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

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/kernel.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/buffer_head.h> /* for invalidate_bdev */
  43 #include <linux/poll.h>
  44 #include <linux/mutex.h>
  45 #include <linux/ctype.h>
  46 #include <linux/freezer.h>
  47
  48 #include <linux/init.h>
  49
  50 #include <linux/file.h>
  51
  52 #ifdef CONFIG_KMOD
  53 #include <linux/kmod.h>
  54 #endif
  55
  56 #include <asm/unaligned.h>
  57
  58 #define MAJOR_NR MD_MAJOR
  59 #define MD_DRIVER
  60
  61 /* 63 partitions with the alternate major number (mdp) */
  62 #define MdpMinorShift 6
  63
  64 #define DEBUG 0
  65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  66
  67
  68 #ifndef MODULE
  69 static void autostart_arrays (int part);
  70 #endif
  71
  72 static LIST_HEAD(pers_list);
  73 static DEFINE_SPINLOCK(pers_lock);
  74
  75 static void md_print_devices(void);
  76
  77 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  78
  79 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  80
  81 /*
  82  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  83  * is 1000 KB/sec, so the extra system load does not show up that much.
  84  * Increase it if you want to have more _guaranteed_ speed. Note that
  85  * the RAID driver will use the maximum available bandwidth if the IO
  86  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  87  * speed limit - in case reconstruction slows down your system despite
  88  * idle IO detection.
  89  *
  90  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  91  * or /sys/block/mdX/md/sync_speed_{min,max}
  92  */
  93
  94 static int sysctl_speed_limit_min = 1000;
  95 static int sysctl_speed_limit_max = 200000;
  96 static inline int speed_min(mddev_t *mddev)
  97 {
  98         return mddev->sync_speed_min ?
  99                 mddev->sync_speed_min : sysctl_speed_limit_min;
 100 }
 101
 102 static inline int speed_max(mddev_t *mddev)
 103 {
 104         return mddev->sync_speed_max ?
 105                 mddev->sync_speed_max : sysctl_speed_limit_max;
 106 }
 107
 108 static struct ctl_table_header *raid_table_header;
 109
 110 static ctl_table raid_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 113                 .procname       = "speed_limit_min",
 114                 .data           = &sysctl_speed_limit_min,
 115                 .maxlen         = sizeof(int),
 116                 .mode           = S_IRUGO|S_IWUSR,
 117                 .proc_handler   = &proc_dointvec,
 118         },
 119         {
 120                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 121                 .procname       = "speed_limit_max",
 122                 .data           = &sysctl_speed_limit_max,
 123                 .maxlen         = sizeof(int),
 124                 .mode           = S_IRUGO|S_IWUSR,
 125                 .proc_handler   = &proc_dointvec,
 126         },
 127         { .ctl_name = 0 }
 128 };
 129
 130 static ctl_table raid_dir_table[] = {
 131         {
 132                 .ctl_name       = DEV_RAID,
 133                 .procname       = "raid",
 134                 .maxlen         = 0,
 135                 .mode           = S_IRUGO|S_IXUGO,
 136                 .child          = raid_table,
 137         },
 138         { .ctl_name = 0 }
 139 };
 140
 141 static ctl_table raid_root_table[] = {
 142         {
 143                 .ctl_name       = CTL_DEV,
 144                 .procname       = "dev",
 145                 .maxlen         = 0,
 146                 .mode           = 0555,
 147                 .child          = raid_dir_table,
 148         },
 149         { .ctl_name = 0 }
 150 };
 151
 152 static struct block_device_operations md_fops;
 153
 154 static int start_readonly;
 155
 156 /*
 157  * We have a system wide 'event count' that is incremented
 158  * on any 'interesting' event, and readers of /proc/mdstat
 159  * can use 'poll' or 'select' to find out when the event
 160  * count increases.
 161  *
 162  * Events are:
 163  *  start array, stop array, error, add device, remove device,
 164  *  start build, activate spare
 165  */
 166 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 167 static atomic_t md_event_count;
 168 void md_new_event(mddev_t *mddev)
 169 {
 170         atomic_inc(&md_event_count);
 171         wake_up(&md_event_waiters);
 172 }
 173 EXPORT_SYMBOL_GPL(md_new_event);
 174
 175 /* Alternate version that can be called from interrupts
 176  * when calling sysfs_notify isn't needed.
 177  */
 178 static void md_new_event_inintr(mddev_t *mddev)
 179 {
 180         atomic_inc(&md_event_count);
 181         wake_up(&md_event_waiters);
 182 }
 183
 184 /*
 185  * Enables to iterate over all existing md arrays
 186  * all_mddevs_lock protects this list.
 187  */
 188 static LIST_HEAD(all_mddevs);
 189 static DEFINE_SPINLOCK(all_mddevs_lock);
 190
 191
 192 /*
 193  * iterates through all used mddevs in the system.
 194  * We take care to grab the all_mddevs_lock whenever navigating
 195  * the list, and to always hold a refcount when unlocked.
 196  * Any code which breaks out of this loop while own
 197  * a reference to the current mddev and must mddev_put it.
 198  */
 199 #define for_each_mddev(mddev,tmp)                                       \
 200                                                                         \
 201         for (({ spin_lock(&all_mddevs_lock);                            \
 202                 tmp = all_mddevs.next;                                  \
 203                 mddev = NULL;});                                        \
 204              ({ if (tmp != &all_mddevs)                                 \
 205                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 206                 spin_unlock(&all_mddevs_lock);                          \
 207                 if (mddev) mddev_put(mddev);                            \
 208                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 209                 tmp != &all_mddevs;});                                  \
 210              ({ spin_lock(&all_mddevs_lock);                            \
 211                 tmp = tmp->next;})                                      \
 212                 )
 213
 214
 215 static int md_fail_request (struct request_queue *q, struct bio *bio)
 216 {
 217         bio_io_error(bio);
 218         return 0;
 219 }
 220
 221 static inline mddev_t *mddev_get(mddev_t *mddev)
 222 {
 223         atomic_inc(&mddev->active);
 224         return mddev;
 225 }
 226
 227 static void mddev_put(mddev_t *mddev)
 228 {
 229         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 230                 return;
 231         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 232                 list_del(&mddev->all_mddevs);
 233                 spin_unlock(&all_mddevs_lock);
 234                 blk_cleanup_queue(mddev->queue);
 235                 kobject_put(&mddev->kobj);
 236         } else
 237                 spin_unlock(&all_mddevs_lock);
 238 }
 239
 240 static mddev_t * mddev_find(dev_t unit)
 241 {
 242         mddev_t *mddev, *new = NULL;
 243
 244  retry:
 245         spin_lock(&all_mddevs_lock);
 246         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 247                 if (mddev->unit == unit) {
 248                         mddev_get(mddev);
 249                         spin_unlock(&all_mddevs_lock);
 250                         kfree(new);
 251                         return mddev;
 252                 }
 253
 254         if (new) {
 255                 list_add(&new->all_mddevs, &all_mddevs);
 256                 spin_unlock(&all_mddevs_lock);
 257                 return new;
 258         }
 259         spin_unlock(&all_mddevs_lock);
 260
 261         new = kzalloc(sizeof(*new), GFP_KERNEL);
 262         if (!new)
 263                 return NULL;
 264
 265         new->unit = unit;
 266         if (MAJOR(unit) == MD_MAJOR)
 267                 new->md_minor = MINOR(unit);
 268         else
 269                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 270
 271         mutex_init(&new->reconfig_mutex);
 272         INIT_LIST_HEAD(&new->disks);
 273         INIT_LIST_HEAD(&new->all_mddevs);
 274         init_timer(&new->safemode_timer);
 275         atomic_set(&new->active, 1);
 276         spin_lock_init(&new->write_lock);
 277         init_waitqueue_head(&new->sb_wait);
 278         init_waitqueue_head(&new->recovery_wait);
 279         new->reshape_position = MaxSector;
 280         new->resync_min = 0;
 281         new->resync_max = MaxSector;
 282         new->level = LEVEL_NONE;
 283
 284         new->queue = blk_alloc_queue(GFP_KERNEL);
 285         if (!new->queue) {
 286                 kfree(new);
 287                 return NULL;
 288         }
 289         /* Can be unlocked because the queue is new: no concurrency */
 290         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
 291
 292         blk_queue_make_request(new->queue, md_fail_request);
 293
 294         goto retry;
 295 }
 296
 297 static inline int mddev_lock(mddev_t * mddev)
 298 {
 299         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 300 }
 301
 302 static inline int mddev_trylock(mddev_t * mddev)
 303 {
 304         return mutex_trylock(&mddev->reconfig_mutex);
 305 }
 306
 307 static inline void mddev_unlock(mddev_t * mddev)
 308 {
 309         mutex_unlock(&mddev->reconfig_mutex);
 310
 311         md_wakeup_thread(mddev->thread);
 312 }
 313
 314 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 315 {
 316         mdk_rdev_t * rdev;
 317         struct list_head *tmp;
 318
 319         rdev_for_each(rdev, tmp, mddev) {
 320                 if (rdev->desc_nr == nr)
 321                         return rdev;
 322         }
 323         return NULL;
 324 }
 325
 326 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 327 {
 328         struct list_head *tmp;
 329         mdk_rdev_t *rdev;
 330
 331         rdev_for_each(rdev, tmp, mddev) {
 332                 if (rdev->bdev->bd_dev == dev)
 333                         return rdev;
 334         }
 335         return NULL;
 336 }
 337
 338 static struct mdk_personality *find_pers(int level, char *clevel)
 339 {
 340         struct mdk_personality *pers;
 341         list_for_each_entry(pers, &pers_list, list) {
 342                 if (level != LEVEL_NONE && pers->level == level)
 343                         return pers;
 344                 if (strcmp(pers->name, clevel)==0)
 345                         return pers;
 346         }
 347         return NULL;
 348 }
 349
 350 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 351 {
 352         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 353         return MD_NEW_SIZE_BLOCKS(size);
 354 }
 355
 356 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
 357 {
 358         sector_t num_sectors = rdev->sb_offset * 2;
 359
 360         if (chunk_size)
 361                 num_sectors &= ~((sector_t)chunk_size/512 - 1);
 362         return num_sectors;
 363 }
 364
 365 static int alloc_disk_sb(mdk_rdev_t * rdev)
 366 {
 367         if (rdev->sb_page)
 368                 MD_BUG();
 369
 370         rdev->sb_page = alloc_page(GFP_KERNEL);
 371         if (!rdev->sb_page) {
 372                 printk(KERN_ALERT "md: out of memory.\n");
 373                 return -ENOMEM;
 374         }
 375
 376         return 0;
 377 }
 378
 379 static void free_disk_sb(mdk_rdev_t * rdev)
 380 {
 381         if (rdev->sb_page) {
 382                 put_page(rdev->sb_page);
 383                 rdev->sb_loaded = 0;
 384                 rdev->sb_page = NULL;
 385                 rdev->sb_offset = 0;
 386                 rdev->size = 0;
 387         }
 388 }
 389
 390
 391 static void super_written(struct bio *bio, int error)
 392 {
 393         mdk_rdev_t *rdev = bio->bi_private;
 394         mddev_t *mddev = rdev->mddev;
 395
 396         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 397                 printk("md: super_written gets error=%d, uptodate=%d\n",
 398                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 399                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 400                 md_error(mddev, rdev);
 401         }
 402
 403         if (atomic_dec_and_test(&mddev->pending_writes))
 404                 wake_up(&mddev->sb_wait);
 405         bio_put(bio);
 406 }
 407
 408 static void super_written_barrier(struct bio *bio, int error)
 409 {
 410         struct bio *bio2 = bio->bi_private;
 411         mdk_rdev_t *rdev = bio2->bi_private;
 412         mddev_t *mddev = rdev->mddev;
 413
 414         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 415             error == -EOPNOTSUPP) {
 416                 unsigned long flags;
 417                 /* barriers don't appear to be supported :-( */
 418                 set_bit(BarriersNotsupp, &rdev->flags);
 419                 mddev->barriers_work = 0;
 420                 spin_lock_irqsave(&mddev->write_lock, flags);
 421                 bio2->bi_next = mddev->biolist;
 422                 mddev->biolist = bio2;
 423                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 424                 wake_up(&mddev->sb_wait);
 425                 bio_put(bio);
 426         } else {
 427                 bio_put(bio2);
 428                 bio->bi_private = rdev;
 429                 super_written(bio, error);
 430         }
 431 }
 432
 433 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 434                    sector_t sector, int size, struct page *page)
 435 {
 436         /* write first size bytes of page to sector of rdev
 437          * Increment mddev->pending_writes before returning
 438          * and decrement it on completion, waking up sb_wait
 439          * if zero is reached.
 440          * If an error occurred, call md_error
 441          *
 442          * As we might need to resubmit the request if BIO_RW_BARRIER
 443          * causes ENOTSUPP, we allocate a spare bio...
 444          */
 445         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 446         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 447
 448         bio->bi_bdev = rdev->bdev;
 449         bio->bi_sector = sector;
 450         bio_add_page(bio, page, size, 0);
 451         bio->bi_private = rdev;
 452         bio->bi_end_io = super_written;
 453         bio->bi_rw = rw;
 454
 455         atomic_inc(&mddev->pending_writes);
 456         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 457                 struct bio *rbio;
 458                 rw |= (1<<BIO_RW_BARRIER);
 459                 rbio = bio_clone(bio, GFP_NOIO);
 460                 rbio->bi_private = bio;
 461                 rbio->bi_end_io = super_written_barrier;
 462                 submit_bio(rw, rbio);
 463         } else
 464                 submit_bio(rw, bio);
 465 }
 466
 467 void md_super_wait(mddev_t *mddev)
 468 {
 469         /* wait for all superblock writes that were scheduled to complete.
 470          * if any had to be retried (due to BARRIER problems), retry them
 471          */
 472         DEFINE_WAIT(wq);
 473         for(;;) {
 474                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 475                 if (atomic_read(&mddev->pending_writes)==0)
 476                         break;
 477                 while (mddev->biolist) {
 478                         struct bio *bio;
 479                         spin_lock_irq(&mddev->write_lock);
 480                         bio = mddev->biolist;
 481                         mddev->biolist = bio->bi_next ;
 482                         bio->bi_next = NULL;
 483                         spin_unlock_irq(&mddev->write_lock);
 484                         submit_bio(bio->bi_rw, bio);
 485                 }
 486                 schedule();
 487         }
 488         finish_wait(&mddev->sb_wait, &wq);
 489 }
 490
 491 static void bi_complete(struct bio *bio, int error)
 492 {
 493         complete((struct completion*)bio->bi_private);
 494 }
 495
 496 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 497                    struct page *page, int rw)
 498 {
 499         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 500         struct completion event;
 501         int ret;
 502
 503         rw |= (1 << BIO_RW_SYNC);
 504
 505         bio->bi_bdev = bdev;
 506         bio->bi_sector = sector;
 507         bio_add_page(bio, page, size, 0);
 508         init_completion(&event);
 509         bio->bi_private = &event;
 510         bio->bi_end_io = bi_complete;
 511         submit_bio(rw, bio);
 512         wait_for_completion(&event);
 513
 514         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 515         bio_put(bio);
 516         return ret;
 517 }
 518 EXPORT_SYMBOL_GPL(sync_page_io);
 519
 520 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 521 {
 522         char b[BDEVNAME_SIZE];
 523         if (!rdev->sb_page) {
 524                 MD_BUG();
 525                 return -EINVAL;
 526         }
 527         if (rdev->sb_loaded)
 528                 return 0;
 529
 530
 531         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 532                 goto fail;
 533         rdev->sb_loaded = 1;
 534         return 0;
 535
 536 fail:
 537         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 538                 bdevname(rdev->bdev,b));
 539         return -EINVAL;
 540 }
 541
 542 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 543 {
 544         return  sb1->set_uuid0 == sb2->set_uuid0 &&
 545                 sb1->set_uuid1 == sb2->set_uuid1 &&
 546                 sb1->set_uuid2 == sb2->set_uuid2 &&
 547                 sb1->set_uuid3 == sb2->set_uuid3;
 548 }
 549
 550 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 551 {
 552         int ret;
 553         mdp_super_t *tmp1, *tmp2;
 554
 555         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 556         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 557
 558         if (!tmp1 || !tmp2) {
 559                 ret = 0;
 560                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 561                 goto abort;
 562         }
 563
 564         *tmp1 = *sb1;
 565         *tmp2 = *sb2;
 566
 567         /*
 568          * nr_disks is not constant
 569          */
 570         tmp1->nr_disks = 0;
 571         tmp2->nr_disks = 0;
 572
 573         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 574 abort:
 575         kfree(tmp1);
 576         kfree(tmp2);
 577         return ret;
 578 }
 579
 580
 581 static u32 md_csum_fold(u32 csum)
 582 {
 583         csum = (csum & 0xffff) + (csum >> 16);
 584         return (csum & 0xffff) + (csum >> 16);
 585 }
 586
 587 static unsigned int calc_sb_csum(mdp_super_t * sb)
 588 {
 589         u64 newcsum = 0;
 590         u32 *sb32 = (u32*)sb;
 591         int i;
 592         unsigned int disk_csum, csum;
 593
 594         disk_csum = sb->sb_csum;
 595         sb->sb_csum = 0;
 596
 597         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 598                 newcsum += sb32[i];
 599         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 600
 601
 602 #ifdef CONFIG_ALPHA
 603         /* This used to use csum_partial, which was wrong for several
 604          * reasons including that different results are returned on
 605          * different architectures.  It isn't critical that we get exactly
 606          * the same return value as before (we always csum_fold before
 607          * testing, and that removes any differences).  However as we
 608          * know that csum_partial always returned a 16bit value on
 609          * alphas, do a fold to maximise conformity to previous behaviour.
 610          */
 611         sb->sb_csum = md_csum_fold(disk_csum);
 612 #else
 613         sb->sb_csum = disk_csum;
 614 #endif
 615         return csum;
 616 }
 617
 618
 619 /*
 620  * Handle superblock details.
 621  * We want to be able to handle multiple superblock formats
 622  * so we have a common interface to them all, and an array of
 623  * different handlers.
 624  * We rely on user-space to write the initial superblock, and support
 625  * reading and updating of superblocks.
 626  * Interface methods are:
 627  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 628  *      loads and validates a superblock on dev.
 629  *      if refdev != NULL, compare superblocks on both devices
 630  *    Return:
 631  *      0 - dev has a superblock that is compatible with refdev
 632  *      1 - dev has a superblock that is compatible and newer than refdev
 633  *          so dev should be used as the refdev in future
 634  *     -EINVAL superblock incompatible or invalid
 635  *     -othererror e.g. -EIO
 636  *
 637  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 638  *      Verify that dev is acceptable into mddev.
 639  *       The first time, mddev->raid_disks will be 0, and data from
 640  *       dev should be merged in.  Subsequent calls check that dev
 641  *       is new enough.  Return 0 or -EINVAL
 642  *
 643  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 644  *     Update the superblock for rdev with data in mddev
 645  *     This does not write to disc.
 646  *
 647  */
 648
 649 struct super_type  {
 650         char                *name;
 651         struct module       *owner;
 652         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
 653                                           int minor_version);
 654         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 655         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 656         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
 657                                                 unsigned long long size);
 658 };
 659
 660 /*
 661  * load_super for 0.90.0
 662  */
 663 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 664 {
 665         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 666         mdp_super_t *sb;
 667         int ret;
 668         sector_t sb_offset;
 669
 670         /*
 671          * Calculate the position of the superblock,
 672          * it's at the end of the disk.
 673          *
 674          * It also happens to be a multiple of 4Kb.
 675          */
 676         sb_offset = calc_dev_sboffset(rdev->bdev);
 677         rdev->sb_offset = sb_offset;
 678
 679         ret = read_disk_sb(rdev, MD_SB_BYTES);
 680         if (ret) return ret;
 681
 682         ret = -EINVAL;
 683
 684         bdevname(rdev->bdev, b);
 685         sb = (mdp_super_t*)page_address(rdev->sb_page);
 686
 687         if (sb->md_magic != MD_SB_MAGIC) {
 688                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 689                        b);
 690                 goto abort;
 691         }
 692
 693         if (sb->major_version != 0 ||
 694             sb->minor_version < 90 ||
 695             sb->minor_version > 91) {
 696                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 697                         sb->major_version, sb->minor_version,
 698                         b);
 699                 goto abort;
 700         }
 701
 702         if (sb->raid_disks <= 0)
 703                 goto abort;
 704
 705         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 706                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 707                         b);
 708                 goto abort;
 709         }
 710
 711         rdev->preferred_minor = sb->md_minor;
 712         rdev->data_offset = 0;
 713         rdev->sb_size = MD_SB_BYTES;
 714
 715         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 716                 if (sb->level != 1 && sb->level != 4
 717                     && sb->level != 5 && sb->level != 6
 718                     && sb->level != 10) {
 719                         /* FIXME use a better test */
 720                         printk(KERN_WARNING
 721                                "md: bitmaps not supported for this level.\n");
 722                         goto abort;
 723                 }
 724         }
 725
 726         if (sb->level == LEVEL_MULTIPATH)
 727                 rdev->desc_nr = -1;
 728         else
 729                 rdev->desc_nr = sb->this_disk.number;
 730
 731         if (!refdev) {
 732                 ret = 1;
 733         } else {
 734                 __u64 ev1, ev2;
 735                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 736                 if (!uuid_equal(refsb, sb)) {
 737                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 738                                 b, bdevname(refdev->bdev,b2));
 739                         goto abort;
 740                 }
 741                 if (!sb_equal(refsb, sb)) {
 742                         printk(KERN_WARNING "md: %s has same UUID"
 743                                " but different superblock to %s\n",
 744                                b, bdevname(refdev->bdev, b2));
 745                         goto abort;
 746                 }
 747                 ev1 = md_event(sb);
 748                 ev2 = md_event(refsb);
 749                 if (ev1 > ev2)
 750                         ret = 1;
 751                 else
 752                         ret = 0;
 753         }
 754         rdev->size = calc_num_sectors(rdev, sb->chunk_size) / 2;
 755
 756         if (rdev->size < sb->size && sb->level > 1)
 757                 /* "this cannot possibly happen" ... */
 758                 ret = -EINVAL;
 759
 760  abort:
 761         return ret;
 762 }
 763
 764 /*
 765  * validate_super for 0.90.0
 766  */
 767 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 768 {
 769         mdp_disk_t *desc;
 770         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 771         __u64 ev1 = md_event(sb);
 772
 773         rdev->raid_disk = -1;
 774         clear_bit(Faulty, &rdev->flags);
 775         clear_bit(In_sync, &rdev->flags);
 776         clear_bit(WriteMostly, &rdev->flags);
 777         clear_bit(BarriersNotsupp, &rdev->flags);
 778
 779         if (mddev->raid_disks == 0) {
 780                 mddev->major_version = 0;
 781                 mddev->minor_version = sb->minor_version;
 782                 mddev->patch_version = sb->patch_version;
 783                 mddev->external = 0;
 784                 mddev->chunk_size = sb->chunk_size;
 785                 mddev->ctime = sb->ctime;
 786                 mddev->utime = sb->utime;
 787                 mddev->level = sb->level;
 788                 mddev->clevel[0] = 0;
 789                 mddev->layout = sb->layout;
 790                 mddev->raid_disks = sb->raid_disks;
 791                 mddev->size = sb->size;
 792                 mddev->events = ev1;
 793                 mddev->bitmap_offset = 0;
 794                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 795
 796                 if (mddev->minor_version >= 91) {
 797                         mddev->reshape_position = sb->reshape_position;
 798                         mddev->delta_disks = sb->delta_disks;
 799                         mddev->new_level = sb->new_level;
 800                         mddev->new_layout = sb->new_layout;
 801                         mddev->new_chunk = sb->new_chunk;
 802                 } else {
 803                         mddev->reshape_position = MaxSector;
 804                         mddev->delta_disks = 0;
 805                         mddev->new_level = mddev->level;
 806                         mddev->new_layout = mddev->layout;
 807                         mddev->new_chunk = mddev->chunk_size;
 808                 }
 809
 810                 if (sb->state & (1<<MD_SB_CLEAN))
 811                         mddev->recovery_cp = MaxSector;
 812                 else {
 813                         if (sb->events_hi == sb->cp_events_hi &&
 814                                 sb->events_lo == sb->cp_events_lo) {
 815                                 mddev->recovery_cp = sb->recovery_cp;
 816                         } else
 817                                 mddev->recovery_cp = 0;
 818                 }
 819
 820                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 821                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 822                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 823                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 824
 825                 mddev->max_disks = MD_SB_DISKS;
 826
 827                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 828                     mddev->bitmap_file == NULL)
 829                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 830
 831         } else if (mddev->pers == NULL) {
 832                 /* Insist on good event counter while assembling */
 833                 ++ev1;
 834                 if (ev1 < mddev->events)
 835                         return -EINVAL;
 836         } else if (mddev->bitmap) {
 837                 /* if adding to array with a bitmap, then we can accept an
 838                  * older device ... but not too old.
 839                  */
 840                 if (ev1 < mddev->bitmap->events_cleared)
 841                         return 0;
 842         } else {
 843                 if (ev1 < mddev->events)
 844                         /* just a hot-add of a new device, leave raid_disk at -1 */
 845                         return 0;
 846         }
 847
 848         if (mddev->level != LEVEL_MULTIPATH) {
 849                 desc = sb->disks + rdev->desc_nr;
 850
 851                 if (desc->state & (1<<MD_DISK_FAULTY))
 852                         set_bit(Faulty, &rdev->flags);
 853                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
 854                             desc->raid_disk < mddev->raid_disks */) {
 855                         set_bit(In_sync, &rdev->flags);
 856                         rdev->raid_disk = desc->raid_disk;
 857                 }
 858                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 859                         set_bit(WriteMostly, &rdev->flags);
 860         } else /* MULTIPATH are always insync */
 861                 set_bit(In_sync, &rdev->flags);
 862         return 0;
 863 }
 864
 865 /*
 866  * sync_super for 0.90.0
 867  */
 868 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 869 {
 870         mdp_super_t *sb;
 871         struct list_head *tmp;
 872         mdk_rdev_t *rdev2;
 873         int next_spare = mddev->raid_disks;
 874
 875
 876         /* make rdev->sb match mddev data..
 877          *
 878          * 1/ zero out disks
 879          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 880          * 3/ any empty disks < next_spare become removed
 881          *
 882          * disks[0] gets initialised to REMOVED because
 883          * we cannot be sure from other fields if it has
 884          * been initialised or not.
 885          */
 886         int i;
 887         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 888
 889         rdev->sb_size = MD_SB_BYTES;
 890
 891         sb = (mdp_super_t*)page_address(rdev->sb_page);
 892
 893         memset(sb, 0, sizeof(*sb));
 894
 895         sb->md_magic = MD_SB_MAGIC;
 896         sb->major_version = mddev->major_version;
 897         sb->patch_version = mddev->patch_version;
 898         sb->gvalid_words  = 0; /* ignored */
 899         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 900         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 901         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 902         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 903
 904         sb->ctime = mddev->ctime;
 905         sb->level = mddev->level;
 906         sb->size  = mddev->size;
 907         sb->raid_disks = mddev->raid_disks;
 908         sb->md_minor = mddev->md_minor;
 909         sb->not_persistent = 0;
 910         sb->utime = mddev->utime;
 911         sb->state = 0;
 912         sb->events_hi = (mddev->events>>32);
 913         sb->events_lo = (u32)mddev->events;
 914
 915         if (mddev->reshape_position == MaxSector)
 916                 sb->minor_version = 90;
 917         else {
 918                 sb->minor_version = 91;
 919                 sb->reshape_position = mddev->reshape_position;
 920                 sb->new_level = mddev->new_level;
 921                 sb->delta_disks = mddev->delta_disks;
 922                 sb->new_layout = mddev->new_layout;
 923                 sb->new_chunk = mddev->new_chunk;
 924         }
 925         mddev->minor_version = sb->minor_version;
 926         if (mddev->in_sync)
 927         {
 928                 sb->recovery_cp = mddev->recovery_cp;
 929                 sb->cp_events_hi = (mddev->events>>32);
 930                 sb->cp_events_lo = (u32)mddev->events;
 931                 if (mddev->recovery_cp == MaxSector)
 932                         sb->state = (1<< MD_SB_CLEAN);
 933         } else
 934                 sb->recovery_cp = 0;
 935
 936         sb->layout = mddev->layout;
 937         sb->chunk_size = mddev->chunk_size;
 938
 939         if (mddev->bitmap && mddev->bitmap_file == NULL)
 940                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 941
 942         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 943         rdev_for_each(rdev2, tmp, mddev) {
 944                 mdp_disk_t *d;
 945                 int desc_nr;
 946                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 947                     && !test_bit(Faulty, &rdev2->flags))
 948                         desc_nr = rdev2->raid_disk;
 949                 else
 950                         desc_nr = next_spare++;
 951                 rdev2->desc_nr = desc_nr;
 952                 d = &sb->disks[rdev2->desc_nr];
 953                 nr_disks++;
 954                 d->number = rdev2->desc_nr;
 955                 d->major = MAJOR(rdev2->bdev->bd_dev);
 956                 d->minor = MINOR(rdev2->bdev->bd_dev);
 957                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 958                     && !test_bit(Faulty, &rdev2->flags))
 959                         d->raid_disk = rdev2->raid_disk;
 960                 else
 961                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 962                 if (test_bit(Faulty, &rdev2->flags))
 963                         d->state = (1<<MD_DISK_FAULTY);
 964                 else if (test_bit(In_sync, &rdev2->flags)) {
 965                         d->state = (1<<MD_DISK_ACTIVE);
 966                         d->state |= (1<<MD_DISK_SYNC);
 967                         active++;
 968                         working++;
 969                 } else {
 970                         d->state = 0;
 971                         spare++;
 972                         working++;
 973                 }
 974                 if (test_bit(WriteMostly, &rdev2->flags))
 975                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 976         }
 977         /* now set the "removed" and "faulty" bits on any missing devices */
 978         for (i=0 ; i < mddev->raid_disks ; i++) {
 979                 mdp_disk_t *d = &sb->disks[i];
 980                 if (d->state == 0 && d->number == 0) {
 981                         d->number = i;
 982                         d->raid_disk = i;
 983                         d->state = (1<<MD_DISK_REMOVED);
 984                         d->state |= (1<<MD_DISK_FAULTY);
 985                         failed++;
 986                 }
 987         }
 988         sb->nr_disks = nr_disks;
 989         sb->active_disks = active;
 990         sb->working_disks = working;
 991         sb->failed_disks = failed;
 992         sb->spare_disks = spare;
 993
 994         sb->this_disk = sb->disks[rdev->desc_nr];
 995         sb->sb_csum = calc_sb_csum(sb);
 996 }
 997
 998 /*
 999  * rdev_size_change for 0.90.0
1000  */
1001 static unsigned long long
1002 super_90_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1003 {
1004         if (size && size < rdev->mddev->size)
1005                 return 0; /* component must fit device */
1006         size *= 2; /* convert to sectors */
1007         if (rdev->mddev->bitmap_offset)
1008                 return 0; /* can't move bitmap */
1009         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
1010         if (!size || size > rdev->sb_offset*2)
1011                 size = rdev->sb_offset*2;
1012         md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size,
1013                        rdev->sb_page);
1014         md_super_wait(rdev->mddev);
1015         return size/2; /* kB for sysfs */
1016 }
1017
1018
1019 /*
1020  * version 1 superblock
1021  */
1022
1023 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1024 {
1025         __le32 disk_csum;
1026         u32 csum;
1027         unsigned long long newcsum;
1028         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1029         __le32 *isuper = (__le32*)sb;
1030         int i;
1031
1032         disk_csum = sb->sb_csum;
1033         sb->sb_csum = 0;
1034         newcsum = 0;
1035         for (i=0; size>=4; size -= 4 )
1036                 newcsum += le32_to_cpu(*isuper++);
1037
1038         if (size == 2)
1039                 newcsum += le16_to_cpu(*(__le16*) isuper);
1040
1041         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1042         sb->sb_csum = disk_csum;
1043         return cpu_to_le32(csum);
1044 }
1045
1046 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1047 {
1048         struct mdp_superblock_1 *sb;
1049         int ret;
1050         sector_t sb_offset;
1051         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1052         int bmask;
1053
1054         /*
1055          * Calculate the position of the superblock.
1056          * It is always aligned to a 4K boundary and
1057          * depeding on minor_version, it can be:
1058          * 0: At least 8K, but less than 12K, from end of device
1059          * 1: At start of device
1060          * 2: 4K from start of device.
1061          */
1062         switch(minor_version) {
1063         case 0:
1064                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1065                 sb_offset -= 8*2;
1066                 sb_offset &= ~(sector_t)(4*2-1);
1067                 /* convert from sectors to K */
1068                 sb_offset /= 2;
1069                 break;
1070         case 1:
1071                 sb_offset = 0;
1072                 break;
1073         case 2:
1074                 sb_offset = 4;
1075                 break;
1076         default:
1077                 return -EINVAL;
1078         }
1079         rdev->sb_offset = sb_offset;
1080
1081         /* superblock is rarely larger than 1K, but it can be larger,
1082          * and it is safe to read 4k, so we do that
1083          */
1084         ret = read_disk_sb(rdev, 4096);
1085         if (ret) return ret;
1086
1087
1088         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1089
1090         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1091             sb->major_version != cpu_to_le32(1) ||
1092             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1093             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1094             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1095                 return -EINVAL;
1096
1097         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1098                 printk("md: invalid superblock checksum on %s\n",
1099                         bdevname(rdev->bdev,b));
1100                 return -EINVAL;
1101         }
1102         if (le64_to_cpu(sb->data_size) < 10) {
1103                 printk("md: data_size too small on %s\n",
1104                        bdevname(rdev->bdev,b));
1105                 return -EINVAL;
1106         }
1107         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1108                 if (sb->level != cpu_to_le32(1) &&
1109                     sb->level != cpu_to_le32(4) &&
1110                     sb->level != cpu_to_le32(5) &&
1111                     sb->level != cpu_to_le32(6) &&
1112                     sb->level != cpu_to_le32(10)) {
1113                         printk(KERN_WARNING
1114                                "md: bitmaps not supported for this level.\n");
1115                         return -EINVAL;
1116                 }
1117         }
1118
1119         rdev->preferred_minor = 0xffff;
1120         rdev->data_offset = le64_to_cpu(sb->data_offset);
1121         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1122
1123         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1124         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1125         if (rdev->sb_size & bmask)
1126                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1127
1128         if (minor_version
1129             && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1130                 return -EINVAL;
1131
1132         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1133                 rdev->desc_nr = -1;
1134         else
1135                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1136
1137         if (!refdev) {
1138                 ret = 1;
1139         } else {
1140                 __u64 ev1, ev2;
1141                 struct mdp_superblock_1 *refsb =
1142                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1143
1144                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1145                     sb->level != refsb->level ||
1146                     sb->layout != refsb->layout ||
1147                     sb->chunksize != refsb->chunksize) {
1148                         printk(KERN_WARNING "md: %s has strangely different"
1149                                 " superblock to %s\n",
1150                                 bdevname(rdev->bdev,b),
1151                                 bdevname(refdev->bdev,b2));
1152                         return -EINVAL;
1153                 }
1154                 ev1 = le64_to_cpu(sb->events);
1155                 ev2 = le64_to_cpu(refsb->events);
1156
1157                 if (ev1 > ev2)
1158                         ret = 1;
1159                 else
1160                         ret = 0;
1161         }
1162         if (minor_version)
1163                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1164         else
1165                 rdev->size = rdev->sb_offset;
1166         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1167                 return -EINVAL;
1168         rdev->size = le64_to_cpu(sb->data_size)/2;
1169         if (le32_to_cpu(sb->chunksize))
1170                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1171
1172         if (le64_to_cpu(sb->size) > rdev->size*2)
1173                 return -EINVAL;
1174         return ret;
1175 }
1176
1177 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1178 {
1179         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1180         __u64 ev1 = le64_to_cpu(sb->events);
1181
1182         rdev->raid_disk = -1;
1183         clear_bit(Faulty, &rdev->flags);
1184         clear_bit(In_sync, &rdev->flags);
1185         clear_bit(WriteMostly, &rdev->flags);
1186         clear_bit(BarriersNotsupp, &rdev->flags);
1187
1188         if (mddev->raid_disks == 0) {
1189                 mddev->major_version = 1;
1190                 mddev->patch_version = 0;
1191                 mddev->external = 0;
1192                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1193                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1194                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1195                 mddev->level = le32_to_cpu(sb->level);
1196                 mddev->clevel[0] = 0;
1197                 mddev->layout = le32_to_cpu(sb->layout);
1198                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1199                 mddev->size = le64_to_cpu(sb->size)/2;
1200                 mddev->events = ev1;
1201                 mddev->bitmap_offset = 0;
1202                 mddev->default_bitmap_offset = 1024 >> 9;
1203
1204                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1205                 memcpy(mddev->uuid, sb->set_uuid, 16);
1206
1207                 mddev->max_disks =  (4096-256)/2;
1208
1209                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1210                     mddev->bitmap_file == NULL )
1211                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1212
1213                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1214                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1215                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1216                         mddev->new_level = le32_to_cpu(sb->new_level);
1217                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1218                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1219                 } else {
1220                         mddev->reshape_position = MaxSector;
1221                         mddev->delta_disks = 0;
1222                         mddev->new_level = mddev->level;
1223                         mddev->new_layout = mddev->layout;
1224                         mddev->new_chunk = mddev->chunk_size;
1225                 }
1226
1227         } else if (mddev->pers == NULL) {
1228                 /* Insist of good event counter while assembling */
1229                 ++ev1;
1230                 if (ev1 < mddev->events)
1231                         return -EINVAL;
1232         } else if (mddev->bitmap) {
1233                 /* If adding to array with a bitmap, then we can accept an
1234                  * older device, but not too old.
1235                  */
1236                 if (ev1 < mddev->bitmap->events_cleared)
1237                         return 0;
1238         } else {
1239                 if (ev1 < mddev->events)
1240                         /* just a hot-add of a new device, leave raid_disk at -1 */
1241                         return 0;
1242         }
1243         if (mddev->level != LEVEL_MULTIPATH) {
1244                 int role;
1245                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1246                 switch(role) {
1247                 case 0xffff: /* spare */
1248                         break;
1249                 case 0xfffe: /* faulty */
1250                         set_bit(Faulty, &rdev->flags);
1251                         break;
1252                 default:
1253                         if ((le32_to_cpu(sb->feature_map) &
1254                              MD_FEATURE_RECOVERY_OFFSET))
1255                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1256                         else
1257                                 set_bit(In_sync, &rdev->flags);
1258                         rdev->raid_disk = role;
1259                         break;
1260                 }
1261                 if (sb->devflags & WriteMostly1)
1262                         set_bit(WriteMostly, &rdev->flags);
1263         } else /* MULTIPATH are always insync */
1264                 set_bit(In_sync, &rdev->flags);
1265
1266         return 0;
1267 }
1268
1269 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1270 {
1271         struct mdp_superblock_1 *sb;
1272         struct list_head *tmp;
1273         mdk_rdev_t *rdev2;
1274         int max_dev, i;
1275         /* make rdev->sb match mddev and rdev data. */
1276
1277         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1278
1279         sb->feature_map = 0;
1280         sb->pad0 = 0;
1281         sb->recovery_offset = cpu_to_le64(0);
1282         memset(sb->pad1, 0, sizeof(sb->pad1));
1283         memset(sb->pad2, 0, sizeof(sb->pad2));
1284         memset(sb->pad3, 0, sizeof(sb->pad3));
1285
1286         sb->utime = cpu_to_le64((__u64)mddev->utime);
1287         sb->events = cpu_to_le64(mddev->events);
1288         if (mddev->in_sync)
1289                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1290         else
1291                 sb->resync_offset = cpu_to_le64(0);
1292
1293         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1294
1295         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1296         sb->size = cpu_to_le64(mddev->size<<1);
1297
1298         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1299                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1300                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1301         }
1302
1303         if (rdev->raid_disk >= 0 &&
1304             !test_bit(In_sync, &rdev->flags) &&
1305             rdev->recovery_offset > 0) {
1306                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1307                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1308         }
1309
1310         if (mddev->reshape_position != MaxSector) {
1311                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1312                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1313                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1314                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1315                 sb->new_level = cpu_to_le32(mddev->new_level);
1316                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1317         }
1318
1319         max_dev = 0;
1320         rdev_for_each(rdev2, tmp, mddev)
1321                 if (rdev2->desc_nr+1 > max_dev)
1322                         max_dev = rdev2->desc_nr+1;
1323
1324         if (max_dev > le32_to_cpu(sb->max_dev))
1325                 sb->max_dev = cpu_to_le32(max_dev);
1326         for (i=0; i<max_dev;i++)
1327                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1328
1329         rdev_for_each(rdev2, tmp, mddev) {
1330                 i = rdev2->desc_nr;
1331                 if (test_bit(Faulty, &rdev2->flags))
1332                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1333                 else if (test_bit(In_sync, &rdev2->flags))
1334                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1335                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1336                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1337                 else
1338                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1339         }
1340
1341         sb->sb_csum = calc_sb_1_csum(sb);
1342 }
1343
1344 static unsigned long long
1345 super_1_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1346 {
1347         struct mdp_superblock_1 *sb;
1348         unsigned long long max_size;
1349         if (size && size < rdev->mddev->size)
1350                 return 0; /* component must fit device */
1351         size *= 2; /* convert to sectors */
1352         if (rdev->sb_offset < rdev->data_offset/2) {
1353                 /* minor versions 1 and 2; superblock before data */
1354                 max_size = (rdev->bdev->bd_inode->i_size >> 9);
1355                 max_size -= rdev->data_offset;
1356                 if (!size || size > max_size)
1357                         size = max_size;
1358         } else if (rdev->mddev->bitmap_offset) {
1359                 /* minor version 0 with bitmap we can't move */
1360                 return 0;
1361         } else {
1362                 /* minor version 0; superblock after data */
1363                 sector_t sb_offset;
1364                 sb_offset = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1365                 sb_offset &= ~(sector_t)(4*2 - 1);
1366                 max_size = rdev->size*2 + sb_offset - rdev->sb_offset*2;
1367                 if (!size || size > max_size)
1368                         size = max_size;
1369                 rdev->sb_offset = sb_offset/2;
1370         }
1371         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1372         sb->data_size = cpu_to_le64(size);
1373         sb->super_offset = rdev->sb_offset*2;
1374         sb->sb_csum = calc_sb_1_csum(sb);
1375         md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size,
1376                        rdev->sb_page);
1377         md_super_wait(rdev->mddev);
1378         return size/2; /* kB for sysfs */
1379 }
1380
1381 static struct super_type super_types[] = {
1382         [0] = {
1383                 .name   = "0.90.0",
1384                 .owner  = THIS_MODULE,
1385                 .load_super         = super_90_load,
1386                 .validate_super     = super_90_validate,
1387                 .sync_super         = super_90_sync,
1388                 .rdev_size_change   = super_90_rdev_size_change,
1389         },
1390         [1] = {
1391                 .name   = "md-1",
1392                 .owner  = THIS_MODULE,
1393                 .load_super         = super_1_load,
1394                 .validate_super     = super_1_validate,
1395                 .sync_super         = super_1_sync,
1396                 .rdev_size_change   = super_1_rdev_size_change,
1397         },
1398 };
1399
1400 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1401 {
1402         struct list_head *tmp, *tmp2;
1403         mdk_rdev_t *rdev, *rdev2;
1404
1405         rdev_for_each(rdev, tmp, mddev1)
1406                 rdev_for_each(rdev2, tmp2, mddev2)
1407                         if (rdev->bdev->bd_contains ==
1408                             rdev2->bdev->bd_contains)
1409                                 return 1;
1410
1411         return 0;
1412 }
1413
1414 static LIST_HEAD(pending_raid_disks);
1415
1416 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1417 {
1418         char b[BDEVNAME_SIZE];
1419         struct kobject *ko;
1420         char *s;
1421         int err;
1422
1423         if (rdev->mddev) {
1424                 MD_BUG();
1425                 return -EINVAL;
1426         }
1427
1428         /* prevent duplicates */
1429         if (find_rdev(mddev, rdev->bdev->bd_dev))
1430                 return -EEXIST;
1431
1432         /* make sure rdev->size exceeds mddev->size */
1433         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1434                 if (mddev->pers) {
1435                         /* Cannot change size, so fail
1436                          * If mddev->level <= 0, then we don't care
1437                          * about aligning sizes (e.g. linear)
1438                          */
1439                         if (mddev->level > 0)
1440                                 return -ENOSPC;
1441                 } else
1442                         mddev->size = rdev->size;
1443         }
1444
1445         /* Verify rdev->desc_nr is unique.
1446          * If it is -1, assign a free number, else
1447          * check number is not in use
1448          */
1449         if (rdev->desc_nr < 0) {
1450                 int choice = 0;
1451                 if (mddev->pers) choice = mddev->raid_disks;
1452                 while (find_rdev_nr(mddev, choice))
1453                         choice++;
1454                 rdev->desc_nr = choice;
1455         } else {
1456                 if (find_rdev_nr(mddev, rdev->desc_nr))
1457                         return -EBUSY;
1458         }
1459         bdevname(rdev->bdev,b);
1460         while ( (s=strchr(b, '/')) != NULL)
1461                 *s = '!';
1462
1463         rdev->mddev = mddev;
1464         printk(KERN_INFO "md: bind<%s>\n", b);
1465
1466         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1467                 goto fail;
1468
1469         if (rdev->bdev->bd_part)
1470                 ko = &rdev->bdev->bd_part->dev.kobj;
1471         else
1472                 ko = &rdev->bdev->bd_disk->dev.kobj;
1473         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1474                 kobject_del(&rdev->kobj);
1475                 goto fail;
1476         }
1477         list_add(&rdev->same_set, &mddev->disks);
1478         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1479         return 0;
1480
1481  fail:
1482         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1483                b, mdname(mddev));
1484         return err;
1485 }
1486
1487 static void md_delayed_delete(struct work_struct *ws)
1488 {
1489         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1490         kobject_del(&rdev->kobj);
1491         kobject_put(&rdev->kobj);
1492 }
1493
1494 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1495 {
1496         char b[BDEVNAME_SIZE];
1497         if (!rdev->mddev) {
1498                 MD_BUG();
1499                 return;
1500         }
1501         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1502         list_del_init(&rdev->same_set);
1503         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1504         rdev->mddev = NULL;
1505         sysfs_remove_link(&rdev->kobj, "block");
1506
1507         /* We need to delay this, otherwise we can deadlock when
1508          * writing to 'remove' to "dev/state"
1509          */
1510         INIT_WORK(&rdev->del_work, md_delayed_delete);
1511         kobject_get(&rdev->kobj);
1512         schedule_work(&rdev->del_work);
1513 }
1514
1515 /*
1516  * prevent the device from being mounted, repartitioned or
1517  * otherwise reused by a RAID array (or any other kernel
1518  * subsystem), by bd_claiming the device.
1519  */
1520 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1521 {
1522         int err = 0;
1523         struct block_device *bdev;
1524         char b[BDEVNAME_SIZE];
1525
1526         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1527         if (IS_ERR(bdev)) {
1528                 printk(KERN_ERR "md: could not open %s.\n",
1529                         __bdevname(dev, b));
1530                 return PTR_ERR(bdev);
1531         }
1532         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1533         if (err) {
1534                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1535                         bdevname(bdev, b));
1536                 blkdev_put(bdev);
1537                 return err;
1538         }
1539         if (!shared)
1540                 set_bit(AllReserved, &rdev->flags);
1541         rdev->bdev = bdev;
1542         return err;
1543 }
1544
1545 static void unlock_rdev(mdk_rdev_t *rdev)
1546 {
1547         struct block_device *bdev = rdev->bdev;
1548         rdev->bdev = NULL;
1549         if (!bdev)
1550                 MD_BUG();
1551         bd_release(bdev);
1552         blkdev_put(bdev);
1553 }
1554
1555 void md_autodetect_dev(dev_t dev);
1556
1557 static void export_rdev(mdk_rdev_t * rdev)
1558 {
1559         char b[BDEVNAME_SIZE];
1560         printk(KERN_INFO "md: export_rdev(%s)\n",
1561                 bdevname(rdev->bdev,b));
1562         if (rdev->mddev)
1563                 MD_BUG();
1564         free_disk_sb(rdev);
1565         list_del_init(&rdev->same_set);
1566 #ifndef MODULE
1567         if (test_bit(AutoDetected, &rdev->flags))
1568                 md_autodetect_dev(rdev->bdev->bd_dev);
1569 #endif
1570         unlock_rdev(rdev);
1571         kobject_put(&rdev->kobj);
1572 }
1573
1574 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1575 {
1576         unbind_rdev_from_array(rdev);
1577         export_rdev(rdev);
1578 }
1579
1580 static void export_array(mddev_t *mddev)
1581 {
1582         struct list_head *tmp;
1583         mdk_rdev_t *rdev;
1584
1585         rdev_for_each(rdev, tmp, mddev) {
1586                 if (!rdev->mddev) {
1587                         MD_BUG();
1588                         continue;
1589                 }
1590                 kick_rdev_from_array(rdev);
1591         }
1592         if (!list_empty(&mddev->disks))
1593                 MD_BUG();
1594         mddev->raid_disks = 0;
1595         mddev->major_version = 0;
1596 }
1597
1598 static void print_desc(mdp_disk_t *desc)
1599 {
1600         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1601                 desc->major,desc->minor,desc->raid_disk,desc->state);
1602 }
1603
1604 static void print_sb(mdp_super_t *sb)
1605 {
1606         int i;
1607
1608         printk(KERN_INFO
1609                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1610                 sb->major_version, sb->minor_version, sb->patch_version,
1611                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1612                 sb->ctime);
1613         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1614                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1615                 sb->md_minor, sb->layout, sb->chunk_size);
1616         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1617                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1618                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1619                 sb->failed_disks, sb->spare_disks,
1620                 sb->sb_csum, (unsigned long)sb->events_lo);
1621
1622         printk(KERN_INFO);
1623         for (i = 0; i < MD_SB_DISKS; i++) {
1624                 mdp_disk_t *desc;
1625
1626                 desc = sb->disks + i;
1627                 if (desc->number || desc->major || desc->minor ||
1628                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1629                         printk("     D %2d: ", i);
1630                         print_desc(desc);
1631                 }
1632         }
1633         printk(KERN_INFO "md:     THIS: ");
1634         print_desc(&sb->this_disk);
1635
1636 }
1637
1638 static void print_rdev(mdk_rdev_t *rdev)
1639 {
1640         char b[BDEVNAME_SIZE];
1641         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1642                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1643                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1644                 rdev->desc_nr);
1645         if (rdev->sb_loaded) {
1646                 printk(KERN_INFO "md: rdev superblock:\n");
1647                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1648         } else
1649                 printk(KERN_INFO "md: no rdev superblock!\n");
1650 }
1651
1652 static void md_print_devices(void)
1653 {
1654         struct list_head *tmp, *tmp2;
1655         mdk_rdev_t *rdev;
1656         mddev_t *mddev;
1657         char b[BDEVNAME_SIZE];
1658
1659         printk("\n");
1660         printk("md:     **********************************\n");
1661         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1662         printk("md:     **********************************\n");
1663         for_each_mddev(mddev, tmp) {
1664
1665                 if (mddev->bitmap)
1666                         bitmap_print_sb(mddev->bitmap);
1667                 else
1668                         printk("%s: ", mdname(mddev));
1669                 rdev_for_each(rdev, tmp2, mddev)
1670                         printk("<%s>", bdevname(rdev->bdev,b));
1671                 printk("\n");
1672
1673                 rdev_for_each(rdev, tmp2, mddev)
1674                         print_rdev(rdev);
1675         }
1676         printk("md:     **********************************\n");
1677         printk("\n");
1678 }
1679
1680
1681 static void sync_sbs(mddev_t * mddev, int nospares)
1682 {
1683         /* Update each superblock (in-memory image), but
1684          * if we are allowed to, skip spares which already
1685          * have the right event counter, or have one earlier
1686          * (which would mean they aren't being marked as dirty
1687          * with the rest of the array)
1688          */
1689         mdk_rdev_t *rdev;
1690         struct list_head *tmp;
1691
1692         rdev_for_each(rdev, tmp, mddev) {
1693                 if (rdev->sb_events == mddev->events ||
1694                     (nospares &&
1695                      rdev->raid_disk < 0 &&
1696                      (rdev->sb_events&1)==0 &&
1697                      rdev->sb_events+1 == mddev->events)) {
1698                         /* Don't update this superblock */
1699                         rdev->sb_loaded = 2;
1700                 } else {
1701                         super_types[mddev->major_version].
1702                                 sync_super(mddev, rdev);
1703                         rdev->sb_loaded = 1;
1704                 }
1705         }
1706 }
1707
1708 static void md_update_sb(mddev_t * mddev, int force_change)
1709 {
1710         struct list_head *tmp;
1711         mdk_rdev_t *rdev;
1712         int sync_req;
1713         int nospares = 0;
1714
1715         if (mddev->external)
1716                 return;
1717 repeat:
1718         spin_lock_irq(&mddev->write_lock);
1719
1720         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1721         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1722                 force_change = 1;
1723         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1724                 /* just a clean<-> dirty transition, possibly leave spares alone,
1725                  * though if events isn't the right even/odd, we will have to do
1726                  * spares after all
1727                  */
1728                 nospares = 1;
1729         if (force_change)
1730                 nospares = 0;
1731         if (mddev->degraded)
1732                 /* If the array is degraded, then skipping spares is both
1733                  * dangerous and fairly pointless.
1734                  * Dangerous because a device that was removed from the array
1735                  * might have a event_count that still looks up-to-date,
1736                  * so it can be re-added without a resync.
1737                  * Pointless because if there are any spares to skip,
1738                  * then a recovery will happen and soon that array won't
1739                  * be degraded any more and the spare can go back to sleep then.
1740                  */
1741                 nospares = 0;
1742
1743         sync_req = mddev->in_sync;
1744         mddev->utime = get_seconds();
1745
1746         /* If this is just a dirty<->clean transition, and the array is clean
1747          * and 'events' is odd, we can roll back to the previous clean state */
1748         if (nospares
1749             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1750             && (mddev->events & 1)
1751             && mddev->events != 1)
1752                 mddev->events--;
1753         else {
1754                 /* otherwise we have to go forward and ... */
1755                 mddev->events ++;
1756                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1757                         /* .. if the array isn't clean, insist on an odd 'events' */
1758                         if ((mddev->events&1)==0) {
1759                                 mddev->events++;
1760                                 nospares = 0;
1761                         }
1762                 } else {
1763                         /* otherwise insist on an even 'events' (for clean states) */
1764                         if ((mddev->events&1)) {
1765                                 mddev->events++;
1766                                 nospares = 0;
1767                         }
1768                 }
1769         }
1770
1771         if (!mddev->events) {
1772                 /*
1773                  * oops, this 64-bit counter should never wrap.
1774                  * Either we are in around ~1 trillion A.C., assuming
1775                  * 1 reboot per second, or we have a bug:
1776                  */
1777                 MD_BUG();
1778                 mddev->events --;
1779         }
1780
1781         /*
1782          * do not write anything to disk if using
1783          * nonpersistent superblocks
1784          */
1785         if (!mddev->persistent) {
1786                 if (!mddev->external)
1787                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1788
1789                 spin_unlock_irq(&mddev->write_lock);
1790                 wake_up(&mddev->sb_wait);
1791                 return;
1792         }
1793         sync_sbs(mddev, nospares);
1794         spin_unlock_irq(&mddev->write_lock);
1795
1796         dprintk(KERN_INFO
1797                 "md: updating %s RAID superblock on device (in sync %d)\n",
1798                 mdname(mddev),mddev->in_sync);
1799
1800         bitmap_update_sb(mddev->bitmap);
1801         rdev_for_each(rdev, tmp, mddev) {
1802                 char b[BDEVNAME_SIZE];
1803                 dprintk(KERN_INFO "md: ");
1804                 if (rdev->sb_loaded != 1)
1805                         continue; /* no noise on spare devices */
1806                 if (test_bit(Faulty, &rdev->flags))
1807                         dprintk("(skipping faulty ");
1808
1809                 dprintk("%s ", bdevname(rdev->bdev,b));
1810                 if (!test_bit(Faulty, &rdev->flags)) {
1811                         md_super_write(mddev,rdev,
1812                                        rdev->sb_offset<<1, rdev->sb_size,
1813                                        rdev->sb_page);
1814                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1815                                 bdevname(rdev->bdev,b),
1816                                 (unsigned long long)rdev->sb_offset);
1817                         rdev->sb_events = mddev->events;
1818
1819                 } else
1820                         dprintk(")\n");
1821                 if (mddev->level == LEVEL_MULTIPATH)
1822                         /* only need to write one superblock... */
1823                         break;
1824         }
1825         md_super_wait(mddev);
1826         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1827
1828         spin_lock_irq(&mddev->write_lock);
1829         if (mddev->in_sync != sync_req ||
1830             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1831                 /* have to write it out again */
1832                 spin_unlock_irq(&mddev->write_lock);
1833                 goto repeat;
1834         }
1835         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1836         spin_unlock_irq(&mddev->write_lock);
1837         wake_up(&mddev->sb_wait);
1838
1839 }
1840
1841 /* words written to sysfs files may, or may not, be \n terminated.
1842  * We want to accept with case. For this we use cmd_match.
1843  */
1844 static int cmd_match(const char *cmd, const char *str)
1845 {
1846         /* See if cmd, written into a sysfs file, matches
1847          * str.  They must either be the same, or cmd can
1848          * have a trailing newline
1849          */
1850         while (*cmd && *str && *cmd == *str) {
1851                 cmd++;
1852                 str++;
1853         }
1854         if (*cmd == '\n')
1855                 cmd++;
1856         if (*str || *cmd)
1857                 return 0;
1858         return 1;
1859 }
1860
1861 struct rdev_sysfs_entry {
1862         struct attribute attr;
1863         ssize_t (*show)(mdk_rdev_t *, char *);
1864         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1865 };
1866
1867 static ssize_t
1868 state_show(mdk_rdev_t *rdev, char *page)
1869 {
1870         char *sep = "";
1871         size_t len = 0;
1872
1873         if (test_bit(Faulty, &rdev->flags)) {
1874                 len+= sprintf(page+len, "%sfaulty",sep);
1875                 sep = ",";
1876         }
1877         if (test_bit(In_sync, &rdev->flags)) {
1878                 len += sprintf(page+len, "%sin_sync",sep);
1879                 sep = ",";
1880         }
1881         if (test_bit(WriteMostly, &rdev->flags)) {
1882                 len += sprintf(page+len, "%swrite_mostly",sep);
1883                 sep = ",";
1884         }
1885         if (test_bit(Blocked, &rdev->flags)) {
1886                 len += sprintf(page+len, "%sblocked", sep);
1887                 sep = ",";
1888         }
1889         if (!test_bit(Faulty, &rdev->flags) &&
1890             !test_bit(In_sync, &rdev->flags)) {
1891                 len += sprintf(page+len, "%sspare", sep);
1892                 sep = ",";
1893         }
1894         return len+sprintf(page+len, "\n");
1895 }
1896
1897 static ssize_t
1898 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1899 {
1900         /* can write
1901          *  faulty  - simulates and error
1902          *  remove  - disconnects the device
1903          *  writemostly - sets write_mostly
1904          *  -writemostly - clears write_mostly
1905          *  blocked - sets the Blocked flag
1906          *  -blocked - clears the Blocked flag
1907          */
1908         int err = -EINVAL;
1909         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1910                 md_error(rdev->mddev, rdev);
1911                 err = 0;
1912         } else if (cmd_match(buf, "remove")) {
1913                 if (rdev->raid_disk >= 0)
1914                         err = -EBUSY;
1915                 else {
1916                         mddev_t *mddev = rdev->mddev;
1917                         kick_rdev_from_array(rdev);
1918                         if (mddev->pers)
1919                                 md_update_sb(mddev, 1);
1920                         md_new_event(mddev);
1921                         err = 0;
1922                 }
1923         } else if (cmd_match(buf, "writemostly")) {
1924                 set_bit(WriteMostly, &rdev->flags);
1925                 err = 0;
1926         } else if (cmd_match(buf, "-writemostly")) {
1927                 clear_bit(WriteMostly, &rdev->flags);
1928                 err = 0;
1929         } else if (cmd_match(buf, "blocked")) {
1930                 set_bit(Blocked, &rdev->flags);
1931                 err = 0;
1932         } else if (cmd_match(buf, "-blocked")) {
1933                 clear_bit(Blocked, &rdev->flags);
1934                 wake_up(&rdev->blocked_wait);
1935                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1936                 md_wakeup_thread(rdev->mddev->thread);
1937
1938                 err = 0;
1939         }
1940         if (!err)
1941                 sysfs_notify(&rdev->kobj, NULL, "state");
1942         return err ? err : len;
1943 }
1944 static struct rdev_sysfs_entry rdev_state =
1945 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1946
1947 static ssize_t
1948 errors_show(mdk_rdev_t *rdev, char *page)
1949 {
1950         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1951 }
1952
1953 static ssize_t
1954 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1955 {
1956         char *e;
1957         unsigned long n = simple_strtoul(buf, &e, 10);
1958         if (*buf && (*e == 0 || *e == '\n')) {
1959                 atomic_set(&rdev->corrected_errors, n);
1960                 return len;
1961         }
1962         return -EINVAL;
1963 }
1964 static struct rdev_sysfs_entry rdev_errors =
1965 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1966
1967 static ssize_t
1968 slot_show(mdk_rdev_t *rdev, char *page)
1969 {
1970         if (rdev->raid_disk < 0)
1971                 return sprintf(page, "none\n");
1972         else
1973                 return sprintf(page, "%d\n", rdev->raid_disk);
1974 }
1975
1976 static ssize_t
1977 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1978 {
1979         char *e;
1980         int err;
1981         char nm[20];
1982         int slot = simple_strtoul(buf, &e, 10);
1983         if (strncmp(buf, "none", 4)==0)
1984                 slot = -1;
1985         else if (e==buf || (*e && *e!= '\n'))
1986                 return -EINVAL;
1987         if (rdev->mddev->pers && slot == -1) {
1988                 /* Setting 'slot' on an active array requires also
1989                  * updating the 'rd%d' link, and communicating
1990                  * with the personality with ->hot_*_disk.
1991                  * For now we only support removing
1992                  * failed/spare devices.  This normally happens automatically,
1993                  * but not when the metadata is externally managed.
1994                  */
1995                 if (rdev->raid_disk == -1)
1996                         return -EEXIST;
1997                 /* personality does all needed checks */
1998                 if (rdev->mddev->pers->hot_add_disk == NULL)
1999                         return -EINVAL;
2000                 err = rdev->mddev->pers->
2001                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2002                 if (err)
2003                         return err;
2004                 sprintf(nm, "rd%d", rdev->raid_disk);
2005                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2006                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2007                 md_wakeup_thread(rdev->mddev->thread);
2008         } else if (rdev->mddev->pers) {
2009                 mdk_rdev_t *rdev2;
2010                 struct list_head *tmp;
2011                 /* Activating a spare .. or possibly reactivating
2012                  * if we every get bitmaps working here.
2013                  */
2014
2015                 if (rdev->raid_disk != -1)
2016                         return -EBUSY;
2017
2018                 if (rdev->mddev->pers->hot_add_disk == NULL)
2019                         return -EINVAL;
2020
2021                 rdev_for_each(rdev2, tmp, rdev->mddev)
2022                         if (rdev2->raid_disk == slot)
2023                                 return -EEXIST;
2024
2025                 rdev->raid_disk = slot;
2026                 if (test_bit(In_sync, &rdev->flags))
2027                         rdev->saved_raid_disk = slot;
2028                 else
2029                         rdev->saved_raid_disk = -1;
2030                 err = rdev->mddev->pers->
2031                         hot_add_disk(rdev->mddev, rdev);
2032                 if (err) {
2033                         rdev->raid_disk = -1;
2034                         return err;
2035                 } else
2036                         sysfs_notify(&rdev->kobj, NULL, "state");
2037                 sprintf(nm, "rd%d", rdev->raid_disk);
2038                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2039                         printk(KERN_WARNING
2040                                "md: cannot register "
2041                                "%s for %s\n",
2042                                nm, mdname(rdev->mddev));
2043
2044                 /* don't wakeup anyone, leave that to userspace. */
2045         } else {
2046                 if (slot >= rdev->mddev->raid_disks)
2047                         return -ENOSPC;
2048                 rdev->raid_disk = slot;
2049                 /* assume it is working */
2050                 clear_bit(Faulty, &rdev->flags);
2051                 clear_bit(WriteMostly, &rdev->flags);
2052                 set_bit(In_sync, &rdev->flags);
2053                 sysfs_notify(&rdev->kobj, NULL, "state");
2054         }
2055         return len;
2056 }
2057
2058
2059 static struct rdev_sysfs_entry rdev_slot =
2060 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2061
2062 static ssize_t
2063 offset_show(mdk_rdev_t *rdev, char *page)
2064 {
2065         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2066 }
2067
2068 static ssize_t
2069 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2070 {
2071         char *e;
2072         unsigned long long offset = simple_strtoull(buf, &e, 10);
2073         if (e==buf || (*e && *e != '\n'))
2074                 return -EINVAL;
2075         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2076                 return -EBUSY;
2077         if (rdev->size && rdev->mddev->external)
2078                 /* Must set offset before size, so overlap checks
2079                  * can be sane */
2080                 return -EBUSY;
2081         rdev->data_offset = offset;
2082         return len;
2083 }
2084
2085 static struct rdev_sysfs_entry rdev_offset =
2086 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2087
2088 static ssize_t
2089 rdev_size_show(mdk_rdev_t *rdev, char *page)
2090 {
2091         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2092 }
2093
2094 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2095 {
2096         /* check if two start/length pairs overlap */
2097         if (s1+l1 <= s2)
2098                 return 0;
2099         if (s2+l2 <= s1)
2100                 return 0;
2101         return 1;
2102 }
2103
2104 static ssize_t
2105 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2106 {
2107         char *e;
2108         unsigned long long size = simple_strtoull(buf, &e, 10);
2109         unsigned long long oldsize = rdev->size;
2110         mddev_t *my_mddev = rdev->mddev;
2111
2112         if (e==buf || (*e && *e != '\n'))
2113                 return -EINVAL;
2114         if (my_mddev->pers && rdev->raid_disk >= 0) {
2115                 if (rdev->mddev->persistent) {
2116                         size = super_types[rdev->mddev->major_version].
2117                                 rdev_size_change(rdev, size);
2118                         if (!size)
2119                                 return -EBUSY;
2120                 } else if (!size) {
2121                         size = (rdev->bdev->bd_inode->i_size >> 10);
2122                         size -= rdev->data_offset/2;
2123                 }
2124                 if (size < rdev->mddev->size)
2125                         return -EINVAL; /* component must fit device */
2126         }
2127
2128         rdev->size = size;
2129         if (size > oldsize && rdev->mddev->external) {
2130                 /* need to check that all other rdevs with the same ->bdev
2131                  * do not overlap.  We need to unlock the mddev to avoid
2132                  * a deadlock.  We have already changed rdev->size, and if
2133                  * we have to change it back, we will have the lock again.
2134                  */
2135                 mddev_t *mddev;
2136                 int overlap = 0;
2137                 struct list_head *tmp, *tmp2;
2138
2139                 mddev_unlock(my_mddev);
2140                 for_each_mddev(mddev, tmp) {
2141                         mdk_rdev_t *rdev2;
2142
2143                         mddev_lock(mddev);
2144                         rdev_for_each(rdev2, tmp2, mddev)
2145                                 if (test_bit(AllReserved, &rdev2->flags) ||
2146                                     (rdev->bdev == rdev2->bdev &&
2147                                      rdev != rdev2 &&
2148                                      overlaps(rdev->data_offset, rdev->size,
2149                                             rdev2->data_offset, rdev2->size))) {
2150                                         overlap = 1;
2151                                         break;
2152                                 }
2153                         mddev_unlock(mddev);
2154                         if (overlap) {
2155                                 mddev_put(mddev);
2156                                 break;
2157                         }
2158                 }
2159                 mddev_lock(my_mddev);
2160                 if (overlap) {
2161                         /* Someone else could have slipped in a size
2162                          * change here, but doing so is just silly.
2163                          * We put oldsize back because we *know* it is
2164                          * safe, and trust userspace not to race with
2165                          * itself
2166                          */
2167                         rdev->size = oldsize;
2168                         return -EBUSY;
2169                 }
2170         }
2171         if (size < my_mddev->size || my_mddev->size == 0)
2172                 my_mddev->size = size;
2173         return len;
2174 }
2175
2176 static struct rdev_sysfs_entry rdev_size =
2177 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2178
2179 static struct attribute *rdev_default_attrs[] = {
2180         &rdev_state.attr,
2181         &rdev_errors.attr,
2182         &rdev_slot.attr,
2183         &rdev_offset.attr,
2184         &rdev_size.attr,
2185         NULL,
2186 };
2187 static ssize_t
2188 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2189 {
2190         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2191         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2192         mddev_t *mddev = rdev->mddev;
2193         ssize_t rv;
2194
2195         if (!entry->show)
2196                 return -EIO;
2197
2198         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2199         if (!rv) {
2200                 if (rdev->mddev == NULL)
2201                         rv = -EBUSY;
2202                 else
2203                         rv = entry->show(rdev, page);
2204                 mddev_unlock(mddev);
2205         }
2206         return rv;
2207 }
2208
2209 static ssize_t
2210 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2211               const char *page, size_t length)
2212 {
2213         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2214         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2215         ssize_t rv;
2216         mddev_t *mddev = rdev->mddev;
2217
2218         if (!entry->store)
2219                 return -EIO;
2220         if (!capable(CAP_SYS_ADMIN))
2221                 return -EACCES;
2222         rv = mddev ? mddev_lock(mddev): -EBUSY;
2223         if (!rv) {
2224                 if (rdev->mddev == NULL)
2225                         rv = -EBUSY;
2226                 else
2227                         rv = entry->store(rdev, page, length);
2228                 mddev_unlock(mddev);
2229         }
2230         return rv;
2231 }
2232
2233 static void rdev_free(struct kobject *ko)
2234 {
2235         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2236         kfree(rdev);
2237 }
2238 static struct sysfs_ops rdev_sysfs_ops = {
2239         .show           = rdev_attr_show,
2240         .store          = rdev_attr_store,
2241 };
2242 static struct kobj_type rdev_ktype = {
2243         .release        = rdev_free,
2244         .sysfs_ops      = &rdev_sysfs_ops,
2245         .default_attrs  = rdev_default_attrs,
2246 };
2247
2248 /*
2249  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2250  *
2251  * mark the device faulty if:
2252  *
2253  *   - the device is nonexistent (zero size)
2254  *   - the device has no valid superblock
2255  *
2256  * a faulty rdev _never_ has rdev->sb set.
2257  */
2258 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2259 {
2260         char b[BDEVNAME_SIZE];
2261         int err;
2262         mdk_rdev_t *rdev;
2263         sector_t size;
2264
2265         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2266         if (!rdev) {
2267                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2268                 return ERR_PTR(-ENOMEM);
2269         }
2270
2271         if ((err = alloc_disk_sb(rdev)))
2272                 goto abort_free;
2273
2274         err = lock_rdev(rdev, newdev, super_format == -2);
2275         if (err)
2276                 goto abort_free;
2277
2278         kobject_init(&rdev->kobj, &rdev_ktype);
2279
2280         rdev->desc_nr = -1;
2281         rdev->saved_raid_disk = -1;
2282         rdev->raid_disk = -1;
2283         rdev->flags = 0;
2284         rdev->data_offset = 0;
2285         rdev->sb_events = 0;
2286         atomic_set(&rdev->nr_pending, 0);
2287         atomic_set(&rdev->read_errors, 0);
2288         atomic_set(&rdev->corrected_errors, 0);
2289
2290         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2291         if (!size) {
2292                 printk(KERN_WARNING
2293                         "md: %s has zero or unknown size, marking faulty!\n",
2294                         bdevname(rdev->bdev,b));
2295                 err = -EINVAL;
2296                 goto abort_free;
2297         }
2298
2299         if (super_format >= 0) {
2300                 err = super_types[super_format].
2301                         load_super(rdev, NULL, super_minor);
2302                 if (err == -EINVAL) {
2303                         printk(KERN_WARNING
2304                                 "md: %s does not have a valid v%d.%d "
2305                                "superblock, not importing!\n",
2306                                 bdevname(rdev->bdev,b),
2307                                super_format, super_minor);
2308                         goto abort_free;
2309                 }
2310                 if (err < 0) {
2311                         printk(KERN_WARNING
2312                                 "md: could not read %s's sb, not importing!\n",
2313                                 bdevname(rdev->bdev,b));
2314                         goto abort_free;
2315                 }
2316         }
2317
2318         INIT_LIST_HEAD(&rdev->same_set);
2319         init_waitqueue_head(&rdev->blocked_wait);
2320
2321         return rdev;
2322
2323 abort_free:
2324         if (rdev->sb_page) {
2325                 if (rdev->bdev)
2326                         unlock_rdev(rdev);
2327                 free_disk_sb(rdev);
2328         }
2329         kfree(rdev);
2330         return ERR_PTR(err);
2331 }
2332
2333 /*
2334  * Check a full RAID array for plausibility
2335  */
2336
2337
2338 static void analyze_sbs(mddev_t * mddev)
2339 {
2340         int i;
2341         struct list_head *tmp;
2342         mdk_rdev_t *rdev, *freshest;
2343         char b[BDEVNAME_SIZE];
2344
2345         freshest = NULL;
2346         rdev_for_each(rdev, tmp, mddev)
2347                 switch (super_types[mddev->major_version].
2348                         load_super(rdev, freshest, mddev->minor_version)) {
2349                 case 1:
2350                         freshest = rdev;
2351                         break;
2352                 case 0:
2353                         break;
2354                 default:
2355                         printk( KERN_ERR \
2356                                 "md: fatal superblock inconsistency in %s"
2357                                 " -- removing from array\n",
2358                                 bdevname(rdev->bdev,b));
2359                         kick_rdev_from_array(rdev);
2360                 }
2361
2362
2363         super_types[mddev->major_version].
2364                 validate_super(mddev, freshest);
2365
2366         i = 0;
2367         rdev_for_each(rdev, tmp, mddev) {
2368                 if (rdev != freshest)
2369                         if (super_types[mddev->major_version].
2370                             validate_super(mddev, rdev)) {
2371                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2372                                         " from array!\n",
2373                                         bdevname(rdev->bdev,b));
2374                                 kick_rdev_from_array(rdev);
2375                                 continue;
2376                         }
2377                 if (mddev->level == LEVEL_MULTIPATH) {
2378                         rdev->desc_nr = i++;
2379                         rdev->raid_disk = rdev->desc_nr;
2380                         set_bit(In_sync, &rdev->flags);
2381                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2382                         rdev->raid_disk = -1;
2383                         clear_bit(In_sync, &rdev->flags);
2384                 }
2385         }
2386
2387
2388
2389         if (mddev->recovery_cp != MaxSector &&
2390             mddev->level >= 1)
2391                 printk(KERN_ERR "md: %s: raid array is not clean"
2392                        " -- starting background reconstruction\n",
2393                        mdname(mddev));
2394
2395 }
2396
2397 static ssize_t
2398 safe_delay_show(mddev_t *mddev, char *page)
2399 {
2400         int msec = (mddev->safemode_delay*1000)/HZ;
2401         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2402 }
2403 static ssize_t
2404 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2405 {
2406         int scale=1;
2407         int dot=0;
2408         int i;
2409         unsigned long msec;
2410         char buf[30];
2411         char *e;
2412         /* remove a period, and count digits after it */
2413         if (len >= sizeof(buf))
2414                 return -EINVAL;
2415         strlcpy(buf, cbuf, len);
2416         buf[len] = 0;
2417         for (i=0; i<len; i++) {
2418                 if (dot) {
2419                         if (isdigit(buf[i])) {
2420                                 buf[i-1] = buf[i];
2421                                 scale *= 10;
2422                         }
2423                         buf[i] = 0;
2424                 } else if (buf[i] == '.') {
2425                         dot=1;
2426                         buf[i] = 0;
2427                 }
2428         }
2429         msec = simple_strtoul(buf, &e, 10);
2430         if (e == buf || (*e && *e != '\n'))
2431                 return -EINVAL;
2432         msec = (msec * 1000) / scale;
2433         if (msec == 0)
2434                 mddev->safemode_delay = 0;
2435         else {
2436                 mddev->safemode_delay = (msec*HZ)/1000;
2437                 if (mddev->safemode_delay == 0)
2438                         mddev->safemode_delay = 1;
2439         }
2440         return len;
2441 }
2442 static struct md_sysfs_entry md_safe_delay =
2443 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2444
2445 static ssize_t
2446 level_show(mddev_t *mddev, char *page)
2447 {
2448         struct mdk_personality *p = mddev->pers;
2449         if (p)
2450                 return sprintf(page, "%s\n", p->name);
2451         else if (mddev->clevel[0])
2452                 return sprintf(page, "%s\n", mddev->clevel);
2453         else if (mddev->level != LEVEL_NONE)
2454                 return sprintf(page, "%d\n", mddev->level);
2455         else
2456                 return 0;
2457 }
2458
2459 static ssize_t
2460 level_store(mddev_t *mddev, const char *buf, size_t len)
2461 {
2462         ssize_t rv = len;
2463         if (mddev->pers)
2464                 return -EBUSY;
2465         if (len == 0)
2466                 return 0;
2467         if (len >= sizeof(mddev->clevel))
2468                 return -ENOSPC;
2469         strncpy(mddev->clevel, buf, len);
2470         if (mddev->clevel[len-1] == '\n')
2471                 len--;
2472         mddev->clevel[len] = 0;
2473         mddev->level = LEVEL_NONE;
2474         return rv;
2475 }
2476
2477 static struct md_sysfs_entry md_level =
2478 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2479
2480
2481 static ssize_t
2482 layout_show(mddev_t *mddev, char *page)
2483 {
2484         /* just a number, not meaningful for all levels */
2485         if (mddev->reshape_position != MaxSector &&
2486             mddev->layout != mddev->new_layout)
2487                 return sprintf(page, "%d (%d)\n",
2488                                mddev->new_layout, mddev->layout);
2489         return sprintf(page, "%d\n", mddev->layout);
2490 }
2491
2492 static ssize_t
2493 layout_store(mddev_t *mddev, const char *buf, size_t len)
2494 {
2495         char *e;
2496         unsigned long n = simple_strtoul(buf, &e, 10);
2497
2498         if (!*buf || (*e && *e != '\n'))
2499                 return -EINVAL;
2500
2501         if (mddev->pers)
2502                 return -EBUSY;
2503         if (mddev->reshape_position != MaxSector)
2504                 mddev->new_layout = n;
2505         else
2506                 mddev->layout = n;
2507         return len;
2508 }
2509 static struct md_sysfs_entry md_layout =
2510 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2511
2512
2513 static ssize_t
2514 raid_disks_show(mddev_t *mddev, char *page)
2515 {
2516         if (mddev->raid_disks == 0)
2517                 return 0;
2518         if (mddev->reshape_position != MaxSector &&
2519             mddev->delta_disks != 0)
2520                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2521                                mddev->raid_disks - mddev->delta_disks);
2522         return sprintf(page, "%d\n", mddev->raid_disks);
2523 }
2524
2525 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2526
2527 static ssize_t
2528 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2529 {
2530         char *e;
2531         int rv = 0;
2532         unsigned long n = simple_strtoul(buf, &e, 10);
2533
2534         if (!*buf || (*e && *e != '\n'))
2535                 return -EINVAL;
2536
2537         if (mddev->pers)
2538                 rv = update_raid_disks(mddev, n);
2539         else if (mddev->reshape_position != MaxSector) {
2540                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2541                 mddev->delta_disks = n - olddisks;
2542                 mddev->raid_disks = n;
2543         } else
2544                 mddev->raid_disks = n;
2545         return rv ? rv : len;
2546 }
2547 static struct md_sysfs_entry md_raid_disks =
2548 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2549
2550 static ssize_t
2551 chunk_size_show(mddev_t *mddev, char *page)
2552 {
2553         if (mddev->reshape_position != MaxSector &&
2554             mddev->chunk_size != mddev->new_chunk)
2555                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2556                                mddev->chunk_size);
2557         return sprintf(page, "%d\n", mddev->chunk_size);
2558 }
2559
2560 static ssize_t
2561 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2562 {
2563         /* can only set chunk_size if array is not yet active */
2564         char *e;
2565         unsigned long n = simple_strtoul(buf, &e, 10);
2566
2567         if (!*buf || (*e && *e != '\n'))
2568                 return -EINVAL;
2569
2570         if (mddev->pers)
2571                 return -EBUSY;
2572         else if (mddev->reshape_position != MaxSector)
2573                 mddev->new_chunk = n;
2574         else
2575                 mddev->chunk_size = n;
2576         return len;
2577 }
2578 static struct md_sysfs_entry md_chunk_size =
2579 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2580
2581 static ssize_t
2582 resync_start_show(mddev_t *mddev, char *page)
2583 {
2584         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2585 }
2586
2587 static ssize_t
2588 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2589 {
2590         char *e;
2591         unsigned long long n = simple_strtoull(buf, &e, 10);
2592
2593         if (mddev->pers)
2594                 return -EBUSY;
2595         if (!*buf || (*e && *e != '\n'))
2596                 return -EINVAL;
2597
2598         mddev->recovery_cp = n;
2599         return len;
2600 }
2601 static struct md_sysfs_entry md_resync_start =
2602 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2603
2604 /*
2605  * The array state can be:
2606  *
2607  * clear
2608  *     No devices, no size, no level
2609  *     Equivalent to STOP_ARRAY ioctl
2610  * inactive
2611  *     May have some settings, but array is not active
2612  *        all IO results in error
2613  *     When written, doesn't tear down array, but just stops it
2614  * suspended (not supported yet)
2615  *     All IO requests will block. The array can be reconfigured.
2616  *     Writing this, if accepted, will block until array is quiescent
2617  * readonly
2618  *     no resync can happen.  no superblocks get written.
2619  *     write requests fail
2620  * read-auto
2621  *     like readonly, but behaves like 'clean' on a write request.
2622  *
2623  * clean - no pending writes, but otherwise active.
2624  *     When written to inactive array, starts without resync
2625  *     If a write request arrives then
2626  *       if metadata is known, mark 'dirty' and switch to 'active'.
2627  *       if not known, block and switch to write-pending
2628  *     If written to an active array that has pending writes, then fails.
2629  * active
2630  *     fully active: IO and resync can be happening.
2631  *     When written to inactive array, starts with resync
2632  *
2633  * write-pending
2634  *     clean, but writes are blocked waiting for 'active' to be written.
2635  *
2636  * active-idle
2637  *     like active, but no writes have been seen for a while (100msec).
2638  *
2639  */
2640 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2641                    write_pending, active_idle, bad_word};
2642 static char *array_states[] = {
2643         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2644         "write-pending", "active-idle", NULL };
2645
2646 static int match_word(const char *word, char **list)
2647 {
2648         int n;
2649         for (n=0; list[n]; n++)
2650                 if (cmd_match(word, list[n]))
2651                         break;
2652         return n;
2653 }
2654
2655 static ssize_t
2656 array_state_show(mddev_t *mddev, char *page)
2657 {
2658         enum array_state st = inactive;
2659
2660         if (mddev->pers)
2661                 switch(mddev->ro) {
2662                 case 1:
2663                         st = readonly;
2664                         break;
2665                 case 2:
2666                         st = read_auto;
2667                         break;
2668                 case 0:
2669                         if (mddev->in_sync)
2670                                 st = clean;
2671                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2672                                 st = write_pending;
2673                         else if (mddev->safemode)
2674                                 st = active_idle;
2675                         else
2676                                 st = active;
2677                 }
2678         else {
2679                 if (list_empty(&mddev->disks) &&
2680                     mddev->raid_disks == 0 &&
2681                     mddev->size == 0)
2682                         st = clear;
2683                 else
2684                         st = inactive;
2685         }
2686         return sprintf(page, "%s\n", array_states[st]);
2687 }
2688
2689 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2690 static int do_md_run(mddev_t * mddev);
2691 static int restart_array(mddev_t *mddev);
2692
2693 static ssize_t
2694 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2695 {
2696         int err = -EINVAL;
2697         enum array_state st = match_word(buf, array_states);
2698         switch(st) {
2699         case bad_word:
2700                 break;
2701         case clear:
2702                 /* stopping an active array */
2703                 if (atomic_read(&mddev->active) > 1)
2704                         return -EBUSY;
2705                 err = do_md_stop(mddev, 0, 0);
2706                 break;
2707         case inactive:
2708                 /* stopping an active array */
2709                 if (mddev->pers) {
2710                         if (atomic_read(&mddev->active) > 1)
2711                                 return -EBUSY;
2712                         err = do_md_stop(mddev, 2, 0);
2713                 } else
2714                         err = 0; /* already inactive */
2715                 break;
2716         case suspended:
2717                 break; /* not supported yet */
2718         case readonly:
2719                 if (mddev->pers)
2720                         err = do_md_stop(mddev, 1, 0);
2721                 else {
2722                         mddev->ro = 1;
2723                         set_disk_ro(mddev->gendisk, 1);
2724                         err = do_md_run(mddev);
2725                 }
2726                 break;
2727         case read_auto:
2728                 if (mddev->pers) {
2729                         if (mddev->ro != 1)
2730                                 err = do_md_stop(mddev, 1, 0);
2731                         else
2732                                 err = restart_array(mddev);
2733                         if (err == 0) {
2734                                 mddev->ro = 2;
2735                                 set_disk_ro(mddev->gendisk, 0);
2736                         }
2737                 } else {
2738                         mddev->ro = 2;
2739                         err = do_md_run(mddev);
2740                 }
2741                 break;
2742         case clean:
2743                 if (mddev->pers) {
2744                         restart_array(mddev);
2745                         spin_lock_irq(&mddev->write_lock);
2746                         if (atomic_read(&mddev->writes_pending) == 0) {
2747                                 if (mddev->in_sync == 0) {
2748                                         mddev->in_sync = 1;
2749                                         if (mddev->safemode == 1)
2750                                                 mddev->safemode = 0;
2751                                         if (mddev->persistent)
2752                                                 set_bit(MD_CHANGE_CLEAN,
2753                                                         &mddev->flags);
2754                                 }
2755                                 err = 0;
2756                         } else
2757                                 err = -EBUSY;
2758                         spin_unlock_irq(&mddev->write_lock);
2759                 } else {
2760                         mddev->ro = 0;
2761                         mddev->recovery_cp = MaxSector;
2762                         err = do_md_run(mddev);
2763                 }
2764                 break;
2765         case active:
2766                 if (mddev->pers) {
2767                         restart_array(mddev);
2768                         if (mddev->external)
2769                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2770                         wake_up(&mddev->sb_wait);
2771                         err = 0;
2772                 } else {
2773                         mddev->ro = 0;
2774                         set_disk_ro(mddev->gendisk, 0);
2775                         err = do_md_run(mddev);
2776                 }
2777                 break;
2778         case write_pending:
2779         case active_idle:
2780                 /* these cannot be set */
2781                 break;
2782         }
2783         if (err)
2784                 return err;
2785         else {
2786                 sysfs_notify(&mddev->kobj, NULL, "array_state");
2787                 return len;
2788         }
2789 }
2790 static struct md_sysfs_entry md_array_state =
2791 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2792
2793 static ssize_t
2794 null_show(mddev_t *mddev, char *page)
2795 {
2796         return -EINVAL;
2797 }
2798
2799 static ssize_t
2800 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2801 {
2802         /* buf must be %d:%d\n? giving major and minor numbers */
2803         /* The new device is added to the array.
2804          * If the array has a persistent superblock, we read the
2805          * superblock to initialise info and check validity.
2806          * Otherwise, only checking done is that in bind_rdev_to_array,
2807          * which mainly checks size.
2808          */
2809         char *e;
2810         int major = simple_strtoul(buf, &e, 10);
2811         int minor;
2812         dev_t dev;
2813         mdk_rdev_t *rdev;
2814         int err;
2815
2816         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2817                 return -EINVAL;
2818         minor = simple_strtoul(e+1, &e, 10);
2819         if (*e && *e != '\n')
2820                 return -EINVAL;
2821         dev = MKDEV(major, minor);
2822         if (major != MAJOR(dev) ||
2823             minor != MINOR(dev))
2824                 return -EOVERFLOW;
2825
2826
2827         if (mddev->persistent) {
2828                 rdev = md_import_device(dev, mddev->major_version,
2829                                         mddev->minor_version);
2830                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2831                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2832                                                        mdk_rdev_t, same_set);
2833                         err = super_types[mddev->major_version]
2834                                 .load_super(rdev, rdev0, mddev->minor_version);
2835                         if (err < 0)
2836                                 goto out;
2837                 }
2838         } else if (mddev->external)
2839                 rdev = md_import_device(dev, -2, -1);
2840         else
2841                 rdev = md_import_device(dev, -1, -1);
2842
2843         if (IS_ERR(rdev))
2844                 return PTR_ERR(rdev);
2845         err = bind_rdev_to_array(rdev, mddev);
2846  out:
2847         if (err)
2848                 export_rdev(rdev);
2849         return err ? err : len;
2850 }
2851
2852 static struct md_sysfs_entry md_new_device =
2853 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2854
2855 static ssize_t
2856 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2857 {
2858         char *end;
2859         unsigned long chunk, end_chunk;
2860
2861         if (!mddev->bitmap)
2862                 goto out;
2863         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2864         while (*buf) {
2865                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2866                 if (buf == end) break;
2867                 if (*end == '-') { /* range */
2868                         buf = end + 1;
2869                         end_chunk = simple_strtoul(buf, &end, 0);
2870                         if (buf == end) break;
2871                 }
2872                 if (*end && !isspace(*end)) break;
2873                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2874                 buf = end;
2875                 while (isspace(*buf)) buf++;
2876         }
2877         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2878 out:
2879         return len;
2880 }
2881
2882 static struct md_sysfs_entry md_bitmap =
2883 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2884
2885 static ssize_t
2886 size_show(mddev_t *mddev, char *page)
2887 {
2888         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2889 }
2890
2891 static int update_size(mddev_t *mddev, sector_t num_sectors);
2892
2893 static ssize_t
2894 size_store(mddev_t *mddev, const char *buf, size_t len)
2895 {
2896         /* If array is inactive, we can reduce the component size, but
2897          * not increase it (except from 0).
2898          * If array is active, we can try an on-line resize
2899          */
2900         char *e;
2901         int err = 0;
2902         unsigned long long size = simple_strtoull(buf, &e, 10);
2903         if (!*buf || *buf == '\n' ||
2904             (*e && *e != '\n'))
2905                 return -EINVAL;
2906
2907         if (mddev->pers) {
2908                 err = update_size(mddev, size * 2);
2909                 md_update_sb(mddev, 1);
2910         } else {
2911                 if (mddev->size == 0 ||
2912                     mddev->size > size)
2913                         mddev->size = size;
2914                 else
2915                         err = -ENOSPC;
2916         }
2917         return err ? err : len;
2918 }
2919
2920 static struct md_sysfs_entry md_size =
2921 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2922
2923
2924 /* Metdata version.
2925  * This is one of
2926  *   'none' for arrays with no metadata (good luck...)
2927  *   'external' for arrays with externally managed metadata,
2928  * or N.M for internally known formats
2929  */
2930 static ssize_t
2931 metadata_show(mddev_t *mddev, char *page)
2932 {
2933         if (mddev->persistent)
2934                 return sprintf(page, "%d.%d\n",
2935                                mddev->major_version, mddev->minor_version);
2936         else if (mddev->external)
2937                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2938         else
2939                 return sprintf(page, "none\n");
2940 }
2941
2942 static ssize_t
2943 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2944 {
2945         int major, minor;
2946         char *e;
2947         if (!list_empty(&mddev->disks))
2948                 return -EBUSY;
2949
2950         if (cmd_match(buf, "none")) {
2951                 mddev->persistent = 0;
2952                 mddev->external = 0;
2953                 mddev->major_version = 0;
2954                 mddev->minor_version = 90;
2955                 return len;
2956         }
2957         if (strncmp(buf, "external:", 9) == 0) {
2958                 size_t namelen = len-9;
2959                 if (namelen >= sizeof(mddev->metadata_type))
2960                         namelen = sizeof(mddev->metadata_type)-1;
2961                 strncpy(mddev->metadata_type, buf+9, namelen);
2962                 mddev->metadata_type[namelen] = 0;
2963                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2964                         mddev->metadata_type[--namelen] = 0;
2965                 mddev->persistent = 0;
2966                 mddev->external = 1;
2967                 mddev->major_version = 0;
2968                 mddev->minor_version = 90;
2969                 return len;
2970         }
2971         major = simple_strtoul(buf, &e, 10);
2972         if (e==buf || *e != '.')
2973                 return -EINVAL;
2974         buf = e+1;
2975         minor = simple_strtoul(buf, &e, 10);
2976         if (e==buf || (*e && *e != '\n') )
2977                 return -EINVAL;
2978         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2979                 return -ENOENT;
2980         mddev->major_version = major;
2981         mddev->minor_version = minor;
2982         mddev->persistent = 1;
2983         mddev->external = 0;
2984         return len;
2985 }
2986
2987 static struct md_sysfs_entry md_metadata =
2988 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2989
2990 static ssize_t
2991 action_show(mddev_t *mddev, char *page)
2992 {
2993         char *type = "idle";
2994         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2995             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2996                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2997                         type = "reshape";
2998                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2999                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3000                                 type = "resync";
3001                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3002                                 type = "check";
3003                         else
3004                                 type = "repair";
3005                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3006                         type = "recover";
3007         }
3008         return sprintf(page, "%s\n", type);
3009 }
3010
3011 static ssize_t
3012 action_store(mddev_t *mddev, const char *page, size_t len)
3013 {
3014         if (!mddev->pers || !mddev->pers->sync_request)
3015                 return -EINVAL;
3016
3017         if (cmd_match(page, "idle")) {
3018                 if (mddev->sync_thread) {
3019                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3020                         md_unregister_thread(mddev->sync_thread);
3021                         mddev->sync_thread = NULL;
3022                         mddev->recovery = 0;
3023                 }
3024         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3025                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3026                 return -EBUSY;
3027         else if (cmd_match(page, "resync"))
3028                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3029         else if (cmd_match(page, "recover")) {
3030                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3031                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3032         } else if (cmd_match(page, "reshape")) {
3033                 int err;
3034                 if (mddev->pers->start_reshape == NULL)
3035                         return -EINVAL;
3036                 err = mddev->pers->start_reshape(mddev);
3037                 if (err)
3038                         return err;
3039                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3040         } else {
3041                 if (cmd_match(page, "check"))
3042                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3043                 else if (!cmd_match(page, "repair"))
3044                         return -EINVAL;
3045                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3046                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3047         }
3048         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3049         md_wakeup_thread(mddev->thread);
3050         sysfs_notify(&mddev->kobj, NULL, "sync_action");
3051         return len;
3052 }
3053
3054 static ssize_t
3055 mismatch_cnt_show(mddev_t *mddev, char *page)
3056 {
3057         return sprintf(page, "%llu\n",
3058                        (unsigned long long) mddev->resync_mismatches);
3059 }
3060
3061 static struct md_sysfs_entry md_scan_mode =
3062 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3063
3064
3065 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3066
3067 static ssize_t
3068 sync_min_show(mddev_t *mddev, char *page)
3069 {
3070         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3071                        mddev->sync_speed_min ? "local": "system");
3072 }
3073
3074 static ssize_t
3075 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3076 {
3077         int min;
3078         char *e;
3079         if (strncmp(buf, "system", 6)==0) {
3080                 mddev->sync_speed_min = 0;
3081                 return len;
3082         }
3083         min = simple_strtoul(buf, &e, 10);
3084         if (buf == e || (*e && *e != '\n') || min <= 0)
3085                 return -EINVAL;
3086         mddev->sync_speed_min = min;
3087         return len;
3088 }
3089
3090 static struct md_sysfs_entry md_sync_min =
3091 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3092
3093 static ssize_t
3094 sync_max_show(mddev_t *mddev, char *page)
3095 {
3096         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3097                        mddev->sync_speed_max ? "local": "system");
3098 }
3099
3100 static ssize_t
3101 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3102 {
3103         int max;
3104         char *e;
3105         if (strncmp(buf, "system", 6)==0) {
3106                 mddev->sync_speed_max = 0;
3107                 return len;
3108         }
3109         max = simple_strtoul(buf, &e, 10);
3110         if (buf == e || (*e && *e != '\n') || max <= 0)
3111                 return -EINVAL;
3112         mddev->sync_speed_max = max;
3113         return len;
3114 }
3115
3116 static struct md_sysfs_entry md_sync_max =
3117 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3118
3119 static ssize_t
3120 degraded_show(mddev_t *mddev, char *page)
3121 {
3122         return sprintf(page, "%d\n", mddev->degraded);
3123 }
3124 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3125
3126 static ssize_t
3127 sync_force_parallel_show(mddev_t *mddev, char *page)
3128 {
3129         return sprintf(page, "%d\n", mddev->parallel_resync);
3130 }
3131
3132 static ssize_t
3133 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3134 {
3135         long n;
3136
3137         if (strict_strtol(buf, 10, &n))
3138                 return -EINVAL;
3139
3140         if (n != 0 && n != 1)
3141                 return -EINVAL;
3142
3143         mddev->parallel_resync = n;
3144
3145         if (mddev->sync_thread)
3146                 wake_up(&resync_wait);
3147
3148         return len;
3149 }
3150
3151 /* force parallel resync, even with shared block devices */
3152 static struct md_sysfs_entry md_sync_force_parallel =
3153 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3154        sync_force_parallel_show, sync_force_parallel_store);
3155
3156 static ssize_t
3157 sync_speed_show(mddev_t *mddev, char *page)
3158 {
3159         unsigned long resync, dt, db;
3160         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3161         dt = (jiffies - mddev->resync_mark) / HZ;
3162         if (!dt) dt++;
3163         db = resync - mddev->resync_mark_cnt;
3164         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3165 }
3166
3167 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3168
3169 static ssize_t
3170 sync_completed_show(mddev_t *mddev, char *page)
3171 {
3172         unsigned long max_blocks, resync;
3173
3174         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3175                 max_blocks = mddev->resync_max_sectors;
3176         else
3177                 max_blocks = mddev->size << 1;
3178
3179         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3180         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3181 }
3182
3183 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3184
3185 static ssize_t
3186 min_sync_show(mddev_t *mddev, char *page)
3187 {
3188         return sprintf(page, "%llu\n",
3189                        (unsigned long long)mddev->resync_min);
3190 }
3191 static ssize_t
3192 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3193 {
3194         unsigned long long min;
3195         if (strict_strtoull(buf, 10, &min))
3196                 return -EINVAL;
3197         if (min > mddev->resync_max)
3198                 return -EINVAL;
3199         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3200                 return -EBUSY;
3201
3202         /* Must be a multiple of chunk_size */
3203         if (mddev->chunk_size) {
3204                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3205                         return -EINVAL;
3206         }
3207         mddev->resync_min = min;
3208
3209         return len;
3210 }
3211
3212 static struct md_sysfs_entry md_min_sync =
3213 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3214
3215 static ssize_t
3216 max_sync_show(mddev_t *mddev, char *page)
3217 {
3218         if (mddev->resync_max == MaxSector)
3219                 return sprintf(page, "max\n");
3220         else
3221                 return sprintf(page, "%llu\n",
3222                                (unsigned long long)mddev->resync_max);
3223 }
3224 static ssize_t
3225 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3226 {
3227         if (strncmp(buf, "max", 3) == 0)
3228                 mddev->resync_max = MaxSector;
3229         else {
3230                 unsigned long long max;
3231                 if (strict_strtoull(buf, 10, &max))
3232                         return -EINVAL;
3233                 if (max < mddev->resync_min)
3234                         return -EINVAL;
3235                 if (max < mddev->resync_max &&
3236                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3237                         return -EBUSY;
3238
3239                 /* Must be a multiple of chunk_size */
3240                 if (mddev->chunk_size) {
3241                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3242                                 return -EINVAL;
3243                 }
3244                 mddev->resync_max = max;
3245         }
3246         wake_up(&mddev->recovery_wait);
3247         return len;
3248 }
3249
3250 static struct md_sysfs_entry md_max_sync =
3251 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3252
3253 static ssize_t
3254 suspend_lo_show(mddev_t *mddev, char *page)
3255 {
3256         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3257 }
3258
3259 static ssize_t
3260 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3261 {
3262         char *e;
3263         unsigned long long new = simple_strtoull(buf, &e, 10);
3264
3265         if (mddev->pers->quiesce == NULL)
3266                 return -EINVAL;
3267         if (buf == e || (*e && *e != '\n'))
3268                 return -EINVAL;
3269         if (new >= mddev->suspend_hi ||
3270             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3271                 mddev->suspend_lo = new;
3272                 mddev->pers->quiesce(mddev, 2);
3273                 return len;
3274         } else
3275                 return -EINVAL;
3276 }
3277 static struct md_sysfs_entry md_suspend_lo =
3278 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3279
3280
3281 static ssize_t
3282 suspend_hi_show(mddev_t *mddev, char *page)
3283 {
3284         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3285 }
3286
3287 static ssize_t
3288 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3289 {
3290         char *e;
3291         unsigned long long new = simple_strtoull(buf, &e, 10);
3292
3293         if (mddev->pers->quiesce == NULL)
3294                 return -EINVAL;
3295         if (buf == e || (*e && *e != '\n'))
3296                 return -EINVAL;
3297         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3298             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3299                 mddev->suspend_hi = new;
3300                 mddev->pers->quiesce(mddev, 1);
3301                 mddev->pers->quiesce(mddev, 0);
3302                 return len;
3303         } else
3304                 return -EINVAL;
3305 }
3306 static struct md_sysfs_entry md_suspend_hi =
3307 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3308
3309 static ssize_t
3310 reshape_position_show(mddev_t *mddev, char *page)
3311 {
3312         if (mddev->reshape_position != MaxSector)
3313                 return sprintf(page, "%llu\n",
3314                                (unsigned long long)mddev->reshape_position);
3315         strcpy(page, "none\n");
3316         return 5;
3317 }
3318
3319 static ssize_t
3320 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3321 {
3322         char *e;
3323         unsigned long long new = simple_strtoull(buf, &e, 10);
3324         if (mddev->pers)
3325                 return -EBUSY;
3326         if (buf == e || (*e && *e != '\n'))
3327                 return -EINVAL;
3328         mddev->reshape_position = new;
3329         mddev->delta_disks = 0;
3330         mddev->new_level = mddev->level;
3331         mddev->new_layout = mddev->layout;
3332         mddev->new_chunk = mddev->chunk_size;
3333         return len;
3334 }
3335
3336 static struct md_sysfs_entry md_reshape_position =
3337 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3338        reshape_position_store);
3339
3340
3341 static struct attribute *md_default_attrs[] = {
3342         &md_level.attr,
3343         &md_layout.attr,
3344         &md_raid_disks.attr,
3345         &md_chunk_size.attr,
3346         &md_size.attr,
3347         &md_resync_start.attr,
3348         &md_metadata.attr,
3349         &md_new_device.attr,
3350         &md_safe_delay.attr,
3351         &md_array_state.attr,
3352         &md_reshape_position.attr,
3353         NULL,
3354 };
3355
3356 static struct attribute *md_redundancy_attrs[] = {
3357         &md_scan_mode.attr,
3358         &md_mismatches.attr,
3359         &md_sync_min.attr,
3360         &md_sync_max.attr,
3361         &md_sync_speed.attr,
3362         &md_sync_force_parallel.attr,
3363         &md_sync_completed.attr,
3364         &md_min_sync.attr,
3365         &md_max_sync.attr,
3366         &md_suspend_lo.attr,
3367         &md_suspend_hi.attr,
3368         &md_bitmap.attr,
3369         &md_degraded.attr,
3370         NULL,
3371 };
3372 static struct attribute_group md_redundancy_group = {
3373         .name = NULL,
3374         .attrs = md_redundancy_attrs,
3375 };
3376
3377
3378 static ssize_t
3379 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3380 {
3381         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3382         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3383         ssize_t rv;
3384
3385         if (!entry->show)
3386                 return -EIO;
3387         rv = mddev_lock(mddev);
3388         if (!rv) {
3389                 rv = entry->show(mddev, page);
3390                 mddev_unlock(mddev);
3391         }
3392         return rv;
3393 }
3394
3395 static ssize_t
3396 md_attr_store(struct kobject *kobj, struct attribute *attr,
3397               const char *page, size_t length)
3398 {
3399         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3400         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3401         ssize_t rv;
3402
3403         if (!entry->store)
3404                 return -EIO;
3405         if (!capable(CAP_SYS_ADMIN))
3406                 return -EACCES;
3407         rv = mddev_lock(mddev);
3408         if (!rv) {
3409                 rv = entry->store(mddev, page, length);
3410                 mddev_unlock(mddev);
3411         }
3412         return rv;
3413 }
3414
3415 static void md_free(struct kobject *ko)
3416 {
3417         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3418         kfree(mddev);
3419 }
3420
3421 static struct sysfs_ops md_sysfs_ops = {
3422         .show   = md_attr_show,
3423         .store  = md_attr_store,
3424 };
3425 static struct kobj_type md_ktype = {
3426         .release        = md_free,
3427         .sysfs_ops      = &md_sysfs_ops,
3428         .default_attrs  = md_default_attrs,
3429 };
3430
3431 int mdp_major = 0;
3432
3433 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3434 {
3435         static DEFINE_MUTEX(disks_mutex);
3436         mddev_t *mddev = mddev_find(dev);
3437         struct gendisk *disk;
3438         int partitioned = (MAJOR(dev) != MD_MAJOR);
3439         int shift = partitioned ? MdpMinorShift : 0;
3440         int unit = MINOR(dev) >> shift;
3441         int error;
3442
3443         if (!mddev)
3444                 return NULL;
3445
3446         mutex_lock(&disks_mutex);
3447         if (mddev->gendisk) {
3448                 mutex_unlock(&disks_mutex);
3449                 mddev_put(mddev);
3450                 return NULL;
3451         }
3452         disk = alloc_disk(1 << shift);
3453         if (!disk) {
3454                 mutex_unlock(&disks_mutex);
3455                 mddev_put(mddev);
3456                 return NULL;
3457         }
3458         disk->major = MAJOR(dev);
3459         disk->first_minor = unit << shift;
3460         if (partitioned)
3461                 sprintf(disk->disk_name, "md_d%d", unit);
3462         else
3463                 sprintf(disk->disk_name, "md%d", unit);
3464         disk->fops = &md_fops;
3465         disk->private_data = mddev;
3466         disk->queue = mddev->queue;
3467         add_disk(disk);
3468         mddev->gendisk = disk;
3469         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3470                                      "%s", "md");
3471         mutex_unlock(&disks_mutex);
3472         if (error)
3473                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3474                        disk->disk_name);
3475         else
3476                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3477         return NULL;
3478 }
3479
3480 static void md_safemode_timeout(unsigned long data)
3481 {
3482         mddev_t *mddev = (mddev_t *) data;
3483
3484         if (!atomic_read(&mddev->writes_pending)) {
3485                 mddev->safemode = 1;
3486                 if (mddev->external)
3487                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3488         }
3489         md_wakeup_thread(mddev->thread);
3490 }
3491
3492 static int start_dirty_degraded;
3493
3494 static int do_md_run(mddev_t * mddev)
3495 {
3496         int err;
3497         int chunk_size;
3498         struct list_head *tmp;
3499         mdk_rdev_t *rdev;
3500         struct gendisk *disk;
3501         struct mdk_personality *pers;
3502         char b[BDEVNAME_SIZE];
3503
3504         if (list_empty(&mddev->disks))
3505                 /* cannot run an array with no devices.. */
3506                 return -EINVAL;
3507
3508         if (mddev->pers)
3509                 return -EBUSY;
3510
3511         /*
3512          * Analyze all RAID superblock(s)
3513          */
3514         if (!mddev->raid_disks) {
3515                 if (!mddev->persistent)
3516                         return -EINVAL;
3517                 analyze_sbs(mddev);
3518         }
3519
3520         chunk_size = mddev->chunk_size;
3521
3522         if (chunk_size) {
3523                 if (chunk_size > MAX_CHUNK_SIZE) {
3524                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3525                                 chunk_size, MAX_CHUNK_SIZE);
3526                         return -EINVAL;
3527                 }
3528                 /*
3529                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3530                  */
3531                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3532                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3533                         return -EINVAL;
3534                 }
3535                 if (chunk_size < PAGE_SIZE) {
3536                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3537                                 chunk_size, PAGE_SIZE);
3538                         return -EINVAL;
3539                 }
3540
3541                 /* devices must have minimum size of one chunk */
3542                 rdev_for_each(rdev, tmp, mddev) {
3543                         if (test_bit(Faulty, &rdev->flags))
3544                                 continue;
3545                         if (rdev->size < chunk_size / 1024) {
3546                                 printk(KERN_WARNING
3547                                         "md: Dev %s smaller than chunk_size:"
3548                                         " %lluk < %dk\n",
3549                                         bdevname(rdev->bdev,b),
3550                                         (unsigned long long)rdev->size,
3551                                         chunk_size / 1024);
3552                                 return -EINVAL;
3553                         }
3554                 }
3555         }
3556
3557 #ifdef CONFIG_KMOD
3558         if (mddev->level != LEVEL_NONE)
3559                 request_module("md-level-%d", mddev->level);
3560         else if (mddev->clevel[0])
3561                 request_module("md-%s", mddev->clevel);
3562 #endif
3563
3564         /*
3565          * Drop all container device buffers, from now on
3566          * the only valid external interface is through the md
3567          * device.
3568          */
3569         rdev_for_each(rdev, tmp, mddev) {
3570                 if (test_bit(Faulty, &rdev->flags))
3571                         continue;
3572                 sync_blockdev(rdev->bdev);
3573                 invalidate_bdev(rdev->bdev);
3574
3575                 /* perform some consistency tests on the device.
3576                  * We don't want the data to overlap the metadata,
3577                  * Internal Bitmap issues has handled elsewhere.
3578                  */
3579                 if (rdev->data_offset < rdev->sb_offset) {
3580                         if (mddev->size &&
3581                             rdev->data_offset + mddev->size*2
3582                             > rdev->sb_offset*2) {
3583                                 printk("md: %s: data overlaps metadata\n",
3584                                        mdname(mddev));
3585                                 return -EINVAL;
3586                         }
3587                 } else {
3588                         if (rdev->sb_offset*2 + rdev->sb_size/512
3589                             > rdev->data_offset) {
3590                                 printk("md: %s: metadata overlaps data\n",
3591                                        mdname(mddev));
3592                                 return -EINVAL;
3593                         }
3594                 }
3595                 sysfs_notify(&rdev->kobj, NULL, "state");
3596         }
3597
3598         md_probe(mddev->unit, NULL, NULL);
3599         disk = mddev->gendisk;
3600         if (!disk)
3601                 return -ENOMEM;
3602
3603         spin_lock(&pers_lock);
3604         pers = find_pers(mddev->level, mddev->clevel);
3605         if (!pers || !try_module_get(pers->owner)) {
3606                 spin_unlock(&pers_lock);
3607                 if (mddev->level != LEVEL_NONE)
3608                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3609                                mddev->level);
3610                 else
3611                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3612                                mddev->clevel);
3613                 return -EINVAL;
3614         }
3615         mddev->pers = pers;
3616         spin_unlock(&pers_lock);
3617         mddev->level = pers->level;
3618         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3619
3620         if (mddev->reshape_position != MaxSector &&
3621             pers->start_reshape == NULL) {
3622                 /* This personality cannot handle reshaping... */
3623                 mddev->pers = NULL;
3624                 module_put(pers->owner);
3625                 return -EINVAL;
3626         }
3627
3628         if (pers->sync_request) {
3629                 /* Warn if this is a potentially silly
3630                  * configuration.
3631                  */
3632                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3633                 mdk_rdev_t *rdev2;
3634                 struct list_head *tmp2;
3635                 int warned = 0;
3636                 rdev_for_each(rdev, tmp, mddev) {
3637                         rdev_for_each(rdev2, tmp2, mddev) {
3638                                 if (rdev < rdev2 &&
3639                                     rdev->bdev->bd_contains ==
3640                                     rdev2->bdev->bd_contains) {
3641                                         printk(KERN_WARNING
3642                                                "%s: WARNING: %s appears to be"
3643                                                " on the same physical disk as"
3644                                                " %s.\n",
3645                                                mdname(mddev),
3646                                                bdevname(rdev->bdev,b),
3647                                                bdevname(rdev2->bdev,b2));
3648                                         warned = 1;
3649                                 }
3650                         }
3651                 }
3652                 if (warned)
3653                         printk(KERN_WARNING
3654                                "True protection against single-disk"
3655                                " failure might be compromised.\n");
3656         }
3657
3658         mddev->recovery = 0;
3659         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3660         mddev->barriers_work = 1;
3661         mddev->ok_start_degraded = start_dirty_degraded;
3662
3663         if (start_readonly)
3664                 mddev->ro = 2; /* read-only, but switch on first write */
3665
3666         err = mddev->pers->run(mddev);
3667         if (err)
3668                 printk(KERN_ERR "md: pers->run() failed ...\n");
3669         else if (mddev->pers->sync_request) {
3670                 err = bitmap_create(mddev);
3671                 if (err) {
3672                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3673                                mdname(mddev), err);
3674                         mddev->pers->stop(mddev);
3675                 }
3676         }
3677         if (err) {
3678                 module_put(mddev->pers->owner);
3679                 mddev->pers = NULL;
3680                 bitmap_destroy(mddev);
3681                 return err;
3682         }
3683         if (mddev->pers->sync_request) {
3684                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3685                         printk(KERN_WARNING
3686                                "md: cannot register extra attributes for %s\n",
3687                                mdname(mddev));
3688         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3689                 mddev->ro = 0;
3690
3691         atomic_set(&mddev->writes_pending,0);
3692         mddev->safemode = 0;
3693         mddev->safemode_timer.function = md_safemode_timeout;
3694         mddev->safemode_timer.data = (unsigned long) mddev;
3695         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3696         mddev->in_sync = 1;
3697
3698         rdev_for_each(rdev, tmp, mddev)
3699                 if (rdev->raid_disk >= 0) {
3700                         char nm[20];
3701                         sprintf(nm, "rd%d", rdev->raid_disk);
3702                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3703                                 printk("md: cannot register %s for %s\n",
3704                                        nm, mdname(mddev));
3705                 }
3706
3707         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3708
3709         if (mddev->flags)
3710                 md_update_sb(mddev, 0);
3711
3712         set_capacity(disk, mddev->array_size<<1);
3713
3714         /* If we call blk_queue_make_request here, it will
3715          * re-initialise max_sectors etc which may have been
3716          * refined inside -> run.  So just set the bits we need to set.
3717          * Most initialisation happended when we called
3718          * blk_queue_make_request(..., md_fail_request)
3719          * earlier.
3720          */
3721         mddev->queue->queuedata = mddev;
3722         mddev->queue->make_request_fn = mddev->pers->make_request;
3723
3724         /* If there is a partially-recovered drive we need to
3725          * start recovery here.  If we leave it to md_check_recovery,
3726          * it will remove the drives and not do the right thing
3727          */
3728         if (mddev->degraded && !mddev->sync_thread) {
3729                 struct list_head *rtmp;
3730                 int spares = 0;
3731                 rdev_for_each(rdev, rtmp, mddev)
3732                         if (rdev->raid_disk >= 0 &&
3733                             !test_bit(In_sync, &rdev->flags) &&
3734                             !test_bit(Faulty, &rdev->flags))
3735                                 /* complete an interrupted recovery */
3736                                 spares++;
3737                 if (spares && mddev->pers->sync_request) {
3738                         mddev->recovery = 0;
3739                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3740                         mddev->sync_thread = md_register_thread(md_do_sync,
3741                                                                 mddev,
3742                                                                 "%s_resync");
3743                         if (!mddev->sync_thread) {
3744                                 printk(KERN_ERR "%s: could not start resync"
3745                                        " thread...\n",
3746                                        mdname(mddev));
3747                                 /* leave the spares where they are, it shouldn't hurt */
3748                                 mddev->recovery = 0;
3749                         }
3750                 }
3751         }
3752         md_wakeup_thread(mddev->thread);
3753         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3754
3755         mddev->changed = 1;
3756         md_new_event(mddev);
3757         sysfs_notify(&mddev->kobj, NULL, "array_state");
3758         sysfs_notify(&mddev->kobj, NULL, "sync_action");
3759         sysfs_notify(&mddev->kobj, NULL, "degraded");
3760         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3761         return 0;
3762 }
3763
3764 static int restart_array(mddev_t *mddev)
3765 {
3766         struct gendisk *disk = mddev->gendisk;
3767
3768         /* Complain if it has no devices */
3769         if (list_empty(&mddev->disks))
3770                 return -ENXIO;
3771         if (!mddev->pers)
3772                 return -EINVAL;
3773         if (!mddev->ro)
3774                 return -EBUSY;
3775         mddev->safemode = 0;
3776         mddev->ro = 0;
3777         set_disk_ro(disk, 0);
3778         printk(KERN_INFO "md: %s switched to read-write mode.\n",
3779                 mdname(mddev));
3780         /* Kick recovery or resync if necessary */
3781         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3782         md_wakeup_thread(mddev->thread);
3783         md_wakeup_thread(mddev->sync_thread);
3784         sysfs_notify(&mddev->kobj, NULL, "array_state");
3785         return 0;
3786 }
3787
3788 /* similar to deny_write_access, but accounts for our holding a reference
3789  * to the file ourselves */
3790 static int deny_bitmap_write_access(struct file * file)
3791 {
3792         struct inode *inode = file->f_mapping->host;
3793
3794         spin_lock(&inode->i_lock);
3795         if (atomic_read(&inode->i_writecount) > 1) {
3796                 spin_unlock(&inode->i_lock);
3797                 return -ETXTBSY;
3798         }
3799         atomic_set(&inode->i_writecount, -1);
3800         spin_unlock(&inode->i_lock);
3801
3802         return 0;
3803 }
3804
3805 static void restore_bitmap_write_access(struct file *file)
3806 {
3807         struct inode *inode = file->f_mapping->host;
3808
3809         spin_lock(&inode->i_lock);
3810         atomic_set(&inode->i_writecount, 1);
3811         spin_unlock(&inode->i_lock);
3812 }
3813
3814 /* mode:
3815  *   0 - completely stop and dis-assemble array
3816  *   1 - switch to readonly
3817  *   2 - stop but do not disassemble array
3818  */
3819 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3820 {
3821         int err = 0;
3822         struct gendisk *disk = mddev->gendisk;
3823
3824         if (atomic_read(&mddev->active) > 1 + is_open) {
3825                 printk("md: %s still in use.\n",mdname(mddev));
3826                 return -EBUSY;
3827         }
3828
3829         if (mddev->pers) {
3830
3831                 if (mddev->sync_thread) {
3832                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3833                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3834                         md_unregister_thread(mddev->sync_thread);
3835                         mddev->sync_thread = NULL;
3836                 }
3837
3838                 del_timer_sync(&mddev->safemode_timer);
3839
3840                 invalidate_partition(disk, 0);
3841
3842                 switch(mode) {
3843                 case 1: /* readonly */
3844                         err  = -ENXIO;
3845                         if (mddev->ro==1)
3846                                 goto out;
3847                         mddev->ro = 1;
3848                         break;
3849                 case 0: /* disassemble */
3850                 case 2: /* stop */
3851                         bitmap_flush(mddev);
3852                         md_super_wait(mddev);
3853                         if (mddev->ro)
3854                                 set_disk_ro(disk, 0);
3855                         blk_queue_make_request(mddev->queue, md_fail_request);
3856                         mddev->pers->stop(mddev);
3857                         mddev->queue->merge_bvec_fn = NULL;
3858                         mddev->queue->unplug_fn = NULL;
3859                         mddev->queue->backing_dev_info.congested_fn = NULL;
3860                         if (mddev->pers->sync_request)
3861                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3862
3863                         module_put(mddev->pers->owner);
3864                         mddev->pers = NULL;
3865                         /* tell userspace to handle 'inactive' */
3866                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3867
3868                         set_capacity(disk, 0);
3869                         mddev->changed = 1;
3870
3871                         if (mddev->ro)
3872                                 mddev->ro = 0;
3873                 }
3874                 if (!mddev->in_sync || mddev->flags) {
3875                         /* mark array as shutdown cleanly */
3876                         mddev->in_sync = 1;
3877                         md_update_sb(mddev, 1);
3878                 }
3879                 if (mode == 1)
3880                         set_disk_ro(disk, 1);
3881                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3882         }
3883
3884         /*
3885          * Free resources if final stop
3886          */
3887         if (mode == 0) {
3888                 mdk_rdev_t *rdev;
3889                 struct list_head *tmp;
3890
3891                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3892
3893                 bitmap_destroy(mddev);
3894                 if (mddev->bitmap_file) {
3895                         restore_bitmap_write_access(mddev->bitmap_file);
3896                         fput(mddev->bitmap_file);
3897                         mddev->bitmap_file = NULL;
3898                 }
3899                 mddev->bitmap_offset = 0;
3900
3901                 rdev_for_each(rdev, tmp, mddev)
3902                         if (rdev->raid_disk >= 0) {
3903                                 char nm[20];
3904                                 sprintf(nm, "rd%d", rdev->raid_disk);
3905                                 sysfs_remove_link(&mddev->kobj, nm);
3906                         }
3907
3908                 /* make sure all md_delayed_delete calls have finished */
3909                 flush_scheduled_work();
3910
3911                 export_array(mddev);
3912
3913                 mddev->array_size = 0;
3914                 mddev->size = 0;
3915                 mddev->raid_disks = 0;
3916                 mddev->recovery_cp = 0;
3917                 mddev->resync_min = 0;
3918                 mddev->resync_max = MaxSector;
3919                 mddev->reshape_position = MaxSector;
3920                 mddev->external = 0;
3921                 mddev->persistent = 0;
3922                 mddev->level = LEVEL_NONE;
3923                 mddev->clevel[0] = 0;
3924                 mddev->flags = 0;
3925                 mddev->ro = 0;
3926                 mddev->metadata_type[0] = 0;
3927                 mddev->chunk_size = 0;
3928                 mddev->ctime = mddev->utime = 0;
3929                 mddev->layout = 0;
3930                 mddev->max_disks = 0;
3931                 mddev->events = 0;
3932                 mddev->delta_disks = 0;
3933                 mddev->new_level = LEVEL_NONE;
3934                 mddev->new_layout = 0;
3935                 mddev->new_chunk = 0;
3936                 mddev->curr_resync = 0;
3937                 mddev->resync_mismatches = 0;
3938                 mddev->suspend_lo = mddev->suspend_hi = 0;
3939                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3940                 mddev->recovery = 0;
3941                 mddev->in_sync = 0;
3942                 mddev->changed = 0;
3943                 mddev->degraded = 0;
3944                 mddev->barriers_work = 0;
3945                 mddev->safemode = 0;
3946
3947         } else if (mddev->pers)
3948                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3949                         mdname(mddev));
3950         err = 0;
3951         md_new_event(mddev);
3952         sysfs_notify(&mddev->kobj, NULL, "array_state");
3953 out:
3954         return err;
3955 }
3956
3957 #ifndef MODULE
3958 static void autorun_array(mddev_t *mddev)
3959 {
3960         mdk_rdev_t *rdev;
3961         struct list_head *tmp;
3962         int err;
3963
3964         if (list_empty(&mddev->disks))
3965                 return;
3966
3967         printk(KERN_INFO "md: running: ");
3968
3969         rdev_for_each(rdev, tmp, mddev) {
3970                 char b[BDEVNAME_SIZE];
3971                 printk("<%s>", bdevname(rdev->bdev,b));
3972         }
3973         printk("\n");
3974
3975         err = do_md_run (mddev);
3976         if (err) {
3977                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3978                 do_md_stop (mddev, 0, 0);
3979         }
3980 }
3981
3982 /*
3983  * lets try to run arrays based on all disks that have arrived
3984  * until now. (those are in pending_raid_disks)
3985  *
3986  * the method: pick the first pending disk, collect all disks with
3987  * the same UUID, remove all from the pending list and put them into
3988  * the 'same_array' list. Then order this list based on superblock
3989  * update time (freshest comes first), kick out 'old' disks and
3990  * compare superblocks. If everything's fine then run it.
3991  *
3992  * If "unit" is allocated, then bump its reference count
3993  */
3994 static void autorun_devices(int part)
3995 {
3996         struct list_head *tmp;
3997         mdk_rdev_t *rdev0, *rdev;
3998         mddev_t *mddev;
3999         char b[BDEVNAME_SIZE];
4000
4001         printk(KERN_INFO "md: autorun ...\n");
4002         while (!list_empty(&pending_raid_disks)) {
4003                 int unit;
4004                 dev_t dev;
4005                 LIST_HEAD(candidates);
4006                 rdev0 = list_entry(pending_raid_disks.next,
4007                                          mdk_rdev_t, same_set);
4008
4009                 printk(KERN_INFO "md: considering %s ...\n",
4010                         bdevname(rdev0->bdev,b));
4011                 INIT_LIST_HEAD(&candidates);
4012                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
4013                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4014                                 printk(KERN_INFO "md:  adding %s ...\n",
4015                                         bdevname(rdev->bdev,b));
4016                                 list_move(&rdev->same_set, &candidates);
4017                         }
4018                 /*
4019                  * now we have a set of devices, with all of them having
4020                  * mostly sane superblocks. It's time to allocate the
4021                  * mddev.
4022                  */
4023                 if (part) {
4024                         dev = MKDEV(mdp_major,
4025                                     rdev0->preferred_minor << MdpMinorShift);
4026                         unit = MINOR(dev) >> MdpMinorShift;
4027                 } else {
4028                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4029                         unit = MINOR(dev);
4030                 }
4031                 if (rdev0->preferred_minor != unit) {
4032                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4033                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4034                         break;
4035                 }
4036
4037                 md_probe(dev, NULL, NULL);
4038                 mddev = mddev_find(dev);
4039                 if (!mddev || !mddev->gendisk) {
4040                         if (mddev)
4041                                 mddev_put(mddev);
4042                         printk(KERN_ERR
4043                                 "md: cannot allocate memory for md drive.\n");
4044                         break;
4045                 }
4046                 if (mddev_lock(mddev))
4047                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4048                                mdname(mddev));
4049                 else if (mddev->raid_disks || mddev->major_version
4050                          || !list_empty(&mddev->disks)) {
4051                         printk(KERN_WARNING
4052                                 "md: %s already running, cannot run %s\n",
4053                                 mdname(mddev), bdevname(rdev0->bdev,b));
4054                         mddev_unlock(mddev);
4055                 } else {
4056                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4057                         mddev->persistent = 1;
4058                         rdev_for_each_list(rdev, tmp, candidates) {
4059                                 list_del_init(&rdev->same_set);
4060                                 if (bind_rdev_to_array(rdev, mddev))
4061                                         export_rdev(rdev);
4062                         }
4063                         autorun_array(mddev);
4064                         mddev_unlock(mddev);
4065                 }
4066                 /* on success, candidates will be empty, on error
4067                  * it won't...
4068                  */
4069                 rdev_for_each_list(rdev, tmp, candidates)
4070                         export_rdev(rdev);
4071                 mddev_put(mddev);
4072         }
4073         printk(KERN_INFO "md: ... autorun DONE.\n");
4074 }
4075 #endif /* !MODULE */
4076
4077 static int get_version(void __user * arg)
4078 {
4079         mdu_version_t ver;
4080
4081         ver.major = MD_MAJOR_VERSION;
4082         ver.minor = MD_MINOR_VERSION;
4083         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4084
4085         if (copy_to_user(arg, &ver, sizeof(ver)))
4086                 return -EFAULT;
4087
4088         return 0;
4089 }
4090
4091 static int get_array_info(mddev_t * mddev, void __user * arg)
4092 {
4093         mdu_array_info_t info;
4094         int nr,working,active,failed,spare;
4095         mdk_rdev_t *rdev;
4096         struct list_head *tmp;
4097
4098         nr=working=active=failed=spare=0;
4099         rdev_for_each(rdev, tmp, mddev) {
4100                 nr++;
4101                 if (test_bit(Faulty, &rdev->flags))
4102                         failed++;
4103                 else {
4104                         working++;
4105                         if (test_bit(In_sync, &rdev->flags))
4106                                 active++;
4107                         else
4108                                 spare++;
4109                 }
4110         }
4111
4112         info.major_version = mddev->major_version;
4113         info.minor_version = mddev->minor_version;
4114         info.patch_version = MD_PATCHLEVEL_VERSION;
4115         info.ctime         = mddev->ctime;
4116         info.level         = mddev->level;
4117         info.size          = mddev->size;
4118         if (info.size != mddev->size) /* overflow */
4119                 info.size = -1;
4120         info.nr_disks      = nr;
4121         info.raid_disks    = mddev->raid_disks;
4122         info.md_minor      = mddev->md_minor;
4123         info.not_persistent= !mddev->persistent;
4124
4125         info.utime         = mddev->utime;
4126         info.state         = 0;
4127         if (mddev->in_sync)
4128                 info.state = (1<<MD_SB_CLEAN);
4129         if (mddev->bitmap && mddev->bitmap_offset)
4130                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4131         info.active_disks  = active;
4132         info.working_disks = working;
4133         info.failed_disks  = failed;
4134         info.spare_disks   = spare;
4135
4136         info.layout        = mddev->layout;
4137         info.chunk_size    = mddev->chunk_size;
4138
4139         if (copy_to_user(arg, &info, sizeof(info)))
4140                 return -EFAULT;
4141
4142         return 0;
4143 }
4144
4145 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4146 {
4147         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4148         char *ptr, *buf = NULL;
4149         int err = -ENOMEM;
4150
4151         if (md_allow_write(mddev))
4152                 file = kmalloc(sizeof(*file), GFP_NOIO);
4153         else
4154                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4155
4156         if (!file)
4157                 goto out;
4158
4159         /* bitmap disabled, zero the first byte and copy out */
4160         if (!mddev->bitmap || !mddev->bitmap->file) {
4161                 file->pathname[0] = '\0';
4162                 goto copy_out;
4163         }
4164
4165         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4166         if (!buf)
4167                 goto out;
4168
4169         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4170         if (IS_ERR(ptr))
4171                 goto out;
4172
4173         strcpy(file->pathname, ptr);
4174
4175 copy_out:
4176         err = 0;
4177         if (copy_to_user(arg, file, sizeof(*file)))
4178                 err = -EFAULT;
4179 out:
4180         kfree(buf);
4181         kfree(file);
4182         return err;
4183 }
4184
4185 static int get_disk_info(mddev_t * mddev, void __user * arg)
4186 {
4187         mdu_disk_info_t info;
4188         mdk_rdev_t *rdev;
4189
4190         if (copy_from_user(&info, arg, sizeof(info)))
4191                 return -EFAULT;
4192
4193         rdev = find_rdev_nr(mddev, info.number);
4194         if (rdev) {
4195                 info.major = MAJOR(rdev->bdev->bd_dev);
4196                 info.minor = MINOR(rdev->bdev->bd_dev);
4197                 info.raid_disk = rdev->raid_disk;
4198                 info.state = 0;
4199                 if (test_bit(Faulty, &rdev->flags))
4200                         info.state |= (1<<MD_DISK_FAULTY);
4201                 else if (test_bit(In_sync, &rdev->flags)) {
4202                         info.state |= (1<<MD_DISK_ACTIVE);
4203                         info.state |= (1<<MD_DISK_SYNC);
4204                 }
4205                 if (test_bit(WriteMostly, &rdev->flags))
4206                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4207         } else {
4208                 info.major = info.minor = 0;
4209                 info.raid_disk = -1;
4210                 info.state = (1<<MD_DISK_REMOVED);
4211         }
4212
4213         if (copy_to_user(arg, &info, sizeof(info)))
4214                 return -EFAULT;
4215
4216         return 0;
4217 }
4218
4219 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4220 {
4221         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4222         mdk_rdev_t *rdev;
4223         dev_t dev = MKDEV(info->major,info->minor);
4224
4225         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4226                 return -EOVERFLOW;
4227
4228         if (!mddev->raid_disks) {
4229                 int err;
4230                 /* expecting a device which has a superblock */
4231                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4232                 if (IS_ERR(rdev)) {
4233                         printk(KERN_WARNING
4234                                 "md: md_import_device returned %ld\n",
4235                                 PTR_ERR(rdev));
4236                         return PTR_ERR(rdev);
4237                 }
4238                 if (!list_empty(&mddev->disks)) {
4239                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4240                                                         mdk_rdev_t, same_set);
4241                         int err = super_types[mddev->major_version]
4242                                 .load_super(rdev, rdev0, mddev->minor_version);
4243                         if (err < 0) {
4244                                 printk(KERN_WARNING
4245                                         "md: %s has different UUID to %s\n",
4246                                         bdevname(rdev->bdev,b),
4247                                         bdevname(rdev0->bdev,b2));
4248                                 export_rdev(rdev);
4249                                 return -EINVAL;
4250                         }
4251                 }
4252                 err = bind_rdev_to_array(rdev, mddev);
4253                 if (err)
4254                         export_rdev(rdev);
4255                 return err;
4256         }
4257
4258         /*
4259          * add_new_disk can be used once the array is assembled
4260          * to add "hot spares".  They must already have a superblock
4261          * written
4262          */
4263         if (mddev->pers) {
4264                 int err;
4265                 if (!mddev->pers->hot_add_disk) {
4266                         printk(KERN_WARNING
4267                                 "%s: personality does not support diskops!\n",
4268                                mdname(mddev));
4269                         return -EINVAL;
4270                 }
4271                 if (mddev->persistent)
4272                         rdev = md_import_device(dev, mddev->major_version,
4273                                                 mddev->minor_version);
4274                 else
4275                         rdev = md_import_device(dev, -1, -1);
4276                 if (IS_ERR(rdev)) {
4277                         printk(KERN_WARNING
4278                                 "md: md_import_device returned %ld\n",
4279                                 PTR_ERR(rdev));
4280                         return PTR_ERR(rdev);
4281                 }
4282                 /* set save_raid_disk if appropriate */
4283                 if (!mddev->persistent) {
4284                         if (info->state & (1<<MD_DISK_SYNC)  &&
4285                             info->raid_disk < mddev->raid_disks)
4286                                 rdev->raid_disk = info->raid_disk;
4287                         else
4288                                 rdev->raid_disk = -1;
4289                 } else
4290                         super_types[mddev->major_version].
4291                                 validate_super(mddev, rdev);
4292                 rdev->saved_raid_disk = rdev->raid_disk;
4293
4294                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4295                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4296                         set_bit(WriteMostly, &rdev->flags);
4297
4298                 rdev->raid_disk = -1;
4299                 err = bind_rdev_to_array(rdev, mddev);
4300                 if (!err && !mddev->pers->hot_remove_disk) {
4301                         /* If there is hot_add_disk but no hot_remove_disk
4302                          * then added disks for geometry changes,
4303                          * and should be added immediately.
4304                          */
4305                         super_types[mddev->major_version].
4306                                 validate_super(mddev, rdev);
4307                         err = mddev->pers->hot_add_disk(mddev, rdev);
4308                         if (err)
4309                                 unbind_rdev_from_array(rdev);
4310                 }
4311                 if (err)
4312                         export_rdev(rdev);
4313                 else
4314                         sysfs_notify(&rdev->kobj, NULL, "state");
4315
4316                 md_update_sb(mddev, 1);
4317                 if (mddev->degraded)
4318                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4319                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4320                 md_wakeup_thread(mddev->thread);
4321                 return err;
4322         }
4323
4324         /* otherwise, add_new_disk is only allowed
4325          * for major_version==0 superblocks
4326          */
4327         if (mddev->major_version != 0) {
4328                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4329                        mdname(mddev));
4330                 return -EINVAL;
4331         }
4332
4333         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4334                 int err;
4335                 rdev = md_import_device (dev, -1, 0);
4336                 if (IS_ERR(rdev)) {
4337                         printk(KERN_WARNING
4338                                 "md: error, md_import_device() returned %ld\n",
4339                                 PTR_ERR(rdev));
4340                         return PTR_ERR(rdev);
4341                 }
4342                 rdev->desc_nr = info->number;
4343                 if (info->raid_disk < mddev->raid_disks)
4344                         rdev->raid_disk = info->raid_disk;
4345                 else
4346                         rdev->raid_disk = -1;
4347
4348                 if (rdev->raid_disk < mddev->raid_disks)
4349                         if (info->state & (1<<MD_DISK_SYNC))
4350                                 set_bit(In_sync, &rdev->flags);
4351
4352                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4353                         set_bit(WriteMostly, &rdev->flags);
4354
4355                 if (!mddev->persistent) {
4356                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4357                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4358                 } else
4359                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4360                 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4361
4362                 err = bind_rdev_to_array(rdev, mddev);
4363                 if (err) {
4364                         export_rdev(rdev);
4365                         return err;
4366                 }
4367         }
4368
4369         return 0;
4370 }
4371
4372 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4373 {
4374         char b[BDEVNAME_SIZE];
4375         mdk_rdev_t *rdev;
4376
4377         rdev = find_rdev(mddev, dev);
4378         if (!rdev)
4379                 return -ENXIO;
4380
4381         if (rdev->raid_disk >= 0)
4382                 goto busy;
4383
4384         kick_rdev_from_array(rdev);
4385         md_update_sb(mddev, 1);
4386         md_new_event(mddev);
4387
4388         return 0;
4389 busy:
4390         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4391                 bdevname(rdev->bdev,b), mdname(mddev));
4392         return -EBUSY;
4393 }
4394
4395 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4396 {
4397         char b[BDEVNAME_SIZE];
4398         int err;
4399         mdk_rdev_t *rdev;
4400
4401         if (!mddev->pers)
4402                 return -ENODEV;
4403
4404         if (mddev->major_version != 0) {
4405                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4406                         " version-0 superblocks.\n",
4407                         mdname(mddev));
4408                 return -EINVAL;
4409         }
4410         if (!mddev->pers->hot_add_disk) {
4411                 printk(KERN_WARNING
4412                         "%s: personality does not support diskops!\n",
4413                         mdname(mddev));
4414                 return -EINVAL;
4415         }
4416
4417         rdev = md_import_device (dev, -1, 0);
4418         if (IS_ERR(rdev)) {
4419                 printk(KERN_WARNING
4420                         "md: error, md_import_device() returned %ld\n",
4421                         PTR_ERR(rdev));
4422                 return -EINVAL;
4423         }
4424
4425         if (mddev->persistent)
4426                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4427         else
4428                 rdev->sb_offset =
4429                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4430
4431         rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4432
4433         if (test_bit(Faulty, &rdev->flags)) {
4434                 printk(KERN_WARNING
4435                         "md: can not hot-add faulty %s disk to %s!\n",
4436                         bdevname(rdev->bdev,b), mdname(mddev));
4437                 err = -EINVAL;
4438                 goto abort_export;
4439         }
4440         clear_bit(In_sync, &rdev->flags);
4441         rdev->desc_nr = -1;
4442         rdev->saved_raid_disk = -1;
4443         err = bind_rdev_to_array(rdev, mddev);
4444         if (err)
4445                 goto abort_export;
4446
4447         /*
4448          * The rest should better be atomic, we can have disk failures
4449          * noticed in interrupt contexts ...
4450          */
4451
4452         if (rdev->desc_nr == mddev->max_disks) {
4453                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4454                         mdname(mddev));
4455                 err = -EBUSY;
4456                 goto abort_unbind_export;
4457         }
4458
4459         rdev->raid_disk = -1;
4460
4461         md_update_sb(mddev, 1);
4462
4463         /*
4464          * Kick recovery, maybe this spare has to be added to the
4465          * array immediately.
4466          */
4467         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4468         md_wakeup_thread(mddev->thread);
4469         md_new_event(mddev);
4470         return 0;
4471
4472 abort_unbind_export:
4473         unbind_rdev_from_array(rdev);
4474
4475 abort_export:
4476         export_rdev(rdev);
4477         return err;
4478 }
4479
4480 static int set_bitmap_file(mddev_t *mddev, int fd)
4481 {
4482         int err;
4483
4484         if (mddev->pers) {
4485                 if (!mddev->pers->quiesce)
4486                         return -EBUSY;
4487                 if (mddev->recovery || mddev->sync_thread)
4488                         return -EBUSY;
4489                 /* we should be able to change the bitmap.. */
4490         }
4491
4492
4493         if (fd >= 0) {
4494                 if (mddev->bitmap)
4495                         return -EEXIST; /* cannot add when bitmap is present */
4496                 mddev->bitmap_file = fget(fd);
4497
4498                 if (mddev->bitmap_file == NULL) {
4499                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4500                                mdname(mddev));
4501                         return -EBADF;
4502                 }
4503
4504                 err = deny_bitmap_write_access(mddev->bitmap_file);
4505                 if (err) {
4506                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4507                                mdname(mddev));
4508                         fput(mddev->bitmap_file);
4509                         mddev->bitmap_file = NULL;
4510                         return err;
4511                 }
4512                 mddev->bitmap_offset = 0; /* file overrides offset */
4513         } else if (mddev->bitmap == NULL)
4514                 return -ENOENT; /* cannot remove what isn't there */
4515         err = 0;
4516         if (mddev->pers) {
4517                 mddev->pers->quiesce(mddev, 1);
4518                 if (fd >= 0)
4519                         err = bitmap_create(mddev);
4520                 if (fd < 0 || err) {
4521                         bitmap_destroy(mddev);
4522                         fd = -1; /* make sure to put the file */
4523                 }
4524                 mddev->pers->quiesce(mddev, 0);
4525         }
4526         if (fd < 0) {
4527                 if (mddev->bitmap_file) {
4528                         restore_bitmap_write_access(mddev->bitmap_file);
4529                         fput(mddev->bitmap_file);
4530                 }
4531                 mddev->bitmap_file = NULL;
4532         }
4533
4534         return err;
4535 }
4536
4537 /*
4538  * set_array_info is used two different ways
4539  * The original usage is when creating a new array.
4540  * In this usage, raid_disks is > 0 and it together with
4541  *  level, size, not_persistent,layout,chunksize determine the
4542  *  shape of the array.
4543  *  This will always create an array with a type-0.90.0 superblock.
4544  * The newer usage is when assembling an array.
4545  *  In this case raid_disks will be 0, and the major_version field is
4546  *  use to determine which style super-blocks are to be found on the devices.
4547  *  The minor and patch _version numbers are also kept incase the
4548  *  super_block handler wishes to interpret them.
4549  */
4550 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4551 {
4552
4553         if (info->raid_disks == 0) {
4554                 /* just setting version number for superblock loading */
4555                 if (info->major_version < 0 ||
4556                     info->major_version >= ARRAY_SIZE(super_types) ||
4557                     super_types[info->major_version].name == NULL) {
4558                         /* maybe try to auto-load a module? */
4559                         printk(KERN_INFO
4560                                 "md: superblock version %d not known\n",
4561                                 info->major_version);
4562                         return -EINVAL;
4563                 }
4564                 mddev->major_version = info->major_version;
4565                 mddev->minor_version = info->minor_version;
4566                 mddev->patch_version = info->patch_version;
4567                 mddev->persistent = !info->not_persistent;
4568                 return 0;
4569         }
4570         mddev->major_version = MD_MAJOR_VERSION;
4571         mddev->minor_version = MD_MINOR_VERSION;
4572         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4573         mddev->ctime         = get_seconds();
4574
4575         mddev->level         = info->level;
4576         mddev->clevel[0]     = 0;
4577         mddev->size          = info->size;
4578         mddev->raid_disks    = info->raid_disks;
4579         /* don't set md_minor, it is determined by which /dev/md* was
4580          * openned
4581          */
4582         if (info->state & (1<<MD_SB_CLEAN))
4583                 mddev->recovery_cp = MaxSector;
4584         else
4585                 mddev->recovery_cp = 0;
4586         mddev->persistent    = ! info->not_persistent;
4587         mddev->external      = 0;
4588
4589         mddev->layout        = info->layout;
4590         mddev->chunk_size    = info->chunk_size;
4591
4592         mddev->max_disks     = MD_SB_DISKS;
4593
4594         if (mddev->persistent)
4595                 mddev->flags         = 0;
4596         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4597
4598         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4599         mddev->bitmap_offset = 0;
4600
4601         mddev->reshape_position = MaxSector;
4602
4603         /*
4604          * Generate a 128 bit UUID
4605          */
4606         get_random_bytes(mddev->uuid, 16);
4607
4608         mddev->new_level = mddev->level;
4609         mddev->new_chunk = mddev->chunk_size;
4610         mddev->new_layout = mddev->layout;
4611         mddev->delta_disks = 0;
4612
4613         return 0;
4614 }
4615
4616 static int update_size(mddev_t *mddev, sector_t num_sectors)
4617 {
4618         mdk_rdev_t * rdev;
4619         int rv;
4620         struct list_head *tmp;
4621         int fit = (num_sectors == 0);
4622
4623         if (mddev->pers->resize == NULL)
4624                 return -EINVAL;
4625         /* The "num_sectors" is the number of sectors of each device that
4626          * is used.  This can only make sense for arrays with redundancy.
4627          * linear and raid0 always use whatever space is available. We can only
4628          * consider changing this number if no resync or reconstruction is
4629          * happening, and if the new size is acceptable. It must fit before the
4630          * sb_offset or, if that is <data_offset, it must fit before the size
4631          * of each device.  If num_sectors is zero, we find the largest size
4632          * that fits.
4633
4634          */
4635         if (mddev->sync_thread)
4636                 return -EBUSY;
4637         rdev_for_each(rdev, tmp, mddev) {
4638                 sector_t avail;
4639                 avail = rdev->size * 2;
4640
4641                 if (fit && (num_sectors == 0 || num_sectors > avail))
4642                         num_sectors = avail;
4643                 if (avail < num_sectors)
4644                         return -ENOSPC;
4645         }
4646         rv = mddev->pers->resize(mddev, num_sectors);
4647         if (!rv) {
4648                 struct block_device *bdev;
4649
4650                 bdev = bdget_disk(mddev->gendisk, 0);
4651                 if (bdev) {
4652                         mutex_lock(&bdev->bd_inode->i_mutex);
4653                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4654                         mutex_unlock(&bdev->bd_inode->i_mutex);
4655                         bdput(bdev);
4656                 }
4657         }
4658         return rv;
4659 }
4660
4661 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4662 {
4663         int rv;
4664         /* change the number of raid disks */
4665         if (mddev->pers->check_reshape == NULL)
4666                 return -EINVAL;
4667         if (raid_disks <= 0 ||
4668             raid_disks >= mddev->max_disks)
4669                 return -EINVAL;
4670         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4671                 return -EBUSY;
4672         mddev->delta_disks = raid_disks - mddev->raid_disks;
4673
4674         rv = mddev->pers->check_reshape(mddev);
4675         return rv;
4676 }
4677
4678
4679 /*
4680  * update_array_info is used to change the configuration of an
4681  * on-line array.
4682  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4683  * fields in the info are checked against the array.
4684  * Any differences that cannot be handled will cause an error.
4685  * Normally, only one change can be managed at a time.
4686  */
4687 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4688 {
4689         int rv = 0;
4690         int cnt = 0;
4691         int state = 0;
4692
4693         /* calculate expected state,ignoring low bits */
4694         if (mddev->bitmap && mddev->bitmap_offset)
4695                 state |= (1 << MD_SB_BITMAP_PRESENT);
4696
4697         if (mddev->major_version != info->major_version ||
4698             mddev->minor_version != info->minor_version ||
4699 /*          mddev->patch_version != info->patch_version || */
4700             mddev->ctime         != info->ctime         ||
4701             mddev->level         != info->level         ||
4702 /*          mddev->layout        != info->layout        || */
4703             !mddev->persistent   != info->not_persistent||
4704             mddev->chunk_size    != info->chunk_size    ||
4705             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4706             ((state^info->state) & 0xfffffe00)
4707                 )
4708                 return -EINVAL;
4709         /* Check there is only one change */
4710         if (info->size >= 0 && mddev->size != info->size) cnt++;
4711         if (mddev->raid_disks != info->raid_disks) cnt++;
4712         if (mddev->layout != info->layout) cnt++;
4713         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4714         if (cnt == 0) return 0;
4715         if (cnt > 1) return -EINVAL;
4716
4717         if (mddev->layout != info->layout) {
4718                 /* Change layout
4719                  * we don't need to do anything at the md level, the
4720                  * personality will take care of it all.
4721                  */
4722                 if (mddev->pers->reconfig == NULL)
4723                         return -EINVAL;
4724                 else
4725                         return mddev->pers->reconfig(mddev, info->layout, -1);
4726         }
4727         if (info->size >= 0 && mddev->size != info->size)
4728                 rv = update_size(mddev, (sector_t)info->size * 2);
4729
4730         if (mddev->raid_disks    != info->raid_disks)
4731                 rv = update_raid_disks(mddev, info->raid_disks);
4732
4733         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4734                 if (mddev->pers->quiesce == NULL)
4735                         return -EINVAL;
4736                 if (mddev->recovery || mddev->sync_thread)
4737                         return -EBUSY;
4738                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4739                         /* add the bitmap */
4740                         if (mddev->bitmap)
4741                                 return -EEXIST;
4742                         if (mddev->default_bitmap_offset == 0)
4743                                 return -EINVAL;
4744                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4745                         mddev->pers->quiesce(mddev, 1);
4746                         rv = bitmap_create(mddev);
4747                         if (rv)
4748                                 bitmap_destroy(mddev);
4749                         mddev->pers->quiesce(mddev, 0);
4750                 } else {
4751                         /* remove the bitmap */
4752                         if (!mddev->bitmap)
4753                                 return -ENOENT;
4754                         if (mddev->bitmap->file)
4755                                 return -EINVAL;
4756                         mddev->pers->quiesce(mddev, 1);
4757                         bitmap_destroy(mddev);
4758                         mddev->pers->quiesce(mddev, 0);
4759                         mddev->bitmap_offset = 0;
4760                 }
4761         }
4762         md_update_sb(mddev, 1);
4763         return rv;
4764 }
4765
4766 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4767 {
4768         mdk_rdev_t *rdev;
4769
4770         if (mddev->pers == NULL)
4771                 return -ENODEV;
4772
4773         rdev = find_rdev(mddev, dev);
4774         if (!rdev)
4775                 return -ENODEV;
4776
4777         md_error(mddev, rdev);
4778         return 0;
4779 }
4780
4781 /*
4782  * We have a problem here : there is no easy way to give a CHS
4783  * virtual geometry. We currently pretend that we have a 2 heads
4784  * 4 sectors (with a BIG number of cylinders...). This drives
4785  * dosfs just mad... ;-)
4786  */
4787 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4788 {
4789         mddev_t *mddev = bdev->bd_disk->private_data;
4790
4791         geo->heads = 2;
4792         geo->sectors = 4;
4793         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4794         return 0;
4795 }
4796
4797 static int md_ioctl(struct inode *inode, struct file *file,
4798                         unsigned int cmd, unsigned long arg)
4799 {
4800         int err = 0;
4801         void __user *argp = (void __user *)arg;
4802         mddev_t *mddev = NULL;
4803
4804         if (!capable(CAP_SYS_ADMIN))
4805                 return -EACCES;
4806
4807         /*
4808          * Commands dealing with the RAID driver but not any
4809          * particular array:
4810          */
4811         switch (cmd)
4812         {
4813                 case RAID_VERSION:
4814                         err = get_version(argp);
4815                         goto done;
4816
4817                 case PRINT_RAID_DEBUG:
4818                         err = 0;
4819                         md_print_devices();
4820                         goto done;
4821
4822 #ifndef MODULE
4823                 case RAID_AUTORUN:
4824                         err = 0;
4825                         autostart_arrays(arg);
4826                         goto done;
4827 #endif
4828                 default:;
4829         }
4830
4831         /*
4832          * Commands creating/starting a new array:
4833          */
4834
4835         mddev = inode->i_bdev->bd_disk->private_data;
4836
4837         if (!mddev) {
4838                 BUG();
4839                 goto abort;
4840         }
4841
4842         err = mddev_lock(mddev);
4843         if (err) {
4844                 printk(KERN_INFO
4845                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4846                         err, cmd);
4847                 goto abort;
4848         }
4849
4850         switch (cmd)
4851         {
4852                 case SET_ARRAY_INFO:
4853                         {
4854                                 mdu_array_info_t info;
4855                                 if (!arg)
4856                                         memset(&info, 0, sizeof(info));
4857                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4858                                         err = -EFAULT;
4859                                         goto abort_unlock;
4860                                 }
4861                                 if (mddev->pers) {
4862                                         err = update_array_info(mddev, &info);
4863                                         if (err) {
4864                                                 printk(KERN_WARNING "md: couldn't update"
4865                                                        " array info. %d\n", err);
4866                                                 goto abort_unlock;
4867                                         }
4868                                         goto done_unlock;
4869                                 }
4870                                 if (!list_empty(&mddev->disks)) {
4871                                         printk(KERN_WARNING
4872                                                "md: array %s already has disks!\n",
4873                                                mdname(mddev));
4874                                         err = -EBUSY;
4875                                         goto abort_unlock;
4876                                 }
4877                                 if (mddev->raid_disks) {
4878                                         printk(KERN_WARNING
4879                                                "md: array %s already initialised!\n",
4880                                                mdname(mddev));
4881                                         err = -EBUSY;
4882                                         goto abort_unlock;
4883                                 }
4884                                 err = set_array_info(mddev, &info);
4885                                 if (err) {
4886                                         printk(KERN_WARNING "md: couldn't set"
4887                                                " array info. %d\n", err);
4888                                         goto abort_unlock;
4889                                 }
4890                         }
4891                         goto done_unlock;
4892
4893                 default:;
4894         }
4895
4896         /*
4897          * Commands querying/configuring an existing array:
4898          */
4899         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4900          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4901         if ((!mddev->raid_disks && !mddev->external)
4902             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4903             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4904             && cmd != GET_BITMAP_FILE) {
4905                 err = -ENODEV;
4906                 goto abort_unlock;
4907         }
4908
4909         /*
4910          * Commands even a read-only array can execute:
4911          */
4912         switch (cmd)
4913         {
4914                 case GET_ARRAY_INFO:
4915                         err = get_array_info(mddev, argp);
4916                         goto done_unlock;
4917
4918                 case GET_BITMAP_FILE:
4919                         err = get_bitmap_file(mddev, argp);
4920                         goto done_unlock;
4921
4922                 case GET_DISK_INFO:
4923                         err = get_disk_info(mddev, argp);
4924                         goto done_unlock;
4925
4926                 case RESTART_ARRAY_RW:
4927                         err = restart_array(mddev);
4928                         goto done_unlock;
4929
4930                 case STOP_ARRAY:
4931                         err = do_md_stop (mddev, 0, 1);
4932                         goto done_unlock;
4933
4934                 case STOP_ARRAY_RO:
4935                         err = do_md_stop (mddev, 1, 1);
4936                         goto done_unlock;
4937
4938         }
4939
4940         /*
4941          * The remaining ioctls are changing the state of the
4942          * superblock, so we do not allow them on read-only arrays.
4943          * However non-MD ioctls (e.g. get-size) will still come through
4944          * here and hit the 'default' below, so only disallow
4945          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4946          */
4947         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
4948                 if (mddev->ro == 2) {
4949                         mddev->ro = 0;
4950                         sysfs_notify(&mddev->kobj, NULL, "array_state");
4951                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4952                         md_wakeup_thread(mddev->thread);
4953                 } else {
4954                         err = -EROFS;
4955                         goto abort_unlock;
4956                 }
4957         }
4958
4959         switch (cmd)
4960         {
4961                 case ADD_NEW_DISK:
4962                 {
4963                         mdu_disk_info_t info;
4964                         if (copy_from_user(&info, argp, sizeof(info)))
4965                                 err = -EFAULT;
4966                         else
4967                                 err = add_new_disk(mddev, &info);
4968                         goto done_unlock;
4969                 }
4970
4971                 case HOT_REMOVE_DISK:
4972                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4973                         goto done_unlock;
4974
4975                 case HOT_ADD_DISK:
4976                         err = hot_add_disk(mddev, new_decode_dev(arg));
4977                         goto done_unlock;
4978
4979                 case SET_DISK_FAULTY:
4980                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4981                         goto done_unlock;
4982
4983                 case RUN_ARRAY:
4984                         err = do_md_run (mddev);
4985                         goto done_unlock;
4986
4987                 case SET_BITMAP_FILE:
4988                         err = set_bitmap_file(mddev, (int)arg);
4989                         goto done_unlock;
4990
4991                 default:
4992                         err = -EINVAL;
4993                         goto abort_unlock;
4994         }
4995
4996 done_unlock:
4997 abort_unlock:
4998         mddev_unlock(mddev);
4999
5000         return err;
5001 done:
5002         if (err)
5003                 MD_BUG();
5004 abort:
5005         return err;
5006 }
5007
5008 static int md_open(struct inode *inode, struct file *file)
5009 {
5010         /*
5011          * Succeed if we can lock the mddev, which confirms that
5012          * it isn't being stopped right now.
5013          */
5014         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5015         int err;
5016
5017         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5018                 goto out;
5019
5020         err = 0;
5021         mddev_get(mddev);
5022         mddev_unlock(mddev);
5023
5024         check_disk_change(inode->i_bdev);
5025  out:
5026         return err;
5027 }
5028
5029 static int md_release(struct inode *inode, struct file * file)
5030 {
5031         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5032
5033         BUG_ON(!mddev);
5034         mddev_put(mddev);
5035
5036         return 0;
5037 }
5038
5039 static int md_media_changed(struct gendisk *disk)
5040 {
5041         mddev_t *mddev = disk->private_data;
5042
5043         return mddev->changed;
5044 }
5045
5046 static int md_revalidate(struct gendisk *disk)
5047 {
5048         mddev_t *mddev = disk->private_data;
5049
5050         mddev->changed = 0;
5051         return 0;
5052 }
5053 static struct block_device_operations md_fops =
5054 {
5055         .owner          = THIS_MODULE,
5056         .open           = md_open,
5057         .release        = md_release,
5058         .ioctl          = md_ioctl,
5059         .getgeo         = md_getgeo,
5060         .media_changed  = md_media_changed,
5061         .revalidate_disk= md_revalidate,
5062 };
5063
5064 static int md_thread(void * arg)
5065 {
5066         mdk_thread_t *thread = arg;
5067
5068         /*
5069          * md_thread is a 'system-thread', it's priority should be very
5070          * high. We avoid resource deadlocks individually in each
5071          * raid personality. (RAID5 does preallocation) We also use RR and
5072          * the very same RT priority as kswapd, thus we will never get
5073          * into a priority inversion deadlock.
5074          *
5075          * we definitely have to have equal or higher priority than
5076          * bdflush, otherwise bdflush will deadlock if there are too
5077          * many dirty RAID5 blocks.
5078          */
5079
5080         allow_signal(SIGKILL);
5081         while (!kthread_should_stop()) {
5082
5083                 /* We need to wait INTERRUPTIBLE so that
5084                  * we don't add to the load-average.
5085                  * That means we need to be sure no signals are
5086                  * pending
5087                  */
5088                 if (signal_pending(current))
5089                         flush_signals(current);
5090
5091                 wait_event_interruptible_timeout
5092                         (thread->wqueue,
5093                          test_bit(THREAD_WAKEUP, &thread->flags)
5094                          || kthread_should_stop(),
5095                          thread->timeout);
5096
5097                 clear_bit(THREAD_WAKEUP, &thread->flags);
5098
5099                 thread->run(thread->mddev);
5100         }
5101
5102         return 0;
5103 }
5104
5105 void md_wakeup_thread(mdk_thread_t *thread)
5106 {
5107         if (thread) {
5108                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5109                 set_bit(THREAD_WAKEUP, &thread->flags);
5110                 wake_up(&thread->wqueue);
5111         }
5112 }
5113
5114 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5115                                  const char *name)
5116 {
5117         mdk_thread_t *thread;
5118
5119         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5120         if (!thread)
5121                 return NULL;
5122
5123         init_waitqueue_head(&thread->wqueue);
5124
5125         thread->run = run;
5126         thread->mddev = mddev;
5127         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5128         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5129         if (IS_ERR(thread->tsk)) {
5130                 kfree(thread);
5131                 return NULL;
5132         }
5133         return thread;
5134 }
5135
5136 void md_unregister_thread(mdk_thread_t *thread)
5137 {
5138         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5139
5140         kthread_stop(thread->tsk);
5141         kfree(thread);
5142 }
5143
5144 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5145 {
5146         if (!mddev) {
5147                 MD_BUG();
5148                 return;
5149         }
5150
5151         if (!rdev || test_bit(Faulty, &rdev->flags))
5152                 return;
5153
5154         if (mddev->external)
5155                 set_bit(Blocked, &rdev->flags);
5156 /*
5157         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5158                 mdname(mddev),
5159                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5160                 __builtin_return_address(0),__builtin_return_address(1),
5161                 __builtin_return_address(2),__builtin_return_address(3));
5162 */
5163         if (!mddev->pers)
5164                 return;
5165         if (!mddev->pers->error_handler)
5166                 return;
5167         mddev->pers->error_handler(mddev,rdev);
5168         if (mddev->degraded)
5169                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5170         set_bit(StateChanged, &rdev->flags);
5171         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5172         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5173         md_wakeup_thread(mddev->thread);
5174         md_new_event_inintr(mddev);
5175 }
5176
5177 /* seq_file implementation /proc/mdstat */
5178
5179 static void status_unused(struct seq_file *seq)
5180 {
5181         int i = 0;
5182         mdk_rdev_t *rdev;
5183         struct list_head *tmp;
5184
5185         seq_printf(seq, "unused devices: ");
5186
5187         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5188                 char b[BDEVNAME_SIZE];
5189                 i++;
5190                 seq_printf(seq, "%s ",
5191                               bdevname(rdev->bdev,b));
5192         }
5193         if (!i)
5194                 seq_printf(seq, "<none>");
5195
5196         seq_printf(seq, "\n");
5197 }
5198
5199
5200 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5201 {
5202         sector_t max_blocks, resync, res;
5203         unsigned long dt, db, rt;
5204         int scale;
5205         unsigned int per_milli;
5206
5207         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5208
5209         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5210                 max_blocks = mddev->resync_max_sectors >> 1;
5211         else
5212                 max_blocks = mddev->size;
5213
5214         /*
5215          * Should not happen.
5216          */
5217         if (!max_blocks) {
5218                 MD_BUG();
5219                 return;
5220         }
5221         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5222          * in a sector_t, and (max_blocks>>scale) will fit in a
5223          * u32, as those are the requirements for sector_div.
5224          * Thus 'scale' must be at least 10
5225          */
5226         scale = 10;
5227         if (sizeof(sector_t) > sizeof(unsigned long)) {
5228                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5229                         scale++;
5230         }
5231         res = (resync>>scale)*1000;
5232         sector_div(res, (u32)((max_blocks>>scale)+1));
5233
5234         per_milli = res;
5235         {
5236                 int i, x = per_milli/50, y = 20-x;
5237                 seq_printf(seq, "[");
5238                 for (i = 0; i < x; i++)
5239                         seq_printf(seq, "=");
5240                 seq_printf(seq, ">");
5241                 for (i = 0; i < y; i++)
5242                         seq_printf(seq, ".");
5243                 seq_printf(seq, "] ");
5244         }
5245         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5246                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5247                     "reshape" :
5248                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5249                      "check" :
5250                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5251                       "resync" : "recovery"))),
5252                    per_milli/10, per_milli % 10,
5253                    (unsigned long long) resync,
5254                    (unsigned long long) max_blocks);
5255
5256         /*
5257          * We do not want to overflow, so the order of operands and
5258          * the * 100 / 100 trick are important. We do a +1 to be
5259          * safe against division by zero. We only estimate anyway.
5260          *
5261          * dt: time from mark until now
5262          * db: blocks written from mark until now
5263          * rt: remaining time
5264          */
5265         dt = ((jiffies - mddev->resync_mark) / HZ);
5266         if (!dt) dt++;
5267         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5268                 - mddev->resync_mark_cnt;
5269         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5270
5271         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5272
5273         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5274 }
5275
5276 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5277 {
5278         struct list_head *tmp;
5279         loff_t l = *pos;
5280         mddev_t *mddev;
5281
5282         if (l >= 0x10000)
5283                 return NULL;
5284         if (!l--)
5285                 /* header */
5286                 return (void*)1;
5287
5288         spin_lock(&all_mddevs_lock);
5289         list_for_each(tmp,&all_mddevs)
5290                 if (!l--) {
5291                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5292                         mddev_get(mddev);
5293                         spin_unlock(&all_mddevs_lock);
5294                         return mddev;
5295                 }
5296         spin_unlock(&all_mddevs_lock);
5297         if (!l--)
5298                 return (void*)2;/* tail */
5299         return NULL;
5300 }
5301
5302 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5303 {
5304         struct list_head *tmp;
5305         mddev_t *next_mddev, *mddev = v;
5306
5307         ++*pos;
5308         if (v == (void*)2)
5309                 return NULL;
5310
5311         spin_lock(&all_mddevs_lock);
5312         if (v == (void*)1)
5313                 tmp = all_mddevs.next;
5314         else
5315                 tmp = mddev->all_mddevs.next;
5316         if (tmp != &all_mddevs)
5317                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5318         else {
5319                 next_mddev = (void*)2;
5320                 *pos = 0x10000;
5321         }
5322         spin_unlock(&all_mddevs_lock);
5323
5324         if (v != (void*)1)
5325                 mddev_put(mddev);
5326         return next_mddev;
5327
5328 }
5329
5330 static void md_seq_stop(struct seq_file *seq, void *v)
5331 {
5332         mddev_t *mddev = v;
5333
5334         if (mddev && v != (void*)1 && v != (void*)2)
5335                 mddev_put(mddev);
5336 }
5337
5338 struct mdstat_info {
5339         int event;
5340 };
5341
5342 static int md_seq_show(struct seq_file *seq, void *v)
5343 {
5344         mddev_t *mddev = v;
5345         sector_t size;
5346         struct list_head *tmp2;
5347         mdk_rdev_t *rdev;
5348         struct mdstat_info *mi = seq->private;
5349         struct bitmap *bitmap;
5350
5351         if (v == (void*)1) {
5352                 struct mdk_personality *pers;
5353                 seq_printf(seq, "Personalities : ");
5354                 spin_lock(&pers_lock);
5355                 list_for_each_entry(pers, &pers_list, list)
5356                         seq_printf(seq, "[%s] ", pers->name);
5357
5358                 spin_unlock(&pers_lock);
5359                 seq_printf(seq, "\n");
5360                 mi->event = atomic_read(&md_event_count);
5361                 return 0;
5362         }
5363         if (v == (void*)2) {
5364                 status_unused(seq);
5365                 return 0;
5366         }
5367
5368         if (mddev_lock(mddev) < 0)
5369                 return -EINTR;
5370
5371         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5372                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5373                                                 mddev->pers ? "" : "in");
5374                 if (mddev->pers) {
5375                         if (mddev->ro==1)
5376                                 seq_printf(seq, " (read-only)");
5377                         if (mddev->ro==2)
5378                                 seq_printf(seq, " (auto-read-only)");
5379                         seq_printf(seq, " %s", mddev->pers->name);
5380                 }
5381
5382                 size = 0;
5383                 rdev_for_each(rdev, tmp2, mddev) {
5384                         char b[BDEVNAME_SIZE];
5385                         seq_printf(seq, " %s[%d]",
5386                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5387                         if (test_bit(WriteMostly, &rdev->flags))
5388                                 seq_printf(seq, "(W)");
5389                         if (test_bit(Faulty, &rdev->flags)) {
5390                                 seq_printf(seq, "(F)");
5391                                 continue;
5392                         } else if (rdev->raid_disk < 0)
5393                                 seq_printf(seq, "(S)"); /* spare */
5394                         size += rdev->size;
5395                 }
5396
5397                 if (!list_empty(&mddev->disks)) {
5398                         if (mddev->pers)
5399                                 seq_printf(seq, "\n      %llu blocks",
5400                                         (unsigned long long)mddev->array_size);
5401                         else
5402                                 seq_printf(seq, "\n      %llu blocks",
5403                                         (unsigned long long)size);
5404                 }
5405                 if (mddev->persistent) {
5406                         if (mddev->major_version != 0 ||
5407                             mddev->minor_version != 90) {
5408                                 seq_printf(seq," super %d.%d",
5409                                            mddev->major_version,
5410                                            mddev->minor_version);
5411                         }
5412                 } else if (mddev->external)
5413                         seq_printf(seq, " super external:%s",
5414                                    mddev->metadata_type);
5415                 else
5416                         seq_printf(seq, " super non-persistent");
5417
5418                 if (mddev->pers) {
5419                         mddev->pers->status (seq, mddev);
5420                         seq_printf(seq, "\n      ");
5421                         if (mddev->pers->sync_request) {
5422                                 if (mddev->curr_resync > 2) {
5423                                         status_resync (seq, mddev);
5424                                         seq_printf(seq, "\n      ");
5425                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5426                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5427                                 else if (mddev->recovery_cp < MaxSector)
5428                                         seq_printf(seq, "\tresync=PENDING\n      ");
5429                         }
5430                 } else
5431                         seq_printf(seq, "\n       ");
5432
5433                 if ((bitmap = mddev->bitmap)) {
5434                         unsigned long chunk_kb;
5435                         unsigned long flags;
5436                         spin_lock_irqsave(&bitmap->lock, flags);
5437                         chunk_kb = bitmap->chunksize >> 10;
5438                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5439                                 "%lu%s chunk",
5440                                 bitmap->pages - bitmap->missing_pages,
5441                                 bitmap->pages,
5442                                 (bitmap->pages - bitmap->missing_pages)
5443                                         << (PAGE_SHIFT - 10),
5444                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5445                                 chunk_kb ? "KB" : "B");
5446                         if (bitmap->file) {
5447                                 seq_printf(seq, ", file: ");
5448                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5449                         }
5450
5451                         seq_printf(seq, "\n");
5452                         spin_unlock_irqrestore(&bitmap->lock, flags);
5453                 }
5454
5455                 seq_printf(seq, "\n");
5456         }
5457         mddev_unlock(mddev);
5458
5459         return 0;
5460 }
5461
5462 static struct seq_operations md_seq_ops = {
5463         .start  = md_seq_start,
5464         .next   = md_seq_next,
5465         .stop   = md_seq_stop,
5466         .show   = md_seq_show,
5467 };
5468
5469 static int md_seq_open(struct inode *inode, struct file *file)
5470 {
5471         int error;
5472         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5473         if (mi == NULL)
5474                 return -ENOMEM;
5475
5476         error = seq_open(file, &md_seq_ops);
5477         if (error)
5478                 kfree(mi);
5479         else {
5480                 struct seq_file *p = file->private_data;
5481                 p->private = mi;
5482                 mi->event = atomic_read(&md_event_count);
5483         }
5484         return error;
5485 }
5486
5487 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5488 {
5489         struct seq_file *m = filp->private_data;
5490         struct mdstat_info *mi = m->private;
5491         int mask;
5492
5493         poll_wait(filp, &md_event_waiters, wait);
5494
5495         /* always allow read */
5496         mask = POLLIN | POLLRDNORM;
5497
5498         if (mi->event != atomic_read(&md_event_count))
5499                 mask |= POLLERR | POLLPRI;
5500         return mask;
5501 }
5502
5503 static const struct file_operations md_seq_fops = {
5504         .owner          = THIS_MODULE,
5505         .open           = md_seq_open,
5506         .read           = seq_read,
5507         .llseek         = seq_lseek,
5508         .release        = seq_release_private,
5509         .poll           = mdstat_poll,
5510 };
5511
5512 int register_md_personality(struct mdk_personality *p)
5513 {
5514         spin_lock(&pers_lock);
5515         list_add_tail(&p->list, &pers_list);
5516         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5517         spin_unlock(&pers_lock);
5518         return 0;
5519 }
5520
5521 int unregister_md_personality(struct mdk_personality *p)
5522 {
5523         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5524         spin_lock(&pers_lock);
5525         list_del_init(&p->list);
5526         spin_unlock(&pers_lock);
5527         return 0;
5528 }
5529
5530 static int is_mddev_idle(mddev_t *mddev)
5531 {
5532         mdk_rdev_t * rdev;
5533         struct list_head *tmp;
5534         int idle;
5535         long curr_events;
5536
5537         idle = 1;
5538         rdev_for_each(rdev, tmp, mddev) {
5539                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5540                 curr_events = disk_stat_read(disk, sectors[0]) +
5541                                 disk_stat_read(disk, sectors[1]) -
5542                                 atomic_read(&disk->sync_io);
5543                 /* sync IO will cause sync_io to increase before the disk_stats
5544                  * as sync_io is counted when a request starts, and
5545                  * disk_stats is counted when it completes.
5546                  * So resync activity will cause curr_events to be smaller than
5547                  * when there was no such activity.
5548                  * non-sync IO will cause disk_stat to increase without
5549                  * increasing sync_io so curr_events will (eventually)
5550                  * be larger than it was before.  Once it becomes
5551                  * substantially larger, the test below will cause
5552                  * the array to appear non-idle, and resync will slow
5553                  * down.
5554                  * If there is a lot of outstanding resync activity when
5555                  * we set last_event to curr_events, then all that activity
5556                  * completing might cause the array to appear non-idle
5557                  * and resync will be slowed down even though there might
5558                  * not have been non-resync activity.  This will only
5559                  * happen once though.  'last_events' will soon reflect
5560                  * the state where there is little or no outstanding
5561                  * resync requests, and further resync activity will
5562                  * always make curr_events less than last_events.
5563                  *
5564                  */
5565                 if (curr_events - rdev->last_events > 4096) {
5566                         rdev->last_events = curr_events;
5567                         idle = 0;
5568                 }
5569         }
5570         return idle;
5571 }
5572
5573 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5574 {
5575         /* another "blocks" (512byte) blocks have been synced */
5576         atomic_sub(blocks, &mddev->recovery_active);
5577         wake_up(&mddev->recovery_wait);
5578         if (!ok) {
5579                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5580                 md_wakeup_thread(mddev->thread);
5581                 // stop recovery, signal do_sync ....
5582         }
5583 }
5584
5585
5586 /* md_write_start(mddev, bi)
5587  * If we need to update some array metadata (e.g. 'active' flag
5588  * in superblock) before writing, schedule a superblock update
5589  * and wait for it to complete.
5590  */
5591 void md_write_start(mddev_t *mddev, struct bio *bi)
5592 {
5593         int did_change = 0;
5594         if (bio_data_dir(bi) != WRITE)
5595                 return;
5596
5597         BUG_ON(mddev->ro == 1);
5598         if (mddev->ro == 2) {
5599                 /* need to switch to read/write */
5600                 mddev->ro = 0;
5601                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5602                 md_wakeup_thread(mddev->thread);
5603                 md_wakeup_thread(mddev->sync_thread);
5604                 did_change = 1;
5605         }
5606         atomic_inc(&mddev->writes_pending);
5607         if (mddev->safemode == 1)
5608                 mddev->safemode = 0;
5609         if (mddev->in_sync) {
5610                 spin_lock_irq(&mddev->write_lock);
5611                 if (mddev->in_sync) {
5612                         mddev->in_sync = 0;
5613                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5614                         md_wakeup_thread(mddev->thread);
5615                         did_change = 1;
5616                 }
5617                 spin_unlock_irq(&mddev->write_lock);
5618         }
5619         if (did_change)
5620                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5621         wait_event(mddev->sb_wait,
5622                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5623                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5624 }
5625
5626 void md_write_end(mddev_t *mddev)
5627 {
5628         if (atomic_dec_and_test(&mddev->writes_pending)) {
5629                 if (mddev->safemode == 2)
5630                         md_wakeup_thread(mddev->thread);
5631                 else if (mddev->safemode_delay)
5632                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5633         }
5634 }
5635
5636 /* md_allow_write(mddev)
5637  * Calling this ensures that the array is marked 'active' so that writes
5638  * may proceed without blocking.  It is important to call this before
5639  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5640  * Must be called with mddev_lock held.
5641  *
5642  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5643  * is dropped, so return -EAGAIN after notifying userspace.
5644  */
5645 int md_allow_write(mddev_t *mddev)
5646 {
5647         if (!mddev->pers)
5648                 return 0;
5649         if (mddev->ro)
5650                 return 0;
5651         if (!mddev->pers->sync_request)
5652                 return 0;
5653
5654         spin_lock_irq(&mddev->write_lock);
5655         if (mddev->in_sync) {
5656                 mddev->in_sync = 0;
5657                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5658                 if (mddev->safemode_delay &&
5659                     mddev->safemode == 0)
5660                         mddev->safemode = 1;
5661                 spin_unlock_irq(&mddev->write_lock);
5662                 md_update_sb(mddev, 0);
5663                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5664         } else
5665                 spin_unlock_irq(&mddev->write_lock);
5666
5667         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5668                 return -EAGAIN;
5669         else
5670                 return 0;
5671 }
5672 EXPORT_SYMBOL_GPL(md_allow_write);
5673
5674 #define SYNC_MARKS      10
5675 #define SYNC_MARK_STEP  (3*HZ)
5676 void md_do_sync(mddev_t *mddev)
5677 {
5678         mddev_t *mddev2;
5679         unsigned int currspeed = 0,
5680                  window;
5681         sector_t max_sectors,j, io_sectors;
5682         unsigned long mark[SYNC_MARKS];
5683         sector_t mark_cnt[SYNC_MARKS];
5684         int last_mark,m;
5685         struct list_head *tmp;
5686         sector_t last_check;
5687         int skipped = 0;
5688         struct list_head *rtmp;
5689         mdk_rdev_t *rdev;
5690         char *desc;
5691
5692         /* just incase thread restarts... */
5693         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5694                 return;
5695         if (mddev->ro) /* never try to sync a read-only array */
5696                 return;
5697
5698         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5699                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5700                         desc = "data-check";
5701                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5702                         desc = "requested-resync";
5703                 else
5704                         desc = "resync";
5705         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5706                 desc = "reshape";
5707         else
5708                 desc = "recovery";
5709
5710         /* we overload curr_resync somewhat here.
5711          * 0 == not engaged in resync at all
5712          * 2 == checking that there is no conflict with another sync
5713          * 1 == like 2, but have yielded to allow conflicting resync to
5714          *              commense
5715          * other == active in resync - this many blocks
5716          *
5717          * Before starting a resync we must have set curr_resync to
5718          * 2, and then checked that every "conflicting" array has curr_resync
5719          * less than ours.  When we find one that is the same or higher
5720          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5721          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5722          * This will mean we have to start checking from the beginning again.
5723          *
5724          */
5725
5726         do {
5727                 mddev->curr_resync = 2;
5728
5729         try_again:
5730                 if (kthread_should_stop()) {
5731                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5732                         goto skip;
5733                 }
5734                 for_each_mddev(mddev2, tmp) {
5735                         if (mddev2 == mddev)
5736                                 continue;
5737                         if (!mddev->parallel_resync
5738                         &&  mddev2->curr_resync
5739                         &&  match_mddev_units(mddev, mddev2)) {
5740                                 DEFINE_WAIT(wq);
5741                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5742                                         /* arbitrarily yield */
5743                                         mddev->curr_resync = 1;
5744                                         wake_up(&resync_wait);
5745                                 }
5746                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5747                                         /* no need to wait here, we can wait the next
5748                                          * time 'round when curr_resync == 2
5749                                          */
5750                                         continue;
5751                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5752                                 if (!kthread_should_stop() &&
5753                                     mddev2->curr_resync >= mddev->curr_resync) {
5754                                         printk(KERN_INFO "md: delaying %s of %s"
5755                                                " until %s has finished (they"
5756                                                " share one or more physical units)\n",
5757                                                desc, mdname(mddev), mdname(mddev2));
5758                                         mddev_put(mddev2);
5759                                         schedule();
5760                                         finish_wait(&resync_wait, &wq);
5761                                         goto try_again;
5762                                 }
5763                                 finish_wait(&resync_wait, &wq);
5764                         }
5765                 }
5766         } while (mddev->curr_resync < 2);
5767
5768         j = 0;
5769         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5770                 /* resync follows the size requested by the personality,
5771                  * which defaults to physical size, but can be virtual size
5772                  */
5773                 max_sectors = mddev->resync_max_sectors;
5774                 mddev->resync_mismatches = 0;
5775                 /* we don't use the checkpoint if there's a bitmap */
5776                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5777                         j = mddev->resync_min;
5778                 else if (!mddev->bitmap)
5779                         j = mddev->recovery_cp;
5780
5781         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5782                 max_sectors = mddev->size << 1;
5783         else {
5784                 /* recovery follows the physical size of devices */
5785                 max_sectors = mddev->size << 1;
5786                 j = MaxSector;
5787                 rdev_for_each(rdev, rtmp, mddev)
5788                         if (rdev->raid_disk >= 0 &&
5789                             !test_bit(Faulty, &rdev->flags) &&
5790                             !test_bit(In_sync, &rdev->flags) &&
5791                             rdev->recovery_offset < j)
5792                                 j = rdev->recovery_offset;
5793         }
5794
5795         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5796         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5797                 " %d KB/sec/disk.\n", speed_min(mddev));
5798         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5799                "(but not more than %d KB/sec) for %s.\n",
5800                speed_max(mddev), desc);
5801
5802         is_mddev_idle(mddev); /* this also initializes IO event counters */
5803
5804         io_sectors = 0;
5805         for (m = 0; m < SYNC_MARKS; m++) {
5806                 mark[m] = jiffies;
5807                 mark_cnt[m] = io_sectors;
5808         }
5809         last_mark = 0;
5810         mddev->resync_mark = mark[last_mark];
5811         mddev->resync_mark_cnt = mark_cnt[last_mark];
5812
5813         /*
5814          * Tune reconstruction:
5815          */
5816         window = 32*(PAGE_SIZE/512);
5817         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5818                 window/2,(unsigned long long) max_sectors/2);
5819
5820         atomic_set(&mddev->recovery_active, 0);
5821         last_check = 0;
5822
5823         if (j>2) {
5824                 printk(KERN_INFO
5825                        "md: resuming %s of %s from checkpoint.\n",
5826                        desc, mdname(mddev));
5827                 mddev->curr_resync = j;
5828         }
5829
5830         while (j < max_sectors) {
5831                 sector_t sectors;
5832
5833                 skipped = 0;
5834                 if (j >= mddev->resync_max) {
5835                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5836                         wait_event(mddev->recovery_wait,
5837                                    mddev->resync_max > j
5838                                    || kthread_should_stop());
5839                 }
5840                 if (kthread_should_stop())
5841                         goto interrupted;
5842                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5843                                                   currspeed < speed_min(mddev));
5844                 if (sectors == 0) {
5845                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5846                         goto out;
5847                 }
5848
5849                 if (!skipped) { /* actual IO requested */
5850                         io_sectors += sectors;
5851                         atomic_add(sectors, &mddev->recovery_active);
5852                 }
5853
5854                 j += sectors;
5855                 if (j>1) mddev->curr_resync = j;
5856                 mddev->curr_mark_cnt = io_sectors;
5857                 if (last_check == 0)
5858                         /* this is the earliers that rebuilt will be
5859                          * visible in /proc/mdstat
5860                          */
5861                         md_new_event(mddev);
5862
5863                 if (last_check + window > io_sectors || j == max_sectors)
5864                         continue;
5865
5866                 last_check = io_sectors;
5867
5868                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5869                         break;
5870
5871         repeat:
5872                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5873                         /* step marks */
5874                         int next = (last_mark+1) % SYNC_MARKS;
5875
5876                         mddev->resync_mark = mark[next];
5877                         mddev->resync_mark_cnt = mark_cnt[next];
5878                         mark[next] = jiffies;
5879                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5880                         last_mark = next;
5881                 }
5882
5883
5884                 if (kthread_should_stop())
5885                         goto interrupted;
5886
5887
5888                 /*
5889                  * this loop exits only if either when we are slower than
5890                  * the 'hard' speed limit, or the system was IO-idle for
5891                  * a jiffy.
5892                  * the system might be non-idle CPU-wise, but we only care
5893                  * about not overloading the IO subsystem. (things like an
5894                  * e2fsck being done on the RAID array should execute fast)
5895                  */
5896                 blk_unplug(mddev->queue);
5897                 cond_resched();
5898
5899                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5900                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5901
5902                 if (currspeed > speed_min(mddev)) {
5903                         if ((currspeed > speed_max(mddev)) ||
5904                                         !is_mddev_idle(mddev)) {
5905                                 msleep(500);
5906                                 goto repeat;
5907                         }
5908                 }
5909         }
5910         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5911         /*
5912          * this also signals 'finished resyncing' to md_stop
5913          */
5914  out:
5915         blk_unplug(mddev->queue);
5916
5917         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5918
5919         /* tell personality that we are finished */
5920         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5921
5922         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5923             mddev->curr_resync > 2) {
5924                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5925                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5926                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5927                                         printk(KERN_INFO
5928                                                "md: checkpointing %s of %s.\n",
5929                                                desc, mdname(mddev));
5930                                         mddev->recovery_cp = mddev->curr_resync;
5931                                 }
5932                         } else
5933                                 mddev->recovery_cp = MaxSector;
5934                 } else {
5935                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5936                                 mddev->curr_resync = MaxSector;
5937                         rdev_for_each(rdev, rtmp, mddev)
5938                                 if (rdev->raid_disk >= 0 &&
5939                                     !test_bit(Faulty, &rdev->flags) &&
5940                                     !test_bit(In_sync, &rdev->flags) &&
5941                                     rdev->recovery_offset < mddev->curr_resync)
5942                                         rdev->recovery_offset = mddev->curr_resync;
5943                 }
5944         }
5945         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5946
5947  skip:
5948         mddev->curr_resync = 0;
5949         mddev->resync_min = 0;
5950         mddev->resync_max = MaxSector;
5951         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5952         wake_up(&resync_wait);
5953         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5954         md_wakeup_thread(mddev->thread);
5955         return;
5956
5957  interrupted:
5958         /*
5959          * got a signal, exit.
5960          */
5961         printk(KERN_INFO
5962                "md: md_do_sync() got signal ... exiting\n");
5963         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5964         goto out;
5965
5966 }
5967 EXPORT_SYMBOL_GPL(md_do_sync);
5968
5969
5970 static int remove_and_add_spares(mddev_t *mddev)
5971 {
5972         mdk_rdev_t *rdev;
5973         struct list_head *rtmp;
5974         int spares = 0;
5975
5976         rdev_for_each(rdev, rtmp, mddev)
5977                 if (rdev->raid_disk >= 0 &&
5978                     !test_bit(Blocked, &rdev->flags) &&
5979                     (test_bit(Faulty, &rdev->flags) ||
5980                      ! test_bit(In_sync, &rdev->flags)) &&
5981                     atomic_read(&rdev->nr_pending)==0) {
5982                         if (mddev->pers->hot_remove_disk(
5983                                     mddev, rdev->raid_disk)==0) {
5984                                 char nm[20];
5985                                 sprintf(nm,"rd%d", rdev->raid_disk);
5986                                 sysfs_remove_link(&mddev->kobj, nm);
5987                                 rdev->raid_disk = -1;
5988                         }
5989                 }
5990
5991         if (mddev->degraded) {
5992                 rdev_for_each(rdev, rtmp, mddev) {
5993                         if (rdev->raid_disk >= 0 &&
5994                             !test_bit(In_sync, &rdev->flags))
5995                                 spares++;
5996                         if (rdev->raid_disk < 0
5997                             && !test_bit(Faulty, &rdev->flags)) {
5998                                 rdev->recovery_offset = 0;
5999                                 if (mddev->pers->
6000                                     hot_add_disk(mddev, rdev) == 0) {
6001                                         char nm[20];
6002                                         sprintf(nm, "rd%d", rdev->raid_disk);
6003                                         if (sysfs_create_link(&mddev->kobj,
6004                                                               &rdev->kobj, nm))
6005                                                 printk(KERN_WARNING
6006                                                        "md: cannot register "
6007                                                        "%s for %s\n",
6008                                                        nm, mdname(mddev));
6009                                         spares++;
6010                                         md_new_event(mddev);
6011                                 } else
6012                                         break;
6013                         }
6014                 }
6015         }
6016         return spares;
6017 }
6018 /*
6019  * This routine is regularly called by all per-raid-array threads to
6020  * deal with generic issues like resync and super-block update.
6021  * Raid personalities that don't have a thread (linear/raid0) do not
6022  * need this as they never do any recovery or update the superblock.
6023  *
6024  * It does not do any resync itself, but rather "forks" off other threads
6025  * to do that as needed.
6026  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6027  * "->recovery" and create a thread at ->sync_thread.
6028  * When the thread finishes it sets MD_RECOVERY_DONE
6029  * and wakeups up this thread which will reap the thread and finish up.
6030  * This thread also removes any faulty devices (with nr_pending == 0).
6031  *
6032  * The overall approach is:
6033  *  1/ if the superblock needs updating, update it.
6034  *  2/ If a recovery thread is running, don't do anything else.
6035  *  3/ If recovery has finished, clean up, possibly marking spares active.
6036  *  4/ If there are any faulty devices, remove them.
6037  *  5/ If array is degraded, try to add spares devices
6038  *  6/ If array has spares or is not in-sync, start a resync thread.
6039  */
6040 void md_check_recovery(mddev_t *mddev)
6041 {
6042         mdk_rdev_t *rdev;
6043         struct list_head *rtmp;
6044
6045
6046         if (mddev->bitmap)
6047                 bitmap_daemon_work(mddev->bitmap);
6048
6049         if (mddev->ro)
6050                 return;
6051
6052         if (signal_pending(current)) {
6053                 if (mddev->pers->sync_request && !mddev->external) {
6054                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6055                                mdname(mddev));
6056                         mddev->safemode = 2;
6057                 }
6058                 flush_signals(current);
6059         }
6060
6061         if ( ! (
6062                 (mddev->flags && !mddev->external) ||
6063                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6064                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6065                 (mddev->external == 0 && mddev->safemode == 1) ||
6066                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6067                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6068                 ))
6069                 return;
6070
6071         if (mddev_trylock(mddev)) {
6072                 int spares = 0;
6073
6074                 if (!mddev->external) {
6075                         int did_change = 0;
6076                         spin_lock_irq(&mddev->write_lock);
6077                         if (mddev->safemode &&
6078                             !atomic_read(&mddev->writes_pending) &&
6079                             !mddev->in_sync &&
6080                             mddev->recovery_cp == MaxSector) {
6081                                 mddev->in_sync = 1;
6082                                 did_change = 1;
6083                                 if (mddev->persistent)
6084                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6085                         }
6086                         if (mddev->safemode == 1)
6087                                 mddev->safemode = 0;
6088                         spin_unlock_irq(&mddev->write_lock);
6089                         if (did_change)
6090                                 sysfs_notify(&mddev->kobj, NULL, "array_state");
6091                 }
6092
6093                 if (mddev->flags)
6094                         md_update_sb(mddev, 0);
6095
6096                 rdev_for_each(rdev, rtmp, mddev)
6097                         if (test_and_clear_bit(StateChanged, &rdev->flags))
6098                                 sysfs_notify(&rdev->kobj, NULL, "state");
6099
6100
6101                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6102                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6103                         /* resync/recovery still happening */
6104                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6105                         goto unlock;
6106                 }
6107                 if (mddev->sync_thread) {
6108                         /* resync has finished, collect result */
6109                         md_unregister_thread(mddev->sync_thread);
6110                         mddev->sync_thread = NULL;
6111                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6112                                 /* success...*/
6113                                 /* activate any spares */
6114                                 if (mddev->pers->spare_active(mddev))
6115                                         sysfs_notify(&mddev->kobj, NULL,
6116                                                      "degraded");
6117                         }
6118                         md_update_sb(mddev, 1);
6119
6120                         /* if array is no-longer degraded, then any saved_raid_disk
6121                          * information must be scrapped
6122                          */
6123                         if (!mddev->degraded)
6124                                 rdev_for_each(rdev, rtmp, mddev)
6125                                         rdev->saved_raid_disk = -1;
6126
6127                         mddev->recovery = 0;
6128                         /* flag recovery needed just to double check */
6129                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6130                         sysfs_notify(&mddev->kobj, NULL, "sync_action");
6131                         md_new_event(mddev);
6132                         goto unlock;
6133                 }
6134                 /* Set RUNNING before clearing NEEDED to avoid
6135                  * any transients in the value of "sync_action".
6136                  */
6137                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6138                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6139                 /* Clear some bits that don't mean anything, but
6140                  * might be left set
6141                  */
6142                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6143                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6144
6145                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6146                         goto unlock;
6147                 /* no recovery is running.
6148                  * remove any failed drives, then
6149                  * add spares if possible.
6150                  * Spare are also removed and re-added, to allow
6151                  * the personality to fail the re-add.
6152                  */
6153
6154                 if (mddev->reshape_position != MaxSector) {
6155                         if (mddev->pers->check_reshape(mddev) != 0)
6156                                 /* Cannot proceed */
6157                                 goto unlock;
6158                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6159                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6160                 } else if ((spares = remove_and_add_spares(mddev))) {
6161                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6162                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6163                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6164                 } else if (mddev->recovery_cp < MaxSector) {
6165                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6166                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6167                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6168                         /* nothing to be done ... */
6169                         goto unlock;
6170
6171                 if (mddev->pers->sync_request) {
6172                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6173                                 /* We are adding a device or devices to an array
6174                                  * which has the bitmap stored on all devices.
6175                                  * So make sure all bitmap pages get written
6176                                  */
6177                                 bitmap_write_all(mddev->bitmap);
6178                         }
6179                         mddev->sync_thread = md_register_thread(md_do_sync,
6180                                                                 mddev,
6181                                                                 "%s_resync");
6182                         if (!mddev->sync_thread) {
6183                                 printk(KERN_ERR "%s: could not start resync"
6184                                         " thread...\n",
6185                                         mdname(mddev));
6186                                 /* leave the spares where they are, it shouldn't hurt */
6187                                 mddev->recovery = 0;
6188                         } else
6189                                 md_wakeup_thread(mddev->sync_thread);
6190                         sysfs_notify(&mddev->kobj, NULL, "sync_action");
6191                         md_new_event(mddev);
6192                 }
6193         unlock:
6194                 if (!mddev->sync_thread) {
6195                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6196                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6197                                                &mddev->recovery))
6198                                 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6199                 }
6200                 mddev_unlock(mddev);
6201         }
6202 }
6203
6204 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6205 {
6206         sysfs_notify(&rdev->kobj, NULL, "state");
6207         wait_event_timeout(rdev->blocked_wait,
6208                            !test_bit(Blocked, &rdev->flags),
6209                            msecs_to_jiffies(5000));
6210         rdev_dec_pending(rdev, mddev);
6211 }
6212 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6213
6214 static int md_notify_reboot(struct notifier_block *this,
6215                             unsigned long code, void *x)
6216 {
6217         struct list_head *tmp;
6218         mddev_t *mddev;
6219
6220         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6221
6222                 printk(KERN_INFO "md: stopping all md devices.\n");
6223
6224                 for_each_mddev(mddev, tmp)
6225                         if (mddev_trylock(mddev)) {
6226                                 do_md_stop (mddev, 1, 0);
6227                                 mddev_unlock(mddev);
6228                         }
6229                 /*
6230                  * certain more exotic SCSI devices are known to be
6231                  * volatile wrt too early system reboots. While the
6232                  * right place to handle this issue is the given
6233                  * driver, we do want to have a safe RAID driver ...
6234                  */
6235                 mdelay(1000*1);
6236         }
6237         return NOTIFY_DONE;
6238 }
6239
6240 static struct notifier_block md_notifier = {
6241         .notifier_call  = md_notify_reboot,
6242         .next           = NULL,
6243         .priority       = INT_MAX, /* before any real devices */
6244 };
6245
6246 static void md_geninit(void)
6247 {
6248         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6249
6250         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6251 }
6252
6253 static int __init md_init(void)
6254 {
6255         if (register_blkdev(MAJOR_NR, "md"))
6256                 return -1;
6257         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6258                 unregister_blkdev(MAJOR_NR, "md");
6259                 return -1;
6260         }
6261         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6262                             md_probe, NULL, NULL);
6263         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6264                             md_probe, NULL, NULL);
6265
6266         register_reboot_notifier(&md_notifier);
6267         raid_table_header = register_sysctl_table(raid_root_table);
6268
6269         md_geninit();
6270         return (0);
6271 }
6272
6273
6274 #ifndef MODULE
6275
6276 /*
6277  * Searches all registered partitions for autorun RAID arrays
6278  * at boot time.
6279  */
6280
6281 static LIST_HEAD(all_detected_devices);
6282 struct detected_devices_node {
6283         struct list_head list;
6284         dev_t dev;
6285 };
6286
6287 void md_autodetect_dev(dev_t dev)
6288 {
6289         struct detected_devices_node *node_detected_dev;
6290
6291         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6292         if (node_detected_dev) {
6293                 node_detected_dev->dev = dev;
6294                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6295         } else {
6296                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6297                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6298         }
6299 }
6300
6301
6302 static void autostart_arrays(int part)
6303 {
6304         mdk_rdev_t *rdev;
6305         struct detected_devices_node *node_detected_dev;
6306         dev_t dev;
6307         int i_scanned, i_passed;
6308
6309         i_scanned = 0;
6310         i_passed = 0;
6311
6312         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6313
6314         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6315                 i_scanned++;
6316                 node_detected_dev = list_entry(all_detected_devices.next,
6317                                         struct detected_devices_node, list);
6318                 list_del(&node_detected_dev->list);
6319                 dev = node_detected_dev->dev;
6320                 kfree(node_detected_dev);
6321                 rdev = md_import_device(dev,0, 90);
6322                 if (IS_ERR(rdev))
6323                         continue;
6324
6325                 if (test_bit(Faulty, &rdev->flags)) {
6326                         MD_BUG();
6327                         continue;
6328                 }
6329                 set_bit(AutoDetected, &rdev->flags);
6330                 list_add(&rdev->same_set, &pending_raid_disks);
6331                 i_passed++;
6332         }
6333
6334         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6335                                                 i_scanned, i_passed);
6336
6337         autorun_devices(part);
6338 }
6339
6340 #endif /* !MODULE */
6341
6342 static __exit void md_exit(void)
6343 {
6344         mddev_t *mddev;
6345         struct list_head *tmp;
6346
6347         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6348         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6349
6350         unregister_blkdev(MAJOR_NR,"md");
6351         unregister_blkdev(mdp_major, "mdp");
6352         unregister_reboot_notifier(&md_notifier);
6353         unregister_sysctl_table(raid_table_header);
6354         remove_proc_entry("mdstat", NULL);
6355         for_each_mddev(mddev, tmp) {
6356                 struct gendisk *disk = mddev->gendisk;
6357                 if (!disk)
6358                         continue;
6359                 export_array(mddev);
6360                 del_gendisk(disk);
6361                 put_disk(disk);
6362                 mddev->gendisk = NULL;
6363                 mddev_put(mddev);
6364         }
6365 }
6366
6367 subsys_initcall(md_init);
6368 module_exit(md_exit)
6369
6370 static int get_ro(char *buffer, struct kernel_param *kp)
6371 {
6372         return sprintf(buffer, "%d", start_readonly);
6373 }
6374 static int set_ro(const char *val, struct kernel_param *kp)
6375 {
6376         char *e;
6377         int num = simple_strtoul(val, &e, 10);
6378         if (*val && (*e == '\0' || *e == '\n')) {
6379                 start_readonly = num;
6380                 return 0;
6381         }
6382         return -EINVAL;
6383 }
6384
6385 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6386 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6387
6388
6389 EXPORT_SYMBOL(register_md_personality);
6390 EXPORT_SYMBOL(unregister_md_personality);
6391 EXPORT_SYMBOL(md_error);
6392 EXPORT_SYMBOL(md_done_sync);
6393 EXPORT_SYMBOL(md_write_start);
6394 EXPORT_SYMBOL(md_write_end);
6395 EXPORT_SYMBOL(md_register_thread);
6396 EXPORT_SYMBOL(md_unregister_thread);
6397 EXPORT_SYMBOL(md_wakeup_thread);
6398 EXPORT_SYMBOL(md_check_recovery);
6399 MODULE_LICENSE("GPL");
6400 MODULE_ALIAS("md");
6401 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);