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