Pileus Git - ~andy/linux/blob - 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/config.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/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45
  46 #include <linux/init.h>
  47
  48 #include <linux/file.h>
  49
  50 #ifdef CONFIG_KMOD
  51 #include <linux/kmod.h>
  52 #endif
  53
  54 #include <asm/unaligned.h>
  55
  56 #define MAJOR_NR MD_MAJOR
  57 #define MD_DRIVER
  58
  59 /* 63 partitions with the alternate major number (mdp) */
  60 #define MdpMinorShift 6
  61
  62 #define DEBUG 0
  63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  64
  65
  66 #ifndef MODULE
  67 static void autostart_arrays (int part);
  68 #endif
  69
  70 static mdk_personality_t *pers[MAX_PERSONALITY];
  71 static DEFINE_SPINLOCK(pers_lock);
  72
  73 /*
  74  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  75  * is 1000 KB/sec, so the extra system load does not show up that much.
  76  * Increase it if you want to have more _guaranteed_ speed. Note that
  77  * the RAID driver will use the maximum available bandwidth if the IO
  78  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  79  * speed limit - in case reconstruction slows down your system despite
  80  * idle IO detection.
  81  *
  82  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  83  */
  84
  85 static int sysctl_speed_limit_min = 1000;
  86 static int sysctl_speed_limit_max = 200000;
  87
  88 static struct ctl_table_header *raid_table_header;
  89
  90 static ctl_table raid_table[] = {
  91         {
  92                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  93                 .procname       = "speed_limit_min",
  94                 .data           = &sysctl_speed_limit_min,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec,
  98         },
  99         {
 100                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 101                 .procname       = "speed_limit_max",
 102                 .data           = &sysctl_speed_limit_max,
 103                 .maxlen         = sizeof(int),
 104                 .mode           = 0644,
 105                 .proc_handler   = &proc_dointvec,
 106         },
 107         { .ctl_name = 0 }
 108 };
 109
 110 static ctl_table raid_dir_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID,
 113                 .procname       = "raid",
 114                 .maxlen         = 0,
 115                 .mode           = 0555,
 116                 .child          = raid_table,
 117         },
 118         { .ctl_name = 0 }
 119 };
 120
 121 static ctl_table raid_root_table[] = {
 122         {
 123                 .ctl_name       = CTL_DEV,
 124                 .procname       = "dev",
 125                 .maxlen         = 0,
 126                 .mode           = 0555,
 127                 .child          = raid_dir_table,
 128         },
 129         { .ctl_name = 0 }
 130 };
 131
 132 static struct block_device_operations md_fops;
 133
 134 /*
 135  * Enables to iterate over all existing md arrays
 136  * all_mddevs_lock protects this list.
 137  */
 138 static LIST_HEAD(all_mddevs);
 139 static DEFINE_SPINLOCK(all_mddevs_lock);
 140
 141
 142 /*
 143  * iterates through all used mddevs in the system.
 144  * We take care to grab the all_mddevs_lock whenever navigating
 145  * the list, and to always hold a refcount when unlocked.
 146  * Any code which breaks out of this loop while own
 147  * a reference to the current mddev and must mddev_put it.
 148  */
 149 #define ITERATE_MDDEV(mddev,tmp)                                        \
 150                                                                         \
 151         for (({ spin_lock(&all_mddevs_lock);                            \
 152                 tmp = all_mddevs.next;                                  \
 153                 mddev = NULL;});                                        \
 154              ({ if (tmp != &all_mddevs)                                 \
 155                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 156                 spin_unlock(&all_mddevs_lock);                          \
 157                 if (mddev) mddev_put(mddev);                            \
 158                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 159                 tmp != &all_mddevs;});                                  \
 160              ({ spin_lock(&all_mddevs_lock);                            \
 161                 tmp = tmp->next;})                                      \
 162                 )
 163
 164
 165 static int md_fail_request (request_queue_t *q, struct bio *bio)
 166 {
 167         bio_io_error(bio, bio->bi_size);
 168         return 0;
 169 }
 170
 171 static inline mddev_t *mddev_get(mddev_t *mddev)
 172 {
 173         atomic_inc(&mddev->active);
 174         return mddev;
 175 }
 176
 177 static void mddev_put(mddev_t *mddev)
 178 {
 179         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 180                 return;
 181         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 182                 list_del(&mddev->all_mddevs);
 183                 blk_put_queue(mddev->queue);
 184                 kobject_unregister(&mddev->kobj);
 185         }
 186         spin_unlock(&all_mddevs_lock);
 187 }
 188
 189 static mddev_t * mddev_find(dev_t unit)
 190 {
 191         mddev_t *mddev, *new = NULL;
 192
 193  retry:
 194         spin_lock(&all_mddevs_lock);
 195         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 196                 if (mddev->unit == unit) {
 197                         mddev_get(mddev);
 198                         spin_unlock(&all_mddevs_lock);
 199                         kfree(new);
 200                         return mddev;
 201                 }
 202
 203         if (new) {
 204                 list_add(&new->all_mddevs, &all_mddevs);
 205                 spin_unlock(&all_mddevs_lock);
 206                 return new;
 207         }
 208         spin_unlock(&all_mddevs_lock);
 209
 210         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 211         if (!new)
 212                 return NULL;
 213
 214         memset(new, 0, sizeof(*new));
 215
 216         new->unit = unit;
 217         if (MAJOR(unit) == MD_MAJOR)
 218                 new->md_minor = MINOR(unit);
 219         else
 220                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 221
 222         init_MUTEX(&new->reconfig_sem);
 223         INIT_LIST_HEAD(&new->disks);
 224         INIT_LIST_HEAD(&new->all_mddevs);
 225         init_timer(&new->safemode_timer);
 226         atomic_set(&new->active, 1);
 227         spin_lock_init(&new->write_lock);
 228         init_waitqueue_head(&new->sb_wait);
 229
 230         new->queue = blk_alloc_queue(GFP_KERNEL);
 231         if (!new->queue) {
 232                 kfree(new);
 233                 return NULL;
 234         }
 235
 236         blk_queue_make_request(new->queue, md_fail_request);
 237
 238         goto retry;
 239 }
 240
 241 static inline int mddev_lock(mddev_t * mddev)
 242 {
 243         return down_interruptible(&mddev->reconfig_sem);
 244 }
 245
 246 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 247 {
 248         down(&mddev->reconfig_sem);
 249 }
 250
 251 static inline int mddev_trylock(mddev_t * mddev)
 252 {
 253         return down_trylock(&mddev->reconfig_sem);
 254 }
 255
 256 static inline void mddev_unlock(mddev_t * mddev)
 257 {
 258         up(&mddev->reconfig_sem);
 259
 260         md_wakeup_thread(mddev->thread);
 261 }
 262
 263 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 264 {
 265         mdk_rdev_t * rdev;
 266         struct list_head *tmp;
 267
 268         ITERATE_RDEV(mddev,rdev,tmp) {
 269                 if (rdev->desc_nr == nr)
 270                         return rdev;
 271         }
 272         return NULL;
 273 }
 274
 275 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 276 {
 277         struct list_head *tmp;
 278         mdk_rdev_t *rdev;
 279
 280         ITERATE_RDEV(mddev,rdev,tmp) {
 281                 if (rdev->bdev->bd_dev == dev)
 282                         return rdev;
 283         }
 284         return NULL;
 285 }
 286
 287 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 288 {
 289         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 290         return MD_NEW_SIZE_BLOCKS(size);
 291 }
 292
 293 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 294 {
 295         sector_t size;
 296
 297         size = rdev->sb_offset;
 298
 299         if (chunk_size)
 300                 size &= ~((sector_t)chunk_size/1024 - 1);
 301         return size;
 302 }
 303
 304 static int alloc_disk_sb(mdk_rdev_t * rdev)
 305 {
 306         if (rdev->sb_page)
 307                 MD_BUG();
 308
 309         rdev->sb_page = alloc_page(GFP_KERNEL);
 310         if (!rdev->sb_page) {
 311                 printk(KERN_ALERT "md: out of memory.\n");
 312                 return -EINVAL;
 313         }
 314
 315         return 0;
 316 }
 317
 318 static void free_disk_sb(mdk_rdev_t * rdev)
 319 {
 320         if (rdev->sb_page) {
 321                 page_cache_release(rdev->sb_page);
 322                 rdev->sb_loaded = 0;
 323                 rdev->sb_page = NULL;
 324                 rdev->sb_offset = 0;
 325                 rdev->size = 0;
 326         }
 327 }
 328
 329
 330 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 331 {
 332         mdk_rdev_t *rdev = bio->bi_private;
 333         if (bio->bi_size)
 334                 return 1;
 335
 336         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 337                 md_error(rdev->mddev, rdev);
 338
 339         if (atomic_dec_and_test(&rdev->mddev->pending_writes))
 340                 wake_up(&rdev->mddev->sb_wait);
 341         bio_put(bio);
 342         return 0;
 343 }
 344
 345 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 346                    sector_t sector, int size, struct page *page)
 347 {
 348         /* write first size bytes of page to sector of rdev
 349          * Increment mddev->pending_writes before returning
 350          * and decrement it on completion, waking up sb_wait
 351          * if zero is reached.
 352          * If an error occurred, call md_error
 353          */
 354         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 355
 356         bio->bi_bdev = rdev->bdev;
 357         bio->bi_sector = sector;
 358         bio_add_page(bio, page, size, 0);
 359         bio->bi_private = rdev;
 360         bio->bi_end_io = super_written;
 361         atomic_inc(&mddev->pending_writes);
 362         submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
 363 }
 364
 365 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 366 {
 367         if (bio->bi_size)
 368                 return 1;
 369
 370         complete((struct completion*)bio->bi_private);
 371         return 0;
 372 }
 373
 374 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 375                    struct page *page, int rw)
 376 {
 377         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 378         struct completion event;
 379         int ret;
 380
 381         rw |= (1 << BIO_RW_SYNC);
 382
 383         bio->bi_bdev = bdev;
 384         bio->bi_sector = sector;
 385         bio_add_page(bio, page, size, 0);
 386         init_completion(&event);
 387         bio->bi_private = &event;
 388         bio->bi_end_io = bi_complete;
 389         submit_bio(rw, bio);
 390         wait_for_completion(&event);
 391
 392         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 393         bio_put(bio);
 394         return ret;
 395 }
 396
 397 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 398 {
 399         char b[BDEVNAME_SIZE];
 400         if (!rdev->sb_page) {
 401                 MD_BUG();
 402                 return -EINVAL;
 403         }
 404         if (rdev->sb_loaded)
 405                 return 0;
 406
 407
 408         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 409                 goto fail;
 410         rdev->sb_loaded = 1;
 411         return 0;
 412
 413 fail:
 414         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 415                 bdevname(rdev->bdev,b));
 416         return -EINVAL;
 417 }
 418
 419 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 420 {
 421         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 422                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 423                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 424                 (sb1->set_uuid3 == sb2->set_uuid3))
 425
 426                 return 1;
 427
 428         return 0;
 429 }
 430
 431
 432 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 433 {
 434         int ret;
 435         mdp_super_t *tmp1, *tmp2;
 436
 437         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 438         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 439
 440         if (!tmp1 || !tmp2) {
 441                 ret = 0;
 442                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 443                 goto abort;
 444         }
 445
 446         *tmp1 = *sb1;
 447         *tmp2 = *sb2;
 448
 449         /*
 450          * nr_disks is not constant
 451          */
 452         tmp1->nr_disks = 0;
 453         tmp2->nr_disks = 0;
 454
 455         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 456                 ret = 0;
 457         else
 458                 ret = 1;
 459
 460 abort:
 461         kfree(tmp1);
 462         kfree(tmp2);
 463         return ret;
 464 }
 465
 466 static unsigned int calc_sb_csum(mdp_super_t * sb)
 467 {
 468         unsigned int disk_csum, csum;
 469
 470         disk_csum = sb->sb_csum;
 471         sb->sb_csum = 0;
 472         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 473         sb->sb_csum = disk_csum;
 474         return csum;
 475 }
 476
 477
 478 /*
 479  * Handle superblock details.
 480  * We want to be able to handle multiple superblock formats
 481  * so we have a common interface to them all, and an array of
 482  * different handlers.
 483  * We rely on user-space to write the initial superblock, and support
 484  * reading and updating of superblocks.
 485  * Interface methods are:
 486  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 487  *      loads and validates a superblock on dev.
 488  *      if refdev != NULL, compare superblocks on both devices
 489  *    Return:
 490  *      0 - dev has a superblock that is compatible with refdev
 491  *      1 - dev has a superblock that is compatible and newer than refdev
 492  *          so dev should be used as the refdev in future
 493  *     -EINVAL superblock incompatible or invalid
 494  *     -othererror e.g. -EIO
 495  *
 496  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 497  *      Verify that dev is acceptable into mddev.
 498  *       The first time, mddev->raid_disks will be 0, and data from
 499  *       dev should be merged in.  Subsequent calls check that dev
 500  *       is new enough.  Return 0 or -EINVAL
 501  *
 502  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 503  *     Update the superblock for rdev with data in mddev
 504  *     This does not write to disc.
 505  *
 506  */
 507
 508 struct super_type  {
 509         char            *name;
 510         struct module   *owner;
 511         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 512         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 513         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 514 };
 515
 516 /*
 517  * load_super for 0.90.0
 518  */
 519 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 520 {
 521         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 522         mdp_super_t *sb;
 523         int ret;
 524         sector_t sb_offset;
 525
 526         /*
 527          * Calculate the position of the superblock,
 528          * it's at the end of the disk.
 529          *
 530          * It also happens to be a multiple of 4Kb.
 531          */
 532         sb_offset = calc_dev_sboffset(rdev->bdev);
 533         rdev->sb_offset = sb_offset;
 534
 535         ret = read_disk_sb(rdev, MD_SB_BYTES);
 536         if (ret) return ret;
 537
 538         ret = -EINVAL;
 539
 540         bdevname(rdev->bdev, b);
 541         sb = (mdp_super_t*)page_address(rdev->sb_page);
 542
 543         if (sb->md_magic != MD_SB_MAGIC) {
 544                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 545                        b);
 546                 goto abort;
 547         }
 548
 549         if (sb->major_version != 0 ||
 550             sb->minor_version != 90) {
 551                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 552                         sb->major_version, sb->minor_version,
 553                         b);
 554                 goto abort;
 555         }
 556
 557         if (sb->raid_disks <= 0)
 558                 goto abort;
 559
 560         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 561                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 562                         b);
 563                 goto abort;
 564         }
 565
 566         rdev->preferred_minor = sb->md_minor;
 567         rdev->data_offset = 0;
 568         rdev->sb_size = MD_SB_BYTES;
 569
 570         if (sb->level == LEVEL_MULTIPATH)
 571                 rdev->desc_nr = -1;
 572         else
 573                 rdev->desc_nr = sb->this_disk.number;
 574
 575         if (refdev == 0)
 576                 ret = 1;
 577         else {
 578                 __u64 ev1, ev2;
 579                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 580                 if (!uuid_equal(refsb, sb)) {
 581                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 582                                 b, bdevname(refdev->bdev,b2));
 583                         goto abort;
 584                 }
 585                 if (!sb_equal(refsb, sb)) {
 586                         printk(KERN_WARNING "md: %s has same UUID"
 587                                " but different superblock to %s\n",
 588                                b, bdevname(refdev->bdev, b2));
 589                         goto abort;
 590                 }
 591                 ev1 = md_event(sb);
 592                 ev2 = md_event(refsb);
 593                 if (ev1 > ev2)
 594                         ret = 1;
 595                 else
 596                         ret = 0;
 597         }
 598         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 599
 600  abort:
 601         return ret;
 602 }
 603
 604 /*
 605  * validate_super for 0.90.0
 606  */
 607 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 608 {
 609         mdp_disk_t *desc;
 610         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 611
 612         rdev->raid_disk = -1;
 613         rdev->in_sync = 0;
 614         if (mddev->raid_disks == 0) {
 615                 mddev->major_version = 0;
 616                 mddev->minor_version = sb->minor_version;
 617                 mddev->patch_version = sb->patch_version;
 618                 mddev->persistent = ! sb->not_persistent;
 619                 mddev->chunk_size = sb->chunk_size;
 620                 mddev->ctime = sb->ctime;
 621                 mddev->utime = sb->utime;
 622                 mddev->level = sb->level;
 623                 mddev->layout = sb->layout;
 624                 mddev->raid_disks = sb->raid_disks;
 625                 mddev->size = sb->size;
 626                 mddev->events = md_event(sb);
 627                 mddev->bitmap_offset = 0;
 628                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 629
 630                 if (sb->state & (1<<MD_SB_CLEAN))
 631                         mddev->recovery_cp = MaxSector;
 632                 else {
 633                         if (sb->events_hi == sb->cp_events_hi &&
 634                                 sb->events_lo == sb->cp_events_lo) {
 635                                 mddev->recovery_cp = sb->recovery_cp;
 636                         } else
 637                                 mddev->recovery_cp = 0;
 638                 }
 639
 640                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 641                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 642                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 643                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 644
 645                 mddev->max_disks = MD_SB_DISKS;
 646
 647                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 648                     mddev->bitmap_file == NULL) {
 649                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
 650                                 /* FIXME use a better test */
 651                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
 652                                 return -EINVAL;
 653                         }
 654                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 655                 }
 656
 657         } else if (mddev->pers == NULL) {
 658                 /* Insist on good event counter while assembling */
 659                 __u64 ev1 = md_event(sb);
 660                 ++ev1;
 661                 if (ev1 < mddev->events)
 662                         return -EINVAL;
 663         } else if (mddev->bitmap) {
 664                 /* if adding to array with a bitmap, then we can accept an
 665                  * older device ... but not too old.
 666                  */
 667                 __u64 ev1 = md_event(sb);
 668                 if (ev1 < mddev->bitmap->events_cleared)
 669                         return 0;
 670         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 671                 return 0;
 672
 673         if (mddev->level != LEVEL_MULTIPATH) {
 674                 rdev->faulty = 0;
 675                 rdev->flags = 0;
 676                 desc = sb->disks + rdev->desc_nr;
 677
 678                 if (desc->state & (1<<MD_DISK_FAULTY))
 679                         rdev->faulty = 1;
 680                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 681                          desc->raid_disk < mddev->raid_disks) {
 682                         rdev->in_sync = 1;
 683                         rdev->raid_disk = desc->raid_disk;
 684                 }
 685                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 686                         set_bit(WriteMostly, &rdev->flags);
 687         } else /* MULTIPATH are always insync */
 688                 rdev->in_sync = 1;
 689         return 0;
 690 }
 691
 692 /*
 693  * sync_super for 0.90.0
 694  */
 695 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 696 {
 697         mdp_super_t *sb;
 698         struct list_head *tmp;
 699         mdk_rdev_t *rdev2;
 700         int next_spare = mddev->raid_disks;
 701         char nm[20];
 702
 703         /* make rdev->sb match mddev data..
 704          *
 705          * 1/ zero out disks
 706          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 707          * 3/ any empty disks < next_spare become removed
 708          *
 709          * disks[0] gets initialised to REMOVED because
 710          * we cannot be sure from other fields if it has
 711          * been initialised or not.
 712          */
 713         int i;
 714         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 715         unsigned int fixdesc=0;
 716
 717         rdev->sb_size = MD_SB_BYTES;
 718
 719         sb = (mdp_super_t*)page_address(rdev->sb_page);
 720
 721         memset(sb, 0, sizeof(*sb));
 722
 723         sb->md_magic = MD_SB_MAGIC;
 724         sb->major_version = mddev->major_version;
 725         sb->minor_version = mddev->minor_version;
 726         sb->patch_version = mddev->patch_version;
 727         sb->gvalid_words  = 0; /* ignored */
 728         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 729         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 730         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 731         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 732
 733         sb->ctime = mddev->ctime;
 734         sb->level = mddev->level;
 735         sb->size  = mddev->size;
 736         sb->raid_disks = mddev->raid_disks;
 737         sb->md_minor = mddev->md_minor;
 738         sb->not_persistent = !mddev->persistent;
 739         sb->utime = mddev->utime;
 740         sb->state = 0;
 741         sb->events_hi = (mddev->events>>32);
 742         sb->events_lo = (u32)mddev->events;
 743
 744         if (mddev->in_sync)
 745         {
 746                 sb->recovery_cp = mddev->recovery_cp;
 747                 sb->cp_events_hi = (mddev->events>>32);
 748                 sb->cp_events_lo = (u32)mddev->events;
 749                 if (mddev->recovery_cp == MaxSector)
 750                         sb->state = (1<< MD_SB_CLEAN);
 751         } else
 752                 sb->recovery_cp = 0;
 753
 754         sb->layout = mddev->layout;
 755         sb->chunk_size = mddev->chunk_size;
 756
 757         if (mddev->bitmap && mddev->bitmap_file == NULL)
 758                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 759
 760         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 761         ITERATE_RDEV(mddev,rdev2,tmp) {
 762                 mdp_disk_t *d;
 763                 int desc_nr;
 764                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 765                         desc_nr = rdev2->raid_disk;
 766                 else
 767                         desc_nr = next_spare++;
 768                 if (desc_nr != rdev2->desc_nr) {
 769                         fixdesc |= (1 << desc_nr);
 770                         rdev2->desc_nr = desc_nr;
 771                         if (rdev2->raid_disk >= 0) {
 772                                 sprintf(nm, "rd%d", rdev2->raid_disk);
 773                                 sysfs_remove_link(&mddev->kobj, nm);
 774                         }
 775                         sysfs_remove_link(&rdev2->kobj, "block");
 776                         kobject_del(&rdev2->kobj);
 777                 }
 778                 d = &sb->disks[rdev2->desc_nr];
 779                 nr_disks++;
 780                 d->number = rdev2->desc_nr;
 781                 d->major = MAJOR(rdev2->bdev->bd_dev);
 782                 d->minor = MINOR(rdev2->bdev->bd_dev);
 783                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 784                         d->raid_disk = rdev2->raid_disk;
 785                 else
 786                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 787                 if (rdev2->faulty) {
 788                         d->state = (1<<MD_DISK_FAULTY);
 789                         failed++;
 790                 } else if (rdev2->in_sync) {
 791                         d->state = (1<<MD_DISK_ACTIVE);
 792                         d->state |= (1<<MD_DISK_SYNC);
 793                         active++;
 794                         working++;
 795                 } else {
 796                         d->state = 0;
 797                         spare++;
 798                         working++;
 799                 }
 800                 if (test_bit(WriteMostly, &rdev2->flags))
 801                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 802         }
 803         if (fixdesc)
 804                 ITERATE_RDEV(mddev,rdev2,tmp)
 805                         if (fixdesc & (1<<rdev2->desc_nr)) {
 806                                 snprintf(rdev2->kobj.name, KOBJ_NAME_LEN, "dev%d",
 807                                          rdev2->desc_nr);
 808                                 /* kobject_add gets a ref on the parent, so
 809                                  * we have to drop the one we already have
 810                                  */
 811                                 kobject_add(&rdev2->kobj);
 812                                 kobject_put(rdev->kobj.parent);
 813                                 sysfs_create_link(&rdev2->kobj,
 814                                                   &rdev2->bdev->bd_disk->kobj,
 815                                                   "block");
 816                                 if (rdev2->raid_disk >= 0) {
 817                                         sprintf(nm, "rd%d", rdev2->raid_disk);
 818                                         sysfs_create_link(&mddev->kobj,
 819                                                           &rdev2->kobj, nm);
 820                                 }
 821                         }
 822         /* now set the "removed" and "faulty" bits on any missing devices */
 823         for (i=0 ; i < mddev->raid_disks ; i++) {
 824                 mdp_disk_t *d = &sb->disks[i];
 825                 if (d->state == 0 && d->number == 0) {
 826                         d->number = i;
 827                         d->raid_disk = i;
 828                         d->state = (1<<MD_DISK_REMOVED);
 829                         d->state |= (1<<MD_DISK_FAULTY);
 830                         failed++;
 831                 }
 832         }
 833         sb->nr_disks = nr_disks;
 834         sb->active_disks = active;
 835         sb->working_disks = working;
 836         sb->failed_disks = failed;
 837         sb->spare_disks = spare;
 838
 839         sb->this_disk = sb->disks[rdev->desc_nr];
 840         sb->sb_csum = calc_sb_csum(sb);
 841 }
 842
 843 /*
 844  * version 1 superblock
 845  */
 846
 847 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 848 {
 849         unsigned int disk_csum, csum;
 850         unsigned long long newcsum;
 851         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 852         unsigned int *isuper = (unsigned int*)sb;
 853         int i;
 854
 855         disk_csum = sb->sb_csum;
 856         sb->sb_csum = 0;
 857         newcsum = 0;
 858         for (i=0; size>=4; size -= 4 )
 859                 newcsum += le32_to_cpu(*isuper++);
 860
 861         if (size == 2)
 862                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 863
 864         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 865         sb->sb_csum = disk_csum;
 866         return cpu_to_le32(csum);
 867 }
 868
 869 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 870 {
 871         struct mdp_superblock_1 *sb;
 872         int ret;
 873         sector_t sb_offset;
 874         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 875         int bmask;
 876
 877         /*
 878          * Calculate the position of the superblock.
 879          * It is always aligned to a 4K boundary and
 880          * depeding on minor_version, it can be:
 881          * 0: At least 8K, but less than 12K, from end of device
 882          * 1: At start of device
 883          * 2: 4K from start of device.
 884          */
 885         switch(minor_version) {
 886         case 0:
 887                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 888                 sb_offset -= 8*2;
 889                 sb_offset &= ~(sector_t)(4*2-1);
 890                 /* convert from sectors to K */
 891                 sb_offset /= 2;
 892                 break;
 893         case 1:
 894                 sb_offset = 0;
 895                 break;
 896         case 2:
 897                 sb_offset = 4;
 898                 break;
 899         default:
 900                 return -EINVAL;
 901         }
 902         rdev->sb_offset = sb_offset;
 903
 904         /* superblock is rarely larger than 1K, but it can be larger,
 905          * and it is safe to read 4k, so we do that
 906          */
 907         ret = read_disk_sb(rdev, 4096);
 908         if (ret) return ret;
 909
 910
 911         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 912
 913         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 914             sb->major_version != cpu_to_le32(1) ||
 915             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 916             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 917             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 918                 return -EINVAL;
 919
 920         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 921                 printk("md: invalid superblock checksum on %s\n",
 922                         bdevname(rdev->bdev,b));
 923                 return -EINVAL;
 924         }
 925         if (le64_to_cpu(sb->data_size) < 10) {
 926                 printk("md: data_size too small on %s\n",
 927                        bdevname(rdev->bdev,b));
 928                 return -EINVAL;
 929         }
 930         rdev->preferred_minor = 0xffff;
 931         rdev->data_offset = le64_to_cpu(sb->data_offset);
 932
 933         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 934         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 935         if (rdev->sb_size & bmask)
 936                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 937
 938         if (refdev == 0)
 939                 return 1;
 940         else {
 941                 __u64 ev1, ev2;
 942                 struct mdp_superblock_1 *refsb =
 943                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 944
 945                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 946                     sb->level != refsb->level ||
 947                     sb->layout != refsb->layout ||
 948                     sb->chunksize != refsb->chunksize) {
 949                         printk(KERN_WARNING "md: %s has strangely different"
 950                                 " superblock to %s\n",
 951                                 bdevname(rdev->bdev,b),
 952                                 bdevname(refdev->bdev,b2));
 953                         return -EINVAL;
 954                 }
 955                 ev1 = le64_to_cpu(sb->events);
 956                 ev2 = le64_to_cpu(refsb->events);
 957
 958                 if (ev1 > ev2)
 959                         return 1;
 960         }
 961         if (minor_version)
 962                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 963         else
 964                 rdev->size = rdev->sb_offset;
 965         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 966                 return -EINVAL;
 967         rdev->size = le64_to_cpu(sb->data_size)/2;
 968         if (le32_to_cpu(sb->chunksize))
 969                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 970         return 0;
 971 }
 972
 973 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 974 {
 975         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 976
 977         rdev->raid_disk = -1;
 978         rdev->in_sync = 0;
 979         if (mddev->raid_disks == 0) {
 980                 mddev->major_version = 1;
 981                 mddev->patch_version = 0;
 982                 mddev->persistent = 1;
 983                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 984                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 985                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 986                 mddev->level = le32_to_cpu(sb->level);
 987                 mddev->layout = le32_to_cpu(sb->layout);
 988                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 989                 mddev->size = le64_to_cpu(sb->size)/2;
 990                 mddev->events = le64_to_cpu(sb->events);
 991                 mddev->bitmap_offset = 0;
 992                 mddev->default_bitmap_offset = 0;
 993                 mddev->default_bitmap_offset = 1024;
 994
 995                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 996                 memcpy(mddev->uuid, sb->set_uuid, 16);
 997
 998                 mddev->max_disks =  (4096-256)/2;
 999
1000                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1001                     mddev->bitmap_file == NULL ) {
1002                         if (mddev->level != 1) {
1003                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1004                                 return -EINVAL;
1005                         }
1006                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1007                 }
1008         } else if (mddev->pers == NULL) {
1009                 /* Insist of good event counter while assembling */
1010                 __u64 ev1 = le64_to_cpu(sb->events);
1011                 ++ev1;
1012                 if (ev1 < mddev->events)
1013                         return -EINVAL;
1014         } else if (mddev->bitmap) {
1015                 /* If adding to array with a bitmap, then we can accept an
1016                  * older device, but not too old.
1017                  */
1018                 __u64 ev1 = le64_to_cpu(sb->events);
1019                 if (ev1 < mddev->bitmap->events_cleared)
1020                         return 0;
1021         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1022                 return 0;
1023
1024         if (mddev->level != LEVEL_MULTIPATH) {
1025                 int role;
1026                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1027                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1028                 switch(role) {
1029                 case 0xffff: /* spare */
1030                         rdev->faulty = 0;
1031                         break;
1032                 case 0xfffe: /* faulty */
1033                         rdev->faulty = 1;
1034                         break;
1035                 default:
1036                         rdev->in_sync = 1;
1037                         rdev->faulty = 0;
1038                         rdev->raid_disk = role;
1039                         break;
1040                 }
1041                 rdev->flags = 0;
1042                 if (sb->devflags & WriteMostly1)
1043                         set_bit(WriteMostly, &rdev->flags);
1044         } else /* MULTIPATH are always insync */
1045                 rdev->in_sync = 1;
1046
1047         return 0;
1048 }
1049
1050 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1051 {
1052         struct mdp_superblock_1 *sb;
1053         struct list_head *tmp;
1054         mdk_rdev_t *rdev2;
1055         int max_dev, i;
1056         /* make rdev->sb match mddev and rdev data. */
1057
1058         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1059
1060         sb->feature_map = 0;
1061         sb->pad0 = 0;
1062         memset(sb->pad1, 0, sizeof(sb->pad1));
1063         memset(sb->pad2, 0, sizeof(sb->pad2));
1064         memset(sb->pad3, 0, sizeof(sb->pad3));
1065
1066         sb->utime = cpu_to_le64((__u64)mddev->utime);
1067         sb->events = cpu_to_le64(mddev->events);
1068         if (mddev->in_sync)
1069                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1070         else
1071                 sb->resync_offset = cpu_to_le64(0);
1072
1073         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1074                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1075                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1076         }
1077
1078         max_dev = 0;
1079         ITERATE_RDEV(mddev,rdev2,tmp)
1080                 if (rdev2->desc_nr+1 > max_dev)
1081                         max_dev = rdev2->desc_nr+1;
1082
1083         sb->max_dev = cpu_to_le32(max_dev);
1084         for (i=0; i<max_dev;i++)
1085                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1086
1087         ITERATE_RDEV(mddev,rdev2,tmp) {
1088                 i = rdev2->desc_nr;
1089                 if (rdev2->faulty)
1090                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1091                 else if (rdev2->in_sync)
1092                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1093                 else
1094                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1095         }
1096
1097         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1098         sb->sb_csum = calc_sb_1_csum(sb);
1099 }
1100
1101
1102 static struct super_type super_types[] = {
1103         [0] = {
1104                 .name   = "0.90.0",
1105                 .owner  = THIS_MODULE,
1106                 .load_super     = super_90_load,
1107                 .validate_super = super_90_validate,
1108                 .sync_super     = super_90_sync,
1109         },
1110         [1] = {
1111                 .name   = "md-1",
1112                 .owner  = THIS_MODULE,
1113                 .load_super     = super_1_load,
1114                 .validate_super = super_1_validate,
1115                 .sync_super     = super_1_sync,
1116         },
1117 };
1118
1119 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1120 {
1121         struct list_head *tmp;
1122         mdk_rdev_t *rdev;
1123
1124         ITERATE_RDEV(mddev,rdev,tmp)
1125                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1126                         return rdev;
1127
1128         return NULL;
1129 }
1130
1131 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1132 {
1133         struct list_head *tmp;
1134         mdk_rdev_t *rdev;
1135
1136         ITERATE_RDEV(mddev1,rdev,tmp)
1137                 if (match_dev_unit(mddev2, rdev))
1138                         return 1;
1139
1140         return 0;
1141 }
1142
1143 static LIST_HEAD(pending_raid_disks);
1144
1145 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1146 {
1147         mdk_rdev_t *same_pdev;
1148         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1149
1150         if (rdev->mddev) {
1151                 MD_BUG();
1152                 return -EINVAL;
1153         }
1154         same_pdev = match_dev_unit(mddev, rdev);
1155         if (same_pdev)
1156                 printk(KERN_WARNING
1157                         "%s: WARNING: %s appears to be on the same physical"
1158                         " disk as %s. True\n     protection against single-disk"
1159                         " failure might be compromised.\n",
1160                         mdname(mddev), bdevname(rdev->bdev,b),
1161                         bdevname(same_pdev->bdev,b2));
1162
1163         /* Verify rdev->desc_nr is unique.
1164          * If it is -1, assign a free number, else
1165          * check number is not in use
1166          */
1167         if (rdev->desc_nr < 0) {
1168                 int choice = 0;
1169                 if (mddev->pers) choice = mddev->raid_disks;
1170                 while (find_rdev_nr(mddev, choice))
1171                         choice++;
1172                 rdev->desc_nr = choice;
1173         } else {
1174                 if (find_rdev_nr(mddev, rdev->desc_nr))
1175                         return -EBUSY;
1176         }
1177
1178         list_add(&rdev->same_set, &mddev->disks);
1179         rdev->mddev = mddev;
1180         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1181
1182         rdev->kobj.k_name = NULL;
1183         snprintf(rdev->kobj.name, KOBJ_NAME_LEN, "dev%d", rdev->desc_nr);
1184         rdev->kobj.parent = &mddev->kobj;
1185         kobject_add(&rdev->kobj);
1186
1187         sysfs_create_link(&rdev->kobj, &rdev->bdev->bd_disk->kobj, "block");
1188         return 0;
1189 }
1190
1191 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1192 {
1193         char b[BDEVNAME_SIZE];
1194         if (!rdev->mddev) {
1195                 MD_BUG();
1196                 return;
1197         }
1198         list_del_init(&rdev->same_set);
1199         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1200         rdev->mddev = NULL;
1201         sysfs_remove_link(&rdev->kobj, "block");
1202         kobject_del(&rdev->kobj);
1203 }
1204
1205 /*
1206  * prevent the device from being mounted, repartitioned or
1207  * otherwise reused by a RAID array (or any other kernel
1208  * subsystem), by bd_claiming the device.
1209  */
1210 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1211 {
1212         int err = 0;
1213         struct block_device *bdev;
1214         char b[BDEVNAME_SIZE];
1215
1216         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1217         if (IS_ERR(bdev)) {
1218                 printk(KERN_ERR "md: could not open %s.\n",
1219                         __bdevname(dev, b));
1220                 return PTR_ERR(bdev);
1221         }
1222         err = bd_claim(bdev, rdev);
1223         if (err) {
1224                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1225                         bdevname(bdev, b));
1226                 blkdev_put(bdev);
1227                 return err;
1228         }
1229         rdev->bdev = bdev;
1230         return err;
1231 }
1232
1233 static void unlock_rdev(mdk_rdev_t *rdev)
1234 {
1235         struct block_device *bdev = rdev->bdev;
1236         rdev->bdev = NULL;
1237         if (!bdev)
1238                 MD_BUG();
1239         bd_release(bdev);
1240         blkdev_put(bdev);
1241 }
1242
1243 void md_autodetect_dev(dev_t dev);
1244
1245 static void export_rdev(mdk_rdev_t * rdev)
1246 {
1247         char b[BDEVNAME_SIZE];
1248         printk(KERN_INFO "md: export_rdev(%s)\n",
1249                 bdevname(rdev->bdev,b));
1250         if (rdev->mddev)
1251                 MD_BUG();
1252         free_disk_sb(rdev);
1253         list_del_init(&rdev->same_set);
1254 #ifndef MODULE
1255         md_autodetect_dev(rdev->bdev->bd_dev);
1256 #endif
1257         unlock_rdev(rdev);
1258         kobject_put(&rdev->kobj);
1259 }
1260
1261 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1262 {
1263         unbind_rdev_from_array(rdev);
1264         export_rdev(rdev);
1265 }
1266
1267 static void export_array(mddev_t *mddev)
1268 {
1269         struct list_head *tmp;
1270         mdk_rdev_t *rdev;
1271
1272         ITERATE_RDEV(mddev,rdev,tmp) {
1273                 if (!rdev->mddev) {
1274                         MD_BUG();
1275                         continue;
1276                 }
1277                 kick_rdev_from_array(rdev);
1278         }
1279         if (!list_empty(&mddev->disks))
1280                 MD_BUG();
1281         mddev->raid_disks = 0;
1282         mddev->major_version = 0;
1283 }
1284
1285 static void print_desc(mdp_disk_t *desc)
1286 {
1287         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1288                 desc->major,desc->minor,desc->raid_disk,desc->state);
1289 }
1290
1291 static void print_sb(mdp_super_t *sb)
1292 {
1293         int i;
1294
1295         printk(KERN_INFO
1296                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1297                 sb->major_version, sb->minor_version, sb->patch_version,
1298                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1299                 sb->ctime);
1300         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1301                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1302                 sb->md_minor, sb->layout, sb->chunk_size);
1303         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1304                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1305                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1306                 sb->failed_disks, sb->spare_disks,
1307                 sb->sb_csum, (unsigned long)sb->events_lo);
1308
1309         printk(KERN_INFO);
1310         for (i = 0; i < MD_SB_DISKS; i++) {
1311                 mdp_disk_t *desc;
1312
1313                 desc = sb->disks + i;
1314                 if (desc->number || desc->major || desc->minor ||
1315                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1316                         printk("     D %2d: ", i);
1317                         print_desc(desc);
1318                 }
1319         }
1320         printk(KERN_INFO "md:     THIS: ");
1321         print_desc(&sb->this_disk);
1322
1323 }
1324
1325 static void print_rdev(mdk_rdev_t *rdev)
1326 {
1327         char b[BDEVNAME_SIZE];
1328         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1329                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1330                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1331         if (rdev->sb_loaded) {
1332                 printk(KERN_INFO "md: rdev superblock:\n");
1333                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1334         } else
1335                 printk(KERN_INFO "md: no rdev superblock!\n");
1336 }
1337
1338 void md_print_devices(void)
1339 {
1340         struct list_head *tmp, *tmp2;
1341         mdk_rdev_t *rdev;
1342         mddev_t *mddev;
1343         char b[BDEVNAME_SIZE];
1344
1345         printk("\n");
1346         printk("md:     **********************************\n");
1347         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1348         printk("md:     **********************************\n");
1349         ITERATE_MDDEV(mddev,tmp) {
1350
1351                 if (mddev->bitmap)
1352                         bitmap_print_sb(mddev->bitmap);
1353                 else
1354                         printk("%s: ", mdname(mddev));
1355                 ITERATE_RDEV(mddev,rdev,tmp2)
1356                         printk("<%s>", bdevname(rdev->bdev,b));
1357                 printk("\n");
1358
1359                 ITERATE_RDEV(mddev,rdev,tmp2)
1360                         print_rdev(rdev);
1361         }
1362         printk("md:     **********************************\n");
1363         printk("\n");
1364 }
1365
1366
1367 static void sync_sbs(mddev_t * mddev)
1368 {
1369         mdk_rdev_t *rdev;
1370         struct list_head *tmp;
1371
1372         ITERATE_RDEV(mddev,rdev,tmp) {
1373                 super_types[mddev->major_version].
1374                         sync_super(mddev, rdev);
1375                 rdev->sb_loaded = 1;
1376         }
1377 }
1378
1379 static void md_update_sb(mddev_t * mddev)
1380 {
1381         int err;
1382         struct list_head *tmp;
1383         mdk_rdev_t *rdev;
1384         int sync_req;
1385
1386 repeat:
1387         spin_lock(&mddev->write_lock);
1388         sync_req = mddev->in_sync;
1389         mddev->utime = get_seconds();
1390         mddev->events ++;
1391
1392         if (!mddev->events) {
1393                 /*
1394                  * oops, this 64-bit counter should never wrap.
1395                  * Either we are in around ~1 trillion A.C., assuming
1396                  * 1 reboot per second, or we have a bug:
1397                  */
1398                 MD_BUG();
1399                 mddev->events --;
1400         }
1401         mddev->sb_dirty = 2;
1402         sync_sbs(mddev);
1403
1404         /*
1405          * do not write anything to disk if using
1406          * nonpersistent superblocks
1407          */
1408         if (!mddev->persistent) {
1409                 mddev->sb_dirty = 0;
1410                 spin_unlock(&mddev->write_lock);
1411                 wake_up(&mddev->sb_wait);
1412                 return;
1413         }
1414         spin_unlock(&mddev->write_lock);
1415
1416         dprintk(KERN_INFO
1417                 "md: updating %s RAID superblock on device (in sync %d)\n",
1418                 mdname(mddev),mddev->in_sync);
1419
1420         err = bitmap_update_sb(mddev->bitmap);
1421         ITERATE_RDEV(mddev,rdev,tmp) {
1422                 char b[BDEVNAME_SIZE];
1423                 dprintk(KERN_INFO "md: ");
1424                 if (rdev->faulty)
1425                         dprintk("(skipping faulty ");
1426
1427                 dprintk("%s ", bdevname(rdev->bdev,b));
1428                 if (!rdev->faulty) {
1429                         md_super_write(mddev,rdev,
1430                                        rdev->sb_offset<<1, rdev->sb_size,
1431                                        rdev->sb_page);
1432                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1433                                 bdevname(rdev->bdev,b),
1434                                 (unsigned long long)rdev->sb_offset);
1435
1436                 } else
1437                         dprintk(")\n");
1438                 if (mddev->level == LEVEL_MULTIPATH)
1439                         /* only need to write one superblock... */
1440                         break;
1441         }
1442         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1443         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1444
1445         spin_lock(&mddev->write_lock);
1446         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1447                 /* have to write it out again */
1448                 spin_unlock(&mddev->write_lock);
1449                 goto repeat;
1450         }
1451         mddev->sb_dirty = 0;
1452         spin_unlock(&mddev->write_lock);
1453         wake_up(&mddev->sb_wait);
1454
1455 }
1456
1457 struct rdev_sysfs_entry {
1458         struct attribute attr;
1459         ssize_t (*show)(mdk_rdev_t *, char *);
1460         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1461 };
1462
1463 static ssize_t
1464 rdev_show_state(mdk_rdev_t *rdev, char *page)
1465 {
1466         char *sep = "";
1467         int len=0;
1468
1469         if (rdev->faulty) {
1470                 len+= sprintf(page+len, "%sfaulty",sep);
1471                 sep = ",";
1472         }
1473         if (rdev->in_sync) {
1474                 len += sprintf(page+len, "%sin_sync",sep);
1475                 sep = ",";
1476         }
1477         if (!rdev->faulty && !rdev->in_sync) {
1478                 len += sprintf(page+len, "%sspare", sep);
1479                 sep = ",";
1480         }
1481         return len+sprintf(page+len, "\n");
1482 }
1483
1484 static struct rdev_sysfs_entry rdev_state = {
1485         .attr = {.name = "state", .mode = S_IRUGO },
1486         .show = rdev_show_state,
1487 };
1488
1489 static ssize_t
1490 rdev_show_super(mdk_rdev_t *rdev, char *page)
1491 {
1492         if (rdev->sb_loaded && rdev->sb_size) {
1493                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1494                 return rdev->sb_size;
1495         } else
1496                 return 0;
1497 }
1498 static struct rdev_sysfs_entry rdev_super = {
1499         .attr = {.name = "super", .mode = S_IRUGO },
1500         .show = rdev_show_super,
1501 };
1502 static struct attribute *rdev_default_attrs[] = {
1503         &rdev_state.attr,
1504         &rdev_super.attr,
1505         NULL,
1506 };
1507 static ssize_t
1508 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1509 {
1510         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1511         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1512
1513         if (!entry->show)
1514                 return -EIO;
1515         return entry->show(rdev, page);
1516 }
1517
1518 static ssize_t
1519 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1520               const char *page, size_t length)
1521 {
1522         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1523         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1524
1525         if (!entry->store)
1526                 return -EIO;
1527         return entry->store(rdev, page, length);
1528 }
1529
1530 static void rdev_free(struct kobject *ko)
1531 {
1532         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1533         kfree(rdev);
1534 }
1535 static struct sysfs_ops rdev_sysfs_ops = {
1536         .show           = rdev_attr_show,
1537         .store          = rdev_attr_store,
1538 };
1539 static struct kobj_type rdev_ktype = {
1540         .release        = rdev_free,
1541         .sysfs_ops      = &rdev_sysfs_ops,
1542         .default_attrs  = rdev_default_attrs,
1543 };
1544
1545 /*
1546  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1547  *
1548  * mark the device faulty if:
1549  *
1550  *   - the device is nonexistent (zero size)
1551  *   - the device has no valid superblock
1552  *
1553  * a faulty rdev _never_ has rdev->sb set.
1554  */
1555 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1556 {
1557         char b[BDEVNAME_SIZE];
1558         int err;
1559         mdk_rdev_t *rdev;
1560         sector_t size;
1561
1562         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1563         if (!rdev) {
1564                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1565                 return ERR_PTR(-ENOMEM);
1566         }
1567         memset(rdev, 0, sizeof(*rdev));
1568
1569         if ((err = alloc_disk_sb(rdev)))
1570                 goto abort_free;
1571
1572         err = lock_rdev(rdev, newdev);
1573         if (err)
1574                 goto abort_free;
1575
1576         rdev->kobj.parent = NULL;
1577         rdev->kobj.ktype = &rdev_ktype;
1578         kobject_init(&rdev->kobj);
1579
1580         rdev->desc_nr = -1;
1581         rdev->faulty = 0;
1582         rdev->in_sync = 0;
1583         rdev->data_offset = 0;
1584         atomic_set(&rdev->nr_pending, 0);
1585         atomic_set(&rdev->read_errors, 0);
1586
1587         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1588         if (!size) {
1589                 printk(KERN_WARNING
1590                         "md: %s has zero or unknown size, marking faulty!\n",
1591                         bdevname(rdev->bdev,b));
1592                 err = -EINVAL;
1593                 goto abort_free;
1594         }
1595
1596         if (super_format >= 0) {
1597                 err = super_types[super_format].
1598                         load_super(rdev, NULL, super_minor);
1599                 if (err == -EINVAL) {
1600                         printk(KERN_WARNING
1601                                 "md: %s has invalid sb, not importing!\n",
1602                                 bdevname(rdev->bdev,b));
1603                         goto abort_free;
1604                 }
1605                 if (err < 0) {
1606                         printk(KERN_WARNING
1607                                 "md: could not read %s's sb, not importing!\n",
1608                                 bdevname(rdev->bdev,b));
1609                         goto abort_free;
1610                 }
1611         }
1612         INIT_LIST_HEAD(&rdev->same_set);
1613
1614         return rdev;
1615
1616 abort_free:
1617         if (rdev->sb_page) {
1618                 if (rdev->bdev)
1619                         unlock_rdev(rdev);
1620                 free_disk_sb(rdev);
1621         }
1622         kfree(rdev);
1623         return ERR_PTR(err);
1624 }
1625
1626 /*
1627  * Check a full RAID array for plausibility
1628  */
1629
1630
1631 static void analyze_sbs(mddev_t * mddev)
1632 {
1633         int i;
1634         struct list_head *tmp;
1635         mdk_rdev_t *rdev, *freshest;
1636         char b[BDEVNAME_SIZE];
1637
1638         freshest = NULL;
1639         ITERATE_RDEV(mddev,rdev,tmp)
1640                 switch (super_types[mddev->major_version].
1641                         load_super(rdev, freshest, mddev->minor_version)) {
1642                 case 1:
1643                         freshest = rdev;
1644                         break;
1645                 case 0:
1646                         break;
1647                 default:
1648                         printk( KERN_ERR \
1649                                 "md: fatal superblock inconsistency in %s"
1650                                 " -- removing from array\n",
1651                                 bdevname(rdev->bdev,b));
1652                         kick_rdev_from_array(rdev);
1653                 }
1654
1655
1656         super_types[mddev->major_version].
1657                 validate_super(mddev, freshest);
1658
1659         i = 0;
1660         ITERATE_RDEV(mddev,rdev,tmp) {
1661                 if (rdev != freshest)
1662                         if (super_types[mddev->major_version].
1663                             validate_super(mddev, rdev)) {
1664                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1665                                         " from array!\n",
1666                                         bdevname(rdev->bdev,b));
1667                                 kick_rdev_from_array(rdev);
1668                                 continue;
1669                         }
1670                 if (mddev->level == LEVEL_MULTIPATH) {
1671                         rdev->desc_nr = i++;
1672                         rdev->raid_disk = rdev->desc_nr;
1673                         rdev->in_sync = 1;
1674                 }
1675         }
1676
1677
1678
1679         if (mddev->recovery_cp != MaxSector &&
1680             mddev->level >= 1)
1681                 printk(KERN_ERR "md: %s: raid array is not clean"
1682                        " -- starting background reconstruction\n",
1683                        mdname(mddev));
1684
1685 }
1686
1687 static ssize_t
1688 md_show_level(mddev_t *mddev, char *page)
1689 {
1690         mdk_personality_t *p = mddev->pers;
1691         if (p == NULL)
1692                 return 0;
1693         if (mddev->level >= 0)
1694                 return sprintf(page, "RAID-%d\n", mddev->level);
1695         else
1696                 return sprintf(page, "%s\n", p->name);
1697 }
1698
1699 static struct md_sysfs_entry md_level = {
1700         .attr = {.name = "level", .mode = S_IRUGO },
1701         .show = md_show_level,
1702 };
1703
1704 static ssize_t
1705 md_show_rdisks(mddev_t *mddev, char *page)
1706 {
1707         return sprintf(page, "%d\n", mddev->raid_disks);
1708 }
1709
1710 static struct md_sysfs_entry md_raid_disks = {
1711         .attr = {.name = "raid_disks", .mode = S_IRUGO },
1712         .show = md_show_rdisks,
1713 };
1714
1715 static ssize_t
1716 md_show_scan(mddev_t *mddev, char *page)
1717 {
1718         char *type = "none";
1719         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1720             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1721                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1722                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1723                                 type = "resync";
1724                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1725                                 type = "check";
1726                         else
1727                                 type = "repair";
1728                 } else
1729                         type = "recover";
1730         }
1731         return sprintf(page, "%s\n", type);
1732 }
1733
1734 static ssize_t
1735 md_store_scan(mddev_t *mddev, const char *page, size_t len)
1736 {
1737         int canscan=0;
1738
1739         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1740             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1741                 return -EBUSY;
1742         down(&mddev->reconfig_sem);
1743         if (mddev->pers && mddev->pers->sync_request)
1744                 canscan=1;
1745         up(&mddev->reconfig_sem);
1746         if (!canscan)
1747                 return -EINVAL;
1748
1749         if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1750                 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1751         else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1752                 return -EINVAL;
1753         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1754         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1755         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1756         md_wakeup_thread(mddev->thread);
1757         return len;
1758 }
1759
1760 static ssize_t
1761 md_show_mismatch(mddev_t *mddev, char *page)
1762 {
1763         return sprintf(page, "%llu\n",
1764                        (unsigned long long) mddev->resync_mismatches);
1765 }
1766
1767 static struct md_sysfs_entry md_scan_mode = {
1768         .attr = {.name = "scan_mode", .mode = S_IRUGO|S_IWUSR },
1769         .show = md_show_scan,
1770         .store = md_store_scan,
1771 };
1772
1773 static struct md_sysfs_entry md_mismatches = {
1774         .attr = {.name = "mismatch_cnt", .mode = S_IRUGO },
1775         .show = md_show_mismatch,
1776 };
1777
1778 static struct attribute *md_default_attrs[] = {
1779         &md_level.attr,
1780         &md_raid_disks.attr,
1781         &md_scan_mode.attr,
1782         &md_mismatches.attr,
1783         NULL,
1784 };
1785
1786 static ssize_t
1787 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1788 {
1789         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1790         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1791
1792         if (!entry->show)
1793                 return -EIO;
1794         return entry->show(mddev, page);
1795 }
1796
1797 static ssize_t
1798 md_attr_store(struct kobject *kobj, struct attribute *attr,
1799               const char *page, size_t length)
1800 {
1801         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1802         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1803
1804         if (!entry->store)
1805                 return -EIO;
1806         return entry->store(mddev, page, length);
1807 }
1808
1809 static void md_free(struct kobject *ko)
1810 {
1811         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1812         kfree(mddev);
1813 }
1814
1815 static struct sysfs_ops md_sysfs_ops = {
1816         .show   = md_attr_show,
1817         .store  = md_attr_store,
1818 };
1819 static struct kobj_type md_ktype = {
1820         .release        = md_free,
1821         .sysfs_ops      = &md_sysfs_ops,
1822         .default_attrs  = md_default_attrs,
1823 };
1824
1825 int mdp_major = 0;
1826
1827 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1828 {
1829         static DECLARE_MUTEX(disks_sem);
1830         mddev_t *mddev = mddev_find(dev);
1831         struct gendisk *disk;
1832         int partitioned = (MAJOR(dev) != MD_MAJOR);
1833         int shift = partitioned ? MdpMinorShift : 0;
1834         int unit = MINOR(dev) >> shift;
1835
1836         if (!mddev)
1837                 return NULL;
1838
1839         down(&disks_sem);
1840         if (mddev->gendisk) {
1841                 up(&disks_sem);
1842                 mddev_put(mddev);
1843                 return NULL;
1844         }
1845         disk = alloc_disk(1 << shift);
1846         if (!disk) {
1847                 up(&disks_sem);
1848                 mddev_put(mddev);
1849                 return NULL;
1850         }
1851         disk->major = MAJOR(dev);
1852         disk->first_minor = unit << shift;
1853         if (partitioned) {
1854                 sprintf(disk->disk_name, "md_d%d", unit);
1855                 sprintf(disk->devfs_name, "md/d%d", unit);
1856         } else {
1857                 sprintf(disk->disk_name, "md%d", unit);
1858                 sprintf(disk->devfs_name, "md/%d", unit);
1859         }
1860         disk->fops = &md_fops;
1861         disk->private_data = mddev;
1862         disk->queue = mddev->queue;
1863         add_disk(disk);
1864         mddev->gendisk = disk;
1865         up(&disks_sem);
1866         mddev->kobj.parent = &disk->kobj;
1867         mddev->kobj.k_name = NULL;
1868         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1869         mddev->kobj.ktype = &md_ktype;
1870         kobject_register(&mddev->kobj);
1871         return NULL;
1872 }
1873
1874 void md_wakeup_thread(mdk_thread_t *thread);
1875
1876 static void md_safemode_timeout(unsigned long data)
1877 {
1878         mddev_t *mddev = (mddev_t *) data;
1879
1880         mddev->safemode = 1;
1881         md_wakeup_thread(mddev->thread);
1882 }
1883
1884
1885 static int do_md_run(mddev_t * mddev)
1886 {
1887         int pnum, err;
1888         int chunk_size;
1889         struct list_head *tmp;
1890         mdk_rdev_t *rdev;
1891         struct gendisk *disk;
1892         char b[BDEVNAME_SIZE];
1893
1894         if (list_empty(&mddev->disks))
1895                 /* cannot run an array with no devices.. */
1896                 return -EINVAL;
1897
1898         if (mddev->pers)
1899                 return -EBUSY;
1900
1901         /*
1902          * Analyze all RAID superblock(s)
1903          */
1904         if (!mddev->raid_disks)
1905                 analyze_sbs(mddev);
1906
1907         chunk_size = mddev->chunk_size;
1908         pnum = level_to_pers(mddev->level);
1909
1910         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1911                 if (!chunk_size) {
1912                         /*
1913                          * 'default chunksize' in the old md code used to
1914                          * be PAGE_SIZE, baaad.
1915                          * we abort here to be on the safe side. We don't
1916                          * want to continue the bad practice.
1917                          */
1918                         printk(KERN_ERR
1919                                 "no chunksize specified, see 'man raidtab'\n");
1920                         return -EINVAL;
1921                 }
1922                 if (chunk_size > MAX_CHUNK_SIZE) {
1923                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1924                                 chunk_size, MAX_CHUNK_SIZE);
1925                         return -EINVAL;
1926                 }
1927                 /*
1928                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1929                  */
1930                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1931                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1932                         return -EINVAL;
1933                 }
1934                 if (chunk_size < PAGE_SIZE) {
1935                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1936                                 chunk_size, PAGE_SIZE);
1937                         return -EINVAL;
1938                 }
1939
1940                 /* devices must have minimum size of one chunk */
1941                 ITERATE_RDEV(mddev,rdev,tmp) {
1942                         if (rdev->faulty)
1943                                 continue;
1944                         if (rdev->size < chunk_size / 1024) {
1945                                 printk(KERN_WARNING
1946                                         "md: Dev %s smaller than chunk_size:"
1947                                         " %lluk < %dk\n",
1948                                         bdevname(rdev->bdev,b),
1949                                         (unsigned long long)rdev->size,
1950                                         chunk_size / 1024);
1951                                 return -EINVAL;
1952                         }
1953                 }
1954         }
1955
1956 #ifdef CONFIG_KMOD
1957         if (!pers[pnum])
1958         {
1959                 request_module("md-personality-%d", pnum);
1960         }
1961 #endif
1962
1963         /*
1964          * Drop all container device buffers, from now on
1965          * the only valid external interface is through the md
1966          * device.
1967          * Also find largest hardsector size
1968          */
1969         ITERATE_RDEV(mddev,rdev,tmp) {
1970                 if (rdev->faulty)
1971                         continue;
1972                 sync_blockdev(rdev->bdev);
1973                 invalidate_bdev(rdev->bdev, 0);
1974         }
1975
1976         md_probe(mddev->unit, NULL, NULL);
1977         disk = mddev->gendisk;
1978         if (!disk)
1979                 return -ENOMEM;
1980
1981         spin_lock(&pers_lock);
1982         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1983                 spin_unlock(&pers_lock);
1984                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1985                        pnum);
1986                 return -EINVAL;
1987         }
1988
1989         mddev->pers = pers[pnum];
1990         spin_unlock(&pers_lock);
1991
1992         mddev->recovery = 0;
1993         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1994
1995         /* before we start the array running, initialise the bitmap */
1996         err = bitmap_create(mddev);
1997         if (err)
1998                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1999                         mdname(mddev), err);
2000         else
2001                 err = mddev->pers->run(mddev);
2002         if (err) {
2003                 printk(KERN_ERR "md: pers->run() failed ...\n");
2004                 module_put(mddev->pers->owner);
2005                 mddev->pers = NULL;
2006                 bitmap_destroy(mddev);
2007                 return err;
2008         }
2009         atomic_set(&mddev->writes_pending,0);
2010         mddev->safemode = 0;
2011         mddev->safemode_timer.function = md_safemode_timeout;
2012         mddev->safemode_timer.data = (unsigned long) mddev;
2013         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2014         mddev->in_sync = 1;
2015
2016         ITERATE_RDEV(mddev,rdev,tmp)
2017                 if (rdev->raid_disk >= 0) {
2018                         char nm[20];
2019                         sprintf(nm, "rd%d", rdev->raid_disk);
2020                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2021                 }
2022
2023         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2024         md_wakeup_thread(mddev->thread);
2025
2026         if (mddev->sb_dirty)
2027                 md_update_sb(mddev);
2028
2029         set_capacity(disk, mddev->array_size<<1);
2030
2031         /* If we call blk_queue_make_request here, it will
2032          * re-initialise max_sectors etc which may have been
2033          * refined inside -> run.  So just set the bits we need to set.
2034          * Most initialisation happended when we called
2035          * blk_queue_make_request(..., md_fail_request)
2036          * earlier.
2037          */
2038         mddev->queue->queuedata = mddev;
2039         mddev->queue->make_request_fn = mddev->pers->make_request;
2040
2041         mddev->changed = 1;
2042         return 0;
2043 }
2044
2045 static int restart_array(mddev_t *mddev)
2046 {
2047         struct gendisk *disk = mddev->gendisk;
2048         int err;
2049
2050         /*
2051          * Complain if it has no devices
2052          */
2053         err = -ENXIO;
2054         if (list_empty(&mddev->disks))
2055                 goto out;
2056
2057         if (mddev->pers) {
2058                 err = -EBUSY;
2059                 if (!mddev->ro)
2060                         goto out;
2061
2062                 mddev->safemode = 0;
2063                 mddev->ro = 0;
2064                 set_disk_ro(disk, 0);
2065
2066                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2067                         mdname(mddev));
2068                 /*
2069                  * Kick recovery or resync if necessary
2070                  */
2071                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2072                 md_wakeup_thread(mddev->thread);
2073                 err = 0;
2074         } else {
2075                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2076                         mdname(mddev));
2077                 err = -EINVAL;
2078         }
2079
2080 out:
2081         return err;
2082 }
2083
2084 static int do_md_stop(mddev_t * mddev, int ro)
2085 {
2086         int err = 0;
2087         struct gendisk *disk = mddev->gendisk;
2088
2089         if (mddev->pers) {
2090                 if (atomic_read(&mddev->active)>2) {
2091                         printk("md: %s still in use.\n",mdname(mddev));
2092                         return -EBUSY;
2093                 }
2094
2095                 if (mddev->sync_thread) {
2096                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2097                         md_unregister_thread(mddev->sync_thread);
2098                         mddev->sync_thread = NULL;
2099                 }
2100
2101                 del_timer_sync(&mddev->safemode_timer);
2102
2103                 invalidate_partition(disk, 0);
2104
2105                 if (ro) {
2106                         err  = -ENXIO;
2107                         if (mddev->ro)
2108                                 goto out;
2109                         mddev->ro = 1;
2110                 } else {
2111                         bitmap_flush(mddev);
2112                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
2113                         if (mddev->ro)
2114                                 set_disk_ro(disk, 0);
2115                         blk_queue_make_request(mddev->queue, md_fail_request);
2116                         mddev->pers->stop(mddev);
2117                         module_put(mddev->pers->owner);
2118                         mddev->pers = NULL;
2119                         if (mddev->ro)
2120                                 mddev->ro = 0;
2121                 }
2122                 if (!mddev->in_sync) {
2123                         /* mark array as shutdown cleanly */
2124                         mddev->in_sync = 1;
2125                         md_update_sb(mddev);
2126                 }
2127                 if (ro)
2128                         set_disk_ro(disk, 1);
2129         }
2130
2131         bitmap_destroy(mddev);
2132         if (mddev->bitmap_file) {
2133                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2134                 fput(mddev->bitmap_file);
2135                 mddev->bitmap_file = NULL;
2136         }
2137         mddev->bitmap_offset = 0;
2138
2139         /*
2140          * Free resources if final stop
2141          */
2142         if (!ro) {
2143                 mdk_rdev_t *rdev;
2144                 struct list_head *tmp;
2145                 struct gendisk *disk;
2146                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2147
2148                 ITERATE_RDEV(mddev,rdev,tmp)
2149                         if (rdev->raid_disk >= 0) {
2150                                 char nm[20];
2151                                 sprintf(nm, "rd%d", rdev->raid_disk);
2152                                 sysfs_remove_link(&mddev->kobj, nm);
2153                         }
2154
2155                 export_array(mddev);
2156
2157                 mddev->array_size = 0;
2158                 disk = mddev->gendisk;
2159                 if (disk)
2160                         set_capacity(disk, 0);
2161                 mddev->changed = 1;
2162         } else
2163                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2164                         mdname(mddev));
2165         err = 0;
2166 out:
2167         return err;
2168 }
2169
2170 static void autorun_array(mddev_t *mddev)
2171 {
2172         mdk_rdev_t *rdev;
2173         struct list_head *tmp;
2174         int err;
2175
2176         if (list_empty(&mddev->disks))
2177                 return;
2178
2179         printk(KERN_INFO "md: running: ");
2180
2181         ITERATE_RDEV(mddev,rdev,tmp) {
2182                 char b[BDEVNAME_SIZE];
2183                 printk("<%s>", bdevname(rdev->bdev,b));
2184         }
2185         printk("\n");
2186
2187         err = do_md_run (mddev);
2188         if (err) {
2189                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2190                 do_md_stop (mddev, 0);
2191         }
2192 }
2193
2194 /*
2195  * lets try to run arrays based on all disks that have arrived
2196  * until now. (those are in pending_raid_disks)
2197  *
2198  * the method: pick the first pending disk, collect all disks with
2199  * the same UUID, remove all from the pending list and put them into
2200  * the 'same_array' list. Then order this list based on superblock
2201  * update time (freshest comes first), kick out 'old' disks and
2202  * compare superblocks. If everything's fine then run it.
2203  *
2204  * If "unit" is allocated, then bump its reference count
2205  */
2206 static void autorun_devices(int part)
2207 {
2208         struct list_head candidates;
2209         struct list_head *tmp;
2210         mdk_rdev_t *rdev0, *rdev;
2211         mddev_t *mddev;
2212         char b[BDEVNAME_SIZE];
2213
2214         printk(KERN_INFO "md: autorun ...\n");
2215         while (!list_empty(&pending_raid_disks)) {
2216                 dev_t dev;
2217                 rdev0 = list_entry(pending_raid_disks.next,
2218                                          mdk_rdev_t, same_set);
2219
2220                 printk(KERN_INFO "md: considering %s ...\n",
2221                         bdevname(rdev0->bdev,b));
2222                 INIT_LIST_HEAD(&candidates);
2223                 ITERATE_RDEV_PENDING(rdev,tmp)
2224                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2225                                 printk(KERN_INFO "md:  adding %s ...\n",
2226                                         bdevname(rdev->bdev,b));
2227                                 list_move(&rdev->same_set, &candidates);
2228                         }
2229                 /*
2230                  * now we have a set of devices, with all of them having
2231                  * mostly sane superblocks. It's time to allocate the
2232                  * mddev.
2233                  */
2234                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2235                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2236                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2237                         break;
2238                 }
2239                 if (part)
2240                         dev = MKDEV(mdp_major,
2241                                     rdev0->preferred_minor << MdpMinorShift);
2242                 else
2243                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2244
2245                 md_probe(dev, NULL, NULL);
2246                 mddev = mddev_find(dev);
2247                 if (!mddev) {
2248                         printk(KERN_ERR
2249                                 "md: cannot allocate memory for md drive.\n");
2250                         break;
2251                 }
2252                 if (mddev_lock(mddev))
2253                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2254                                mdname(mddev));
2255                 else if (mddev->raid_disks || mddev->major_version
2256                          || !list_empty(&mddev->disks)) {
2257                         printk(KERN_WARNING
2258                                 "md: %s already running, cannot run %s\n",
2259                                 mdname(mddev), bdevname(rdev0->bdev,b));
2260                         mddev_unlock(mddev);
2261                 } else {
2262                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2263                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2264                                 list_del_init(&rdev->same_set);
2265                                 if (bind_rdev_to_array(rdev, mddev))
2266                                         export_rdev(rdev);
2267                         }
2268                         autorun_array(mddev);
2269                         mddev_unlock(mddev);
2270                 }
2271                 /* on success, candidates will be empty, on error
2272                  * it won't...
2273                  */
2274                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2275                         export_rdev(rdev);
2276                 mddev_put(mddev);
2277         }
2278         printk(KERN_INFO "md: ... autorun DONE.\n");
2279 }
2280
2281 /*
2282  * import RAID devices based on one partition
2283  * if possible, the array gets run as well.
2284  */
2285
2286 static int autostart_array(dev_t startdev)
2287 {
2288         char b[BDEVNAME_SIZE];
2289         int err = -EINVAL, i;
2290         mdp_super_t *sb = NULL;
2291         mdk_rdev_t *start_rdev = NULL, *rdev;
2292
2293         start_rdev = md_import_device(startdev, 0, 0);
2294         if (IS_ERR(start_rdev))
2295                 return err;
2296
2297
2298         /* NOTE: this can only work for 0.90.0 superblocks */
2299         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2300         if (sb->major_version != 0 ||
2301             sb->minor_version != 90 ) {
2302                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2303                 export_rdev(start_rdev);
2304                 return err;
2305         }
2306
2307         if (start_rdev->faulty) {
2308                 printk(KERN_WARNING
2309                         "md: can not autostart based on faulty %s!\n",
2310                         bdevname(start_rdev->bdev,b));
2311                 export_rdev(start_rdev);
2312                 return err;
2313         }
2314         list_add(&start_rdev->same_set, &pending_raid_disks);
2315
2316         for (i = 0; i < MD_SB_DISKS; i++) {
2317                 mdp_disk_t *desc = sb->disks + i;
2318                 dev_t dev = MKDEV(desc->major, desc->minor);
2319
2320                 if (!dev)
2321                         continue;
2322                 if (dev == startdev)
2323                         continue;
2324                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2325                         continue;
2326                 rdev = md_import_device(dev, 0, 0);
2327                 if (IS_ERR(rdev))
2328                         continue;
2329
2330                 list_add(&rdev->same_set, &pending_raid_disks);
2331         }
2332
2333         /*
2334          * possibly return codes
2335          */
2336         autorun_devices(0);
2337         return 0;
2338
2339 }
2340
2341
2342 static int get_version(void __user * arg)
2343 {
2344         mdu_version_t ver;
2345
2346         ver.major = MD_MAJOR_VERSION;
2347         ver.minor = MD_MINOR_VERSION;
2348         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2349
2350         if (copy_to_user(arg, &ver, sizeof(ver)))
2351                 return -EFAULT;
2352
2353         return 0;
2354 }
2355
2356 static int get_array_info(mddev_t * mddev, void __user * arg)
2357 {
2358         mdu_array_info_t info;
2359         int nr,working,active,failed,spare;
2360         mdk_rdev_t *rdev;
2361         struct list_head *tmp;
2362
2363         nr=working=active=failed=spare=0;
2364         ITERATE_RDEV(mddev,rdev,tmp) {
2365                 nr++;
2366                 if (rdev->faulty)
2367                         failed++;
2368                 else {
2369                         working++;
2370                         if (rdev->in_sync)
2371                                 active++;
2372                         else
2373                                 spare++;
2374                 }
2375         }
2376
2377         info.major_version = mddev->major_version;
2378         info.minor_version = mddev->minor_version;
2379         info.patch_version = MD_PATCHLEVEL_VERSION;
2380         info.ctime         = mddev->ctime;
2381         info.level         = mddev->level;
2382         info.size          = mddev->size;
2383         info.nr_disks      = nr;
2384         info.raid_disks    = mddev->raid_disks;
2385         info.md_minor      = mddev->md_minor;
2386         info.not_persistent= !mddev->persistent;
2387
2388         info.utime         = mddev->utime;
2389         info.state         = 0;
2390         if (mddev->in_sync)
2391                 info.state = (1<<MD_SB_CLEAN);
2392         if (mddev->bitmap && mddev->bitmap_offset)
2393                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2394         info.active_disks  = active;
2395         info.working_disks = working;
2396         info.failed_disks  = failed;
2397         info.spare_disks   = spare;
2398
2399         info.layout        = mddev->layout;
2400         info.chunk_size    = mddev->chunk_size;
2401
2402         if (copy_to_user(arg, &info, sizeof(info)))
2403                 return -EFAULT;
2404
2405         return 0;
2406 }
2407
2408 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2409 {
2410         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2411         char *ptr, *buf = NULL;
2412         int err = -ENOMEM;
2413
2414         file = kmalloc(sizeof(*file), GFP_KERNEL);
2415         if (!file)
2416                 goto out;
2417
2418         /* bitmap disabled, zero the first byte and copy out */
2419         if (!mddev->bitmap || !mddev->bitmap->file) {
2420                 file->pathname[0] = '\0';
2421                 goto copy_out;
2422         }
2423
2424         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2425         if (!buf)
2426                 goto out;
2427
2428         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2429         if (!ptr)
2430                 goto out;
2431
2432         strcpy(file->pathname, ptr);
2433
2434 copy_out:
2435         err = 0;
2436         if (copy_to_user(arg, file, sizeof(*file)))
2437                 err = -EFAULT;
2438 out:
2439         kfree(buf);
2440         kfree(file);
2441         return err;
2442 }
2443
2444 static int get_disk_info(mddev_t * mddev, void __user * arg)
2445 {
2446         mdu_disk_info_t info;
2447         unsigned int nr;
2448         mdk_rdev_t *rdev;
2449
2450         if (copy_from_user(&info, arg, sizeof(info)))
2451                 return -EFAULT;
2452
2453         nr = info.number;
2454
2455         rdev = find_rdev_nr(mddev, nr);
2456         if (rdev) {
2457                 info.major = MAJOR(rdev->bdev->bd_dev);
2458                 info.minor = MINOR(rdev->bdev->bd_dev);
2459                 info.raid_disk = rdev->raid_disk;
2460                 info.state = 0;
2461                 if (rdev->faulty)
2462                         info.state |= (1<<MD_DISK_FAULTY);
2463                 else if (rdev->in_sync) {
2464                         info.state |= (1<<MD_DISK_ACTIVE);
2465                         info.state |= (1<<MD_DISK_SYNC);
2466                 }
2467                 if (test_bit(WriteMostly, &rdev->flags))
2468                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2469         } else {
2470                 info.major = info.minor = 0;
2471                 info.raid_disk = -1;
2472                 info.state = (1<<MD_DISK_REMOVED);
2473         }
2474
2475         if (copy_to_user(arg, &info, sizeof(info)))
2476                 return -EFAULT;
2477
2478         return 0;
2479 }
2480
2481 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2482 {
2483         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2484         mdk_rdev_t *rdev;
2485         dev_t dev = MKDEV(info->major,info->minor);
2486
2487         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2488                 return -EOVERFLOW;
2489
2490         if (!mddev->raid_disks) {
2491                 int err;
2492                 /* expecting a device which has a superblock */
2493                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2494                 if (IS_ERR(rdev)) {
2495                         printk(KERN_WARNING
2496                                 "md: md_import_device returned %ld\n",
2497                                 PTR_ERR(rdev));
2498                         return PTR_ERR(rdev);
2499                 }
2500                 if (!list_empty(&mddev->disks)) {
2501                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2502                                                         mdk_rdev_t, same_set);
2503                         int err = super_types[mddev->major_version]
2504                                 .load_super(rdev, rdev0, mddev->minor_version);
2505                         if (err < 0) {
2506                                 printk(KERN_WARNING
2507                                         "md: %s has different UUID to %s\n",
2508                                         bdevname(rdev->bdev,b),
2509                                         bdevname(rdev0->bdev,b2));
2510                                 export_rdev(rdev);
2511                                 return -EINVAL;
2512                         }
2513                 }
2514                 err = bind_rdev_to_array(rdev, mddev);
2515                 if (err)
2516                         export_rdev(rdev);
2517                 return err;
2518         }
2519
2520         /*
2521          * add_new_disk can be used once the array is assembled
2522          * to add "hot spares".  They must already have a superblock
2523          * written
2524          */
2525         if (mddev->pers) {
2526                 int err;
2527                 if (!mddev->pers->hot_add_disk) {
2528                         printk(KERN_WARNING
2529                                 "%s: personality does not support diskops!\n",
2530                                mdname(mddev));
2531                         return -EINVAL;
2532                 }
2533                 if (mddev->persistent)
2534                         rdev = md_import_device(dev, mddev->major_version,
2535                                                 mddev->minor_version);
2536                 else
2537                         rdev = md_import_device(dev, -1, -1);
2538                 if (IS_ERR(rdev)) {
2539                         printk(KERN_WARNING
2540                                 "md: md_import_device returned %ld\n",
2541                                 PTR_ERR(rdev));
2542                         return PTR_ERR(rdev);
2543                 }
2544                 /* set save_raid_disk if appropriate */
2545                 if (!mddev->persistent) {
2546                         if (info->state & (1<<MD_DISK_SYNC)  &&
2547                             info->raid_disk < mddev->raid_disks)
2548                                 rdev->raid_disk = info->raid_disk;
2549                         else
2550                                 rdev->raid_disk = -1;
2551                 } else
2552                         super_types[mddev->major_version].
2553                                 validate_super(mddev, rdev);
2554                 rdev->saved_raid_disk = rdev->raid_disk;
2555
2556                 rdev->in_sync = 0; /* just to be sure */
2557                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2558                         set_bit(WriteMostly, &rdev->flags);
2559
2560                 rdev->raid_disk = -1;
2561                 err = bind_rdev_to_array(rdev, mddev);
2562                 if (err)
2563                         export_rdev(rdev);
2564
2565                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2566                 md_wakeup_thread(mddev->thread);
2567                 return err;
2568         }
2569
2570         /* otherwise, add_new_disk is only allowed
2571          * for major_version==0 superblocks
2572          */
2573         if (mddev->major_version != 0) {
2574                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2575                        mdname(mddev));
2576                 return -EINVAL;
2577         }
2578
2579         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2580                 int err;
2581                 rdev = md_import_device (dev, -1, 0);
2582                 if (IS_ERR(rdev)) {
2583                         printk(KERN_WARNING
2584                                 "md: error, md_import_device() returned %ld\n",
2585                                 PTR_ERR(rdev));
2586                         return PTR_ERR(rdev);
2587                 }
2588                 rdev->desc_nr = info->number;
2589                 if (info->raid_disk < mddev->raid_disks)
2590                         rdev->raid_disk = info->raid_disk;
2591                 else
2592                         rdev->raid_disk = -1;
2593
2594                 rdev->faulty = 0;
2595                 if (rdev->raid_disk < mddev->raid_disks)
2596                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2597                 else
2598                         rdev->in_sync = 0;
2599
2600                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2601                         set_bit(WriteMostly, &rdev->flags);
2602
2603                 err = bind_rdev_to_array(rdev, mddev);
2604                 if (err) {
2605                         export_rdev(rdev);
2606                         return err;
2607                 }
2608
2609                 if (!mddev->persistent) {
2610                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2611                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2612                 } else
2613                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2614                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2615
2616                 if (!mddev->size || (mddev->size > rdev->size))
2617                         mddev->size = rdev->size;
2618         }
2619
2620         return 0;
2621 }
2622
2623 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2624 {
2625         char b[BDEVNAME_SIZE];
2626         mdk_rdev_t *rdev;
2627
2628         if (!mddev->pers)
2629                 return -ENODEV;
2630
2631         rdev = find_rdev(mddev, dev);
2632         if (!rdev)
2633                 return -ENXIO;
2634
2635         if (rdev->raid_disk >= 0)
2636                 goto busy;
2637
2638         kick_rdev_from_array(rdev);
2639         md_update_sb(mddev);
2640
2641         return 0;
2642 busy:
2643         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2644                 bdevname(rdev->bdev,b), mdname(mddev));
2645         return -EBUSY;
2646 }
2647
2648 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2649 {
2650         char b[BDEVNAME_SIZE];
2651         int err;
2652         unsigned int size;
2653         mdk_rdev_t *rdev;
2654
2655         if (!mddev->pers)
2656                 return -ENODEV;
2657
2658         if (mddev->major_version != 0) {
2659                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2660                         " version-0 superblocks.\n",
2661                         mdname(mddev));
2662                 return -EINVAL;
2663         }
2664         if (!mddev->pers->hot_add_disk) {
2665                 printk(KERN_WARNING
2666                         "%s: personality does not support diskops!\n",
2667                         mdname(mddev));
2668                 return -EINVAL;
2669         }
2670
2671         rdev = md_import_device (dev, -1, 0);
2672         if (IS_ERR(rdev)) {
2673                 printk(KERN_WARNING
2674                         "md: error, md_import_device() returned %ld\n",
2675                         PTR_ERR(rdev));
2676                 return -EINVAL;
2677         }
2678
2679         if (mddev->persistent)
2680                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2681         else
2682                 rdev->sb_offset =
2683                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2684
2685         size = calc_dev_size(rdev, mddev->chunk_size);
2686         rdev->size = size;
2687
2688         if (size < mddev->size) {
2689                 printk(KERN_WARNING
2690                         "%s: disk size %llu blocks < array size %llu\n",
2691                         mdname(mddev), (unsigned long long)size,
2692                         (unsigned long long)mddev->size);
2693                 err = -ENOSPC;
2694                 goto abort_export;
2695         }
2696
2697         if (rdev->faulty) {
2698                 printk(KERN_WARNING
2699                         "md: can not hot-add faulty %s disk to %s!\n",
2700                         bdevname(rdev->bdev,b), mdname(mddev));
2701                 err = -EINVAL;
2702                 goto abort_export;
2703         }
2704         rdev->in_sync = 0;
2705         rdev->desc_nr = -1;
2706         bind_rdev_to_array(rdev, mddev);
2707
2708         /*
2709          * The rest should better be atomic, we can have disk failures
2710          * noticed in interrupt contexts ...
2711          */
2712
2713         if (rdev->desc_nr == mddev->max_disks) {
2714                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2715                         mdname(mddev));
2716                 err = -EBUSY;
2717                 goto abort_unbind_export;
2718         }
2719
2720         rdev->raid_disk = -1;
2721
2722         md_update_sb(mddev);
2723
2724         /*
2725          * Kick recovery, maybe this spare has to be added to the
2726          * array immediately.
2727          */
2728         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2729         md_wakeup_thread(mddev->thread);
2730
2731         return 0;
2732
2733 abort_unbind_export:
2734         unbind_rdev_from_array(rdev);
2735
2736 abort_export:
2737         export_rdev(rdev);
2738         return err;
2739 }
2740
2741 /* similar to deny_write_access, but accounts for our holding a reference
2742  * to the file ourselves */
2743 static int deny_bitmap_write_access(struct file * file)
2744 {
2745         struct inode *inode = file->f_mapping->host;
2746
2747         spin_lock(&inode->i_lock);
2748         if (atomic_read(&inode->i_writecount) > 1) {
2749                 spin_unlock(&inode->i_lock);
2750                 return -ETXTBSY;
2751         }
2752         atomic_set(&inode->i_writecount, -1);
2753         spin_unlock(&inode->i_lock);
2754
2755         return 0;
2756 }
2757
2758 static int set_bitmap_file(mddev_t *mddev, int fd)
2759 {
2760         int err;
2761
2762         if (mddev->pers) {
2763                 if (!mddev->pers->quiesce)
2764                         return -EBUSY;
2765                 if (mddev->recovery || mddev->sync_thread)
2766                         return -EBUSY;
2767                 /* we should be able to change the bitmap.. */
2768         }
2769
2770
2771         if (fd >= 0) {
2772                 if (mddev->bitmap)
2773                         return -EEXIST; /* cannot add when bitmap is present */
2774                 mddev->bitmap_file = fget(fd);
2775
2776                 if (mddev->bitmap_file == NULL) {
2777                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2778                                mdname(mddev));
2779                         return -EBADF;
2780                 }
2781
2782                 err = deny_bitmap_write_access(mddev->bitmap_file);
2783                 if (err) {
2784                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2785                                mdname(mddev));
2786                         fput(mddev->bitmap_file);
2787                         mddev->bitmap_file = NULL;
2788                         return err;
2789                 }
2790                 mddev->bitmap_offset = 0; /* file overrides offset */
2791         } else if (mddev->bitmap == NULL)
2792                 return -ENOENT; /* cannot remove what isn't there */
2793         err = 0;
2794         if (mddev->pers) {
2795                 mddev->pers->quiesce(mddev, 1);
2796                 if (fd >= 0)
2797                         err = bitmap_create(mddev);
2798                 if (fd < 0 || err)
2799                         bitmap_destroy(mddev);
2800                 mddev->pers->quiesce(mddev, 0);
2801         } else if (fd < 0) {
2802                 if (mddev->bitmap_file)
2803                         fput(mddev->bitmap_file);
2804                 mddev->bitmap_file = NULL;
2805         }
2806
2807         return err;
2808 }
2809
2810 /*
2811  * set_array_info is used two different ways
2812  * The original usage is when creating a new array.
2813  * In this usage, raid_disks is > 0 and it together with
2814  *  level, size, not_persistent,layout,chunksize determine the
2815  *  shape of the array.
2816  *  This will always create an array with a type-0.90.0 superblock.
2817  * The newer usage is when assembling an array.
2818  *  In this case raid_disks will be 0, and the major_version field is
2819  *  use to determine which style super-blocks are to be found on the devices.
2820  *  The minor and patch _version numbers are also kept incase the
2821  *  super_block handler wishes to interpret them.
2822  */
2823 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2824 {
2825
2826         if (info->raid_disks == 0) {
2827                 /* just setting version number for superblock loading */
2828                 if (info->major_version < 0 ||
2829                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2830                     super_types[info->major_version].name == NULL) {
2831                         /* maybe try to auto-load a module? */
2832                         printk(KERN_INFO
2833                                 "md: superblock version %d not known\n",
2834                                 info->major_version);
2835                         return -EINVAL;
2836                 }
2837                 mddev->major_version = info->major_version;
2838                 mddev->minor_version = info->minor_version;
2839                 mddev->patch_version = info->patch_version;
2840                 return 0;
2841         }
2842         mddev->major_version = MD_MAJOR_VERSION;
2843         mddev->minor_version = MD_MINOR_VERSION;
2844         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2845         mddev->ctime         = get_seconds();
2846
2847         mddev->level         = info->level;
2848         mddev->size          = info->size;
2849         mddev->raid_disks    = info->raid_disks;
2850         /* don't set md_minor, it is determined by which /dev/md* was
2851          * openned
2852          */
2853         if (info->state & (1<<MD_SB_CLEAN))
2854                 mddev->recovery_cp = MaxSector;
2855         else
2856                 mddev->recovery_cp = 0;
2857         mddev->persistent    = ! info->not_persistent;
2858
2859         mddev->layout        = info->layout;
2860         mddev->chunk_size    = info->chunk_size;
2861
2862         mddev->max_disks     = MD_SB_DISKS;
2863
2864         mddev->sb_dirty      = 1;
2865
2866         /*
2867          * Generate a 128 bit UUID
2868          */
2869         get_random_bytes(mddev->uuid, 16);
2870
2871         return 0;
2872 }
2873
2874 /*
2875  * update_array_info is used to change the configuration of an
2876  * on-line array.
2877  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2878  * fields in the info are checked against the array.
2879  * Any differences that cannot be handled will cause an error.
2880  * Normally, only one change can be managed at a time.
2881  */
2882 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2883 {
2884         int rv = 0;
2885         int cnt = 0;
2886         int state = 0;
2887
2888         /* calculate expected state,ignoring low bits */
2889         if (mddev->bitmap && mddev->bitmap_offset)
2890                 state |= (1 << MD_SB_BITMAP_PRESENT);
2891
2892         if (mddev->major_version != info->major_version ||
2893             mddev->minor_version != info->minor_version ||
2894 /*          mddev->patch_version != info->patch_version || */
2895             mddev->ctime         != info->ctime         ||
2896             mddev->level         != info->level         ||
2897 /*          mddev->layout        != info->layout        || */
2898             !mddev->persistent   != info->not_persistent||
2899             mddev->chunk_size    != info->chunk_size    ||
2900             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2901             ((state^info->state) & 0xfffffe00)
2902                 )
2903                 return -EINVAL;
2904         /* Check there is only one change */
2905         if (mddev->size != info->size) cnt++;
2906         if (mddev->raid_disks != info->raid_disks) cnt++;
2907         if (mddev->layout != info->layout) cnt++;
2908         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2909         if (cnt == 0) return 0;
2910         if (cnt > 1) return -EINVAL;
2911
2912         if (mddev->layout != info->layout) {
2913                 /* Change layout
2914                  * we don't need to do anything at the md level, the
2915                  * personality will take care of it all.
2916                  */
2917                 if (mddev->pers->reconfig == NULL)
2918                         return -EINVAL;
2919                 else
2920                         return mddev->pers->reconfig(mddev, info->layout, -1);
2921         }
2922         if (mddev->size != info->size) {
2923                 mdk_rdev_t * rdev;
2924                 struct list_head *tmp;
2925                 if (mddev->pers->resize == NULL)
2926                         return -EINVAL;
2927                 /* The "size" is the amount of each device that is used.
2928                  * This can only make sense for arrays with redundancy.
2929                  * linear and raid0 always use whatever space is available
2930                  * We can only consider changing the size if no resync
2931                  * or reconstruction is happening, and if the new size
2932                  * is acceptable. It must fit before the sb_offset or,
2933                  * if that is <data_offset, it must fit before the
2934                  * size of each device.
2935                  * If size is zero, we find the largest size that fits.
2936                  */
2937                 if (mddev->sync_thread)
2938                         return -EBUSY;
2939                 ITERATE_RDEV(mddev,rdev,tmp) {
2940                         sector_t avail;
2941                         int fit = (info->size == 0);
2942                         if (rdev->sb_offset > rdev->data_offset)
2943                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2944                         else
2945                                 avail = get_capacity(rdev->bdev->bd_disk)
2946                                         - rdev->data_offset;
2947                         if (fit && (info->size == 0 || info->size > avail/2))
2948                                 info->size = avail/2;
2949                         if (avail < ((sector_t)info->size << 1))
2950                                 return -ENOSPC;
2951                 }
2952                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2953                 if (!rv) {
2954                         struct block_device *bdev;
2955
2956                         bdev = bdget_disk(mddev->gendisk, 0);
2957                         if (bdev) {
2958                                 down(&bdev->bd_inode->i_sem);
2959                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2960                                 up(&bdev->bd_inode->i_sem);
2961                                 bdput(bdev);
2962                         }
2963                 }
2964         }
2965         if (mddev->raid_disks    != info->raid_disks) {
2966                 /* change the number of raid disks */
2967                 if (mddev->pers->reshape == NULL)
2968                         return -EINVAL;
2969                 if (info->raid_disks <= 0 ||
2970                     info->raid_disks >= mddev->max_disks)
2971                         return -EINVAL;
2972                 if (mddev->sync_thread)
2973                         return -EBUSY;
2974                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2975                 if (!rv) {
2976                         struct block_device *bdev;
2977
2978                         bdev = bdget_disk(mddev->gendisk, 0);
2979                         if (bdev) {
2980                                 down(&bdev->bd_inode->i_sem);
2981                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2982                                 up(&bdev->bd_inode->i_sem);
2983                                 bdput(bdev);
2984                         }
2985                 }
2986         }
2987         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2988                 if (mddev->pers->quiesce == NULL)
2989                         return -EINVAL;
2990                 if (mddev->recovery || mddev->sync_thread)
2991                         return -EBUSY;
2992                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2993                         /* add the bitmap */
2994                         if (mddev->bitmap)
2995                                 return -EEXIST;
2996                         if (mddev->default_bitmap_offset == 0)
2997                                 return -EINVAL;
2998                         mddev->bitmap_offset = mddev->default_bitmap_offset;
2999                         mddev->pers->quiesce(mddev, 1);
3000                         rv = bitmap_create(mddev);
3001                         if (rv)
3002                                 bitmap_destroy(mddev);
3003                         mddev->pers->quiesce(mddev, 0);
3004                 } else {
3005                         /* remove the bitmap */
3006                         if (!mddev->bitmap)
3007                                 return -ENOENT;
3008                         if (mddev->bitmap->file)
3009                                 return -EINVAL;
3010                         mddev->pers->quiesce(mddev, 1);
3011                         bitmap_destroy(mddev);
3012                         mddev->pers->quiesce(mddev, 0);
3013                         mddev->bitmap_offset = 0;
3014                 }
3015         }
3016         md_update_sb(mddev);
3017         return rv;
3018 }
3019
3020 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3021 {
3022         mdk_rdev_t *rdev;
3023
3024         if (mddev->pers == NULL)
3025                 return -ENODEV;
3026
3027         rdev = find_rdev(mddev, dev);
3028         if (!rdev)
3029                 return -ENODEV;
3030
3031         md_error(mddev, rdev);
3032         return 0;
3033 }
3034
3035 static int md_ioctl(struct inode *inode, struct file *file,
3036                         unsigned int cmd, unsigned long arg)
3037 {
3038         int err = 0;
3039         void __user *argp = (void __user *)arg;
3040         struct hd_geometry __user *loc = argp;
3041         mddev_t *mddev = NULL;
3042
3043         if (!capable(CAP_SYS_ADMIN))
3044                 return -EACCES;
3045
3046         /*
3047          * Commands dealing with the RAID driver but not any
3048          * particular array:
3049          */
3050         switch (cmd)
3051         {
3052                 case RAID_VERSION:
3053                         err = get_version(argp);
3054                         goto done;
3055
3056                 case PRINT_RAID_DEBUG:
3057                         err = 0;
3058                         md_print_devices();
3059                         goto done;
3060
3061 #ifndef MODULE
3062                 case RAID_AUTORUN:
3063                         err = 0;
3064                         autostart_arrays(arg);
3065                         goto done;
3066 #endif
3067                 default:;
3068         }
3069
3070         /*
3071          * Commands creating/starting a new array:
3072          */
3073
3074         mddev = inode->i_bdev->bd_disk->private_data;
3075
3076         if (!mddev) {
3077                 BUG();
3078                 goto abort;
3079         }
3080
3081
3082         if (cmd == START_ARRAY) {
3083                 /* START_ARRAY doesn't need to lock the array as autostart_array
3084                  * does the locking, and it could even be a different array
3085                  */
3086                 static int cnt = 3;
3087                 if (cnt > 0 ) {
3088                         printk(KERN_WARNING
3089                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3090                                "This will not be supported beyond 2.6\n",
3091                                current->comm, current->pid);
3092                         cnt--;
3093                 }
3094                 err = autostart_array(new_decode_dev(arg));
3095                 if (err) {
3096                         printk(KERN_WARNING "md: autostart failed!\n");
3097                         goto abort;
3098                 }
3099                 goto done;
3100         }
3101
3102         err = mddev_lock(mddev);
3103         if (err) {
3104                 printk(KERN_INFO
3105                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3106                         err, cmd);
3107                 goto abort;
3108         }
3109
3110         switch (cmd)
3111         {
3112                 case SET_ARRAY_INFO:
3113                         {
3114                                 mdu_array_info_t info;
3115                                 if (!arg)
3116                                         memset(&info, 0, sizeof(info));
3117                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3118                                         err = -EFAULT;
3119                                         goto abort_unlock;
3120                                 }
3121                                 if (mddev->pers) {
3122                                         err = update_array_info(mddev, &info);
3123                                         if (err) {
3124                                                 printk(KERN_WARNING "md: couldn't update"
3125                                                        " array info. %d\n", err);
3126                                                 goto abort_unlock;
3127                                         }
3128                                         goto done_unlock;
3129                                 }
3130                                 if (!list_empty(&mddev->disks)) {
3131                                         printk(KERN_WARNING
3132                                                "md: array %s already has disks!\n",
3133                                                mdname(mddev));
3134                                         err = -EBUSY;
3135                                         goto abort_unlock;
3136                                 }
3137                                 if (mddev->raid_disks) {
3138                                         printk(KERN_WARNING
3139                                                "md: array %s already initialised!\n",
3140                                                mdname(mddev));
3141                                         err = -EBUSY;
3142                                         goto abort_unlock;
3143                                 }
3144                                 err = set_array_info(mddev, &info);
3145                                 if (err) {
3146                                         printk(KERN_WARNING "md: couldn't set"
3147                                                " array info. %d\n", err);
3148                                         goto abort_unlock;
3149                                 }
3150                         }
3151                         goto done_unlock;
3152
3153                 default:;
3154         }
3155
3156         /*
3157          * Commands querying/configuring an existing array:
3158          */
3159         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3160          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3161         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3162                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3163                 err = -ENODEV;
3164                 goto abort_unlock;
3165         }
3166
3167         /*
3168          * Commands even a read-only array can execute:
3169          */
3170         switch (cmd)
3171         {
3172                 case GET_ARRAY_INFO:
3173                         err = get_array_info(mddev, argp);
3174                         goto done_unlock;
3175
3176                 case GET_BITMAP_FILE:
3177                         err = get_bitmap_file(mddev, argp);
3178                         goto done_unlock;
3179
3180                 case GET_DISK_INFO:
3181                         err = get_disk_info(mddev, argp);
3182                         goto done_unlock;
3183
3184                 case RESTART_ARRAY_RW:
3185                         err = restart_array(mddev);
3186                         goto done_unlock;
3187
3188                 case STOP_ARRAY:
3189                         err = do_md_stop (mddev, 0);
3190                         goto done_unlock;
3191
3192                 case STOP_ARRAY_RO:
3193                         err = do_md_stop (mddev, 1);
3194                         goto done_unlock;
3195
3196         /*
3197          * We have a problem here : there is no easy way to give a CHS
3198          * virtual geometry. We currently pretend that we have a 2 heads
3199          * 4 sectors (with a BIG number of cylinders...). This drives
3200          * dosfs just mad... ;-)
3201          */
3202                 case HDIO_GETGEO:
3203                         if (!loc) {
3204                                 err = -EINVAL;
3205                                 goto abort_unlock;
3206                         }
3207                         err = put_user (2, (char __user *) &loc->heads);
3208                         if (err)
3209                                 goto abort_unlock;
3210                         err = put_user (4, (char __user *) &loc->sectors);
3211                         if (err)
3212                                 goto abort_unlock;
3213                         err = put_user(get_capacity(mddev->gendisk)/8,
3214                                         (short __user *) &loc->cylinders);
3215                         if (err)
3216                                 goto abort_unlock;
3217                         err = put_user (get_start_sect(inode->i_bdev),
3218                                                 (long __user *) &loc->start);
3219                         goto done_unlock;
3220         }
3221
3222         /*
3223          * The remaining ioctls are changing the state of the
3224          * superblock, so we do not allow read-only arrays
3225          * here:
3226          */
3227         if (mddev->ro) {
3228                 err = -EROFS;
3229                 goto abort_unlock;
3230         }
3231
3232         switch (cmd)
3233         {
3234                 case ADD_NEW_DISK:
3235                 {
3236                         mdu_disk_info_t info;
3237                         if (copy_from_user(&info, argp, sizeof(info)))
3238                                 err = -EFAULT;
3239                         else
3240                                 err = add_new_disk(mddev, &info);
3241                         goto done_unlock;
3242                 }
3243
3244                 case HOT_REMOVE_DISK:
3245                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3246                         goto done_unlock;
3247
3248                 case HOT_ADD_DISK:
3249                         err = hot_add_disk(mddev, new_decode_dev(arg));
3250                         goto done_unlock;
3251
3252                 case SET_DISK_FAULTY:
3253                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3254                         goto done_unlock;
3255
3256                 case RUN_ARRAY:
3257                         err = do_md_run (mddev);
3258                         goto done_unlock;
3259
3260                 case SET_BITMAP_FILE:
3261                         err = set_bitmap_file(mddev, (int)arg);
3262                         goto done_unlock;
3263
3264                 default:
3265                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3266                                 printk(KERN_WARNING "md: %s(pid %d) used"
3267                                         " obsolete MD ioctl, upgrade your"
3268                                         " software to use new ictls.\n",
3269                                         current->comm, current->pid);
3270                         err = -EINVAL;
3271                         goto abort_unlock;
3272         }
3273
3274 done_unlock:
3275 abort_unlock:
3276         mddev_unlock(mddev);
3277
3278         return err;
3279 done:
3280         if (err)
3281                 MD_BUG();
3282 abort:
3283         return err;
3284 }
3285
3286 static int md_open(struct inode *inode, struct file *file)
3287 {
3288         /*
3289          * Succeed if we can lock the mddev, which confirms that
3290          * it isn't being stopped right now.
3291          */
3292         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3293         int err;
3294
3295         if ((err = mddev_lock(mddev)))
3296                 goto out;
3297
3298         err = 0;
3299         mddev_get(mddev);
3300         mddev_unlock(mddev);
3301
3302         check_disk_change(inode->i_bdev);
3303  out:
3304         return err;
3305 }
3306
3307 static int md_release(struct inode *inode, struct file * file)
3308 {
3309         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3310
3311         if (!mddev)
3312                 BUG();
3313         mddev_put(mddev);
3314
3315         return 0;
3316 }
3317
3318 static int md_media_changed(struct gendisk *disk)
3319 {
3320         mddev_t *mddev = disk->private_data;
3321
3322         return mddev->changed;
3323 }
3324
3325 static int md_revalidate(struct gendisk *disk)
3326 {
3327         mddev_t *mddev = disk->private_data;
3328
3329         mddev->changed = 0;
3330         return 0;
3331 }
3332 static struct block_device_operations md_fops =
3333 {
3334         .owner          = THIS_MODULE,
3335         .open           = md_open,
3336         .release        = md_release,
3337         .ioctl          = md_ioctl,
3338         .media_changed  = md_media_changed,
3339         .revalidate_disk= md_revalidate,
3340 };
3341
3342 static int md_thread(void * arg)
3343 {
3344         mdk_thread_t *thread = arg;
3345
3346         /*
3347          * md_thread is a 'system-thread', it's priority should be very
3348          * high. We avoid resource deadlocks individually in each
3349          * raid personality. (RAID5 does preallocation) We also use RR and
3350          * the very same RT priority as kswapd, thus we will never get
3351          * into a priority inversion deadlock.
3352          *
3353          * we definitely have to have equal or higher priority than
3354          * bdflush, otherwise bdflush will deadlock if there are too
3355          * many dirty RAID5 blocks.
3356          */
3357
3358         allow_signal(SIGKILL);
3359         complete(thread->event);
3360         while (!kthread_should_stop()) {
3361                 void (*run)(mddev_t *);
3362
3363                 wait_event_interruptible_timeout(thread->wqueue,
3364                                                  test_bit(THREAD_WAKEUP, &thread->flags)
3365                                                  || kthread_should_stop(),
3366                                                  thread->timeout);
3367                 try_to_freeze();
3368
3369                 clear_bit(THREAD_WAKEUP, &thread->flags);
3370
3371                 run = thread->run;
3372                 if (run)
3373                         run(thread->mddev);
3374         }
3375
3376         return 0;
3377 }
3378
3379 void md_wakeup_thread(mdk_thread_t *thread)
3380 {
3381         if (thread) {
3382                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3383                 set_bit(THREAD_WAKEUP, &thread->flags);
3384                 wake_up(&thread->wqueue);
3385         }
3386 }
3387
3388 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3389                                  const char *name)
3390 {
3391         mdk_thread_t *thread;
3392         struct completion event;
3393
3394         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3395         if (!thread)
3396                 return NULL;
3397
3398         memset(thread, 0, sizeof(mdk_thread_t));
3399         init_waitqueue_head(&thread->wqueue);
3400
3401         init_completion(&event);
3402         thread->event = &event;
3403         thread->run = run;
3404         thread->mddev = mddev;
3405         thread->name = name;
3406         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3407         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3408         if (IS_ERR(thread->tsk)) {
3409                 kfree(thread);
3410                 return NULL;
3411         }
3412         wait_for_completion(&event);
3413         return thread;
3414 }
3415
3416 void md_unregister_thread(mdk_thread_t *thread)
3417 {
3418         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3419
3420         kthread_stop(thread->tsk);
3421         kfree(thread);
3422 }
3423
3424 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3425 {
3426         if (!mddev) {
3427                 MD_BUG();
3428                 return;
3429         }
3430
3431         if (!rdev || rdev->faulty)
3432                 return;
3433 /*
3434         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3435                 mdname(mddev),
3436                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3437                 __builtin_return_address(0),__builtin_return_address(1),
3438                 __builtin_return_address(2),__builtin_return_address(3));
3439 */
3440         if (!mddev->pers->error_handler)
3441                 return;
3442         mddev->pers->error_handler(mddev,rdev);
3443         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3444         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3445         md_wakeup_thread(mddev->thread);
3446 }
3447
3448 /* seq_file implementation /proc/mdstat */
3449
3450 static void status_unused(struct seq_file *seq)
3451 {
3452         int i = 0;
3453         mdk_rdev_t *rdev;
3454         struct list_head *tmp;
3455
3456         seq_printf(seq, "unused devices: ");
3457
3458         ITERATE_RDEV_PENDING(rdev,tmp) {
3459                 char b[BDEVNAME_SIZE];
3460                 i++;
3461                 seq_printf(seq, "%s ",
3462                               bdevname(rdev->bdev,b));
3463         }
3464         if (!i)
3465                 seq_printf(seq, "<none>");
3466
3467         seq_printf(seq, "\n");
3468 }
3469
3470
3471 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3472 {
3473         unsigned long max_blocks, resync, res, dt, db, rt;
3474
3475         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3476
3477         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3478                 max_blocks = mddev->resync_max_sectors >> 1;
3479         else
3480                 max_blocks = mddev->size;
3481
3482         /*
3483          * Should not happen.
3484          */
3485         if (!max_blocks) {
3486                 MD_BUG();
3487                 return;
3488         }
3489         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3490         {
3491                 int i, x = res/50, y = 20-x;
3492                 seq_printf(seq, "[");
3493                 for (i = 0; i < x; i++)
3494                         seq_printf(seq, "=");
3495                 seq_printf(seq, ">");
3496                 for (i = 0; i < y; i++)
3497                         seq_printf(seq, ".");
3498                 seq_printf(seq, "] ");
3499         }
3500         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3501                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3502                        "resync" : "recovery"),
3503                       res/10, res % 10, resync, max_blocks);
3504
3505         /*
3506          * We do not want to overflow, so the order of operands and
3507          * the * 100 / 100 trick are important. We do a +1 to be
3508          * safe against division by zero. We only estimate anyway.
3509          *
3510          * dt: time from mark until now
3511          * db: blocks written from mark until now
3512          * rt: remaining time
3513          */
3514         dt = ((jiffies - mddev->resync_mark) / HZ);
3515         if (!dt) dt++;
3516         db = resync - (mddev->resync_mark_cnt/2);
3517         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3518
3519         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3520
3521         seq_printf(seq, " speed=%ldK/sec", db/dt);
3522 }
3523
3524 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3525 {
3526         struct list_head *tmp;
3527         loff_t l = *pos;
3528         mddev_t *mddev;
3529
3530         if (l >= 0x10000)
3531                 return NULL;
3532         if (!l--)
3533                 /* header */
3534                 return (void*)1;
3535
3536         spin_lock(&all_mddevs_lock);
3537         list_for_each(tmp,&all_mddevs)
3538                 if (!l--) {
3539                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3540                         mddev_get(mddev);
3541                         spin_unlock(&all_mddevs_lock);
3542                         return mddev;
3543                 }
3544         spin_unlock(&all_mddevs_lock);
3545         if (!l--)
3546                 return (void*)2;/* tail */
3547         return NULL;
3548 }
3549
3550 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3551 {
3552         struct list_head *tmp;
3553         mddev_t *next_mddev, *mddev = v;
3554
3555         ++*pos;
3556         if (v == (void*)2)
3557                 return NULL;
3558
3559         spin_lock(&all_mddevs_lock);
3560         if (v == (void*)1)
3561                 tmp = all_mddevs.next;
3562         else
3563                 tmp = mddev->all_mddevs.next;
3564         if (tmp != &all_mddevs)
3565                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3566         else {
3567                 next_mddev = (void*)2;
3568                 *pos = 0x10000;
3569         }
3570         spin_unlock(&all_mddevs_lock);
3571
3572         if (v != (void*)1)
3573                 mddev_put(mddev);
3574         return next_mddev;
3575
3576 }
3577
3578 static void md_seq_stop(struct seq_file *seq, void *v)
3579 {
3580         mddev_t *mddev = v;
3581
3582         if (mddev && v != (void*)1 && v != (void*)2)
3583                 mddev_put(mddev);
3584 }
3585
3586 static int md_seq_show(struct seq_file *seq, void *v)
3587 {
3588         mddev_t *mddev = v;
3589         sector_t size;
3590         struct list_head *tmp2;
3591         mdk_rdev_t *rdev;
3592         int i;
3593         struct bitmap *bitmap;
3594
3595         if (v == (void*)1) {
3596                 seq_printf(seq, "Personalities : ");
3597                 spin_lock(&pers_lock);
3598                 for (i = 0; i < MAX_PERSONALITY; i++)
3599                         if (pers[i])
3600                                 seq_printf(seq, "[%s] ", pers[i]->name);
3601
3602                 spin_unlock(&pers_lock);
3603                 seq_printf(seq, "\n");
3604                 return 0;
3605         }
3606         if (v == (void*)2) {
3607                 status_unused(seq);
3608                 return 0;
3609         }
3610
3611         if (mddev_lock(mddev)!=0)
3612                 return -EINTR;
3613         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3614                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3615                                                 mddev->pers ? "" : "in");
3616                 if (mddev->pers) {
3617                         if (mddev->ro)
3618                                 seq_printf(seq, " (read-only)");
3619                         seq_printf(seq, " %s", mddev->pers->name);
3620                 }
3621
3622                 size = 0;
3623                 ITERATE_RDEV(mddev,rdev,tmp2) {
3624                         char b[BDEVNAME_SIZE];
3625                         seq_printf(seq, " %s[%d]",
3626                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3627                         if (test_bit(WriteMostly, &rdev->flags))
3628                                 seq_printf(seq, "(W)");
3629                         if (rdev->faulty) {
3630                                 seq_printf(seq, "(F)");
3631                                 continue;
3632                         } else if (rdev->raid_disk < 0)
3633                                 seq_printf(seq, "(S)"); /* spare */
3634                         size += rdev->size;
3635                 }
3636
3637                 if (!list_empty(&mddev->disks)) {
3638                         if (mddev->pers)
3639                                 seq_printf(seq, "\n      %llu blocks",
3640                                         (unsigned long long)mddev->array_size);
3641                         else
3642                                 seq_printf(seq, "\n      %llu blocks",
3643                                         (unsigned long long)size);
3644                 }
3645                 if (mddev->persistent) {
3646                         if (mddev->major_version != 0 ||
3647                             mddev->minor_version != 90) {
3648                                 seq_printf(seq," super %d.%d",
3649                                            mddev->major_version,
3650                                            mddev->minor_version);
3651                         }
3652                 } else
3653                         seq_printf(seq, " super non-persistent");
3654
3655                 if (mddev->pers) {
3656                         mddev->pers->status (seq, mddev);
3657                         seq_printf(seq, "\n      ");
3658                         if (mddev->curr_resync > 2) {
3659                                 status_resync (seq, mddev);
3660                                 seq_printf(seq, "\n      ");
3661                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3662                                 seq_printf(seq, "       resync=DELAYED\n      ");
3663                 } else
3664                         seq_printf(seq, "\n       ");
3665
3666                 if ((bitmap = mddev->bitmap)) {
3667                         unsigned long chunk_kb;
3668                         unsigned long flags;
3669                         spin_lock_irqsave(&bitmap->lock, flags);
3670                         chunk_kb = bitmap->chunksize >> 10;
3671                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3672                                 "%lu%s chunk",
3673                                 bitmap->pages - bitmap->missing_pages,
3674                                 bitmap->pages,
3675                                 (bitmap->pages - bitmap->missing_pages)
3676                                         << (PAGE_SHIFT - 10),
3677                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3678                                 chunk_kb ? "KB" : "B");
3679                         if (bitmap->file) {
3680                                 seq_printf(seq, ", file: ");
3681                                 seq_path(seq, bitmap->file->f_vfsmnt,
3682                                          bitmap->file->f_dentry," \t\n");
3683                         }
3684
3685                         seq_printf(seq, "\n");
3686                         spin_unlock_irqrestore(&bitmap->lock, flags);
3687                 }
3688
3689                 seq_printf(seq, "\n");
3690         }
3691         mddev_unlock(mddev);
3692
3693         return 0;
3694 }
3695
3696 static struct seq_operations md_seq_ops = {
3697         .start  = md_seq_start,
3698         .next   = md_seq_next,
3699         .stop   = md_seq_stop,
3700         .show   = md_seq_show,
3701 };
3702
3703 static int md_seq_open(struct inode *inode, struct file *file)
3704 {
3705         int error;
3706
3707         error = seq_open(file, &md_seq_ops);
3708         return error;
3709 }
3710
3711 static struct file_operations md_seq_fops = {
3712         .open           = md_seq_open,
3713         .read           = seq_read,
3714         .llseek         = seq_lseek,
3715         .release        = seq_release,
3716 };
3717
3718 int register_md_personality(int pnum, mdk_personality_t *p)
3719 {
3720         if (pnum >= MAX_PERSONALITY) {
3721                 printk(KERN_ERR
3722                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3723                        p->name, pnum, MAX_PERSONALITY-1);
3724                 return -EINVAL;
3725         }
3726
3727         spin_lock(&pers_lock);
3728         if (pers[pnum]) {
3729                 spin_unlock(&pers_lock);
3730                 return -EBUSY;
3731         }
3732
3733         pers[pnum] = p;
3734         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3735         spin_unlock(&pers_lock);
3736         return 0;
3737 }
3738
3739 int unregister_md_personality(int pnum)
3740 {
3741         if (pnum >= MAX_PERSONALITY)
3742                 return -EINVAL;
3743
3744         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3745         spin_lock(&pers_lock);
3746         pers[pnum] = NULL;
3747         spin_unlock(&pers_lock);
3748         return 0;
3749 }
3750
3751 static int is_mddev_idle(mddev_t *mddev)
3752 {
3753         mdk_rdev_t * rdev;
3754         struct list_head *tmp;
3755         int idle;
3756         unsigned long curr_events;
3757
3758         idle = 1;
3759         ITERATE_RDEV(mddev,rdev,tmp) {
3760                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3761                 curr_events = disk_stat_read(disk, sectors[0]) +
3762                                 disk_stat_read(disk, sectors[1]) -
3763                                 atomic_read(&disk->sync_io);
3764                 /* Allow some slack between valud of curr_events and last_events,
3765                  * as there are some uninteresting races.
3766                  * Note: the following is an unsigned comparison.
3767                  */
3768                 if ((curr_events - rdev->last_events + 32) > 64) {
3769                         rdev->last_events = curr_events;
3770                         idle = 0;
3771                 }
3772         }
3773         return idle;
3774 }
3775
3776 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3777 {
3778         /* another "blocks" (512byte) blocks have been synced */
3779         atomic_sub(blocks, &mddev->recovery_active);
3780         wake_up(&mddev->recovery_wait);
3781         if (!ok) {
3782                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3783                 md_wakeup_thread(mddev->thread);
3784                 // stop recovery, signal do_sync ....
3785         }
3786 }
3787
3788
3789 /* md_write_start(mddev, bi)
3790  * If we need to update some array metadata (e.g. 'active' flag
3791  * in superblock) before writing, schedule a superblock update
3792  * and wait for it to complete.
3793  */
3794 void md_write_start(mddev_t *mddev, struct bio *bi)
3795 {
3796         if (bio_data_dir(bi) != WRITE)
3797                 return;
3798
3799         atomic_inc(&mddev->writes_pending);
3800         if (mddev->in_sync) {
3801                 spin_lock(&mddev->write_lock);
3802                 if (mddev->in_sync) {
3803                         mddev->in_sync = 0;
3804                         mddev->sb_dirty = 1;
3805                         md_wakeup_thread(mddev->thread);
3806                 }
3807                 spin_unlock(&mddev->write_lock);
3808         }
3809         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3810 }
3811
3812 void md_write_end(mddev_t *mddev)
3813 {
3814         if (atomic_dec_and_test(&mddev->writes_pending)) {
3815                 if (mddev->safemode == 2)
3816                         md_wakeup_thread(mddev->thread);
3817                 else
3818                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3819         }
3820 }
3821
3822 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3823
3824 #define SYNC_MARKS      10
3825 #define SYNC_MARK_STEP  (3*HZ)
3826 static void md_do_sync(mddev_t *mddev)
3827 {
3828         mddev_t *mddev2;
3829         unsigned int currspeed = 0,
3830                  window;
3831         sector_t max_sectors,j, io_sectors;
3832         unsigned long mark[SYNC_MARKS];
3833         sector_t mark_cnt[SYNC_MARKS];
3834         int last_mark,m;
3835         struct list_head *tmp;
3836         sector_t last_check;
3837         int skipped = 0;
3838
3839         /* just incase thread restarts... */
3840         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3841                 return;
3842
3843         /* we overload curr_resync somewhat here.
3844          * 0 == not engaged in resync at all
3845          * 2 == checking that there is no conflict with another sync
3846          * 1 == like 2, but have yielded to allow conflicting resync to
3847          *              commense
3848          * other == active in resync - this many blocks
3849          *
3850          * Before starting a resync we must have set curr_resync to
3851          * 2, and then checked that every "conflicting" array has curr_resync
3852          * less than ours.  When we find one that is the same or higher
3853          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3854          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3855          * This will mean we have to start checking from the beginning again.
3856          *
3857          */
3858
3859         do {
3860                 mddev->curr_resync = 2;
3861
3862         try_again:
3863                 if (signal_pending(current) ||
3864                     kthread_should_stop()) {
3865                         flush_signals(current);
3866                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3867                         goto skip;
3868                 }
3869                 ITERATE_MDDEV(mddev2,tmp) {
3870                         if (mddev2 == mddev)
3871                                 continue;
3872                         if (mddev2->curr_resync &&
3873                             match_mddev_units(mddev,mddev2)) {
3874                                 DEFINE_WAIT(wq);
3875                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3876                                         /* arbitrarily yield */
3877                                         mddev->curr_resync = 1;
3878                                         wake_up(&resync_wait);
3879                                 }
3880                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3881                                         /* no need to wait here, we can wait the next
3882                                          * time 'round when curr_resync == 2
3883                                          */
3884                                         continue;
3885                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3886                                 if (!signal_pending(current) &&
3887                                     !kthread_should_stop() &&
3888                                     mddev2->curr_resync >= mddev->curr_resync) {
3889                                         printk(KERN_INFO "md: delaying resync of %s"
3890                                                " until %s has finished resync (they"
3891                                                " share one or more physical units)\n",
3892                                                mdname(mddev), mdname(mddev2));
3893                                         mddev_put(mddev2);
3894                                         schedule();
3895                                         finish_wait(&resync_wait, &wq);
3896                                         goto try_again;
3897                                 }
3898                                 finish_wait(&resync_wait, &wq);
3899                         }
3900                 }
3901         } while (mddev->curr_resync < 2);
3902
3903         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3904                 /* resync follows the size requested by the personality,
3905                  * which defaults to physical size, but can be virtual size
3906                  */
3907                 max_sectors = mddev->resync_max_sectors;
3908                 mddev->resync_mismatches = 0;
3909         } else
3910                 /* recovery follows the physical size of devices */
3911                 max_sectors = mddev->size << 1;
3912
3913         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3914         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3915                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3916         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3917                "(but not more than %d KB/sec) for reconstruction.\n",
3918                sysctl_speed_limit_max);
3919
3920         is_mddev_idle(mddev); /* this also initializes IO event counters */
3921         /* we don't use the checkpoint if there's a bitmap */
3922         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
3923             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3924                 j = mddev->recovery_cp;
3925         else
3926                 j = 0;
3927         io_sectors = 0;
3928         for (m = 0; m < SYNC_MARKS; m++) {
3929                 mark[m] = jiffies;
3930                 mark_cnt[m] = io_sectors;
3931         }
3932         last_mark = 0;
3933         mddev->resync_mark = mark[last_mark];
3934         mddev->resync_mark_cnt = mark_cnt[last_mark];
3935
3936         /*
3937          * Tune reconstruction:
3938          */
3939         window = 32*(PAGE_SIZE/512);
3940         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3941                 window/2,(unsigned long long) max_sectors/2);
3942
3943         atomic_set(&mddev->recovery_active, 0);
3944         init_waitqueue_head(&mddev->recovery_wait);
3945         last_check = 0;
3946
3947         if (j>2) {
3948                 printk(KERN_INFO
3949                         "md: resuming recovery of %s from checkpoint.\n",
3950                         mdname(mddev));
3951                 mddev->curr_resync = j;
3952         }
3953
3954         while (j < max_sectors) {
3955                 sector_t sectors;
3956
3957                 skipped = 0;
3958                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3959                                             currspeed < sysctl_speed_limit_min);
3960                 if (sectors == 0) {
3961                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3962                         goto out;
3963                 }
3964
3965                 if (!skipped) { /* actual IO requested */
3966                         io_sectors += sectors;
3967                         atomic_add(sectors, &mddev->recovery_active);
3968                 }
3969
3970                 j += sectors;
3971                 if (j>1) mddev->curr_resync = j;
3972
3973
3974                 if (last_check + window > io_sectors || j == max_sectors)
3975                         continue;
3976
3977                 last_check = io_sectors;
3978
3979                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3980                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3981                         break;
3982
3983         repeat:
3984                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3985                         /* step marks */
3986                         int next = (last_mark+1) % SYNC_MARKS;
3987
3988                         mddev->resync_mark = mark[next];
3989                         mddev->resync_mark_cnt = mark_cnt[next];
3990                         mark[next] = jiffies;
3991                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3992                         last_mark = next;
3993                 }
3994
3995
3996                 if (signal_pending(current) || kthread_should_stop()) {
3997                         /*
3998                          * got a signal, exit.
3999                          */
4000                         printk(KERN_INFO
4001                                 "md: md_do_sync() got signal ... exiting\n");
4002                         flush_signals(current);
4003                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4004                         goto out;
4005                 }
4006
4007                 /*
4008                  * this loop exits only if either when we are slower than
4009                  * the 'hard' speed limit, or the system was IO-idle for
4010                  * a jiffy.
4011                  * the system might be non-idle CPU-wise, but we only care
4012                  * about not overloading the IO subsystem. (things like an
4013                  * e2fsck being done on the RAID array should execute fast)
4014                  */
4015                 mddev->queue->unplug_fn(mddev->queue);
4016                 cond_resched();
4017
4018                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4019                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4020
4021                 if (currspeed > sysctl_speed_limit_min) {
4022                         if ((currspeed > sysctl_speed_limit_max) ||
4023                                         !is_mddev_idle(mddev)) {
4024                                 msleep_interruptible(250);
4025                                 goto repeat;
4026                         }
4027                 }
4028         }
4029         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4030         /*
4031          * this also signals 'finished resyncing' to md_stop
4032          */
4033  out:
4034         mddev->queue->unplug_fn(mddev->queue);
4035
4036         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4037
4038         /* tell personality that we are finished */
4039         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4040
4041         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4042             mddev->curr_resync > 2 &&
4043             mddev->curr_resync >= mddev->recovery_cp) {
4044                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4045                         printk(KERN_INFO
4046                                 "md: checkpointing recovery of %s.\n",
4047                                 mdname(mddev));
4048                         mddev->recovery_cp = mddev->curr_resync;
4049                 } else
4050                         mddev->recovery_cp = MaxSector;
4051         }
4052
4053  skip:
4054         mddev->curr_resync = 0;
4055         wake_up(&resync_wait);
4056         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4057         md_wakeup_thread(mddev->thread);
4058 }
4059
4060
4061 /*
4062  * This routine is regularly called by all per-raid-array threads to
4063  * deal with generic issues like resync and super-block update.
4064  * Raid personalities that don't have a thread (linear/raid0) do not
4065  * need this as they never do any recovery or update the superblock.
4066  *
4067  * It does not do any resync itself, but rather "forks" off other threads
4068  * to do that as needed.
4069  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4070  * "->recovery" and create a thread at ->sync_thread.
4071  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4072  * and wakeups up this thread which will reap the thread and finish up.
4073  * This thread also removes any faulty devices (with nr_pending == 0).
4074  *
4075  * The overall approach is:
4076  *  1/ if the superblock needs updating, update it.
4077  *  2/ If a recovery thread is running, don't do anything else.
4078  *  3/ If recovery has finished, clean up, possibly marking spares active.
4079  *  4/ If there are any faulty devices, remove them.
4080  *  5/ If array is degraded, try to add spares devices
4081  *  6/ If array has spares or is not in-sync, start a resync thread.
4082  */
4083 void md_check_recovery(mddev_t *mddev)
4084 {
4085         mdk_rdev_t *rdev;
4086         struct list_head *rtmp;
4087
4088
4089         if (mddev->bitmap)
4090                 bitmap_daemon_work(mddev->bitmap);
4091
4092         if (mddev->ro)
4093                 return;
4094
4095         if (signal_pending(current)) {
4096                 if (mddev->pers->sync_request) {
4097                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4098                                mdname(mddev));
4099                         mddev->safemode = 2;
4100                 }
4101                 flush_signals(current);
4102         }
4103
4104         if ( ! (
4105                 mddev->sb_dirty ||
4106                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4107                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4108                 (mddev->safemode == 1) ||
4109                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4110                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4111                 ))
4112                 return;
4113
4114         if (mddev_trylock(mddev)==0) {
4115                 int spares =0;
4116
4117                 spin_lock(&mddev->write_lock);
4118                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4119                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4120                         mddev->in_sync = 1;
4121                         mddev->sb_dirty = 1;
4122                 }
4123                 if (mddev->safemode == 1)
4124                         mddev->safemode = 0;
4125                 spin_unlock(&mddev->write_lock);
4126
4127                 if (mddev->sb_dirty)
4128                         md_update_sb(mddev);
4129
4130
4131                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4132                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4133                         /* resync/recovery still happening */
4134                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4135                         goto unlock;
4136                 }
4137                 if (mddev->sync_thread) {
4138                         /* resync has finished, collect result */
4139                         md_unregister_thread(mddev->sync_thread);
4140                         mddev->sync_thread = NULL;
4141                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4142                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4143                                 /* success...*/
4144                                 /* activate any spares */
4145                                 mddev->pers->spare_active(mddev);
4146                         }
4147                         md_update_sb(mddev);
4148
4149                         /* if array is no-longer degraded, then any saved_raid_disk
4150                          * information must be scrapped
4151                          */
4152                         if (!mddev->degraded)
4153                                 ITERATE_RDEV(mddev,rdev,rtmp)
4154                                         rdev->saved_raid_disk = -1;
4155
4156                         mddev->recovery = 0;
4157                         /* flag recovery needed just to double check */
4158                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4159                         goto unlock;
4160                 }
4161                 /* Clear some bits that don't mean anything, but
4162                  * might be left set
4163                  */
4164                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4165                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4166                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4167                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4168
4169                 /* no recovery is running.
4170                  * remove any failed drives, then
4171                  * add spares if possible.
4172                  * Spare are also removed and re-added, to allow
4173                  * the personality to fail the re-add.
4174                  */
4175                 ITERATE_RDEV(mddev,rdev,rtmp)
4176                         if (rdev->raid_disk >= 0 &&
4177                             (rdev->faulty || ! rdev->in_sync) &&
4178                             atomic_read(&rdev->nr_pending)==0) {
4179                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4180                                         char nm[20];
4181                                         sprintf(nm,"rd%d", rdev->raid_disk);
4182                                         sysfs_remove_link(&mddev->kobj, nm);
4183                                         rdev->raid_disk = -1;
4184                                 }
4185                         }
4186
4187                 if (mddev->degraded) {
4188                         ITERATE_RDEV(mddev,rdev,rtmp)
4189                                 if (rdev->raid_disk < 0
4190                                     && !rdev->faulty) {
4191                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4192                                                 char nm[20];
4193                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4194                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4195                                                 spares++;
4196                                         } else
4197                                                 break;
4198                                 }
4199                 }
4200
4201                 if (spares) {
4202                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4203                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4204                 } else if (mddev->recovery_cp < MaxSector) {
4205                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4206                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4207                         /* nothing to be done ... */
4208                         goto unlock;
4209
4210                 if (mddev->pers->sync_request) {
4211                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4212                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4213                                 /* We are adding a device or devices to an array
4214                                  * which has the bitmap stored on all devices.
4215                                  * So make sure all bitmap pages get written
4216                                  */
4217                                 bitmap_write_all(mddev->bitmap);
4218                         }
4219                         mddev->sync_thread = md_register_thread(md_do_sync,
4220                                                                 mddev,
4221                                                                 "%s_resync");
4222                         if (!mddev->sync_thread) {
4223                                 printk(KERN_ERR "%s: could not start resync"
4224                                         " thread...\n",
4225                                         mdname(mddev));
4226                                 /* leave the spares where they are, it shouldn't hurt */
4227                                 mddev->recovery = 0;
4228                         } else {
4229                                 md_wakeup_thread(mddev->sync_thread);
4230                         }
4231                 }
4232         unlock:
4233                 mddev_unlock(mddev);
4234         }
4235 }
4236
4237 static int md_notify_reboot(struct notifier_block *this,
4238                             unsigned long code, void *x)
4239 {
4240         struct list_head *tmp;
4241         mddev_t *mddev;
4242
4243         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4244
4245                 printk(KERN_INFO "md: stopping all md devices.\n");
4246
4247                 ITERATE_MDDEV(mddev,tmp)
4248                         if (mddev_trylock(mddev)==0)
4249                                 do_md_stop (mddev, 1);
4250                 /*
4251                  * certain more exotic SCSI devices are known to be
4252                  * volatile wrt too early system reboots. While the
4253                  * right place to handle this issue is the given
4254                  * driver, we do want to have a safe RAID driver ...
4255                  */
4256                 mdelay(1000*1);
4257         }
4258         return NOTIFY_DONE;
4259 }
4260
4261 static struct notifier_block md_notifier = {
4262         .notifier_call  = md_notify_reboot,
4263         .next           = NULL,
4264         .priority       = INT_MAX, /* before any real devices */
4265 };
4266
4267 static void md_geninit(void)
4268 {
4269         struct proc_dir_entry *p;
4270
4271         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4272
4273         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4274         if (p)
4275                 p->proc_fops = &md_seq_fops;
4276 }
4277
4278 static int __init md_init(void)
4279 {
4280         int minor;
4281
4282         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4283                         " MD_SB_DISKS=%d\n",
4284                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4285                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4286         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
4287                         BITMAP_MINOR);
4288
4289         if (register_blkdev(MAJOR_NR, "md"))
4290                 return -1;
4291         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4292                 unregister_blkdev(MAJOR_NR, "md");
4293                 return -1;
4294         }
4295         devfs_mk_dir("md");
4296         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4297                                 md_probe, NULL, NULL);
4298         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4299                             md_probe, NULL, NULL);
4300
4301         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4302                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4303                                 S_IFBLK|S_IRUSR|S_IWUSR,
4304                                 "md/%d", minor);
4305
4306         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4307                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4308                               S_IFBLK|S_IRUSR|S_IWUSR,
4309                               "md/mdp%d", minor);
4310
4311
4312         register_reboot_notifier(&md_notifier);
4313         raid_table_header = register_sysctl_table(raid_root_table, 1);
4314
4315         md_geninit();
4316         return (0);
4317 }
4318
4319
4320 #ifndef MODULE
4321
4322 /*
4323  * Searches all registered partitions for autorun RAID arrays
4324  * at boot time.
4325  */
4326 static dev_t detected_devices[128];
4327 static int dev_cnt;
4328
4329 void md_autodetect_dev(dev_t dev)
4330 {
4331         if (dev_cnt >= 0 && dev_cnt < 127)
4332                 detected_devices[dev_cnt++] = dev;
4333 }
4334
4335
4336 static void autostart_arrays(int part)
4337 {
4338         mdk_rdev_t *rdev;
4339         int i;
4340
4341         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4342
4343         for (i = 0; i < dev_cnt; i++) {
4344                 dev_t dev = detected_devices[i];
4345
4346                 rdev = md_import_device(dev,0, 0);
4347                 if (IS_ERR(rdev))
4348                         continue;
4349
4350                 if (rdev->faulty) {
4351                         MD_BUG();
4352                         continue;
4353                 }
4354                 list_add(&rdev->same_set, &pending_raid_disks);
4355         }
4356         dev_cnt = 0;
4357
4358         autorun_devices(part);
4359 }
4360
4361 #endif
4362
4363 static __exit void md_exit(void)
4364 {
4365         mddev_t *mddev;
4366         struct list_head *tmp;
4367         int i;
4368         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4369         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4370         for (i=0; i < MAX_MD_DEVS; i++)
4371                 devfs_remove("md/%d", i);
4372         for (i=0; i < MAX_MD_DEVS; i++)
4373                 devfs_remove("md/d%d", i);
4374
4375         devfs_remove("md");
4376
4377         unregister_blkdev(MAJOR_NR,"md");
4378         unregister_blkdev(mdp_major, "mdp");
4379         unregister_reboot_notifier(&md_notifier);
4380         unregister_sysctl_table(raid_table_header);
4381         remove_proc_entry("mdstat", NULL);
4382         ITERATE_MDDEV(mddev,tmp) {
4383                 struct gendisk *disk = mddev->gendisk;
4384                 if (!disk)
4385                         continue;
4386                 export_array(mddev);
4387                 del_gendisk(disk);
4388                 put_disk(disk);
4389                 mddev->gendisk = NULL;
4390                 mddev_put(mddev);
4391         }
4392 }
4393
4394 module_init(md_init)
4395 module_exit(md_exit)
4396
4397 EXPORT_SYMBOL(register_md_personality);
4398 EXPORT_SYMBOL(unregister_md_personality);
4399 EXPORT_SYMBOL(md_error);
4400 EXPORT_SYMBOL(md_done_sync);
4401 EXPORT_SYMBOL(md_write_start);
4402 EXPORT_SYMBOL(md_write_end);
4403 EXPORT_SYMBOL(md_register_thread);
4404 EXPORT_SYMBOL(md_unregister_thread);
4405 EXPORT_SYMBOL(md_wakeup_thread);
4406 EXPORT_SYMBOL(md_print_devices);
4407 EXPORT_SYMBOL(md_check_recovery);
4408 MODULE_LICENSE("GPL");
4409 MODULE_ALIAS("md");
4410 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);