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