mddev_t *mddev)
5708 {
5709         mdk_rdev_t *rdev;
5710         struct list_head *rtmp;
5711         int spares = 0;
5712
5713         rdev_for_each(rdev, rtmp, mddev)
5714                 if (rdev->raid_disk >= 0 &&
5715                     !mddev->external &&
5716                     (test_bit(Faulty, &rdev->flags) ||
5717                      ! test_bit(In_sync, &rdev->flags)) &&
5718                     atomic_read(&rdev->nr_pending)==0) {
5719                         if (mddev->pers->hot_remove_disk(
5720                                     mddev, rdev->raid_disk)==0) {
5721                                 char nm[20];
5722                                 sprintf(nm,"rd%d", rdev->raid_disk);
5723                                 sysfs_remove_link(&mddev->kobj, nm);
5724                                 rdev->raid_disk = -1;
5725                         }
5726                 }
5727
5728         if (mddev->degraded) {
5729                 rdev_for_each(rdev, rtmp, mddev)
5730                         if (rdev->raid_disk < 0
5731                             && !test_bit(Faulty, &rdev->flags)) {
5732                                 rdev->recovery_offset = 0;
5733                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5734                                         char nm[20];
5735                                         sprintf(nm, "rd%d", rdev->raid_disk);
5736                                         if (sysfs_create_link(&mddev->kobj,
5737                                                               &rdev->kobj, nm))
5738                                                 printk(KERN_WARNING
5739                                                        "md: cannot register "
5740                                                        "%s for %s\n",
5741                                                        nm, mdname(mddev));
5742                                         spares++;
5743                                         md_new_event(mddev);
5744                                 } else
5745                                         break;
5746                         }
5747         }
5748         return spares;
5749 }
5750 /*
5751  * This routine is regularly called by all per-raid-array threads to
5752  * deal with generic issues like resync and super-block update.
5753  * Raid personalities that don't have a thread (linear/raid0) do not
5754  * need this as they never do any recovery or update the superblock.
5755  *
5756  * It does not do any resync itself, but rather "forks" off other threads
5757  * to do that as needed.
5758  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5759  * "->recovery" and create a thread at ->sync_thread.
5760  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5761  * and wakeups up this thread which will reap the thread and finish up.
5762  * This thread also removes any faulty devices (with nr_pending == 0).
5763  *
5764  * The overall approach is:
5765  *  1/ if the superblock needs updating, update it.
5766  *  2/ If a recovery thread is running, don't do anything else.
5767  *  3/ If recovery has finished, clean up, possibly marking spares active.
5768  *  4/ If there are any faulty devices, remove them.
5769  *  5/ If array is degraded, try to add spares devices
5770  *  6/ If array has spares or is not in-sync, start a resync thread.
5771  */
5772 void md_check_recovery(mddev_t *mddev)
5773 {
5774         mdk_rdev_t *rdev;
5775         struct list_head *rtmp;
5776
5777
5778         if (mddev->bitmap)
5779                 bitmap_daemon_work(mddev->bitmap);
5780
5781         if (mddev->ro)
5782                 return;
5783
5784         if (signal_pending(current)) {
5785                 if (mddev->pers->sync_request) {
5786                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5787                                mdname(mddev));
5788                         mddev->safemode = 2;
5789                 }
5790                 flush_signals(current);
5791         }
5792
5793         if ( ! (
5794                 (mddev->flags && !mddev->external) ||
5795                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5796                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5797                 (mddev->safemode == 1) ||
5798                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5799                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5800                 ))
5801                 return;
5802
5803         if (mddev_trylock(mddev)) {
5804                 int spares = 0;
5805
5806                 spin_lock_irq(&mddev->write_lock);
5807                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5808                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5809                         mddev->in_sync = 1;
5810                         if (mddev->persistent)
5811                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5812                 }
5813                 if (mddev->safemode == 1)
5814                         mddev->safemode = 0;
5815                 spin_unlock_irq(&mddev->write_lock);
5816
5817                 if (mddev->flags)
5818                         md_update_sb(mddev, 0);
5819
5820
5821                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5822                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5823                         /* resync/recovery still happening */
5824                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5825                         goto unlock;
5826                 }
5827                 if (mddev->sync_thread) {
5828                         /* resync has finished, collect result */
5829                         md_unregister_thread(mddev->sync_thread);
5830                         mddev->sync_thread = NULL;
5831                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5832                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5833                                 /* success...*/
5834                                 /* activate any spares */
5835                                 mddev->pers->spare_active(mddev);
5836                         }
5837                         md_update_sb(mddev, 1);
5838
5839                         /* if array is no-longer degraded, then any saved_raid_disk
5840                          * information must be scrapped
5841                          */
5842                         if (!mddev->degraded)
5843                                 rdev_for_each(rdev, rtmp, mddev)
5844                                         rdev->saved_raid_disk = -1;
5845
5846                         mddev->recovery = 0;
5847                         /* flag recovery needed just to double check */
5848                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5849                         md_new_event(mddev);
5850                         goto unlock;
5851                 }
5852                 /* Clear some bits that don't mean anything, but
5853                  * might be left set
5854                  */
5855                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5856                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5857                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5858                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5859
5860                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5861                         goto unlock;
5862                 /* no recovery is running.
5863                  * remove any failed drives, then
5864                  * add spares if possible.
5865                  * Spare are also removed and re-added, to allow
5866                  * the personality to fail the re-add.
5867                  */
5868
5869                 if (mddev->reshape_position != MaxSector) {
5870                         if (mddev->pers->check_reshape(mddev) != 0)
5871                                 /* Cannot proceed */
5872                                 goto unlock;
5873                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5874                 } else if ((spares = remove_and_add_spares(mddev))) {
5875                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5876                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5877                 } else if (mddev->recovery_cp < MaxSector) {
5878                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5879                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5880                         /* nothing to be done ... */
5881                         goto unlock;
5882
5883                 if (mddev->pers->sync_request) {
5884                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5885                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5886                                 /* We are adding a device or devices to an array
5887                                  * which has the bitmap stored on all devices.
5888                                  * So make sure all bitmap pages get written
5889                                  */
5890                                 bitmap_write_all(mddev->bitmap);
5891                         }
5892                         mddev->sync_thread = md_register_thread(md_do_sync,
5893                                                                 mddev,
5894                                                                 "%s_resync");
5895                         if (!mddev->sync_thread) {
5896                                 printk(KERN_ERR "%s: could not start resync"
5897                                         " thread...\n",
5898                                         mdname(mddev));
5899                                 /* leave the spares where they are, it shouldn't hurt */
5900                                 mddev->recovery = 0;
5901                         } else
5902                                 md_wakeup_thread(mddev->sync_thread);
5903                         md_new_event(mddev);
5904                 }
5905         unlock:
5906                 mddev_unlock(mddev);
5907         }
5908 }
5909
5910 static int md_notify_reboot(struct notifier_block *this,
5911                             unsigned long code, void *x)
5912 {
5913         struct list_head *tmp;
5914         mddev_t *mddev;
5915
5916         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5917
5918                 printk(KERN_INFO "md: stopping all md devices.\n");
5919
5920                 for_each_mddev(mddev, tmp)
5921                         if (mddev_trylock(mddev)) {
5922                                 do_md_stop (mddev, 1);
5923                                 mddev_unlock(mddev);
5924                         }
5925                 /*
5926                  * certain more exotic SCSI devices are known to be
5927                  * volatile wrt too early system reboots. While the
5928                  * right place to handle this issue is the given
5929                  * driver, we do want to have a safe RAID driver ...
5930                  */
5931                 mdelay(1000*1);
5932         }
5933         return NOTIFY_DONE;
5934 }
5935
5936 static struct notifier_block md_notifier = {
5937         .notifier_call  = md_notify_reboot,
5938         .next           = NULL,
5939         .priority       = INT_MAX, /* before any real devices */
5940 };
5941
5942 static void md_geninit(void)
5943 {
5944         struct proc_dir_entry *p;
5945
5946         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5947
5948         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5949         if (p)
5950                 p->proc_fops = &md_seq_fops;
5951 }
5952
5953 static int __init md_init(void)
5954 {
5955         if (register_blkdev(MAJOR_NR, "md"))
5956                 return -1;
5957         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5958                 unregister_blkdev(MAJOR_NR, "md");
5959                 return -1;
5960         }
5961         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5962                             md_probe, NULL, NULL);
5963         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5964                             md_probe, NULL, NULL);
5965
5966         register_reboot_notifier(&md_notifier);
5967         raid_table_header = register_sysctl_table(raid_root_table);
5968
5969         md_geninit();
5970         return (0);
5971 }
5972
5973
5974 #ifndef MODULE
5975
5976 /*
5977  * Searches all registered partitions for autorun RAID arrays
5978  * at boot time.
5979  */
5980
5981 static LIST_HEAD(all_detected_devices);
5982 struct detected_devices_node {
5983         struct list_head list;
5984         dev_t dev;
5985 };
5986
5987 void md_autodetect_dev(dev_t dev)
5988 {
5989         struct detected_devices_node *node_detected_dev;
5990
5991         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5992         if (node_detected_dev) {
5993                 node_detected_dev->dev = dev;
5994                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5995         } else {
5996                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5997                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
5998         }
5999 }
6000
6001
6002 static void autostart_arrays(int part)
6003 {
6004         mdk_rdev_t *rdev;
6005         struct detected_devices_node *node_detected_dev;
6006         dev_t dev;
6007         int i_scanned, i_passed;
6008
6009         i_scanned = 0;
6010         i_passed = 0;
6011
6012         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6013
6014         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6015                 i_scanned++;
6016                 node_detected_dev = list_entry(all_detected_devices.next,
6017                                         struct detected_devices_node, list);
6018                 list_del(&node_detected_dev->list);
6019                 dev = node_detected_dev->dev;
6020                 kfree(node_detected_dev);
6021                 rdev = md_import_device(dev,0, 90);
6022                 if (IS_ERR(rdev))
6023                         continue;
6024
6025                 if (test_bit(Faulty, &rdev->flags)) {
6026                         MD_BUG();
6027                         continue;
6028                 }
6029                 set_bit(AutoDetected, &rdev->flags);
6030                 list_add(&rdev->same_set, &pending_raid_disks);
6031                 i_passed++;
6032         }
6033
6034         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6035                                                 i_scanned, i_passed);
6036
6037         autorun_devices(part);
6038 }
6039
6040 #endif /* !MODULE */
6041
6042 static __exit void md_exit(void)
6043 {
6044         mddev_t *mddev;
6045         struct list_head *tmp;
6046
6047         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6048         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6049
6050         unregister_blkdev(MAJOR_NR,"md");
6051         unregister_blkdev(mdp_major, "mdp");
6052         unregister_reboot_notifier(&md_notifier);
6053         unregister_sysctl_table(raid_table_header);
6054         remove_proc_entry("mdstat", NULL);
6055         for_each_mddev(mddev, tmp) {
6056                 struct gendisk *disk = mddev->gendisk;
6057                 if (!disk)
6058                         continue;
6059                 export_array(mddev);
6060                 del_gendisk(disk);
6061                 put_disk(disk);
6062                 mddev->gendisk = NULL;
6063                 mddev_put(mddev);
6064         }
6065 }
6066
6067 subsys_initcall(md_init);
6068 module_exit(md_exit)
6069
6070 static int get_ro(char *buffer, struct kernel_param *kp)
6071 {
6072         return sprintf(buffer, "%d", start_readonly);
6073 }
6074 static int set_ro(const char *val, struct kernel_param *kp)
6075 {
6076         char *e;
6077         int num = simple_strtoul(val, &e, 10);
6078         if (*val && (*e == '\0' || *e == '\n')) {
6079                 start_readonly = num;
6080                 return 0;
6081         }
6082         return -EINVAL;
6083 }
6084
6085 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6086 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6087
6088
6089 EXPORT_SYMBOL(register_md_personality);
6090 EXPORT_SYMBOL(unregister_md_personality);
6091 EXPORT_SYMBOL(md_error);
6092 EXPORT_SYMBOL(md_done_sync);
6093 EXPORT_SYMBOL(md_write_start);
6094 EXPORT_SYMBOL(md_write_end);
6095 EXPORT_SYMBOL(md_register_thread);
6096 EXPORT_SYMBOL(md_unregister_thread);
6097 EXPORT_SYMBOL(md_wakeup_thread);
6098 EXPORT_SYMBOL(md_check_recovery);
6099 MODULE_LICENSE("GPL");
6100 MODULE_ALIAS("md");
6101 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);