drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45
  46 #include <linux/init.h>
  47
  48 #include <linux/file.h>
  49
  50 #ifdef CONFIG_KMOD
  51 #include <linux/kmod.h>
  52 #endif
  53
  54 #include <asm/unaligned.h>
  55
  56 #define MAJOR_NR MD_MAJOR
  57 #define MD_DRIVER
  58
  59 /* 63 partitions with the alternate major number (mdp) */
  60 #define MdpMinorShift 6
  61
  62 #define DEBUG 0
  63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  64
  65
  66 #ifndef MODULE
  67 static void autostart_arrays (int part);
  68 #endif
  69
  70 static mdk_personality_t *pers[MAX_PERSONALITY];
  71 static DEFINE_SPINLOCK(pers_lock);
  72
  73 /*
  74  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  75  * is 1000 KB/sec, so the extra system load does not show up that much.
  76  * Increase it if you want to have more _guaranteed_ speed. Note that
  77  * the RAID driver will use the maximum available bandwidth if the IO
  78  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  79  * speed limit - in case reconstruction slows down your system despite
  80  * idle IO detection.
  81  *
  82  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  83  */
  84
  85 static int sysctl_speed_limit_min = 1000;
  86 static int sysctl_speed_limit_max = 200000;
  87
  88 static struct ctl_table_header *raid_table_header;
  89
  90 static ctl_table raid_table[] = {
  91         {
  92                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  93                 .procname       = "speed_limit_min",
  94                 .data           = &sysctl_speed_limit_min,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec,
  98         },
  99         {
 100                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 101                 .procname       = "speed_limit_max",
 102                 .data           = &sysctl_speed_limit_max,
 103                 .maxlen         = sizeof(int),
 104                 .mode           = 0644,
 105                 .proc_handler   = &proc_dointvec,
 106         },
 107         { .ctl_name = 0 }
 108 };
 109
 110 static ctl_table raid_dir_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID,
 113                 .procname       = "raid",
 114                 .maxlen         = 0,
 115                 .mode           = 0555,
 116                 .child          = raid_table,
 117         },
 118         { .ctl_name = 0 }
 119 };
 120
 121 static ctl_table raid_root_table[] = {
 122         {
 123                 .ctl_name       = CTL_DEV,
 124                 .procname       = "dev",
 125                 .maxlen         = 0,
 126                 .mode           = 0555,
 127                 .child          = raid_dir_table,
 128         },
 129         { .ctl_name = 0 }
 130 };
 131
 132 static struct block_device_operations md_fops;
 133
 134 /*
 135  * Enables to iterate over all existing md arrays
 136  * all_mddevs_lock protects this list.
 137  */
 138 static LIST_HEAD(all_mddevs);
 139 static DEFINE_SPINLOCK(all_mddevs_lock);
 140
 141
 142 /*
 143  * iterates through all used mddevs in the system.
 144  * We take care to grab the all_mddevs_lock whenever navigating
 145  * the list, and to always hold a refcount when unlocked.
 146  * Any code which breaks out of this loop while own
 147  * a reference to the current mddev and must mddev_put it.
 148  */
 149 #define ITERATE_MDDEV(mddev,tmp)                                        \
 150                                                                         \
 151         for (({ spin_lock(&all_mddevs_lock);                            \
 152                 tmp = all_mddevs.next;                                  \
 153                 mddev = NULL;});                                        \
 154              ({ if (tmp != &all_mddevs)                                 \
 155                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 156                 spin_unlock(&all_mddevs_lock);                          \
 157                 if (mddev) mddev_put(mddev);                            \
 158                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 159                 tmp != &all_mddevs;});                                  \
 160              ({ spin_lock(&all_mddevs_lock);                            \
 161                 tmp = tmp->next;})                                      \
 162                 )
 163
 164
 165 static int md_fail_request (request_queue_t *q, struct bio *bio)
 166 {
 167         bio_io_error(bio, bio->bi_size);
 168         return 0;
 169 }
 170
 171 static inline mddev_t *mddev_get(mddev_t *mddev)
 172 {
 173         atomic_inc(&mddev->active);
 174         return mddev;
 175 }
 176
 177 static void mddev_put(mddev_t *mddev)
 178 {
 179         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 180                 return;
 181         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 182                 list_del(&mddev->all_mddevs);
 183                 blk_put_queue(mddev->queue);
 184                 kobject_unregister(&mddev->kobj);
 185         }
 186         spin_unlock(&all_mddevs_lock);
 187 }
 188
 189 static mddev_t * mddev_find(dev_t unit)
 190 {
 191         mddev_t *mddev, *new = NULL;
 192
 193  retry:
 194         spin_lock(&all_mddevs_lock);
 195         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 196                 if (mddev->unit == unit) {
 197                         mddev_get(mddev);
 198                         spin_unlock(&all_mddevs_lock);
 199                         kfree(new);
 200                         return mddev;
 201                 }
 202
 203         if (new) {
 204                 list_add(&new->all_mddevs, &all_mddevs);
 205                 spin_unlock(&all_mddevs_lock);
 206                 return new;
 207         }
 208         spin_unlock(&all_mddevs_lock);
 209
 210         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 211         if (!new)
 212                 return NULL;
 213
 214         memset(new, 0, sizeof(*new));
 215
 216         new->unit = unit;
 217         if (MAJOR(unit) == MD_MAJOR)
 218                 new->md_minor = MINOR(unit);
 219         else
 220                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 221
 222         init_MUTEX(&new->reconfig_sem);
 223         INIT_LIST_HEAD(&new->disks);
 224         INIT_LIST_HEAD(&new->all_mddevs);
 225         init_timer(&new->safemode_timer);
 226         atomic_set(&new->active, 1);
 227         spin_lock_init(&new->write_lock);
 228         init_waitqueue_head(&new->sb_wait);
 229
 230         new->queue = blk_alloc_queue(GFP_KERNEL);
 231         if (!new->queue) {
 232                 kfree(new);
 233                 return NULL;
 234         }
 235
 236         blk_queue_make_request(new->queue, md_fail_request);
 237
 238         goto retry;
 239 }
 240
 241 static inline int mddev_lock(mddev_t * mddev)
 242 {
 243         return down_interruptible(&mddev->reconfig_sem);
 244 }
 245
 246 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 247 {
 248         down(&mddev->reconfig_sem);
 249 }
 250
 251 static inline int mddev_trylock(mddev_t * mddev)
 252 {
 253         return down_trylock(&mddev->reconfig_sem);
 254 }
 255
 256 static inline void mddev_unlock(mddev_t * mddev)
 257 {
 258         up(&mddev->reconfig_sem);
 259
 260         md_wakeup_thread(mddev->thread);
 261 }
 262
 263 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 264 {
 265         mdk_rdev_t * rdev;
 266         struct list_head *tmp;
 267
 268         ITERATE_RDEV(mddev,rdev,tmp) {
 269                 if (rdev->desc_nr == nr)
 270                         return rdev;
 271         }
 272         return NULL;
 273 }
 274
 275 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 276 {
 277         struct list_head *tmp;
 278         mdk_rdev_t *rdev;
 279
 280         ITERATE_RDEV(mddev,rdev,tmp) {
 281                 if (rdev->bdev->bd_dev == dev)
 282                         return rdev;
 283         }
 284         return NULL;
 285 }
 286
 287 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 288 {
 289         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 290         return MD_NEW_SIZE_BLOCKS(size);
 291 }
 292
 293 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 294 {
 295         sector_t size;
 296
 297         size = rdev->sb_offset;
 298
 299         if (chunk_size)
 300                 size &= ~((sector_t)chunk_size/1024 - 1);
 301         return size;
 302 }
 303
 304 static int alloc_disk_sb(mdk_rdev_t * rdev)
 305 {
 306         if (rdev->sb_page)
 307                 MD_BUG();
 308
 309         rdev->sb_page = alloc_page(GFP_KERNEL);
 310         if (!rdev->sb_page) {
 311                 printk(KERN_ALERT "md: out of memory.\n");
 312                 return -EINVAL;
 313         }
 314
 315         return 0;
 316 }
 317
 318 static void free_disk_sb(mdk_rdev_t * rdev)
 319 {
 320         if (rdev->sb_page) {
 321                 page_cache_release(rdev->sb_page);
 322                 rdev->sb_loaded = 0;
 323                 rdev->sb_page = NULL;
 324                 rdev->sb_offset = 0;
 325                 rdev->size = 0;
 326         }
 327 }
 328
 329
 330 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 331 {
 332         mdk_rdev_t *rdev = bio->bi_private;
 333         mddev_t *mddev = rdev->mddev;
 334         if (bio->bi_size)
 335                 return 1;
 336
 337         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 338                 md_error(mddev, rdev);
 339
 340         if (atomic_dec_and_test(&mddev->pending_writes))
 341                 wake_up(&mddev->sb_wait);
 342         bio_put(bio);
 343         return 0;
 344 }
 345
 346 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 347 {
 348         struct bio *bio2 = bio->bi_private;
 349         mdk_rdev_t *rdev = bio2->bi_private;
 350         mddev_t *mddev = rdev->mddev;
 351         if (bio->bi_size)
 352                 return 1;
 353
 354         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 355             error == -EOPNOTSUPP) {
 356                 unsigned long flags;
 357                 /* barriers don't appear to be supported :-( */
 358                 set_bit(BarriersNotsupp, &rdev->flags);
 359                 mddev->barriers_work = 0;
 360                 spin_lock_irqsave(&mddev->write_lock, flags);
 361                 bio2->bi_next = mddev->biolist;
 362                 mddev->biolist = bio2;
 363                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 364                 wake_up(&mddev->sb_wait);
 365                 bio_put(bio);
 366                 return 0;
 367         }
 368         bio_put(bio2);
 369         bio->bi_private = rdev;
 370         return super_written(bio, bytes_done, error);
 371 }
 372
 373 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 374                    sector_t sector, int size, struct page *page)
 375 {
 376         /* write first size bytes of page to sector of rdev
 377          * Increment mddev->pending_writes before returning
 378          * and decrement it on completion, waking up sb_wait
 379          * if zero is reached.
 380          * If an error occurred, call md_error
 381          *
 382          * As we might need to resubmit the request if BIO_RW_BARRIER
 383          * causes ENOTSUPP, we allocate a spare bio...
 384          */
 385         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 386         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 387
 388         bio->bi_bdev = rdev->bdev;
 389         bio->bi_sector = sector;
 390         bio_add_page(bio, page, size, 0);
 391         bio->bi_private = rdev;
 392         bio->bi_end_io = super_written;
 393         bio->bi_rw = rw;
 394
 395         atomic_inc(&mddev->pending_writes);
 396         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 397                 struct bio *rbio;
 398                 rw |= (1<<BIO_RW_BARRIER);
 399                 rbio = bio_clone(bio, GFP_NOIO);
 400                 rbio->bi_private = bio;
 401                 rbio->bi_end_io = super_written_barrier;
 402                 submit_bio(rw, rbio);
 403         } else
 404                 submit_bio(rw, bio);
 405 }
 406
 407 void md_super_wait(mddev_t *mddev)
 408 {
 409         /* wait for all superblock writes that were scheduled to complete.
 410          * if any had to be retried (due to BARRIER problems), retry them
 411          */
 412         DEFINE_WAIT(wq);
 413         for(;;) {
 414                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 415                 if (atomic_read(&mddev->pending_writes)==0)
 416                         break;
 417                 while (mddev->biolist) {
 418                         struct bio *bio;
 419                         spin_lock_irq(&mddev->write_lock);
 420                         bio = mddev->biolist;
 421                         mddev->biolist = bio->bi_next ;
 422                         bio->bi_next = NULL;
 423                         spin_unlock_irq(&mddev->write_lock);
 424                         submit_bio(bio->bi_rw, bio);
 425                 }
 426                 schedule();
 427         }
 428         finish_wait(&mddev->sb_wait, &wq);
 429 }
 430
 431 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 432 {
 433         if (bio->bi_size)
 434                 return 1;
 435
 436         complete((struct completion*)bio->bi_private);
 437         return 0;
 438 }
 439
 440 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 441                    struct page *page, int rw)
 442 {
 443         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 444         struct completion event;
 445         int ret;
 446
 447         rw |= (1 << BIO_RW_SYNC);
 448
 449         bio->bi_bdev = bdev;
 450         bio->bi_sector = sector;
 451         bio_add_page(bio, page, size, 0);
 452         init_completion(&event);
 453         bio->bi_private = &event;
 454         bio->bi_end_io = bi_complete;
 455         submit_bio(rw, bio);
 456         wait_for_completion(&event);
 457
 458         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 459         bio_put(bio);
 460         return ret;
 461 }
 462
 463 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 464 {
 465         char b[BDEVNAME_SIZE];
 466         if (!rdev->sb_page) {
 467                 MD_BUG();
 468                 return -EINVAL;
 469         }
 470         if (rdev->sb_loaded)
 471                 return 0;
 472
 473
 474         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 475                 goto fail;
 476         rdev->sb_loaded = 1;
 477         return 0;
 478
 479 fail:
 480         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 481                 bdevname(rdev->bdev,b));
 482         return -EINVAL;
 483 }
 484
 485 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 486 {
 487         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 488                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 489                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 490                 (sb1->set_uuid3 == sb2->set_uuid3))
 491
 492                 return 1;
 493
 494         return 0;
 495 }
 496
 497
 498 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 499 {
 500         int ret;
 501         mdp_super_t *tmp1, *tmp2;
 502
 503         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 504         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 505
 506         if (!tmp1 || !tmp2) {
 507                 ret = 0;
 508                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 509                 goto abort;
 510         }
 511
 512         *tmp1 = *sb1;
 513         *tmp2 = *sb2;
 514
 515         /*
 516          * nr_disks is not constant
 517          */
 518         tmp1->nr_disks = 0;
 519         tmp2->nr_disks = 0;
 520
 521         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 522                 ret = 0;
 523         else
 524                 ret = 1;
 525
 526 abort:
 527         kfree(tmp1);
 528         kfree(tmp2);
 529         return ret;
 530 }
 531
 532 static unsigned int calc_sb_csum(mdp_super_t * sb)
 533 {
 534         unsigned int disk_csum, csum;
 535
 536         disk_csum = sb->sb_csum;
 537         sb->sb_csum = 0;
 538         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 539         sb->sb_csum = disk_csum;
 540         return csum;
 541 }
 542
 543
 544 /*
 545  * Handle superblock details.
 546  * We want to be able to handle multiple superblock formats
 547  * so we have a common interface to them all, and an array of
 548  * different handlers.
 549  * We rely on user-space to write the initial superblock, and support
 550  * reading and updating of superblocks.
 551  * Interface methods are:
 552  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 553  *      loads and validates a superblock on dev.
 554  *      if refdev != NULL, compare superblocks on both devices
 555  *    Return:
 556  *      0 - dev has a superblock that is compatible with refdev
 557  *      1 - dev has a superblock that is compatible and newer than refdev
 558  *          so dev should be used as the refdev in future
 559  *     -EINVAL superblock incompatible or invalid
 560  *     -othererror e.g. -EIO
 561  *
 562  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 563  *      Verify that dev is acceptable into mddev.
 564  *       The first time, mddev->raid_disks will be 0, and data from
 565  *       dev should be merged in.  Subsequent calls check that dev
 566  *       is new enough.  Return 0 or -EINVAL
 567  *
 568  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 569  *     Update the superblock for rdev with data in mddev
 570  *     This does not write to disc.
 571  *
 572  */
 573
 574 struct super_type  {
 575         char            *name;
 576         struct module   *owner;
 577         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 578         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 579         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 580 };
 581
 582 /*
 583  * load_super for 0.90.0
 584  */
 585 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 586 {
 587         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 588         mdp_super_t *sb;
 589         int ret;
 590         sector_t sb_offset;
 591
 592         /*
 593          * Calculate the position of the superblock,
 594          * it's at the end of the disk.
 595          *
 596          * It also happens to be a multiple of 4Kb.
 597          */
 598         sb_offset = calc_dev_sboffset(rdev->bdev);
 599         rdev->sb_offset = sb_offset;
 600
 601         ret = read_disk_sb(rdev, MD_SB_BYTES);
 602         if (ret) return ret;
 603
 604         ret = -EINVAL;
 605
 606         bdevname(rdev->bdev, b);
 607         sb = (mdp_super_t*)page_address(rdev->sb_page);
 608
 609         if (sb->md_magic != MD_SB_MAGIC) {
 610                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 611                        b);
 612                 goto abort;
 613         }
 614
 615         if (sb->major_version != 0 ||
 616             sb->minor_version != 90) {
 617                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 618                         sb->major_version, sb->minor_version,
 619                         b);
 620                 goto abort;
 621         }
 622
 623         if (sb->raid_disks <= 0)
 624                 goto abort;
 625
 626         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 627                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 628                         b);
 629                 goto abort;
 630         }
 631
 632         rdev->preferred_minor = sb->md_minor;
 633         rdev->data_offset = 0;
 634         rdev->sb_size = MD_SB_BYTES;
 635
 636         if (sb->level == LEVEL_MULTIPATH)
 637                 rdev->desc_nr = -1;
 638         else
 639                 rdev->desc_nr = sb->this_disk.number;
 640
 641         if (refdev == 0)
 642                 ret = 1;
 643         else {
 644                 __u64 ev1, ev2;
 645                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 646                 if (!uuid_equal(refsb, sb)) {
 647                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 648                                 b, bdevname(refdev->bdev,b2));
 649                         goto abort;
 650                 }
 651                 if (!sb_equal(refsb, sb)) {
 652                         printk(KERN_WARNING "md: %s has same UUID"
 653                                " but different superblock to %s\n",
 654                                b, bdevname(refdev->bdev, b2));
 655                         goto abort;
 656                 }
 657                 ev1 = md_event(sb);
 658                 ev2 = md_event(refsb);
 659                 if (ev1 > ev2)
 660                         ret = 1;
 661                 else
 662                         ret = 0;
 663         }
 664         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 665
 666  abort:
 667         return ret;
 668 }
 669
 670 /*
 671  * validate_super for 0.90.0
 672  */
 673 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 674 {
 675         mdp_disk_t *desc;
 676         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 677
 678         rdev->raid_disk = -1;
 679         rdev->flags = 0;
 680         if (mddev->raid_disks == 0) {
 681                 mddev->major_version = 0;
 682                 mddev->minor_version = sb->minor_version;
 683                 mddev->patch_version = sb->patch_version;
 684                 mddev->persistent = ! sb->not_persistent;
 685                 mddev->chunk_size = sb->chunk_size;
 686                 mddev->ctime = sb->ctime;
 687                 mddev->utime = sb->utime;
 688                 mddev->level = sb->level;
 689                 mddev->layout = sb->layout;
 690                 mddev->raid_disks = sb->raid_disks;
 691                 mddev->size = sb->size;
 692                 mddev->events = md_event(sb);
 693                 mddev->bitmap_offset = 0;
 694                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 695
 696                 if (sb->state & (1<<MD_SB_CLEAN))
 697                         mddev->recovery_cp = MaxSector;
 698                 else {
 699                         if (sb->events_hi == sb->cp_events_hi &&
 700                                 sb->events_lo == sb->cp_events_lo) {
 701                                 mddev->recovery_cp = sb->recovery_cp;
 702                         } else
 703                                 mddev->recovery_cp = 0;
 704                 }
 705
 706                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 707                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 708                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 709                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 710
 711                 mddev->max_disks = MD_SB_DISKS;
 712
 713                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 714                     mddev->bitmap_file == NULL) {
 715                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
 716                                 /* FIXME use a better test */
 717                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
 718                                 return -EINVAL;
 719                         }
 720                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 721                 }
 722
 723         } else if (mddev->pers == NULL) {
 724                 /* Insist on good event counter while assembling */
 725                 __u64 ev1 = md_event(sb);
 726                 ++ev1;
 727                 if (ev1 < mddev->events)
 728                         return -EINVAL;
 729         } else if (mddev->bitmap) {
 730                 /* if adding to array with a bitmap, then we can accept an
 731                  * older device ... but not too old.
 732                  */
 733                 __u64 ev1 = md_event(sb);
 734                 if (ev1 < mddev->bitmap->events_cleared)
 735                         return 0;
 736         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 737                 return 0;
 738
 739         if (mddev->level != LEVEL_MULTIPATH) {
 740                 desc = sb->disks + rdev->desc_nr;
 741
 742                 if (desc->state & (1<<MD_DISK_FAULTY))
 743                         set_bit(Faulty, &rdev->flags);
 744                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 745                          desc->raid_disk < mddev->raid_disks) {
 746                         set_bit(In_sync, &rdev->flags);
 747                         rdev->raid_disk = desc->raid_disk;
 748                 }
 749                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 750                         set_bit(WriteMostly, &rdev->flags);
 751         } else /* MULTIPATH are always insync */
 752                 set_bit(In_sync, &rdev->flags);
 753         return 0;
 754 }
 755
 756 /*
 757  * sync_super for 0.90.0
 758  */
 759 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 760 {
 761         mdp_super_t *sb;
 762         struct list_head *tmp;
 763         mdk_rdev_t *rdev2;
 764         int next_spare = mddev->raid_disks;
 765
 766
 767         /* make rdev->sb match mddev data..
 768          *
 769          * 1/ zero out disks
 770          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 771          * 3/ any empty disks < next_spare become removed
 772          *
 773          * disks[0] gets initialised to REMOVED because
 774          * we cannot be sure from other fields if it has
 775          * been initialised or not.
 776          */
 777         int i;
 778         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 779
 780         rdev->sb_size = MD_SB_BYTES;
 781
 782         sb = (mdp_super_t*)page_address(rdev->sb_page);
 783
 784         memset(sb, 0, sizeof(*sb));
 785
 786         sb->md_magic = MD_SB_MAGIC;
 787         sb->major_version = mddev->major_version;
 788         sb->minor_version = mddev->minor_version;
 789         sb->patch_version = mddev->patch_version;
 790         sb->gvalid_words  = 0; /* ignored */
 791         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 792         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 793         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 794         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 795
 796         sb->ctime = mddev->ctime;
 797         sb->level = mddev->level;
 798         sb->size  = mddev->size;
 799         sb->raid_disks = mddev->raid_disks;
 800         sb->md_minor = mddev->md_minor;
 801         sb->not_persistent = !mddev->persistent;
 802         sb->utime = mddev->utime;
 803         sb->state = 0;
 804         sb->events_hi = (mddev->events>>32);
 805         sb->events_lo = (u32)mddev->events;
 806
 807         if (mddev->in_sync)
 808         {
 809                 sb->recovery_cp = mddev->recovery_cp;
 810                 sb->cp_events_hi = (mddev->events>>32);
 811                 sb->cp_events_lo = (u32)mddev->events;
 812                 if (mddev->recovery_cp == MaxSector)
 813                         sb->state = (1<< MD_SB_CLEAN);
 814         } else
 815                 sb->recovery_cp = 0;
 816
 817         sb->layout = mddev->layout;
 818         sb->chunk_size = mddev->chunk_size;
 819
 820         if (mddev->bitmap && mddev->bitmap_file == NULL)
 821                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 822
 823         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 824         ITERATE_RDEV(mddev,rdev2,tmp) {
 825                 mdp_disk_t *d;
 826                 int desc_nr;
 827                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 828                     && !test_bit(Faulty, &rdev2->flags))
 829                         desc_nr = rdev2->raid_disk;
 830                 else
 831                         desc_nr = next_spare++;
 832                 rdev2->desc_nr = desc_nr;
 833                 d = &sb->disks[rdev2->desc_nr];
 834                 nr_disks++;
 835                 d->number = rdev2->desc_nr;
 836                 d->major = MAJOR(rdev2->bdev->bd_dev);
 837                 d->minor = MINOR(rdev2->bdev->bd_dev);
 838                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 839                     && !test_bit(Faulty, &rdev2->flags))
 840                         d->raid_disk = rdev2->raid_disk;
 841                 else
 842                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 843                 if (test_bit(Faulty, &rdev2->flags)) {
 844                         d->state = (1<<MD_DISK_FAULTY);
 845                         failed++;
 846                 } else if (test_bit(In_sync, &rdev2->flags)) {
 847                         d->state = (1<<MD_DISK_ACTIVE);
 848                         d->state |= (1<<MD_DISK_SYNC);
 849                         active++;
 850                         working++;
 851                 } else {
 852                         d->state = 0;
 853                         spare++;
 854                         working++;
 855                 }
 856                 if (test_bit(WriteMostly, &rdev2->flags))
 857                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 858         }
 859         /* now set the "removed" and "faulty" bits on any missing devices */
 860         for (i=0 ; i < mddev->raid_disks ; i++) {
 861                 mdp_disk_t *d = &sb->disks[i];
 862                 if (d->state == 0 && d->number == 0) {
 863                         d->number = i;
 864                         d->raid_disk = i;
 865                         d->state = (1<<MD_DISK_REMOVED);
 866                         d->state |= (1<<MD_DISK_FAULTY);
 867                         failed++;
 868                 }
 869         }
 870         sb->nr_disks = nr_disks;
 871         sb->active_disks = active;
 872         sb->working_disks = working;
 873         sb->failed_disks = failed;
 874         sb->spare_disks = spare;
 875
 876         sb->this_disk = sb->disks[rdev->desc_nr];
 877         sb->sb_csum = calc_sb_csum(sb);
 878 }
 879
 880 /*
 881  * version 1 superblock
 882  */
 883
 884 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 885 {
 886         unsigned int disk_csum, csum;
 887         unsigned long long newcsum;
 888         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 889         unsigned int *isuper = (unsigned int*)sb;
 890         int i;
 891
 892         disk_csum = sb->sb_csum;
 893         sb->sb_csum = 0;
 894         newcsum = 0;
 895         for (i=0; size>=4; size -= 4 )
 896                 newcsum += le32_to_cpu(*isuper++);
 897
 898         if (size == 2)
 899                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 900
 901         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 902         sb->sb_csum = disk_csum;
 903         return cpu_to_le32(csum);
 904 }
 905
 906 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 907 {
 908         struct mdp_superblock_1 *sb;
 909         int ret;
 910         sector_t sb_offset;
 911         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 912         int bmask;
 913
 914         /*
 915          * Calculate the position of the superblock.
 916          * It is always aligned to a 4K boundary and
 917          * depeding on minor_version, it can be:
 918          * 0: At least 8K, but less than 12K, from end of device
 919          * 1: At start of device
 920          * 2: 4K from start of device.
 921          */
 922         switch(minor_version) {
 923         case 0:
 924                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 925                 sb_offset -= 8*2;
 926                 sb_offset &= ~(sector_t)(4*2-1);
 927                 /* convert from sectors to K */
 928                 sb_offset /= 2;
 929                 break;
 930         case 1:
 931                 sb_offset = 0;
 932                 break;
 933         case 2:
 934                 sb_offset = 4;
 935                 break;
 936         default:
 937                 return -EINVAL;
 938         }
 939         rdev->sb_offset = sb_offset;
 940
 941         /* superblock is rarely larger than 1K, but it can be larger,
 942          * and it is safe to read 4k, so we do that
 943          */
 944         ret = read_disk_sb(rdev, 4096);
 945         if (ret) return ret;
 946
 947
 948         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 949
 950         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 951             sb->major_version != cpu_to_le32(1) ||
 952             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 953             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 954             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 955                 return -EINVAL;
 956
 957         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 958                 printk("md: invalid superblock checksum on %s\n",
 959                         bdevname(rdev->bdev,b));
 960                 return -EINVAL;
 961         }
 962         if (le64_to_cpu(sb->data_size) < 10) {
 963                 printk("md: data_size too small on %s\n",
 964                        bdevname(rdev->bdev,b));
 965                 return -EINVAL;
 966         }
 967         rdev->preferred_minor = 0xffff;
 968         rdev->data_offset = le64_to_cpu(sb->data_offset);
 969
 970         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 971         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 972         if (rdev->sb_size & bmask)
 973                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 974
 975         if (refdev == 0)
 976                 return 1;
 977         else {
 978                 __u64 ev1, ev2;
 979                 struct mdp_superblock_1 *refsb =
 980                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 981
 982                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 983                     sb->level != refsb->level ||
 984                     sb->layout != refsb->layout ||
 985                     sb->chunksize != refsb->chunksize) {
 986                         printk(KERN_WARNING "md: %s has strangely different"
 987                                 " superblock to %s\n",
 988                                 bdevname(rdev->bdev,b),
 989                                 bdevname(refdev->bdev,b2));
 990                         return -EINVAL;
 991                 }
 992                 ev1 = le64_to_cpu(sb->events);
 993                 ev2 = le64_to_cpu(refsb->events);
 994
 995                 if (ev1 > ev2)
 996                         return 1;
 997         }
 998         if (minor_version)
 999                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1000         else
1001                 rdev->size = rdev->sb_offset;
1002         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1003                 return -EINVAL;
1004         rdev->size = le64_to_cpu(sb->data_size)/2;
1005         if (le32_to_cpu(sb->chunksize))
1006                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1007         return 0;
1008 }
1009
1010 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1011 {
1012         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1013
1014         rdev->raid_disk = -1;
1015         rdev->flags = 0;
1016         if (mddev->raid_disks == 0) {
1017                 mddev->major_version = 1;
1018                 mddev->patch_version = 0;
1019                 mddev->persistent = 1;
1020                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1021                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1022                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1023                 mddev->level = le32_to_cpu(sb->level);
1024                 mddev->layout = le32_to_cpu(sb->layout);
1025                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1026                 mddev->size = le64_to_cpu(sb->size)/2;
1027                 mddev->events = le64_to_cpu(sb->events);
1028                 mddev->bitmap_offset = 0;
1029                 mddev->default_bitmap_offset = 0;
1030                 mddev->default_bitmap_offset = 1024;
1031
1032                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1033                 memcpy(mddev->uuid, sb->set_uuid, 16);
1034
1035                 mddev->max_disks =  (4096-256)/2;
1036
1037                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1038                     mddev->bitmap_file == NULL ) {
1039                         if (mddev->level != 1) {
1040                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1041                                 return -EINVAL;
1042                         }
1043                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1044                 }
1045         } else if (mddev->pers == NULL) {
1046                 /* Insist of good event counter while assembling */
1047                 __u64 ev1 = le64_to_cpu(sb->events);
1048                 ++ev1;
1049                 if (ev1 < mddev->events)
1050                         return -EINVAL;
1051         } else if (mddev->bitmap) {
1052                 /* If adding to array with a bitmap, then we can accept an
1053                  * older device, but not too old.
1054                  */
1055                 __u64 ev1 = le64_to_cpu(sb->events);
1056                 if (ev1 < mddev->bitmap->events_cleared)
1057                         return 0;
1058         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1059                 return 0;
1060
1061         if (mddev->level != LEVEL_MULTIPATH) {
1062                 int role;
1063                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1064                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1065                 switch(role) {
1066                 case 0xffff: /* spare */
1067                         break;
1068                 case 0xfffe: /* faulty */
1069                         set_bit(Faulty, &rdev->flags);
1070                         break;
1071                 default:
1072                         set_bit(In_sync, &rdev->flags);
1073                         rdev->raid_disk = role;
1074                         break;
1075                 }
1076                 if (sb->devflags & WriteMostly1)
1077                         set_bit(WriteMostly, &rdev->flags);
1078         } else /* MULTIPATH are always insync */
1079                 set_bit(In_sync, &rdev->flags);
1080
1081         return 0;
1082 }
1083
1084 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1085 {
1086         struct mdp_superblock_1 *sb;
1087         struct list_head *tmp;
1088         mdk_rdev_t *rdev2;
1089         int max_dev, i;
1090         /* make rdev->sb match mddev and rdev data. */
1091
1092         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1093
1094         sb->feature_map = 0;
1095         sb->pad0 = 0;
1096         memset(sb->pad1, 0, sizeof(sb->pad1));
1097         memset(sb->pad2, 0, sizeof(sb->pad2));
1098         memset(sb->pad3, 0, sizeof(sb->pad3));
1099
1100         sb->utime = cpu_to_le64((__u64)mddev->utime);
1101         sb->events = cpu_to_le64(mddev->events);
1102         if (mddev->in_sync)
1103                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1104         else
1105                 sb->resync_offset = cpu_to_le64(0);
1106
1107         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1108                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1109                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1110         }
1111
1112         max_dev = 0;
1113         ITERATE_RDEV(mddev,rdev2,tmp)
1114                 if (rdev2->desc_nr+1 > max_dev)
1115                         max_dev = rdev2->desc_nr+1;
1116
1117         sb->max_dev = cpu_to_le32(max_dev);
1118         for (i=0; i<max_dev;i++)
1119                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1120
1121         ITERATE_RDEV(mddev,rdev2,tmp) {
1122                 i = rdev2->desc_nr;
1123                 if (test_bit(Faulty, &rdev2->flags))
1124                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1125                 else if (test_bit(In_sync, &rdev2->flags))
1126                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1127                 else
1128                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1129         }
1130
1131         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1132         sb->sb_csum = calc_sb_1_csum(sb);
1133 }
1134
1135
1136 static struct super_type super_types[] = {
1137         [0] = {
1138                 .name   = "0.90.0",
1139                 .owner  = THIS_MODULE,
1140                 .load_super     = super_90_load,
1141                 .validate_super = super_90_validate,
1142                 .sync_super     = super_90_sync,
1143         },
1144         [1] = {
1145                 .name   = "md-1",
1146                 .owner  = THIS_MODULE,
1147                 .load_super     = super_1_load,
1148                 .validate_super = super_1_validate,
1149                 .sync_super     = super_1_sync,
1150         },
1151 };
1152
1153 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1154 {
1155         struct list_head *tmp;
1156         mdk_rdev_t *rdev;
1157
1158         ITERATE_RDEV(mddev,rdev,tmp)
1159                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1160                         return rdev;
1161
1162         return NULL;
1163 }
1164
1165 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1166 {
1167         struct list_head *tmp;
1168         mdk_rdev_t *rdev;
1169
1170         ITERATE_RDEV(mddev1,rdev,tmp)
1171                 if (match_dev_unit(mddev2, rdev))
1172                         return 1;
1173
1174         return 0;
1175 }
1176
1177 static LIST_HEAD(pending_raid_disks);
1178
1179 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1180 {
1181         mdk_rdev_t *same_pdev;
1182         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1183
1184         if (rdev->mddev) {
1185                 MD_BUG();
1186                 return -EINVAL;
1187         }
1188         same_pdev = match_dev_unit(mddev, rdev);
1189         if (same_pdev)
1190                 printk(KERN_WARNING
1191                         "%s: WARNING: %s appears to be on the same physical"
1192                         " disk as %s. True\n     protection against single-disk"
1193                         " failure might be compromised.\n",
1194                         mdname(mddev), bdevname(rdev->bdev,b),
1195                         bdevname(same_pdev->bdev,b2));
1196
1197         /* Verify rdev->desc_nr is unique.
1198          * If it is -1, assign a free number, else
1199          * check number is not in use
1200          */
1201         if (rdev->desc_nr < 0) {
1202                 int choice = 0;
1203                 if (mddev->pers) choice = mddev->raid_disks;
1204                 while (find_rdev_nr(mddev, choice))
1205                         choice++;
1206                 rdev->desc_nr = choice;
1207         } else {
1208                 if (find_rdev_nr(mddev, rdev->desc_nr))
1209                         return -EBUSY;
1210         }
1211         bdevname(rdev->bdev,b);
1212         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1213                 return -ENOMEM;
1214
1215         list_add(&rdev->same_set, &mddev->disks);
1216         rdev->mddev = mddev;
1217         printk(KERN_INFO "md: bind<%s>\n", b);
1218
1219         rdev->kobj.parent = &mddev->kobj;
1220         kobject_add(&rdev->kobj);
1221
1222         sysfs_create_link(&rdev->kobj, &rdev->bdev->bd_disk->kobj, "block");
1223         return 0;
1224 }
1225
1226 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1227 {
1228         char b[BDEVNAME_SIZE];
1229         if (!rdev->mddev) {
1230                 MD_BUG();
1231                 return;
1232         }
1233         list_del_init(&rdev->same_set);
1234         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1235         rdev->mddev = NULL;
1236         sysfs_remove_link(&rdev->kobj, "block");
1237         kobject_del(&rdev->kobj);
1238 }
1239
1240 /*
1241  * prevent the device from being mounted, repartitioned or
1242  * otherwise reused by a RAID array (or any other kernel
1243  * subsystem), by bd_claiming the device.
1244  */
1245 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1246 {
1247         int err = 0;
1248         struct block_device *bdev;
1249         char b[BDEVNAME_SIZE];
1250
1251         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1252         if (IS_ERR(bdev)) {
1253                 printk(KERN_ERR "md: could not open %s.\n",
1254                         __bdevname(dev, b));
1255                 return PTR_ERR(bdev);
1256         }
1257         err = bd_claim(bdev, rdev);
1258         if (err) {
1259                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1260                         bdevname(bdev, b));
1261                 blkdev_put(bdev);
1262                 return err;
1263         }
1264         rdev->bdev = bdev;
1265         return err;
1266 }
1267
1268 static void unlock_rdev(mdk_rdev_t *rdev)
1269 {
1270         struct block_device *bdev = rdev->bdev;
1271         rdev->bdev = NULL;
1272         if (!bdev)
1273                 MD_BUG();
1274         bd_release(bdev);
1275         blkdev_put(bdev);
1276 }
1277
1278 void md_autodetect_dev(dev_t dev);
1279
1280 static void export_rdev(mdk_rdev_t * rdev)
1281 {
1282         char b[BDEVNAME_SIZE];
1283         printk(KERN_INFO "md: export_rdev(%s)\n",
1284                 bdevname(rdev->bdev,b));
1285         if (rdev->mddev)
1286                 MD_BUG();
1287         free_disk_sb(rdev);
1288         list_del_init(&rdev->same_set);
1289 #ifndef MODULE
1290         md_autodetect_dev(rdev->bdev->bd_dev);
1291 #endif
1292         unlock_rdev(rdev);
1293         kobject_put(&rdev->kobj);
1294 }
1295
1296 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1297 {
1298         unbind_rdev_from_array(rdev);
1299         export_rdev(rdev);
1300 }
1301
1302 static void export_array(mddev_t *mddev)
1303 {
1304         struct list_head *tmp;
1305         mdk_rdev_t *rdev;
1306
1307         ITERATE_RDEV(mddev,rdev,tmp) {
1308                 if (!rdev->mddev) {
1309                         MD_BUG();
1310                         continue;
1311                 }
1312                 kick_rdev_from_array(rdev);
1313         }
1314         if (!list_empty(&mddev->disks))
1315                 MD_BUG();
1316         mddev->raid_disks = 0;
1317         mddev->major_version = 0;
1318 }
1319
1320 static void print_desc(mdp_disk_t *desc)
1321 {
1322         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1323                 desc->major,desc->minor,desc->raid_disk,desc->state);
1324 }
1325
1326 static void print_sb(mdp_super_t *sb)
1327 {
1328         int i;
1329
1330         printk(KERN_INFO
1331                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1332                 sb->major_version, sb->minor_version, sb->patch_version,
1333                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1334                 sb->ctime);
1335         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1336                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1337                 sb->md_minor, sb->layout, sb->chunk_size);
1338         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1339                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1340                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1341                 sb->failed_disks, sb->spare_disks,
1342                 sb->sb_csum, (unsigned long)sb->events_lo);
1343
1344         printk(KERN_INFO);
1345         for (i = 0; i < MD_SB_DISKS; i++) {
1346                 mdp_disk_t *desc;
1347
1348                 desc = sb->disks + i;
1349                 if (desc->number || desc->major || desc->minor ||
1350                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1351                         printk("     D %2d: ", i);
1352                         print_desc(desc);
1353                 }
1354         }
1355         printk(KERN_INFO "md:     THIS: ");
1356         print_desc(&sb->this_disk);
1357
1358 }
1359
1360 static void print_rdev(mdk_rdev_t *rdev)
1361 {
1362         char b[BDEVNAME_SIZE];
1363         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1364                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1365                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1366                 rdev->desc_nr);
1367         if (rdev->sb_loaded) {
1368                 printk(KERN_INFO "md: rdev superblock:\n");
1369                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1370         } else
1371                 printk(KERN_INFO "md: no rdev superblock!\n");
1372 }
1373
1374 void md_print_devices(void)
1375 {
1376         struct list_head *tmp, *tmp2;
1377         mdk_rdev_t *rdev;
1378         mddev_t *mddev;
1379         char b[BDEVNAME_SIZE];
1380
1381         printk("\n");
1382         printk("md:     **********************************\n");
1383         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1384         printk("md:     **********************************\n");
1385         ITERATE_MDDEV(mddev,tmp) {
1386
1387                 if (mddev->bitmap)
1388                         bitmap_print_sb(mddev->bitmap);
1389                 else
1390                         printk("%s: ", mdname(mddev));
1391                 ITERATE_RDEV(mddev,rdev,tmp2)
1392                         printk("<%s>", bdevname(rdev->bdev,b));
1393                 printk("\n");
1394
1395                 ITERATE_RDEV(mddev,rdev,tmp2)
1396                         print_rdev(rdev);
1397         }
1398         printk("md:     **********************************\n");
1399         printk("\n");
1400 }
1401
1402
1403 static void sync_sbs(mddev_t * mddev)
1404 {
1405         mdk_rdev_t *rdev;
1406         struct list_head *tmp;
1407
1408         ITERATE_RDEV(mddev,rdev,tmp) {
1409                 super_types[mddev->major_version].
1410                         sync_super(mddev, rdev);
1411                 rdev->sb_loaded = 1;
1412         }
1413 }
1414
1415 static void md_update_sb(mddev_t * mddev)
1416 {
1417         int err;
1418         struct list_head *tmp;
1419         mdk_rdev_t *rdev;
1420         int sync_req;
1421
1422 repeat:
1423         spin_lock_irq(&mddev->write_lock);
1424         sync_req = mddev->in_sync;
1425         mddev->utime = get_seconds();
1426         mddev->events ++;
1427
1428         if (!mddev->events) {
1429                 /*
1430                  * oops, this 64-bit counter should never wrap.
1431                  * Either we are in around ~1 trillion A.C., assuming
1432                  * 1 reboot per second, or we have a bug:
1433                  */
1434                 MD_BUG();
1435                 mddev->events --;
1436         }
1437         mddev->sb_dirty = 2;
1438         sync_sbs(mddev);
1439
1440         /*
1441          * do not write anything to disk if using
1442          * nonpersistent superblocks
1443          */
1444         if (!mddev->persistent) {
1445                 mddev->sb_dirty = 0;
1446                 spin_unlock_irq(&mddev->write_lock);
1447                 wake_up(&mddev->sb_wait);
1448                 return;
1449         }
1450         spin_unlock_irq(&mddev->write_lock);
1451
1452         dprintk(KERN_INFO
1453                 "md: updating %s RAID superblock on device (in sync %d)\n",
1454                 mdname(mddev),mddev->in_sync);
1455
1456         err = bitmap_update_sb(mddev->bitmap);
1457         ITERATE_RDEV(mddev,rdev,tmp) {
1458                 char b[BDEVNAME_SIZE];
1459                 dprintk(KERN_INFO "md: ");
1460                 if (test_bit(Faulty, &rdev->flags))
1461                         dprintk("(skipping faulty ");
1462
1463                 dprintk("%s ", bdevname(rdev->bdev,b));
1464                 if (!test_bit(Faulty, &rdev->flags)) {
1465                         md_super_write(mddev,rdev,
1466                                        rdev->sb_offset<<1, rdev->sb_size,
1467                                        rdev->sb_page);
1468                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1469                                 bdevname(rdev->bdev,b),
1470                                 (unsigned long long)rdev->sb_offset);
1471
1472                 } else
1473                         dprintk(")\n");
1474                 if (mddev->level == LEVEL_MULTIPATH)
1475                         /* only need to write one superblock... */
1476                         break;
1477         }
1478         md_super_wait(mddev);
1479         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1480
1481         spin_lock_irq(&mddev->write_lock);
1482         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1483                 /* have to write it out again */
1484                 spin_unlock_irq(&mddev->write_lock);
1485                 goto repeat;
1486         }
1487         mddev->sb_dirty = 0;
1488         spin_unlock_irq(&mddev->write_lock);
1489         wake_up(&mddev->sb_wait);
1490
1491 }
1492
1493 struct rdev_sysfs_entry {
1494         struct attribute attr;
1495         ssize_t (*show)(mdk_rdev_t *, char *);
1496         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1497 };
1498
1499 static ssize_t
1500 rdev_show_state(mdk_rdev_t *rdev, char *page)
1501 {
1502         char *sep = "";
1503         int len=0;
1504
1505         if (test_bit(Faulty, &rdev->flags)) {
1506                 len+= sprintf(page+len, "%sfaulty",sep);
1507                 sep = ",";
1508         }
1509         if (test_bit(In_sync, &rdev->flags)) {
1510                 len += sprintf(page+len, "%sin_sync",sep);
1511                 sep = ",";
1512         }
1513         if (!test_bit(Faulty, &rdev->flags) &&
1514             !test_bit(In_sync, &rdev->flags)) {
1515                 len += sprintf(page+len, "%sspare", sep);
1516                 sep = ",";
1517         }
1518         return len+sprintf(page+len, "\n");
1519 }
1520
1521 static struct rdev_sysfs_entry rdev_state = {
1522         .attr = {.name = "state", .mode = S_IRUGO },
1523         .show = rdev_show_state,
1524 };
1525
1526 static ssize_t
1527 rdev_show_super(mdk_rdev_t *rdev, char *page)
1528 {
1529         if (rdev->sb_loaded && rdev->sb_size) {
1530                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1531                 return rdev->sb_size;
1532         } else
1533                 return 0;
1534 }
1535 static struct rdev_sysfs_entry rdev_super = {
1536         .attr = {.name = "super", .mode = S_IRUGO },
1537         .show = rdev_show_super,
1538 };
1539 static struct attribute *rdev_default_attrs[] = {
1540         &rdev_state.attr,
1541         &rdev_super.attr,
1542         NULL,
1543 };
1544 static ssize_t
1545 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1546 {
1547         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1548         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1549
1550         if (!entry->show)
1551                 return -EIO;
1552         return entry->show(rdev, page);
1553 }
1554
1555 static ssize_t
1556 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1557               const char *page, size_t length)
1558 {
1559         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1560         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1561
1562         if (!entry->store)
1563                 return -EIO;
1564         return entry->store(rdev, page, length);
1565 }
1566
1567 static void rdev_free(struct kobject *ko)
1568 {
1569         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1570         kfree(rdev);
1571 }
1572 static struct sysfs_ops rdev_sysfs_ops = {
1573         .show           = rdev_attr_show,
1574         .store          = rdev_attr_store,
1575 };
1576 static struct kobj_type rdev_ktype = {
1577         .release        = rdev_free,
1578         .sysfs_ops      = &rdev_sysfs_ops,
1579         .default_attrs  = rdev_default_attrs,
1580 };
1581
1582 /*
1583  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1584  *
1585  * mark the device faulty if:
1586  *
1587  *   - the device is nonexistent (zero size)
1588  *   - the device has no valid superblock
1589  *
1590  * a faulty rdev _never_ has rdev->sb set.
1591  */
1592 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1593 {
1594         char b[BDEVNAME_SIZE];
1595         int err;
1596         mdk_rdev_t *rdev;
1597         sector_t size;
1598
1599         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1600         if (!rdev) {
1601                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1602                 return ERR_PTR(-ENOMEM);
1603         }
1604         memset(rdev, 0, sizeof(*rdev));
1605
1606         if ((err = alloc_disk_sb(rdev)))
1607                 goto abort_free;
1608
1609         err = lock_rdev(rdev, newdev);
1610         if (err)
1611                 goto abort_free;
1612
1613         rdev->kobj.parent = NULL;
1614         rdev->kobj.ktype = &rdev_ktype;
1615         kobject_init(&rdev->kobj);
1616
1617         rdev->desc_nr = -1;
1618         rdev->flags = 0;
1619         rdev->data_offset = 0;
1620         atomic_set(&rdev->nr_pending, 0);
1621         atomic_set(&rdev->read_errors, 0);
1622
1623         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1624         if (!size) {
1625                 printk(KERN_WARNING
1626                         "md: %s has zero or unknown size, marking faulty!\n",
1627                         bdevname(rdev->bdev,b));
1628                 err = -EINVAL;
1629                 goto abort_free;
1630         }
1631
1632         if (super_format >= 0) {
1633                 err = super_types[super_format].
1634                         load_super(rdev, NULL, super_minor);
1635                 if (err == -EINVAL) {
1636                         printk(KERN_WARNING
1637                                 "md: %s has invalid sb, not importing!\n",
1638                                 bdevname(rdev->bdev,b));
1639                         goto abort_free;
1640                 }
1641                 if (err < 0) {
1642                         printk(KERN_WARNING
1643                                 "md: could not read %s's sb, not importing!\n",
1644                                 bdevname(rdev->bdev,b));
1645                         goto abort_free;
1646                 }
1647         }
1648         INIT_LIST_HEAD(&rdev->same_set);
1649
1650         return rdev;
1651
1652 abort_free:
1653         if (rdev->sb_page) {
1654                 if (rdev->bdev)
1655                         unlock_rdev(rdev);
1656                 free_disk_sb(rdev);
1657         }
1658         kfree(rdev);
1659         return ERR_PTR(err);
1660 }
1661
1662 /*
1663  * Check a full RAID array for plausibility
1664  */
1665
1666
1667 static void analyze_sbs(mddev_t * mddev)
1668 {
1669         int i;
1670         struct list_head *tmp;
1671         mdk_rdev_t *rdev, *freshest;
1672         char b[BDEVNAME_SIZE];
1673
1674         freshest = NULL;
1675         ITERATE_RDEV(mddev,rdev,tmp)
1676                 switch (super_types[mddev->major_version].
1677                         load_super(rdev, freshest, mddev->minor_version)) {
1678                 case 1:
1679                         freshest = rdev;
1680                         break;
1681                 case 0:
1682                         break;
1683                 default:
1684                         printk( KERN_ERR \
1685                                 "md: fatal superblock inconsistency in %s"
1686                                 " -- removing from array\n",
1687                                 bdevname(rdev->bdev,b));
1688                         kick_rdev_from_array(rdev);
1689                 }
1690
1691
1692         super_types[mddev->major_version].
1693                 validate_super(mddev, freshest);
1694
1695         i = 0;
1696         ITERATE_RDEV(mddev,rdev,tmp) {
1697                 if (rdev != freshest)
1698                         if (super_types[mddev->major_version].
1699                             validate_super(mddev, rdev)) {
1700                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1701                                         " from array!\n",
1702                                         bdevname(rdev->bdev,b));
1703                                 kick_rdev_from_array(rdev);
1704                                 continue;
1705                         }
1706                 if (mddev->level == LEVEL_MULTIPATH) {
1707                         rdev->desc_nr = i++;
1708                         rdev->raid_disk = rdev->desc_nr;
1709                         set_bit(In_sync, &rdev->flags);
1710                 }
1711         }
1712
1713
1714
1715         if (mddev->recovery_cp != MaxSector &&
1716             mddev->level >= 1)
1717                 printk(KERN_ERR "md: %s: raid array is not clean"
1718                        " -- starting background reconstruction\n",
1719                        mdname(mddev));
1720
1721 }
1722
1723 static ssize_t
1724 md_show_level(mddev_t *mddev, char *page)
1725 {
1726         mdk_personality_t *p = mddev->pers;
1727         if (p == NULL)
1728                 return 0;
1729         if (mddev->level >= 0)
1730                 return sprintf(page, "RAID-%d\n", mddev->level);
1731         else
1732                 return sprintf(page, "%s\n", p->name);
1733 }
1734
1735 static struct md_sysfs_entry md_level = {
1736         .attr = {.name = "level", .mode = S_IRUGO },
1737         .show = md_show_level,
1738 };
1739
1740 static ssize_t
1741 md_show_rdisks(mddev_t *mddev, char *page)
1742 {
1743         return sprintf(page, "%d\n", mddev->raid_disks);
1744 }
1745
1746 static struct md_sysfs_entry md_raid_disks = {
1747         .attr = {.name = "raid_disks", .mode = S_IRUGO },
1748         .show = md_show_rdisks,
1749 };
1750
1751 static ssize_t
1752 md_show_scan(mddev_t *mddev, char *page)
1753 {
1754         char *type = "none";
1755         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1756             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1757                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1758                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1759                                 type = "resync";
1760                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1761                                 type = "check";
1762                         else
1763                                 type = "repair";
1764                 } else
1765                         type = "recover";
1766         }
1767         return sprintf(page, "%s\n", type);
1768 }
1769
1770 static ssize_t
1771 md_store_scan(mddev_t *mddev, const char *page, size_t len)
1772 {
1773         int canscan=0;
1774
1775         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1776             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1777                 return -EBUSY;
1778         down(&mddev->reconfig_sem);
1779         if (mddev->pers && mddev->pers->sync_request)
1780                 canscan=1;
1781         up(&mddev->reconfig_sem);
1782         if (!canscan)
1783                 return -EINVAL;
1784
1785         if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1786                 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1787         else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1788                 return -EINVAL;
1789         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1790         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1791         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1792         md_wakeup_thread(mddev->thread);
1793         return len;
1794 }
1795
1796 static ssize_t
1797 md_show_mismatch(mddev_t *mddev, char *page)
1798 {
1799         return sprintf(page, "%llu\n",
1800                        (unsigned long long) mddev->resync_mismatches);
1801 }
1802
1803 static struct md_sysfs_entry md_scan_mode = {
1804         .attr = {.name = "scan_mode", .mode = S_IRUGO|S_IWUSR },
1805         .show = md_show_scan,
1806         .store = md_store_scan,
1807 };
1808
1809 static struct md_sysfs_entry md_mismatches = {
1810         .attr = {.name = "mismatch_cnt", .mode = S_IRUGO },
1811         .show = md_show_mismatch,
1812 };
1813
1814 static struct attribute *md_default_attrs[] = {
1815         &md_level.attr,
1816         &md_raid_disks.attr,
1817         &md_scan_mode.attr,
1818         &md_mismatches.attr,
1819         NULL,
1820 };
1821
1822 static ssize_t
1823 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1824 {
1825         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1826         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1827
1828         if (!entry->show)
1829                 return -EIO;
1830         return entry->show(mddev, page);
1831 }
1832
1833 static ssize_t
1834 md_attr_store(struct kobject *kobj, struct attribute *attr,
1835               const char *page, size_t length)
1836 {
1837         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1838         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1839
1840         if (!entry->store)
1841                 return -EIO;
1842         return entry->store(mddev, page, length);
1843 }
1844
1845 static void md_free(struct kobject *ko)
1846 {
1847         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1848         kfree(mddev);
1849 }
1850
1851 static struct sysfs_ops md_sysfs_ops = {
1852         .show   = md_attr_show,
1853         .store  = md_attr_store,
1854 };
1855 static struct kobj_type md_ktype = {
1856         .release        = md_free,
1857         .sysfs_ops      = &md_sysfs_ops,
1858         .default_attrs  = md_default_attrs,
1859 };
1860
1861 int mdp_major = 0;
1862
1863 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1864 {
1865         static DECLARE_MUTEX(disks_sem);
1866         mddev_t *mddev = mddev_find(dev);
1867         struct gendisk *disk;
1868         int partitioned = (MAJOR(dev) != MD_MAJOR);
1869         int shift = partitioned ? MdpMinorShift : 0;
1870         int unit = MINOR(dev) >> shift;
1871
1872         if (!mddev)
1873                 return NULL;
1874
1875         down(&disks_sem);
1876         if (mddev->gendisk) {
1877                 up(&disks_sem);
1878                 mddev_put(mddev);
1879                 return NULL;
1880         }
1881         disk = alloc_disk(1 << shift);
1882         if (!disk) {
1883                 up(&disks_sem);
1884                 mddev_put(mddev);
1885                 return NULL;
1886         }
1887         disk->major = MAJOR(dev);
1888         disk->first_minor = unit << shift;
1889         if (partitioned) {
1890                 sprintf(disk->disk_name, "md_d%d", unit);
1891                 sprintf(disk->devfs_name, "md/d%d", unit);
1892         } else {
1893                 sprintf(disk->disk_name, "md%d", unit);
1894                 sprintf(disk->devfs_name, "md/%d", unit);
1895         }
1896         disk->fops = &md_fops;
1897         disk->private_data = mddev;
1898         disk->queue = mddev->queue;
1899         add_disk(disk);
1900         mddev->gendisk = disk;
1901         up(&disks_sem);
1902         mddev->kobj.parent = &disk->kobj;
1903         mddev->kobj.k_name = NULL;
1904         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1905         mddev->kobj.ktype = &md_ktype;
1906         kobject_register(&mddev->kobj);
1907         return NULL;
1908 }
1909
1910 void md_wakeup_thread(mdk_thread_t *thread);
1911
1912 static void md_safemode_timeout(unsigned long data)
1913 {
1914         mddev_t *mddev = (mddev_t *) data;
1915
1916         mddev->safemode = 1;
1917         md_wakeup_thread(mddev->thread);
1918 }
1919
1920
1921 static int do_md_run(mddev_t * mddev)
1922 {
1923         int pnum, err;
1924         int chunk_size;
1925         struct list_head *tmp;
1926         mdk_rdev_t *rdev;
1927         struct gendisk *disk;
1928         char b[BDEVNAME_SIZE];
1929
1930         if (list_empty(&mddev->disks))
1931                 /* cannot run an array with no devices.. */
1932                 return -EINVAL;
1933
1934         if (mddev->pers)
1935                 return -EBUSY;
1936
1937         /*
1938          * Analyze all RAID superblock(s)
1939          */
1940         if (!mddev->raid_disks)
1941                 analyze_sbs(mddev);
1942
1943         chunk_size = mddev->chunk_size;
1944         pnum = level_to_pers(mddev->level);
1945
1946         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1947                 if (!chunk_size) {
1948                         /*
1949                          * 'default chunksize' in the old md code used to
1950                          * be PAGE_SIZE, baaad.
1951                          * we abort here to be on the safe side. We don't
1952                          * want to continue the bad practice.
1953                          */
1954                         printk(KERN_ERR
1955                                 "no chunksize specified, see 'man raidtab'\n");
1956                         return -EINVAL;
1957                 }
1958                 if (chunk_size > MAX_CHUNK_SIZE) {
1959                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1960                                 chunk_size, MAX_CHUNK_SIZE);
1961                         return -EINVAL;
1962                 }
1963                 /*
1964                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1965                  */
1966                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1967                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1968                         return -EINVAL;
1969                 }
1970                 if (chunk_size < PAGE_SIZE) {
1971                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1972                                 chunk_size, PAGE_SIZE);
1973                         return -EINVAL;
1974                 }
1975
1976                 /* devices must have minimum size of one chunk */
1977                 ITERATE_RDEV(mddev,rdev,tmp) {
1978                         if (test_bit(Faulty, &rdev->flags))
1979                                 continue;
1980                         if (rdev->size < chunk_size / 1024) {
1981                                 printk(KERN_WARNING
1982                                         "md: Dev %s smaller than chunk_size:"
1983                                         " %lluk < %dk\n",
1984                                         bdevname(rdev->bdev,b),
1985                                         (unsigned long long)rdev->size,
1986                                         chunk_size / 1024);
1987                                 return -EINVAL;
1988                         }
1989                 }
1990         }
1991
1992 #ifdef CONFIG_KMOD
1993         if (!pers[pnum])
1994         {
1995                 request_module("md-personality-%d", pnum);
1996         }
1997 #endif
1998
1999         /*
2000          * Drop all container device buffers, from now on
2001          * the only valid external interface is through the md
2002          * device.
2003          * Also find largest hardsector size
2004          */
2005         ITERATE_RDEV(mddev,rdev,tmp) {
2006                 if (test_bit(Faulty, &rdev->flags))
2007                         continue;
2008                 sync_blockdev(rdev->bdev);
2009                 invalidate_bdev(rdev->bdev, 0);
2010         }
2011
2012         md_probe(mddev->unit, NULL, NULL);
2013         disk = mddev->gendisk;
2014         if (!disk)
2015                 return -ENOMEM;
2016
2017         spin_lock(&pers_lock);
2018         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2019                 spin_unlock(&pers_lock);
2020                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2021                        pnum);
2022                 return -EINVAL;
2023         }
2024
2025         mddev->pers = pers[pnum];
2026         spin_unlock(&pers_lock);
2027
2028         mddev->recovery = 0;
2029         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2030         mddev->barriers_work = 1;
2031
2032         /* before we start the array running, initialise the bitmap */
2033         err = bitmap_create(mddev);
2034         if (err)
2035                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2036                         mdname(mddev), err);
2037         else
2038                 err = mddev->pers->run(mddev);
2039         if (err) {
2040                 printk(KERN_ERR "md: pers->run() failed ...\n");
2041                 module_put(mddev->pers->owner);
2042                 mddev->pers = NULL;
2043                 bitmap_destroy(mddev);
2044                 return err;
2045         }
2046         atomic_set(&mddev->writes_pending,0);
2047         mddev->safemode = 0;
2048         mddev->safemode_timer.function = md_safemode_timeout;
2049         mddev->safemode_timer.data = (unsigned long) mddev;
2050         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2051         mddev->in_sync = 1;
2052
2053         ITERATE_RDEV(mddev,rdev,tmp)
2054                 if (rdev->raid_disk >= 0) {
2055                         char nm[20];
2056                         sprintf(nm, "rd%d", rdev->raid_disk);
2057                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2058                 }
2059
2060         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2061         md_wakeup_thread(mddev->thread);
2062
2063         if (mddev->sb_dirty)
2064                 md_update_sb(mddev);
2065
2066         set_capacity(disk, mddev->array_size<<1);
2067
2068         /* If we call blk_queue_make_request here, it will
2069          * re-initialise max_sectors etc which may have been
2070          * refined inside -> run.  So just set the bits we need to set.
2071          * Most initialisation happended when we called
2072          * blk_queue_make_request(..., md_fail_request)
2073          * earlier.
2074          */
2075         mddev->queue->queuedata = mddev;
2076         mddev->queue->make_request_fn = mddev->pers->make_request;
2077
2078         mddev->changed = 1;
2079         return 0;
2080 }
2081
2082 static int restart_array(mddev_t *mddev)
2083 {
2084         struct gendisk *disk = mddev->gendisk;
2085         int err;
2086
2087         /*
2088          * Complain if it has no devices
2089          */
2090         err = -ENXIO;
2091         if (list_empty(&mddev->disks))
2092                 goto out;
2093
2094         if (mddev->pers) {
2095                 err = -EBUSY;
2096                 if (!mddev->ro)
2097                         goto out;
2098
2099                 mddev->safemode = 0;
2100                 mddev->ro = 0;
2101                 set_disk_ro(disk, 0);
2102
2103                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2104                         mdname(mddev));
2105                 /*
2106                  * Kick recovery or resync if necessary
2107                  */
2108                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2109                 md_wakeup_thread(mddev->thread);
2110                 err = 0;
2111         } else {
2112                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2113                         mdname(mddev));
2114                 err = -EINVAL;
2115         }
2116
2117 out:
2118         return err;
2119 }
2120
2121 static int do_md_stop(mddev_t * mddev, int ro)
2122 {
2123         int err = 0;
2124         struct gendisk *disk = mddev->gendisk;
2125
2126         if (mddev->pers) {
2127                 if (atomic_read(&mddev->active)>2) {
2128                         printk("md: %s still in use.\n",mdname(mddev));
2129                         return -EBUSY;
2130                 }
2131
2132                 if (mddev->sync_thread) {
2133                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2134                         md_unregister_thread(mddev->sync_thread);
2135                         mddev->sync_thread = NULL;
2136                 }
2137
2138                 del_timer_sync(&mddev->safemode_timer);
2139
2140                 invalidate_partition(disk, 0);
2141
2142                 if (ro) {
2143                         err  = -ENXIO;
2144                         if (mddev->ro)
2145                                 goto out;
2146                         mddev->ro = 1;
2147                 } else {
2148                         bitmap_flush(mddev);
2149                         md_super_wait(mddev);
2150                         if (mddev->ro)
2151                                 set_disk_ro(disk, 0);
2152                         blk_queue_make_request(mddev->queue, md_fail_request);
2153                         mddev->pers->stop(mddev);
2154                         module_put(mddev->pers->owner);
2155                         mddev->pers = NULL;
2156                         if (mddev->ro)
2157                                 mddev->ro = 0;
2158                 }
2159                 if (!mddev->in_sync) {
2160                         /* mark array as shutdown cleanly */
2161                         mddev->in_sync = 1;
2162                         md_update_sb(mddev);
2163                 }
2164                 if (ro)
2165                         set_disk_ro(disk, 1);
2166         }
2167
2168         bitmap_destroy(mddev);
2169         if (mddev->bitmap_file) {
2170                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2171                 fput(mddev->bitmap_file);
2172                 mddev->bitmap_file = NULL;
2173         }
2174         mddev->bitmap_offset = 0;
2175
2176         /*
2177          * Free resources if final stop
2178          */
2179         if (!ro) {
2180                 mdk_rdev_t *rdev;
2181                 struct list_head *tmp;
2182                 struct gendisk *disk;
2183                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2184
2185                 ITERATE_RDEV(mddev,rdev,tmp)
2186                         if (rdev->raid_disk >= 0) {
2187                                 char nm[20];
2188                                 sprintf(nm, "rd%d", rdev->raid_disk);
2189                                 sysfs_remove_link(&mddev->kobj, nm);
2190                         }
2191
2192                 export_array(mddev);
2193
2194                 mddev->array_size = 0;
2195                 disk = mddev->gendisk;
2196                 if (disk)
2197                         set_capacity(disk, 0);
2198                 mddev->changed = 1;
2199         } else
2200                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2201                         mdname(mddev));
2202         err = 0;
2203 out:
2204         return err;
2205 }
2206
2207 static void autorun_array(mddev_t *mddev)
2208 {
2209         mdk_rdev_t *rdev;
2210         struct list_head *tmp;
2211         int err;
2212
2213         if (list_empty(&mddev->disks))
2214                 return;
2215
2216         printk(KERN_INFO "md: running: ");
2217
2218         ITERATE_RDEV(mddev,rdev,tmp) {
2219                 char b[BDEVNAME_SIZE];
2220                 printk("<%s>", bdevname(rdev->bdev,b));
2221         }
2222         printk("\n");
2223
2224         err = do_md_run (mddev);
2225         if (err) {
2226                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2227                 do_md_stop (mddev, 0);
2228         }
2229 }
2230
2231 /*
2232  * lets try to run arrays based on all disks that have arrived
2233  * until now. (those are in pending_raid_disks)
2234  *
2235  * the method: pick the first pending disk, collect all disks with
2236  * the same UUID, remove all from the pending list and put them into
2237  * the 'same_array' list. Then order this list based on superblock
2238  * update time (freshest comes first), kick out 'old' disks and
2239  * compare superblocks. If everything's fine then run it.
2240  *
2241  * If "unit" is allocated, then bump its reference count
2242  */
2243 static void autorun_devices(int part)
2244 {
2245         struct list_head candidates;
2246         struct list_head *tmp;
2247         mdk_rdev_t *rdev0, *rdev;
2248         mddev_t *mddev;
2249         char b[BDEVNAME_SIZE];
2250
2251         printk(KERN_INFO "md: autorun ...\n");
2252         while (!list_empty(&pending_raid_disks)) {
2253                 dev_t dev;
2254                 rdev0 = list_entry(pending_raid_disks.next,
2255                                          mdk_rdev_t, same_set);
2256
2257                 printk(KERN_INFO "md: considering %s ...\n",
2258                         bdevname(rdev0->bdev,b));
2259                 INIT_LIST_HEAD(&candidates);
2260                 ITERATE_RDEV_PENDING(rdev,tmp)
2261                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2262                                 printk(KERN_INFO "md:  adding %s ...\n",
2263                                         bdevname(rdev->bdev,b));
2264                                 list_move(&rdev->same_set, &candidates);
2265                         }
2266                 /*
2267                  * now we have a set of devices, with all of them having
2268                  * mostly sane superblocks. It's time to allocate the
2269                  * mddev.
2270                  */
2271                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2272                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2273                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2274                         break;
2275                 }
2276                 if (part)
2277                         dev = MKDEV(mdp_major,
2278                                     rdev0->preferred_minor << MdpMinorShift);
2279                 else
2280                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2281
2282                 md_probe(dev, NULL, NULL);
2283                 mddev = mddev_find(dev);
2284                 if (!mddev) {
2285                         printk(KERN_ERR
2286                                 "md: cannot allocate memory for md drive.\n");
2287                         break;
2288                 }
2289                 if (mddev_lock(mddev))
2290                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2291                                mdname(mddev));
2292                 else if (mddev->raid_disks || mddev->major_version
2293                          || !list_empty(&mddev->disks)) {
2294                         printk(KERN_WARNING
2295                                 "md: %s already running, cannot run %s\n",
2296                                 mdname(mddev), bdevname(rdev0->bdev,b));
2297                         mddev_unlock(mddev);
2298                 } else {
2299                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2300                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2301                                 list_del_init(&rdev->same_set);
2302                                 if (bind_rdev_to_array(rdev, mddev))
2303                                         export_rdev(rdev);
2304                         }
2305                         autorun_array(mddev);
2306                         mddev_unlock(mddev);
2307                 }
2308                 /* on success, candidates will be empty, on error
2309                  * it won't...
2310                  */
2311                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2312                         export_rdev(rdev);
2313                 mddev_put(mddev);
2314         }
2315         printk(KERN_INFO "md: ... autorun DONE.\n");
2316 }
2317
2318 /*
2319  * import RAID devices based on one partition
2320  * if possible, the array gets run as well.
2321  */
2322
2323 static int autostart_array(dev_t startdev)
2324 {
2325         char b[BDEVNAME_SIZE];
2326         int err = -EINVAL, i;
2327         mdp_super_t *sb = NULL;
2328         mdk_rdev_t *start_rdev = NULL, *rdev;
2329
2330         start_rdev = md_import_device(startdev, 0, 0);
2331         if (IS_ERR(start_rdev))
2332                 return err;
2333
2334
2335         /* NOTE: this can only work for 0.90.0 superblocks */
2336         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2337         if (sb->major_version != 0 ||
2338             sb->minor_version != 90 ) {
2339                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2340                 export_rdev(start_rdev);
2341                 return err;
2342         }
2343
2344         if (test_bit(Faulty, &start_rdev->flags)) {
2345                 printk(KERN_WARNING
2346                         "md: can not autostart based on faulty %s!\n",
2347                         bdevname(start_rdev->bdev,b));
2348                 export_rdev(start_rdev);
2349                 return err;
2350         }
2351         list_add(&start_rdev->same_set, &pending_raid_disks);
2352
2353         for (i = 0; i < MD_SB_DISKS; i++) {
2354                 mdp_disk_t *desc = sb->disks + i;
2355                 dev_t dev = MKDEV(desc->major, desc->minor);
2356
2357                 if (!dev)
2358                         continue;
2359                 if (dev == startdev)
2360                         continue;
2361                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2362                         continue;
2363                 rdev = md_import_device(dev, 0, 0);
2364                 if (IS_ERR(rdev))
2365                         continue;
2366
2367                 list_add(&rdev->same_set, &pending_raid_disks);
2368         }
2369
2370         /*
2371          * possibly return codes
2372          */
2373         autorun_devices(0);
2374         return 0;
2375
2376 }
2377
2378
2379 static int get_version(void __user * arg)
2380 {
2381         mdu_version_t ver;
2382
2383         ver.major = MD_MAJOR_VERSION;
2384         ver.minor = MD_MINOR_VERSION;
2385         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2386
2387         if (copy_to_user(arg, &ver, sizeof(ver)))
2388                 return -EFAULT;
2389
2390         return 0;
2391 }
2392
2393 static int get_array_info(mddev_t * mddev, void __user * arg)
2394 {
2395         mdu_array_info_t info;
2396         int nr,working,active,failed,spare;
2397         mdk_rdev_t *rdev;
2398         struct list_head *tmp;
2399
2400         nr=working=active=failed=spare=0;
2401         ITERATE_RDEV(mddev,rdev,tmp) {
2402                 nr++;
2403                 if (test_bit(Faulty, &rdev->flags))
2404                         failed++;
2405                 else {
2406                         working++;
2407                         if (test_bit(In_sync, &rdev->flags))
2408                                 active++;
2409                         else
2410                                 spare++;
2411                 }
2412         }
2413
2414         info.major_version = mddev->major_version;
2415         info.minor_version = mddev->minor_version;
2416         info.patch_version = MD_PATCHLEVEL_VERSION;
2417         info.ctime         = mddev->ctime;
2418         info.level         = mddev->level;
2419         info.size          = mddev->size;
2420         info.nr_disks      = nr;
2421         info.raid_disks    = mddev->raid_disks;
2422         info.md_minor      = mddev->md_minor;
2423         info.not_persistent= !mddev->persistent;
2424
2425         info.utime         = mddev->utime;
2426         info.state         = 0;
2427         if (mddev->in_sync)
2428                 info.state = (1<<MD_SB_CLEAN);
2429         if (mddev->bitmap && mddev->bitmap_offset)
2430                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2431         info.active_disks  = active;
2432         info.working_disks = working;
2433         info.failed_disks  = failed;
2434         info.spare_disks   = spare;
2435
2436         info.layout        = mddev->layout;
2437         info.chunk_size    = mddev->chunk_size;
2438
2439         if (copy_to_user(arg, &info, sizeof(info)))
2440                 return -EFAULT;
2441
2442         return 0;
2443 }
2444
2445 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2446 {
2447         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2448         char *ptr, *buf = NULL;
2449         int err = -ENOMEM;
2450
2451         file = kmalloc(sizeof(*file), GFP_KERNEL);
2452         if (!file)
2453                 goto out;
2454
2455         /* bitmap disabled, zero the first byte and copy out */
2456         if (!mddev->bitmap || !mddev->bitmap->file) {
2457                 file->pathname[0] = '\0';
2458                 goto copy_out;
2459         }
2460
2461         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2462         if (!buf)
2463                 goto out;
2464
2465         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2466         if (!ptr)
2467                 goto out;
2468
2469         strcpy(file->pathname, ptr);
2470
2471 copy_out:
2472         err = 0;
2473         if (copy_to_user(arg, file, sizeof(*file)))
2474                 err = -EFAULT;
2475 out:
2476         kfree(buf);
2477         kfree(file);
2478         return err;
2479 }
2480
2481 static int get_disk_info(mddev_t * mddev, void __user * arg)
2482 {
2483         mdu_disk_info_t info;
2484         unsigned int nr;
2485         mdk_rdev_t *rdev;
2486
2487         if (copy_from_user(&info, arg, sizeof(info)))
2488                 return -EFAULT;
2489
2490         nr = info.number;
2491
2492         rdev = find_rdev_nr(mddev, nr);
2493         if (rdev) {
2494                 info.major = MAJOR(rdev->bdev->bd_dev);
2495                 info.minor = MINOR(rdev->bdev->bd_dev);
2496                 info.raid_disk = rdev->raid_disk;
2497                 info.state = 0;
2498                 if (test_bit(Faulty, &rdev->flags))
2499                         info.state |= (1<<MD_DISK_FAULTY);
2500                 else if (test_bit(In_sync, &rdev->flags)) {
2501                         info.state |= (1<<MD_DISK_ACTIVE);
2502                         info.state |= (1<<MD_DISK_SYNC);
2503                 }
2504                 if (test_bit(WriteMostly, &rdev->flags))
2505                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2506         } else {
2507                 info.major = info.minor = 0;
2508                 info.raid_disk = -1;
2509                 info.state = (1<<MD_DISK_REMOVED);
2510         }
2511
2512         if (copy_to_user(arg, &info, sizeof(info)))
2513                 return -EFAULT;
2514
2515         return 0;
2516 }
2517
2518 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2519 {
2520         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2521         mdk_rdev_t *rdev;
2522         dev_t dev = MKDEV(info->major,info->minor);
2523
2524         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2525                 return -EOVERFLOW;
2526
2527         if (!mddev->raid_disks) {
2528                 int err;
2529                 /* expecting a device which has a superblock */
2530                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2531                 if (IS_ERR(rdev)) {
2532                         printk(KERN_WARNING
2533                                 "md: md_import_device returned %ld\n",
2534                                 PTR_ERR(rdev));
2535                         return PTR_ERR(rdev);
2536                 }
2537                 if (!list_empty(&mddev->disks)) {
2538                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2539                                                         mdk_rdev_t, same_set);
2540                         int err = super_types[mddev->major_version]
2541                                 .load_super(rdev, rdev0, mddev->minor_version);
2542                         if (err < 0) {
2543                                 printk(KERN_WARNING
2544                                         "md: %s has different UUID to %s\n",
2545                                         bdevname(rdev->bdev,b),
2546                                         bdevname(rdev0->bdev,b2));
2547                                 export_rdev(rdev);
2548                                 return -EINVAL;
2549                         }
2550                 }
2551                 err = bind_rdev_to_array(rdev, mddev);
2552                 if (err)
2553                         export_rdev(rdev);
2554                 return err;
2555         }
2556
2557         /*
2558          * add_new_disk can be used once the array is assembled
2559          * to add "hot spares".  They must already have a superblock
2560          * written
2561          */
2562         if (mddev->pers) {
2563                 int err;
2564                 if (!mddev->pers->hot_add_disk) {
2565                         printk(KERN_WARNING
2566                                 "%s: personality does not support diskops!\n",
2567                                mdname(mddev));
2568                         return -EINVAL;
2569                 }
2570                 if (mddev->persistent)
2571                         rdev = md_import_device(dev, mddev->major_version,
2572                                                 mddev->minor_version);
2573                 else
2574                         rdev = md_import_device(dev, -1, -1);
2575                 if (IS_ERR(rdev)) {
2576                         printk(KERN_WARNING
2577                                 "md: md_import_device returned %ld\n",
2578                                 PTR_ERR(rdev));
2579                         return PTR_ERR(rdev);
2580                 }
2581                 /* set save_raid_disk if appropriate */
2582                 if (!mddev->persistent) {
2583                         if (info->state & (1<<MD_DISK_SYNC)  &&
2584                             info->raid_disk < mddev->raid_disks)
2585                                 rdev->raid_disk = info->raid_disk;
2586                         else
2587                                 rdev->raid_disk = -1;
2588                 } else
2589                         super_types[mddev->major_version].
2590                                 validate_super(mddev, rdev);
2591                 rdev->saved_raid_disk = rdev->raid_disk;
2592
2593                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2594                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2595                         set_bit(WriteMostly, &rdev->flags);
2596
2597                 rdev->raid_disk = -1;
2598                 err = bind_rdev_to_array(rdev, mddev);
2599                 if (err)
2600                         export_rdev(rdev);
2601
2602                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2603                 md_wakeup_thread(mddev->thread);
2604                 return err;
2605         }
2606
2607         /* otherwise, add_new_disk is only allowed
2608          * for major_version==0 superblocks
2609          */
2610         if (mddev->major_version != 0) {
2611                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2612                        mdname(mddev));
2613                 return -EINVAL;
2614         }
2615
2616         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2617                 int err;
2618                 rdev = md_import_device (dev, -1, 0);
2619                 if (IS_ERR(rdev)) {
2620                         printk(KERN_WARNING
2621                                 "md: error, md_import_device() returned %ld\n",
2622                                 PTR_ERR(rdev));
2623                         return PTR_ERR(rdev);
2624                 }
2625                 rdev->desc_nr = info->number;
2626                 if (info->raid_disk < mddev->raid_disks)
2627                         rdev->raid_disk = info->raid_disk;
2628                 else
2629                         rdev->raid_disk = -1;
2630
2631                 rdev->flags = 0;
2632
2633                 if (rdev->raid_disk < mddev->raid_disks)
2634                         if (info->state & (1<<MD_DISK_SYNC))
2635                                 set_bit(In_sync, &rdev->flags);
2636
2637                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2638                         set_bit(WriteMostly, &rdev->flags);
2639
2640                 err = bind_rdev_to_array(rdev, mddev);
2641                 if (err) {
2642                         export_rdev(rdev);
2643                         return err;
2644                 }
2645
2646                 if (!mddev->persistent) {
2647                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2648                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2649                 } else
2650                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2651                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2652
2653                 if (!mddev->size || (mddev->size > rdev->size))
2654                         mddev->size = rdev->size;
2655         }
2656
2657         return 0;
2658 }
2659
2660 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2661 {
2662         char b[BDEVNAME_SIZE];
2663         mdk_rdev_t *rdev;
2664
2665         if (!mddev->pers)
2666                 return -ENODEV;
2667
2668         rdev = find_rdev(mddev, dev);
2669         if (!rdev)
2670                 return -ENXIO;
2671
2672         if (rdev->raid_disk >= 0)
2673                 goto busy;
2674
2675         kick_rdev_from_array(rdev);
2676         md_update_sb(mddev);
2677
2678         return 0;
2679 busy:
2680         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2681                 bdevname(rdev->bdev,b), mdname(mddev));
2682         return -EBUSY;
2683 }
2684
2685 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2686 {
2687         char b[BDEVNAME_SIZE];
2688         int err;
2689         unsigned int size;
2690         mdk_rdev_t *rdev;
2691
2692         if (!mddev->pers)
2693                 return -ENODEV;
2694
2695         if (mddev->major_version != 0) {
2696                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2697                         " version-0 superblocks.\n",
2698                         mdname(mddev));
2699                 return -EINVAL;
2700         }
2701         if (!mddev->pers->hot_add_disk) {
2702                 printk(KERN_WARNING
2703                         "%s: personality does not support diskops!\n",
2704                         mdname(mddev));
2705                 return -EINVAL;
2706         }
2707
2708         rdev = md_import_device (dev, -1, 0);
2709         if (IS_ERR(rdev)) {
2710                 printk(KERN_WARNING
2711                         "md: error, md_import_device() returned %ld\n",
2712                         PTR_ERR(rdev));
2713                 return -EINVAL;
2714         }
2715
2716         if (mddev->persistent)
2717                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2718         else
2719                 rdev->sb_offset =
2720                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2721
2722         size = calc_dev_size(rdev, mddev->chunk_size);
2723         rdev->size = size;
2724
2725         if (size < mddev->size) {
2726                 printk(KERN_WARNING
2727                         "%s: disk size %llu blocks < array size %llu\n",
2728                         mdname(mddev), (unsigned long long)size,
2729                         (unsigned long long)mddev->size);
2730                 err = -ENOSPC;
2731                 goto abort_export;
2732         }
2733
2734         if (test_bit(Faulty, &rdev->flags)) {
2735                 printk(KERN_WARNING
2736                         "md: can not hot-add faulty %s disk to %s!\n",
2737                         bdevname(rdev->bdev,b), mdname(mddev));
2738                 err = -EINVAL;
2739                 goto abort_export;
2740         }
2741         clear_bit(In_sync, &rdev->flags);
2742         rdev->desc_nr = -1;
2743         bind_rdev_to_array(rdev, mddev);
2744
2745         /*
2746          * The rest should better be atomic, we can have disk failures
2747          * noticed in interrupt contexts ...
2748          */
2749
2750         if (rdev->desc_nr == mddev->max_disks) {
2751                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2752                         mdname(mddev));
2753                 err = -EBUSY;
2754                 goto abort_unbind_export;
2755         }
2756
2757         rdev->raid_disk = -1;
2758
2759         md_update_sb(mddev);
2760
2761         /*
2762          * Kick recovery, maybe this spare has to be added to the
2763          * array immediately.
2764          */
2765         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2766         md_wakeup_thread(mddev->thread);
2767
2768         return 0;
2769
2770 abort_unbind_export:
2771         unbind_rdev_from_array(rdev);
2772
2773 abort_export:
2774         export_rdev(rdev);
2775         return err;
2776 }
2777
2778 /* similar to deny_write_access, but accounts for our holding a reference
2779  * to the file ourselves */
2780 static int deny_bitmap_write_access(struct file * file)
2781 {
2782         struct inode *inode = file->f_mapping->host;
2783
2784         spin_lock(&inode->i_lock);
2785         if (atomic_read(&inode->i_writecount) > 1) {
2786                 spin_unlock(&inode->i_lock);
2787                 return -ETXTBSY;
2788         }
2789         atomic_set(&inode->i_writecount, -1);
2790         spin_unlock(&inode->i_lock);
2791
2792         return 0;
2793 }
2794
2795 static int set_bitmap_file(mddev_t *mddev, int fd)
2796 {
2797         int err;
2798
2799         if (mddev->pers) {
2800                 if (!mddev->pers->quiesce)
2801                         return -EBUSY;
2802                 if (mddev->recovery || mddev->sync_thread)
2803                         return -EBUSY;
2804                 /* we should be able to change the bitmap.. */
2805         }
2806
2807
2808         if (fd >= 0) {
2809                 if (mddev->bitmap)
2810                         return -EEXIST; /* cannot add when bitmap is present */
2811                 mddev->bitmap_file = fget(fd);
2812
2813                 if (mddev->bitmap_file == NULL) {
2814                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2815                                mdname(mddev));
2816                         return -EBADF;
2817                 }
2818
2819                 err = deny_bitmap_write_access(mddev->bitmap_file);
2820                 if (err) {
2821                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2822                                mdname(mddev));
2823                         fput(mddev->bitmap_file);
2824                         mddev->bitmap_file = NULL;
2825                         return err;
2826                 }
2827                 mddev->bitmap_offset = 0; /* file overrides offset */
2828         } else if (mddev->bitmap == NULL)
2829                 return -ENOENT; /* cannot remove what isn't there */
2830         err = 0;
2831         if (mddev->pers) {
2832                 mddev->pers->quiesce(mddev, 1);
2833                 if (fd >= 0)
2834                         err = bitmap_create(mddev);
2835                 if (fd < 0 || err)
2836                         bitmap_destroy(mddev);
2837                 mddev->pers->quiesce(mddev, 0);
2838         } else if (fd < 0) {
2839                 if (mddev->bitmap_file)
2840                         fput(mddev->bitmap_file);
2841                 mddev->bitmap_file = NULL;
2842         }
2843
2844         return err;
2845 }
2846
2847 /*
2848  * set_array_info is used two different ways
2849  * The original usage is when creating a new array.
2850  * In this usage, raid_disks is > 0 and it together with
2851  *  level, size, not_persistent,layout,chunksize determine the
2852  *  shape of the array.
2853  *  This will always create an array with a type-0.90.0 superblock.
2854  * The newer usage is when assembling an array.
2855  *  In this case raid_disks will be 0, and the major_version field is
2856  *  use to determine which style super-blocks are to be found on the devices.
2857  *  The minor and patch _version numbers are also kept incase the
2858  *  super_block handler wishes to interpret them.
2859  */
2860 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2861 {
2862
2863         if (info->raid_disks == 0) {
2864                 /* just setting version number for superblock loading */
2865                 if (info->major_version < 0 ||
2866                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2867                     super_types[info->major_version].name == NULL) {
2868                         /* maybe try to auto-load a module? */
2869                         printk(KERN_INFO
2870                                 "md: superblock version %d not known\n",
2871                                 info->major_version);
2872                         return -EINVAL;
2873                 }
2874                 mddev->major_version = info->major_version;
2875                 mddev->minor_version = info->minor_version;
2876                 mddev->patch_version = info->patch_version;
2877                 return 0;
2878         }
2879         mddev->major_version = MD_MAJOR_VERSION;
2880         mddev->minor_version = MD_MINOR_VERSION;
2881         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2882         mddev->ctime         = get_seconds();
2883
2884         mddev->level         = info->level;
2885         mddev->size          = info->size;
2886         mddev->raid_disks    = info->raid_disks;
2887         /* don't set md_minor, it is determined by which /dev/md* was
2888          * openned
2889          */
2890         if (info->state & (1<<MD_SB_CLEAN))
2891                 mddev->recovery_cp = MaxSector;
2892         else
2893                 mddev->recovery_cp = 0;
2894         mddev->persistent    = ! info->not_persistent;
2895
2896         mddev->layout        = info->layout;
2897         mddev->chunk_size    = info->chunk_size;
2898
2899         mddev->max_disks     = MD_SB_DISKS;
2900
2901         mddev->sb_dirty      = 1;
2902
2903         /*
2904          * Generate a 128 bit UUID
2905          */
2906         get_random_bytes(mddev->uuid, 16);
2907
2908         return 0;
2909 }
2910
2911 /*
2912  * update_array_info is used to change the configuration of an
2913  * on-line array.
2914  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2915  * fields in the info are checked against the array.
2916  * Any differences that cannot be handled will cause an error.
2917  * Normally, only one change can be managed at a time.
2918  */
2919 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2920 {
2921         int rv = 0;
2922         int cnt = 0;
2923         int state = 0;
2924
2925         /* calculate expected state,ignoring low bits */
2926         if (mddev->bitmap && mddev->bitmap_offset)
2927                 state |= (1 << MD_SB_BITMAP_PRESENT);
2928
2929         if (mddev->major_version != info->major_version ||
2930             mddev->minor_version != info->minor_version ||
2931 /*          mddev->patch_version != info->patch_version || */
2932             mddev->ctime         != info->ctime         ||
2933             mddev->level         != info->level         ||
2934 /*          mddev->layout        != info->layout        || */
2935             !mddev->persistent   != info->not_persistent||
2936             mddev->chunk_size    != info->chunk_size    ||
2937             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2938             ((state^info->state) & 0xfffffe00)
2939                 )
2940                 return -EINVAL;
2941         /* Check there is only one change */
2942         if (mddev->size != info->size) cnt++;
2943         if (mddev->raid_disks != info->raid_disks) cnt++;
2944         if (mddev->layout != info->layout) cnt++;
2945         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2946         if (cnt == 0) return 0;
2947         if (cnt > 1) return -EINVAL;
2948
2949         if (mddev->layout != info->layout) {
2950                 /* Change layout
2951                  * we don't need to do anything at the md level, the
2952                  * personality will take care of it all.
2953                  */
2954                 if (mddev->pers->reconfig == NULL)
2955                         return -EINVAL;
2956                 else
2957                         return mddev->pers->reconfig(mddev, info->layout, -1);
2958         }
2959         if (mddev->size != info->size) {
2960                 mdk_rdev_t * rdev;
2961                 struct list_head *tmp;
2962                 if (mddev->pers->resize == NULL)
2963                         return -EINVAL;
2964                 /* The "size" is the amount of each device that is used.
2965                  * This can only make sense for arrays with redundancy.
2966                  * linear and raid0 always use whatever space is available
2967                  * We can only consider changing the size if no resync
2968                  * or reconstruction is happening, and if the new size
2969                  * is acceptable. It must fit before the sb_offset or,
2970                  * if that is <data_offset, it must fit before the
2971                  * size of each device.
2972                  * If size is zero, we find the largest size that fits.
2973                  */
2974                 if (mddev->sync_thread)
2975                         return -EBUSY;
2976                 ITERATE_RDEV(mddev,rdev,tmp) {
2977                         sector_t avail;
2978                         int fit = (info->size == 0);
2979                         if (rdev->sb_offset > rdev->data_offset)
2980                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2981                         else
2982                                 avail = get_capacity(rdev->bdev->bd_disk)
2983                                         - rdev->data_offset;
2984                         if (fit && (info->size == 0 || info->size > avail/2))
2985                                 info->size = avail/2;
2986                         if (avail < ((sector_t)info->size << 1))
2987                                 return -ENOSPC;
2988                 }
2989                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2990                 if (!rv) {
2991                         struct block_device *bdev;
2992
2993                         bdev = bdget_disk(mddev->gendisk, 0);
2994                         if (bdev) {
2995                                 down(&bdev->bd_inode->i_sem);
2996                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2997                                 up(&bdev->bd_inode->i_sem);
2998                                 bdput(bdev);
2999                         }
3000                 }
3001         }
3002         if (mddev->raid_disks    != info->raid_disks) {
3003                 /* change the number of raid disks */
3004                 if (mddev->pers->reshape == NULL)
3005                         return -EINVAL;
3006                 if (info->raid_disks <= 0 ||
3007                     info->raid_disks >= mddev->max_disks)
3008                         return -EINVAL;
3009                 if (mddev->sync_thread)
3010                         return -EBUSY;
3011                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3012                 if (!rv) {
3013                         struct block_device *bdev;
3014
3015                         bdev = bdget_disk(mddev->gendisk, 0);
3016                         if (bdev) {
3017                                 down(&bdev->bd_inode->i_sem);
3018                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3019                                 up(&bdev->bd_inode->i_sem);
3020                                 bdput(bdev);
3021                         }
3022                 }
3023         }
3024         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3025                 if (mddev->pers->quiesce == NULL)
3026                         return -EINVAL;
3027                 if (mddev->recovery || mddev->sync_thread)
3028                         return -EBUSY;
3029                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3030                         /* add the bitmap */
3031                         if (mddev->bitmap)
3032                                 return -EEXIST;
3033                         if (mddev->default_bitmap_offset == 0)
3034                                 return -EINVAL;
3035                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3036                         mddev->pers->quiesce(mddev, 1);
3037                         rv = bitmap_create(mddev);
3038                         if (rv)
3039                                 bitmap_destroy(mddev);
3040                         mddev->pers->quiesce(mddev, 0);
3041                 } else {
3042                         /* remove the bitmap */
3043                         if (!mddev->bitmap)
3044                                 return -ENOENT;
3045                         if (mddev->bitmap->file)
3046                                 return -EINVAL;
3047                         mddev->pers->quiesce(mddev, 1);
3048                         bitmap_destroy(mddev);
3049                         mddev->pers->quiesce(mddev, 0);
3050                         mddev->bitmap_offset = 0;
3051                 }
3052         }
3053         md_update_sb(mddev);
3054         return rv;
3055 }
3056
3057 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3058 {
3059         mdk_rdev_t *rdev;
3060
3061         if (mddev->pers == NULL)
3062                 return -ENODEV;
3063
3064         rdev = find_rdev(mddev, dev);
3065         if (!rdev)
3066                 return -ENODEV;
3067
3068         md_error(mddev, rdev);
3069         return 0;
3070 }
3071
3072 static int md_ioctl(struct inode *inode, struct file *file,
3073                         unsigned int cmd, unsigned long arg)
3074 {
3075         int err = 0;
3076         void __user *argp = (void __user *)arg;
3077         struct hd_geometry __user *loc = argp;
3078         mddev_t *mddev = NULL;
3079
3080         if (!capable(CAP_SYS_ADMIN))
3081                 return -EACCES;
3082
3083         /*
3084          * Commands dealing with the RAID driver but not any
3085          * particular array:
3086          */
3087         switch (cmd)
3088         {
3089                 case RAID_VERSION:
3090                         err = get_version(argp);
3091                         goto done;
3092
3093                 case PRINT_RAID_DEBUG:
3094                         err = 0;
3095                         md_print_devices();
3096                         goto done;
3097
3098 #ifndef MODULE
3099                 case RAID_AUTORUN:
3100                         err = 0;
3101                         autostart_arrays(arg);
3102                         goto done;
3103 #endif
3104                 default:;
3105         }
3106
3107         /*
3108          * Commands creating/starting a new array:
3109          */
3110
3111         mddev = inode->i_bdev->bd_disk->private_data;
3112
3113         if (!mddev) {
3114                 BUG();
3115                 goto abort;
3116         }
3117
3118
3119         if (cmd == START_ARRAY) {
3120                 /* START_ARRAY doesn't need to lock the array as autostart_array
3121                  * does the locking, and it could even be a different array
3122                  */
3123                 static int cnt = 3;
3124                 if (cnt > 0 ) {
3125                         printk(KERN_WARNING
3126                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3127                                "This will not be supported beyond 2.6\n",
3128                                current->comm, current->pid);
3129                         cnt--;
3130                 }
3131                 err = autostart_array(new_decode_dev(arg));
3132                 if (err) {
3133                         printk(KERN_WARNING "md: autostart failed!\n");
3134                         goto abort;
3135                 }
3136                 goto done;
3137         }
3138
3139         err = mddev_lock(mddev);
3140         if (err) {
3141                 printk(KERN_INFO
3142                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3143                         err, cmd);
3144                 goto abort;
3145         }
3146
3147         switch (cmd)
3148         {
3149                 case SET_ARRAY_INFO:
3150                         {
3151                                 mdu_array_info_t info;
3152                                 if (!arg)
3153                                         memset(&info, 0, sizeof(info));
3154                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3155                                         err = -EFAULT;
3156                                         goto abort_unlock;
3157                                 }
3158                                 if (mddev->pers) {
3159                                         err = update_array_info(mddev, &info);
3160                                         if (err) {
3161                                                 printk(KERN_WARNING "md: couldn't update"
3162                                                        " array info. %d\n", err);
3163                                                 goto abort_unlock;
3164                                         }
3165                                         goto done_unlock;
3166                                 }
3167                                 if (!list_empty(&mddev->disks)) {
3168                                         printk(KERN_WARNING
3169                                                "md: array %s already has disks!\n",
3170                                                mdname(mddev));
3171                                         err = -EBUSY;
3172                                         goto abort_unlock;
3173                                 }
3174                                 if (mddev->raid_disks) {
3175                                         printk(KERN_WARNING
3176                                                "md: array %s already initialised!\n",
3177                                                mdname(mddev));
3178                                         err = -EBUSY;
3179                                         goto abort_unlock;
3180                                 }
3181                                 err = set_array_info(mddev, &info);
3182                                 if (err) {
3183                                         printk(KERN_WARNING "md: couldn't set"
3184                                                " array info. %d\n", err);
3185                                         goto abort_unlock;
3186                                 }
3187                         }
3188                         goto done_unlock;
3189
3190                 default:;
3191         }
3192
3193         /*
3194          * Commands querying/configuring an existing array:
3195          */
3196         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3197          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3198         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3199                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3200                 err = -ENODEV;
3201                 goto abort_unlock;
3202         }
3203
3204         /*
3205          * Commands even a read-only array can execute:
3206          */
3207         switch (cmd)
3208         {
3209                 case GET_ARRAY_INFO:
3210                         err = get_array_info(mddev, argp);
3211                         goto done_unlock;
3212
3213                 case GET_BITMAP_FILE:
3214                         err = get_bitmap_file(mddev, argp);
3215                         goto done_unlock;
3216
3217                 case GET_DISK_INFO:
3218                         err = get_disk_info(mddev, argp);
3219                         goto done_unlock;
3220
3221                 case RESTART_ARRAY_RW:
3222                         err = restart_array(mddev);
3223                         goto done_unlock;
3224
3225                 case STOP_ARRAY:
3226                         err = do_md_stop (mddev, 0);
3227                         goto done_unlock;
3228
3229                 case STOP_ARRAY_RO:
3230                         err = do_md_stop (mddev, 1);
3231                         goto done_unlock;
3232
3233         /*
3234          * We have a problem here : there is no easy way to give a CHS
3235          * virtual geometry. We currently pretend that we have a 2 heads
3236          * 4 sectors (with a BIG number of cylinders...). This drives
3237          * dosfs just mad... ;-)
3238          */
3239                 case HDIO_GETGEO:
3240                         if (!loc) {
3241                                 err = -EINVAL;
3242                                 goto abort_unlock;
3243                         }
3244                         err = put_user (2, (char __user *) &loc->heads);
3245                         if (err)
3246                                 goto abort_unlock;
3247                         err = put_user (4, (char __user *) &loc->sectors);
3248                         if (err)
3249                                 goto abort_unlock;
3250                         err = put_user(get_capacity(mddev->gendisk)/8,
3251                                         (short __user *) &loc->cylinders);
3252                         if (err)
3253                                 goto abort_unlock;
3254                         err = put_user (get_start_sect(inode->i_bdev),
3255                                                 (long __user *) &loc->start);
3256                         goto done_unlock;
3257         }
3258
3259         /*
3260          * The remaining ioctls are changing the state of the
3261          * superblock, so we do not allow read-only arrays
3262          * here:
3263          */
3264         if (mddev->ro) {
3265                 err = -EROFS;
3266                 goto abort_unlock;
3267         }
3268
3269         switch (cmd)
3270         {
3271                 case ADD_NEW_DISK:
3272                 {
3273                         mdu_disk_info_t info;
3274                         if (copy_from_user(&info, argp, sizeof(info)))
3275                                 err = -EFAULT;
3276                         else
3277                                 err = add_new_disk(mddev, &info);
3278                         goto done_unlock;
3279                 }
3280
3281                 case HOT_REMOVE_DISK:
3282                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3283                         goto done_unlock;
3284
3285                 case HOT_ADD_DISK:
3286                         err = hot_add_disk(mddev, new_decode_dev(arg));
3287                         goto done_unlock;
3288
3289                 case SET_DISK_FAULTY:
3290                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3291                         goto done_unlock;
3292
3293                 case RUN_ARRAY:
3294                         err = do_md_run (mddev);
3295                         goto done_unlock;
3296
3297                 case SET_BITMAP_FILE:
3298                         err = set_bitmap_file(mddev, (int)arg);
3299                         goto done_unlock;
3300
3301                 default:
3302                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3303                                 printk(KERN_WARNING "md: %s(pid %d) used"
3304                                         " obsolete MD ioctl, upgrade your"
3305                                         " software to use new ictls.\n",
3306                                         current->comm, current->pid);
3307                         err = -EINVAL;
3308                         goto abort_unlock;
3309         }
3310
3311 done_unlock:
3312 abort_unlock:
3313         mddev_unlock(mddev);
3314
3315         return err;
3316 done:
3317         if (err)
3318                 MD_BUG();
3319 abort:
3320         return err;
3321 }
3322
3323 static int md_open(struct inode *inode, struct file *file)
3324 {
3325         /*
3326          * Succeed if we can lock the mddev, which confirms that
3327          * it isn't being stopped right now.
3328          */
3329         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3330         int err;
3331
3332         if ((err = mddev_lock(mddev)))
3333                 goto out;
3334
3335         err = 0;
3336         mddev_get(mddev);
3337         mddev_unlock(mddev);
3338
3339         check_disk_change(inode->i_bdev);
3340  out:
3341         return err;
3342 }
3343
3344 static int md_release(struct inode *inode, struct file * file)
3345 {
3346         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3347
3348         if (!mddev)
3349                 BUG();
3350         mddev_put(mddev);
3351
3352         return 0;
3353 }
3354
3355 static int md_media_changed(struct gendisk *disk)
3356 {
3357         mddev_t *mddev = disk->private_data;
3358
3359         return mddev->changed;
3360 }
3361
3362 static int md_revalidate(struct gendisk *disk)
3363 {
3364         mddev_t *mddev = disk->private_data;
3365
3366         mddev->changed = 0;
3367         return 0;
3368 }
3369 static struct block_device_operations md_fops =
3370 {
3371         .owner          = THIS_MODULE,
3372         .open           = md_open,
3373         .release        = md_release,
3374         .ioctl          = md_ioctl,
3375         .media_changed  = md_media_changed,
3376         .revalidate_disk= md_revalidate,
3377 };
3378
3379 static int md_thread(void * arg)
3380 {
3381         mdk_thread_t *thread = arg;
3382
3383         /*
3384          * md_thread is a 'system-thread', it's priority should be very
3385          * high. We avoid resource deadlocks individually in each
3386          * raid personality. (RAID5 does preallocation) We also use RR and
3387          * the very same RT priority as kswapd, thus we will never get
3388          * into a priority inversion deadlock.
3389          *
3390          * we definitely have to have equal or higher priority than
3391          * bdflush, otherwise bdflush will deadlock if there are too
3392          * many dirty RAID5 blocks.
3393          */
3394
3395         allow_signal(SIGKILL);
3396         complete(thread->event);
3397         while (!kthread_should_stop()) {
3398                 void (*run)(mddev_t *);
3399
3400                 wait_event_interruptible_timeout(thread->wqueue,
3401                                                  test_bit(THREAD_WAKEUP, &thread->flags)
3402                                                  || kthread_should_stop(),
3403                                                  thread->timeout);
3404                 try_to_freeze();
3405
3406                 clear_bit(THREAD_WAKEUP, &thread->flags);
3407
3408                 run = thread->run;
3409                 if (run)
3410                         run(thread->mddev);
3411         }
3412
3413         return 0;
3414 }
3415
3416 void md_wakeup_thread(mdk_thread_t *thread)
3417 {
3418         if (thread) {
3419                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3420                 set_bit(THREAD_WAKEUP, &thread->flags);
3421                 wake_up(&thread->wqueue);
3422         }
3423 }
3424
3425 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3426                                  const char *name)
3427 {
3428         mdk_thread_t *thread;
3429         struct completion event;
3430
3431         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3432         if (!thread)
3433                 return NULL;
3434
3435         memset(thread, 0, sizeof(mdk_thread_t));
3436         init_waitqueue_head(&thread->wqueue);
3437
3438         init_completion(&event);
3439         thread->event = &event;
3440         thread->run = run;
3441         thread->mddev = mddev;
3442         thread->name = name;
3443         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3444         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3445         if (IS_ERR(thread->tsk)) {
3446                 kfree(thread);
3447                 return NULL;
3448         }
3449         wait_for_completion(&event);
3450         return thread;
3451 }
3452
3453 void md_unregister_thread(mdk_thread_t *thread)
3454 {
3455         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3456
3457         kthread_stop(thread->tsk);
3458         kfree(thread);
3459 }
3460
3461 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3462 {
3463         if (!mddev) {
3464                 MD_BUG();
3465                 return;
3466         }
3467
3468         if (!rdev || test_bit(Faulty, &rdev->flags))
3469                 return;
3470 /*
3471         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3472                 mdname(mddev),
3473                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3474                 __builtin_return_address(0),__builtin_return_address(1),
3475                 __builtin_return_address(2),__builtin_return_address(3));
3476 */
3477         if (!mddev->pers->error_handler)
3478                 return;
3479         mddev->pers->error_handler(mddev,rdev);
3480         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3481         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3482         md_wakeup_thread(mddev->thread);
3483 }
3484
3485 /* seq_file implementation /proc/mdstat */
3486
3487 static void status_unused(struct seq_file *seq)
3488 {
3489         int i = 0;
3490         mdk_rdev_t *rdev;
3491         struct list_head *tmp;
3492
3493         seq_printf(seq, "unused devices: ");
3494
3495         ITERATE_RDEV_PENDING(rdev,tmp) {
3496                 char b[BDEVNAME_SIZE];
3497                 i++;
3498                 seq_printf(seq, "%s ",
3499                               bdevname(rdev->bdev,b));
3500         }
3501         if (!i)
3502                 seq_printf(seq, "<none>");
3503
3504         seq_printf(seq, "\n");
3505 }
3506
3507
3508 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3509 {
3510         unsigned long max_blocks, resync, res, dt, db, rt;
3511
3512         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3513
3514         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3515                 max_blocks = mddev->resync_max_sectors >> 1;
3516         else
3517                 max_blocks = mddev->size;
3518
3519         /*
3520          * Should not happen.
3521          */
3522         if (!max_blocks) {
3523                 MD_BUG();
3524                 return;
3525         }
3526         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3527         {
3528                 int i, x = res/50, y = 20-x;
3529                 seq_printf(seq, "[");
3530                 for (i = 0; i < x; i++)
3531                         seq_printf(seq, "=");
3532                 seq_printf(seq, ">");
3533                 for (i = 0; i < y; i++)
3534                         seq_printf(seq, ".");
3535                 seq_printf(seq, "] ");
3536         }
3537         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3538                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3539                        "resync" : "recovery"),
3540                       res/10, res % 10, resync, max_blocks);
3541
3542         /*
3543          * We do not want to overflow, so the order of operands and
3544          * the * 100 / 100 trick are important. We do a +1 to be
3545          * safe against division by zero. We only estimate anyway.
3546          *
3547          * dt: time from mark until now
3548          * db: blocks written from mark until now
3549          * rt: remaining time
3550          */
3551         dt = ((jiffies - mddev->resync_mark) / HZ);
3552         if (!dt) dt++;
3553         db = resync - (mddev->resync_mark_cnt/2);
3554         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3555
3556         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3557
3558         seq_printf(seq, " speed=%ldK/sec", db/dt);
3559 }
3560
3561 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3562 {
3563         struct list_head *tmp;
3564         loff_t l = *pos;
3565         mddev_t *mddev;
3566
3567         if (l >= 0x10000)
3568                 return NULL;
3569         if (!l--)
3570                 /* header */
3571                 return (void*)1;
3572
3573         spin_lock(&all_mddevs_lock);
3574         list_for_each(tmp,&all_mddevs)
3575                 if (!l--) {
3576                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3577                         mddev_get(mddev);
3578                         spin_unlock(&all_mddevs_lock);
3579                         return mddev;
3580                 }
3581         spin_unlock(&all_mddevs_lock);
3582         if (!l--)
3583                 return (void*)2;/* tail */
3584         return NULL;
3585 }
3586
3587 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3588 {
3589         struct list_head *tmp;
3590         mddev_t *next_mddev, *mddev = v;
3591
3592         ++*pos;
3593         if (v == (void*)2)
3594                 return NULL;
3595
3596         spin_lock(&all_mddevs_lock);
3597         if (v == (void*)1)
3598                 tmp = all_mddevs.next;
3599         else
3600                 tmp = mddev->all_mddevs.next;
3601         if (tmp != &all_mddevs)
3602                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3603         else {
3604                 next_mddev = (void*)2;
3605                 *pos = 0x10000;
3606         }
3607         spin_unlock(&all_mddevs_lock);
3608
3609         if (v != (void*)1)
3610                 mddev_put(mddev);
3611         return next_mddev;
3612
3613 }
3614
3615 static void md_seq_stop(struct seq_file *seq, void *v)
3616 {
3617         mddev_t *mddev = v;
3618
3619         if (mddev && v != (void*)1 && v != (void*)2)
3620                 mddev_put(mddev);
3621 }
3622
3623 static int md_seq_show(struct seq_file *seq, void *v)
3624 {
3625         mddev_t *mddev = v;
3626         sector_t size;
3627         struct list_head *tmp2;
3628         mdk_rdev_t *rdev;
3629         int i;
3630         struct bitmap *bitmap;
3631
3632         if (v == (void*)1) {
3633                 seq_printf(seq, "Personalities : ");
3634                 spin_lock(&pers_lock);
3635                 for (i = 0; i < MAX_PERSONALITY; i++)
3636                         if (pers[i])
3637                                 seq_printf(seq, "[%s] ", pers[i]->name);
3638
3639                 spin_unlock(&pers_lock);
3640                 seq_printf(seq, "\n");
3641                 return 0;
3642         }
3643         if (v == (void*)2) {
3644                 status_unused(seq);
3645                 return 0;
3646         }
3647
3648         if (mddev_lock(mddev)!=0)
3649                 return -EINTR;
3650         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3651                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3652                                                 mddev->pers ? "" : "in");
3653                 if (mddev->pers) {
3654                         if (mddev->ro)
3655                                 seq_printf(seq, " (read-only)");
3656                         seq_printf(seq, " %s", mddev->pers->name);
3657                 }
3658
3659                 size = 0;
3660                 ITERATE_RDEV(mddev,rdev,tmp2) {
3661                         char b[BDEVNAME_SIZE];
3662                         seq_printf(seq, " %s[%d]",
3663                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3664                         if (test_bit(WriteMostly, &rdev->flags))
3665                                 seq_printf(seq, "(W)");
3666                         if (test_bit(Faulty, &rdev->flags)) {
3667                                 seq_printf(seq, "(F)");
3668                                 continue;
3669                         } else if (rdev->raid_disk < 0)
3670                                 seq_printf(seq, "(S)"); /* spare */
3671                         size += rdev->size;
3672                 }
3673
3674                 if (!list_empty(&mddev->disks)) {
3675                         if (mddev->pers)
3676                                 seq_printf(seq, "\n      %llu blocks",
3677                                         (unsigned long long)mddev->array_size);
3678                         else
3679                                 seq_printf(seq, "\n      %llu blocks",
3680                                         (unsigned long long)size);
3681                 }
3682                 if (mddev->persistent) {
3683                         if (mddev->major_version != 0 ||
3684                             mddev->minor_version != 90) {
3685                                 seq_printf(seq," super %d.%d",
3686                                            mddev->major_version,
3687                                            mddev->minor_version);
3688                         }
3689                 } else
3690                         seq_printf(seq, " super non-persistent");
3691
3692                 if (mddev->pers) {
3693                         mddev->pers->status (seq, mddev);
3694                         seq_printf(seq, "\n      ");
3695                         if (mddev->curr_resync > 2) {
3696                                 status_resync (seq, mddev);
3697                                 seq_printf(seq, "\n      ");
3698                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3699                                 seq_printf(seq, "       resync=DELAYED\n      ");
3700                 } else
3701                         seq_printf(seq, "\n       ");
3702
3703                 if ((bitmap = mddev->bitmap)) {
3704                         unsigned long chunk_kb;
3705                         unsigned long flags;
3706                         spin_lock_irqsave(&bitmap->lock, flags);
3707                         chunk_kb = bitmap->chunksize >> 10;
3708                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3709                                 "%lu%s chunk",
3710                                 bitmap->pages - bitmap->missing_pages,
3711                                 bitmap->pages,
3712                                 (bitmap->pages - bitmap->missing_pages)
3713                                         << (PAGE_SHIFT - 10),
3714                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3715                                 chunk_kb ? "KB" : "B");
3716                         if (bitmap->file) {
3717                                 seq_printf(seq, ", file: ");
3718                                 seq_path(seq, bitmap->file->f_vfsmnt,
3719                                          bitmap->file->f_dentry," \t\n");
3720                         }
3721
3722                         seq_printf(seq, "\n");
3723                         spin_unlock_irqrestore(&bitmap->lock, flags);
3724                 }
3725
3726                 seq_printf(seq, "\n");
3727         }
3728         mddev_unlock(mddev);
3729
3730         return 0;
3731 }
3732
3733 static struct seq_operations md_seq_ops = {
3734         .start  = md_seq_start,
3735         .next   = md_seq_next,
3736         .stop   = md_seq_stop,
3737         .show   = md_seq_show,
3738 };
3739
3740 static int md_seq_open(struct inode *inode, struct file *file)
3741 {
3742         int error;
3743
3744         error = seq_open(file, &md_seq_ops);
3745         return error;
3746 }
3747
3748 static struct file_operations md_seq_fops = {
3749         .open           = md_seq_open,
3750         .read           = seq_read,
3751         .llseek         = seq_lseek,
3752         .release        = seq_release,
3753 };
3754
3755 int register_md_personality(int pnum, mdk_personality_t *p)
3756 {
3757         if (pnum >= MAX_PERSONALITY) {
3758                 printk(KERN_ERR
3759                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3760                        p->name, pnum, MAX_PERSONALITY-1);
3761                 return -EINVAL;
3762         }
3763
3764         spin_lock(&pers_lock);
3765         if (pers[pnum]) {
3766                 spin_unlock(&pers_lock);
3767                 return -EBUSY;
3768         }
3769
3770         pers[pnum] = p;
3771         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3772         spin_unlock(&pers_lock);
3773         return 0;
3774 }
3775
3776 int unregister_md_personality(int pnum)
3777 {
3778         if (pnum >= MAX_PERSONALITY)
3779                 return -EINVAL;
3780
3781         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3782         spin_lock(&pers_lock);
3783         pers[pnum] = NULL;
3784         spin_unlock(&pers_lock);
3785         return 0;
3786 }
3787
3788 static int is_mddev_idle(mddev_t *mddev)
3789 {
3790         mdk_rdev_t * rdev;
3791         struct list_head *tmp;
3792         int idle;
3793         unsigned long curr_events;
3794
3795         idle = 1;
3796         ITERATE_RDEV(mddev,rdev,tmp) {
3797                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3798                 curr_events = disk_stat_read(disk, sectors[0]) +
3799                                 disk_stat_read(disk, sectors[1]) -
3800                                 atomic_read(&disk->sync_io);
3801                 /* Allow some slack between valud of curr_events and last_events,
3802                  * as there are some uninteresting races.
3803                  * Note: the following is an unsigned comparison.
3804                  */
3805                 if ((curr_events - rdev->last_events + 32) > 64) {
3806                         rdev->last_events = curr_events;
3807                         idle = 0;
3808                 }
3809         }
3810         return idle;
3811 }
3812
3813 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3814 {
3815         /* another "blocks" (512byte) blocks have been synced */
3816         atomic_sub(blocks, &mddev->recovery_active);
3817         wake_up(&mddev->recovery_wait);
3818         if (!ok) {
3819                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3820                 md_wakeup_thread(mddev->thread);
3821                 // stop recovery, signal do_sync ....
3822         }
3823 }
3824
3825
3826 /* md_write_start(mddev, bi)
3827  * If we need to update some array metadata (e.g. 'active' flag
3828  * in superblock) before writing, schedule a superblock update
3829  * and wait for it to complete.
3830  */
3831 void md_write_start(mddev_t *mddev, struct bio *bi)
3832 {
3833         if (bio_data_dir(bi) != WRITE)
3834                 return;
3835
3836         atomic_inc(&mddev->writes_pending);
3837         if (mddev->in_sync) {
3838                 spin_lock_irq(&mddev->write_lock);
3839                 if (mddev->in_sync) {
3840                         mddev->in_sync = 0;
3841                         mddev->sb_dirty = 1;
3842                         md_wakeup_thread(mddev->thread);
3843                 }
3844                 spin_unlock_irq(&mddev->write_lock);
3845         }
3846         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3847 }
3848
3849 void md_write_end(mddev_t *mddev)
3850 {
3851         if (atomic_dec_and_test(&mddev->writes_pending)) {
3852                 if (mddev->safemode == 2)
3853                         md_wakeup_thread(mddev->thread);
3854                 else
3855                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3856         }
3857 }
3858
3859 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3860
3861 #define SYNC_MARKS      10
3862 #define SYNC_MARK_STEP  (3*HZ)
3863 static void md_do_sync(mddev_t *mddev)
3864 {
3865         mddev_t *mddev2;
3866         unsigned int currspeed = 0,
3867                  window;
3868         sector_t max_sectors,j, io_sectors;
3869         unsigned long mark[SYNC_MARKS];
3870         sector_t mark_cnt[SYNC_MARKS];
3871         int last_mark,m;
3872         struct list_head *tmp;
3873         sector_t last_check;
3874         int skipped = 0;
3875
3876         /* just incase thread restarts... */
3877         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3878                 return;
3879
3880         /* we overload curr_resync somewhat here.
3881          * 0 == not engaged in resync at all
3882          * 2 == checking that there is no conflict with another sync
3883          * 1 == like 2, but have yielded to allow conflicting resync to
3884          *              commense
3885          * other == active in resync - this many blocks
3886          *
3887          * Before starting a resync we must have set curr_resync to
3888          * 2, and then checked that every "conflicting" array has curr_resync
3889          * less than ours.  When we find one that is the same or higher
3890          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3891          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3892          * This will mean we have to start checking from the beginning again.
3893          *
3894          */
3895
3896         do {
3897                 mddev->curr_resync = 2;
3898
3899         try_again:
3900                 if (signal_pending(current) ||
3901                     kthread_should_stop()) {
3902                         flush_signals(current);
3903                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3904                         goto skip;
3905                 }
3906                 ITERATE_MDDEV(mddev2,tmp) {
3907                         if (mddev2 == mddev)
3908                                 continue;
3909                         if (mddev2->curr_resync &&
3910                             match_mddev_units(mddev,mddev2)) {
3911                                 DEFINE_WAIT(wq);
3912                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3913                                         /* arbitrarily yield */
3914                                         mddev->curr_resync = 1;
3915                                         wake_up(&resync_wait);
3916                                 }
3917                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3918                                         /* no need to wait here, we can wait the next
3919                                          * time 'round when curr_resync == 2
3920                                          */
3921                                         continue;
3922                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3923                                 if (!signal_pending(current) &&
3924                                     !kthread_should_stop() &&
3925                                     mddev2->curr_resync >= mddev->curr_resync) {
3926                                         printk(KERN_INFO "md: delaying resync of %s"
3927                                                " until %s has finished resync (they"
3928                                                " share one or more physical units)\n",
3929                                                mdname(mddev), mdname(mddev2));
3930                                         mddev_put(mddev2);
3931                                         schedule();
3932                                         finish_wait(&resync_wait, &wq);
3933                                         goto try_again;
3934                                 }
3935                                 finish_wait(&resync_wait, &wq);
3936                         }
3937                 }
3938         } while (mddev->curr_resync < 2);
3939
3940         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3941                 /* resync follows the size requested by the personality,
3942                  * which defaults to physical size, but can be virtual size
3943                  */
3944                 max_sectors = mddev->resync_max_sectors;
3945                 mddev->resync_mismatches = 0;
3946         } else
3947                 /* recovery follows the physical size of devices */
3948                 max_sectors = mddev->size << 1;
3949
3950         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3951         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3952                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3953         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3954                "(but not more than %d KB/sec) for reconstruction.\n",
3955                sysctl_speed_limit_max);
3956
3957         is_mddev_idle(mddev); /* this also initializes IO event counters */
3958         /* we don't use the checkpoint if there's a bitmap */
3959         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
3960             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3961                 j = mddev->recovery_cp;
3962         else
3963                 j = 0;
3964         io_sectors = 0;
3965         for (m = 0; m < SYNC_MARKS; m++) {
3966                 mark[m] = jiffies;
3967                 mark_cnt[m] = io_sectors;
3968         }
3969         last_mark = 0;
3970         mddev->resync_mark = mark[last_mark];
3971         mddev->resync_mark_cnt = mark_cnt[last_mark];
3972
3973         /*
3974          * Tune reconstruction:
3975          */
3976         window = 32*(PAGE_SIZE/512);
3977         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3978                 window/2,(unsigned long long) max_sectors/2);
3979
3980         atomic_set(&mddev->recovery_active, 0);
3981         init_waitqueue_head(&mddev->recovery_wait);
3982         last_check = 0;
3983
3984         if (j>2) {
3985                 printk(KERN_INFO
3986                         "md: resuming recovery of %s from checkpoint.\n",
3987                         mdname(mddev));
3988                 mddev->curr_resync = j;
3989         }
3990
3991         while (j < max_sectors) {
3992                 sector_t sectors;
3993
3994                 skipped = 0;
3995                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3996                                             currspeed < sysctl_speed_limit_min);
3997                 if (sectors == 0) {
3998                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3999                         goto out;
4000                 }
4001
4002                 if (!skipped) { /* actual IO requested */
4003                         io_sectors += sectors;
4004                         atomic_add(sectors, &mddev->recovery_active);
4005                 }
4006
4007                 j += sectors;
4008                 if (j>1) mddev->curr_resync = j;
4009
4010
4011                 if (last_check + window > io_sectors || j == max_sectors)
4012                         continue;
4013
4014                 last_check = io_sectors;
4015
4016                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4017                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4018                         break;
4019
4020         repeat:
4021                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4022                         /* step marks */
4023                         int next = (last_mark+1) % SYNC_MARKS;
4024
4025                         mddev->resync_mark = mark[next];
4026                         mddev->resync_mark_cnt = mark_cnt[next];
4027                         mark[next] = jiffies;
4028                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4029                         last_mark = next;
4030                 }
4031
4032
4033                 if (signal_pending(current) || kthread_should_stop()) {
4034                         /*
4035                          * got a signal, exit.
4036                          */
4037                         printk(KERN_INFO
4038                                 "md: md_do_sync() got signal ... exiting\n");
4039                         flush_signals(current);
4040                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4041                         goto out;
4042                 }
4043
4044                 /*
4045                  * this loop exits only if either when we are slower than
4046                  * the 'hard' speed limit, or the system was IO-idle for
4047                  * a jiffy.
4048                  * the system might be non-idle CPU-wise, but we only care
4049                  * about not overloading the IO subsystem. (things like an
4050                  * e2fsck being done on the RAID array should execute fast)
4051                  */
4052                 mddev->queue->unplug_fn(mddev->queue);
4053                 cond_resched();
4054
4055                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4056                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4057
4058                 if (currspeed > sysctl_speed_limit_min) {
4059                         if ((currspeed > sysctl_speed_limit_max) ||
4060                                         !is_mddev_idle(mddev)) {
4061                                 msleep_interruptible(250);
4062                                 goto repeat;
4063                         }
4064                 }
4065         }
4066         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4067         /*
4068          * this also signals 'finished resyncing' to md_stop
4069          */
4070  out:
4071         mddev->queue->unplug_fn(mddev->queue);
4072
4073         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4074
4075         /* tell personality that we are finished */
4076         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4077
4078         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4079             mddev->curr_resync > 2 &&
4080             mddev->curr_resync >= mddev->recovery_cp) {
4081                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4082                         printk(KERN_INFO
4083                                 "md: checkpointing recovery of %s.\n",
4084                                 mdname(mddev));
4085                         mddev->recovery_cp = mddev->curr_resync;
4086                 } else
4087                         mddev->recovery_cp = MaxSector;
4088         }
4089
4090  skip:
4091         mddev->curr_resync = 0;
4092         wake_up(&resync_wait);
4093         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4094         md_wakeup_thread(mddev->thread);
4095 }
4096
4097
4098 /*
4099  * This routine is regularly called by all per-raid-array threads to
4100  * deal with generic issues like resync and super-block update.
4101  * Raid personalities that don't have a thread (linear/raid0) do not
4102  * need this as they never do any recovery or update the superblock.
4103  *
4104  * It does not do any resync itself, but rather "forks" off other threads
4105  * to do that as needed.
4106  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4107  * "->recovery" and create a thread at ->sync_thread.
4108  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4109  * and wakeups up this thread which will reap the thread and finish up.
4110  * This thread also removes any faulty devices (with nr_pending == 0).
4111  *
4112  * The overall approach is:
4113  *  1/ if the superblock needs updating, update it.
4114  *  2/ If a recovery thread is running, don't do anything else.
4115  *  3/ If recovery has finished, clean up, possibly marking spares active.
4116  *  4/ If there are any faulty devices, remove them.
4117  *  5/ If array is degraded, try to add spares devices
4118  *  6/ If array has spares or is not in-sync, start a resync thread.
4119  */
4120 void md_check_recovery(mddev_t *mddev)
4121 {
4122         mdk_rdev_t *rdev;
4123         struct list_head *rtmp;
4124
4125
4126         if (mddev->bitmap)
4127                 bitmap_daemon_work(mddev->bitmap);
4128
4129         if (mddev->ro)
4130                 return;
4131
4132         if (signal_pending(current)) {
4133                 if (mddev->pers->sync_request) {
4134                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4135                                mdname(mddev));
4136                         mddev->safemode = 2;
4137                 }
4138                 flush_signals(current);
4139         }
4140
4141         if ( ! (
4142                 mddev->sb_dirty ||
4143                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4144                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4145                 (mddev->safemode == 1) ||
4146                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4147                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4148                 ))
4149                 return;
4150
4151         if (mddev_trylock(mddev)==0) {
4152                 int spares =0;
4153
4154                 spin_lock_irq(&mddev->write_lock);
4155                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4156                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4157                         mddev->in_sync = 1;
4158                         mddev->sb_dirty = 1;
4159                 }
4160                 if (mddev->safemode == 1)
4161                         mddev->safemode = 0;
4162                 spin_unlock_irq(&mddev->write_lock);
4163
4164                 if (mddev->sb_dirty)
4165                         md_update_sb(mddev);
4166
4167
4168                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4169                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4170                         /* resync/recovery still happening */
4171                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4172                         goto unlock;
4173                 }
4174                 if (mddev->sync_thread) {
4175                         /* resync has finished, collect result */
4176                         md_unregister_thread(mddev->sync_thread);
4177                         mddev->sync_thread = NULL;
4178                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4179                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4180                                 /* success...*/
4181                                 /* activate any spares */
4182                                 mddev->pers->spare_active(mddev);
4183                         }
4184                         md_update_sb(mddev);
4185
4186                         /* if array is no-longer degraded, then any saved_raid_disk
4187                          * information must be scrapped
4188                          */
4189                         if (!mddev->degraded)
4190                                 ITERATE_RDEV(mddev,rdev,rtmp)
4191                                         rdev->saved_raid_disk = -1;
4192
4193                         mddev->recovery = 0;
4194                         /* flag recovery needed just to double check */
4195                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4196                         goto unlock;
4197                 }
4198                 /* Clear some bits that don't mean anything, but
4199                  * might be left set
4200                  */
4201                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4202                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4203                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4204                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4205
4206                 /* no recovery is running.
4207                  * remove any failed drives, then
4208                  * add spares if possible.
4209                  * Spare are also removed and re-added, to allow
4210                  * the personality to fail the re-add.
4211                  */
4212                 ITERATE_RDEV(mddev,rdev,rtmp)
4213                         if (rdev->raid_disk >= 0 &&
4214                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4215                             atomic_read(&rdev->nr_pending)==0) {
4216                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4217                                         char nm[20];
4218                                         sprintf(nm,"rd%d", rdev->raid_disk);
4219                                         sysfs_remove_link(&mddev->kobj, nm);
4220                                         rdev->raid_disk = -1;
4221                                 }
4222                         }
4223
4224                 if (mddev->degraded) {
4225                         ITERATE_RDEV(mddev,rdev,rtmp)
4226                                 if (rdev->raid_disk < 0
4227                                     && !test_bit(Faulty, &rdev->flags)) {
4228                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4229                                                 char nm[20];
4230                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4231                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4232                                                 spares++;
4233                                         } else
4234                                                 break;
4235                                 }
4236                 }
4237
4238                 if (spares) {
4239                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4240                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4241                 } else if (mddev->recovery_cp < MaxSector) {
4242                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4243                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4244                         /* nothing to be done ... */
4245                         goto unlock;
4246
4247                 if (mddev->pers->sync_request) {
4248                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4249                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4250                                 /* We are adding a device or devices to an array
4251                                  * which has the bitmap stored on all devices.
4252                                  * So make sure all bitmap pages get written
4253                                  */
4254                                 bitmap_write_all(mddev->bitmap);
4255                         }
4256                         mddev->sync_thread = md_register_thread(md_do_sync,
4257                                                                 mddev,
4258                                                                 "%s_resync");
4259                         if (!mddev->sync_thread) {
4260                                 printk(KERN_ERR "%s: could not start resync"
4261                                         " thread...\n",
4262                                         mdname(mddev));
4263                                 /* leave the spares where they are, it shouldn't hurt */
4264                                 mddev->recovery = 0;
4265                         } else {
4266                                 md_wakeup_thread(mddev->sync_thread);
4267                         }
4268                 }
4269         unlock:
4270                 mddev_unlock(mddev);
4271         }
4272 }
4273
4274 static int md_notify_reboot(struct notifier_block *this,
4275                             unsigned long code, void *x)
4276 {
4277         struct list_head *tmp;
4278         mddev_t *mddev;
4279
4280         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4281
4282                 printk(KERN_INFO "md: stopping all md devices.\n");
4283
4284                 ITERATE_MDDEV(mddev,tmp)
4285                         if (mddev_trylock(mddev)==0)
4286                                 do_md_stop (mddev, 1);
4287                 /*
4288                  * certain more exotic SCSI devices are known to be
4289                  * volatile wrt too early system reboots. While the
4290                  * right place to handle this issue is the given
4291                  * driver, we do want to have a safe RAID driver ...
4292                  */
4293                 mdelay(1000*1);
4294         }
4295         return NOTIFY_DONE;
4296 }
4297
4298 static struct notifier_block md_notifier = {
4299         .notifier_call  = md_notify_reboot,
4300         .next           = NULL,
4301         .priority       = INT_MAX, /* before any real devices */
4302 };
4303
4304 static void md_geninit(void)
4305 {
4306         struct proc_dir_entry *p;
4307
4308         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4309
4310         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4311         if (p)
4312                 p->proc_fops = &md_seq_fops;
4313 }
4314
4315 static int __init md_init(void)
4316 {
4317         int minor;
4318
4319         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4320                         " MD_SB_DISKS=%d\n",
4321                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4322                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4323         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4324                         BITMAP_MINOR);
4325
4326         if (register_blkdev(MAJOR_NR, "md"))
4327                 return -1;
4328         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4329                 unregister_blkdev(MAJOR_NR, "md");
4330                 return -1;
4331         }
4332         devfs_mk_dir("md");
4333         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4334                                 md_probe, NULL, NULL);
4335         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4336                             md_probe, NULL, NULL);
4337
4338         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4339                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4340                                 S_IFBLK|S_IRUSR|S_IWUSR,
4341                                 "md/%d", minor);
4342
4343         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4344                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4345                               S_IFBLK|S_IRUSR|S_IWUSR,
4346                               "md/mdp%d", minor);
4347
4348
4349         register_reboot_notifier(&md_notifier);
4350         raid_table_header = register_sysctl_table(raid_root_table, 1);
4351
4352         md_geninit();
4353         return (0);
4354 }
4355
4356
4357 #ifndef MODULE
4358
4359 /*
4360  * Searches all registered partitions for autorun RAID arrays
4361  * at boot time.
4362  */
4363 static dev_t detected_devices[128];
4364 static int dev_cnt;
4365
4366 void md_autodetect_dev(dev_t dev)
4367 {
4368         if (dev_cnt >= 0 && dev_cnt < 127)
4369                 detected_devices[dev_cnt++] = dev;
4370 }
4371
4372
4373 static void autostart_arrays(int part)
4374 {
4375         mdk_rdev_t *rdev;
4376         int i;
4377
4378         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4379
4380         for (i = 0; i < dev_cnt; i++) {
4381                 dev_t dev = detected_devices[i];
4382
4383                 rdev = md_import_device(dev,0, 0);
4384                 if (IS_ERR(rdev))
4385                         continue;
4386
4387                 if (test_bit(Faulty, &rdev->flags)) {
4388                         MD_BUG();
4389                         continue;
4390                 }
4391                 list_add(&rdev->same_set, &pending_raid_disks);
4392         }
4393         dev_cnt = 0;
4394
4395         autorun_devices(part);
4396 }
4397
4398 #endif
4399
4400 static __exit void md_exit(void)
4401 {
4402         mddev_t *mddev;
4403         struct list_head *tmp;
4404         int i;
4405         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4406         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4407         for (i=0; i < MAX_MD_DEVS; i++)
4408                 devfs_remove("md/%d", i);
4409         for (i=0; i < MAX_MD_DEVS; i++)
4410                 devfs_remove("md/d%d", i);
4411
4412         devfs_remove("md");
4413
4414         unregister_blkdev(MAJOR_NR,"md");
4415         unregister_blkdev(mdp_major, "mdp");
4416         unregister_reboot_notifier(&md_notifier);
4417         unregister_sysctl_table(raid_table_header);
4418         remove_proc_entry("mdstat", NULL);
4419         ITERATE_MDDEV(mddev,tmp) {
4420                 struct gendisk *disk = mddev->gendisk;
4421                 if (!disk)
4422                         continue;
4423                 export_array(mddev);
4424                 del_gendisk(disk);
4425                 put_disk(disk);
4426                 mddev->gendisk = NULL;
4427                 mddev_put(mddev);
4428         }
4429 }
4430
4431 module_init(md_init)
4432 module_exit(md_exit)
4433
4434 EXPORT_SYMBOL(register_md_personality);
4435 EXPORT_SYMBOL(unregister_md_personality);
4436 EXPORT_SYMBOL(md_error);
4437 EXPORT_SYMBOL(md_done_sync);
4438 EXPORT_SYMBOL(md_write_start);
4439 EXPORT_SYMBOL(md_write_end);
4440 EXPORT_SYMBOL(md_register_thread);
4441 EXPORT_SYMBOL(md_unregister_thread);
4442 EXPORT_SYMBOL(md_wakeup_thread);
4443 EXPORT_SYMBOL(md_print_devices);
4444 EXPORT_SYMBOL(md_check_recovery);
4445 MODULE_LICENSE("GPL");
4446 MODULE_ALIAS("md");
4447 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);