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
 702         /* make rdev->sb match mddev data..
 703          *
 704          * 1/ zero out disks
 705          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 706          * 3/ any empty disks < next_spare become removed
 707          *
 708          * disks[0] gets initialised to REMOVED because
 709          * we cannot be sure from other fields if it has
 710          * been initialised or not.
 711          */
 712         int i;
 713         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 714         unsigned int fixdesc=0;
 715
 716         rdev->sb_size = MD_SB_BYTES;
 717
 718         sb = (mdp_super_t*)page_address(rdev->sb_page);
 719
 720         memset(sb, 0, sizeof(*sb));
 721
 722         sb->md_magic = MD_SB_MAGIC;
 723         sb->major_version = mddev->major_version;
 724         sb->minor_version = mddev->minor_version;
 725         sb->patch_version = mddev->patch_version;
 726         sb->gvalid_words  = 0; /* ignored */
 727         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 728         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 729         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 730         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 731
 732         sb->ctime = mddev->ctime;
 733         sb->level = mddev->level;
 734         sb->size  = mddev->size;
 735         sb->raid_disks = mddev->raid_disks;
 736         sb->md_minor = mddev->md_minor;
 737         sb->not_persistent = !mddev->persistent;
 738         sb->utime = mddev->utime;
 739         sb->state = 0;
 740         sb->events_hi = (mddev->events>>32);
 741         sb->events_lo = (u32)mddev->events;
 742
 743         if (mddev->in_sync)
 744         {
 745                 sb->recovery_cp = mddev->recovery_cp;
 746                 sb->cp_events_hi = (mddev->events>>32);
 747                 sb->cp_events_lo = (u32)mddev->events;
 748                 if (mddev->recovery_cp == MaxSector)
 749                         sb->state = (1<< MD_SB_CLEAN);
 750         } else
 751                 sb->recovery_cp = 0;
 752
 753         sb->layout = mddev->layout;
 754         sb->chunk_size = mddev->chunk_size;
 755
 756         if (mddev->bitmap && mddev->bitmap_file == NULL)
 757                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 758
 759         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 760         ITERATE_RDEV(mddev,rdev2,tmp) {
 761                 mdp_disk_t *d;
 762                 int desc_nr;
 763                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 764                         desc_nr = rdev2->raid_disk;
 765                 else
 766                         desc_nr = next_spare++;
 767                 if (desc_nr != rdev2->desc_nr) {
 768                         fixdesc |= (1 << desc_nr);
 769                         rdev2->desc_nr = desc_nr;
 770                         if (rdev2->raid_disk >= 0) {
 771                                 char nm[20];
 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(&rdev2->kobj);
 809                                 sysfs_create_link(&rdev2->kobj,
 810                                                   &rdev2->bdev->bd_disk->kobj,
 811                                                   "block");
 812                                 if (rdev2->raid_disk >= 0) {
 813                                         char nm[20];
 814                                         sprintf(nm, "rd%d", rdev2->raid_disk);
 815                                         sysfs_create_link(&mddev->kobj,
 816                                                           &rdev2->kobj, nm);
 817                                 }
 818                         }
 819         /* now set the "removed" and "faulty" bits on any missing devices */
 820         for (i=0 ; i < mddev->raid_disks ; i++) {
 821                 mdp_disk_t *d = &sb->disks[i];
 822                 if (d->state == 0 && d->number == 0) {
 823                         d->number = i;
 824                         d->raid_disk = i;
 825                         d->state = (1<<MD_DISK_REMOVED);
 826                         d->state |= (1<<MD_DISK_FAULTY);
 827                         failed++;
 828                 }
 829         }
 830         sb->nr_disks = nr_disks;
 831         sb->active_disks = active;
 832         sb->working_disks = working;
 833         sb->failed_disks = failed;
 834         sb->spare_disks = spare;
 835
 836         sb->this_disk = sb->disks[rdev->desc_nr];
 837         sb->sb_csum = calc_sb_csum(sb);
 838 }
 839
 840 /*
 841  * version 1 superblock
 842  */
 843
 844 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 845 {
 846         unsigned int disk_csum, csum;
 847         unsigned long long newcsum;
 848         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 849         unsigned int *isuper = (unsigned int*)sb;
 850         int i;
 851
 852         disk_csum = sb->sb_csum;
 853         sb->sb_csum = 0;
 854         newcsum = 0;
 855         for (i=0; size>=4; size -= 4 )
 856                 newcsum += le32_to_cpu(*isuper++);
 857
 858         if (size == 2)
 859                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 860
 861         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 862         sb->sb_csum = disk_csum;
 863         return cpu_to_le32(csum);
 864 }
 865
 866 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 867 {
 868         struct mdp_superblock_1 *sb;
 869         int ret;
 870         sector_t sb_offset;
 871         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 872         int bmask;
 873
 874         /*
 875          * Calculate the position of the superblock.
 876          * It is always aligned to a 4K boundary and
 877          * depeding on minor_version, it can be:
 878          * 0: At least 8K, but less than 12K, from end of device
 879          * 1: At start of device
 880          * 2: 4K from start of device.
 881          */
 882         switch(minor_version) {
 883         case 0:
 884                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 885                 sb_offset -= 8*2;
 886                 sb_offset &= ~(sector_t)(4*2-1);
 887                 /* convert from sectors to K */
 888                 sb_offset /= 2;
 889                 break;
 890         case 1:
 891                 sb_offset = 0;
 892                 break;
 893         case 2:
 894                 sb_offset = 4;
 895                 break;
 896         default:
 897                 return -EINVAL;
 898         }
 899         rdev->sb_offset = sb_offset;
 900
 901         /* superblock is rarely larger than 1K, but it can be larger,
 902          * and it is safe to read 4k, so we do that
 903          */
 904         ret = read_disk_sb(rdev, 4096);
 905         if (ret) return ret;
 906
 907
 908         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 909
 910         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 911             sb->major_version != cpu_to_le32(1) ||
 912             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 913             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 914             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 915                 return -EINVAL;
 916
 917         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 918                 printk("md: invalid superblock checksum on %s\n",
 919                         bdevname(rdev->bdev,b));
 920                 return -EINVAL;
 921         }
 922         if (le64_to_cpu(sb->data_size) < 10) {
 923                 printk("md: data_size too small on %s\n",
 924                        bdevname(rdev->bdev,b));
 925                 return -EINVAL;
 926         }
 927         rdev->preferred_minor = 0xffff;
 928         rdev->data_offset = le64_to_cpu(sb->data_offset);
 929
 930         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 931         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 932         if (rdev->sb_size & bmask)
 933                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 934
 935         if (refdev == 0)
 936                 return 1;
 937         else {
 938                 __u64 ev1, ev2;
 939                 struct mdp_superblock_1 *refsb =
 940                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 941
 942                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 943                     sb->level != refsb->level ||
 944                     sb->layout != refsb->layout ||
 945                     sb->chunksize != refsb->chunksize) {
 946                         printk(KERN_WARNING "md: %s has strangely different"
 947                                 " superblock to %s\n",
 948                                 bdevname(rdev->bdev,b),
 949                                 bdevname(refdev->bdev,b2));
 950                         return -EINVAL;
 951                 }
 952                 ev1 = le64_to_cpu(sb->events);
 953                 ev2 = le64_to_cpu(refsb->events);
 954
 955                 if (ev1 > ev2)
 956                         return 1;
 957         }
 958         if (minor_version)
 959                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 960         else
 961                 rdev->size = rdev->sb_offset;
 962         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 963                 return -EINVAL;
 964         rdev->size = le64_to_cpu(sb->data_size)/2;
 965         if (le32_to_cpu(sb->chunksize))
 966                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 967         return 0;
 968 }
 969
 970 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 971 {
 972         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 973
 974         rdev->raid_disk = -1;
 975         rdev->in_sync = 0;
 976         if (mddev->raid_disks == 0) {
 977                 mddev->major_version = 1;
 978                 mddev->patch_version = 0;
 979                 mddev->persistent = 1;
 980                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 981                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 982                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 983                 mddev->level = le32_to_cpu(sb->level);
 984                 mddev->layout = le32_to_cpu(sb->layout);
 985                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 986                 mddev->size = le64_to_cpu(sb->size)/2;
 987                 mddev->events = le64_to_cpu(sb->events);
 988                 mddev->bitmap_offset = 0;
 989                 mddev->default_bitmap_offset = 0;
 990                 mddev->default_bitmap_offset = 1024;
 991
 992                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 993                 memcpy(mddev->uuid, sb->set_uuid, 16);
 994
 995                 mddev->max_disks =  (4096-256)/2;
 996
 997                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
 998                     mddev->bitmap_file == NULL ) {
 999                         if (mddev->level != 1) {
1000                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1001                                 return -EINVAL;
1002                         }
1003                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1004                 }
1005         } else if (mddev->pers == NULL) {
1006                 /* Insist of good event counter while assembling */
1007                 __u64 ev1 = le64_to_cpu(sb->events);
1008                 ++ev1;
1009                 if (ev1 < mddev->events)
1010                         return -EINVAL;
1011         } else if (mddev->bitmap) {
1012                 /* If adding to array with a bitmap, then we can accept an
1013                  * older device, but not too old.
1014                  */
1015                 __u64 ev1 = le64_to_cpu(sb->events);
1016                 if (ev1 < mddev->bitmap->events_cleared)
1017                         return 0;
1018         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1019                 return 0;
1020
1021         if (mddev->level != LEVEL_MULTIPATH) {
1022                 int role;
1023                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1024                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1025                 switch(role) {
1026                 case 0xffff: /* spare */
1027                         rdev->faulty = 0;
1028                         break;
1029                 case 0xfffe: /* faulty */
1030                         rdev->faulty = 1;
1031                         break;
1032                 default:
1033                         rdev->in_sync = 1;
1034                         rdev->faulty = 0;
1035                         rdev->raid_disk = role;
1036                         break;
1037                 }
1038                 rdev->flags = 0;
1039                 if (sb->devflags & WriteMostly1)
1040                         set_bit(WriteMostly, &rdev->flags);
1041         } else /* MULTIPATH are always insync */
1042                 rdev->in_sync = 1;
1043
1044         return 0;
1045 }
1046
1047 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1048 {
1049         struct mdp_superblock_1 *sb;
1050         struct list_head *tmp;
1051         mdk_rdev_t *rdev2;
1052         int max_dev, i;
1053         /* make rdev->sb match mddev and rdev data. */
1054
1055         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1056
1057         sb->feature_map = 0;
1058         sb->pad0 = 0;
1059         memset(sb->pad1, 0, sizeof(sb->pad1));
1060         memset(sb->pad2, 0, sizeof(sb->pad2));
1061         memset(sb->pad3, 0, sizeof(sb->pad3));
1062
1063         sb->utime = cpu_to_le64((__u64)mddev->utime);
1064         sb->events = cpu_to_le64(mddev->events);
1065         if (mddev->in_sync)
1066                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1067         else
1068                 sb->resync_offset = cpu_to_le64(0);
1069
1070         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1071                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1072                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1073         }
1074
1075         max_dev = 0;
1076         ITERATE_RDEV(mddev,rdev2,tmp)
1077                 if (rdev2->desc_nr+1 > max_dev)
1078                         max_dev = rdev2->desc_nr+1;
1079
1080         sb->max_dev = cpu_to_le32(max_dev);
1081         for (i=0; i<max_dev;i++)
1082                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1083
1084         ITERATE_RDEV(mddev,rdev2,tmp) {
1085                 i = rdev2->desc_nr;
1086                 if (rdev2->faulty)
1087                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1088                 else if (rdev2->in_sync)
1089                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1090                 else
1091                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1092         }
1093
1094         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1095         sb->sb_csum = calc_sb_1_csum(sb);
1096 }
1097
1098
1099 static struct super_type super_types[] = {
1100         [0] = {
1101                 .name   = "0.90.0",
1102                 .owner  = THIS_MODULE,
1103                 .load_super     = super_90_load,
1104                 .validate_super = super_90_validate,
1105                 .sync_super     = super_90_sync,
1106         },
1107         [1] = {
1108                 .name   = "md-1",
1109                 .owner  = THIS_MODULE,
1110                 .load_super     = super_1_load,
1111                 .validate_super = super_1_validate,
1112                 .sync_super     = super_1_sync,
1113         },
1114 };
1115
1116 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1117 {
1118         struct list_head *tmp;
1119         mdk_rdev_t *rdev;
1120
1121         ITERATE_RDEV(mddev,rdev,tmp)
1122                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1123                         return rdev;
1124
1125         return NULL;
1126 }
1127
1128 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1129 {
1130         struct list_head *tmp;
1131         mdk_rdev_t *rdev;
1132
1133         ITERATE_RDEV(mddev1,rdev,tmp)
1134                 if (match_dev_unit(mddev2, rdev))
1135                         return 1;
1136
1137         return 0;
1138 }
1139
1140 static LIST_HEAD(pending_raid_disks);
1141
1142 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1143 {
1144         mdk_rdev_t *same_pdev;
1145         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1146
1147         if (rdev->mddev) {
1148                 MD_BUG();
1149                 return -EINVAL;
1150         }
1151         same_pdev = match_dev_unit(mddev, rdev);
1152         if (same_pdev)
1153                 printk(KERN_WARNING
1154                         "%s: WARNING: %s appears to be on the same physical"
1155                         " disk as %s. True\n     protection against single-disk"
1156                         " failure might be compromised.\n",
1157                         mdname(mddev), bdevname(rdev->bdev,b),
1158                         bdevname(same_pdev->bdev,b2));
1159
1160         /* Verify rdev->desc_nr is unique.
1161          * If it is -1, assign a free number, else
1162          * check number is not in use
1163          */
1164         if (rdev->desc_nr < 0) {
1165                 int choice = 0;
1166                 if (mddev->pers) choice = mddev->raid_disks;
1167                 while (find_rdev_nr(mddev, choice))
1168                         choice++;
1169                 rdev->desc_nr = choice;
1170         } else {
1171                 if (find_rdev_nr(mddev, rdev->desc_nr))
1172                         return -EBUSY;
1173         }
1174
1175         list_add(&rdev->same_set, &mddev->disks);
1176         rdev->mddev = mddev;
1177         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1178
1179         rdev->kobj.k_name = NULL;
1180         snprintf(rdev->kobj.name, KOBJ_NAME_LEN, "dev%d", rdev->desc_nr);
1181         rdev->kobj.parent = kobject_get(&mddev->kobj);
1182         kobject_add(&rdev->kobj);
1183
1184         sysfs_create_link(&rdev->kobj, &rdev->bdev->bd_disk->kobj, "block");
1185         return 0;
1186 }
1187
1188 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1189 {
1190         char b[BDEVNAME_SIZE];
1191         if (!rdev->mddev) {
1192                 MD_BUG();
1193                 return;
1194         }
1195         list_del_init(&rdev->same_set);
1196         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1197         rdev->mddev = NULL;
1198         sysfs_remove_link(&rdev->kobj, "block");
1199         kobject_del(&rdev->kobj);
1200 }
1201
1202 /*
1203  * prevent the device from being mounted, repartitioned or
1204  * otherwise reused by a RAID array (or any other kernel
1205  * subsystem), by bd_claiming the device.
1206  */
1207 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1208 {
1209         int err = 0;
1210         struct block_device *bdev;
1211         char b[BDEVNAME_SIZE];
1212
1213         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1214         if (IS_ERR(bdev)) {
1215                 printk(KERN_ERR "md: could not open %s.\n",
1216                         __bdevname(dev, b));
1217                 return PTR_ERR(bdev);
1218         }
1219         err = bd_claim(bdev, rdev);
1220         if (err) {
1221                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1222                         bdevname(bdev, b));
1223                 blkdev_put(bdev);
1224                 return err;
1225         }
1226         rdev->bdev = bdev;
1227         return err;
1228 }
1229
1230 static void unlock_rdev(mdk_rdev_t *rdev)
1231 {
1232         struct block_device *bdev = rdev->bdev;
1233         rdev->bdev = NULL;
1234         if (!bdev)
1235                 MD_BUG();
1236         bd_release(bdev);
1237         blkdev_put(bdev);
1238 }
1239
1240 void md_autodetect_dev(dev_t dev);
1241
1242 static void export_rdev(mdk_rdev_t * rdev)
1243 {
1244         char b[BDEVNAME_SIZE];
1245         printk(KERN_INFO "md: export_rdev(%s)\n",
1246                 bdevname(rdev->bdev,b));
1247         if (rdev->mddev)
1248                 MD_BUG();
1249         free_disk_sb(rdev);
1250         list_del_init(&rdev->same_set);
1251 #ifndef MODULE
1252         md_autodetect_dev(rdev->bdev->bd_dev);
1253 #endif
1254         unlock_rdev(rdev);
1255         kobject_put(&rdev->kobj);
1256 }
1257
1258 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1259 {
1260         unbind_rdev_from_array(rdev);
1261         export_rdev(rdev);
1262 }
1263
1264 static void export_array(mddev_t *mddev)
1265 {
1266         struct list_head *tmp;
1267         mdk_rdev_t *rdev;
1268
1269         ITERATE_RDEV(mddev,rdev,tmp) {
1270                 if (!rdev->mddev) {
1271                         MD_BUG();
1272                         continue;
1273                 }
1274                 kick_rdev_from_array(rdev);
1275         }
1276         if (!list_empty(&mddev->disks))
1277                 MD_BUG();
1278         mddev->raid_disks = 0;
1279         mddev->major_version = 0;
1280 }
1281
1282 static void print_desc(mdp_disk_t *desc)
1283 {
1284         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1285                 desc->major,desc->minor,desc->raid_disk,desc->state);
1286 }
1287
1288 static void print_sb(mdp_super_t *sb)
1289 {
1290         int i;
1291
1292         printk(KERN_INFO
1293                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1294                 sb->major_version, sb->minor_version, sb->patch_version,
1295                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1296                 sb->ctime);
1297         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1298                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1299                 sb->md_minor, sb->layout, sb->chunk_size);
1300         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1301                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1302                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1303                 sb->failed_disks, sb->spare_disks,
1304                 sb->sb_csum, (unsigned long)sb->events_lo);
1305
1306         printk(KERN_INFO);
1307         for (i = 0; i < MD_SB_DISKS; i++) {
1308                 mdp_disk_t *desc;
1309
1310                 desc = sb->disks + i;
1311                 if (desc->number || desc->major || desc->minor ||
1312                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1313                         printk("     D %2d: ", i);
1314                         print_desc(desc);
1315                 }
1316         }
1317         printk(KERN_INFO "md:     THIS: ");
1318         print_desc(&sb->this_disk);
1319
1320 }
1321
1322 static void print_rdev(mdk_rdev_t *rdev)
1323 {
1324         char b[BDEVNAME_SIZE];
1325         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1326                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1327                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1328         if (rdev->sb_loaded) {
1329                 printk(KERN_INFO "md: rdev superblock:\n");
1330                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1331         } else
1332                 printk(KERN_INFO "md: no rdev superblock!\n");
1333 }
1334
1335 void md_print_devices(void)
1336 {
1337         struct list_head *tmp, *tmp2;
1338         mdk_rdev_t *rdev;
1339         mddev_t *mddev;
1340         char b[BDEVNAME_SIZE];
1341
1342         printk("\n");
1343         printk("md:     **********************************\n");
1344         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1345         printk("md:     **********************************\n");
1346         ITERATE_MDDEV(mddev,tmp) {
1347
1348                 if (mddev->bitmap)
1349                         bitmap_print_sb(mddev->bitmap);
1350                 else
1351                         printk("%s: ", mdname(mddev));
1352                 ITERATE_RDEV(mddev,rdev,tmp2)
1353                         printk("<%s>", bdevname(rdev->bdev,b));
1354                 printk("\n");
1355
1356                 ITERATE_RDEV(mddev,rdev,tmp2)
1357                         print_rdev(rdev);
1358         }
1359         printk("md:     **********************************\n");
1360         printk("\n");
1361 }
1362
1363
1364 static void sync_sbs(mddev_t * mddev)
1365 {
1366         mdk_rdev_t *rdev;
1367         struct list_head *tmp;
1368
1369         ITERATE_RDEV(mddev,rdev,tmp) {
1370                 super_types[mddev->major_version].
1371                         sync_super(mddev, rdev);
1372                 rdev->sb_loaded = 1;
1373         }
1374 }
1375
1376 static void md_update_sb(mddev_t * mddev)
1377 {
1378         int err;
1379         struct list_head *tmp;
1380         mdk_rdev_t *rdev;
1381         int sync_req;
1382
1383 repeat:
1384         spin_lock(&mddev->write_lock);
1385         sync_req = mddev->in_sync;
1386         mddev->utime = get_seconds();
1387         mddev->events ++;
1388
1389         if (!mddev->events) {
1390                 /*
1391                  * oops, this 64-bit counter should never wrap.
1392                  * Either we are in around ~1 trillion A.C., assuming
1393                  * 1 reboot per second, or we have a bug:
1394                  */
1395                 MD_BUG();
1396                 mddev->events --;
1397         }
1398         mddev->sb_dirty = 2;
1399         sync_sbs(mddev);
1400
1401         /*
1402          * do not write anything to disk if using
1403          * nonpersistent superblocks
1404          */
1405         if (!mddev->persistent) {
1406                 mddev->sb_dirty = 0;
1407                 spin_unlock(&mddev->write_lock);
1408                 wake_up(&mddev->sb_wait);
1409                 return;
1410         }
1411         spin_unlock(&mddev->write_lock);
1412
1413         dprintk(KERN_INFO
1414                 "md: updating %s RAID superblock on device (in sync %d)\n",
1415                 mdname(mddev),mddev->in_sync);
1416
1417         err = bitmap_update_sb(mddev->bitmap);
1418         ITERATE_RDEV(mddev,rdev,tmp) {
1419                 char b[BDEVNAME_SIZE];
1420                 dprintk(KERN_INFO "md: ");
1421                 if (rdev->faulty)
1422                         dprintk("(skipping faulty ");
1423
1424                 dprintk("%s ", bdevname(rdev->bdev,b));
1425                 if (!rdev->faulty) {
1426                         md_super_write(mddev,rdev,
1427                                        rdev->sb_offset<<1, rdev->sb_size,
1428                                        rdev->sb_page);
1429                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1430                                 bdevname(rdev->bdev,b),
1431                                 (unsigned long long)rdev->sb_offset);
1432
1433                 } else
1434                         dprintk(")\n");
1435                 if (mddev->level == LEVEL_MULTIPATH)
1436                         /* only need to write one superblock... */
1437                         break;
1438         }
1439         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1440         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1441
1442         spin_lock(&mddev->write_lock);
1443         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1444                 /* have to write it out again */
1445                 spin_unlock(&mddev->write_lock);
1446                 goto repeat;
1447         }
1448         mddev->sb_dirty = 0;
1449         spin_unlock(&mddev->write_lock);
1450         wake_up(&mddev->sb_wait);
1451
1452 }
1453
1454 struct rdev_sysfs_entry {
1455         struct attribute attr;
1456         ssize_t (*show)(mdk_rdev_t *, char *);
1457         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1458 };
1459
1460 static ssize_t
1461 rdev_show_state(mdk_rdev_t *rdev, char *page)
1462 {
1463         char *sep = "";
1464         int len=0;
1465
1466         if (rdev->faulty) {
1467                 len+= sprintf(page+len, "%sfaulty",sep);
1468                 sep = ",";
1469         }
1470         if (rdev->in_sync) {
1471                 len += sprintf(page+len, "%sin_sync",sep);
1472                 sep = ",";
1473         }
1474         if (!rdev->faulty && !rdev->in_sync) {
1475                 len += sprintf(page+len, "%sspare", sep);
1476                 sep = ",";
1477         }
1478         return len+sprintf(page+len, "\n");
1479 }
1480
1481 static struct rdev_sysfs_entry rdev_state = {
1482         .attr = {.name = "state", .mode = S_IRUGO },
1483         .show = rdev_show_state,
1484 };
1485
1486 static ssize_t
1487 rdev_show_super(mdk_rdev_t *rdev, char *page)
1488 {
1489         if (rdev->sb_loaded && rdev->sb_size) {
1490                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1491                 return rdev->sb_size;
1492         } else
1493                 return 0;
1494 }
1495 static struct rdev_sysfs_entry rdev_super = {
1496         .attr = {.name = "super", .mode = S_IRUGO },
1497         .show = rdev_show_super,
1498 };
1499 static struct attribute *rdev_default_attrs[] = {
1500         &rdev_state.attr,
1501         &rdev_super.attr,
1502         NULL,
1503 };
1504 static ssize_t
1505 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1506 {
1507         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1508         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1509
1510         if (!entry->show)
1511                 return -EIO;
1512         return entry->show(rdev, page);
1513 }
1514
1515 static ssize_t
1516 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1517               const char *page, size_t length)
1518 {
1519         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1520         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1521
1522         if (!entry->store)
1523                 return -EIO;
1524         return entry->store(rdev, page, length);
1525 }
1526
1527 static void rdev_free(struct kobject *ko)
1528 {
1529         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1530         kfree(rdev);
1531 }
1532 static struct sysfs_ops rdev_sysfs_ops = {
1533         .show           = rdev_attr_show,
1534         .store          = rdev_attr_store,
1535 };
1536 static struct kobj_type rdev_ktype = {
1537         .release        = rdev_free,
1538         .sysfs_ops      = &rdev_sysfs_ops,
1539         .default_attrs  = rdev_default_attrs,
1540 };
1541
1542 /*
1543  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1544  *
1545  * mark the device faulty if:
1546  *
1547  *   - the device is nonexistent (zero size)
1548  *   - the device has no valid superblock
1549  *
1550  * a faulty rdev _never_ has rdev->sb set.
1551  */
1552 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1553 {
1554         char b[BDEVNAME_SIZE];
1555         int err;
1556         mdk_rdev_t *rdev;
1557         sector_t size;
1558
1559         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1560         if (!rdev) {
1561                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1562                 return ERR_PTR(-ENOMEM);
1563         }
1564         memset(rdev, 0, sizeof(*rdev));
1565
1566         if ((err = alloc_disk_sb(rdev)))
1567                 goto abort_free;
1568
1569         err = lock_rdev(rdev, newdev);
1570         if (err)
1571                 goto abort_free;
1572
1573         rdev->kobj.parent = NULL;
1574         rdev->kobj.ktype = &rdev_ktype;
1575         kobject_init(&rdev->kobj);
1576
1577         rdev->desc_nr = -1;
1578         rdev->faulty = 0;
1579         rdev->in_sync = 0;
1580         rdev->data_offset = 0;
1581         atomic_set(&rdev->nr_pending, 0);
1582
1583         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1584         if (!size) {
1585                 printk(KERN_WARNING
1586                         "md: %s has zero or unknown size, marking faulty!\n",
1587                         bdevname(rdev->bdev,b));
1588                 err = -EINVAL;
1589                 goto abort_free;
1590         }
1591
1592         if (super_format >= 0) {
1593                 err = super_types[super_format].
1594                         load_super(rdev, NULL, super_minor);
1595                 if (err == -EINVAL) {
1596                         printk(KERN_WARNING
1597                                 "md: %s has invalid sb, not importing!\n",
1598                                 bdevname(rdev->bdev,b));
1599                         goto abort_free;
1600                 }
1601                 if (err < 0) {
1602                         printk(KERN_WARNING
1603                                 "md: could not read %s's sb, not importing!\n",
1604                                 bdevname(rdev->bdev,b));
1605                         goto abort_free;
1606                 }
1607         }
1608         INIT_LIST_HEAD(&rdev->same_set);
1609
1610         return rdev;
1611
1612 abort_free:
1613         if (rdev->sb_page) {
1614                 if (rdev->bdev)
1615                         unlock_rdev(rdev);
1616                 free_disk_sb(rdev);
1617         }
1618         kfree(rdev);
1619         return ERR_PTR(err);
1620 }
1621
1622 /*
1623  * Check a full RAID array for plausibility
1624  */
1625
1626
1627 static void analyze_sbs(mddev_t * mddev)
1628 {
1629         int i;
1630         struct list_head *tmp;
1631         mdk_rdev_t *rdev, *freshest;
1632         char b[BDEVNAME_SIZE];
1633
1634         freshest = NULL;
1635         ITERATE_RDEV(mddev,rdev,tmp)
1636                 switch (super_types[mddev->major_version].
1637                         load_super(rdev, freshest, mddev->minor_version)) {
1638                 case 1:
1639                         freshest = rdev;
1640                         break;
1641                 case 0:
1642                         break;
1643                 default:
1644                         printk( KERN_ERR \
1645                                 "md: fatal superblock inconsistency in %s"
1646                                 " -- removing from array\n",
1647                                 bdevname(rdev->bdev,b));
1648                         kick_rdev_from_array(rdev);
1649                 }
1650
1651
1652         super_types[mddev->major_version].
1653                 validate_super(mddev, freshest);
1654
1655         i = 0;
1656         ITERATE_RDEV(mddev,rdev,tmp) {
1657                 if (rdev != freshest)
1658                         if (super_types[mddev->major_version].
1659                             validate_super(mddev, rdev)) {
1660                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1661                                         " from array!\n",
1662                                         bdevname(rdev->bdev,b));
1663                                 kick_rdev_from_array(rdev);
1664                                 continue;
1665                         }
1666                 if (mddev->level == LEVEL_MULTIPATH) {
1667                         rdev->desc_nr = i++;
1668                         rdev->raid_disk = rdev->desc_nr;
1669                         rdev->in_sync = 1;
1670                 }
1671         }
1672
1673
1674
1675         if (mddev->recovery_cp != MaxSector &&
1676             mddev->level >= 1)
1677                 printk(KERN_ERR "md: %s: raid array is not clean"
1678                        " -- starting background reconstruction\n",
1679                        mdname(mddev));
1680
1681 }
1682
1683 struct md_sysfs_entry {
1684         struct attribute attr;
1685         ssize_t (*show)(mddev_t *, char *);
1686         ssize_t (*store)(mddev_t *, const char *, size_t);
1687 };
1688
1689 static ssize_t
1690 md_show_level(mddev_t *mddev, char *page)
1691 {
1692         mdk_personality_t *p = mddev->pers;
1693         if (p == NULL)
1694                 return 0;
1695         if (mddev->level >= 0)
1696                 return sprintf(page, "RAID-%d\n", mddev->level);
1697         else
1698                 return sprintf(page, "%s\n", p->name);
1699 }
1700
1701 static struct md_sysfs_entry md_level = {
1702         .attr = {.name = "level", .mode = S_IRUGO },
1703         .show = md_show_level,
1704 };
1705
1706 static ssize_t
1707 md_show_rdisks(mddev_t *mddev, char *page)
1708 {
1709         return sprintf(page, "%d\n", mddev->raid_disks);
1710 }
1711
1712 static struct md_sysfs_entry md_raid_disks = {
1713         .attr = {.name = "raid_disks", .mode = S_IRUGO },
1714         .show = md_show_rdisks,
1715 };
1716
1717 static ssize_t
1718 md_show_scan(mddev_t *mddev, char *page)
1719 {
1720         char *type = "none";
1721         if (mddev->recovery &
1722             ((1<<MD_RECOVERY_RUNNING) || (1<<MD_RECOVERY_NEEDED))) {
1723                 if (mddev->recovery & (1<<MD_RECOVERY_SYNC)) {
1724                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1725                                 type = "resync";
1726                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1727                                 type = "check";
1728                         else
1729                                 type = "repair";
1730                 } else
1731                         type = "recover";
1732         }
1733         return sprintf(page, "%s\n", type);
1734 }
1735
1736 static ssize_t
1737 md_store_scan(mddev_t *mddev, const char *page, size_t len)
1738 {
1739         int canscan=0;
1740         if (mddev->recovery &
1741             ((1<<MD_RECOVERY_RUNNING) || (1<<MD_RECOVERY_NEEDED)))
1742                 return -EBUSY;
1743         down(&mddev->reconfig_sem);
1744         if (mddev->pers && mddev->pers->sync_request)
1745                 canscan=1;
1746         up(&mddev->reconfig_sem);
1747         if (!canscan)
1748                 return -EINVAL;
1749
1750         if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1751                 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1752         else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1753                 return -EINVAL;
1754         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1755         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1756         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1757         md_wakeup_thread(mddev->thread);
1758         return len;
1759 }
1760
1761 static ssize_t
1762 md_show_mismatch(mddev_t *mddev, char *page)
1763 {
1764         return sprintf(page, "%llu\n",
1765                        (unsigned long long) mddev->resync_mismatches);
1766 }
1767
1768 static struct md_sysfs_entry md_scan_mode = {
1769         .attr = {.name = "scan_mode", .mode = S_IRUGO|S_IWUSR },
1770         .show = md_show_scan,
1771         .store = md_store_scan,
1772 };
1773
1774 static struct md_sysfs_entry md_mismatches = {
1775         .attr = {.name = "mismatch_cnt", .mode = S_IRUGO },
1776         .show = md_show_mismatch,
1777 };
1778
1779 static struct attribute *md_default_attrs[] = {
1780         &md_level.attr,
1781         &md_raid_disks.attr,
1782         &md_scan_mode.attr,
1783         &md_mismatches.attr,
1784         NULL,
1785 };
1786
1787 static ssize_t
1788 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1789 {
1790         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1791         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1792
1793         if (!entry->show)
1794                 return -EIO;
1795         return entry->show(mddev, page);
1796 }
1797
1798 static ssize_t
1799 md_attr_store(struct kobject *kobj, struct attribute *attr,
1800               const char *page, size_t length)
1801 {
1802         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1803         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1804
1805         if (!entry->store)
1806                 return -EIO;
1807         return entry->store(mddev, page, length);
1808 }
1809
1810 static void md_free(struct kobject *ko)
1811 {
1812         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1813         kfree(mddev);
1814 }
1815
1816 static struct sysfs_ops md_sysfs_ops = {
1817         .show   = md_attr_show,
1818         .store  = md_attr_store,
1819 };
1820 static struct kobj_type md_ktype = {
1821         .release        = md_free,
1822         .sysfs_ops      = &md_sysfs_ops,
1823         .default_attrs  = md_default_attrs,
1824 };
1825
1826 int mdp_major = 0;
1827
1828 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1829 {
1830         static DECLARE_MUTEX(disks_sem);
1831         mddev_t *mddev = mddev_find(dev);
1832         struct gendisk *disk;
1833         int partitioned = (MAJOR(dev) != MD_MAJOR);
1834         int shift = partitioned ? MdpMinorShift : 0;
1835         int unit = MINOR(dev) >> shift;
1836
1837         if (!mddev)
1838                 return NULL;
1839
1840         down(&disks_sem);
1841         if (mddev->gendisk) {
1842                 up(&disks_sem);
1843                 mddev_put(mddev);
1844                 return NULL;
1845         }
1846         disk = alloc_disk(1 << shift);
1847         if (!disk) {
1848                 up(&disks_sem);
1849                 mddev_put(mddev);
1850                 return NULL;
1851         }
1852         disk->major = MAJOR(dev);
1853         disk->first_minor = unit << shift;
1854         if (partitioned) {
1855                 sprintf(disk->disk_name, "md_d%d", unit);
1856                 sprintf(disk->devfs_name, "md/d%d", unit);
1857         } else {
1858                 sprintf(disk->disk_name, "md%d", unit);
1859                 sprintf(disk->devfs_name, "md/%d", unit);
1860         }
1861         disk->fops = &md_fops;
1862         disk->private_data = mddev;
1863         disk->queue = mddev->queue;
1864         add_disk(disk);
1865         mddev->gendisk = disk;
1866         up(&disks_sem);
1867         mddev->kobj.parent = kobject_get(&disk->kobj);
1868         mddev->kobj.k_name = NULL;
1869         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1870         mddev->kobj.ktype = &md_ktype;
1871         kobject_register(&mddev->kobj);
1872         return NULL;
1873 }
1874
1875 void md_wakeup_thread(mdk_thread_t *thread);
1876
1877 static void md_safemode_timeout(unsigned long data)
1878 {
1879         mddev_t *mddev = (mddev_t *) data;
1880
1881         mddev->safemode = 1;
1882         md_wakeup_thread(mddev->thread);
1883 }
1884
1885
1886 static int do_md_run(mddev_t * mddev)
1887 {
1888         int pnum, err;
1889         int chunk_size;
1890         struct list_head *tmp;
1891         mdk_rdev_t *rdev;
1892         struct gendisk *disk;
1893         char b[BDEVNAME_SIZE];
1894
1895         if (list_empty(&mddev->disks))
1896                 /* cannot run an array with no devices.. */
1897                 return -EINVAL;
1898
1899         if (mddev->pers)
1900                 return -EBUSY;
1901
1902         /*
1903          * Analyze all RAID superblock(s)
1904          */
1905         if (!mddev->raid_disks)
1906                 analyze_sbs(mddev);
1907
1908         chunk_size = mddev->chunk_size;
1909         pnum = level_to_pers(mddev->level);
1910
1911         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1912                 if (!chunk_size) {
1913                         /*
1914                          * 'default chunksize' in the old md code used to
1915                          * be PAGE_SIZE, baaad.
1916                          * we abort here to be on the safe side. We don't
1917                          * want to continue the bad practice.
1918                          */
1919                         printk(KERN_ERR
1920                                 "no chunksize specified, see 'man raidtab'\n");
1921                         return -EINVAL;
1922                 }
1923                 if (chunk_size > MAX_CHUNK_SIZE) {
1924                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1925                                 chunk_size, MAX_CHUNK_SIZE);
1926                         return -EINVAL;
1927                 }
1928                 /*
1929                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1930                  */
1931                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1932                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1933                         return -EINVAL;
1934                 }
1935                 if (chunk_size < PAGE_SIZE) {
1936                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1937                                 chunk_size, PAGE_SIZE);
1938                         return -EINVAL;
1939                 }
1940
1941                 /* devices must have minimum size of one chunk */
1942                 ITERATE_RDEV(mddev,rdev,tmp) {
1943                         if (rdev->faulty)
1944                                 continue;
1945                         if (rdev->size < chunk_size / 1024) {
1946                                 printk(KERN_WARNING
1947                                         "md: Dev %s smaller than chunk_size:"
1948                                         " %lluk < %dk\n",
1949                                         bdevname(rdev->bdev,b),
1950                                         (unsigned long long)rdev->size,
1951                                         chunk_size / 1024);
1952                                 return -EINVAL;
1953                         }
1954                 }
1955         }
1956
1957 #ifdef CONFIG_KMOD
1958         if (!pers[pnum])
1959         {
1960                 request_module("md-personality-%d", pnum);
1961         }
1962 #endif
1963
1964         /*
1965          * Drop all container device buffers, from now on
1966          * the only valid external interface is through the md
1967          * device.
1968          * Also find largest hardsector size
1969          */
1970         ITERATE_RDEV(mddev,rdev,tmp) {
1971                 if (rdev->faulty)
1972                         continue;
1973                 sync_blockdev(rdev->bdev);
1974                 invalidate_bdev(rdev->bdev, 0);
1975         }
1976
1977         md_probe(mddev->unit, NULL, NULL);
1978         disk = mddev->gendisk;
1979         if (!disk)
1980                 return -ENOMEM;
1981
1982         spin_lock(&pers_lock);
1983         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1984                 spin_unlock(&pers_lock);
1985                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1986                        pnum);
1987                 return -EINVAL;
1988         }
1989
1990         mddev->pers = pers[pnum];
1991         spin_unlock(&pers_lock);
1992
1993         mddev->recovery = 0;
1994         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1995
1996         /* before we start the array running, initialise the bitmap */
1997         err = bitmap_create(mddev);
1998         if (err)
1999                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2000                         mdname(mddev), err);
2001         else
2002                 err = mddev->pers->run(mddev);
2003         if (err) {
2004                 printk(KERN_ERR "md: pers->run() failed ...\n");
2005                 module_put(mddev->pers->owner);
2006                 mddev->pers = NULL;
2007                 bitmap_destroy(mddev);
2008                 return err;
2009         }
2010         atomic_set(&mddev->writes_pending,0);
2011         mddev->safemode = 0;
2012         mddev->safemode_timer.function = md_safemode_timeout;
2013         mddev->safemode_timer.data = (unsigned long) mddev;
2014         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2015         mddev->in_sync = 1;
2016
2017         ITERATE_RDEV(mddev,rdev,tmp)
2018                 if (rdev->raid_disk >= 0) {
2019                         char nm[20];
2020                         sprintf(nm, "rd%d", rdev->raid_disk);
2021                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2022                 }
2023
2024         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2025         md_wakeup_thread(mddev->thread);
2026
2027         if (mddev->sb_dirty)
2028                 md_update_sb(mddev);
2029
2030         set_capacity(disk, mddev->array_size<<1);
2031
2032         /* If we call blk_queue_make_request here, it will
2033          * re-initialise max_sectors etc which may have been
2034          * refined inside -> run.  So just set the bits we need to set.
2035          * Most initialisation happended when we called
2036          * blk_queue_make_request(..., md_fail_request)
2037          * earlier.
2038          */
2039         mddev->queue->queuedata = mddev;
2040         mddev->queue->make_request_fn = mddev->pers->make_request;
2041
2042         mddev->changed = 1;
2043         return 0;
2044 }
2045
2046 static int restart_array(mddev_t *mddev)
2047 {
2048         struct gendisk *disk = mddev->gendisk;
2049         int err;
2050
2051         /*
2052          * Complain if it has no devices
2053          */
2054         err = -ENXIO;
2055         if (list_empty(&mddev->disks))
2056                 goto out;
2057
2058         if (mddev->pers) {
2059                 err = -EBUSY;
2060                 if (!mddev->ro)
2061                         goto out;
2062
2063                 mddev->safemode = 0;
2064                 mddev->ro = 0;
2065                 set_disk_ro(disk, 0);
2066
2067                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2068                         mdname(mddev));
2069                 /*
2070                  * Kick recovery or resync if necessary
2071                  */
2072                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2073                 md_wakeup_thread(mddev->thread);
2074                 err = 0;
2075         } else {
2076                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2077                         mdname(mddev));
2078                 err = -EINVAL;
2079         }
2080
2081 out:
2082         return err;
2083 }
2084
2085 static int do_md_stop(mddev_t * mddev, int ro)
2086 {
2087         int err = 0;
2088         struct gendisk *disk = mddev->gendisk;
2089
2090         if (mddev->pers) {
2091                 if (atomic_read(&mddev->active)>2) {
2092                         printk("md: %s still in use.\n",mdname(mddev));
2093                         return -EBUSY;
2094                 }
2095
2096                 if (mddev->sync_thread) {
2097                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2098                         md_unregister_thread(mddev->sync_thread);
2099                         mddev->sync_thread = NULL;
2100                 }
2101
2102                 del_timer_sync(&mddev->safemode_timer);
2103
2104                 invalidate_partition(disk, 0);
2105
2106                 if (ro) {
2107                         err  = -ENXIO;
2108                         if (mddev->ro)
2109                                 goto out;
2110                         mddev->ro = 1;
2111                 } else {
2112                         bitmap_flush(mddev);
2113                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
2114                         if (mddev->ro)
2115                                 set_disk_ro(disk, 0);
2116                         blk_queue_make_request(mddev->queue, md_fail_request);
2117                         mddev->pers->stop(mddev);
2118                         module_put(mddev->pers->owner);
2119                         mddev->pers = NULL;
2120                         if (mddev->ro)
2121                                 mddev->ro = 0;
2122                 }
2123                 if (!mddev->in_sync) {
2124                         /* mark array as shutdown cleanly */
2125                         mddev->in_sync = 1;
2126                         md_update_sb(mddev);
2127                 }
2128                 if (ro)
2129                         set_disk_ro(disk, 1);
2130         }
2131
2132         bitmap_destroy(mddev);
2133         if (mddev->bitmap_file) {
2134                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2135                 fput(mddev->bitmap_file);
2136                 mddev->bitmap_file = NULL;
2137         }
2138         mddev->bitmap_offset = 0;
2139
2140         /*
2141          * Free resources if final stop
2142          */
2143         if (!ro) {
2144                 mdk_rdev_t *rdev;
2145                 struct list_head *tmp;
2146                 struct gendisk *disk;
2147                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2148
2149                 ITERATE_RDEV(mddev,rdev,tmp)
2150                         if (rdev->raid_disk >= 0) {
2151                                 char nm[20];
2152                                 sprintf(nm, "rd%d", rdev->raid_disk);
2153                                 sysfs_remove_link(&mddev->kobj, nm);
2154                         }
2155
2156                 export_array(mddev);
2157
2158                 mddev->array_size = 0;
2159                 disk = mddev->gendisk;
2160                 if (disk)
2161                         set_capacity(disk, 0);
2162                 mddev->changed = 1;
2163         } else
2164                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2165                         mdname(mddev));
2166         err = 0;
2167 out:
2168         return err;
2169 }
2170
2171 static void autorun_array(mddev_t *mddev)
2172 {
2173         mdk_rdev_t *rdev;
2174         struct list_head *tmp;
2175         int err;
2176
2177         if (list_empty(&mddev->disks))
2178                 return;
2179
2180         printk(KERN_INFO "md: running: ");
2181
2182         ITERATE_RDEV(mddev,rdev,tmp) {
2183                 char b[BDEVNAME_SIZE];
2184                 printk("<%s>", bdevname(rdev->bdev,b));
2185         }
2186         printk("\n");
2187
2188         err = do_md_run (mddev);
2189         if (err) {
2190                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2191                 do_md_stop (mddev, 0);
2192         }
2193 }
2194
2195 /*
2196  * lets try to run arrays based on all disks that have arrived
2197  * until now. (those are in pending_raid_disks)
2198  *
2199  * the method: pick the first pending disk, collect all disks with
2200  * the same UUID, remove all from the pending list and put them into
2201  * the 'same_array' list. Then order this list based on superblock
2202  * update time (freshest comes first), kick out 'old' disks and
2203  * compare superblocks. If everything's fine then run it.
2204  *
2205  * If "unit" is allocated, then bump its reference count
2206  */
2207 static void autorun_devices(int part)
2208 {
2209         struct list_head candidates;
2210         struct list_head *tmp;
2211         mdk_rdev_t *rdev0, *rdev;
2212         mddev_t *mddev;
2213         char b[BDEVNAME_SIZE];
2214
2215         printk(KERN_INFO "md: autorun ...\n");
2216         while (!list_empty(&pending_raid_disks)) {
2217                 dev_t dev;
2218                 rdev0 = list_entry(pending_raid_disks.next,
2219                                          mdk_rdev_t, same_set);
2220
2221                 printk(KERN_INFO "md: considering %s ...\n",
2222                         bdevname(rdev0->bdev,b));
2223                 INIT_LIST_HEAD(&candidates);
2224                 ITERATE_RDEV_PENDING(rdev,tmp)
2225                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2226                                 printk(KERN_INFO "md:  adding %s ...\n",
2227                                         bdevname(rdev->bdev,b));
2228                                 list_move(&rdev->same_set, &candidates);
2229                         }
2230                 /*
2231                  * now we have a set of devices, with all of them having
2232                  * mostly sane superblocks. It's time to allocate the
2233                  * mddev.
2234                  */
2235                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2236                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2237                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2238                         break;
2239                 }
2240                 if (part)
2241                         dev = MKDEV(mdp_major,
2242                                     rdev0->preferred_minor << MdpMinorShift);
2243                 else
2244                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2245
2246                 md_probe(dev, NULL, NULL);
2247                 mddev = mddev_find(dev);
2248                 if (!mddev) {
2249                         printk(KERN_ERR
2250                                 "md: cannot allocate memory for md drive.\n");
2251                         break;
2252                 }
2253                 if (mddev_lock(mddev))
2254                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2255                                mdname(mddev));
2256                 else if (mddev->raid_disks || mddev->major_version
2257                          || !list_empty(&mddev->disks)) {
2258                         printk(KERN_WARNING
2259                                 "md: %s already running, cannot run %s\n",
2260                                 mdname(mddev), bdevname(rdev0->bdev,b));
2261                         mddev_unlock(mddev);
2262                 } else {
2263                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2264                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2265                                 list_del_init(&rdev->same_set);
2266                                 if (bind_rdev_to_array(rdev, mddev))
2267                                         export_rdev(rdev);
2268                         }
2269                         autorun_array(mddev);
2270                         mddev_unlock(mddev);
2271                 }
2272                 /* on success, candidates will be empty, on error
2273                  * it won't...
2274                  */
2275                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2276                         export_rdev(rdev);
2277                 mddev_put(mddev);
2278         }
2279         printk(KERN_INFO "md: ... autorun DONE.\n");
2280 }
2281
2282 /*
2283  * import RAID devices based on one partition
2284  * if possible, the array gets run as well.
2285  */
2286
2287 static int autostart_array(dev_t startdev)
2288 {
2289         char b[BDEVNAME_SIZE];
2290         int err = -EINVAL, i;
2291         mdp_super_t *sb = NULL;
2292         mdk_rdev_t *start_rdev = NULL, *rdev;
2293
2294         start_rdev = md_import_device(startdev, 0, 0);
2295         if (IS_ERR(start_rdev))
2296                 return err;
2297
2298
2299         /* NOTE: this can only work for 0.90.0 superblocks */
2300         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2301         if (sb->major_version != 0 ||
2302             sb->minor_version != 90 ) {
2303                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2304                 export_rdev(start_rdev);
2305                 return err;
2306         }
2307
2308         if (start_rdev->faulty) {
2309                 printk(KERN_WARNING
2310                         "md: can not autostart based on faulty %s!\n",
2311                         bdevname(start_rdev->bdev,b));
2312                 export_rdev(start_rdev);
2313                 return err;
2314         }
2315         list_add(&start_rdev->same_set, &pending_raid_disks);
2316
2317         for (i = 0; i < MD_SB_DISKS; i++) {
2318                 mdp_disk_t *desc = sb->disks + i;
2319                 dev_t dev = MKDEV(desc->major, desc->minor);
2320
2321                 if (!dev)
2322                         continue;
2323                 if (dev == startdev)
2324                         continue;
2325                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2326                         continue;
2327                 rdev = md_import_device(dev, 0, 0);
2328                 if (IS_ERR(rdev))
2329                         continue;
2330
2331                 list_add(&rdev->same_set, &pending_raid_disks);
2332         }
2333
2334         /*
2335          * possibly return codes
2336          */
2337         autorun_devices(0);
2338         return 0;
2339
2340 }
2341
2342
2343 static int get_version(void __user * arg)
2344 {
2345         mdu_version_t ver;
2346
2347         ver.major = MD_MAJOR_VERSION;
2348         ver.minor = MD_MINOR_VERSION;
2349         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2350
2351         if (copy_to_user(arg, &ver, sizeof(ver)))
2352                 return -EFAULT;
2353
2354         return 0;
2355 }
2356
2357 static int get_array_info(mddev_t * mddev, void __user * arg)
2358 {
2359         mdu_array_info_t info;
2360         int nr,working,active,failed,spare;
2361         mdk_rdev_t *rdev;
2362         struct list_head *tmp;
2363
2364         nr=working=active=failed=spare=0;
2365         ITERATE_RDEV(mddev,rdev,tmp) {
2366                 nr++;
2367                 if (rdev->faulty)
2368                         failed++;
2369                 else {
2370                         working++;
2371                         if (rdev->in_sync)
2372                                 active++;
2373                         else
2374                                 spare++;
2375                 }
2376         }
2377
2378         info.major_version = mddev->major_version;
2379         info.minor_version = mddev->minor_version;
2380         info.patch_version = MD_PATCHLEVEL_VERSION;
2381         info.ctime         = mddev->ctime;
2382         info.level         = mddev->level;
2383         info.size          = mddev->size;
2384         info.nr_disks      = nr;
2385         info.raid_disks    = mddev->raid_disks;
2386         info.md_minor      = mddev->md_minor;
2387         info.not_persistent= !mddev->persistent;
2388
2389         info.utime         = mddev->utime;
2390         info.state         = 0;
2391         if (mddev->in_sync)
2392                 info.state = (1<<MD_SB_CLEAN);
2393         if (mddev->bitmap && mddev->bitmap_offset)
2394                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2395         info.active_disks  = active;
2396         info.working_disks = working;
2397         info.failed_disks  = failed;
2398         info.spare_disks   = spare;
2399
2400         info.layout        = mddev->layout;
2401         info.chunk_size    = mddev->chunk_size;
2402
2403         if (copy_to_user(arg, &info, sizeof(info)))
2404                 return -EFAULT;
2405
2406         return 0;
2407 }
2408
2409 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2410 {
2411         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2412         char *ptr, *buf = NULL;
2413         int err = -ENOMEM;
2414
2415         file = kmalloc(sizeof(*file), GFP_KERNEL);
2416         if (!file)
2417                 goto out;
2418
2419         /* bitmap disabled, zero the first byte and copy out */
2420         if (!mddev->bitmap || !mddev->bitmap->file) {
2421                 file->pathname[0] = '\0';
2422                 goto copy_out;
2423         }
2424
2425         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2426         if (!buf)
2427                 goto out;
2428
2429         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2430         if (!ptr)
2431                 goto out;
2432
2433         strcpy(file->pathname, ptr);
2434
2435 copy_out:
2436         err = 0;
2437         if (copy_to_user(arg, file, sizeof(*file)))
2438                 err = -EFAULT;
2439 out:
2440         kfree(buf);
2441         kfree(file);
2442         return err;
2443 }
2444
2445 static int get_disk_info(mddev_t * mddev, void __user * arg)
2446 {
2447         mdu_disk_info_t info;
2448         unsigned int nr;
2449         mdk_rdev_t *rdev;
2450
2451         if (copy_from_user(&info, arg, sizeof(info)))
2452                 return -EFAULT;
2453
2454         nr = info.number;
2455
2456         rdev = find_rdev_nr(mddev, nr);
2457         if (rdev) {
2458                 info.major = MAJOR(rdev->bdev->bd_dev);
2459                 info.minor = MINOR(rdev->bdev->bd_dev);
2460                 info.raid_disk = rdev->raid_disk;
2461                 info.state = 0;
2462                 if (rdev->faulty)
2463                         info.state |= (1<<MD_DISK_FAULTY);
2464                 else if (rdev->in_sync) {
2465                         info.state |= (1<<MD_DISK_ACTIVE);
2466                         info.state |= (1<<MD_DISK_SYNC);
2467                 }
2468                 if (test_bit(WriteMostly, &rdev->flags))
2469                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2470         } else {
2471                 info.major = info.minor = 0;
2472                 info.raid_disk = -1;
2473                 info.state = (1<<MD_DISK_REMOVED);
2474         }
2475
2476         if (copy_to_user(arg, &info, sizeof(info)))
2477                 return -EFAULT;
2478
2479         return 0;
2480 }
2481
2482 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2483 {
2484         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2485         mdk_rdev_t *rdev;
2486         dev_t dev = MKDEV(info->major,info->minor);
2487
2488         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2489                 return -EOVERFLOW;
2490
2491         if (!mddev->raid_disks) {
2492                 int err;
2493                 /* expecting a device which has a superblock */
2494                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2495                 if (IS_ERR(rdev)) {
2496                         printk(KERN_WARNING
2497                                 "md: md_import_device returned %ld\n",
2498                                 PTR_ERR(rdev));
2499                         return PTR_ERR(rdev);
2500                 }
2501                 if (!list_empty(&mddev->disks)) {
2502                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2503                                                         mdk_rdev_t, same_set);
2504                         int err = super_types[mddev->major_version]
2505                                 .load_super(rdev, rdev0, mddev->minor_version);
2506                         if (err < 0) {
2507                                 printk(KERN_WARNING
2508                                         "md: %s has different UUID to %s\n",
2509                                         bdevname(rdev->bdev,b),
2510                                         bdevname(rdev0->bdev,b2));
2511                                 export_rdev(rdev);
2512                                 return -EINVAL;
2513                         }
2514                 }
2515                 err = bind_rdev_to_array(rdev, mddev);
2516                 if (err)
2517                         export_rdev(rdev);
2518                 return err;
2519         }
2520
2521         /*
2522          * add_new_disk can be used once the array is assembled
2523          * to add "hot spares".  They must already have a superblock
2524          * written
2525          */
2526         if (mddev->pers) {
2527                 int err;
2528                 if (!mddev->pers->hot_add_disk) {
2529                         printk(KERN_WARNING
2530                                 "%s: personality does not support diskops!\n",
2531                                mdname(mddev));
2532                         return -EINVAL;
2533                 }
2534                 if (mddev->persistent)
2535                         rdev = md_import_device(dev, mddev->major_version,
2536                                                 mddev->minor_version);
2537                 else
2538                         rdev = md_import_device(dev, -1, -1);
2539                 if (IS_ERR(rdev)) {
2540                         printk(KERN_WARNING
2541                                 "md: md_import_device returned %ld\n",
2542                                 PTR_ERR(rdev));
2543                         return PTR_ERR(rdev);
2544                 }
2545                 /* set save_raid_disk if appropriate */
2546                 if (!mddev->persistent) {
2547                         if (info->state & (1<<MD_DISK_SYNC)  &&
2548                             info->raid_disk < mddev->raid_disks)
2549                                 rdev->raid_disk = info->raid_disk;
2550                         else
2551                                 rdev->raid_disk = -1;
2552                 } else
2553                         super_types[mddev->major_version].
2554                                 validate_super(mddev, rdev);
2555                 rdev->saved_raid_disk = rdev->raid_disk;
2556
2557                 rdev->in_sync = 0; /* just to be sure */
2558                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2559                         set_bit(WriteMostly, &rdev->flags);
2560
2561                 rdev->raid_disk = -1;
2562                 err = bind_rdev_to_array(rdev, mddev);
2563                 if (err)
2564                         export_rdev(rdev);
2565
2566                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2567                 md_wakeup_thread(mddev->thread);
2568                 return err;
2569         }
2570
2571         /* otherwise, add_new_disk is only allowed
2572          * for major_version==0 superblocks
2573          */
2574         if (mddev->major_version != 0) {
2575                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2576                        mdname(mddev));
2577                 return -EINVAL;
2578         }
2579
2580         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2581                 int err;
2582                 rdev = md_import_device (dev, -1, 0);
2583                 if (IS_ERR(rdev)) {
2584                         printk(KERN_WARNING
2585                                 "md: error, md_import_device() returned %ld\n",
2586                                 PTR_ERR(rdev));
2587                         return PTR_ERR(rdev);
2588                 }
2589                 rdev->desc_nr = info->number;
2590                 if (info->raid_disk < mddev->raid_disks)
2591                         rdev->raid_disk = info->raid_disk;
2592                 else
2593                         rdev->raid_disk = -1;
2594
2595                 rdev->faulty = 0;
2596                 if (rdev->raid_disk < mddev->raid_disks)
2597                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2598                 else
2599                         rdev->in_sync = 0;
2600
2601                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2602                         set_bit(WriteMostly, &rdev->flags);
2603
2604                 err = bind_rdev_to_array(rdev, mddev);
2605                 if (err) {
2606                         export_rdev(rdev);
2607                         return err;
2608                 }
2609
2610                 if (!mddev->persistent) {
2611                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2612                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2613                 } else
2614                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2615                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2616
2617                 if (!mddev->size || (mddev->size > rdev->size))
2618                         mddev->size = rdev->size;
2619         }
2620
2621         return 0;
2622 }
2623
2624 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2625 {
2626         char b[BDEVNAME_SIZE];
2627         mdk_rdev_t *rdev;
2628
2629         if (!mddev->pers)
2630                 return -ENODEV;
2631
2632         rdev = find_rdev(mddev, dev);
2633         if (!rdev)
2634                 return -ENXIO;
2635
2636         if (rdev->raid_disk >= 0)
2637                 goto busy;
2638
2639         kick_rdev_from_array(rdev);
2640         md_update_sb(mddev);
2641
2642         return 0;
2643 busy:
2644         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2645                 bdevname(rdev->bdev,b), mdname(mddev));
2646         return -EBUSY;
2647 }
2648
2649 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2650 {
2651         char b[BDEVNAME_SIZE];
2652         int err;
2653         unsigned int size;
2654         mdk_rdev_t *rdev;
2655
2656         if (!mddev->pers)
2657                 return -ENODEV;
2658
2659         if (mddev->major_version != 0) {
2660                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2661                         " version-0 superblocks.\n",
2662                         mdname(mddev));
2663                 return -EINVAL;
2664         }
2665         if (!mddev->pers->hot_add_disk) {
2666                 printk(KERN_WARNING
2667                         "%s: personality does not support diskops!\n",
2668                         mdname(mddev));
2669                 return -EINVAL;
2670         }
2671
2672         rdev = md_import_device (dev, -1, 0);
2673         if (IS_ERR(rdev)) {
2674                 printk(KERN_WARNING
2675                         "md: error, md_import_device() returned %ld\n",
2676                         PTR_ERR(rdev));
2677                 return -EINVAL;
2678         }
2679
2680         if (mddev->persistent)
2681                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2682         else
2683                 rdev->sb_offset =
2684                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2685
2686         size = calc_dev_size(rdev, mddev->chunk_size);
2687         rdev->size = size;
2688
2689         if (size < mddev->size) {
2690                 printk(KERN_WARNING
2691                         "%s: disk size %llu blocks < array size %llu\n",
2692                         mdname(mddev), (unsigned long long)size,
2693                         (unsigned long long)mddev->size);
2694                 err = -ENOSPC;
2695                 goto abort_export;
2696         }
2697
2698         if (rdev->faulty) {
2699                 printk(KERN_WARNING
2700                         "md: can not hot-add faulty %s disk to %s!\n",
2701                         bdevname(rdev->bdev,b), mdname(mddev));
2702                 err = -EINVAL;
2703                 goto abort_export;
2704         }
2705         rdev->in_sync = 0;
2706         rdev->desc_nr = -1;
2707         bind_rdev_to_array(rdev, mddev);
2708
2709         /*
2710          * The rest should better be atomic, we can have disk failures
2711          * noticed in interrupt contexts ...
2712          */
2713
2714         if (rdev->desc_nr == mddev->max_disks) {
2715                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2716                         mdname(mddev));
2717                 err = -EBUSY;
2718                 goto abort_unbind_export;
2719         }
2720
2721         rdev->raid_disk = -1;
2722
2723         md_update_sb(mddev);
2724
2725         /*
2726          * Kick recovery, maybe this spare has to be added to the
2727          * array immediately.
2728          */
2729         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2730         md_wakeup_thread(mddev->thread);
2731
2732         return 0;
2733
2734 abort_unbind_export:
2735         unbind_rdev_from_array(rdev);
2736
2737 abort_export:
2738         export_rdev(rdev);
2739         return err;
2740 }
2741
2742 /* similar to deny_write_access, but accounts for our holding a reference
2743  * to the file ourselves */
2744 static int deny_bitmap_write_access(struct file * file)
2745 {
2746         struct inode *inode = file->f_mapping->host;
2747
2748         spin_lock(&inode->i_lock);
2749         if (atomic_read(&inode->i_writecount) > 1) {
2750                 spin_unlock(&inode->i_lock);
2751                 return -ETXTBSY;
2752         }
2753         atomic_set(&inode->i_writecount, -1);
2754         spin_unlock(&inode->i_lock);
2755
2756         return 0;
2757 }
2758
2759 static int set_bitmap_file(mddev_t *mddev, int fd)
2760 {
2761         int err;
2762
2763         if (mddev->pers) {
2764                 if (!mddev->pers->quiesce)
2765                         return -EBUSY;
2766                 if (mddev->recovery || mddev->sync_thread)
2767                         return -EBUSY;
2768                 /* we should be able to change the bitmap.. */
2769         }
2770
2771
2772         if (fd >= 0) {
2773                 if (mddev->bitmap)
2774                         return -EEXIST; /* cannot add when bitmap is present */
2775                 mddev->bitmap_file = fget(fd);
2776
2777                 if (mddev->bitmap_file == NULL) {
2778                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2779                                mdname(mddev));
2780                         return -EBADF;
2781                 }
2782
2783                 err = deny_bitmap_write_access(mddev->bitmap_file);
2784                 if (err) {
2785                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2786                                mdname(mddev));
2787                         fput(mddev->bitmap_file);
2788                         mddev->bitmap_file = NULL;
2789                         return err;
2790                 }
2791                 mddev->bitmap_offset = 0; /* file overrides offset */
2792         } else if (mddev->bitmap == NULL)
2793                 return -ENOENT; /* cannot remove what isn't there */
2794         err = 0;
2795         if (mddev->pers) {
2796                 mddev->pers->quiesce(mddev, 1);
2797                 if (fd >= 0)
2798                         err = bitmap_create(mddev);
2799                 if (fd < 0 || err)
2800                         bitmap_destroy(mddev);
2801                 mddev->pers->quiesce(mddev, 0);
2802         } else if (fd < 0) {
2803                 if (mddev->bitmap_file)
2804                         fput(mddev->bitmap_file);
2805                 mddev->bitmap_file = NULL;
2806         }
2807
2808         return err;
2809 }
2810
2811 /*
2812  * set_array_info is used two different ways
2813  * The original usage is when creating a new array.
2814  * In this usage, raid_disks is > 0 and it together with
2815  *  level, size, not_persistent,layout,chunksize determine the
2816  *  shape of the array.
2817  *  This will always create an array with a type-0.90.0 superblock.
2818  * The newer usage is when assembling an array.
2819  *  In this case raid_disks will be 0, and the major_version field is
2820  *  use to determine which style super-blocks are to be found on the devices.
2821  *  The minor and patch _version numbers are also kept incase the
2822  *  super_block handler wishes to interpret them.
2823  */
2824 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2825 {
2826
2827         if (info->raid_disks == 0) {
2828                 /* just setting version number for superblock loading */
2829                 if (info->major_version < 0 ||
2830                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2831                     super_types[info->major_version].name == NULL) {
2832                         /* maybe try to auto-load a module? */
2833                         printk(KERN_INFO
2834                                 "md: superblock version %d not known\n",
2835                                 info->major_version);
2836                         return -EINVAL;
2837                 }
2838                 mddev->major_version = info->major_version;
2839                 mddev->minor_version = info->minor_version;
2840                 mddev->patch_version = info->patch_version;
2841                 return 0;
2842         }
2843         mddev->major_version = MD_MAJOR_VERSION;
2844         mddev->minor_version = MD_MINOR_VERSION;
2845         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2846         mddev->ctime         = get_seconds();
2847
2848         mddev->level         = info->level;
2849         mddev->size          = info->size;
2850         mddev->raid_disks    = info->raid_disks;
2851         /* don't set md_minor, it is determined by which /dev/md* was
2852          * openned
2853          */
2854         if (info->state & (1<<MD_SB_CLEAN))
2855                 mddev->recovery_cp = MaxSector;
2856         else
2857                 mddev->recovery_cp = 0;
2858         mddev->persistent    = ! info->not_persistent;
2859
2860         mddev->layout        = info->layout;
2861         mddev->chunk_size    = info->chunk_size;
2862
2863         mddev->max_disks     = MD_SB_DISKS;
2864
2865         mddev->sb_dirty      = 1;
2866
2867         /*
2868          * Generate a 128 bit UUID
2869          */
2870         get_random_bytes(mddev->uuid, 16);
2871
2872         return 0;
2873 }
2874
2875 /*
2876  * update_array_info is used to change the configuration of an
2877  * on-line array.
2878  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2879  * fields in the info are checked against the array.
2880  * Any differences that cannot be handled will cause an error.
2881  * Normally, only one change can be managed at a time.
2882  */
2883 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2884 {
2885         int rv = 0;
2886         int cnt = 0;
2887         int state = 0;
2888
2889         /* calculate expected state,ignoring low bits */
2890         if (mddev->bitmap && mddev->bitmap_offset)
2891                 state |= (1 << MD_SB_BITMAP_PRESENT);
2892
2893         if (mddev->major_version != info->major_version ||
2894             mddev->minor_version != info->minor_version ||
2895 /*          mddev->patch_version != info->patch_version || */
2896             mddev->ctime         != info->ctime         ||
2897             mddev->level         != info->level         ||
2898 /*          mddev->layout        != info->layout        || */
2899             !mddev->persistent   != info->not_persistent||
2900             mddev->chunk_size    != info->chunk_size    ||
2901             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2902             ((state^info->state) & 0xfffffe00)
2903                 )
2904                 return -EINVAL;
2905         /* Check there is only one change */
2906         if (mddev->size != info->size) cnt++;
2907         if (mddev->raid_disks != info->raid_disks) cnt++;
2908         if (mddev->layout != info->layout) cnt++;
2909         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2910         if (cnt == 0) return 0;
2911         if (cnt > 1) return -EINVAL;
2912
2913         if (mddev->layout != info->layout) {
2914                 /* Change layout
2915                  * we don't need to do anything at the md level, the
2916                  * personality will take care of it all.
2917                  */
2918                 if (mddev->pers->reconfig == NULL)
2919                         return -EINVAL;
2920                 else
2921                         return mddev->pers->reconfig(mddev, info->layout, -1);
2922         }
2923         if (mddev->size != info->size) {
2924                 mdk_rdev_t * rdev;
2925                 struct list_head *tmp;
2926                 if (mddev->pers->resize == NULL)
2927                         return -EINVAL;
2928                 /* The "size" is the amount of each device that is used.
2929                  * This can only make sense for arrays with redundancy.
2930                  * linear and raid0 always use whatever space is available
2931                  * We can only consider changing the size if no resync
2932                  * or reconstruction is happening, and if the new size
2933                  * is acceptable. It must fit before the sb_offset or,
2934                  * if that is <data_offset, it must fit before the
2935                  * size of each device.
2936                  * If size is zero, we find the largest size that fits.
2937                  */
2938                 if (mddev->sync_thread)
2939                         return -EBUSY;
2940                 ITERATE_RDEV(mddev,rdev,tmp) {
2941                         sector_t avail;
2942                         int fit = (info->size == 0);
2943                         if (rdev->sb_offset > rdev->data_offset)
2944                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2945                         else
2946                                 avail = get_capacity(rdev->bdev->bd_disk)
2947                                         - rdev->data_offset;
2948                         if (fit && (info->size == 0 || info->size > avail/2))
2949                                 info->size = avail/2;
2950                         if (avail < ((sector_t)info->size << 1))
2951                                 return -ENOSPC;
2952                 }
2953                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2954                 if (!rv) {
2955                         struct block_device *bdev;
2956
2957                         bdev = bdget_disk(mddev->gendisk, 0);
2958                         if (bdev) {
2959                                 down(&bdev->bd_inode->i_sem);
2960                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2961                                 up(&bdev->bd_inode->i_sem);
2962                                 bdput(bdev);
2963                         }
2964                 }
2965         }
2966         if (mddev->raid_disks    != info->raid_disks) {
2967                 /* change the number of raid disks */
2968                 if (mddev->pers->reshape == NULL)
2969                         return -EINVAL;
2970                 if (info->raid_disks <= 0 ||
2971                     info->raid_disks >= mddev->max_disks)
2972                         return -EINVAL;
2973                 if (mddev->sync_thread)
2974                         return -EBUSY;
2975                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2976                 if (!rv) {
2977                         struct block_device *bdev;
2978
2979                         bdev = bdget_disk(mddev->gendisk, 0);
2980                         if (bdev) {
2981                                 down(&bdev->bd_inode->i_sem);
2982                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2983                                 up(&bdev->bd_inode->i_sem);
2984                                 bdput(bdev);
2985                         }
2986                 }
2987         }
2988         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2989                 if (mddev->pers->quiesce == NULL)
2990                         return -EINVAL;
2991                 if (mddev->recovery || mddev->sync_thread)
2992                         return -EBUSY;
2993                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2994                         /* add the bitmap */
2995                         if (mddev->bitmap)
2996                                 return -EEXIST;
2997                         if (mddev->default_bitmap_offset == 0)
2998                                 return -EINVAL;
2999                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3000                         mddev->pers->quiesce(mddev, 1);
3001                         rv = bitmap_create(mddev);
3002                         if (rv)
3003                                 bitmap_destroy(mddev);
3004                         mddev->pers->quiesce(mddev, 0);
3005                 } else {
3006                         /* remove the bitmap */
3007                         if (!mddev->bitmap)
3008                                 return -ENOENT;
3009                         if (mddev->bitmap->file)
3010                                 return -EINVAL;
3011                         mddev->pers->quiesce(mddev, 1);
3012                         bitmap_destroy(mddev);
3013                         mddev->pers->quiesce(mddev, 0);
3014                         mddev->bitmap_offset = 0;
3015                 }
3016         }
3017         md_update_sb(mddev);
3018         return rv;
3019 }
3020
3021 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3022 {
3023         mdk_rdev_t *rdev;
3024
3025         if (mddev->pers == NULL)
3026                 return -ENODEV;
3027
3028         rdev = find_rdev(mddev, dev);
3029         if (!rdev)
3030                 return -ENODEV;
3031
3032         md_error(mddev, rdev);
3033         return 0;
3034 }
3035
3036 static int md_ioctl(struct inode *inode, struct file *file,
3037                         unsigned int cmd, unsigned long arg)
3038 {
3039         int err = 0;
3040         void __user *argp = (void __user *)arg;
3041         struct hd_geometry __user *loc = argp;
3042         mddev_t *mddev = NULL;
3043
3044         if (!capable(CAP_SYS_ADMIN))
3045                 return -EACCES;
3046
3047         /*
3048          * Commands dealing with the RAID driver but not any
3049          * particular array:
3050          */
3051         switch (cmd)
3052         {
3053                 case RAID_VERSION:
3054                         err = get_version(argp);
3055                         goto done;
3056
3057                 case PRINT_RAID_DEBUG:
3058                         err = 0;
3059                         md_print_devices();
3060                         goto done;
3061
3062 #ifndef MODULE
3063                 case RAID_AUTORUN:
3064                         err = 0;
3065                         autostart_arrays(arg);
3066                         goto done;
3067 #endif
3068                 default:;
3069         }
3070
3071         /*
3072          * Commands creating/starting a new array:
3073          */
3074
3075         mddev = inode->i_bdev->bd_disk->private_data;
3076
3077         if (!mddev) {
3078                 BUG();
3079                 goto abort;
3080         }
3081
3082
3083         if (cmd == START_ARRAY) {
3084                 /* START_ARRAY doesn't need to lock the array as autostart_array
3085                  * does the locking, and it could even be a different array
3086                  */
3087                 static int cnt = 3;
3088                 if (cnt > 0 ) {
3089                         printk(KERN_WARNING
3090                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3091                                "This will not be supported beyond 2.6\n",
3092                                current->comm, current->pid);
3093                         cnt--;
3094                 }
3095                 err = autostart_array(new_decode_dev(arg));
3096                 if (err) {
3097                         printk(KERN_WARNING "md: autostart failed!\n");
3098                         goto abort;
3099                 }
3100                 goto done;
3101         }
3102
3103         err = mddev_lock(mddev);
3104         if (err) {
3105                 printk(KERN_INFO
3106                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3107                         err, cmd);
3108                 goto abort;
3109         }
3110
3111         switch (cmd)
3112         {
3113                 case SET_ARRAY_INFO:
3114                         {
3115                                 mdu_array_info_t info;
3116                                 if (!arg)
3117                                         memset(&info, 0, sizeof(info));
3118                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3119                                         err = -EFAULT;
3120                                         goto abort_unlock;
3121                                 }
3122                                 if (mddev->pers) {
3123                                         err = update_array_info(mddev, &info);
3124                                         if (err) {
3125                                                 printk(KERN_WARNING "md: couldn't update"
3126                                                        " array info. %d\n", err);
3127                                                 goto abort_unlock;
3128                                         }
3129                                         goto done_unlock;
3130                                 }
3131                                 if (!list_empty(&mddev->disks)) {
3132                                         printk(KERN_WARNING
3133                                                "md: array %s already has disks!\n",
3134                                                mdname(mddev));
3135                                         err = -EBUSY;
3136                                         goto abort_unlock;
3137                                 }
3138                                 if (mddev->raid_disks) {
3139                                         printk(KERN_WARNING
3140                                                "md: array %s already initialised!\n",
3141                                                mdname(mddev));
3142                                         err = -EBUSY;
3143                                         goto abort_unlock;
3144                                 }
3145                                 err = set_array_info(mddev, &info);
3146                                 if (err) {
3147                                         printk(KERN_WARNING "md: couldn't set"
3148                                                " array info. %d\n", err);
3149                                         goto abort_unlock;
3150                                 }
3151                         }
3152                         goto done_unlock;
3153
3154                 default:;
3155         }
3156
3157         /*
3158          * Commands querying/configuring an existing array:
3159          */
3160         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3161          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3162         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3163                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3164                 err = -ENODEV;
3165                 goto abort_unlock;
3166         }
3167
3168         /*
3169          * Commands even a read-only array can execute:
3170          */
3171         switch (cmd)
3172         {
3173                 case GET_ARRAY_INFO:
3174                         err = get_array_info(mddev, argp);
3175                         goto done_unlock;
3176
3177                 case GET_BITMAP_FILE:
3178                         err = get_bitmap_file(mddev, argp);
3179                         goto done_unlock;
3180
3181                 case GET_DISK_INFO:
3182                         err = get_disk_info(mddev, argp);
3183                         goto done_unlock;
3184
3185                 case RESTART_ARRAY_RW:
3186                         err = restart_array(mddev);
3187                         goto done_unlock;
3188
3189                 case STOP_ARRAY:
3190                         err = do_md_stop (mddev, 0);
3191                         goto done_unlock;
3192
3193                 case STOP_ARRAY_RO:
3194                         err = do_md_stop (mddev, 1);
3195                         goto done_unlock;
3196
3197         /*
3198          * We have a problem here : there is no easy way to give a CHS
3199          * virtual geometry. We currently pretend that we have a 2 heads
3200          * 4 sectors (with a BIG number of cylinders...). This drives
3201          * dosfs just mad... ;-)
3202          */
3203                 case HDIO_GETGEO:
3204                         if (!loc) {
3205                                 err = -EINVAL;
3206                                 goto abort_unlock;
3207                         }
3208                         err = put_user (2, (char __user *) &loc->heads);
3209                         if (err)
3210                                 goto abort_unlock;
3211                         err = put_user (4, (char __user *) &loc->sectors);
3212                         if (err)
3213                                 goto abort_unlock;
3214                         err = put_user(get_capacity(mddev->gendisk)/8,
3215                                         (short __user *) &loc->cylinders);
3216                         if (err)
3217                                 goto abort_unlock;
3218                         err = put_user (get_start_sect(inode->i_bdev),
3219                                                 (long __user *) &loc->start);
3220                         goto done_unlock;
3221         }
3222
3223         /*
3224          * The remaining ioctls are changing the state of the
3225          * superblock, so we do not allow read-only arrays
3226          * here:
3227          */
3228         if (mddev->ro) {
3229                 err = -EROFS;
3230                 goto abort_unlock;
3231         }
3232
3233         switch (cmd)
3234         {
3235                 case ADD_NEW_DISK:
3236                 {
3237                         mdu_disk_info_t info;
3238                         if (copy_from_user(&info, argp, sizeof(info)))
3239                                 err = -EFAULT;
3240                         else
3241                                 err = add_new_disk(mddev, &info);
3242                         goto done_unlock;
3243                 }
3244
3245                 case HOT_REMOVE_DISK:
3246                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3247                         goto done_unlock;
3248
3249                 case HOT_ADD_DISK:
3250                         err = hot_add_disk(mddev, new_decode_dev(arg));
3251                         goto done_unlock;
3252
3253                 case SET_DISK_FAULTY:
3254                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3255                         goto done_unlock;
3256
3257                 case RUN_ARRAY:
3258                         err = do_md_run (mddev);
3259                         goto done_unlock;
3260
3261                 case SET_BITMAP_FILE:
3262                         err = set_bitmap_file(mddev, (int)arg);
3263                         goto done_unlock;
3264
3265                 default:
3266                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3267                                 printk(KERN_WARNING "md: %s(pid %d) used"
3268                                         " obsolete MD ioctl, upgrade your"
3269                                         " software to use new ictls.\n",
3270                                         current->comm, current->pid);
3271                         err = -EINVAL;
3272                         goto abort_unlock;
3273         }
3274
3275 done_unlock:
3276 abort_unlock:
3277         mddev_unlock(mddev);
3278
3279         return err;
3280 done:
3281         if (err)
3282                 MD_BUG();
3283 abort:
3284         return err;
3285 }
3286
3287 static int md_open(struct inode *inode, struct file *file)
3288 {
3289         /*
3290          * Succeed if we can lock the mddev, which confirms that
3291          * it isn't being stopped right now.
3292          */
3293         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3294         int err;
3295
3296         if ((err = mddev_lock(mddev)))
3297                 goto out;
3298
3299         err = 0;
3300         mddev_get(mddev);
3301         mddev_unlock(mddev);
3302
3303         check_disk_change(inode->i_bdev);
3304  out:
3305         return err;
3306 }
3307
3308 static int md_release(struct inode *inode, struct file * file)
3309 {
3310         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3311
3312         if (!mddev)
3313                 BUG();
3314         mddev_put(mddev);
3315
3316         return 0;
3317 }
3318
3319 static int md_media_changed(struct gendisk *disk)
3320 {
3321         mddev_t *mddev = disk->private_data;
3322
3323         return mddev->changed;
3324 }
3325
3326 static int md_revalidate(struct gendisk *disk)
3327 {
3328         mddev_t *mddev = disk->private_data;
3329
3330         mddev->changed = 0;
3331         return 0;
3332 }
3333 static struct block_device_operations md_fops =
3334 {
3335         .owner          = THIS_MODULE,
3336         .open           = md_open,
3337         .release        = md_release,
3338         .ioctl          = md_ioctl,
3339         .media_changed  = md_media_changed,
3340         .revalidate_disk= md_revalidate,
3341 };
3342
3343 static int md_thread(void * arg)
3344 {
3345         mdk_thread_t *thread = arg;
3346
3347         /*
3348          * md_thread is a 'system-thread', it's priority should be very
3349          * high. We avoid resource deadlocks individually in each
3350          * raid personality. (RAID5 does preallocation) We also use RR and
3351          * the very same RT priority as kswapd, thus we will never get
3352          * into a priority inversion deadlock.
3353          *
3354          * we definitely have to have equal or higher priority than
3355          * bdflush, otherwise bdflush will deadlock if there are too
3356          * many dirty RAID5 blocks.
3357          */
3358
3359         allow_signal(SIGKILL);
3360         complete(thread->event);
3361         while (!kthread_should_stop()) {
3362                 void (*run)(mddev_t *);
3363
3364                 wait_event_interruptible_timeout(thread->wqueue,
3365                                                  test_bit(THREAD_WAKEUP, &thread->flags)
3366                                                  || kthread_should_stop(),
3367                                                  thread->timeout);
3368                 try_to_freeze();
3369
3370                 clear_bit(THREAD_WAKEUP, &thread->flags);
3371
3372                 run = thread->run;
3373                 if (run)
3374                         run(thread->mddev);
3375         }
3376
3377         return 0;
3378 }
3379
3380 void md_wakeup_thread(mdk_thread_t *thread)
3381 {
3382         if (thread) {
3383                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3384                 set_bit(THREAD_WAKEUP, &thread->flags);
3385                 wake_up(&thread->wqueue);
3386         }
3387 }
3388
3389 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3390                                  const char *name)
3391 {
3392         mdk_thread_t *thread;
3393         struct completion event;
3394
3395         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3396         if (!thread)
3397                 return NULL;
3398
3399         memset(thread, 0, sizeof(mdk_thread_t));
3400         init_waitqueue_head(&thread->wqueue);
3401
3402         init_completion(&event);
3403         thread->event = &event;
3404         thread->run = run;
3405         thread->mddev = mddev;
3406         thread->name = name;
3407         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3408         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3409         if (IS_ERR(thread->tsk)) {
3410                 kfree(thread);
3411                 return NULL;
3412         }
3413         wait_for_completion(&event);
3414         return thread;
3415 }
3416
3417 void md_unregister_thread(mdk_thread_t *thread)
3418 {
3419         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3420
3421         kthread_stop(thread->tsk);
3422         kfree(thread);
3423 }
3424
3425 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3426 {
3427         if (!mddev) {
3428                 MD_BUG();
3429                 return;
3430         }
3431
3432         if (!rdev || rdev->faulty)
3433                 return;
3434 /*
3435         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3436                 mdname(mddev),
3437                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3438                 __builtin_return_address(0),__builtin_return_address(1),
3439                 __builtin_return_address(2),__builtin_return_address(3));
3440 */
3441         if (!mddev->pers->error_handler)
3442                 return;
3443         mddev->pers->error_handler(mddev,rdev);
3444         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3445         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3446         md_wakeup_thread(mddev->thread);
3447 }
3448
3449 /* seq_file implementation /proc/mdstat */
3450
3451 static void status_unused(struct seq_file *seq)
3452 {
3453         int i = 0;
3454         mdk_rdev_t *rdev;
3455         struct list_head *tmp;
3456
3457         seq_printf(seq, "unused devices: ");
3458
3459         ITERATE_RDEV_PENDING(rdev,tmp) {
3460                 char b[BDEVNAME_SIZE];
3461                 i++;
3462                 seq_printf(seq, "%s ",
3463                               bdevname(rdev->bdev,b));
3464         }
3465         if (!i)
3466                 seq_printf(seq, "<none>");
3467
3468         seq_printf(seq, "\n");
3469 }
3470
3471
3472 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3473 {
3474         unsigned long max_blocks, resync, res, dt, db, rt;
3475
3476         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3477
3478         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3479                 max_blocks = mddev->resync_max_sectors >> 1;
3480         else
3481                 max_blocks = mddev->size;
3482
3483         /*
3484          * Should not happen.
3485          */
3486         if (!max_blocks) {
3487                 MD_BUG();
3488                 return;
3489         }
3490         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3491         {
3492                 int i, x = res/50, y = 20-x;
3493                 seq_printf(seq, "[");
3494                 for (i = 0; i < x; i++)
3495                         seq_printf(seq, "=");
3496                 seq_printf(seq, ">");
3497                 for (i = 0; i < y; i++)
3498                         seq_printf(seq, ".");
3499                 seq_printf(seq, "] ");
3500         }
3501         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3502                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3503                        "resync" : "recovery"),
3504                       res/10, res % 10, resync, max_blocks);
3505
3506         /*
3507          * We do not want to overflow, so the order of operands and
3508          * the * 100 / 100 trick are important. We do a +1 to be
3509          * safe against division by zero. We only estimate anyway.
3510          *
3511          * dt: time from mark until now
3512          * db: blocks written from mark until now
3513          * rt: remaining time
3514          */
3515         dt = ((jiffies - mddev->resync_mark) / HZ);
3516         if (!dt) dt++;
3517         db = resync - (mddev->resync_mark_cnt/2);
3518         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3519
3520         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3521
3522         seq_printf(seq, " speed=%ldK/sec", db/dt);
3523 }
3524
3525 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3526 {
3527         struct list_head *tmp;
3528         loff_t l = *pos;
3529         mddev_t *mddev;
3530
3531         if (l >= 0x10000)
3532                 return NULL;
3533         if (!l--)
3534                 /* header */
3535                 return (void*)1;
3536
3537         spin_lock(&all_mddevs_lock);
3538         list_for_each(tmp,&all_mddevs)
3539                 if (!l--) {
3540                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3541                         mddev_get(mddev);
3542                         spin_unlock(&all_mddevs_lock);
3543                         return mddev;
3544                 }
3545         spin_unlock(&all_mddevs_lock);
3546         if (!l--)
3547                 return (void*)2;/* tail */
3548         return NULL;
3549 }
3550
3551 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3552 {
3553         struct list_head *tmp;
3554         mddev_t *next_mddev, *mddev = v;
3555
3556         ++*pos;
3557         if (v == (void*)2)
3558                 return NULL;
3559
3560         spin_lock(&all_mddevs_lock);
3561         if (v == (void*)1)
3562                 tmp = all_mddevs.next;
3563         else
3564                 tmp = mddev->all_mddevs.next;
3565         if (tmp != &all_mddevs)
3566                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3567         else {
3568                 next_mddev = (void*)2;
3569                 *pos = 0x10000;
3570         }
3571         spin_unlock(&all_mddevs_lock);
3572
3573         if (v != (void*)1)
3574                 mddev_put(mddev);
3575         return next_mddev;
3576
3577 }
3578
3579 static void md_seq_stop(struct seq_file *seq, void *v)
3580 {
3581         mddev_t *mddev = v;
3582
3583         if (mddev && v != (void*)1 && v != (void*)2)
3584                 mddev_put(mddev);
3585 }
3586
3587 static int md_seq_show(struct seq_file *seq, void *v)
3588 {
3589         mddev_t *mddev = v;
3590         sector_t size;
3591         struct list_head *tmp2;
3592         mdk_rdev_t *rdev;
3593         int i;
3594         struct bitmap *bitmap;
3595
3596         if (v == (void*)1) {
3597                 seq_printf(seq, "Personalities : ");
3598                 spin_lock(&pers_lock);
3599                 for (i = 0; i < MAX_PERSONALITY; i++)
3600                         if (pers[i])
3601                                 seq_printf(seq, "[%s] ", pers[i]->name);
3602
3603                 spin_unlock(&pers_lock);
3604                 seq_printf(seq, "\n");
3605                 return 0;
3606         }
3607         if (v == (void*)2) {
3608                 status_unused(seq);
3609                 return 0;
3610         }
3611
3612         if (mddev_lock(mddev)!=0)
3613                 return -EINTR;
3614         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3615                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3616                                                 mddev->pers ? "" : "in");
3617                 if (mddev->pers) {
3618                         if (mddev->ro)
3619                                 seq_printf(seq, " (read-only)");
3620                         seq_printf(seq, " %s", mddev->pers->name);
3621                 }
3622
3623                 size = 0;
3624                 ITERATE_RDEV(mddev,rdev,tmp2) {
3625                         char b[BDEVNAME_SIZE];
3626                         seq_printf(seq, " %s[%d]",
3627                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3628                         if (test_bit(WriteMostly, &rdev->flags))
3629                                 seq_printf(seq, "(W)");
3630                         if (rdev->faulty) {
3631                                 seq_printf(seq, "(F)");
3632                                 continue;
3633                         } else if (rdev->raid_disk < 0)
3634                                 seq_printf(seq, "(S)"); /* spare */
3635                         size += rdev->size;
3636                 }
3637
3638                 if (!list_empty(&mddev->disks)) {
3639                         if (mddev->pers)
3640                                 seq_printf(seq, "\n      %llu blocks",
3641                                         (unsigned long long)mddev->array_size);
3642                         else
3643                                 seq_printf(seq, "\n      %llu blocks",
3644                                         (unsigned long long)size);
3645                 }
3646                 if (mddev->persistent) {
3647                         if (mddev->major_version != 0 ||
3648                             mddev->minor_version != 90) {
3649                                 seq_printf(seq," super %d.%d",
3650                                            mddev->major_version,
3651                                            mddev->minor_version);
3652                         }
3653                 } else
3654                         seq_printf(seq, " super non-persistent");
3655
3656                 if (mddev->pers) {
3657                         mddev->pers->status (seq, mddev);
3658                         seq_printf(seq, "\n      ");
3659                         if (mddev->curr_resync > 2) {
3660                                 status_resync (seq, mddev);
3661                                 seq_printf(seq, "\n      ");
3662                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3663                                 seq_printf(seq, "       resync=DELAYED\n      ");
3664                 } else
3665                         seq_printf(seq, "\n       ");
3666
3667                 if ((bitmap = mddev->bitmap)) {
3668                         unsigned long chunk_kb;
3669                         unsigned long flags;
3670                         spin_lock_irqsave(&bitmap->lock, flags);
3671                         chunk_kb = bitmap->chunksize >> 10;
3672                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3673                                 "%lu%s chunk",
3674                                 bitmap->pages - bitmap->missing_pages,
3675                                 bitmap->pages,
3676                                 (bitmap->pages - bitmap->missing_pages)
3677                                         << (PAGE_SHIFT - 10),
3678                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3679                                 chunk_kb ? "KB" : "B");
3680                         if (bitmap->file) {
3681                                 seq_printf(seq, ", file: ");
3682                                 seq_path(seq, bitmap->file->f_vfsmnt,
3683                                          bitmap->file->f_dentry," \t\n");
3684                         }
3685
3686                         seq_printf(seq, "\n");
3687                         spin_unlock_irqrestore(&bitmap->lock, flags);
3688                 }
3689
3690                 seq_printf(seq, "\n");
3691         }
3692         mddev_unlock(mddev);
3693
3694         return 0;
3695 }
3696
3697 static struct seq_operations md_seq_ops = {
3698         .start  = md_seq_start,
3699         .next   = md_seq_next,
3700         .stop   = md_seq_stop,
3701         .show   = md_seq_show,
3702 };
3703
3704 static int md_seq_open(struct inode *inode, struct file *file)
3705 {
3706         int error;
3707
3708         error = seq_open(file, &md_seq_ops);
3709         return error;
3710 }
3711
3712 static struct file_operations md_seq_fops = {
3713         .open           = md_seq_open,
3714         .read           = seq_read,
3715         .llseek         = seq_lseek,
3716         .release        = seq_release,
3717 };
3718
3719 int register_md_personality(int pnum, mdk_personality_t *p)
3720 {
3721         if (pnum >= MAX_PERSONALITY) {
3722                 printk(KERN_ERR
3723                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3724                        p->name, pnum, MAX_PERSONALITY-1);
3725                 return -EINVAL;
3726         }
3727
3728         spin_lock(&pers_lock);
3729         if (pers[pnum]) {
3730                 spin_unlock(&pers_lock);
3731                 return -EBUSY;
3732         }
3733
3734         pers[pnum] = p;
3735         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3736         spin_unlock(&pers_lock);
3737         return 0;
3738 }
3739
3740 int unregister_md_personality(int pnum)
3741 {
3742         if (pnum >= MAX_PERSONALITY)
3743                 return -EINVAL;
3744
3745         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3746         spin_lock(&pers_lock);
3747         pers[pnum] = NULL;
3748         spin_unlock(&pers_lock);
3749         return 0;
3750 }
3751
3752 static int is_mddev_idle(mddev_t *mddev)
3753 {
3754         mdk_rdev_t * rdev;
3755         struct list_head *tmp;
3756         int idle;
3757         unsigned long curr_events;
3758
3759         idle = 1;
3760         ITERATE_RDEV(mddev,rdev,tmp) {
3761                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3762                 curr_events = disk_stat_read(disk, sectors[0]) +
3763                                 disk_stat_read(disk, sectors[1]) -
3764                                 atomic_read(&disk->sync_io);
3765                 /* Allow some slack between valud of curr_events and last_events,
3766                  * as there are some uninteresting races.
3767                  * Note: the following is an unsigned comparison.
3768                  */
3769                 if ((curr_events - rdev->last_events + 32) > 64) {
3770                         rdev->last_events = curr_events;
3771                         idle = 0;
3772                 }
3773         }
3774         return idle;
3775 }
3776
3777 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3778 {
3779         /* another "blocks" (512byte) blocks have been synced */
3780         atomic_sub(blocks, &mddev->recovery_active);
3781         wake_up(&mddev->recovery_wait);
3782         if (!ok) {
3783                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3784                 md_wakeup_thread(mddev->thread);
3785                 // stop recovery, signal do_sync ....
3786         }
3787 }
3788
3789
3790 /* md_write_start(mddev, bi)
3791  * If we need to update some array metadata (e.g. 'active' flag
3792  * in superblock) before writing, schedule a superblock update
3793  * and wait for it to complete.
3794  */
3795 void md_write_start(mddev_t *mddev, struct bio *bi)
3796 {
3797         if (bio_data_dir(bi) != WRITE)
3798                 return;
3799
3800         atomic_inc(&mddev->writes_pending);
3801         if (mddev->in_sync) {
3802                 spin_lock(&mddev->write_lock);
3803                 if (mddev->in_sync) {
3804                         mddev->in_sync = 0;
3805                         mddev->sb_dirty = 1;
3806                         md_wakeup_thread(mddev->thread);
3807                 }
3808                 spin_unlock(&mddev->write_lock);
3809         }
3810         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3811 }
3812
3813 void md_write_end(mddev_t *mddev)
3814 {
3815         if (atomic_dec_and_test(&mddev->writes_pending)) {
3816                 if (mddev->safemode == 2)
3817                         md_wakeup_thread(mddev->thread);
3818                 else
3819                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3820         }
3821 }
3822
3823 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3824
3825 #define SYNC_MARKS      10
3826 #define SYNC_MARK_STEP  (3*HZ)
3827 static void md_do_sync(mddev_t *mddev)
3828 {
3829         mddev_t *mddev2;
3830         unsigned int currspeed = 0,
3831                  window;
3832         sector_t max_sectors,j, io_sectors;
3833         unsigned long mark[SYNC_MARKS];
3834         sector_t mark_cnt[SYNC_MARKS];
3835         int last_mark,m;
3836         struct list_head *tmp;
3837         sector_t last_check;
3838         int skipped = 0;
3839
3840         /* just incase thread restarts... */
3841         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3842                 return;
3843
3844         /* we overload curr_resync somewhat here.
3845          * 0 == not engaged in resync at all
3846          * 2 == checking that there is no conflict with another sync
3847          * 1 == like 2, but have yielded to allow conflicting resync to
3848          *              commense
3849          * other == active in resync - this many blocks
3850          *
3851          * Before starting a resync we must have set curr_resync to
3852          * 2, and then checked that every "conflicting" array has curr_resync
3853          * less than ours.  When we find one that is the same or higher
3854          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3855          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3856          * This will mean we have to start checking from the beginning again.
3857          *
3858          */
3859
3860         do {
3861                 mddev->curr_resync = 2;
3862
3863         try_again:
3864                 if (signal_pending(current) ||
3865                     kthread_should_stop()) {
3866                         flush_signals(current);
3867                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3868                         goto skip;
3869                 }
3870                 ITERATE_MDDEV(mddev2,tmp) {
3871                         if (mddev2 == mddev)
3872                                 continue;
3873                         if (mddev2->curr_resync &&
3874                             match_mddev_units(mddev,mddev2)) {
3875                                 DEFINE_WAIT(wq);
3876                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3877                                         /* arbitrarily yield */
3878                                         mddev->curr_resync = 1;
3879                                         wake_up(&resync_wait);
3880                                 }
3881                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3882                                         /* no need to wait here, we can wait the next
3883                                          * time 'round when curr_resync == 2
3884                                          */
3885                                         continue;
3886                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3887                                 if (!signal_pending(current) &&
3888                                     !kthread_should_stop() &&
3889                                     mddev2->curr_resync >= mddev->curr_resync) {
3890                                         printk(KERN_INFO "md: delaying resync of %s"
3891                                                " until %s has finished resync (they"
3892                                                " share one or more physical units)\n",
3893                                                mdname(mddev), mdname(mddev2));
3894                                         mddev_put(mddev2);
3895                                         schedule();
3896                                         finish_wait(&resync_wait, &wq);
3897                                         goto try_again;
3898                                 }
3899                                 finish_wait(&resync_wait, &wq);
3900                         }
3901                 }
3902         } while (mddev->curr_resync < 2);
3903
3904         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3905                 /* resync follows the size requested by the personality,
3906                  * which defaults to physical size, but can be virtual size
3907                  */
3908                 max_sectors = mddev->resync_max_sectors;
3909                 mddev->resync_mismatches = 0;
3910         } else
3911                 /* recovery follows the physical size of devices */
3912                 max_sectors = mddev->size << 1;
3913
3914         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3915         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3916                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3917         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3918                "(but not more than %d KB/sec) for reconstruction.\n",
3919                sysctl_speed_limit_max);
3920
3921         is_mddev_idle(mddev); /* this also initializes IO event counters */
3922         /* we don't use the checkpoint if there's a bitmap */
3923         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
3924             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3925                 j = mddev->recovery_cp;
3926         else
3927                 j = 0;
3928         io_sectors = 0;
3929         for (m = 0; m < SYNC_MARKS; m++) {
3930                 mark[m] = jiffies;
3931                 mark_cnt[m] = io_sectors;
3932         }
3933         last_mark = 0;
3934         mddev->resync_mark = mark[last_mark];
3935         mddev->resync_mark_cnt = mark_cnt[last_mark];
3936
3937         /*
3938          * Tune reconstruction:
3939          */
3940         window = 32*(PAGE_SIZE/512);
3941         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3942                 window/2,(unsigned long long) max_sectors/2);
3943
3944         atomic_set(&mddev->recovery_active, 0);
3945         init_waitqueue_head(&mddev->recovery_wait);
3946         last_check = 0;
3947
3948         if (j>2) {
3949                 printk(KERN_INFO
3950                         "md: resuming recovery of %s from checkpoint.\n",
3951                         mdname(mddev));
3952                 mddev->curr_resync = j;
3953         }
3954
3955         while (j < max_sectors) {
3956                 sector_t sectors;
3957
3958                 skipped = 0;
3959                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3960                                             currspeed < sysctl_speed_limit_min);
3961                 if (sectors == 0) {
3962                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3963                         goto out;
3964                 }
3965
3966                 if (!skipped) { /* actual IO requested */
3967                         io_sectors += sectors;
3968                         atomic_add(sectors, &mddev->recovery_active);
3969                 }
3970
3971                 j += sectors;
3972                 if (j>1) mddev->curr_resync = j;
3973
3974
3975                 if (last_check + window > io_sectors || j == max_sectors)
3976                         continue;
3977
3978                 last_check = io_sectors;
3979
3980                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3981                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3982                         break;
3983
3984         repeat:
3985                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3986                         /* step marks */
3987                         int next = (last_mark+1) % SYNC_MARKS;
3988
3989                         mddev->resync_mark = mark[next];
3990                         mddev->resync_mark_cnt = mark_cnt[next];
3991                         mark[next] = jiffies;
3992                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3993                         last_mark = next;
3994                 }
3995
3996
3997                 if (signal_pending(current) || kthread_should_stop()) {
3998                         /*
3999                          * got a signal, exit.
4000                          */
4001                         printk(KERN_INFO
4002                                 "md: md_do_sync() got signal ... exiting\n");
4003                         flush_signals(current);
4004                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4005                         goto out;
4006                 }
4007
4008                 /*
4009                  * this loop exits only if either when we are slower than
4010                  * the 'hard' speed limit, or the system was IO-idle for
4011                  * a jiffy.
4012                  * the system might be non-idle CPU-wise, but we only care
4013                  * about not overloading the IO subsystem. (things like an
4014                  * e2fsck being done on the RAID array should execute fast)
4015                  */
4016                 mddev->queue->unplug_fn(mddev->queue);
4017                 cond_resched();
4018
4019                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4020                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4021
4022                 if (currspeed > sysctl_speed_limit_min) {
4023                         if ((currspeed > sysctl_speed_limit_max) ||
4024                                         !is_mddev_idle(mddev)) {
4025                                 msleep_interruptible(250);
4026                                 goto repeat;
4027                         }
4028                 }
4029         }
4030         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4031         /*
4032          * this also signals 'finished resyncing' to md_stop
4033          */
4034  out:
4035         mddev->queue->unplug_fn(mddev->queue);
4036
4037         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4038
4039         /* tell personality that we are finished */
4040         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4041
4042         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4043             mddev->curr_resync > 2 &&
4044             mddev->curr_resync >= mddev->recovery_cp) {
4045                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4046                         printk(KERN_INFO
4047                                 "md: checkpointing recovery of %s.\n",
4048                                 mdname(mddev));
4049                         mddev->recovery_cp = mddev->curr_resync;
4050                 } else
4051                         mddev->recovery_cp = MaxSector;
4052         }
4053
4054  skip:
4055         mddev->curr_resync = 0;
4056         wake_up(&resync_wait);
4057         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4058         md_wakeup_thread(mddev->thread);
4059 }
4060
4061
4062 /*
4063  * This routine is regularly called by all per-raid-array threads to
4064  * deal with generic issues like resync and super-block update.
4065  * Raid personalities that don't have a thread (linear/raid0) do not
4066  * need this as they never do any recovery or update the superblock.
4067  *
4068  * It does not do any resync itself, but rather "forks" off other threads
4069  * to do that as needed.
4070  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4071  * "->recovery" and create a thread at ->sync_thread.
4072  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4073  * and wakeups up this thread which will reap the thread and finish up.
4074  * This thread also removes any faulty devices (with nr_pending == 0).
4075  *
4076  * The overall approach is:
4077  *  1/ if the superblock needs updating, update it.
4078  *  2/ If a recovery thread is running, don't do anything else.
4079  *  3/ If recovery has finished, clean up, possibly marking spares active.
4080  *  4/ If there are any faulty devices, remove them.
4081  *  5/ If array is degraded, try to add spares devices
4082  *  6/ If array has spares or is not in-sync, start a resync thread.
4083  */
4084 void md_check_recovery(mddev_t *mddev)
4085 {
4086         mdk_rdev_t *rdev;
4087         struct list_head *rtmp;
4088
4089
4090         if (mddev->bitmap)
4091                 bitmap_daemon_work(mddev->bitmap);
4092
4093         if (mddev->ro)
4094                 return;
4095
4096         if (signal_pending(current)) {
4097                 if (mddev->pers->sync_request) {
4098                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4099                                mdname(mddev));
4100                         mddev->safemode = 2;
4101                 }
4102                 flush_signals(current);
4103         }
4104
4105         if ( ! (
4106                 mddev->sb_dirty ||
4107                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4108                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4109                 (mddev->safemode == 1) ||
4110                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4111                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4112                 ))
4113                 return;
4114
4115         if (mddev_trylock(mddev)==0) {
4116                 int spares =0;
4117
4118                 spin_lock(&mddev->write_lock);
4119                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4120                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4121                         mddev->in_sync = 1;
4122                         mddev->sb_dirty = 1;
4123                 }
4124                 if (mddev->safemode == 1)
4125                         mddev->safemode = 0;
4126                 spin_unlock(&mddev->write_lock);
4127
4128                 if (mddev->sb_dirty)
4129                         md_update_sb(mddev);
4130
4131
4132                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4133                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4134                         /* resync/recovery still happening */
4135                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4136                         goto unlock;
4137                 }
4138                 if (mddev->sync_thread) {
4139                         /* resync has finished, collect result */
4140                         md_unregister_thread(mddev->sync_thread);
4141                         mddev->sync_thread = NULL;
4142                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4143                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4144                                 /* success...*/
4145                                 /* activate any spares */
4146                                 mddev->pers->spare_active(mddev);
4147                         }
4148                         md_update_sb(mddev);
4149
4150                         /* if array is no-longer degraded, then any saved_raid_disk
4151                          * information must be scrapped
4152                          */
4153                         if (!mddev->degraded)
4154                                 ITERATE_RDEV(mddev,rdev,rtmp)
4155                                         rdev->saved_raid_disk = -1;
4156
4157                         mddev->recovery = 0;
4158                         /* flag recovery needed just to double check */
4159                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4160                         goto unlock;
4161                 }
4162                 /* Clear some bits that don't mean anything, but
4163                  * might be left set
4164                  */
4165                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4166                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4167                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4168                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4169
4170                 /* no recovery is running.
4171                  * remove any failed drives, then
4172                  * add spares if possible.
4173                  * Spare are also removed and re-added, to allow
4174                  * the personality to fail the re-add.
4175                  */
4176                 ITERATE_RDEV(mddev,rdev,rtmp)
4177                         if (rdev->raid_disk >= 0 &&
4178                             (rdev->faulty || ! rdev->in_sync) &&
4179                             atomic_read(&rdev->nr_pending)==0) {
4180                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4181                                         char nm[20];
4182                                         sprintf(nm,"rd%d", rdev->raid_disk);
4183                                         sysfs_remove_link(&mddev->kobj, nm);
4184                                         rdev->raid_disk = -1;
4185                                 }
4186                         }
4187
4188                 if (mddev->degraded) {
4189                         ITERATE_RDEV(mddev,rdev,rtmp)
4190                                 if (rdev->raid_disk < 0
4191                                     && !rdev->faulty) {
4192                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4193                                                 char nm[20];
4194                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4195                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4196                                                 spares++;
4197                                         } else
4198                                                 break;
4199                                 }
4200                 }
4201
4202                 if (spares) {
4203                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4204                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4205                 } else if (mddev->recovery_cp < MaxSector) {
4206                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4207                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4208                         /* nothing to be done ... */
4209                         goto unlock;
4210
4211                 if (mddev->pers->sync_request) {
4212                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4213                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4214                                 /* We are adding a device or devices to an array
4215                                  * which has the bitmap stored on all devices.
4216                                  * So make sure all bitmap pages get written
4217                                  */
4218                                 bitmap_write_all(mddev->bitmap);
4219                         }
4220                         mddev->sync_thread = md_register_thread(md_do_sync,
4221                                                                 mddev,
4222                                                                 "%s_resync");
4223                         if (!mddev->sync_thread) {
4224                                 printk(KERN_ERR "%s: could not start resync"
4225                                         " thread...\n",
4226                                         mdname(mddev));
4227                                 /* leave the spares where they are, it shouldn't hurt */
4228                                 mddev->recovery = 0;
4229                         } else {
4230                                 md_wakeup_thread(mddev->sync_thread);
4231                         }
4232                 }
4233         unlock:
4234                 mddev_unlock(mddev);
4235         }
4236 }
4237
4238 static int md_notify_reboot(struct notifier_block *this,
4239                             unsigned long code, void *x)
4240 {
4241         struct list_head *tmp;
4242         mddev_t *mddev;
4243
4244         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4245
4246                 printk(KERN_INFO "md: stopping all md devices.\n");
4247
4248                 ITERATE_MDDEV(mddev,tmp)
4249                         if (mddev_trylock(mddev)==0)
4250                                 do_md_stop (mddev, 1);
4251                 /*
4252                  * certain more exotic SCSI devices are known to be
4253                  * volatile wrt too early system reboots. While the
4254                  * right place to handle this issue is the given
4255                  * driver, we do want to have a safe RAID driver ...
4256                  */
4257                 mdelay(1000*1);
4258         }
4259         return NOTIFY_DONE;
4260 }
4261
4262 static struct notifier_block md_notifier = {
4263         .notifier_call  = md_notify_reboot,
4264         .next           = NULL,
4265         .priority       = INT_MAX, /* before any real devices */
4266 };
4267
4268 static void md_geninit(void)
4269 {
4270         struct proc_dir_entry *p;
4271
4272         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4273
4274         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4275         if (p)
4276                 p->proc_fops = &md_seq_fops;
4277 }
4278
4279 static int __init md_init(void)
4280 {
4281         int minor;
4282
4283         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4284                         " MD_SB_DISKS=%d\n",
4285                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4286                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4287         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
4288                         BITMAP_MINOR);
4289
4290         if (register_blkdev(MAJOR_NR, "md"))
4291                 return -1;
4292         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4293                 unregister_blkdev(MAJOR_NR, "md");
4294                 return -1;
4295         }
4296         devfs_mk_dir("md");
4297         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4298                                 md_probe, NULL, NULL);
4299         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4300                             md_probe, NULL, NULL);
4301
4302         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4303                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4304                                 S_IFBLK|S_IRUSR|S_IWUSR,
4305                                 "md/%d", minor);
4306
4307         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4308                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4309                               S_IFBLK|S_IRUSR|S_IWUSR,
4310                               "md/mdp%d", minor);
4311
4312
4313         register_reboot_notifier(&md_notifier);
4314         raid_table_header = register_sysctl_table(raid_root_table, 1);
4315
4316         md_geninit();
4317         return (0);
4318 }
4319
4320
4321 #ifndef MODULE
4322
4323 /*
4324  * Searches all registered partitions for autorun RAID arrays
4325  * at boot time.
4326  */
4327 static dev_t detected_devices[128];
4328 static int dev_cnt;
4329
4330 void md_autodetect_dev(dev_t dev)
4331 {
4332         if (dev_cnt >= 0 && dev_cnt < 127)
4333                 detected_devices[dev_cnt++] = dev;
4334 }
4335
4336
4337 static void autostart_arrays(int part)
4338 {
4339         mdk_rdev_t *rdev;
4340         int i;
4341
4342         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4343
4344         for (i = 0; i < dev_cnt; i++) {
4345                 dev_t dev = detected_devices[i];
4346
4347                 rdev = md_import_device(dev,0, 0);
4348                 if (IS_ERR(rdev))
4349                         continue;
4350
4351                 if (rdev->faulty) {
4352                         MD_BUG();
4353                         continue;
4354                 }
4355                 list_add(&rdev->same_set, &pending_raid_disks);
4356         }
4357         dev_cnt = 0;
4358
4359         autorun_devices(part);
4360 }
4361
4362 #endif
4363
4364 static __exit void md_exit(void)
4365 {
4366         mddev_t *mddev;
4367         struct list_head *tmp;
4368         int i;
4369         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4370         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4371         for (i=0; i < MAX_MD_DEVS; i++)
4372                 devfs_remove("md/%d", i);
4373         for (i=0; i < MAX_MD_DEVS; i++)
4374                 devfs_remove("md/d%d", i);
4375
4376         devfs_remove("md");
4377
4378         unregister_blkdev(MAJOR_NR,"md");
4379         unregister_blkdev(mdp_major, "mdp");
4380         unregister_reboot_notifier(&md_notifier);
4381         unregister_sysctl_table(raid_table_header);
4382         remove_proc_entry("mdstat", NULL);
4383         ITERATE_MDDEV(mddev,tmp) {
4384                 struct gendisk *disk = mddev->gendisk;
4385                 if (!disk)
4386                         continue;
4387                 export_array(mddev);
4388                 del_gendisk(disk);
4389                 put_disk(disk);
4390                 mddev->gendisk = NULL;
4391                 mddev_put(mddev);
4392         }
4393 }
4394
4395 module_init(md_init)
4396 module_exit(md_exit)
4397
4398 EXPORT_SYMBOL(register_md_personality);
4399 EXPORT_SYMBOL(unregister_md_personality);
4400 EXPORT_SYMBOL(md_error);
4401 EXPORT_SYMBOL(md_done_sync);
4402 EXPORT_SYMBOL(md_write_start);
4403 EXPORT_SYMBOL(md_write_end);
4404 EXPORT_SYMBOL(md_register_thread);
4405 EXPORT_SYMBOL(md_unregister_thread);
4406 EXPORT_SYMBOL(md_wakeup_thread);
4407 EXPORT_SYMBOL(md_print_devices);
4408 EXPORT_SYMBOL(md_check_recovery);
4409 MODULE_LICENSE("GPL");
4410 MODULE_ALIAS("md");
4411 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);