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