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         if (!err)
1890                 sysfs_notify(&rdev->kobj, NULL, "state");
1891         return err ? err : len;
1892 }
1893 static struct rdev_sysfs_entry rdev_state =
1894 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1895
1896 static ssize_t
1897 errors_show(mdk_rdev_t *rdev, char *page)
1898 {
1899         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1900 }
1901
1902 static ssize_t
1903 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1904 {
1905         char *e;
1906         unsigned long n = simple_strtoul(buf, &e, 10);
1907         if (*buf && (*e == 0 || *e == '\n')) {
1908                 atomic_set(&rdev->corrected_errors, n);
1909                 return len;
1910         }
1911         return -EINVAL;
1912 }
1913 static struct rdev_sysfs_entry rdev_errors =
1914 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1915
1916 static ssize_t
1917 slot_show(mdk_rdev_t *rdev, char *page)
1918 {
1919         if (rdev->raid_disk < 0)
1920                 return sprintf(page, "none\n");
1921         else
1922                 return sprintf(page, "%d\n", rdev->raid_disk);
1923 }
1924
1925 static ssize_t
1926 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1927 {
1928         char *e;
1929         int err;
1930         char nm[20];
1931         int slot = simple_strtoul(buf, &e, 10);
1932         if (strncmp(buf, "none", 4)==0)
1933                 slot = -1;
1934         else if (e==buf || (*e && *e!= '\n'))
1935                 return -EINVAL;
1936         if (rdev->mddev->pers && slot == -1) {
1937                 /* Setting 'slot' on an active array requires also
1938                  * updating the 'rd%d' link, and communicating
1939                  * with the personality with ->hot_*_disk.
1940                  * For now we only support removing
1941                  * failed/spare devices.  This normally happens automatically,
1942                  * but not when the metadata is externally managed.
1943                  */
1944                 if (rdev->raid_disk == -1)
1945                         return -EEXIST;
1946                 /* personality does all needed checks */
1947                 if (rdev->mddev->pers->hot_add_disk == NULL)
1948                         return -EINVAL;
1949                 err = rdev->mddev->pers->
1950                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
1951                 if (err)
1952                         return err;
1953                 sprintf(nm, "rd%d", rdev->raid_disk);
1954                 sysfs_remove_link(&rdev->mddev->kobj, nm);
1955                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1956                 md_wakeup_thread(rdev->mddev->thread);
1957         } else if (rdev->mddev->pers) {
1958                 mdk_rdev_t *rdev2;
1959                 struct list_head *tmp;
1960                 /* Activating a spare .. or possibly reactivating
1961                  * if we every get bitmaps working here.
1962                  */
1963
1964                 if (rdev->raid_disk != -1)
1965                         return -EBUSY;
1966
1967                 if (rdev->mddev->pers->hot_add_disk == NULL)
1968                         return -EINVAL;
1969
1970                 rdev_for_each(rdev2, tmp, rdev->mddev)
1971                         if (rdev2->raid_disk == slot)
1972                                 return -EEXIST;
1973
1974                 rdev->raid_disk = slot;
1975                 if (test_bit(In_sync, &rdev->flags))
1976                         rdev->saved_raid_disk = slot;
1977                 else
1978                         rdev->saved_raid_disk = -1;
1979                 err = rdev->mddev->pers->
1980                         hot_add_disk(rdev->mddev, rdev);
1981                 if (err) {
1982                         rdev->raid_disk = -1;
1983                         return err;
1984                 } else
1985                         sysfs_notify(&rdev->kobj, NULL, "state");
1986                 sprintf(nm, "rd%d", rdev->raid_disk);
1987                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
1988                         printk(KERN_WARNING
1989                                "md: cannot register "
1990                                "%s for %s\n",
1991                                nm, mdname(rdev->mddev));
1992
1993                 /* don't wakeup anyone, leave that to userspace. */
1994         } else {
1995                 if (slot >= rdev->mddev->raid_disks)
1996                         return -ENOSPC;
1997                 rdev->raid_disk = slot;
1998                 /* assume it is working */
1999                 clear_bit(Faulty, &rdev->flags);
2000                 clear_bit(WriteMostly, &rdev->flags);
2001                 set_bit(In_sync, &rdev->flags);
2002                 sysfs_notify(&rdev->kobj, NULL, "state");
2003         }
2004         return len;
2005 }
2006
2007
2008 static struct rdev_sysfs_entry rdev_slot =
2009 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2010
2011 static ssize_t
2012 offset_show(mdk_rdev_t *rdev, char *page)
2013 {
2014         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2015 }
2016
2017 static ssize_t
2018 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2019 {
2020         char *e;
2021         unsigned long long offset = simple_strtoull(buf, &e, 10);
2022         if (e==buf || (*e && *e != '\n'))
2023                 return -EINVAL;
2024         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2025                 return -EBUSY;
2026         if (rdev->size && rdev->mddev->external)
2027                 /* Must set offset before size, so overlap checks
2028                  * can be sane */
2029                 return -EBUSY;
2030         rdev->data_offset = offset;
2031         return len;
2032 }
2033
2034 static struct rdev_sysfs_entry rdev_offset =
2035 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2036
2037 static ssize_t
2038 rdev_size_show(mdk_rdev_t *rdev, char *page)
2039 {
2040         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2041 }
2042
2043 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2044 {
2045         /* check if two start/length pairs overlap */
2046         if (s1+l1 <= s2)
2047                 return 0;
2048         if (s2+l2 <= s1)
2049                 return 0;
2050         return 1;
2051 }
2052
2053 static ssize_t
2054 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2055 {
2056         char *e;
2057         unsigned long long size = simple_strtoull(buf, &e, 10);
2058         unsigned long long oldsize = rdev->size;
2059         mddev_t *my_mddev = rdev->mddev;
2060
2061         if (e==buf || (*e && *e != '\n'))
2062                 return -EINVAL;
2063         if (my_mddev->pers && rdev->raid_disk >= 0)
2064                 return -EBUSY;
2065         rdev->size = size;
2066         if (size > oldsize && rdev->mddev->external) {
2067                 /* need to check that all other rdevs with the same ->bdev
2068                  * do not overlap.  We need to unlock the mddev to avoid
2069                  * a deadlock.  We have already changed rdev->size, and if
2070                  * we have to change it back, we will have the lock again.
2071                  */
2072                 mddev_t *mddev;
2073                 int overlap = 0;
2074                 struct list_head *tmp, *tmp2;
2075
2076                 mddev_unlock(my_mddev);
2077                 for_each_mddev(mddev, tmp) {
2078                         mdk_rdev_t *rdev2;
2079
2080                         mddev_lock(mddev);
2081                         rdev_for_each(rdev2, tmp2, mddev)
2082                                 if (test_bit(AllReserved, &rdev2->flags) ||
2083                                     (rdev->bdev == rdev2->bdev &&
2084                                      rdev != rdev2 &&
2085                                      overlaps(rdev->data_offset, rdev->size,
2086                                             rdev2->data_offset, rdev2->size))) {
2087                                         overlap = 1;
2088                                         break;
2089                                 }
2090                         mddev_unlock(mddev);
2091                         if (overlap) {
2092                                 mddev_put(mddev);
2093                                 break;
2094                         }
2095                 }
2096                 mddev_lock(my_mddev);
2097                 if (overlap) {
2098                         /* Someone else could have slipped in a size
2099                          * change here, but doing so is just silly.
2100                          * We put oldsize back because we *know* it is
2101                          * safe, and trust userspace not to race with
2102                          * itself
2103                          */
2104                         rdev->size = oldsize;
2105                         return -EBUSY;
2106                 }
2107         }
2108         if (size < my_mddev->size || my_mddev->size == 0)
2109                 my_mddev->size = size;
2110         return len;
2111 }
2112
2113 static struct rdev_sysfs_entry rdev_size =
2114 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2115
2116 static struct attribute *rdev_default_attrs[] = {
2117         &rdev_state.attr,
2118         &rdev_errors.attr,
2119         &rdev_slot.attr,
2120         &rdev_offset.attr,
2121         &rdev_size.attr,
2122         NULL,
2123 };
2124 static ssize_t
2125 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2126 {
2127         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2128         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2129         mddev_t *mddev = rdev->mddev;
2130         ssize_t rv;
2131
2132         if (!entry->show)
2133                 return -EIO;
2134
2135         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2136         if (!rv) {
2137                 if (rdev->mddev == NULL)
2138                         rv = -EBUSY;
2139                 else
2140                         rv = entry->show(rdev, page);
2141                 mddev_unlock(mddev);
2142         }
2143         return rv;
2144 }
2145
2146 static ssize_t
2147 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2148               const char *page, size_t length)
2149 {
2150         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2151         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2152         ssize_t rv;
2153         mddev_t *mddev = rdev->mddev;
2154
2155         if (!entry->store)
2156                 return -EIO;
2157         if (!capable(CAP_SYS_ADMIN))
2158                 return -EACCES;
2159         rv = mddev ? mddev_lock(mddev): -EBUSY;
2160         if (!rv) {
2161                 if (rdev->mddev == NULL)
2162                         rv = -EBUSY;
2163                 else
2164                         rv = entry->store(rdev, page, length);
2165                 mddev_unlock(mddev);
2166         }
2167         return rv;
2168 }
2169
2170 static void rdev_free(struct kobject *ko)
2171 {
2172         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2173         kfree(rdev);
2174 }
2175 static struct sysfs_ops rdev_sysfs_ops = {
2176         .show           = rdev_attr_show,
2177         .store          = rdev_attr_store,
2178 };
2179 static struct kobj_type rdev_ktype = {
2180         .release        = rdev_free,
2181         .sysfs_ops      = &rdev_sysfs_ops,
2182         .default_attrs  = rdev_default_attrs,
2183 };
2184
2185 /*
2186  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2187  *
2188  * mark the device faulty if:
2189  *
2190  *   - the device is nonexistent (zero size)
2191  *   - the device has no valid superblock
2192  *
2193  * a faulty rdev _never_ has rdev->sb set.
2194  */
2195 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2196 {
2197         char b[BDEVNAME_SIZE];
2198         int err;
2199         mdk_rdev_t *rdev;
2200         sector_t size;
2201
2202         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2203         if (!rdev) {
2204                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2205                 return ERR_PTR(-ENOMEM);
2206         }
2207
2208         if ((err = alloc_disk_sb(rdev)))
2209                 goto abort_free;
2210
2211         err = lock_rdev(rdev, newdev, super_format == -2);
2212         if (err)
2213                 goto abort_free;
2214
2215         kobject_init(&rdev->kobj, &rdev_ktype);
2216
2217         rdev->desc_nr = -1;
2218         rdev->saved_raid_disk = -1;
2219         rdev->raid_disk = -1;
2220         rdev->flags = 0;
2221         rdev->data_offset = 0;
2222         rdev->sb_events = 0;
2223         atomic_set(&rdev->nr_pending, 0);
2224         atomic_set(&rdev->read_errors, 0);
2225         atomic_set(&rdev->corrected_errors, 0);
2226
2227         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2228         if (!size) {
2229                 printk(KERN_WARNING
2230                         "md: %s has zero or unknown size, marking faulty!\n",
2231                         bdevname(rdev->bdev,b));
2232                 err = -EINVAL;
2233                 goto abort_free;
2234         }
2235
2236         if (super_format >= 0) {
2237                 err = super_types[super_format].
2238                         load_super(rdev, NULL, super_minor);
2239                 if (err == -EINVAL) {
2240                         printk(KERN_WARNING
2241                                 "md: %s does not have a valid v%d.%d "
2242                                "superblock, not importing!\n",
2243                                 bdevname(rdev->bdev,b),
2244                                super_format, super_minor);
2245                         goto abort_free;
2246                 }
2247                 if (err < 0) {
2248                         printk(KERN_WARNING
2249                                 "md: could not read %s's sb, not importing!\n",
2250                                 bdevname(rdev->bdev,b));
2251                         goto abort_free;
2252                 }
2253         }
2254
2255         INIT_LIST_HEAD(&rdev->same_set);
2256         init_waitqueue_head(&rdev->blocked_wait);
2257
2258         return rdev;
2259
2260 abort_free:
2261         if (rdev->sb_page) {
2262                 if (rdev->bdev)
2263                         unlock_rdev(rdev);
2264                 free_disk_sb(rdev);
2265         }
2266         kfree(rdev);
2267         return ERR_PTR(err);
2268 }
2269
2270 /*
2271  * Check a full RAID array for plausibility
2272  */
2273
2274
2275 static void analyze_sbs(mddev_t * mddev)
2276 {
2277         int i;
2278         struct list_head *tmp;
2279         mdk_rdev_t *rdev, *freshest;
2280         char b[BDEVNAME_SIZE];
2281
2282         freshest = NULL;
2283         rdev_for_each(rdev, tmp, mddev)
2284                 switch (super_types[mddev->major_version].
2285                         load_super(rdev, freshest, mddev->minor_version)) {
2286                 case 1:
2287                         freshest = rdev;
2288                         break;
2289                 case 0:
2290                         break;
2291                 default:
2292                         printk( KERN_ERR \
2293                                 "md: fatal superblock inconsistency in %s"
2294                                 " -- removing from array\n",
2295                                 bdevname(rdev->bdev,b));
2296                         kick_rdev_from_array(rdev);
2297                 }
2298
2299
2300         super_types[mddev->major_version].
2301                 validate_super(mddev, freshest);
2302
2303         i = 0;
2304         rdev_for_each(rdev, tmp, mddev) {
2305                 if (rdev != freshest)
2306                         if (super_types[mddev->major_version].
2307                             validate_super(mddev, rdev)) {
2308                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2309                                         " from array!\n",
2310                                         bdevname(rdev->bdev,b));
2311                                 kick_rdev_from_array(rdev);
2312                                 continue;
2313                         }
2314                 if (mddev->level == LEVEL_MULTIPATH) {
2315                         rdev->desc_nr = i++;
2316                         rdev->raid_disk = rdev->desc_nr;
2317                         set_bit(In_sync, &rdev->flags);
2318                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2319                         rdev->raid_disk = -1;
2320                         clear_bit(In_sync, &rdev->flags);
2321                 }
2322         }
2323
2324
2325
2326         if (mddev->recovery_cp != MaxSector &&
2327             mddev->level >= 1)
2328                 printk(KERN_ERR "md: %s: raid array is not clean"
2329                        " -- starting background reconstruction\n",
2330                        mdname(mddev));
2331
2332 }
2333
2334 static ssize_t
2335 safe_delay_show(mddev_t *mddev, char *page)
2336 {
2337         int msec = (mddev->safemode_delay*1000)/HZ;
2338         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2339 }
2340 static ssize_t
2341 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2342 {
2343         int scale=1;
2344         int dot=0;
2345         int i;
2346         unsigned long msec;
2347         char buf[30];
2348         char *e;
2349         /* remove a period, and count digits after it */
2350         if (len >= sizeof(buf))
2351                 return -EINVAL;
2352         strlcpy(buf, cbuf, len);
2353         buf[len] = 0;
2354         for (i=0; i<len; i++) {
2355                 if (dot) {
2356                         if (isdigit(buf[i])) {
2357                                 buf[i-1] = buf[i];
2358                                 scale *= 10;
2359                         }
2360                         buf[i] = 0;
2361                 } else if (buf[i] == '.') {
2362                         dot=1;
2363                         buf[i] = 0;
2364                 }
2365         }
2366         msec = simple_strtoul(buf, &e, 10);
2367         if (e == buf || (*e && *e != '\n'))
2368                 return -EINVAL;
2369         msec = (msec * 1000) / scale;
2370         if (msec == 0)
2371                 mddev->safemode_delay = 0;
2372         else {
2373                 mddev->safemode_delay = (msec*HZ)/1000;
2374                 if (mddev->safemode_delay == 0)
2375                         mddev->safemode_delay = 1;
2376         }
2377         return len;
2378 }
2379 static struct md_sysfs_entry md_safe_delay =
2380 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2381
2382 static ssize_t
2383 level_show(mddev_t *mddev, char *page)
2384 {
2385         struct mdk_personality *p = mddev->pers;
2386         if (p)
2387                 return sprintf(page, "%s\n", p->name);
2388         else if (mddev->clevel[0])
2389                 return sprintf(page, "%s\n", mddev->clevel);
2390         else if (mddev->level != LEVEL_NONE)
2391                 return sprintf(page, "%d\n", mddev->level);
2392         else
2393                 return 0;
2394 }
2395
2396 static ssize_t
2397 level_store(mddev_t *mddev, const char *buf, size_t len)
2398 {
2399         ssize_t rv = len;
2400         if (mddev->pers)
2401                 return -EBUSY;
2402         if (len == 0)
2403                 return 0;
2404         if (len >= sizeof(mddev->clevel))
2405                 return -ENOSPC;
2406         strncpy(mddev->clevel, buf, len);
2407         if (mddev->clevel[len-1] == '\n')
2408                 len--;
2409         mddev->clevel[len] = 0;
2410         mddev->level = LEVEL_NONE;
2411         return rv;
2412 }
2413
2414 static struct md_sysfs_entry md_level =
2415 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2416
2417
2418 static ssize_t
2419 layout_show(mddev_t *mddev, char *page)
2420 {
2421         /* just a number, not meaningful for all levels */
2422         if (mddev->reshape_position != MaxSector &&
2423             mddev->layout != mddev->new_layout)
2424                 return sprintf(page, "%d (%d)\n",
2425                                mddev->new_layout, mddev->layout);
2426         return sprintf(page, "%d\n", mddev->layout);
2427 }
2428
2429 static ssize_t
2430 layout_store(mddev_t *mddev, const char *buf, size_t len)
2431 {
2432         char *e;
2433         unsigned long n = simple_strtoul(buf, &e, 10);
2434
2435         if (!*buf || (*e && *e != '\n'))
2436                 return -EINVAL;
2437
2438         if (mddev->pers)
2439                 return -EBUSY;
2440         if (mddev->reshape_position != MaxSector)
2441                 mddev->new_layout = n;
2442         else
2443                 mddev->layout = n;
2444         return len;
2445 }
2446 static struct md_sysfs_entry md_layout =
2447 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2448
2449
2450 static ssize_t
2451 raid_disks_show(mddev_t *mddev, char *page)
2452 {
2453         if (mddev->raid_disks == 0)
2454                 return 0;
2455         if (mddev->reshape_position != MaxSector &&
2456             mddev->delta_disks != 0)
2457                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2458                                mddev->raid_disks - mddev->delta_disks);
2459         return sprintf(page, "%d\n", mddev->raid_disks);
2460 }
2461
2462 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2463
2464 static ssize_t
2465 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2466 {
2467         char *e;
2468         int rv = 0;
2469         unsigned long n = simple_strtoul(buf, &e, 10);
2470
2471         if (!*buf || (*e && *e != '\n'))
2472                 return -EINVAL;
2473
2474         if (mddev->pers)
2475                 rv = update_raid_disks(mddev, n);
2476         else if (mddev->reshape_position != MaxSector) {
2477                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2478                 mddev->delta_disks = n - olddisks;
2479                 mddev->raid_disks = n;
2480         } else
2481                 mddev->raid_disks = n;
2482         return rv ? rv : len;
2483 }
2484 static struct md_sysfs_entry md_raid_disks =
2485 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2486
2487 static ssize_t
2488 chunk_size_show(mddev_t *mddev, char *page)
2489 {
2490         if (mddev->reshape_position != MaxSector &&
2491             mddev->chunk_size != mddev->new_chunk)
2492                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2493                                mddev->chunk_size);
2494         return sprintf(page, "%d\n", mddev->chunk_size);
2495 }
2496
2497 static ssize_t
2498 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2499 {
2500         /* can only set chunk_size if array is not yet active */
2501         char *e;
2502         unsigned long n = simple_strtoul(buf, &e, 10);
2503
2504         if (!*buf || (*e && *e != '\n'))
2505                 return -EINVAL;
2506
2507         if (mddev->pers)
2508                 return -EBUSY;
2509         else if (mddev->reshape_position != MaxSector)
2510                 mddev->new_chunk = n;
2511         else
2512                 mddev->chunk_size = n;
2513         return len;
2514 }
2515 static struct md_sysfs_entry md_chunk_size =
2516 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2517
2518 static ssize_t
2519 resync_start_show(mddev_t *mddev, char *page)
2520 {
2521         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2522 }
2523
2524 static ssize_t
2525 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2526 {
2527         char *e;
2528         unsigned long long n = simple_strtoull(buf, &e, 10);
2529
2530         if (mddev->pers)
2531                 return -EBUSY;
2532         if (!*buf || (*e && *e != '\n'))
2533                 return -EINVAL;
2534
2535         mddev->recovery_cp = n;
2536         return len;
2537 }
2538 static struct md_sysfs_entry md_resync_start =
2539 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2540
2541 /*
2542  * The array state can be:
2543  *
2544  * clear
2545  *     No devices, no size, no level
2546  *     Equivalent to STOP_ARRAY ioctl
2547  * inactive
2548  *     May have some settings, but array is not active
2549  *        all IO results in error
2550  *     When written, doesn't tear down array, but just stops it
2551  * suspended (not supported yet)
2552  *     All IO requests will block. The array can be reconfigured.
2553  *     Writing this, if accepted, will block until array is quiessent
2554  * readonly
2555  *     no resync can happen.  no superblocks get written.
2556  *     write requests fail
2557  * read-auto
2558  *     like readonly, but behaves like 'clean' on a write request.
2559  *
2560  * clean - no pending writes, but otherwise active.
2561  *     When written to inactive array, starts without resync
2562  *     If a write request arrives then
2563  *       if metadata is known, mark 'dirty' and switch to 'active'.
2564  *       if not known, block and switch to write-pending
2565  *     If written to an active array that has pending writes, then fails.
2566  * active
2567  *     fully active: IO and resync can be happening.
2568  *     When written to inactive array, starts with resync
2569  *
2570  * write-pending
2571  *     clean, but writes are blocked waiting for 'active' to be written.
2572  *
2573  * active-idle
2574  *     like active, but no writes have been seen for a while (100msec).
2575  *
2576  */
2577 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2578                    write_pending, active_idle, bad_word};
2579 static char *array_states[] = {
2580         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2581         "write-pending", "active-idle", NULL };
2582
2583 static int match_word(const char *word, char **list)
2584 {
2585         int n;
2586         for (n=0; list[n]; n++)
2587                 if (cmd_match(word, list[n]))
2588                         break;
2589         return n;
2590 }
2591
2592 static ssize_t
2593 array_state_show(mddev_t *mddev, char *page)
2594 {
2595         enum array_state st = inactive;
2596
2597         if (mddev->pers)
2598                 switch(mddev->ro) {
2599                 case 1:
2600                         st = readonly;
2601                         break;
2602                 case 2:
2603                         st = read_auto;
2604                         break;
2605                 case 0:
2606                         if (mddev->in_sync)
2607                                 st = clean;
2608                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2609                                 st = write_pending;
2610                         else if (mddev->safemode)
2611                                 st = active_idle;
2612                         else
2613                                 st = active;
2614                 }
2615         else {
2616                 if (list_empty(&mddev->disks) &&
2617                     mddev->raid_disks == 0 &&
2618                     mddev->size == 0)
2619                         st = clear;
2620                 else
2621                         st = inactive;
2622         }
2623         return sprintf(page, "%s\n", array_states[st]);
2624 }
2625
2626 static int do_md_stop(mddev_t * mddev, int ro);
2627 static int do_md_run(mddev_t * mddev);
2628 static int restart_array(mddev_t *mddev);
2629
2630 static ssize_t
2631 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2632 {
2633         int err = -EINVAL;
2634         enum array_state st = match_word(buf, array_states);
2635         switch(st) {
2636         case bad_word:
2637                 break;
2638         case clear:
2639                 /* stopping an active array */
2640                 if (atomic_read(&mddev->active) > 1)
2641                         return -EBUSY;
2642                 err = do_md_stop(mddev, 0);
2643                 break;
2644         case inactive:
2645                 /* stopping an active array */
2646                 if (mddev->pers) {
2647                         if (atomic_read(&mddev->active) > 1)
2648                                 return -EBUSY;
2649                         err = do_md_stop(mddev, 2);
2650                 } else
2651                         err = 0; /* already inactive */
2652                 break;
2653         case suspended:
2654                 break; /* not supported yet */
2655         case readonly:
2656                 if (mddev->pers)
2657                         err = do_md_stop(mddev, 1);
2658                 else {
2659                         mddev->ro = 1;
2660                         set_disk_ro(mddev->gendisk, 1);
2661                         err = do_md_run(mddev);
2662                 }
2663                 break;
2664         case read_auto:
2665                 if (mddev->pers) {
2666                         if (mddev->ro != 1)
2667                                 err = do_md_stop(mddev, 1);
2668                         else
2669                                 err = restart_array(mddev);
2670                         if (err == 0) {
2671                                 mddev->ro = 2;
2672                                 set_disk_ro(mddev->gendisk, 0);
2673                         }
2674                 } else {
2675                         mddev->ro = 2;
2676                         err = do_md_run(mddev);
2677                 }
2678                 break;
2679         case clean:
2680                 if (mddev->pers) {
2681                         restart_array(mddev);
2682                         spin_lock_irq(&mddev->write_lock);
2683                         if (atomic_read(&mddev->writes_pending) == 0) {
2684                                 if (mddev->in_sync == 0) {
2685                                         mddev->in_sync = 1;
2686                                         if (mddev->safemode == 1)
2687                                                 mddev->safemode = 0;
2688                                         if (mddev->persistent)
2689                                                 set_bit(MD_CHANGE_CLEAN,
2690                                                         &mddev->flags);
2691                                 }
2692                                 err = 0;
2693                         } else
2694                                 err = -EBUSY;
2695                         spin_unlock_irq(&mddev->write_lock);
2696                 } else {
2697                         mddev->ro = 0;
2698                         mddev->recovery_cp = MaxSector;
2699                         err = do_md_run(mddev);
2700                 }
2701                 break;
2702         case active:
2703                 if (mddev->pers) {
2704                         restart_array(mddev);
2705                         if (mddev->external)
2706                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2707                         wake_up(&mddev->sb_wait);
2708                         err = 0;
2709                 } else {
2710                         mddev->ro = 0;
2711                         set_disk_ro(mddev->gendisk, 0);
2712                         err = do_md_run(mddev);
2713                 }
2714                 break;
2715         case write_pending:
2716         case active_idle:
2717                 /* these cannot be set */
2718                 break;
2719         }
2720         if (err)
2721                 return err;
2722         else {
2723                 sysfs_notify(&mddev->kobj, NULL, "array_state");
2724                 return len;
2725         }
2726 }
2727 static struct md_sysfs_entry md_array_state =
2728 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2729
2730 static ssize_t
2731 null_show(mddev_t *mddev, char *page)
2732 {
2733         return -EINVAL;
2734 }
2735
2736 static ssize_t
2737 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2738 {
2739         /* buf must be %d:%d\n? giving major and minor numbers */
2740         /* The new device is added to the array.
2741          * If the array has a persistent superblock, we read the
2742          * superblock to initialise info and check validity.
2743          * Otherwise, only checking done is that in bind_rdev_to_array,
2744          * which mainly checks size.
2745          */
2746         char *e;
2747         int major = simple_strtoul(buf, &e, 10);
2748         int minor;
2749         dev_t dev;
2750         mdk_rdev_t *rdev;
2751         int err;
2752
2753         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2754                 return -EINVAL;
2755         minor = simple_strtoul(e+1, &e, 10);
2756         if (*e && *e != '\n')
2757                 return -EINVAL;
2758         dev = MKDEV(major, minor);
2759         if (major != MAJOR(dev) ||
2760             minor != MINOR(dev))
2761                 return -EOVERFLOW;
2762
2763
2764         if (mddev->persistent) {
2765                 rdev = md_import_device(dev, mddev->major_version,
2766                                         mddev->minor_version);
2767                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2768                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2769                                                        mdk_rdev_t, same_set);
2770                         err = super_types[mddev->major_version]
2771                                 .load_super(rdev, rdev0, mddev->minor_version);
2772                         if (err < 0)
2773                                 goto out;
2774                 }
2775         } else if (mddev->external)
2776                 rdev = md_import_device(dev, -2, -1);
2777         else
2778                 rdev = md_import_device(dev, -1, -1);
2779
2780         if (IS_ERR(rdev))
2781                 return PTR_ERR(rdev);
2782         err = bind_rdev_to_array(rdev, mddev);
2783  out:
2784         if (err)
2785                 export_rdev(rdev);
2786         return err ? err : len;
2787 }
2788
2789 static struct md_sysfs_entry md_new_device =
2790 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2791
2792 static ssize_t
2793 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2794 {
2795         char *end;
2796         unsigned long chunk, end_chunk;
2797
2798         if (!mddev->bitmap)
2799                 goto out;
2800         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2801         while (*buf) {
2802                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2803                 if (buf == end) break;
2804                 if (*end == '-') { /* range */
2805                         buf = end + 1;
2806                         end_chunk = simple_strtoul(buf, &end, 0);
2807                         if (buf == end) break;
2808                 }
2809                 if (*end && !isspace(*end)) break;
2810                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2811                 buf = end;
2812                 while (isspace(*buf)) buf++;
2813         }
2814         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2815 out:
2816         return len;
2817 }
2818
2819 static struct md_sysfs_entry md_bitmap =
2820 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2821
2822 static ssize_t
2823 size_show(mddev_t *mddev, char *page)
2824 {
2825         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2826 }
2827
2828 static int update_size(mddev_t *mddev, unsigned long size);
2829
2830 static ssize_t
2831 size_store(mddev_t *mddev, const char *buf, size_t len)
2832 {
2833         /* If array is inactive, we can reduce the component size, but
2834          * not increase it (except from 0).
2835          * If array is active, we can try an on-line resize
2836          */
2837         char *e;
2838         int err = 0;
2839         unsigned long long size = simple_strtoull(buf, &e, 10);
2840         if (!*buf || *buf == '\n' ||
2841             (*e && *e != '\n'))
2842                 return -EINVAL;
2843
2844         if (mddev->pers) {
2845                 err = update_size(mddev, size);
2846                 md_update_sb(mddev, 1);
2847         } else {
2848                 if (mddev->size == 0 ||
2849                     mddev->size > size)
2850                         mddev->size = size;
2851                 else
2852                         err = -ENOSPC;
2853         }
2854         return err ? err : len;
2855 }
2856
2857 static struct md_sysfs_entry md_size =
2858 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2859
2860
2861 /* Metdata version.
2862  * This is one of
2863  *   'none' for arrays with no metadata (good luck...)
2864  *   'external' for arrays with externally managed metadata,
2865  * or N.M for internally known formats
2866  */
2867 static ssize_t
2868 metadata_show(mddev_t *mddev, char *page)
2869 {
2870         if (mddev->persistent)
2871                 return sprintf(page, "%d.%d\n",
2872                                mddev->major_version, mddev->minor_version);
2873         else if (mddev->external)
2874                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2875         else
2876                 return sprintf(page, "none\n");
2877 }
2878
2879 static ssize_t
2880 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2881 {
2882         int major, minor;
2883         char *e;
2884         if (!list_empty(&mddev->disks))
2885                 return -EBUSY;
2886
2887         if (cmd_match(buf, "none")) {
2888                 mddev->persistent = 0;
2889                 mddev->external = 0;
2890                 mddev->major_version = 0;
2891                 mddev->minor_version = 90;
2892                 return len;
2893         }
2894         if (strncmp(buf, "external:", 9) == 0) {
2895                 size_t namelen = len-9;
2896                 if (namelen >= sizeof(mddev->metadata_type))
2897                         namelen = sizeof(mddev->metadata_type)-1;
2898                 strncpy(mddev->metadata_type, buf+9, namelen);
2899                 mddev->metadata_type[namelen] = 0;
2900                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2901                         mddev->metadata_type[--namelen] = 0;
2902                 mddev->persistent = 0;
2903                 mddev->external = 1;
2904                 mddev->major_version = 0;
2905                 mddev->minor_version = 90;
2906                 return len;
2907         }
2908         major = simple_strtoul(buf, &e, 10);
2909         if (e==buf || *e != '.')
2910                 return -EINVAL;
2911         buf = e+1;
2912         minor = simple_strtoul(buf, &e, 10);
2913         if (e==buf || (*e && *e != '\n') )
2914                 return -EINVAL;
2915         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2916                 return -ENOENT;
2917         mddev->major_version = major;
2918         mddev->minor_version = minor;
2919         mddev->persistent = 1;
2920         mddev->external = 0;
2921         return len;
2922 }
2923
2924 static struct md_sysfs_entry md_metadata =
2925 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2926
2927 static ssize_t
2928 action_show(mddev_t *mddev, char *page)
2929 {
2930         char *type = "idle";
2931         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2932             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2933                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2934                         type = "reshape";
2935                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2936                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2937                                 type = "resync";
2938                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2939                                 type = "check";
2940                         else
2941                                 type = "repair";
2942                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
2943                         type = "recover";
2944         }
2945         return sprintf(page, "%s\n", type);
2946 }
2947
2948 static ssize_t
2949 action_store(mddev_t *mddev, const char *page, size_t len)
2950 {
2951         if (!mddev->pers || !mddev->pers->sync_request)
2952                 return -EINVAL;
2953
2954         if (cmd_match(page, "idle")) {
2955                 if (mddev->sync_thread) {
2956                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2957                         md_unregister_thread(mddev->sync_thread);
2958                         mddev->sync_thread = NULL;
2959                         mddev->recovery = 0;
2960                 }
2961         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2962                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2963                 return -EBUSY;
2964         else if (cmd_match(page, "resync"))
2965                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2966         else if (cmd_match(page, "recover")) {
2967                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2968                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2969         } else if (cmd_match(page, "reshape")) {
2970                 int err;
2971                 if (mddev->pers->start_reshape == NULL)
2972                         return -EINVAL;
2973                 err = mddev->pers->start_reshape(mddev);
2974                 if (err)
2975                         return err;
2976                 sysfs_notify(&mddev->kobj, NULL, "degraded");
2977         } else {
2978                 if (cmd_match(page, "check"))
2979                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2980                 else if (!cmd_match(page, "repair"))
2981                         return -EINVAL;
2982                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2983                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2984         }
2985         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2986         md_wakeup_thread(mddev->thread);
2987         sysfs_notify(&mddev->kobj, NULL, "sync_action");
2988         return len;
2989 }
2990
2991 static ssize_t
2992 mismatch_cnt_show(mddev_t *mddev, char *page)
2993 {
2994         return sprintf(page, "%llu\n",
2995                        (unsigned long long) mddev->resync_mismatches);
2996 }
2997
2998 static struct md_sysfs_entry md_scan_mode =
2999 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3000
3001
3002 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3003
3004 static ssize_t
3005 sync_min_show(mddev_t *mddev, char *page)
3006 {
3007         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3008                        mddev->sync_speed_min ? "local": "system");
3009 }
3010
3011 static ssize_t
3012 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3013 {
3014         int min;
3015         char *e;
3016         if (strncmp(buf, "system", 6)==0) {
3017                 mddev->sync_speed_min = 0;
3018                 return len;
3019         }
3020         min = simple_strtoul(buf, &e, 10);
3021         if (buf == e || (*e && *e != '\n') || min <= 0)
3022                 return -EINVAL;
3023         mddev->sync_speed_min = min;
3024         return len;
3025 }
3026
3027 static struct md_sysfs_entry md_sync_min =
3028 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3029
3030 static ssize_t
3031 sync_max_show(mddev_t *mddev, char *page)
3032 {
3033         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3034                        mddev->sync_speed_max ? "local": "system");
3035 }
3036
3037 static ssize_t
3038 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3039 {
3040         int max;
3041         char *e;
3042         if (strncmp(buf, "system", 6)==0) {
3043                 mddev->sync_speed_max = 0;
3044                 return len;
3045         }
3046         max = simple_strtoul(buf, &e, 10);
3047         if (buf == e || (*e && *e != '\n') || max <= 0)
3048                 return -EINVAL;
3049         mddev->sync_speed_max = max;
3050         return len;
3051 }
3052
3053 static struct md_sysfs_entry md_sync_max =
3054 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3055
3056 static ssize_t
3057 degraded_show(mddev_t *mddev, char *page)
3058 {
3059         return sprintf(page, "%d\n", mddev->degraded);
3060 }
3061 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3062
3063 static ssize_t
3064 sync_force_parallel_show(mddev_t *mddev, char *page)
3065 {
3066         return sprintf(page, "%d\n", mddev->parallel_resync);
3067 }
3068
3069 static ssize_t
3070 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3071 {
3072         long n;
3073
3074         if (strict_strtol(buf, 10, &n))
3075                 return -EINVAL;
3076
3077         if (n != 0 && n != 1)
3078                 return -EINVAL;
3079
3080         mddev->parallel_resync = n;
3081
3082         if (mddev->sync_thread)
3083                 wake_up(&resync_wait);
3084
3085         return len;
3086 }
3087
3088 /* force parallel resync, even with shared block devices */
3089 static struct md_sysfs_entry md_sync_force_parallel =
3090 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3091        sync_force_parallel_show, sync_force_parallel_store);
3092
3093 static ssize_t
3094 sync_speed_show(mddev_t *mddev, char *page)
3095 {
3096         unsigned long resync, dt, db;
3097         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
3098         dt = ((jiffies - mddev->resync_mark) / HZ);
3099         if (!dt) dt++;
3100         db = resync - (mddev->resync_mark_cnt);
3101         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
3102 }
3103
3104 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3105
3106 static ssize_t
3107 sync_completed_show(mddev_t *mddev, char *page)
3108 {
3109         unsigned long max_blocks, resync;
3110
3111         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3112                 max_blocks = mddev->resync_max_sectors;
3113         else
3114                 max_blocks = mddev->size << 1;
3115
3116         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3117         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3118 }
3119
3120 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3121
3122 static ssize_t
3123 min_sync_show(mddev_t *mddev, char *page)
3124 {
3125         return sprintf(page, "%llu\n",
3126                        (unsigned long long)mddev->resync_min);
3127 }
3128 static ssize_t
3129 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3130 {
3131         unsigned long long min;
3132         if (strict_strtoull(buf, 10, &min))
3133                 return -EINVAL;
3134         if (min > mddev->resync_max)
3135                 return -EINVAL;
3136         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3137                 return -EBUSY;
3138
3139         /* Must be a multiple of chunk_size */
3140         if (mddev->chunk_size) {
3141                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3142                         return -EINVAL;
3143         }
3144         mddev->resync_min = min;
3145
3146         return len;
3147 }
3148
3149 static struct md_sysfs_entry md_min_sync =
3150 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3151
3152 static ssize_t
3153 max_sync_show(mddev_t *mddev, char *page)
3154 {
3155         if (mddev->resync_max == MaxSector)
3156                 return sprintf(page, "max\n");
3157         else
3158                 return sprintf(page, "%llu\n",
3159                                (unsigned long long)mddev->resync_max);
3160 }
3161 static ssize_t
3162 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3163 {
3164         if (strncmp(buf, "max", 3) == 0)
3165                 mddev->resync_max = MaxSector;
3166         else {
3167                 unsigned long long max;
3168                 if (strict_strtoull(buf, 10, &max))
3169                         return -EINVAL;
3170                 if (max < mddev->resync_min)
3171                         return -EINVAL;
3172                 if (max < mddev->resync_max &&
3173                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3174                         return -EBUSY;
3175
3176                 /* Must be a multiple of chunk_size */
3177                 if (mddev->chunk_size) {
3178                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3179                                 return -EINVAL;
3180                 }
3181                 mddev->resync_max = max;
3182         }
3183         wake_up(&mddev->recovery_wait);
3184         return len;
3185 }
3186
3187 static struct md_sysfs_entry md_max_sync =
3188 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3189
3190 static ssize_t
3191 suspend_lo_show(mddev_t *mddev, char *page)
3192 {
3193         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3194 }
3195
3196 static ssize_t
3197 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3198 {
3199         char *e;
3200         unsigned long long new = simple_strtoull(buf, &e, 10);
3201
3202         if (mddev->pers->quiesce == NULL)
3203                 return -EINVAL;
3204         if (buf == e || (*e && *e != '\n'))
3205                 return -EINVAL;
3206         if (new >= mddev->suspend_hi ||
3207             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3208                 mddev->suspend_lo = new;
3209                 mddev->pers->quiesce(mddev, 2);
3210                 return len;
3211         } else
3212                 return -EINVAL;
3213 }
3214 static struct md_sysfs_entry md_suspend_lo =
3215 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3216
3217
3218 static ssize_t
3219 suspend_hi_show(mddev_t *mddev, char *page)
3220 {
3221         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3222 }
3223
3224 static ssize_t
3225 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3226 {
3227         char *e;
3228         unsigned long long new = simple_strtoull(buf, &e, 10);
3229
3230         if (mddev->pers->quiesce == NULL)
3231                 return -EINVAL;
3232         if (buf == e || (*e && *e != '\n'))
3233                 return -EINVAL;
3234         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3235             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3236                 mddev->suspend_hi = new;
3237                 mddev->pers->quiesce(mddev, 1);
3238                 mddev->pers->quiesce(mddev, 0);
3239                 return len;
3240         } else
3241                 return -EINVAL;
3242 }
3243 static struct md_sysfs_entry md_suspend_hi =
3244 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3245
3246 static ssize_t
3247 reshape_position_show(mddev_t *mddev, char *page)
3248 {
3249         if (mddev->reshape_position != MaxSector)
3250                 return sprintf(page, "%llu\n",
3251                                (unsigned long long)mddev->reshape_position);
3252         strcpy(page, "none\n");
3253         return 5;
3254 }
3255
3256 static ssize_t
3257 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3258 {
3259         char *e;
3260         unsigned long long new = simple_strtoull(buf, &e, 10);
3261         if (mddev->pers)
3262                 return -EBUSY;
3263         if (buf == e || (*e && *e != '\n'))
3264                 return -EINVAL;
3265         mddev->reshape_position = new;
3266         mddev->delta_disks = 0;
3267         mddev->new_level = mddev->level;
3268         mddev->new_layout = mddev->layout;
3269         mddev->new_chunk = mddev->chunk_size;
3270         return len;
3271 }
3272
3273 static struct md_sysfs_entry md_reshape_position =
3274 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3275        reshape_position_store);
3276
3277
3278 static struct attribute *md_default_attrs[] = {
3279         &md_level.attr,
3280         &md_layout.attr,
3281         &md_raid_disks.attr,
3282         &md_chunk_size.attr,
3283         &md_size.attr,
3284         &md_resync_start.attr,
3285         &md_metadata.attr,
3286         &md_new_device.attr,
3287         &md_safe_delay.attr,
3288         &md_array_state.attr,
3289         &md_reshape_position.attr,
3290         NULL,
3291 };
3292
3293 static struct attribute *md_redundancy_attrs[] = {
3294         &md_scan_mode.attr,
3295         &md_mismatches.attr,
3296         &md_sync_min.attr,
3297         &md_sync_max.attr,
3298         &md_sync_speed.attr,
3299         &md_sync_force_parallel.attr,
3300         &md_sync_completed.attr,
3301         &md_min_sync.attr,
3302         &md_max_sync.attr,
3303         &md_suspend_lo.attr,
3304         &md_suspend_hi.attr,
3305         &md_bitmap.attr,
3306         &md_degraded.attr,
3307         NULL,
3308 };
3309 static struct attribute_group md_redundancy_group = {
3310         .name = NULL,
3311         .attrs = md_redundancy_attrs,
3312 };
3313
3314
3315 static ssize_t
3316 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3317 {
3318         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3319         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3320         ssize_t rv;
3321
3322         if (!entry->show)
3323                 return -EIO;
3324         rv = mddev_lock(mddev);
3325         if (!rv) {
3326                 rv = entry->show(mddev, page);
3327                 mddev_unlock(mddev);
3328         }
3329         return rv;
3330 }
3331
3332 static ssize_t
3333 md_attr_store(struct kobject *kobj, struct attribute *attr,
3334               const char *page, size_t length)
3335 {
3336         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3337         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3338         ssize_t rv;
3339
3340         if (!entry->store)
3341                 return -EIO;
3342         if (!capable(CAP_SYS_ADMIN))
3343                 return -EACCES;
3344         rv = mddev_lock(mddev);
3345         if (!rv) {
3346                 rv = entry->store(mddev, page, length);
3347                 mddev_unlock(mddev);
3348         }
3349         return rv;
3350 }
3351
3352 static void md_free(struct kobject *ko)
3353 {
3354         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3355         kfree(mddev);
3356 }
3357
3358 static struct sysfs_ops md_sysfs_ops = {
3359         .show   = md_attr_show,
3360         .store  = md_attr_store,
3361 };
3362 static struct kobj_type md_ktype = {
3363         .release        = md_free,
3364         .sysfs_ops      = &md_sysfs_ops,
3365         .default_attrs  = md_default_attrs,
3366 };
3367
3368 int mdp_major = 0;
3369
3370 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3371 {
3372         static DEFINE_MUTEX(disks_mutex);
3373         mddev_t *mddev = mddev_find(dev);
3374         struct gendisk *disk;
3375         int partitioned = (MAJOR(dev) != MD_MAJOR);
3376         int shift = partitioned ? MdpMinorShift : 0;
3377         int unit = MINOR(dev) >> shift;
3378         int error;
3379
3380         if (!mddev)
3381                 return NULL;
3382
3383         mutex_lock(&disks_mutex);
3384         if (mddev->gendisk) {
3385                 mutex_unlock(&disks_mutex);
3386                 mddev_put(mddev);
3387                 return NULL;
3388         }
3389         disk = alloc_disk(1 << shift);
3390         if (!disk) {
3391                 mutex_unlock(&disks_mutex);
3392                 mddev_put(mddev);
3393                 return NULL;
3394         }
3395         disk->major = MAJOR(dev);
3396         disk->first_minor = unit << shift;
3397         if (partitioned)
3398                 sprintf(disk->disk_name, "md_d%d", unit);
3399         else
3400                 sprintf(disk->disk_name, "md%d", unit);
3401         disk->fops = &md_fops;
3402         disk->private_data = mddev;
3403         disk->queue = mddev->queue;
3404         add_disk(disk);
3405         mddev->gendisk = disk;
3406         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3407                                      "%s", "md");
3408         mutex_unlock(&disks_mutex);
3409         if (error)
3410                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3411                        disk->disk_name);
3412         else
3413                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3414         return NULL;
3415 }
3416
3417 static void md_safemode_timeout(unsigned long data)
3418 {
3419         mddev_t *mddev = (mddev_t *) data;
3420
3421         if (!atomic_read(&mddev->writes_pending)) {
3422                 mddev->safemode = 1;
3423                 if (mddev->external)
3424                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3425         }
3426         md_wakeup_thread(mddev->thread);
3427 }
3428
3429 static int start_dirty_degraded;
3430
3431 static int do_md_run(mddev_t * mddev)
3432 {
3433         int err;
3434         int chunk_size;
3435         struct list_head *tmp;
3436         mdk_rdev_t *rdev;
3437         struct gendisk *disk;
3438         struct mdk_personality *pers;
3439         char b[BDEVNAME_SIZE];
3440
3441         if (list_empty(&mddev->disks))
3442                 /* cannot run an array with no devices.. */
3443                 return -EINVAL;
3444
3445         if (mddev->pers)
3446                 return -EBUSY;
3447
3448         /*
3449          * Analyze all RAID superblock(s)
3450          */
3451         if (!mddev->raid_disks) {
3452                 if (!mddev->persistent)
3453                         return -EINVAL;
3454                 analyze_sbs(mddev);
3455         }
3456
3457         chunk_size = mddev->chunk_size;
3458
3459         if (chunk_size) {
3460                 if (chunk_size > MAX_CHUNK_SIZE) {
3461                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3462                                 chunk_size, MAX_CHUNK_SIZE);
3463                         return -EINVAL;
3464                 }
3465                 /*
3466                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3467                  */
3468                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3469                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3470                         return -EINVAL;
3471                 }
3472                 if (chunk_size < PAGE_SIZE) {
3473                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3474                                 chunk_size, PAGE_SIZE);
3475                         return -EINVAL;
3476                 }
3477
3478                 /* devices must have minimum size of one chunk */
3479                 rdev_for_each(rdev, tmp, mddev) {
3480                         if (test_bit(Faulty, &rdev->flags))
3481                                 continue;
3482                         if (rdev->size < chunk_size / 1024) {
3483                                 printk(KERN_WARNING
3484                                         "md: Dev %s smaller than chunk_size:"
3485                                         " %lluk < %dk\n",
3486                                         bdevname(rdev->bdev,b),
3487                                         (unsigned long long)rdev->size,
3488                                         chunk_size / 1024);
3489                                 return -EINVAL;
3490                         }
3491                 }
3492         }
3493
3494 #ifdef CONFIG_KMOD
3495         if (mddev->level != LEVEL_NONE)
3496                 request_module("md-level-%d", mddev->level);
3497         else if (mddev->clevel[0])
3498                 request_module("md-%s", mddev->clevel);
3499 #endif
3500
3501         /*
3502          * Drop all container device buffers, from now on
3503          * the only valid external interface is through the md
3504          * device.
3505          */
3506         rdev_for_each(rdev, tmp, mddev) {
3507                 if (test_bit(Faulty, &rdev->flags))
3508                         continue;
3509                 sync_blockdev(rdev->bdev);
3510                 invalidate_bdev(rdev->bdev);
3511
3512                 /* perform some consistency tests on the device.
3513                  * We don't want the data to overlap the metadata,
3514                  * Internal Bitmap issues has handled elsewhere.
3515                  */
3516                 if (rdev->data_offset < rdev->sb_offset) {
3517                         if (mddev->size &&
3518                             rdev->data_offset + mddev->size*2
3519                             > rdev->sb_offset*2) {
3520                                 printk("md: %s: data overlaps metadata\n",
3521                                        mdname(mddev));
3522                                 return -EINVAL;
3523                         }
3524                 } else {
3525                         if (rdev->sb_offset*2 + rdev->sb_size/512
3526                             > rdev->data_offset) {
3527                                 printk("md: %s: metadata overlaps data\n",
3528                                        mdname(mddev));
3529                                 return -EINVAL;
3530                         }
3531                 }
3532                 sysfs_notify(&rdev->kobj, NULL, "state");
3533         }
3534
3535         md_probe(mddev->unit, NULL, NULL);
3536         disk = mddev->gendisk;
3537         if (!disk)
3538                 return -ENOMEM;
3539
3540         spin_lock(&pers_lock);
3541         pers = find_pers(mddev->level, mddev->clevel);
3542         if (!pers || !try_module_get(pers->owner)) {
3543                 spin_unlock(&pers_lock);
3544                 if (mddev->level != LEVEL_NONE)
3545                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3546                                mddev->level);
3547                 else
3548                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3549                                mddev->clevel);
3550                 return -EINVAL;
3551         }
3552         mddev->pers = pers;
3553         spin_unlock(&pers_lock);
3554         mddev->level = pers->level;
3555         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3556
3557         if (mddev->reshape_position != MaxSector &&
3558             pers->start_reshape == NULL) {
3559                 /* This personality cannot handle reshaping... */
3560                 mddev->pers = NULL;
3561                 module_put(pers->owner);
3562                 return -EINVAL;
3563         }
3564
3565         if (pers->sync_request) {
3566                 /* Warn if this is a potentially silly
3567                  * configuration.
3568                  */
3569                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3570                 mdk_rdev_t *rdev2;
3571                 struct list_head *tmp2;
3572                 int warned = 0;
3573                 rdev_for_each(rdev, tmp, mddev) {
3574                         rdev_for_each(rdev2, tmp2, mddev) {
3575                                 if (rdev < rdev2 &&
3576                                     rdev->bdev->bd_contains ==
3577                                     rdev2->bdev->bd_contains) {
3578                                         printk(KERN_WARNING
3579                                                "%s: WARNING: %s appears to be"
3580                                                " on the same physical disk as"
3581                                                " %s.\n",
3582                                                mdname(mddev),
3583                                                bdevname(rdev->bdev,b),
3584                                                bdevname(rdev2->bdev,b2));
3585                                         warned = 1;
3586                                 }
3587                         }
3588                 }
3589                 if (warned)
3590                         printk(KERN_WARNING
3591                                "True protection against single-disk"
3592                                " failure might be compromised.\n");
3593         }
3594
3595         mddev->recovery = 0;
3596         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3597         mddev->barriers_work = 1;
3598         mddev->ok_start_degraded = start_dirty_degraded;
3599
3600         if (start_readonly)
3601                 mddev->ro = 2; /* read-only, but switch on first write */
3602
3603         err = mddev->pers->run(mddev);
3604         if (!err && mddev->pers->sync_request) {
3605                 err = bitmap_create(mddev);
3606                 if (err) {
3607                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3608                                mdname(mddev), err);
3609                         mddev->pers->stop(mddev);
3610                 }
3611         }
3612         if (err) {
3613                 printk(KERN_ERR "md: pers->run() failed ...\n");
3614                 module_put(mddev->pers->owner);
3615                 mddev->pers = NULL;
3616                 bitmap_destroy(mddev);
3617                 return err;
3618         }
3619         if (mddev->pers->sync_request) {
3620                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3621                         printk(KERN_WARNING
3622                                "md: cannot register extra attributes for %s\n",
3623                                mdname(mddev));
3624         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3625                 mddev->ro = 0;
3626
3627         atomic_set(&mddev->writes_pending,0);
3628         mddev->safemode = 0;
3629         mddev->safemode_timer.function = md_safemode_timeout;
3630         mddev->safemode_timer.data = (unsigned long) mddev;
3631         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3632         mddev->in_sync = 1;
3633
3634         rdev_for_each(rdev, tmp, mddev)
3635                 if (rdev->raid_disk >= 0) {
3636                         char nm[20];
3637                         sprintf(nm, "rd%d", rdev->raid_disk);
3638                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3639                                 printk("md: cannot register %s for %s\n",
3640                                        nm, mdname(mddev));
3641                 }
3642
3643         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3644
3645         if (mddev->flags)
3646                 md_update_sb(mddev, 0);
3647
3648         set_capacity(disk, mddev->array_size<<1);
3649
3650         /* If we call blk_queue_make_request here, it will
3651          * re-initialise max_sectors etc which may have been
3652          * refined inside -> run.  So just set the bits we need to set.
3653          * Most initialisation happended when we called
3654          * blk_queue_make_request(..., md_fail_request)
3655          * earlier.
3656          */
3657         mddev->queue->queuedata = mddev;
3658         mddev->queue->make_request_fn = mddev->pers->make_request;
3659
3660         /* If there is a partially-recovered drive we need to
3661          * start recovery here.  If we leave it to md_check_recovery,
3662          * it will remove the drives and not do the right thing
3663          */
3664         if (mddev->degraded && !mddev->sync_thread) {
3665                 struct list_head *rtmp;
3666                 int spares = 0;
3667                 rdev_for_each(rdev, rtmp, mddev)
3668                         if (rdev->raid_disk >= 0 &&
3669                             !test_bit(In_sync, &rdev->flags) &&
3670                             !test_bit(Faulty, &rdev->flags))
3671                                 /* complete an interrupted recovery */
3672                                 spares++;
3673                 if (spares && mddev->pers->sync_request) {
3674                         mddev->recovery = 0;
3675                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3676                         mddev->sync_thread = md_register_thread(md_do_sync,
3677                                                                 mddev,
3678                                                                 "%s_resync");
3679                         if (!mddev->sync_thread) {
3680                                 printk(KERN_ERR "%s: could not start resync"
3681                                        " thread...\n",
3682                                        mdname(mddev));
3683                                 /* leave the spares where they are, it shouldn't hurt */
3684                                 mddev->recovery = 0;
3685                         }
3686                 }
3687         }
3688         md_wakeup_thread(mddev->thread);
3689         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3690
3691         mddev->changed = 1;
3692         md_new_event(mddev);
3693         sysfs_notify(&mddev->kobj, NULL, "array_state");
3694         sysfs_notify(&mddev->kobj, NULL, "sync_action");
3695         sysfs_notify(&mddev->kobj, NULL, "degraded");
3696         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3697         return 0;
3698 }
3699
3700 static int restart_array(mddev_t *mddev)
3701 {
3702         struct gendisk *disk = mddev->gendisk;
3703         int err;
3704
3705         /*
3706          * Complain if it has no devices
3707          */
3708         err = -ENXIO;
3709         if (list_empty(&mddev->disks))
3710                 goto out;
3711
3712         if (mddev->pers) {
3713                 err = -EBUSY;
3714                 if (!mddev->ro)
3715                         goto out;
3716
3717                 mddev->safemode = 0;
3718                 mddev->ro = 0;
3719                 set_disk_ro(disk, 0);
3720
3721                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3722                         mdname(mddev));
3723                 /*
3724                  * Kick recovery or resync if necessary
3725                  */
3726                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3727                 md_wakeup_thread(mddev->thread);
3728                 md_wakeup_thread(mddev->sync_thread);
3729                 err = 0;
3730                 sysfs_notify(&mddev->kobj, NULL, "array_state");
3731
3732         } else
3733                 err = -EINVAL;
3734
3735 out:
3736         return err;
3737 }
3738
3739 /* similar to deny_write_access, but accounts for our holding a reference
3740  * to the file ourselves */
3741 static int deny_bitmap_write_access(struct file * file)
3742 {
3743         struct inode *inode = file->f_mapping->host;
3744
3745         spin_lock(&inode->i_lock);
3746         if (atomic_read(&inode->i_writecount) > 1) {
3747                 spin_unlock(&inode->i_lock);
3748                 return -ETXTBSY;
3749         }
3750         atomic_set(&inode->i_writecount, -1);
3751         spin_unlock(&inode->i_lock);
3752
3753         return 0;
3754 }
3755
3756 static void restore_bitmap_write_access(struct file *file)
3757 {
3758         struct inode *inode = file->f_mapping->host;
3759
3760         spin_lock(&inode->i_lock);
3761         atomic_set(&inode->i_writecount, 1);
3762         spin_unlock(&inode->i_lock);
3763 }
3764
3765 /* mode:
3766  *   0 - completely stop and dis-assemble array
3767  *   1 - switch to readonly
3768  *   2 - stop but do not disassemble array
3769  */
3770 static int do_md_stop(mddev_t * mddev, int mode)
3771 {
3772         int err = 0;
3773         struct gendisk *disk = mddev->gendisk;
3774
3775         if (mddev->pers) {
3776                 if (atomic_read(&mddev->active)>2) {
3777                         printk("md: %s still in use.\n",mdname(mddev));
3778                         return -EBUSY;
3779                 }
3780
3781                 if (mddev->sync_thread) {
3782                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3783                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3784                         md_unregister_thread(mddev->sync_thread);
3785                         mddev->sync_thread = NULL;
3786                 }
3787
3788                 del_timer_sync(&mddev->safemode_timer);
3789
3790                 invalidate_partition(disk, 0);
3791
3792                 switch(mode) {
3793                 case 1: /* readonly */
3794                         err  = -ENXIO;
3795                         if (mddev->ro==1)
3796                                 goto out;
3797                         mddev->ro = 1;
3798                         break;
3799                 case 0: /* disassemble */
3800                 case 2: /* stop */
3801                         bitmap_flush(mddev);
3802                         md_super_wait(mddev);
3803                         if (mddev->ro)
3804                                 set_disk_ro(disk, 0);
3805                         blk_queue_make_request(mddev->queue, md_fail_request);
3806                         mddev->pers->stop(mddev);
3807                         mddev->queue->merge_bvec_fn = NULL;
3808                         mddev->queue->unplug_fn = NULL;
3809                         mddev->queue->backing_dev_info.congested_fn = NULL;
3810                         if (mddev->pers->sync_request)
3811                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3812
3813                         module_put(mddev->pers->owner);
3814                         mddev->pers = NULL;
3815                         /* tell userspace to handle 'inactive' */
3816                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3817
3818                         set_capacity(disk, 0);
3819                         mddev->changed = 1;
3820
3821                         if (mddev->ro)
3822                                 mddev->ro = 0;
3823                 }
3824                 if (!mddev->in_sync || mddev->flags) {
3825                         /* mark array as shutdown cleanly */
3826                         mddev->in_sync = 1;
3827                         md_update_sb(mddev, 1);
3828                 }
3829                 if (mode == 1)
3830                         set_disk_ro(disk, 1);
3831                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3832         }
3833
3834         /*
3835          * Free resources if final stop
3836          */
3837         if (mode == 0) {
3838                 mdk_rdev_t *rdev;
3839                 struct list_head *tmp;
3840
3841                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3842
3843                 bitmap_destroy(mddev);
3844                 if (mddev->bitmap_file) {
3845                         restore_bitmap_write_access(mddev->bitmap_file);
3846                         fput(mddev->bitmap_file);
3847                         mddev->bitmap_file = NULL;
3848                 }
3849                 mddev->bitmap_offset = 0;
3850
3851                 rdev_for_each(rdev, tmp, mddev)
3852                         if (rdev->raid_disk >= 0) {
3853                                 char nm[20];
3854                                 sprintf(nm, "rd%d", rdev->raid_disk);
3855                                 sysfs_remove_link(&mddev->kobj, nm);
3856                         }
3857
3858                 /* make sure all md_delayed_delete calls have finished */
3859                 flush_scheduled_work();
3860
3861                 export_array(mddev);
3862
3863                 mddev->array_size = 0;
3864                 mddev->size = 0;
3865                 mddev->raid_disks = 0;
3866                 mddev->recovery_cp = 0;
3867                 mddev->resync_min = 0;
3868                 mddev->resync_max = MaxSector;
3869                 mddev->reshape_position = MaxSector;
3870                 mddev->external = 0;
3871                 mddev->persistent = 0;
3872                 mddev->level = LEVEL_NONE;
3873                 mddev->clevel[0] = 0;
3874                 mddev->flags = 0;
3875                 mddev->ro = 0;
3876                 mddev->metadata_type[0] = 0;
3877                 mddev->chunk_size = 0;
3878                 mddev->ctime = mddev->utime = 0;
3879                 mddev->layout = 0;
3880                 mddev->max_disks = 0;
3881                 mddev->events = 0;
3882                 mddev->delta_disks = 0;
3883                 mddev->new_level = LEVEL_NONE;
3884                 mddev->new_layout = 0;
3885                 mddev->new_chunk = 0;
3886                 mddev->curr_resync = 0;
3887                 mddev->resync_mismatches = 0;
3888                 mddev->suspend_lo = mddev->suspend_hi = 0;
3889                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3890                 mddev->recovery = 0;
3891                 mddev->in_sync = 0;
3892                 mddev->changed = 0;
3893                 mddev->degraded = 0;
3894                 mddev->barriers_work = 0;
3895                 mddev->safemode = 0;
3896
3897         } else if (mddev->pers)
3898                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3899                         mdname(mddev));
3900         err = 0;
3901         md_new_event(mddev);
3902         sysfs_notify(&mddev->kobj, NULL, "array_state");
3903 out:
3904         return err;
3905 }
3906
3907 #ifndef MODULE
3908 static void autorun_array(mddev_t *mddev)
3909 {
3910         mdk_rdev_t *rdev;
3911         struct list_head *tmp;
3912         int err;
3913
3914         if (list_empty(&mddev->disks))
3915                 return;
3916
3917         printk(KERN_INFO "md: running: ");
3918
3919         rdev_for_each(rdev, tmp, mddev) {
3920                 char b[BDEVNAME_SIZE];
3921                 printk("<%s>", bdevname(rdev->bdev,b));
3922         }
3923         printk("\n");
3924
3925         err = do_md_run (mddev);
3926         if (err) {
3927                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3928                 do_md_stop (mddev, 0);
3929         }
3930 }
3931
3932 /*
3933  * lets try to run arrays based on all disks that have arrived
3934  * until now. (those are in pending_raid_disks)
3935  *
3936  * the method: pick the first pending disk, collect all disks with
3937  * the same UUID, remove all from the pending list and put them into
3938  * the 'same_array' list. Then order this list based on superblock
3939  * update time (freshest comes first), kick out 'old' disks and
3940  * compare superblocks. If everything's fine then run it.
3941  *
3942  * If "unit" is allocated, then bump its reference count
3943  */
3944 static void autorun_devices(int part)
3945 {
3946         struct list_head *tmp;
3947         mdk_rdev_t *rdev0, *rdev;
3948         mddev_t *mddev;
3949         char b[BDEVNAME_SIZE];
3950
3951         printk(KERN_INFO "md: autorun ...\n");
3952         while (!list_empty(&pending_raid_disks)) {
3953                 int unit;
3954                 dev_t dev;
3955                 LIST_HEAD(candidates);
3956                 rdev0 = list_entry(pending_raid_disks.next,
3957                                          mdk_rdev_t, same_set);
3958
3959                 printk(KERN_INFO "md: considering %s ...\n",
3960                         bdevname(rdev0->bdev,b));
3961                 INIT_LIST_HEAD(&candidates);
3962                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3963                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3964                                 printk(KERN_INFO "md:  adding %s ...\n",
3965                                         bdevname(rdev->bdev,b));
3966                                 list_move(&rdev->same_set, &candidates);
3967                         }
3968                 /*
3969                  * now we have a set of devices, with all of them having
3970                  * mostly sane superblocks. It's time to allocate the
3971                  * mddev.
3972                  */
3973                 if (part) {
3974                         dev = MKDEV(mdp_major,
3975                                     rdev0->preferred_minor << MdpMinorShift);
3976                         unit = MINOR(dev) >> MdpMinorShift;
3977                 } else {
3978                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3979                         unit = MINOR(dev);
3980                 }
3981                 if (rdev0->preferred_minor != unit) {
3982                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3983                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3984                         break;
3985                 }
3986
3987                 md_probe(dev, NULL, NULL);
3988                 mddev = mddev_find(dev);
3989                 if (!mddev || !mddev->gendisk) {
3990                         if (mddev)
3991                                 mddev_put(mddev);
3992                         printk(KERN_ERR
3993                                 "md: cannot allocate memory for md drive.\n");
3994                         break;
3995                 }
3996                 if (mddev_lock(mddev))
3997                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3998                                mdname(mddev));
3999                 else if (mddev->raid_disks || mddev->major_version
4000                          || !list_empty(&mddev->disks)) {
4001                         printk(KERN_WARNING
4002                                 "md: %s already running, cannot run %s\n",
4003                                 mdname(mddev), bdevname(rdev0->bdev,b));
4004                         mddev_unlock(mddev);
4005                 } else {
4006                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4007                         mddev->persistent = 1;
4008                         rdev_for_each_list(rdev, tmp, candidates) {
4009                                 list_del_init(&rdev->same_set);
4010                                 if (bind_rdev_to_array(rdev, mddev))
4011                                         export_rdev(rdev);
4012                         }
4013                         autorun_array(mddev);
4014                         mddev_unlock(mddev);
4015                 }
4016                 /* on success, candidates will be empty, on error
4017                  * it won't...
4018                  */
4019                 rdev_for_each_list(rdev, tmp, candidates)
4020                         export_rdev(rdev);
4021                 mddev_put(mddev);
4022         }
4023         printk(KERN_INFO "md: ... autorun DONE.\n");
4024 }
4025 #endif /* !MODULE */
4026
4027 static int get_version(void __user * arg)
4028 {
4029         mdu_version_t ver;
4030
4031         ver.major = MD_MAJOR_VERSION;
4032         ver.minor = MD_MINOR_VERSION;
4033         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4034
4035         if (copy_to_user(arg, &ver, sizeof(ver)))
4036                 return -EFAULT;
4037
4038         return 0;
4039 }
4040
4041 static int get_array_info(mddev_t * mddev, void __user * arg)
4042 {
4043         mdu_array_info_t info;
4044         int nr,working,active,failed,spare;
4045         mdk_rdev_t *rdev;
4046         struct list_head *tmp;
4047
4048         nr=working=active=failed=spare=0;
4049         rdev_for_each(rdev, tmp, mddev) {
4050                 nr++;
4051                 if (test_bit(Faulty, &rdev->flags))
4052                         failed++;
4053                 else {
4054                         working++;
4055                         if (test_bit(In_sync, &rdev->flags))
4056                                 active++;
4057                         else
4058                                 spare++;
4059                 }
4060         }
4061
4062         info.major_version = mddev->major_version;
4063         info.minor_version = mddev->minor_version;
4064         info.patch_version = MD_PATCHLEVEL_VERSION;
4065         info.ctime         = mddev->ctime;
4066         info.level         = mddev->level;
4067         info.size          = mddev->size;
4068         if (info.size != mddev->size) /* overflow */
4069                 info.size = -1;
4070         info.nr_disks      = nr;
4071         info.raid_disks    = mddev->raid_disks;
4072         info.md_minor      = mddev->md_minor;
4073         info.not_persistent= !mddev->persistent;
4074
4075         info.utime         = mddev->utime;
4076         info.state         = 0;
4077         if (mddev->in_sync)
4078                 info.state = (1<<MD_SB_CLEAN);
4079         if (mddev->bitmap && mddev->bitmap_offset)
4080                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4081         info.active_disks  = active;
4082         info.working_disks = working;
4083         info.failed_disks  = failed;
4084         info.spare_disks   = spare;
4085
4086         info.layout        = mddev->layout;
4087         info.chunk_size    = mddev->chunk_size;
4088
4089         if (copy_to_user(arg, &info, sizeof(info)))
4090                 return -EFAULT;
4091
4092         return 0;
4093 }
4094
4095 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4096 {
4097         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4098         char *ptr, *buf = NULL;
4099         int err = -ENOMEM;
4100
4101         md_allow_write(mddev);
4102
4103         file = kmalloc(sizeof(*file), GFP_KERNEL);
4104         if (!file)
4105                 goto out;
4106
4107         /* bitmap disabled, zero the first byte and copy out */
4108         if (!mddev->bitmap || !mddev->bitmap->file) {
4109                 file->pathname[0] = '\0';
4110                 goto copy_out;
4111         }
4112
4113         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4114         if (!buf)
4115                 goto out;
4116
4117         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4118         if (IS_ERR(ptr))
4119                 goto out;
4120
4121         strcpy(file->pathname, ptr);
4122
4123 copy_out:
4124         err = 0;
4125         if (copy_to_user(arg, file, sizeof(*file)))
4126                 err = -EFAULT;
4127 out:
4128         kfree(buf);
4129         kfree(file);
4130         return err;
4131 }
4132
4133 static int get_disk_info(mddev_t * mddev, void __user * arg)
4134 {
4135         mdu_disk_info_t info;
4136         unsigned int nr;
4137         mdk_rdev_t *rdev;
4138
4139         if (copy_from_user(&info, arg, sizeof(info)))
4140                 return -EFAULT;
4141
4142         nr = info.number;
4143
4144         rdev = find_rdev_nr(mddev, nr);
4145         if (rdev) {
4146                 info.major = MAJOR(rdev->bdev->bd_dev);
4147                 info.minor = MINOR(rdev->bdev->bd_dev);
4148                 info.raid_disk = rdev->raid_disk;
4149                 info.state = 0;
4150                 if (test_bit(Faulty, &rdev->flags))
4151                         info.state |= (1<<MD_DISK_FAULTY);
4152                 else if (test_bit(In_sync, &rdev->flags)) {
4153                         info.state |= (1<<MD_DISK_ACTIVE);
4154                         info.state |= (1<<MD_DISK_SYNC);
4155                 }
4156                 if (test_bit(WriteMostly, &rdev->flags))
4157                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4158         } else {
4159                 info.major = info.minor = 0;
4160                 info.raid_disk = -1;
4161                 info.state = (1<<MD_DISK_REMOVED);
4162         }
4163
4164         if (copy_to_user(arg, &info, sizeof(info)))
4165                 return -EFAULT;
4166
4167         return 0;
4168 }
4169
4170 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4171 {
4172         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4173         mdk_rdev_t *rdev;
4174         dev_t dev = MKDEV(info->major,info->minor);
4175
4176         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4177                 return -EOVERFLOW;
4178
4179         if (!mddev->raid_disks) {
4180                 int err;
4181                 /* expecting a device which has a superblock */
4182                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4183                 if (IS_ERR(rdev)) {
4184                         printk(KERN_WARNING
4185                                 "md: md_import_device returned %ld\n",
4186                                 PTR_ERR(rdev));
4187                         return PTR_ERR(rdev);
4188                 }
4189                 if (!list_empty(&mddev->disks)) {
4190                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4191                                                         mdk_rdev_t, same_set);
4192                         int err = super_types[mddev->major_version]
4193                                 .load_super(rdev, rdev0, mddev->minor_version);
4194                         if (err < 0) {
4195                                 printk(KERN_WARNING
4196                                         "md: %s has different UUID to %s\n",
4197                                         bdevname(rdev->bdev,b),
4198                                         bdevname(rdev0->bdev,b2));
4199                                 export_rdev(rdev);
4200                                 return -EINVAL;
4201                         }
4202                 }
4203                 err = bind_rdev_to_array(rdev, mddev);
4204                 if (err)
4205                         export_rdev(rdev);
4206                 return err;
4207         }
4208
4209         /*
4210          * add_new_disk can be used once the array is assembled
4211          * to add "hot spares".  They must already have a superblock
4212          * written
4213          */
4214         if (mddev->pers) {
4215                 int err;
4216                 if (!mddev->pers->hot_add_disk) {
4217                         printk(KERN_WARNING
4218                                 "%s: personality does not support diskops!\n",
4219                                mdname(mddev));
4220                         return -EINVAL;
4221                 }
4222                 if (mddev->persistent)
4223                         rdev = md_import_device(dev, mddev->major_version,
4224                                                 mddev->minor_version);
4225                 else
4226                         rdev = md_import_device(dev, -1, -1);
4227                 if (IS_ERR(rdev)) {
4228                         printk(KERN_WARNING
4229                                 "md: md_import_device returned %ld\n",
4230                                 PTR_ERR(rdev));
4231                         return PTR_ERR(rdev);
4232                 }
4233                 /* set save_raid_disk if appropriate */
4234                 if (!mddev->persistent) {
4235                         if (info->state & (1<<MD_DISK_SYNC)  &&
4236                             info->raid_disk < mddev->raid_disks)
4237                                 rdev->raid_disk = info->raid_disk;
4238                         else
4239                                 rdev->raid_disk = -1;
4240                 } else
4241                         super_types[mddev->major_version].
4242                                 validate_super(mddev, rdev);
4243                 rdev->saved_raid_disk = rdev->raid_disk;
4244
4245                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4246                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4247                         set_bit(WriteMostly, &rdev->flags);
4248
4249                 rdev->raid_disk = -1;
4250                 err = bind_rdev_to_array(rdev, mddev);
4251                 if (!err && !mddev->pers->hot_remove_disk) {
4252                         /* If there is hot_add_disk but no hot_remove_disk
4253                          * then added disks for geometry changes,
4254                          * and should be added immediately.
4255                          */
4256                         super_types[mddev->major_version].
4257                                 validate_super(mddev, rdev);
4258                         err = mddev->pers->hot_add_disk(mddev, rdev);
4259                         if (err)
4260                                 unbind_rdev_from_array(rdev);
4261                 }
4262                 if (err)
4263                         export_rdev(rdev);
4264                 else
4265                         sysfs_notify(&rdev->kobj, NULL, "state");
4266
4267                 md_update_sb(mddev, 1);
4268                 if (mddev->degraded)
4269                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4270                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4271                 md_wakeup_thread(mddev->thread);
4272                 return err;
4273         }
4274
4275         /* otherwise, add_new_disk is only allowed
4276          * for major_version==0 superblocks
4277          */
4278         if (mddev->major_version != 0) {
4279                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4280                        mdname(mddev));
4281                 return -EINVAL;
4282         }
4283
4284         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4285                 int err;
4286                 rdev = md_import_device (dev, -1, 0);
4287                 if (IS_ERR(rdev)) {
4288                         printk(KERN_WARNING
4289                                 "md: error, md_import_device() returned %ld\n",
4290                                 PTR_ERR(rdev));
4291                         return PTR_ERR(rdev);
4292                 }
4293                 rdev->desc_nr = info->number;
4294                 if (info->raid_disk < mddev->raid_disks)
4295                         rdev->raid_disk = info->raid_disk;
4296                 else
4297                         rdev->raid_disk = -1;
4298
4299                 if (rdev->raid_disk < mddev->raid_disks)
4300                         if (info->state & (1<<MD_DISK_SYNC))
4301                                 set_bit(In_sync, &rdev->flags);
4302
4303                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4304                         set_bit(WriteMostly, &rdev->flags);
4305
4306                 if (!mddev->persistent) {
4307                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4308                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4309                 } else
4310                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4311                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4312
4313                 err = bind_rdev_to_array(rdev, mddev);
4314                 if (err) {
4315                         export_rdev(rdev);
4316                         return err;
4317                 }
4318         }
4319
4320         return 0;
4321 }
4322
4323 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4324 {
4325         char b[BDEVNAME_SIZE];
4326         mdk_rdev_t *rdev;
4327
4328         rdev = find_rdev(mddev, dev);
4329         if (!rdev)
4330                 return -ENXIO;
4331
4332         if (rdev->raid_disk >= 0)
4333                 goto busy;
4334
4335         kick_rdev_from_array(rdev);
4336         md_update_sb(mddev, 1);
4337         md_new_event(mddev);
4338
4339         return 0;
4340 busy:
4341         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4342                 bdevname(rdev->bdev,b), mdname(mddev));
4343         return -EBUSY;
4344 }
4345
4346 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4347 {
4348         char b[BDEVNAME_SIZE];
4349         int err;
4350         unsigned int size;
4351         mdk_rdev_t *rdev;
4352
4353         if (!mddev->pers)
4354                 return -ENODEV;
4355
4356         if (mddev->major_version != 0) {
4357                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4358                         " version-0 superblocks.\n",
4359                         mdname(mddev));
4360                 return -EINVAL;
4361         }
4362         if (!mddev->pers->hot_add_disk) {
4363                 printk(KERN_WARNING
4364                         "%s: personality does not support diskops!\n",
4365                         mdname(mddev));
4366                 return -EINVAL;
4367         }
4368
4369         rdev = md_import_device (dev, -1, 0);
4370         if (IS_ERR(rdev)) {
4371                 printk(KERN_WARNING
4372                         "md: error, md_import_device() returned %ld\n",
4373                         PTR_ERR(rdev));
4374                 return -EINVAL;
4375         }
4376
4377         if (mddev->persistent)
4378                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4379         else
4380                 rdev->sb_offset =
4381                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4382
4383         size = calc_dev_size(rdev, mddev->chunk_size);
4384         rdev->size = size;
4385
4386         if (test_bit(Faulty, &rdev->flags)) {
4387                 printk(KERN_WARNING
4388                         "md: can not hot-add faulty %s disk to %s!\n",
4389                         bdevname(rdev->bdev,b), mdname(mddev));
4390                 err = -EINVAL;
4391                 goto abort_export;
4392         }
4393         clear_bit(In_sync, &rdev->flags);
4394         rdev->desc_nr = -1;
4395         rdev->saved_raid_disk = -1;
4396         err = bind_rdev_to_array(rdev, mddev);
4397         if (err)
4398                 goto abort_export;
4399
4400         /*
4401          * The rest should better be atomic, we can have disk failures
4402          * noticed in interrupt contexts ...
4403          */
4404
4405         if (rdev->desc_nr == mddev->max_disks) {
4406                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4407                         mdname(mddev));
4408                 err = -EBUSY;
4409                 goto abort_unbind_export;
4410         }
4411
4412         rdev->raid_disk = -1;
4413
4414         md_update_sb(mddev, 1);
4415
4416         /*
4417          * Kick recovery, maybe this spare has to be added to the
4418          * array immediately.
4419          */
4420         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4421         md_wakeup_thread(mddev->thread);
4422         md_new_event(mddev);
4423         return 0;
4424
4425 abort_unbind_export:
4426         unbind_rdev_from_array(rdev);
4427
4428 abort_export:
4429         export_rdev(rdev);
4430         return err;
4431 }
4432
4433 static int set_bitmap_file(mddev_t *mddev, int fd)
4434 {
4435         int err;
4436
4437         if (mddev->pers) {
4438                 if (!mddev->pers->quiesce)
4439                         return -EBUSY;
4440                 if (mddev->recovery || mddev->sync_thread)
4441                         return -EBUSY;
4442                 /* we should be able to change the bitmap.. */
4443         }
4444
4445
4446         if (fd >= 0) {
4447                 if (mddev->bitmap)
4448                         return -EEXIST; /* cannot add when bitmap is present */
4449                 mddev->bitmap_file = fget(fd);
4450
4451                 if (mddev->bitmap_file == NULL) {
4452                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4453                                mdname(mddev));
4454                         return -EBADF;
4455                 }
4456
4457                 err = deny_bitmap_write_access(mddev->bitmap_file);
4458                 if (err) {
4459                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4460                                mdname(mddev));
4461                         fput(mddev->bitmap_file);
4462                         mddev->bitmap_file = NULL;
4463                         return err;
4464                 }
4465                 mddev->bitmap_offset = 0; /* file overrides offset */
4466         } else if (mddev->bitmap == NULL)
4467                 return -ENOENT; /* cannot remove what isn't there */
4468         err = 0;
4469         if (mddev->pers) {
4470                 mddev->pers->quiesce(mddev, 1);
4471                 if (fd >= 0)
4472                         err = bitmap_create(mddev);
4473                 if (fd < 0 || err) {
4474                         bitmap_destroy(mddev);
4475                         fd = -1; /* make sure to put the file */
4476                 }
4477                 mddev->pers->quiesce(mddev, 0);
4478         }
4479         if (fd < 0) {
4480                 if (mddev->bitmap_file) {
4481                         restore_bitmap_write_access(mddev->bitmap_file);
4482                         fput(mddev->bitmap_file);
4483                 }
4484                 mddev->bitmap_file = NULL;
4485         }
4486
4487         return err;
4488 }
4489
4490 /*
4491  * set_array_info is used two different ways
4492  * The original usage is when creating a new array.
4493  * In this usage, raid_disks is > 0 and it together with
4494  *  level, size, not_persistent,layout,chunksize determine the
4495  *  shape of the array.
4496  *  This will always create an array with a type-0.90.0 superblock.
4497  * The newer usage is when assembling an array.
4498  *  In this case raid_disks will be 0, and the major_version field is
4499  *  use to determine which style super-blocks are to be found on the devices.
4500  *  The minor and patch _version numbers are also kept incase the
4501  *  super_block handler wishes to interpret them.
4502  */
4503 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4504 {
4505
4506         if (info->raid_disks == 0) {
4507                 /* just setting version number for superblock loading */
4508                 if (info->major_version < 0 ||
4509                     info->major_version >= ARRAY_SIZE(super_types) ||
4510                     super_types[info->major_version].name == NULL) {
4511                         /* maybe try to auto-load a module? */
4512                         printk(KERN_INFO
4513                                 "md: superblock version %d not known\n",
4514                                 info->major_version);
4515                         return -EINVAL;
4516                 }
4517                 mddev->major_version = info->major_version;
4518                 mddev->minor_version = info->minor_version;
4519                 mddev->patch_version = info->patch_version;
4520                 mddev->persistent = !info->not_persistent;
4521                 return 0;
4522         }
4523         mddev->major_version = MD_MAJOR_VERSION;
4524         mddev->minor_version = MD_MINOR_VERSION;
4525         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4526         mddev->ctime         = get_seconds();
4527
4528         mddev->level         = info->level;
4529         mddev->clevel[0]     = 0;
4530         mddev->size          = info->size;
4531         mddev->raid_disks    = info->raid_disks;
4532         /* don't set md_minor, it is determined by which /dev/md* was
4533          * openned
4534          */
4535         if (info->state & (1<<MD_SB_CLEAN))
4536                 mddev->recovery_cp = MaxSector;
4537         else
4538                 mddev->recovery_cp = 0;
4539         mddev->persistent    = ! info->not_persistent;
4540         mddev->external      = 0;
4541
4542         mddev->layout        = info->layout;
4543         mddev->chunk_size    = info->chunk_size;
4544
4545         mddev->max_disks     = MD_SB_DISKS;
4546
4547         if (mddev->persistent)
4548                 mddev->flags         = 0;
4549         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4550
4551         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4552         mddev->bitmap_offset = 0;
4553
4554         mddev->reshape_position = MaxSector;
4555
4556         /*
4557          * Generate a 128 bit UUID
4558          */
4559         get_random_bytes(mddev->uuid, 16);
4560
4561         mddev->new_level = mddev->level;
4562         mddev->new_chunk = mddev->chunk_size;
4563         mddev->new_layout = mddev->layout;
4564         mddev->delta_disks = 0;
4565
4566         return 0;
4567 }
4568
4569 static int update_size(mddev_t *mddev, unsigned long size)
4570 {
4571         mdk_rdev_t * rdev;
4572         int rv;
4573         struct list_head *tmp;
4574         int fit = (size == 0);
4575
4576         if (mddev->pers->resize == NULL)
4577                 return -EINVAL;
4578         /* The "size" is the amount of each device that is used.
4579          * This can only make sense for arrays with redundancy.
4580          * linear and raid0 always use whatever space is available
4581          * We can only consider changing the size if no resync
4582          * or reconstruction is happening, and if the new size
4583          * is acceptable. It must fit before the sb_offset or,
4584          * if that is <data_offset, it must fit before the
4585          * size of each device.
4586          * If size is zero, we find the largest size that fits.
4587          */
4588         if (mddev->sync_thread)
4589                 return -EBUSY;
4590         rdev_for_each(rdev, tmp, mddev) {
4591                 sector_t avail;
4592                 avail = rdev->size * 2;
4593
4594                 if (fit && (size == 0 || size > avail/2))
4595                         size = avail/2;
4596                 if (avail < ((sector_t)size << 1))
4597                         return -ENOSPC;
4598         }
4599         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4600         if (!rv) {
4601                 struct block_device *bdev;
4602
4603                 bdev = bdget_disk(mddev->gendisk, 0);
4604                 if (bdev) {
4605                         mutex_lock(&bdev->bd_inode->i_mutex);
4606                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4607                         mutex_unlock(&bdev->bd_inode->i_mutex);
4608                         bdput(bdev);
4609                 }
4610         }
4611         return rv;
4612 }
4613
4614 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4615 {
4616         int rv;
4617         /* change the number of raid disks */
4618         if (mddev->pers->check_reshape == NULL)
4619                 return -EINVAL;
4620         if (raid_disks <= 0 ||
4621             raid_disks >= mddev->max_disks)
4622                 return -EINVAL;
4623         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4624                 return -EBUSY;
4625         mddev->delta_disks = raid_disks - mddev->raid_disks;
4626
4627         rv = mddev->pers->check_reshape(mddev);
4628         return rv;
4629 }
4630
4631
4632 /*
4633  * update_array_info is used to change the configuration of an
4634  * on-line array.
4635  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4636  * fields in the info are checked against the array.
4637  * Any differences that cannot be handled will cause an error.
4638  * Normally, only one change can be managed at a time.
4639  */
4640 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4641 {
4642         int rv = 0;
4643         int cnt = 0;
4644         int state = 0;
4645
4646         /* calculate expected state,ignoring low bits */
4647         if (mddev->bitmap && mddev->bitmap_offset)
4648                 state |= (1 << MD_SB_BITMAP_PRESENT);
4649
4650         if (mddev->major_version != info->major_version ||
4651             mddev->minor_version != info->minor_version ||
4652 /*          mddev->patch_version != info->patch_version || */
4653             mddev->ctime         != info->ctime         ||
4654             mddev->level         != info->level         ||
4655 /*          mddev->layout        != info->layout        || */
4656             !mddev->persistent   != info->not_persistent||
4657             mddev->chunk_size    != info->chunk_size    ||
4658             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4659             ((state^info->state) & 0xfffffe00)
4660                 )
4661                 return -EINVAL;
4662         /* Check there is only one change */
4663         if (info->size >= 0 && mddev->size != info->size) cnt++;
4664         if (mddev->raid_disks != info->raid_disks) cnt++;
4665         if (mddev->layout != info->layout) cnt++;
4666         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4667         if (cnt == 0) return 0;
4668         if (cnt > 1) return -EINVAL;
4669
4670         if (mddev->layout != info->layout) {
4671                 /* Change layout
4672                  * we don't need to do anything at the md level, the
4673                  * personality will take care of it all.
4674                  */
4675                 if (mddev->pers->reconfig == NULL)
4676                         return -EINVAL;
4677                 else
4678                         return mddev->pers->reconfig(mddev, info->layout, -1);
4679         }
4680         if (info->size >= 0 && mddev->size != info->size)
4681                 rv = update_size(mddev, info->size);
4682
4683         if (mddev->raid_disks    != info->raid_disks)
4684                 rv = update_raid_disks(mddev, info->raid_disks);
4685
4686         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4687                 if (mddev->pers->quiesce == NULL)
4688                         return -EINVAL;
4689                 if (mddev->recovery || mddev->sync_thread)
4690                         return -EBUSY;
4691                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4692                         /* add the bitmap */
4693                         if (mddev->bitmap)
4694                                 return -EEXIST;
4695                         if (mddev->default_bitmap_offset == 0)
4696                                 return -EINVAL;
4697                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4698                         mddev->pers->quiesce(mddev, 1);
4699                         rv = bitmap_create(mddev);
4700                         if (rv)
4701                                 bitmap_destroy(mddev);
4702                         mddev->pers->quiesce(mddev, 0);
4703                 } else {
4704                         /* remove the bitmap */
4705                         if (!mddev->bitmap)
4706                                 return -ENOENT;
4707                         if (mddev->bitmap->file)
4708                                 return -EINVAL;
4709                         mddev->pers->quiesce(mddev, 1);
4710                         bitmap_destroy(mddev);
4711                         mddev->pers->quiesce(mddev, 0);
4712                         mddev->bitmap_offset = 0;
4713                 }
4714         }
4715         md_update_sb(mddev, 1);
4716         return rv;
4717 }
4718
4719 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4720 {
4721         mdk_rdev_t *rdev;
4722
4723         if (mddev->pers == NULL)
4724                 return -ENODEV;
4725
4726         rdev = find_rdev(mddev, dev);
4727         if (!rdev)
4728                 return -ENODEV;
4729
4730         md_error(mddev, rdev);
4731         return 0;
4732 }
4733
4734 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4735 {
4736         mddev_t *mddev = bdev->bd_disk->private_data;
4737
4738         geo->heads = 2;
4739         geo->sectors = 4;
4740         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4741         return 0;
4742 }
4743
4744 static int md_ioctl(struct inode *inode, struct file *file,
4745                         unsigned int cmd, unsigned long arg)
4746 {
4747         int err = 0;
4748         void __user *argp = (void __user *)arg;
4749         mddev_t *mddev = NULL;
4750
4751         if (!capable(CAP_SYS_ADMIN))
4752                 return -EACCES;
4753
4754         /*
4755          * Commands dealing with the RAID driver but not any
4756          * particular array:
4757          */
4758         switch (cmd)
4759         {
4760                 case RAID_VERSION:
4761                         err = get_version(argp);
4762                         goto done;
4763
4764                 case PRINT_RAID_DEBUG:
4765                         err = 0;
4766                         md_print_devices();
4767                         goto done;
4768
4769 #ifndef MODULE
4770                 case RAID_AUTORUN:
4771                         err = 0;
4772                         autostart_arrays(arg);
4773                         goto done;
4774 #endif
4775                 default:;
4776         }
4777
4778         /*
4779          * Commands creating/starting a new array:
4780          */
4781
4782         mddev = inode->i_bdev->bd_disk->private_data;
4783
4784         if (!mddev) {
4785                 BUG();
4786                 goto abort;
4787         }
4788
4789         err = mddev_lock(mddev);
4790         if (err) {
4791                 printk(KERN_INFO
4792                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4793                         err, cmd);
4794                 goto abort;
4795         }
4796
4797         switch (cmd)
4798         {
4799                 case SET_ARRAY_INFO:
4800                         {
4801                                 mdu_array_info_t info;
4802                                 if (!arg)
4803                                         memset(&info, 0, sizeof(info));
4804                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4805                                         err = -EFAULT;
4806                                         goto abort_unlock;
4807                                 }
4808                                 if (mddev->pers) {
4809                                         err = update_array_info(mddev, &info);
4810                                         if (err) {
4811                                                 printk(KERN_WARNING "md: couldn't update"
4812                                                        " array info. %d\n", err);
4813                                                 goto abort_unlock;
4814                                         }
4815                                         goto done_unlock;
4816                                 }
4817                                 if (!list_empty(&mddev->disks)) {
4818                                         printk(KERN_WARNING
4819                                                "md: array %s already has disks!\n",
4820                                                mdname(mddev));
4821                                         err = -EBUSY;
4822                                         goto abort_unlock;
4823                                 }
4824                                 if (mddev->raid_disks) {
4825                                         printk(KERN_WARNING
4826                                                "md: array %s already initialised!\n",
4827                                                mdname(mddev));
4828                                         err = -EBUSY;
4829                                         goto abort_unlock;
4830                                 }
4831                                 err = set_array_info(mddev, &info);
4832                                 if (err) {
4833                                         printk(KERN_WARNING "md: couldn't set"
4834                                                " array info. %d\n", err);
4835                                         goto abort_unlock;
4836                                 }
4837                         }
4838                         goto done_unlock;
4839
4840                 default:;
4841         }
4842
4843         /*
4844          * Commands querying/configuring an existing array:
4845          */
4846         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4847          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4848         if ((!mddev->raid_disks && !mddev->external)
4849             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4850             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4851             && cmd != GET_BITMAP_FILE) {
4852                 err = -ENODEV;
4853                 goto abort_unlock;
4854         }
4855
4856         /*
4857          * Commands even a read-only array can execute:
4858          */
4859         switch (cmd)
4860         {
4861                 case GET_ARRAY_INFO:
4862                         err = get_array_info(mddev, argp);
4863                         goto done_unlock;
4864
4865                 case GET_BITMAP_FILE:
4866                         err = get_bitmap_file(mddev, argp);
4867                         goto done_unlock;
4868
4869                 case GET_DISK_INFO:
4870                         err = get_disk_info(mddev, argp);
4871                         goto done_unlock;
4872
4873                 case RESTART_ARRAY_RW:
4874                         err = restart_array(mddev);
4875                         goto done_unlock;
4876
4877                 case STOP_ARRAY:
4878                         err = do_md_stop (mddev, 0);
4879                         goto done_unlock;
4880
4881                 case STOP_ARRAY_RO:
4882                         err = do_md_stop (mddev, 1);
4883                         goto done_unlock;
4884
4885         /*
4886          * We have a problem here : there is no easy way to give a CHS
4887          * virtual geometry. We currently pretend that we have a 2 heads
4888          * 4 sectors (with a BIG number of cylinders...). This drives
4889          * dosfs just mad... ;-)
4890          */
4891         }
4892
4893         /*
4894          * The remaining ioctls are changing the state of the
4895          * superblock, so we do not allow them on read-only arrays.
4896          * However non-MD ioctls (e.g. get-size) will still come through
4897          * here and hit the 'default' below, so only disallow
4898          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4899          */
4900         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4901             mddev->ro && mddev->pers) {
4902                 if (mddev->ro == 2) {
4903                         mddev->ro = 0;
4904                         sysfs_notify(&mddev->kobj, NULL, "array_state");
4905                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4906                         md_wakeup_thread(mddev->thread);
4907
4908                 } else {
4909                         err = -EROFS;
4910                         goto abort_unlock;
4911                 }
4912         }
4913
4914         switch (cmd)
4915         {
4916                 case ADD_NEW_DISK:
4917                 {
4918                         mdu_disk_info_t info;
4919                         if (copy_from_user(&info, argp, sizeof(info)))
4920                                 err = -EFAULT;
4921                         else
4922                                 err = add_new_disk(mddev, &info);
4923                         goto done_unlock;
4924                 }
4925
4926                 case HOT_REMOVE_DISK:
4927                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4928                         goto done_unlock;
4929
4930                 case HOT_ADD_DISK:
4931                         err = hot_add_disk(mddev, new_decode_dev(arg));
4932                         goto done_unlock;
4933
4934                 case SET_DISK_FAULTY:
4935                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4936                         goto done_unlock;
4937
4938                 case RUN_ARRAY:
4939                         err = do_md_run (mddev);
4940                         goto done_unlock;
4941
4942                 case SET_BITMAP_FILE:
4943                         err = set_bitmap_file(mddev, (int)arg);
4944                         goto done_unlock;
4945
4946                 default:
4947                         err = -EINVAL;
4948                         goto abort_unlock;
4949         }
4950
4951 done_unlock:
4952 abort_unlock:
4953         mddev_unlock(mddev);
4954
4955         return err;
4956 done:
4957         if (err)
4958                 MD_BUG();
4959 abort:
4960         return err;
4961 }
4962
4963 static int md_open(struct inode *inode, struct file *file)
4964 {
4965         /*
4966          * Succeed if we can lock the mddev, which confirms that
4967          * it isn't being stopped right now.
4968          */
4969         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4970         int err;
4971
4972         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4973                 goto out;
4974
4975         err = 0;
4976         mddev_get(mddev);
4977         mddev_unlock(mddev);
4978
4979         check_disk_change(inode->i_bdev);
4980  out:
4981         return err;
4982 }
4983
4984 static int md_release(struct inode *inode, struct file * file)
4985 {
4986         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4987
4988         BUG_ON(!mddev);
4989         mddev_put(mddev);
4990
4991         return 0;
4992 }
4993
4994 static int md_media_changed(struct gendisk *disk)
4995 {
4996         mddev_t *mddev = disk->private_data;
4997
4998         return mddev->changed;
4999 }
5000
5001 static int md_revalidate(struct gendisk *disk)
5002 {
5003         mddev_t *mddev = disk->private_data;
5004
5005         mddev->changed = 0;
5006         return 0;
5007 }
5008 static struct block_device_operations md_fops =
5009 {
5010         .owner          = THIS_MODULE,
5011         .open           = md_open,
5012         .release        = md_release,
5013         .ioctl          = md_ioctl,
5014         .getgeo         = md_getgeo,
5015         .media_changed  = md_media_changed,
5016         .revalidate_disk= md_revalidate,
5017 };
5018
5019 static int md_thread(void * arg)
5020 {
5021         mdk_thread_t *thread = arg;
5022
5023         /*
5024          * md_thread is a 'system-thread', it's priority should be very
5025          * high. We avoid resource deadlocks individually in each
5026          * raid personality. (RAID5 does preallocation) We also use RR and
5027          * the very same RT priority as kswapd, thus we will never get
5028          * into a priority inversion deadlock.
5029          *
5030          * we definitely have to have equal or higher priority than
5031          * bdflush, otherwise bdflush will deadlock if there are too
5032          * many dirty RAID5 blocks.
5033          */
5034
5035         allow_signal(SIGKILL);
5036         while (!kthread_should_stop()) {
5037
5038                 /* We need to wait INTERRUPTIBLE so that
5039                  * we don't add to the load-average.
5040                  * That means we need to be sure no signals are
5041                  * pending
5042                  */
5043                 if (signal_pending(current))
5044                         flush_signals(current);
5045
5046                 wait_event_interruptible_timeout
5047                         (thread->wqueue,
5048                          test_bit(THREAD_WAKEUP, &thread->flags)
5049                          || kthread_should_stop(),
5050                          thread->timeout);
5051
5052                 clear_bit(THREAD_WAKEUP, &thread->flags);
5053
5054                 thread->run(thread->mddev);
5055         }
5056
5057         return 0;
5058 }
5059
5060 void md_wakeup_thread(mdk_thread_t *thread)
5061 {
5062         if (thread) {
5063                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5064                 set_bit(THREAD_WAKEUP, &thread->flags);
5065                 wake_up(&thread->wqueue);
5066         }
5067 }
5068
5069 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5070                                  const char *name)
5071 {
5072         mdk_thread_t *thread;
5073
5074         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5075         if (!thread)
5076                 return NULL;
5077
5078         init_waitqueue_head(&thread->wqueue);
5079
5080         thread->run = run;
5081         thread->mddev = mddev;
5082         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5083         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5084         if (IS_ERR(thread->tsk)) {
5085                 kfree(thread);
5086                 return NULL;
5087         }
5088         return thread;
5089 }
5090
5091 void md_unregister_thread(mdk_thread_t *thread)
5092 {
5093         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5094
5095         kthread_stop(thread->tsk);
5096         kfree(thread);
5097 }
5098
5099 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5100 {
5101         if (!mddev) {
5102                 MD_BUG();
5103                 return;
5104         }
5105
5106         if (!rdev || test_bit(Faulty, &rdev->flags))
5107                 return;
5108
5109         if (mddev->external)
5110                 set_bit(Blocked, &rdev->flags);
5111 /*
5112         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5113                 mdname(mddev),
5114                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5115                 __builtin_return_address(0),__builtin_return_address(1),
5116                 __builtin_return_address(2),__builtin_return_address(3));
5117 */
5118         if (!mddev->pers)
5119                 return;
5120         if (!mddev->pers->error_handler)
5121                 return;
5122         mddev->pers->error_handler(mddev,rdev);
5123         if (mddev->degraded)
5124                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5125         set_bit(StateChanged, &rdev->flags);
5126         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5127         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5128         md_wakeup_thread(mddev->thread);
5129         md_new_event_inintr(mddev);
5130 }
5131
5132 /* seq_file implementation /proc/mdstat */
5133
5134 static void status_unused(struct seq_file *seq)
5135 {
5136         int i = 0;
5137         mdk_rdev_t *rdev;
5138         struct list_head *tmp;
5139
5140         seq_printf(seq, "unused devices: ");
5141
5142         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5143                 char b[BDEVNAME_SIZE];
5144                 i++;
5145                 seq_printf(seq, "%s ",
5146                               bdevname(rdev->bdev,b));
5147         }
5148         if (!i)
5149                 seq_printf(seq, "<none>");
5150
5151         seq_printf(seq, "\n");
5152 }
5153
5154
5155 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5156 {
5157         sector_t max_blocks, resync, res;
5158         unsigned long dt, db, rt;
5159         int scale;
5160         unsigned int per_milli;
5161
5162         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5163
5164         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5165                 max_blocks = mddev->resync_max_sectors >> 1;
5166         else
5167                 max_blocks = mddev->size;
5168
5169         /*
5170          * Should not happen.
5171          */
5172         if (!max_blocks) {
5173                 MD_BUG();
5174                 return;
5175         }
5176         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5177          * in a sector_t, and (max_blocks>>scale) will fit in a
5178          * u32, as those are the requirements for sector_div.
5179          * Thus 'scale' must be at least 10
5180          */
5181         scale = 10;
5182         if (sizeof(sector_t) > sizeof(unsigned long)) {
5183                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5184                         scale++;
5185         }
5186         res = (resync>>scale)*1000;
5187         sector_div(res, (u32)((max_blocks>>scale)+1));
5188
5189         per_milli = res;
5190         {
5191                 int i, x = per_milli/50, y = 20-x;
5192                 seq_printf(seq, "[");
5193                 for (i = 0; i < x; i++)
5194                         seq_printf(seq, "=");
5195                 seq_printf(seq, ">");
5196                 for (i = 0; i < y; i++)
5197                         seq_printf(seq, ".");
5198                 seq_printf(seq, "] ");
5199         }
5200         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5201                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5202                     "reshape" :
5203                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5204                      "check" :
5205                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5206                       "resync" : "recovery"))),
5207                    per_milli/10, per_milli % 10,
5208                    (unsigned long long) resync,
5209                    (unsigned long long) max_blocks);
5210
5211         /*
5212          * We do not want to overflow, so the order of operands and
5213          * the * 100 / 100 trick are important. We do a +1 to be
5214          * safe against division by zero. We only estimate anyway.
5215          *
5216          * dt: time from mark until now
5217          * db: blocks written from mark until now
5218          * rt: remaining time
5219          */
5220         dt = ((jiffies - mddev->resync_mark) / HZ);
5221         if (!dt) dt++;
5222         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5223                 - mddev->resync_mark_cnt;
5224         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5225
5226         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5227
5228         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5229 }
5230
5231 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5232 {
5233         struct list_head *tmp;
5234         loff_t l = *pos;
5235         mddev_t *mddev;
5236
5237         if (l >= 0x10000)
5238                 return NULL;
5239         if (!l--)
5240                 /* header */
5241                 return (void*)1;
5242
5243         spin_lock(&all_mddevs_lock);
5244         list_for_each(tmp,&all_mddevs)
5245                 if (!l--) {
5246                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5247                         mddev_get(mddev);
5248                         spin_unlock(&all_mddevs_lock);
5249                         return mddev;
5250                 }
5251         spin_unlock(&all_mddevs_lock);
5252         if (!l--)
5253                 return (void*)2;/* tail */
5254         return NULL;
5255 }
5256
5257 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5258 {
5259         struct list_head *tmp;
5260         mddev_t *next_mddev, *mddev = v;
5261
5262         ++*pos;
5263         if (v == (void*)2)
5264                 return NULL;
5265
5266         spin_lock(&all_mddevs_lock);
5267         if (v == (void*)1)
5268                 tmp = all_mddevs.next;
5269         else
5270                 tmp = mddev->all_mddevs.next;
5271         if (tmp != &all_mddevs)
5272                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5273         else {
5274                 next_mddev = (void*)2;
5275                 *pos = 0x10000;
5276         }
5277         spin_unlock(&all_mddevs_lock);
5278
5279         if (v != (void*)1)
5280                 mddev_put(mddev);
5281         return next_mddev;
5282
5283 }
5284
5285 static void md_seq_stop(struct seq_file *seq, void *v)
5286 {
5287         mddev_t *mddev = v;
5288
5289         if (mddev && v != (void*)1 && v != (void*)2)
5290                 mddev_put(mddev);
5291 }
5292
5293 struct mdstat_info {
5294         int event;
5295 };
5296
5297 static int md_seq_show(struct seq_file *seq, void *v)
5298 {
5299         mddev_t *mddev = v;
5300         sector_t size;
5301         struct list_head *tmp2;
5302         mdk_rdev_t *rdev;
5303         struct mdstat_info *mi = seq->private;
5304         struct bitmap *bitmap;
5305
5306         if (v == (void*)1) {
5307                 struct mdk_personality *pers;
5308                 seq_printf(seq, "Personalities : ");
5309                 spin_lock(&pers_lock);
5310                 list_for_each_entry(pers, &pers_list, list)
5311                         seq_printf(seq, "[%s] ", pers->name);
5312
5313                 spin_unlock(&pers_lock);
5314                 seq_printf(seq, "\n");
5315                 mi->event = atomic_read(&md_event_count);
5316                 return 0;
5317         }
5318         if (v == (void*)2) {
5319                 status_unused(seq);
5320                 return 0;
5321         }
5322
5323         if (mddev_lock(mddev) < 0)
5324                 return -EINTR;
5325
5326         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5327                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5328                                                 mddev->pers ? "" : "in");
5329                 if (mddev->pers) {
5330                         if (mddev->ro==1)
5331                                 seq_printf(seq, " (read-only)");
5332                         if (mddev->ro==2)
5333                                 seq_printf(seq, " (auto-read-only)");
5334                         seq_printf(seq, " %s", mddev->pers->name);
5335                 }
5336
5337                 size = 0;
5338                 rdev_for_each(rdev, tmp2, mddev) {
5339                         char b[BDEVNAME_SIZE];
5340                         seq_printf(seq, " %s[%d]",
5341                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5342                         if (test_bit(WriteMostly, &rdev->flags))
5343                                 seq_printf(seq, "(W)");
5344                         if (test_bit(Faulty, &rdev->flags)) {
5345                                 seq_printf(seq, "(F)");
5346                                 continue;
5347                         } else if (rdev->raid_disk < 0)
5348                                 seq_printf(seq, "(S)"); /* spare */
5349                         size += rdev->size;
5350                 }
5351
5352                 if (!list_empty(&mddev->disks)) {
5353                         if (mddev->pers)
5354                                 seq_printf(seq, "\n      %llu blocks",
5355                                         (unsigned long long)mddev->array_size);
5356                         else
5357                                 seq_printf(seq, "\n      %llu blocks",
5358                                         (unsigned long long)size);
5359                 }
5360                 if (mddev->persistent) {
5361                         if (mddev->major_version != 0 ||
5362                             mddev->minor_version != 90) {
5363                                 seq_printf(seq," super %d.%d",
5364                                            mddev->major_version,
5365                                            mddev->minor_version);
5366                         }
5367                 } else if (mddev->external)
5368                         seq_printf(seq, " super external:%s",
5369                                    mddev->metadata_type);
5370                 else
5371                         seq_printf(seq, " super non-persistent");
5372
5373                 if (mddev->pers) {
5374                         mddev->pers->status (seq, mddev);
5375                         seq_printf(seq, "\n      ");
5376                         if (mddev->pers->sync_request) {
5377                                 if (mddev->curr_resync > 2) {
5378                                         status_resync (seq, mddev);
5379                                         seq_printf(seq, "\n      ");
5380                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5381                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5382                                 else if (mddev->recovery_cp < MaxSector)
5383                                         seq_printf(seq, "\tresync=PENDING\n      ");
5384                         }
5385                 } else
5386                         seq_printf(seq, "\n       ");
5387
5388                 if ((bitmap = mddev->bitmap)) {
5389                         unsigned long chunk_kb;
5390                         unsigned long flags;
5391                         spin_lock_irqsave(&bitmap->lock, flags);
5392                         chunk_kb = bitmap->chunksize >> 10;
5393                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5394                                 "%lu%s chunk",
5395                                 bitmap->pages - bitmap->missing_pages,
5396                                 bitmap->pages,
5397                                 (bitmap->pages - bitmap->missing_pages)
5398                                         << (PAGE_SHIFT - 10),
5399                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5400                                 chunk_kb ? "KB" : "B");
5401                         if (bitmap->file) {
5402                                 seq_printf(seq, ", file: ");
5403                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5404                         }
5405
5406                         seq_printf(seq, "\n");
5407                         spin_unlock_irqrestore(&bitmap->lock, flags);
5408                 }
5409
5410                 seq_printf(seq, "\n");
5411         }
5412         mddev_unlock(mddev);
5413
5414         return 0;
5415 }
5416
5417 static struct seq_operations md_seq_ops = {
5418         .start  = md_seq_start,
5419         .next   = md_seq_next,
5420         .stop   = md_seq_stop,
5421         .show   = md_seq_show,
5422 };
5423
5424 static int md_seq_open(struct inode *inode, struct file *file)
5425 {
5426         int error;
5427         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5428         if (mi == NULL)
5429                 return -ENOMEM;
5430
5431         error = seq_open(file, &md_seq_ops);
5432         if (error)
5433                 kfree(mi);
5434         else {
5435                 struct seq_file *p = file->private_data;
5436                 p->private = mi;
5437                 mi->event = atomic_read(&md_event_count);
5438         }
5439         return error;
5440 }
5441
5442 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5443 {
5444         struct seq_file *m = filp->private_data;
5445         struct mdstat_info *mi = m->private;
5446         int mask;
5447
5448         poll_wait(filp, &md_event_waiters, wait);
5449
5450         /* always allow read */
5451         mask = POLLIN | POLLRDNORM;
5452
5453         if (mi->event != atomic_read(&md_event_count))
5454                 mask |= POLLERR | POLLPRI;
5455         return mask;
5456 }
5457
5458 static const struct file_operations md_seq_fops = {
5459         .owner          = THIS_MODULE,
5460         .open           = md_seq_open,
5461         .read           = seq_read,
5462         .llseek         = seq_lseek,
5463         .release        = seq_release_private,
5464         .poll           = mdstat_poll,
5465 };
5466
5467 int register_md_personality(struct mdk_personality *p)
5468 {
5469         spin_lock(&pers_lock);
5470         list_add_tail(&p->list, &pers_list);
5471         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5472         spin_unlock(&pers_lock);
5473         return 0;
5474 }
5475
5476 int unregister_md_personality(struct mdk_personality *p)
5477 {
5478         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5479         spin_lock(&pers_lock);
5480         list_del_init(&p->list);
5481         spin_unlock(&pers_lock);
5482         return 0;
5483 }
5484
5485 static int is_mddev_idle(mddev_t *mddev)
5486 {
5487         mdk_rdev_t * rdev;
5488         struct list_head *tmp;
5489         int idle;
5490         long curr_events;
5491
5492         idle = 1;
5493         rdev_for_each(rdev, tmp, mddev) {
5494                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5495                 curr_events = disk_stat_read(disk, sectors[0]) +
5496                                 disk_stat_read(disk, sectors[1]) -
5497                                 atomic_read(&disk->sync_io);
5498                 /* sync IO will cause sync_io to increase before the disk_stats
5499                  * as sync_io is counted when a request starts, and
5500                  * disk_stats is counted when it completes.
5501                  * So resync activity will cause curr_events to be smaller than
5502                  * when there was no such activity.
5503                  * non-sync IO will cause disk_stat to increase without
5504                  * increasing sync_io so curr_events will (eventually)
5505                  * be larger than it was before.  Once it becomes
5506                  * substantially larger, the test below will cause
5507                  * the array to appear non-idle, and resync will slow
5508                  * down.
5509                  * If there is a lot of outstanding resync activity when
5510                  * we set last_event to curr_events, then all that activity
5511                  * completing might cause the array to appear non-idle
5512                  * and resync will be slowed down even though there might
5513                  * not have been non-resync activity.  This will only
5514                  * happen once though.  'last_events' will soon reflect
5515                  * the state where there is little or no outstanding
5516                  * resync requests, and further resync activity will
5517                  * always make curr_events less than last_events.
5518                  *
5519                  */
5520                 if (curr_events - rdev->last_events > 4096) {
5521                         rdev->last_events = curr_events;
5522                         idle = 0;
5523                 }
5524         }
5525         return idle;
5526 }
5527
5528 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5529 {
5530         /* another "blocks" (512byte) blocks have been synced */
5531         atomic_sub(blocks, &mddev->recovery_active);
5532         wake_up(&mddev->recovery_wait);
5533         if (!ok) {
5534                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5535                 md_wakeup_thread(mddev->thread);
5536                 // stop recovery, signal do_sync ....
5537         }
5538 }
5539
5540
5541 /* md_write_start(mddev, bi)
5542  * If we need to update some array metadata (e.g. 'active' flag
5543  * in superblock) before writing, schedule a superblock update
5544  * and wait for it to complete.
5545  */
5546 void md_write_start(mddev_t *mddev, struct bio *bi)
5547 {
5548         int did_change = 0;
5549         if (bio_data_dir(bi) != WRITE)
5550                 return;
5551
5552         BUG_ON(mddev->ro == 1);
5553         if (mddev->ro == 2) {
5554                 /* need to switch to read/write */
5555                 mddev->ro = 0;
5556                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5557                 md_wakeup_thread(mddev->thread);
5558                 md_wakeup_thread(mddev->sync_thread);
5559                 did_change = 1;
5560         }
5561         atomic_inc(&mddev->writes_pending);
5562         if (mddev->safemode == 1)
5563                 mddev->safemode = 0;
5564         if (mddev->in_sync) {
5565                 spin_lock_irq(&mddev->write_lock);
5566                 if (mddev->in_sync) {
5567                         mddev->in_sync = 0;
5568                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5569                         md_wakeup_thread(mddev->thread);
5570                         did_change = 1;
5571                 }
5572                 spin_unlock_irq(&mddev->write_lock);
5573         }
5574         if (did_change)
5575                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5576         wait_event(mddev->sb_wait,
5577                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5578                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5579 }
5580
5581 void md_write_end(mddev_t *mddev)
5582 {
5583         if (atomic_dec_and_test(&mddev->writes_pending)) {
5584                 if (mddev->safemode == 2)
5585                         md_wakeup_thread(mddev->thread);
5586                 else if (mddev->safemode_delay)
5587                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5588         }
5589 }
5590
5591 /* md_allow_write(mddev)
5592  * Calling this ensures that the array is marked 'active' so that writes
5593  * may proceed without blocking.  It is important to call this before
5594  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5595  * Must be called with mddev_lock held.
5596  */
5597 void md_allow_write(mddev_t *mddev)
5598 {
5599         if (!mddev->pers)
5600                 return;
5601         if (mddev->ro)
5602                 return;
5603         if (!mddev->pers->sync_request)
5604                 return;
5605
5606         spin_lock_irq(&mddev->write_lock);
5607         if (mddev->in_sync) {
5608                 mddev->in_sync = 0;
5609                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5610                 if (mddev->safemode_delay &&
5611                     mddev->safemode == 0)
5612                         mddev->safemode = 1;
5613                 spin_unlock_irq(&mddev->write_lock);
5614                 md_update_sb(mddev, 0);
5615
5616                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5617                 /* wait for the dirty state to be recorded in the metadata */
5618                 wait_event(mddev->sb_wait,
5619                            !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5620                            !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5621         } else
5622                 spin_unlock_irq(&mddev->write_lock);
5623 }
5624 EXPORT_SYMBOL_GPL(md_allow_write);
5625
5626 #define SYNC_MARKS      10
5627 #define SYNC_MARK_STEP  (3*HZ)
5628 void md_do_sync(mddev_t *mddev)
5629 {
5630         mddev_t *mddev2;
5631         unsigned int currspeed = 0,
5632                  window;
5633         sector_t max_sectors,j, io_sectors;
5634         unsigned long mark[SYNC_MARKS];
5635         sector_t mark_cnt[SYNC_MARKS];
5636         int last_mark,m;
5637         struct list_head *tmp;
5638         sector_t last_check;
5639         int skipped = 0;
5640         struct list_head *rtmp;
5641         mdk_rdev_t *rdev;
5642         char *desc;
5643
5644         /* just incase thread restarts... */
5645         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5646                 return;
5647         if (mddev->ro) /* never try to sync a read-only array */
5648                 return;
5649
5650         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5651                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5652                         desc = "data-check";
5653                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5654                         desc = "requested-resync";
5655                 else
5656                         desc = "resync";
5657         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5658                 desc = "reshape";
5659         else
5660                 desc = "recovery";
5661
5662         /* we overload curr_resync somewhat here.
5663          * 0 == not engaged in resync at all
5664          * 2 == checking that there is no conflict with another sync
5665          * 1 == like 2, but have yielded to allow conflicting resync to
5666          *              commense
5667          * other == active in resync - this many blocks
5668          *
5669          * Before starting a resync we must have set curr_resync to
5670          * 2, and then checked that every "conflicting" array has curr_resync
5671          * less than ours.  When we find one that is the same or higher
5672          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5673          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5674          * This will mean we have to start checking from the beginning again.
5675          *
5676          */
5677
5678         do {
5679                 mddev->curr_resync = 2;
5680
5681         try_again:
5682                 if (kthread_should_stop()) {
5683                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5684                         goto skip;
5685                 }
5686                 for_each_mddev(mddev2, tmp) {
5687                         if (mddev2 == mddev)
5688                                 continue;
5689                         if (!mddev->parallel_resync
5690                         &&  mddev2->curr_resync
5691                         &&  match_mddev_units(mddev, mddev2)) {
5692                                 DEFINE_WAIT(wq);
5693                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5694                                         /* arbitrarily yield */
5695                                         mddev->curr_resync = 1;
5696                                         wake_up(&resync_wait);
5697                                 }
5698                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5699                                         /* no need to wait here, we can wait the next
5700                                          * time 'round when curr_resync == 2
5701                                          */
5702                                         continue;
5703                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5704                                 if (!kthread_should_stop() &&
5705                                     mddev2->curr_resync >= mddev->curr_resync) {
5706                                         printk(KERN_INFO "md: delaying %s of %s"
5707                                                " until %s has finished (they"
5708                                                " share one or more physical units)\n",
5709                                                desc, mdname(mddev), mdname(mddev2));
5710                                         mddev_put(mddev2);
5711                                         schedule();
5712                                         finish_wait(&resync_wait, &wq);
5713                                         goto try_again;
5714                                 }
5715                                 finish_wait(&resync_wait, &wq);
5716                         }
5717                 }
5718         } while (mddev->curr_resync < 2);
5719
5720         j = 0;
5721         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5722                 /* resync follows the size requested by the personality,
5723                  * which defaults to physical size, but can be virtual size
5724                  */
5725                 max_sectors = mddev->resync_max_sectors;
5726                 mddev->resync_mismatches = 0;
5727                 /* we don't use the checkpoint if there's a bitmap */
5728                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5729                         j = mddev->resync_min;
5730                 else if (!mddev->bitmap)
5731                         j = mddev->recovery_cp;
5732
5733         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5734                 max_sectors = mddev->size << 1;
5735         else {
5736                 /* recovery follows the physical size of devices */
5737                 max_sectors = mddev->size << 1;
5738                 j = MaxSector;
5739                 rdev_for_each(rdev, rtmp, mddev)
5740                         if (rdev->raid_disk >= 0 &&
5741                             !test_bit(Faulty, &rdev->flags) &&
5742                             !test_bit(In_sync, &rdev->flags) &&
5743                             rdev->recovery_offset < j)
5744                                 j = rdev->recovery_offset;
5745         }
5746
5747         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5748         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5749                 " %d KB/sec/disk.\n", speed_min(mddev));
5750         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5751                "(but not more than %d KB/sec) for %s.\n",
5752                speed_max(mddev), desc);
5753
5754         is_mddev_idle(mddev); /* this also initializes IO event counters */
5755
5756         io_sectors = 0;
5757         for (m = 0; m < SYNC_MARKS; m++) {
5758                 mark[m] = jiffies;
5759                 mark_cnt[m] = io_sectors;
5760         }
5761         last_mark = 0;
5762         mddev->resync_mark = mark[last_mark];
5763         mddev->resync_mark_cnt = mark_cnt[last_mark];
5764
5765         /*
5766          * Tune reconstruction:
5767          */
5768         window = 32*(PAGE_SIZE/512);
5769         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5770                 window/2,(unsigned long long) max_sectors/2);
5771
5772         atomic_set(&mddev->recovery_active, 0);
5773         last_check = 0;
5774
5775         if (j>2) {
5776                 printk(KERN_INFO
5777                        "md: resuming %s of %s from checkpoint.\n",
5778                        desc, mdname(mddev));
5779                 mddev->curr_resync = j;
5780         }
5781
5782         while (j < max_sectors) {
5783                 sector_t sectors;
5784
5785                 skipped = 0;
5786                 if (j >= mddev->resync_max) {
5787                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5788                         wait_event(mddev->recovery_wait,
5789                                    mddev->resync_max > j
5790                                    || kthread_should_stop());
5791                 }
5792                 if (kthread_should_stop())
5793                         goto interrupted;
5794                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5795                                                   currspeed < speed_min(mddev));
5796                 if (sectors == 0) {
5797                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5798                         goto out;
5799                 }
5800
5801                 if (!skipped) { /* actual IO requested */
5802                         io_sectors += sectors;
5803                         atomic_add(sectors, &mddev->recovery_active);
5804                 }
5805
5806                 j += sectors;
5807                 if (j>1) mddev->curr_resync = j;
5808                 mddev->curr_mark_cnt = io_sectors;
5809                 if (last_check == 0)
5810                         /* this is the earliers that rebuilt will be
5811                          * visible in /proc/mdstat
5812                          */
5813                         md_new_event(mddev);
5814
5815                 if (last_check + window > io_sectors || j == max_sectors)
5816                         continue;
5817
5818                 last_check = io_sectors;
5819
5820                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5821                         break;
5822
5823         repeat:
5824                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5825                         /* step marks */
5826                         int next = (last_mark+1) % SYNC_MARKS;
5827
5828                         mddev->resync_mark = mark[next];
5829                         mddev->resync_mark_cnt = mark_cnt[next];
5830                         mark[next] = jiffies;
5831                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5832                         last_mark = next;
5833                 }
5834
5835
5836                 if (kthread_should_stop())
5837                         goto interrupted;
5838
5839
5840                 /*
5841                  * this loop exits only if either when we are slower than
5842                  * the 'hard' speed limit, or the system was IO-idle for
5843                  * a jiffy.
5844                  * the system might be non-idle CPU-wise, but we only care
5845                  * about not overloading the IO subsystem. (things like an
5846                  * e2fsck being done on the RAID array should execute fast)
5847                  */
5848                 blk_unplug(mddev->queue);
5849                 cond_resched();
5850
5851                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5852                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5853
5854                 if (currspeed > speed_min(mddev)) {
5855                         if ((currspeed > speed_max(mddev)) ||
5856                                         !is_mddev_idle(mddev)) {
5857                                 msleep(500);
5858                                 goto repeat;
5859                         }
5860                 }
5861         }
5862         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5863         /*
5864          * this also signals 'finished resyncing' to md_stop
5865          */
5866  out:
5867         blk_unplug(mddev->queue);
5868
5869         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5870
5871         /* tell personality that we are finished */
5872         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5873
5874         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5875             mddev->curr_resync > 2) {
5876                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5877                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5878                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5879                                         printk(KERN_INFO
5880                                                "md: checkpointing %s of %s.\n",
5881                                                desc, mdname(mddev));
5882                                         mddev->recovery_cp = mddev->curr_resync;
5883                                 }
5884                         } else
5885                                 mddev->recovery_cp = MaxSector;
5886                 } else {
5887                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5888                                 mddev->curr_resync = MaxSector;
5889                         rdev_for_each(rdev, rtmp, mddev)
5890                                 if (rdev->raid_disk >= 0 &&
5891                                     !test_bit(Faulty, &rdev->flags) &&
5892                                     !test_bit(In_sync, &rdev->flags) &&
5893                                     rdev->recovery_offset < mddev->curr_resync)
5894                                         rdev->recovery_offset = mddev->curr_resync;
5895                 }
5896         }
5897         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5898
5899  skip:
5900         mddev->curr_resync = 0;
5901         mddev->resync_min = 0;
5902         mddev->resync_max = MaxSector;
5903         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5904         wake_up(&resync_wait);
5905         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5906         md_wakeup_thread(mddev->thread);
5907         return;
5908
5909  interrupted:
5910         /*
5911          * got a signal, exit.
5912          */
5913         printk(KERN_INFO
5914                "md: md_do_sync() got signal ... exiting\n");
5915         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5916         goto out;
5917
5918 }
5919 EXPORT_SYMBOL_GPL(md_do_sync);
5920
5921
5922 static int remove_and_add_spares(mddev_t *mddev)
5923 {
5924         mdk_rdev_t *rdev;
5925         struct list_head *rtmp;
5926         int spares = 0;
5927
5928         rdev_for_each(rdev, rtmp, mddev)
5929                 if (rdev->raid_disk >= 0 &&
5930                     !test_bit(Blocked, &rdev->flags) &&
5931                     (test_bit(Faulty, &rdev->flags) ||
5932                      ! test_bit(In_sync, &rdev->flags)) &&
5933                     atomic_read(&rdev->nr_pending)==0) {
5934                         if (mddev->pers->hot_remove_disk(
5935                                     mddev, rdev->raid_disk)==0) {
5936                                 char nm[20];
5937                                 sprintf(nm,"rd%d", rdev->raid_disk);
5938                                 sysfs_remove_link(&mddev->kobj, nm);
5939                                 rdev->raid_disk = -1;
5940                         }
5941                 }
5942
5943         if (mddev->degraded) {
5944                 rdev_for_each(rdev, rtmp, mddev) {
5945                         if (rdev->raid_disk >= 0 &&
5946                             !test_bit(In_sync, &rdev->flags))
5947                                 spares++;
5948                         if (rdev->raid_disk < 0
5949                             && !test_bit(Faulty, &rdev->flags)) {
5950                                 rdev->recovery_offset = 0;
5951                                 if (mddev->pers->
5952                                     hot_add_disk(mddev, rdev) == 0) {
5953                                         char nm[20];
5954                                         sprintf(nm, "rd%d", rdev->raid_disk);
5955                                         if (sysfs_create_link(&mddev->kobj,
5956                                                               &rdev->kobj, nm))
5957                                                 printk(KERN_WARNING
5958                                                        "md: cannot register "
5959                                                        "%s for %s\n",
5960                                                        nm, mdname(mddev));
5961                                         spares++;
5962                                         md_new_event(mddev);
5963                                 } else
5964                                         break;
5965                         }
5966                 }
5967         }
5968         return spares;
5969 }
5970 /*
5971  * This routine is regularly called by all per-raid-array threads to
5972  * deal with generic issues like resync and super-block update.
5973  * Raid personalities that don't have a thread (linear/raid0) do not
5974  * need this as they never do any recovery or update the superblock.
5975  *
5976  * It does not do any resync itself, but rather "forks" off other threads
5977  * to do that as needed.
5978  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5979  * "->recovery" and create a thread at ->sync_thread.
5980  * When the thread finishes it sets MD_RECOVERY_DONE
5981  * and wakeups up this thread which will reap the thread and finish up.
5982  * This thread also removes any faulty devices (with nr_pending == 0).
5983  *
5984  * The overall approach is:
5985  *  1/ if the superblock needs updating, update it.
5986  *  2/ If a recovery thread is running, don't do anything else.
5987  *  3/ If recovery has finished, clean up, possibly marking spares active.
5988  *  4/ If there are any faulty devices, remove them.
5989  *  5/ If array is degraded, try to add spares devices
5990  *  6/ If array has spares or is not in-sync, start a resync thread.
5991  */
5992 void md_check_recovery(mddev_t *mddev)
5993 {
5994         mdk_rdev_t *rdev;
5995         struct list_head *rtmp;
5996
5997
5998         if (mddev->bitmap)
5999                 bitmap_daemon_work(mddev->bitmap);
6000
6001         if (mddev->ro)
6002                 return;
6003
6004         if (signal_pending(current)) {
6005                 if (mddev->pers->sync_request && !mddev->external) {
6006                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6007                                mdname(mddev));
6008                         mddev->safemode = 2;
6009                 }
6010                 flush_signals(current);
6011         }
6012
6013         if ( ! (
6014                 (mddev->flags && !mddev->external) ||
6015                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6016                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6017                 (mddev->external == 0 && mddev->safemode == 1) ||
6018                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6019                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6020                 ))
6021                 return;
6022
6023         if (mddev_trylock(mddev)) {
6024                 int spares = 0;
6025
6026                 if (!mddev->external) {
6027                         int did_change = 0;
6028                         spin_lock_irq(&mddev->write_lock);
6029                         if (mddev->safemode &&
6030                             !atomic_read(&mddev->writes_pending) &&
6031                             !mddev->in_sync &&
6032                             mddev->recovery_cp == MaxSector) {
6033                                 mddev->in_sync = 1;
6034                                 did_change = 1;
6035                                 if (mddev->persistent)
6036                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6037                         }
6038                         if (mddev->safemode == 1)
6039                                 mddev->safemode = 0;
6040                         spin_unlock_irq(&mddev->write_lock);
6041                         if (did_change)
6042                                 sysfs_notify(&mddev->kobj, NULL, "array_state");
6043                 }
6044
6045                 if (mddev->flags)
6046                         md_update_sb(mddev, 0);
6047
6048                 rdev_for_each(rdev, rtmp, mddev)
6049                         if (test_and_clear_bit(StateChanged, &rdev->flags))
6050                                 sysfs_notify(&rdev->kobj, NULL, "state");
6051
6052
6053                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6054                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6055                         /* resync/recovery still happening */
6056                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6057                         goto unlock;
6058                 }
6059                 if (mddev->sync_thread) {
6060                         /* resync has finished, collect result */
6061                         md_unregister_thread(mddev->sync_thread);
6062                         mddev->sync_thread = NULL;
6063                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6064                                 /* success...*/
6065                                 /* activate any spares */
6066                                 if (mddev->pers->spare_active(mddev))
6067                                         sysfs_notify(&mddev->kobj, NULL,
6068                                                      "degraded");
6069                         }
6070                         md_update_sb(mddev, 1);
6071
6072                         /* if array is no-longer degraded, then any saved_raid_disk
6073                          * information must be scrapped
6074                          */
6075                         if (!mddev->degraded)
6076                                 rdev_for_each(rdev, rtmp, mddev)
6077                                         rdev->saved_raid_disk = -1;
6078
6079                         mddev->recovery = 0;
6080                         /* flag recovery needed just to double check */
6081                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6082                         sysfs_notify(&mddev->kobj, NULL, "sync_action");
6083                         md_new_event(mddev);
6084                         goto unlock;
6085                 }
6086                 /* Set RUNNING before clearing NEEDED to avoid
6087                  * any transients in the value of "sync_action".
6088                  */
6089                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6090                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6091                 /* Clear some bits that don't mean anything, but
6092                  * might be left set
6093                  */
6094                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6095                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6096
6097                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6098                         goto unlock;
6099                 /* no recovery is running.
6100                  * remove any failed drives, then
6101                  * add spares if possible.
6102                  * Spare are also removed and re-added, to allow
6103                  * the personality to fail the re-add.
6104                  */
6105
6106                 if (mddev->reshape_position != MaxSector) {
6107                         if (mddev->pers->check_reshape(mddev) != 0)
6108                                 /* Cannot proceed */
6109                                 goto unlock;
6110                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6111                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6112                 } else if ((spares = remove_and_add_spares(mddev))) {
6113                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6114                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6115                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6116                 } else if (mddev->recovery_cp < MaxSector) {
6117                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6118                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6119                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6120                         /* nothing to be done ... */
6121                         goto unlock;
6122
6123                 if (mddev->pers->sync_request) {
6124                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6125                                 /* We are adding a device or devices to an array
6126                                  * which has the bitmap stored on all devices.
6127                                  * So make sure all bitmap pages get written
6128                                  */
6129                                 bitmap_write_all(mddev->bitmap);
6130                         }
6131                         mddev->sync_thread = md_register_thread(md_do_sync,
6132                                                                 mddev,
6133                                                                 "%s_resync");
6134                         if (!mddev->sync_thread) {
6135                                 printk(KERN_ERR "%s: could not start resync"
6136                                         " thread...\n",
6137                                         mdname(mddev));
6138                                 /* leave the spares where they are, it shouldn't hurt */
6139                                 mddev->recovery = 0;
6140                         } else
6141                                 md_wakeup_thread(mddev->sync_thread);
6142                         sysfs_notify(&mddev->kobj, NULL, "sync_action");
6143                         md_new_event(mddev);
6144                 }
6145         unlock:
6146                 if (!mddev->sync_thread) {
6147                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6148                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6149                                                &mddev->recovery))
6150                                 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6151                 }
6152                 mddev_unlock(mddev);
6153         }
6154 }
6155
6156 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6157 {
6158         sysfs_notify(&rdev->kobj, NULL, "state");
6159         wait_event_timeout(rdev->blocked_wait,
6160                            !test_bit(Blocked, &rdev->flags),
6161                            msecs_to_jiffies(5000));
6162         rdev_dec_pending(rdev, mddev);
6163 }
6164 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6165
6166 static int md_notify_reboot(struct notifier_block *this,
6167                             unsigned long code, void *x)
6168 {
6169         struct list_head *tmp;
6170         mddev_t *mddev;
6171
6172         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6173
6174                 printk(KERN_INFO "md: stopping all md devices.\n");
6175
6176                 for_each_mddev(mddev, tmp)
6177                         if (mddev_trylock(mddev)) {
6178                                 do_md_stop (mddev, 1);
6179                                 mddev_unlock(mddev);
6180                         }
6181                 /*
6182                  * certain more exotic SCSI devices are known to be
6183                  * volatile wrt too early system reboots. While the
6184                  * right place to handle this issue is the given
6185                  * driver, we do want to have a safe RAID driver ...
6186                  */
6187                 mdelay(1000*1);
6188         }
6189         return NOTIFY_DONE;
6190 }
6191
6192 static struct notifier_block md_notifier = {
6193         .notifier_call  = md_notify_reboot,
6194         .next           = NULL,
6195         .priority       = INT_MAX, /* before any real devices */
6196 };
6197
6198 static void md_geninit(void)
6199 {
6200         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6201
6202         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6203 }
6204
6205 static int __init md_init(void)
6206 {
6207         if (register_blkdev(MAJOR_NR, "md"))
6208                 return -1;
6209         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6210                 unregister_blkdev(MAJOR_NR, "md");
6211                 return -1;
6212         }
6213         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6214                             md_probe, NULL, NULL);
6215         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6216                             md_probe, NULL, NULL);
6217
6218         register_reboot_notifier(&md_notifier);
6219         raid_table_header = register_sysctl_table(raid_root_table);
6220
6221         md_geninit();
6222         return (0);
6223 }
6224
6225
6226 #ifndef MODULE
6227
6228 /*
6229  * Searches all registered partitions for autorun RAID arrays
6230  * at boot time.
6231  */
6232
6233 static LIST_HEAD(all_detected_devices);
6234 struct detected_devices_node {
6235         struct list_head list;
6236         dev_t dev;
6237 };
6238
6239 void md_autodetect_dev(dev_t dev)
6240 {
6241         struct detected_devices_node *node_detected_dev;
6242
6243         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6244         if (node_detected_dev) {
6245                 node_detected_dev->dev = dev;
6246                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6247         } else {
6248                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6249                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6250         }
6251 }
6252
6253
6254 static void autostart_arrays(int part)
6255 {
6256         mdk_rdev_t *rdev;
6257         struct detected_devices_node *node_detected_dev;
6258         dev_t dev;
6259         int i_scanned, i_passed;
6260
6261         i_scanned = 0;
6262         i_passed = 0;
6263
6264         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6265
6266         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6267                 i_scanned++;
6268                 node_detected_dev = list_entry(all_detected_devices.next,
6269                                         struct detected_devices_node, list);
6270                 list_del(&node_detected_dev->list);
6271                 dev = node_detected_dev->dev;
6272                 kfree(node_detected_dev);
6273                 rdev = md_import_device(dev,0, 90);
6274                 if (IS_ERR(rdev))
6275                         continue;
6276
6277                 if (test_bit(Faulty, &rdev->flags)) {
6278                         MD_BUG();
6279                         continue;
6280                 }
6281                 set_bit(AutoDetected, &rdev->flags);
6282                 list_add(&rdev->same_set, &pending_raid_disks);
6283                 i_passed++;
6284         }
6285
6286         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6287                                                 i_scanned, i_passed);
6288
6289         autorun_devices(part);
6290 }
6291
6292 #endif /* !MODULE */
6293
6294 static __exit void md_exit(void)
6295 {
6296         mddev_t *mddev;
6297         struct list_head *tmp;
6298
6299         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6300         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6301
6302         unregister_blkdev(MAJOR_NR,"md");
6303         unregister_blkdev(mdp_major, "mdp");
6304         unregister_reboot_notifier(&md_notifier);
6305         unregister_sysctl_table(raid_table_header);
6306         remove_proc_entry("mdstat", NULL);
6307         for_each_mddev(mddev, tmp) {
6308                 struct gendisk *disk = mddev->gendisk;
6309                 if (!disk)
6310                         continue;
6311                 export_array(mddev);
6312                 del_gendisk(disk);
6313                 put_disk(disk);
6314                 mddev->gendisk = NULL;
6315                 mddev_put(mddev);
6316         }
6317 }
6318
6319 subsys_initcall(md_init);
6320 module_exit(md_exit)
6321
6322 static int get_ro(char *buffer, struct kernel_param *kp)
6323 {
6324         return sprintf(buffer, "%d", start_readonly);
6325 }
6326 static int set_ro(const char *val, struct kernel_param *kp)
6327 {
6328         char *e;
6329         int num = simple_strtoul(val, &e, 10);
6330         if (*val && (*e == '\0' || *e == '\n')) {
6331                 start_readonly = num;
6332                 return 0;
6333         }
6334         return -EINVAL;
6335 }
6336
6337 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6338 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6339
6340
6341 EXPORT_SYMBOL(register_md_personality);
6342 EXPORT_SYMBOL(unregister_md_personality);
6343 EXPORT_SYMBOL(md_error);
6344 EXPORT_SYMBOL(md_done_sync);
6345 EXPORT_SYMBOL(md_write_start);
6346 EXPORT_SYMBOL(md_write_end);
6347 EXPORT_SYMBOL(md_register_thread);
6348 EXPORT_SYMBOL(md_unregister_thread);
6349 EXPORT_SYMBOL(md_wakeup_thread);
6350 EXPORT_SYMBOL(md_check_recovery);
6351 MODULE_LICENSE("GPL");
6352 MODULE_ALIAS("md");
6353 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);