drivers/md/md.c

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