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