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 static int start_readonly;
 135
 136 /*
 137  * Enables to iterate over all existing md arrays
 138  * all_mddevs_lock protects this list.
 139  */
 140 static LIST_HEAD(all_mddevs);
 141 static DEFINE_SPINLOCK(all_mddevs_lock);
 142
 143
 144 /*
 145  * iterates through all used mddevs in the system.
 146  * We take care to grab the all_mddevs_lock whenever navigating
 147  * the list, and to always hold a refcount when unlocked.
 148  * Any code which breaks out of this loop while own
 149  * a reference to the current mddev and must mddev_put it.
 150  */
 151 #define ITERATE_MDDEV(mddev,tmp)                                        \
 152                                                                         \
 153         for (({ spin_lock(&all_mddevs_lock);                            \
 154                 tmp = all_mddevs.next;                                  \
 155                 mddev = NULL;});                                        \
 156              ({ if (tmp != &all_mddevs)                                 \
 157                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 158                 spin_unlock(&all_mddevs_lock);                          \
 159                 if (mddev) mddev_put(mddev);                            \
 160                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 161                 tmp != &all_mddevs;});                                  \
 162              ({ spin_lock(&all_mddevs_lock);                            \
 163                 tmp = tmp->next;})                                      \
 164                 )
 165
 166
 167 static int md_fail_request (request_queue_t *q, struct bio *bio)
 168 {
 169         bio_io_error(bio, bio->bi_size);
 170         return 0;
 171 }
 172
 173 static inline mddev_t *mddev_get(mddev_t *mddev)
 174 {
 175         atomic_inc(&mddev->active);
 176         return mddev;
 177 }
 178
 179 static void mddev_put(mddev_t *mddev)
 180 {
 181         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 182                 return;
 183         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 184                 list_del(&mddev->all_mddevs);
 185                 blk_put_queue(mddev->queue);
 186                 kobject_unregister(&mddev->kobj);
 187         }
 188         spin_unlock(&all_mddevs_lock);
 189 }
 190
 191 static mddev_t * mddev_find(dev_t unit)
 192 {
 193         mddev_t *mddev, *new = NULL;
 194
 195  retry:
 196         spin_lock(&all_mddevs_lock);
 197         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 198                 if (mddev->unit == unit) {
 199                         mddev_get(mddev);
 200                         spin_unlock(&all_mddevs_lock);
 201                         kfree(new);
 202                         return mddev;
 203                 }
 204
 205         if (new) {
 206                 list_add(&new->all_mddevs, &all_mddevs);
 207                 spin_unlock(&all_mddevs_lock);
 208                 return new;
 209         }
 210         spin_unlock(&all_mddevs_lock);
 211
 212         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 213         if (!new)
 214                 return NULL;
 215
 216         memset(new, 0, sizeof(*new));
 217
 218         new->unit = unit;
 219         if (MAJOR(unit) == MD_MAJOR)
 220                 new->md_minor = MINOR(unit);
 221         else
 222                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 223
 224         init_MUTEX(&new->reconfig_sem);
 225         INIT_LIST_HEAD(&new->disks);
 226         INIT_LIST_HEAD(&new->all_mddevs);
 227         init_timer(&new->safemode_timer);
 228         atomic_set(&new->active, 1);
 229         spin_lock_init(&new->write_lock);
 230         init_waitqueue_head(&new->sb_wait);
 231
 232         new->queue = blk_alloc_queue(GFP_KERNEL);
 233         if (!new->queue) {
 234                 kfree(new);
 235                 return NULL;
 236         }
 237
 238         blk_queue_make_request(new->queue, md_fail_request);
 239
 240         goto retry;
 241 }
 242
 243 static inline int mddev_lock(mddev_t * mddev)
 244 {
 245         return down_interruptible(&mddev->reconfig_sem);
 246 }
 247
 248 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 249 {
 250         down(&mddev->reconfig_sem);
 251 }
 252
 253 static inline int mddev_trylock(mddev_t * mddev)
 254 {
 255         return down_trylock(&mddev->reconfig_sem);
 256 }
 257
 258 static inline void mddev_unlock(mddev_t * mddev)
 259 {
 260         up(&mddev->reconfig_sem);
 261
 262         md_wakeup_thread(mddev->thread);
 263 }
 264
 265 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 266 {
 267         mdk_rdev_t * rdev;
 268         struct list_head *tmp;
 269
 270         ITERATE_RDEV(mddev,rdev,tmp) {
 271                 if (rdev->desc_nr == nr)
 272                         return rdev;
 273         }
 274         return NULL;
 275 }
 276
 277 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 278 {
 279         struct list_head *tmp;
 280         mdk_rdev_t *rdev;
 281
 282         ITERATE_RDEV(mddev,rdev,tmp) {
 283                 if (rdev->bdev->bd_dev == dev)
 284                         return rdev;
 285         }
 286         return NULL;
 287 }
 288
 289 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 290 {
 291         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 292         return MD_NEW_SIZE_BLOCKS(size);
 293 }
 294
 295 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 296 {
 297         sector_t size;
 298
 299         size = rdev->sb_offset;
 300
 301         if (chunk_size)
 302                 size &= ~((sector_t)chunk_size/1024 - 1);
 303         return size;
 304 }
 305
 306 static int alloc_disk_sb(mdk_rdev_t * rdev)
 307 {
 308         if (rdev->sb_page)
 309                 MD_BUG();
 310
 311         rdev->sb_page = alloc_page(GFP_KERNEL);
 312         if (!rdev->sb_page) {
 313                 printk(KERN_ALERT "md: out of memory.\n");
 314                 return -EINVAL;
 315         }
 316
 317         return 0;
 318 }
 319
 320 static void free_disk_sb(mdk_rdev_t * rdev)
 321 {
 322         if (rdev->sb_page) {
 323                 page_cache_release(rdev->sb_page);
 324                 rdev->sb_loaded = 0;
 325                 rdev->sb_page = NULL;
 326                 rdev->sb_offset = 0;
 327                 rdev->size = 0;
 328         }
 329 }
 330
 331
 332 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 333 {
 334         mdk_rdev_t *rdev = bio->bi_private;
 335         mddev_t *mddev = rdev->mddev;
 336         if (bio->bi_size)
 337                 return 1;
 338
 339         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 340                 md_error(mddev, rdev);
 341
 342         if (atomic_dec_and_test(&mddev->pending_writes))
 343                 wake_up(&mddev->sb_wait);
 344         bio_put(bio);
 345         return 0;
 346 }
 347
 348 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 349 {
 350         struct bio *bio2 = bio->bi_private;
 351         mdk_rdev_t *rdev = bio2->bi_private;
 352         mddev_t *mddev = rdev->mddev;
 353         if (bio->bi_size)
 354                 return 1;
 355
 356         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 357             error == -EOPNOTSUPP) {
 358                 unsigned long flags;
 359                 /* barriers don't appear to be supported :-( */
 360                 set_bit(BarriersNotsupp, &rdev->flags);
 361                 mddev->barriers_work = 0;
 362                 spin_lock_irqsave(&mddev->write_lock, flags);
 363                 bio2->bi_next = mddev->biolist;
 364                 mddev->biolist = bio2;
 365                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 366                 wake_up(&mddev->sb_wait);
 367                 bio_put(bio);
 368                 return 0;
 369         }
 370         bio_put(bio2);
 371         bio->bi_private = rdev;
 372         return super_written(bio, bytes_done, error);
 373 }
 374
 375 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 376                    sector_t sector, int size, struct page *page)
 377 {
 378         /* write first size bytes of page to sector of rdev
 379          * Increment mddev->pending_writes before returning
 380          * and decrement it on completion, waking up sb_wait
 381          * if zero is reached.
 382          * If an error occurred, call md_error
 383          *
 384          * As we might need to resubmit the request if BIO_RW_BARRIER
 385          * causes ENOTSUPP, we allocate a spare bio...
 386          */
 387         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 388         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 389
 390         bio->bi_bdev = rdev->bdev;
 391         bio->bi_sector = sector;
 392         bio_add_page(bio, page, size, 0);
 393         bio->bi_private = rdev;
 394         bio->bi_end_io = super_written;
 395         bio->bi_rw = rw;
 396
 397         atomic_inc(&mddev->pending_writes);
 398         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 399                 struct bio *rbio;
 400                 rw |= (1<<BIO_RW_BARRIER);
 401                 rbio = bio_clone(bio, GFP_NOIO);
 402                 rbio->bi_private = bio;
 403                 rbio->bi_end_io = super_written_barrier;
 404                 submit_bio(rw, rbio);
 405         } else
 406                 submit_bio(rw, bio);
 407 }
 408
 409 void md_super_wait(mddev_t *mddev)
 410 {
 411         /* wait for all superblock writes that were scheduled to complete.
 412          * if any had to be retried (due to BARRIER problems), retry them
 413          */
 414         DEFINE_WAIT(wq);
 415         for(;;) {
 416                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 417                 if (atomic_read(&mddev->pending_writes)==0)
 418                         break;
 419                 while (mddev->biolist) {
 420                         struct bio *bio;
 421                         spin_lock_irq(&mddev->write_lock);
 422                         bio = mddev->biolist;
 423                         mddev->biolist = bio->bi_next ;
 424                         bio->bi_next = NULL;
 425                         spin_unlock_irq(&mddev->write_lock);
 426                         submit_bio(bio->bi_rw, bio);
 427                 }
 428                 schedule();
 429         }
 430         finish_wait(&mddev->sb_wait, &wq);
 431 }
 432
 433 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 434 {
 435         if (bio->bi_size)
 436                 return 1;
 437
 438         complete((struct completion*)bio->bi_private);
 439         return 0;
 440 }
 441
 442 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 443                    struct page *page, int rw)
 444 {
 445         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 446         struct completion event;
 447         int ret;
 448
 449         rw |= (1 << BIO_RW_SYNC);
 450
 451         bio->bi_bdev = bdev;
 452         bio->bi_sector = sector;
 453         bio_add_page(bio, page, size, 0);
 454         init_completion(&event);
 455         bio->bi_private = &event;
 456         bio->bi_end_io = bi_complete;
 457         submit_bio(rw, bio);
 458         wait_for_completion(&event);
 459
 460         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 461         bio_put(bio);
 462         return ret;
 463 }
 464
 465 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 466 {
 467         char b[BDEVNAME_SIZE];
 468         if (!rdev->sb_page) {
 469                 MD_BUG();
 470                 return -EINVAL;
 471         }
 472         if (rdev->sb_loaded)
 473                 return 0;
 474
 475
 476         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 477                 goto fail;
 478         rdev->sb_loaded = 1;
 479         return 0;
 480
 481 fail:
 482         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 483                 bdevname(rdev->bdev,b));
 484         return -EINVAL;
 485 }
 486
 487 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 488 {
 489         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 490                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 491                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 492                 (sb1->set_uuid3 == sb2->set_uuid3))
 493
 494                 return 1;
 495
 496         return 0;
 497 }
 498
 499
 500 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 501 {
 502         int ret;
 503         mdp_super_t *tmp1, *tmp2;
 504
 505         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 506         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 507
 508         if (!tmp1 || !tmp2) {
 509                 ret = 0;
 510                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 511                 goto abort;
 512         }
 513
 514         *tmp1 = *sb1;
 515         *tmp2 = *sb2;
 516
 517         /*
 518          * nr_disks is not constant
 519          */
 520         tmp1->nr_disks = 0;
 521         tmp2->nr_disks = 0;
 522
 523         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 524                 ret = 0;
 525         else
 526                 ret = 1;
 527
 528 abort:
 529         kfree(tmp1);
 530         kfree(tmp2);
 531         return ret;
 532 }
 533
 534 static unsigned int calc_sb_csum(mdp_super_t * sb)
 535 {
 536         unsigned int disk_csum, csum;
 537
 538         disk_csum = sb->sb_csum;
 539         sb->sb_csum = 0;
 540         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 541         sb->sb_csum = disk_csum;
 542         return csum;
 543 }
 544
 545
 546 /*
 547  * Handle superblock details.
 548  * We want to be able to handle multiple superblock formats
 549  * so we have a common interface to them all, and an array of
 550  * different handlers.
 551  * We rely on user-space to write the initial superblock, and support
 552  * reading and updating of superblocks.
 553  * Interface methods are:
 554  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 555  *      loads and validates a superblock on dev.
 556  *      if refdev != NULL, compare superblocks on both devices
 557  *    Return:
 558  *      0 - dev has a superblock that is compatible with refdev
 559  *      1 - dev has a superblock that is compatible and newer than refdev
 560  *          so dev should be used as the refdev in future
 561  *     -EINVAL superblock incompatible or invalid
 562  *     -othererror e.g. -EIO
 563  *
 564  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 565  *      Verify that dev is acceptable into mddev.
 566  *       The first time, mddev->raid_disks will be 0, and data from
 567  *       dev should be merged in.  Subsequent calls check that dev
 568  *       is new enough.  Return 0 or -EINVAL
 569  *
 570  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 571  *     Update the superblock for rdev with data in mddev
 572  *     This does not write to disc.
 573  *
 574  */
 575
 576 struct super_type  {
 577         char            *name;
 578         struct module   *owner;
 579         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 580         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 581         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 582 };
 583
 584 /*
 585  * load_super for 0.90.0
 586  */
 587 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 588 {
 589         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 590         mdp_super_t *sb;
 591         int ret;
 592         sector_t sb_offset;
 593
 594         /*
 595          * Calculate the position of the superblock,
 596          * it's at the end of the disk.
 597          *
 598          * It also happens to be a multiple of 4Kb.
 599          */
 600         sb_offset = calc_dev_sboffset(rdev->bdev);
 601         rdev->sb_offset = sb_offset;
 602
 603         ret = read_disk_sb(rdev, MD_SB_BYTES);
 604         if (ret) return ret;
 605
 606         ret = -EINVAL;
 607
 608         bdevname(rdev->bdev, b);
 609         sb = (mdp_super_t*)page_address(rdev->sb_page);
 610
 611         if (sb->md_magic != MD_SB_MAGIC) {
 612                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 613                        b);
 614                 goto abort;
 615         }
 616
 617         if (sb->major_version != 0 ||
 618             sb->minor_version != 90) {
 619                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 620                         sb->major_version, sb->minor_version,
 621                         b);
 622                 goto abort;
 623         }
 624
 625         if (sb->raid_disks <= 0)
 626                 goto abort;
 627
 628         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 629                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 630                         b);
 631                 goto abort;
 632         }
 633
 634         rdev->preferred_minor = sb->md_minor;
 635         rdev->data_offset = 0;
 636         rdev->sb_size = MD_SB_BYTES;
 637
 638         if (sb->level == LEVEL_MULTIPATH)
 639                 rdev->desc_nr = -1;
 640         else
 641                 rdev->desc_nr = sb->this_disk.number;
 642
 643         if (refdev == 0)
 644                 ret = 1;
 645         else {
 646                 __u64 ev1, ev2;
 647                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 648                 if (!uuid_equal(refsb, sb)) {
 649                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 650                                 b, bdevname(refdev->bdev,b2));
 651                         goto abort;
 652                 }
 653                 if (!sb_equal(refsb, sb)) {
 654                         printk(KERN_WARNING "md: %s has same UUID"
 655                                " but different superblock to %s\n",
 656                                b, bdevname(refdev->bdev, b2));
 657                         goto abort;
 658                 }
 659                 ev1 = md_event(sb);
 660                 ev2 = md_event(refsb);
 661                 if (ev1 > ev2)
 662                         ret = 1;
 663                 else
 664                         ret = 0;
 665         }
 666         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 667
 668  abort:
 669         return ret;
 670 }
 671
 672 /*
 673  * validate_super for 0.90.0
 674  */
 675 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 676 {
 677         mdp_disk_t *desc;
 678         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 679
 680         rdev->raid_disk = -1;
 681         rdev->flags = 0;
 682         if (mddev->raid_disks == 0) {
 683                 mddev->major_version = 0;
 684                 mddev->minor_version = sb->minor_version;
 685                 mddev->patch_version = sb->patch_version;
 686                 mddev->persistent = ! sb->not_persistent;
 687                 mddev->chunk_size = sb->chunk_size;
 688                 mddev->ctime = sb->ctime;
 689                 mddev->utime = sb->utime;
 690                 mddev->level = sb->level;
 691                 mddev->layout = sb->layout;
 692                 mddev->raid_disks = sb->raid_disks;
 693                 mddev->size = sb->size;
 694                 mddev->events = md_event(sb);
 695                 mddev->bitmap_offset = 0;
 696                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 697
 698                 if (sb->state & (1<<MD_SB_CLEAN))
 699                         mddev->recovery_cp = MaxSector;
 700                 else {
 701                         if (sb->events_hi == sb->cp_events_hi &&
 702                                 sb->events_lo == sb->cp_events_lo) {
 703                                 mddev->recovery_cp = sb->recovery_cp;
 704                         } else
 705                                 mddev->recovery_cp = 0;
 706                 }
 707
 708                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 709                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 710                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 711                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 712
 713                 mddev->max_disks = MD_SB_DISKS;
 714
 715                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 716                     mddev->bitmap_file == NULL) {
 717                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
 718                                 /* FIXME use a better test */
 719                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
 720                                 return -EINVAL;
 721                         }
 722                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 723                 }
 724
 725         } else if (mddev->pers == NULL) {
 726                 /* Insist on good event counter while assembling */
 727                 __u64 ev1 = md_event(sb);
 728                 ++ev1;
 729                 if (ev1 < mddev->events)
 730                         return -EINVAL;
 731         } else if (mddev->bitmap) {
 732                 /* if adding to array with a bitmap, then we can accept an
 733                  * older device ... but not too old.
 734                  */
 735                 __u64 ev1 = md_event(sb);
 736                 if (ev1 < mddev->bitmap->events_cleared)
 737                         return 0;
 738         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 739                 return 0;
 740
 741         if (mddev->level != LEVEL_MULTIPATH) {
 742                 desc = sb->disks + rdev->desc_nr;
 743
 744                 if (desc->state & (1<<MD_DISK_FAULTY))
 745                         set_bit(Faulty, &rdev->flags);
 746                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 747                          desc->raid_disk < mddev->raid_disks) {
 748                         set_bit(In_sync, &rdev->flags);
 749                         rdev->raid_disk = desc->raid_disk;
 750                 }
 751                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 752                         set_bit(WriteMostly, &rdev->flags);
 753         } else /* MULTIPATH are always insync */
 754                 set_bit(In_sync, &rdev->flags);
 755         return 0;
 756 }
 757
 758 /*
 759  * sync_super for 0.90.0
 760  */
 761 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 762 {
 763         mdp_super_t *sb;
 764         struct list_head *tmp;
 765         mdk_rdev_t *rdev2;
 766         int next_spare = mddev->raid_disks;
 767
 768
 769         /* make rdev->sb match mddev data..
 770          *
 771          * 1/ zero out disks
 772          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 773          * 3/ any empty disks < next_spare become removed
 774          *
 775          * disks[0] gets initialised to REMOVED because
 776          * we cannot be sure from other fields if it has
 777          * been initialised or not.
 778          */
 779         int i;
 780         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 781
 782         rdev->sb_size = MD_SB_BYTES;
 783
 784         sb = (mdp_super_t*)page_address(rdev->sb_page);
 785
 786         memset(sb, 0, sizeof(*sb));
 787
 788         sb->md_magic = MD_SB_MAGIC;
 789         sb->major_version = mddev->major_version;
 790         sb->minor_version = mddev->minor_version;
 791         sb->patch_version = mddev->patch_version;
 792         sb->gvalid_words  = 0; /* ignored */
 793         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 794         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 795         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 796         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 797
 798         sb->ctime = mddev->ctime;
 799         sb->level = mddev->level;
 800         sb->size  = mddev->size;
 801         sb->raid_disks = mddev->raid_disks;
 802         sb->md_minor = mddev->md_minor;
 803         sb->not_persistent = !mddev->persistent;
 804         sb->utime = mddev->utime;
 805         sb->state = 0;
 806         sb->events_hi = (mddev->events>>32);
 807         sb->events_lo = (u32)mddev->events;
 808
 809         if (mddev->in_sync)
 810         {
 811                 sb->recovery_cp = mddev->recovery_cp;
 812                 sb->cp_events_hi = (mddev->events>>32);
 813                 sb->cp_events_lo = (u32)mddev->events;
 814                 if (mddev->recovery_cp == MaxSector)
 815                         sb->state = (1<< MD_SB_CLEAN);
 816         } else
 817                 sb->recovery_cp = 0;
 818
 819         sb->layout = mddev->layout;
 820         sb->chunk_size = mddev->chunk_size;
 821
 822         if (mddev->bitmap && mddev->bitmap_file == NULL)
 823                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 824
 825         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 826         ITERATE_RDEV(mddev,rdev2,tmp) {
 827                 mdp_disk_t *d;
 828                 int desc_nr;
 829                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 830                     && !test_bit(Faulty, &rdev2->flags))
 831                         desc_nr = rdev2->raid_disk;
 832                 else
 833                         desc_nr = next_spare++;
 834                 rdev2->desc_nr = desc_nr;
 835                 d = &sb->disks[rdev2->desc_nr];
 836                 nr_disks++;
 837                 d->number = rdev2->desc_nr;
 838                 d->major = MAJOR(rdev2->bdev->bd_dev);
 839                 d->minor = MINOR(rdev2->bdev->bd_dev);
 840                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 841                     && !test_bit(Faulty, &rdev2->flags))
 842                         d->raid_disk = rdev2->raid_disk;
 843                 else
 844                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 845                 if (test_bit(Faulty, &rdev2->flags)) {
 846                         d->state = (1<<MD_DISK_FAULTY);
 847                         failed++;
 848                 } else if (test_bit(In_sync, &rdev2->flags)) {
 849                         d->state = (1<<MD_DISK_ACTIVE);
 850                         d->state |= (1<<MD_DISK_SYNC);
 851                         active++;
 852                         working++;
 853                 } else {
 854                         d->state = 0;
 855                         spare++;
 856                         working++;
 857                 }
 858                 if (test_bit(WriteMostly, &rdev2->flags))
 859                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 860         }
 861         /* now set the "removed" and "faulty" bits on any missing devices */
 862         for (i=0 ; i < mddev->raid_disks ; i++) {
 863                 mdp_disk_t *d = &sb->disks[i];
 864                 if (d->state == 0 && d->number == 0) {
 865                         d->number = i;
 866                         d->raid_disk = i;
 867                         d->state = (1<<MD_DISK_REMOVED);
 868                         d->state |= (1<<MD_DISK_FAULTY);
 869                         failed++;
 870                 }
 871         }
 872         sb->nr_disks = nr_disks;
 873         sb->active_disks = active;
 874         sb->working_disks = working;
 875         sb->failed_disks = failed;
 876         sb->spare_disks = spare;
 877
 878         sb->this_disk = sb->disks[rdev->desc_nr];
 879         sb->sb_csum = calc_sb_csum(sb);
 880 }
 881
 882 /*
 883  * version 1 superblock
 884  */
 885
 886 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 887 {
 888         unsigned int disk_csum, csum;
 889         unsigned long long newcsum;
 890         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 891         unsigned int *isuper = (unsigned int*)sb;
 892         int i;
 893
 894         disk_csum = sb->sb_csum;
 895         sb->sb_csum = 0;
 896         newcsum = 0;
 897         for (i=0; size>=4; size -= 4 )
 898                 newcsum += le32_to_cpu(*isuper++);
 899
 900         if (size == 2)
 901                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 902
 903         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 904         sb->sb_csum = disk_csum;
 905         return cpu_to_le32(csum);
 906 }
 907
 908 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 909 {
 910         struct mdp_superblock_1 *sb;
 911         int ret;
 912         sector_t sb_offset;
 913         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 914         int bmask;
 915
 916         /*
 917          * Calculate the position of the superblock.
 918          * It is always aligned to a 4K boundary and
 919          * depeding on minor_version, it can be:
 920          * 0: At least 8K, but less than 12K, from end of device
 921          * 1: At start of device
 922          * 2: 4K from start of device.
 923          */
 924         switch(minor_version) {
 925         case 0:
 926                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 927                 sb_offset -= 8*2;
 928                 sb_offset &= ~(sector_t)(4*2-1);
 929                 /* convert from sectors to K */
 930                 sb_offset /= 2;
 931                 break;
 932         case 1:
 933                 sb_offset = 0;
 934                 break;
 935         case 2:
 936                 sb_offset = 4;
 937                 break;
 938         default:
 939                 return -EINVAL;
 940         }
 941         rdev->sb_offset = sb_offset;
 942
 943         /* superblock is rarely larger than 1K, but it can be larger,
 944          * and it is safe to read 4k, so we do that
 945          */
 946         ret = read_disk_sb(rdev, 4096);
 947         if (ret) return ret;
 948
 949
 950         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 951
 952         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 953             sb->major_version != cpu_to_le32(1) ||
 954             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 955             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 956             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 957                 return -EINVAL;
 958
 959         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 960                 printk("md: invalid superblock checksum on %s\n",
 961                         bdevname(rdev->bdev,b));
 962                 return -EINVAL;
 963         }
 964         if (le64_to_cpu(sb->data_size) < 10) {
 965                 printk("md: data_size too small on %s\n",
 966                        bdevname(rdev->bdev,b));
 967                 return -EINVAL;
 968         }
 969         rdev->preferred_minor = 0xffff;
 970         rdev->data_offset = le64_to_cpu(sb->data_offset);
 971
 972         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 973         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 974         if (rdev->sb_size & bmask)
 975                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 976
 977         if (refdev == 0)
 978                 return 1;
 979         else {
 980                 __u64 ev1, ev2;
 981                 struct mdp_superblock_1 *refsb =
 982                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 983
 984                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 985                     sb->level != refsb->level ||
 986                     sb->layout != refsb->layout ||
 987                     sb->chunksize != refsb->chunksize) {
 988                         printk(KERN_WARNING "md: %s has strangely different"
 989                                 " superblock to %s\n",
 990                                 bdevname(rdev->bdev,b),
 991                                 bdevname(refdev->bdev,b2));
 992                         return -EINVAL;
 993                 }
 994                 ev1 = le64_to_cpu(sb->events);
 995                 ev2 = le64_to_cpu(refsb->events);
 996
 997                 if (ev1 > ev2)
 998                         return 1;
 999         }
1000         if (minor_version)
1001                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1002         else
1003                 rdev->size = rdev->sb_offset;
1004         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1005                 return -EINVAL;
1006         rdev->size = le64_to_cpu(sb->data_size)/2;
1007         if (le32_to_cpu(sb->chunksize))
1008                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1009         return 0;
1010 }
1011
1012 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1013 {
1014         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1015
1016         rdev->raid_disk = -1;
1017         rdev->flags = 0;
1018         if (mddev->raid_disks == 0) {
1019                 mddev->major_version = 1;
1020                 mddev->patch_version = 0;
1021                 mddev->persistent = 1;
1022                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1023                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1024                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1025                 mddev->level = le32_to_cpu(sb->level);
1026                 mddev->layout = le32_to_cpu(sb->layout);
1027                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1028                 mddev->size = le64_to_cpu(sb->size)/2;
1029                 mddev->events = le64_to_cpu(sb->events);
1030                 mddev->bitmap_offset = 0;
1031                 mddev->default_bitmap_offset = 0;
1032                 mddev->default_bitmap_offset = 1024;
1033
1034                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1035                 memcpy(mddev->uuid, sb->set_uuid, 16);
1036
1037                 mddev->max_disks =  (4096-256)/2;
1038
1039                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1040                     mddev->bitmap_file == NULL ) {
1041                         if (mddev->level != 1) {
1042                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1043                                 return -EINVAL;
1044                         }
1045                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1046                 }
1047         } else if (mddev->pers == NULL) {
1048                 /* Insist of good event counter while assembling */
1049                 __u64 ev1 = le64_to_cpu(sb->events);
1050                 ++ev1;
1051                 if (ev1 < mddev->events)
1052                         return -EINVAL;
1053         } else if (mddev->bitmap) {
1054                 /* If adding to array with a bitmap, then we can accept an
1055                  * older device, but not too old.
1056                  */
1057                 __u64 ev1 = le64_to_cpu(sb->events);
1058                 if (ev1 < mddev->bitmap->events_cleared)
1059                         return 0;
1060         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1061                 return 0;
1062
1063         if (mddev->level != LEVEL_MULTIPATH) {
1064                 int role;
1065                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1066                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1067                 switch(role) {
1068                 case 0xffff: /* spare */
1069                         break;
1070                 case 0xfffe: /* faulty */
1071                         set_bit(Faulty, &rdev->flags);
1072                         break;
1073                 default:
1074                         set_bit(In_sync, &rdev->flags);
1075                         rdev->raid_disk = role;
1076                         break;
1077                 }
1078                 if (sb->devflags & WriteMostly1)
1079                         set_bit(WriteMostly, &rdev->flags);
1080         } else /* MULTIPATH are always insync */
1081                 set_bit(In_sync, &rdev->flags);
1082
1083         return 0;
1084 }
1085
1086 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1087 {
1088         struct mdp_superblock_1 *sb;
1089         struct list_head *tmp;
1090         mdk_rdev_t *rdev2;
1091         int max_dev, i;
1092         /* make rdev->sb match mddev and rdev data. */
1093
1094         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1095
1096         sb->feature_map = 0;
1097         sb->pad0 = 0;
1098         memset(sb->pad1, 0, sizeof(sb->pad1));
1099         memset(sb->pad2, 0, sizeof(sb->pad2));
1100         memset(sb->pad3, 0, sizeof(sb->pad3));
1101
1102         sb->utime = cpu_to_le64((__u64)mddev->utime);
1103         sb->events = cpu_to_le64(mddev->events);
1104         if (mddev->in_sync)
1105                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1106         else
1107                 sb->resync_offset = cpu_to_le64(0);
1108
1109         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1110                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1111                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1112         }
1113
1114         max_dev = 0;
1115         ITERATE_RDEV(mddev,rdev2,tmp)
1116                 if (rdev2->desc_nr+1 > max_dev)
1117                         max_dev = rdev2->desc_nr+1;
1118
1119         sb->max_dev = cpu_to_le32(max_dev);
1120         for (i=0; i<max_dev;i++)
1121                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1122
1123         ITERATE_RDEV(mddev,rdev2,tmp) {
1124                 i = rdev2->desc_nr;
1125                 if (test_bit(Faulty, &rdev2->flags))
1126                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1127                 else if (test_bit(In_sync, &rdev2->flags))
1128                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1129                 else
1130                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1131         }
1132
1133         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1134         sb->sb_csum = calc_sb_1_csum(sb);
1135 }
1136
1137
1138 static struct super_type super_types[] = {
1139         [0] = {
1140                 .name   = "0.90.0",
1141                 .owner  = THIS_MODULE,
1142                 .load_super     = super_90_load,
1143                 .validate_super = super_90_validate,
1144                 .sync_super     = super_90_sync,
1145         },
1146         [1] = {
1147                 .name   = "md-1",
1148                 .owner  = THIS_MODULE,
1149                 .load_super     = super_1_load,
1150                 .validate_super = super_1_validate,
1151                 .sync_super     = super_1_sync,
1152         },
1153 };
1154
1155 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1156 {
1157         struct list_head *tmp;
1158         mdk_rdev_t *rdev;
1159
1160         ITERATE_RDEV(mddev,rdev,tmp)
1161                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1162                         return rdev;
1163
1164         return NULL;
1165 }
1166
1167 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1168 {
1169         struct list_head *tmp;
1170         mdk_rdev_t *rdev;
1171
1172         ITERATE_RDEV(mddev1,rdev,tmp)
1173                 if (match_dev_unit(mddev2, rdev))
1174                         return 1;
1175
1176         return 0;
1177 }
1178
1179 static LIST_HEAD(pending_raid_disks);
1180
1181 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1182 {
1183         mdk_rdev_t *same_pdev;
1184         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1185         struct kobject *ko;
1186
1187         if (rdev->mddev) {
1188                 MD_BUG();
1189                 return -EINVAL;
1190         }
1191         same_pdev = match_dev_unit(mddev, rdev);
1192         if (same_pdev)
1193                 printk(KERN_WARNING
1194                         "%s: WARNING: %s appears to be on the same physical"
1195                         " disk as %s. True\n     protection against single-disk"
1196                         " failure might be compromised.\n",
1197                         mdname(mddev), bdevname(rdev->bdev,b),
1198                         bdevname(same_pdev->bdev,b2));
1199
1200         /* Verify rdev->desc_nr is unique.
1201          * If it is -1, assign a free number, else
1202          * check number is not in use
1203          */
1204         if (rdev->desc_nr < 0) {
1205                 int choice = 0;
1206                 if (mddev->pers) choice = mddev->raid_disks;
1207                 while (find_rdev_nr(mddev, choice))
1208                         choice++;
1209                 rdev->desc_nr = choice;
1210         } else {
1211                 if (find_rdev_nr(mddev, rdev->desc_nr))
1212                         return -EBUSY;
1213         }
1214         bdevname(rdev->bdev,b);
1215         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1216                 return -ENOMEM;
1217
1218         list_add(&rdev->same_set, &mddev->disks);
1219         rdev->mddev = mddev;
1220         printk(KERN_INFO "md: bind<%s>\n", b);
1221
1222         rdev->kobj.parent = &mddev->kobj;
1223         kobject_add(&rdev->kobj);
1224
1225         if (rdev->bdev->bd_part)
1226                 ko = &rdev->bdev->bd_part->kobj;
1227         else
1228                 ko = &rdev->bdev->bd_disk->kobj;
1229         sysfs_create_link(&rdev->kobj, ko, "block");
1230         return 0;
1231 }
1232
1233 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1234 {
1235         char b[BDEVNAME_SIZE];
1236         if (!rdev->mddev) {
1237                 MD_BUG();
1238                 return;
1239         }
1240         list_del_init(&rdev->same_set);
1241         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1242         rdev->mddev = NULL;
1243         sysfs_remove_link(&rdev->kobj, "block");
1244         kobject_del(&rdev->kobj);
1245 }
1246
1247 /*
1248  * prevent the device from being mounted, repartitioned or
1249  * otherwise reused by a RAID array (or any other kernel
1250  * subsystem), by bd_claiming the device.
1251  */
1252 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1253 {
1254         int err = 0;
1255         struct block_device *bdev;
1256         char b[BDEVNAME_SIZE];
1257
1258         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1259         if (IS_ERR(bdev)) {
1260                 printk(KERN_ERR "md: could not open %s.\n",
1261                         __bdevname(dev, b));
1262                 return PTR_ERR(bdev);
1263         }
1264         err = bd_claim(bdev, rdev);
1265         if (err) {
1266                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1267                         bdevname(bdev, b));
1268                 blkdev_put(bdev);
1269                 return err;
1270         }
1271         rdev->bdev = bdev;
1272         return err;
1273 }
1274
1275 static void unlock_rdev(mdk_rdev_t *rdev)
1276 {
1277         struct block_device *bdev = rdev->bdev;
1278         rdev->bdev = NULL;
1279         if (!bdev)
1280                 MD_BUG();
1281         bd_release(bdev);
1282         blkdev_put(bdev);
1283 }
1284
1285 void md_autodetect_dev(dev_t dev);
1286
1287 static void export_rdev(mdk_rdev_t * rdev)
1288 {
1289         char b[BDEVNAME_SIZE];
1290         printk(KERN_INFO "md: export_rdev(%s)\n",
1291                 bdevname(rdev->bdev,b));
1292         if (rdev->mddev)
1293                 MD_BUG();
1294         free_disk_sb(rdev);
1295         list_del_init(&rdev->same_set);
1296 #ifndef MODULE
1297         md_autodetect_dev(rdev->bdev->bd_dev);
1298 #endif
1299         unlock_rdev(rdev);
1300         kobject_put(&rdev->kobj);
1301 }
1302
1303 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1304 {
1305         unbind_rdev_from_array(rdev);
1306         export_rdev(rdev);
1307 }
1308
1309 static void export_array(mddev_t *mddev)
1310 {
1311         struct list_head *tmp;
1312         mdk_rdev_t *rdev;
1313
1314         ITERATE_RDEV(mddev,rdev,tmp) {
1315                 if (!rdev->mddev) {
1316                         MD_BUG();
1317                         continue;
1318                 }
1319                 kick_rdev_from_array(rdev);
1320         }
1321         if (!list_empty(&mddev->disks))
1322                 MD_BUG();
1323         mddev->raid_disks = 0;
1324         mddev->major_version = 0;
1325 }
1326
1327 static void print_desc(mdp_disk_t *desc)
1328 {
1329         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1330                 desc->major,desc->minor,desc->raid_disk,desc->state);
1331 }
1332
1333 static void print_sb(mdp_super_t *sb)
1334 {
1335         int i;
1336
1337         printk(KERN_INFO
1338                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1339                 sb->major_version, sb->minor_version, sb->patch_version,
1340                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1341                 sb->ctime);
1342         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1343                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1344                 sb->md_minor, sb->layout, sb->chunk_size);
1345         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1346                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1347                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1348                 sb->failed_disks, sb->spare_disks,
1349                 sb->sb_csum, (unsigned long)sb->events_lo);
1350
1351         printk(KERN_INFO);
1352         for (i = 0; i < MD_SB_DISKS; i++) {
1353                 mdp_disk_t *desc;
1354
1355                 desc = sb->disks + i;
1356                 if (desc->number || desc->major || desc->minor ||
1357                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1358                         printk("     D %2d: ", i);
1359                         print_desc(desc);
1360                 }
1361         }
1362         printk(KERN_INFO "md:     THIS: ");
1363         print_desc(&sb->this_disk);
1364
1365 }
1366
1367 static void print_rdev(mdk_rdev_t *rdev)
1368 {
1369         char b[BDEVNAME_SIZE];
1370         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1371                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1372                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1373                 rdev->desc_nr);
1374         if (rdev->sb_loaded) {
1375                 printk(KERN_INFO "md: rdev superblock:\n");
1376                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1377         } else
1378                 printk(KERN_INFO "md: no rdev superblock!\n");
1379 }
1380
1381 void md_print_devices(void)
1382 {
1383         struct list_head *tmp, *tmp2;
1384         mdk_rdev_t *rdev;
1385         mddev_t *mddev;
1386         char b[BDEVNAME_SIZE];
1387
1388         printk("\n");
1389         printk("md:     **********************************\n");
1390         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1391         printk("md:     **********************************\n");
1392         ITERATE_MDDEV(mddev,tmp) {
1393
1394                 if (mddev->bitmap)
1395                         bitmap_print_sb(mddev->bitmap);
1396                 else
1397                         printk("%s: ", mdname(mddev));
1398                 ITERATE_RDEV(mddev,rdev,tmp2)
1399                         printk("<%s>", bdevname(rdev->bdev,b));
1400                 printk("\n");
1401
1402                 ITERATE_RDEV(mddev,rdev,tmp2)
1403                         print_rdev(rdev);
1404         }
1405         printk("md:     **********************************\n");
1406         printk("\n");
1407 }
1408
1409
1410 static void sync_sbs(mddev_t * mddev)
1411 {
1412         mdk_rdev_t *rdev;
1413         struct list_head *tmp;
1414
1415         ITERATE_RDEV(mddev,rdev,tmp) {
1416                 super_types[mddev->major_version].
1417                         sync_super(mddev, rdev);
1418                 rdev->sb_loaded = 1;
1419         }
1420 }
1421
1422 static void md_update_sb(mddev_t * mddev)
1423 {
1424         int err;
1425         struct list_head *tmp;
1426         mdk_rdev_t *rdev;
1427         int sync_req;
1428
1429 repeat:
1430         spin_lock_irq(&mddev->write_lock);
1431         sync_req = mddev->in_sync;
1432         mddev->utime = get_seconds();
1433         mddev->events ++;
1434
1435         if (!mddev->events) {
1436                 /*
1437                  * oops, this 64-bit counter should never wrap.
1438                  * Either we are in around ~1 trillion A.C., assuming
1439                  * 1 reboot per second, or we have a bug:
1440                  */
1441                 MD_BUG();
1442                 mddev->events --;
1443         }
1444         mddev->sb_dirty = 2;
1445         sync_sbs(mddev);
1446
1447         /*
1448          * do not write anything to disk if using
1449          * nonpersistent superblocks
1450          */
1451         if (!mddev->persistent) {
1452                 mddev->sb_dirty = 0;
1453                 spin_unlock_irq(&mddev->write_lock);
1454                 wake_up(&mddev->sb_wait);
1455                 return;
1456         }
1457         spin_unlock_irq(&mddev->write_lock);
1458
1459         dprintk(KERN_INFO
1460                 "md: updating %s RAID superblock on device (in sync %d)\n",
1461                 mdname(mddev),mddev->in_sync);
1462
1463         err = bitmap_update_sb(mddev->bitmap);
1464         ITERATE_RDEV(mddev,rdev,tmp) {
1465                 char b[BDEVNAME_SIZE];
1466                 dprintk(KERN_INFO "md: ");
1467                 if (test_bit(Faulty, &rdev->flags))
1468                         dprintk("(skipping faulty ");
1469
1470                 dprintk("%s ", bdevname(rdev->bdev,b));
1471                 if (!test_bit(Faulty, &rdev->flags)) {
1472                         md_super_write(mddev,rdev,
1473                                        rdev->sb_offset<<1, rdev->sb_size,
1474                                        rdev->sb_page);
1475                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1476                                 bdevname(rdev->bdev,b),
1477                                 (unsigned long long)rdev->sb_offset);
1478
1479                 } else
1480                         dprintk(")\n");
1481                 if (mddev->level == LEVEL_MULTIPATH)
1482                         /* only need to write one superblock... */
1483                         break;
1484         }
1485         md_super_wait(mddev);
1486         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1487
1488         spin_lock_irq(&mddev->write_lock);
1489         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1490                 /* have to write it out again */
1491                 spin_unlock_irq(&mddev->write_lock);
1492                 goto repeat;
1493         }
1494         mddev->sb_dirty = 0;
1495         spin_unlock_irq(&mddev->write_lock);
1496         wake_up(&mddev->sb_wait);
1497
1498 }
1499
1500 struct rdev_sysfs_entry {
1501         struct attribute attr;
1502         ssize_t (*show)(mdk_rdev_t *, char *);
1503         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1504 };
1505
1506 static ssize_t
1507 state_show(mdk_rdev_t *rdev, char *page)
1508 {
1509         char *sep = "";
1510         int len=0;
1511
1512         if (test_bit(Faulty, &rdev->flags)) {
1513                 len+= sprintf(page+len, "%sfaulty",sep);
1514                 sep = ",";
1515         }
1516         if (test_bit(In_sync, &rdev->flags)) {
1517                 len += sprintf(page+len, "%sin_sync",sep);
1518                 sep = ",";
1519         }
1520         if (!test_bit(Faulty, &rdev->flags) &&
1521             !test_bit(In_sync, &rdev->flags)) {
1522                 len += sprintf(page+len, "%sspare", sep);
1523                 sep = ",";
1524         }
1525         return len+sprintf(page+len, "\n");
1526 }
1527
1528 static struct rdev_sysfs_entry
1529 rdev_state = __ATTR_RO(state);
1530
1531 static ssize_t
1532 super_show(mdk_rdev_t *rdev, char *page)
1533 {
1534         if (rdev->sb_loaded && rdev->sb_size) {
1535                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1536                 return rdev->sb_size;
1537         } else
1538                 return 0;
1539 }
1540 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1541
1542 static struct attribute *rdev_default_attrs[] = {
1543         &rdev_state.attr,
1544         &rdev_super.attr,
1545         NULL,
1546 };
1547 static ssize_t
1548 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1549 {
1550         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1551         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1552
1553         if (!entry->show)
1554                 return -EIO;
1555         return entry->show(rdev, page);
1556 }
1557
1558 static ssize_t
1559 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1560               const char *page, size_t length)
1561 {
1562         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1563         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1564
1565         if (!entry->store)
1566                 return -EIO;
1567         return entry->store(rdev, page, length);
1568 }
1569
1570 static void rdev_free(struct kobject *ko)
1571 {
1572         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1573         kfree(rdev);
1574 }
1575 static struct sysfs_ops rdev_sysfs_ops = {
1576         .show           = rdev_attr_show,
1577         .store          = rdev_attr_store,
1578 };
1579 static struct kobj_type rdev_ktype = {
1580         .release        = rdev_free,
1581         .sysfs_ops      = &rdev_sysfs_ops,
1582         .default_attrs  = rdev_default_attrs,
1583 };
1584
1585 /*
1586  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1587  *
1588  * mark the device faulty if:
1589  *
1590  *   - the device is nonexistent (zero size)
1591  *   - the device has no valid superblock
1592  *
1593  * a faulty rdev _never_ has rdev->sb set.
1594  */
1595 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1596 {
1597         char b[BDEVNAME_SIZE];
1598         int err;
1599         mdk_rdev_t *rdev;
1600         sector_t size;
1601
1602         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1603         if (!rdev) {
1604                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1605                 return ERR_PTR(-ENOMEM);
1606         }
1607         memset(rdev, 0, sizeof(*rdev));
1608
1609         if ((err = alloc_disk_sb(rdev)))
1610                 goto abort_free;
1611
1612         err = lock_rdev(rdev, newdev);
1613         if (err)
1614                 goto abort_free;
1615
1616         rdev->kobj.parent = NULL;
1617         rdev->kobj.ktype = &rdev_ktype;
1618         kobject_init(&rdev->kobj);
1619
1620         rdev->desc_nr = -1;
1621         rdev->flags = 0;
1622         rdev->data_offset = 0;
1623         atomic_set(&rdev->nr_pending, 0);
1624         atomic_set(&rdev->read_errors, 0);
1625
1626         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1627         if (!size) {
1628                 printk(KERN_WARNING
1629                         "md: %s has zero or unknown size, marking faulty!\n",
1630                         bdevname(rdev->bdev,b));
1631                 err = -EINVAL;
1632                 goto abort_free;
1633         }
1634
1635         if (super_format >= 0) {
1636                 err = super_types[super_format].
1637                         load_super(rdev, NULL, super_minor);
1638                 if (err == -EINVAL) {
1639                         printk(KERN_WARNING
1640                                 "md: %s has invalid sb, not importing!\n",
1641                                 bdevname(rdev->bdev,b));
1642                         goto abort_free;
1643                 }
1644                 if (err < 0) {
1645                         printk(KERN_WARNING
1646                                 "md: could not read %s's sb, not importing!\n",
1647                                 bdevname(rdev->bdev,b));
1648                         goto abort_free;
1649                 }
1650         }
1651         INIT_LIST_HEAD(&rdev->same_set);
1652
1653         return rdev;
1654
1655 abort_free:
1656         if (rdev->sb_page) {
1657                 if (rdev->bdev)
1658                         unlock_rdev(rdev);
1659                 free_disk_sb(rdev);
1660         }
1661         kfree(rdev);
1662         return ERR_PTR(err);
1663 }
1664
1665 /*
1666  * Check a full RAID array for plausibility
1667  */
1668
1669
1670 static void analyze_sbs(mddev_t * mddev)
1671 {
1672         int i;
1673         struct list_head *tmp;
1674         mdk_rdev_t *rdev, *freshest;
1675         char b[BDEVNAME_SIZE];
1676
1677         freshest = NULL;
1678         ITERATE_RDEV(mddev,rdev,tmp)
1679                 switch (super_types[mddev->major_version].
1680                         load_super(rdev, freshest, mddev->minor_version)) {
1681                 case 1:
1682                         freshest = rdev;
1683                         break;
1684                 case 0:
1685                         break;
1686                 default:
1687                         printk( KERN_ERR \
1688                                 "md: fatal superblock inconsistency in %s"
1689                                 " -- removing from array\n",
1690                                 bdevname(rdev->bdev,b));
1691                         kick_rdev_from_array(rdev);
1692                 }
1693
1694
1695         super_types[mddev->major_version].
1696                 validate_super(mddev, freshest);
1697
1698         i = 0;
1699         ITERATE_RDEV(mddev,rdev,tmp) {
1700                 if (rdev != freshest)
1701                         if (super_types[mddev->major_version].
1702                             validate_super(mddev, rdev)) {
1703                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1704                                         " from array!\n",
1705                                         bdevname(rdev->bdev,b));
1706                                 kick_rdev_from_array(rdev);
1707                                 continue;
1708                         }
1709                 if (mddev->level == LEVEL_MULTIPATH) {
1710                         rdev->desc_nr = i++;
1711                         rdev->raid_disk = rdev->desc_nr;
1712                         set_bit(In_sync, &rdev->flags);
1713                 }
1714         }
1715
1716
1717
1718         if (mddev->recovery_cp != MaxSector &&
1719             mddev->level >= 1)
1720                 printk(KERN_ERR "md: %s: raid array is not clean"
1721                        " -- starting background reconstruction\n",
1722                        mdname(mddev));
1723
1724 }
1725
1726 static ssize_t
1727 level_show(mddev_t *mddev, char *page)
1728 {
1729         mdk_personality_t *p = mddev->pers;
1730         if (p == NULL && mddev->raid_disks == 0)
1731                 return 0;
1732         if (mddev->level >= 0)
1733                 return sprintf(page, "RAID-%d\n", mddev->level);
1734         else
1735                 return sprintf(page, "%s\n", p->name);
1736 }
1737
1738 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1739
1740 static ssize_t
1741 raid_disks_show(mddev_t *mddev, char *page)
1742 {
1743         if (mddev->raid_disks == 0)
1744                 return 0;
1745         return sprintf(page, "%d\n", mddev->raid_disks);
1746 }
1747
1748 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1749
1750 static ssize_t
1751 action_show(mddev_t *mddev, char *page)
1752 {
1753         char *type = "idle";
1754         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1755             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1756                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1757                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1758                                 type = "resync";
1759                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1760                                 type = "check";
1761                         else
1762                                 type = "repair";
1763                 } else
1764                         type = "recover";
1765         }
1766         return sprintf(page, "%s\n", type);
1767 }
1768
1769 static ssize_t
1770 action_store(mddev_t *mddev, const char *page, size_t len)
1771 {
1772         if (!mddev->pers || !mddev->pers->sync_request)
1773                 return -EINVAL;
1774
1775         if (strcmp(page, "idle")==0 || strcmp(page, "idle\n")==0) {
1776                 if (mddev->sync_thread) {
1777                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1778                         md_unregister_thread(mddev->sync_thread);
1779                         mddev->sync_thread = NULL;
1780                         mddev->recovery = 0;
1781                 }
1782                 return len;
1783         }
1784
1785         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1786             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1787                 return -EBUSY;
1788         if (strcmp(page, "resync")==0 || strcmp(page, "resync\n")==0 ||
1789             strcmp(page, "recover")==0 || strcmp(page, "recover\n")==0)
1790                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1791         else {
1792                 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1793                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1794                 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1795                         return -EINVAL;
1796                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1797                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1798                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1799         }
1800         md_wakeup_thread(mddev->thread);
1801         return len;
1802 }
1803
1804 static ssize_t
1805 mismatch_cnt_show(mddev_t *mddev, char *page)
1806 {
1807         return sprintf(page, "%llu\n",
1808                        (unsigned long long) mddev->resync_mismatches);
1809 }
1810
1811 static struct md_sysfs_entry
1812 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1813
1814
1815 static struct md_sysfs_entry
1816 md_mismatches = __ATTR_RO(mismatch_cnt);
1817
1818 static struct attribute *md_default_attrs[] = {
1819         &md_level.attr,
1820         &md_raid_disks.attr,
1821         NULL,
1822 };
1823
1824 static struct attribute *md_redundancy_attrs[] = {
1825         &md_scan_mode.attr,
1826         &md_mismatches.attr,
1827         NULL,
1828 };
1829 static struct attribute_group md_redundancy_group = {
1830         .name = NULL,
1831         .attrs = md_redundancy_attrs,
1832 };
1833
1834
1835 static ssize_t
1836 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1837 {
1838         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1839         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1840         ssize_t rv;
1841
1842         if (!entry->show)
1843                 return -EIO;
1844         mddev_lock(mddev);
1845         rv = entry->show(mddev, page);
1846         mddev_unlock(mddev);
1847         return rv;
1848 }
1849
1850 static ssize_t
1851 md_attr_store(struct kobject *kobj, struct attribute *attr,
1852               const char *page, size_t length)
1853 {
1854         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1855         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1856         ssize_t rv;
1857
1858         if (!entry->store)
1859                 return -EIO;
1860         mddev_lock(mddev);
1861         rv = entry->store(mddev, page, length);
1862         mddev_unlock(mddev);
1863         return rv;
1864 }
1865
1866 static void md_free(struct kobject *ko)
1867 {
1868         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1869         kfree(mddev);
1870 }
1871
1872 static struct sysfs_ops md_sysfs_ops = {
1873         .show   = md_attr_show,
1874         .store  = md_attr_store,
1875 };
1876 static struct kobj_type md_ktype = {
1877         .release        = md_free,
1878         .sysfs_ops      = &md_sysfs_ops,
1879         .default_attrs  = md_default_attrs,
1880 };
1881
1882 int mdp_major = 0;
1883
1884 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1885 {
1886         static DECLARE_MUTEX(disks_sem);
1887         mddev_t *mddev = mddev_find(dev);
1888         struct gendisk *disk;
1889         int partitioned = (MAJOR(dev) != MD_MAJOR);
1890         int shift = partitioned ? MdpMinorShift : 0;
1891         int unit = MINOR(dev) >> shift;
1892
1893         if (!mddev)
1894                 return NULL;
1895
1896         down(&disks_sem);
1897         if (mddev->gendisk) {
1898                 up(&disks_sem);
1899                 mddev_put(mddev);
1900                 return NULL;
1901         }
1902         disk = alloc_disk(1 << shift);
1903         if (!disk) {
1904                 up(&disks_sem);
1905                 mddev_put(mddev);
1906                 return NULL;
1907         }
1908         disk->major = MAJOR(dev);
1909         disk->first_minor = unit << shift;
1910         if (partitioned) {
1911                 sprintf(disk->disk_name, "md_d%d", unit);
1912                 sprintf(disk->devfs_name, "md/d%d", unit);
1913         } else {
1914                 sprintf(disk->disk_name, "md%d", unit);
1915                 sprintf(disk->devfs_name, "md/%d", unit);
1916         }
1917         disk->fops = &md_fops;
1918         disk->private_data = mddev;
1919         disk->queue = mddev->queue;
1920         add_disk(disk);
1921         mddev->gendisk = disk;
1922         up(&disks_sem);
1923         mddev->kobj.parent = &disk->kobj;
1924         mddev->kobj.k_name = NULL;
1925         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1926         mddev->kobj.ktype = &md_ktype;
1927         kobject_register(&mddev->kobj);
1928         return NULL;
1929 }
1930
1931 void md_wakeup_thread(mdk_thread_t *thread);
1932
1933 static void md_safemode_timeout(unsigned long data)
1934 {
1935         mddev_t *mddev = (mddev_t *) data;
1936
1937         mddev->safemode = 1;
1938         md_wakeup_thread(mddev->thread);
1939 }
1940
1941
1942 static int do_md_run(mddev_t * mddev)
1943 {
1944         int pnum, err;
1945         int chunk_size;
1946         struct list_head *tmp;
1947         mdk_rdev_t *rdev;
1948         struct gendisk *disk;
1949         char b[BDEVNAME_SIZE];
1950
1951         if (list_empty(&mddev->disks))
1952                 /* cannot run an array with no devices.. */
1953                 return -EINVAL;
1954
1955         if (mddev->pers)
1956                 return -EBUSY;
1957
1958         /*
1959          * Analyze all RAID superblock(s)
1960          */
1961         if (!mddev->raid_disks)
1962                 analyze_sbs(mddev);
1963
1964         chunk_size = mddev->chunk_size;
1965         pnum = level_to_pers(mddev->level);
1966
1967         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1968                 if (!chunk_size) {
1969                         /*
1970                          * 'default chunksize' in the old md code used to
1971                          * be PAGE_SIZE, baaad.
1972                          * we abort here to be on the safe side. We don't
1973                          * want to continue the bad practice.
1974                          */
1975                         printk(KERN_ERR
1976                                 "no chunksize specified, see 'man raidtab'\n");
1977                         return -EINVAL;
1978                 }
1979                 if (chunk_size > MAX_CHUNK_SIZE) {
1980                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1981                                 chunk_size, MAX_CHUNK_SIZE);
1982                         return -EINVAL;
1983                 }
1984                 /*
1985                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1986                  */
1987                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1988                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1989                         return -EINVAL;
1990                 }
1991                 if (chunk_size < PAGE_SIZE) {
1992                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1993                                 chunk_size, PAGE_SIZE);
1994                         return -EINVAL;
1995                 }
1996
1997                 /* devices must have minimum size of one chunk */
1998                 ITERATE_RDEV(mddev,rdev,tmp) {
1999                         if (test_bit(Faulty, &rdev->flags))
2000                                 continue;
2001                         if (rdev->size < chunk_size / 1024) {
2002                                 printk(KERN_WARNING
2003                                         "md: Dev %s smaller than chunk_size:"
2004                                         " %lluk < %dk\n",
2005                                         bdevname(rdev->bdev,b),
2006                                         (unsigned long long)rdev->size,
2007                                         chunk_size / 1024);
2008                                 return -EINVAL;
2009                         }
2010                 }
2011         }
2012
2013 #ifdef CONFIG_KMOD
2014         if (!pers[pnum])
2015         {
2016                 request_module("md-personality-%d", pnum);
2017         }
2018 #endif
2019
2020         /*
2021          * Drop all container device buffers, from now on
2022          * the only valid external interface is through the md
2023          * device.
2024          * Also find largest hardsector size
2025          */
2026         ITERATE_RDEV(mddev,rdev,tmp) {
2027                 if (test_bit(Faulty, &rdev->flags))
2028                         continue;
2029                 sync_blockdev(rdev->bdev);
2030                 invalidate_bdev(rdev->bdev, 0);
2031         }
2032
2033         md_probe(mddev->unit, NULL, NULL);
2034         disk = mddev->gendisk;
2035         if (!disk)
2036                 return -ENOMEM;
2037
2038         spin_lock(&pers_lock);
2039         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2040                 spin_unlock(&pers_lock);
2041                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2042                        pnum);
2043                 return -EINVAL;
2044         }
2045
2046         mddev->pers = pers[pnum];
2047         spin_unlock(&pers_lock);
2048
2049         mddev->recovery = 0;
2050         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2051         mddev->barriers_work = 1;
2052
2053         if (start_readonly)
2054                 mddev->ro = 2; /* read-only, but switch on first write */
2055
2056         /* before we start the array running, initialise the bitmap */
2057         err = bitmap_create(mddev);
2058         if (err)
2059                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2060                         mdname(mddev), err);
2061         else
2062                 err = mddev->pers->run(mddev);
2063         if (err) {
2064                 printk(KERN_ERR "md: pers->run() failed ...\n");
2065                 module_put(mddev->pers->owner);
2066                 mddev->pers = NULL;
2067                 bitmap_destroy(mddev);
2068                 return err;
2069         }
2070         if (mddev->pers->sync_request)
2071                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2072         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2073                 mddev->ro = 0;
2074
2075         atomic_set(&mddev->writes_pending,0);
2076         mddev->safemode = 0;
2077         mddev->safemode_timer.function = md_safemode_timeout;
2078         mddev->safemode_timer.data = (unsigned long) mddev;
2079         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2080         mddev->in_sync = 1;
2081
2082         ITERATE_RDEV(mddev,rdev,tmp)
2083                 if (rdev->raid_disk >= 0) {
2084                         char nm[20];
2085                         sprintf(nm, "rd%d", rdev->raid_disk);
2086                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2087                 }
2088
2089         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2090         md_wakeup_thread(mddev->thread);
2091
2092         if (mddev->sb_dirty)
2093                 md_update_sb(mddev);
2094
2095         set_capacity(disk, mddev->array_size<<1);
2096
2097         /* If we call blk_queue_make_request here, it will
2098          * re-initialise max_sectors etc which may have been
2099          * refined inside -> run.  So just set the bits we need to set.
2100          * Most initialisation happended when we called
2101          * blk_queue_make_request(..., md_fail_request)
2102          * earlier.
2103          */
2104         mddev->queue->queuedata = mddev;
2105         mddev->queue->make_request_fn = mddev->pers->make_request;
2106
2107         mddev->changed = 1;
2108         return 0;
2109 }
2110
2111 static int restart_array(mddev_t *mddev)
2112 {
2113         struct gendisk *disk = mddev->gendisk;
2114         int err;
2115
2116         /*
2117          * Complain if it has no devices
2118          */
2119         err = -ENXIO;
2120         if (list_empty(&mddev->disks))
2121                 goto out;
2122
2123         if (mddev->pers) {
2124                 err = -EBUSY;
2125                 if (!mddev->ro)
2126                         goto out;
2127
2128                 mddev->safemode = 0;
2129                 mddev->ro = 0;
2130                 set_disk_ro(disk, 0);
2131
2132                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2133                         mdname(mddev));
2134                 /*
2135                  * Kick recovery or resync if necessary
2136                  */
2137                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2138                 md_wakeup_thread(mddev->thread);
2139                 err = 0;
2140         } else {
2141                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2142                         mdname(mddev));
2143                 err = -EINVAL;
2144         }
2145
2146 out:
2147         return err;
2148 }
2149
2150 static int do_md_stop(mddev_t * mddev, int ro)
2151 {
2152         int err = 0;
2153         struct gendisk *disk = mddev->gendisk;
2154
2155         if (mddev->pers) {
2156                 if (atomic_read(&mddev->active)>2) {
2157                         printk("md: %s still in use.\n",mdname(mddev));
2158                         return -EBUSY;
2159                 }
2160
2161                 if (mddev->sync_thread) {
2162                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2163                         md_unregister_thread(mddev->sync_thread);
2164                         mddev->sync_thread = NULL;
2165                 }
2166
2167                 del_timer_sync(&mddev->safemode_timer);
2168
2169                 invalidate_partition(disk, 0);
2170
2171                 if (ro) {
2172                         err  = -ENXIO;
2173                         if (mddev->ro==1)
2174                                 goto out;
2175                         mddev->ro = 1;
2176                 } else {
2177                         bitmap_flush(mddev);
2178                         md_super_wait(mddev);
2179                         if (mddev->ro)
2180                                 set_disk_ro(disk, 0);
2181                         blk_queue_make_request(mddev->queue, md_fail_request);
2182                         mddev->pers->stop(mddev);
2183                         if (mddev->pers->sync_request)
2184                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2185
2186                         module_put(mddev->pers->owner);
2187                         mddev->pers = NULL;
2188                         if (mddev->ro)
2189                                 mddev->ro = 0;
2190                 }
2191                 if (!mddev->in_sync) {
2192                         /* mark array as shutdown cleanly */
2193                         mddev->in_sync = 1;
2194                         md_update_sb(mddev);
2195                 }
2196                 if (ro)
2197                         set_disk_ro(disk, 1);
2198         }
2199
2200         bitmap_destroy(mddev);
2201         if (mddev->bitmap_file) {
2202                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2203                 fput(mddev->bitmap_file);
2204                 mddev->bitmap_file = NULL;
2205         }
2206         mddev->bitmap_offset = 0;
2207
2208         /*
2209          * Free resources if final stop
2210          */
2211         if (!ro) {
2212                 mdk_rdev_t *rdev;
2213                 struct list_head *tmp;
2214                 struct gendisk *disk;
2215                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2216
2217                 ITERATE_RDEV(mddev,rdev,tmp)
2218                         if (rdev->raid_disk >= 0) {
2219                                 char nm[20];
2220                                 sprintf(nm, "rd%d", rdev->raid_disk);
2221                                 sysfs_remove_link(&mddev->kobj, nm);
2222                         }
2223
2224                 export_array(mddev);
2225
2226                 mddev->array_size = 0;
2227                 disk = mddev->gendisk;
2228                 if (disk)
2229                         set_capacity(disk, 0);
2230                 mddev->changed = 1;
2231         } else
2232                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2233                         mdname(mddev));
2234         err = 0;
2235 out:
2236         return err;
2237 }
2238
2239 static void autorun_array(mddev_t *mddev)
2240 {
2241         mdk_rdev_t *rdev;
2242         struct list_head *tmp;
2243         int err;
2244
2245         if (list_empty(&mddev->disks))
2246                 return;
2247
2248         printk(KERN_INFO "md: running: ");
2249
2250         ITERATE_RDEV(mddev,rdev,tmp) {
2251                 char b[BDEVNAME_SIZE];
2252                 printk("<%s>", bdevname(rdev->bdev,b));
2253         }
2254         printk("\n");
2255
2256         err = do_md_run (mddev);
2257         if (err) {
2258                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2259                 do_md_stop (mddev, 0);
2260         }
2261 }
2262
2263 /*
2264  * lets try to run arrays based on all disks that have arrived
2265  * until now. (those are in pending_raid_disks)
2266  *
2267  * the method: pick the first pending disk, collect all disks with
2268  * the same UUID, remove all from the pending list and put them into
2269  * the 'same_array' list. Then order this list based on superblock
2270  * update time (freshest comes first), kick out 'old' disks and
2271  * compare superblocks. If everything's fine then run it.
2272  *
2273  * If "unit" is allocated, then bump its reference count
2274  */
2275 static void autorun_devices(int part)
2276 {
2277         struct list_head candidates;
2278         struct list_head *tmp;
2279         mdk_rdev_t *rdev0, *rdev;
2280         mddev_t *mddev;
2281         char b[BDEVNAME_SIZE];
2282
2283         printk(KERN_INFO "md: autorun ...\n");
2284         while (!list_empty(&pending_raid_disks)) {
2285                 dev_t dev;
2286                 rdev0 = list_entry(pending_raid_disks.next,
2287                                          mdk_rdev_t, same_set);
2288
2289                 printk(KERN_INFO "md: considering %s ...\n",
2290                         bdevname(rdev0->bdev,b));
2291                 INIT_LIST_HEAD(&candidates);
2292                 ITERATE_RDEV_PENDING(rdev,tmp)
2293                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2294                                 printk(KERN_INFO "md:  adding %s ...\n",
2295                                         bdevname(rdev->bdev,b));
2296                                 list_move(&rdev->same_set, &candidates);
2297                         }
2298                 /*
2299                  * now we have a set of devices, with all of them having
2300                  * mostly sane superblocks. It's time to allocate the
2301                  * mddev.
2302                  */
2303                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2304                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2305                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2306                         break;
2307                 }
2308                 if (part)
2309                         dev = MKDEV(mdp_major,
2310                                     rdev0->preferred_minor << MdpMinorShift);
2311                 else
2312                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2313
2314                 md_probe(dev, NULL, NULL);
2315                 mddev = mddev_find(dev);
2316                 if (!mddev) {
2317                         printk(KERN_ERR
2318                                 "md: cannot allocate memory for md drive.\n");
2319                         break;
2320                 }
2321                 if (mddev_lock(mddev))
2322                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2323                                mdname(mddev));
2324                 else if (mddev->raid_disks || mddev->major_version
2325                          || !list_empty(&mddev->disks)) {
2326                         printk(KERN_WARNING
2327                                 "md: %s already running, cannot run %s\n",
2328                                 mdname(mddev), bdevname(rdev0->bdev,b));
2329                         mddev_unlock(mddev);
2330                 } else {
2331                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2332                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2333                                 list_del_init(&rdev->same_set);
2334                                 if (bind_rdev_to_array(rdev, mddev))
2335                                         export_rdev(rdev);
2336                         }
2337                         autorun_array(mddev);
2338                         mddev_unlock(mddev);
2339                 }
2340                 /* on success, candidates will be empty, on error
2341                  * it won't...
2342                  */
2343                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2344                         export_rdev(rdev);
2345                 mddev_put(mddev);
2346         }
2347         printk(KERN_INFO "md: ... autorun DONE.\n");
2348 }
2349
2350 /*
2351  * import RAID devices based on one partition
2352  * if possible, the array gets run as well.
2353  */
2354
2355 static int autostart_array(dev_t startdev)
2356 {
2357         char b[BDEVNAME_SIZE];
2358         int err = -EINVAL, i;
2359         mdp_super_t *sb = NULL;
2360         mdk_rdev_t *start_rdev = NULL, *rdev;
2361
2362         start_rdev = md_import_device(startdev, 0, 0);
2363         if (IS_ERR(start_rdev))
2364                 return err;
2365
2366
2367         /* NOTE: this can only work for 0.90.0 superblocks */
2368         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2369         if (sb->major_version != 0 ||
2370             sb->minor_version != 90 ) {
2371                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2372                 export_rdev(start_rdev);
2373                 return err;
2374         }
2375
2376         if (test_bit(Faulty, &start_rdev->flags)) {
2377                 printk(KERN_WARNING
2378                         "md: can not autostart based on faulty %s!\n",
2379                         bdevname(start_rdev->bdev,b));
2380                 export_rdev(start_rdev);
2381                 return err;
2382         }
2383         list_add(&start_rdev->same_set, &pending_raid_disks);
2384
2385         for (i = 0; i < MD_SB_DISKS; i++) {
2386                 mdp_disk_t *desc = sb->disks + i;
2387                 dev_t dev = MKDEV(desc->major, desc->minor);
2388
2389                 if (!dev)
2390                         continue;
2391                 if (dev == startdev)
2392                         continue;
2393                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2394                         continue;
2395                 rdev = md_import_device(dev, 0, 0);
2396                 if (IS_ERR(rdev))
2397                         continue;
2398
2399                 list_add(&rdev->same_set, &pending_raid_disks);
2400         }
2401
2402         /*
2403          * possibly return codes
2404          */
2405         autorun_devices(0);
2406         return 0;
2407
2408 }
2409
2410
2411 static int get_version(void __user * arg)
2412 {
2413         mdu_version_t ver;
2414
2415         ver.major = MD_MAJOR_VERSION;
2416         ver.minor = MD_MINOR_VERSION;
2417         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2418
2419         if (copy_to_user(arg, &ver, sizeof(ver)))
2420                 return -EFAULT;
2421
2422         return 0;
2423 }
2424
2425 static int get_array_info(mddev_t * mddev, void __user * arg)
2426 {
2427         mdu_array_info_t info;
2428         int nr,working,active,failed,spare;
2429         mdk_rdev_t *rdev;
2430         struct list_head *tmp;
2431
2432         nr=working=active=failed=spare=0;
2433         ITERATE_RDEV(mddev,rdev,tmp) {
2434                 nr++;
2435                 if (test_bit(Faulty, &rdev->flags))
2436                         failed++;
2437                 else {
2438                         working++;
2439                         if (test_bit(In_sync, &rdev->flags))
2440                                 active++;
2441                         else
2442                                 spare++;
2443                 }
2444         }
2445
2446         info.major_version = mddev->major_version;
2447         info.minor_version = mddev->minor_version;
2448         info.patch_version = MD_PATCHLEVEL_VERSION;
2449         info.ctime         = mddev->ctime;
2450         info.level         = mddev->level;
2451         info.size          = mddev->size;
2452         info.nr_disks      = nr;
2453         info.raid_disks    = mddev->raid_disks;
2454         info.md_minor      = mddev->md_minor;
2455         info.not_persistent= !mddev->persistent;
2456
2457         info.utime         = mddev->utime;
2458         info.state         = 0;
2459         if (mddev->in_sync)
2460                 info.state = (1<<MD_SB_CLEAN);
2461         if (mddev->bitmap && mddev->bitmap_offset)
2462                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2463         info.active_disks  = active;
2464         info.working_disks = working;
2465         info.failed_disks  = failed;
2466         info.spare_disks   = spare;
2467
2468         info.layout        = mddev->layout;
2469         info.chunk_size    = mddev->chunk_size;
2470
2471         if (copy_to_user(arg, &info, sizeof(info)))
2472                 return -EFAULT;
2473
2474         return 0;
2475 }
2476
2477 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2478 {
2479         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2480         char *ptr, *buf = NULL;
2481         int err = -ENOMEM;
2482
2483         file = kmalloc(sizeof(*file), GFP_KERNEL);
2484         if (!file)
2485                 goto out;
2486
2487         /* bitmap disabled, zero the first byte and copy out */
2488         if (!mddev->bitmap || !mddev->bitmap->file) {
2489                 file->pathname[0] = '\0';
2490                 goto copy_out;
2491         }
2492
2493         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2494         if (!buf)
2495                 goto out;
2496
2497         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2498         if (!ptr)
2499                 goto out;
2500
2501         strcpy(file->pathname, ptr);
2502
2503 copy_out:
2504         err = 0;
2505         if (copy_to_user(arg, file, sizeof(*file)))
2506                 err = -EFAULT;
2507 out:
2508         kfree(buf);
2509         kfree(file);
2510         return err;
2511 }
2512
2513 static int get_disk_info(mddev_t * mddev, void __user * arg)
2514 {
2515         mdu_disk_info_t info;
2516         unsigned int nr;
2517         mdk_rdev_t *rdev;
2518
2519         if (copy_from_user(&info, arg, sizeof(info)))
2520                 return -EFAULT;
2521
2522         nr = info.number;
2523
2524         rdev = find_rdev_nr(mddev, nr);
2525         if (rdev) {
2526                 info.major = MAJOR(rdev->bdev->bd_dev);
2527                 info.minor = MINOR(rdev->bdev->bd_dev);
2528                 info.raid_disk = rdev->raid_disk;
2529                 info.state = 0;
2530                 if (test_bit(Faulty, &rdev->flags))
2531                         info.state |= (1<<MD_DISK_FAULTY);
2532                 else if (test_bit(In_sync, &rdev->flags)) {
2533                         info.state |= (1<<MD_DISK_ACTIVE);
2534                         info.state |= (1<<MD_DISK_SYNC);
2535                 }
2536                 if (test_bit(WriteMostly, &rdev->flags))
2537                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2538         } else {
2539                 info.major = info.minor = 0;
2540                 info.raid_disk = -1;
2541                 info.state = (1<<MD_DISK_REMOVED);
2542         }
2543
2544         if (copy_to_user(arg, &info, sizeof(info)))
2545                 return -EFAULT;
2546
2547         return 0;
2548 }
2549
2550 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2551 {
2552         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2553         mdk_rdev_t *rdev;
2554         dev_t dev = MKDEV(info->major,info->minor);
2555
2556         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2557                 return -EOVERFLOW;
2558
2559         if (!mddev->raid_disks) {
2560                 int err;
2561                 /* expecting a device which has a superblock */
2562                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2563                 if (IS_ERR(rdev)) {
2564                         printk(KERN_WARNING
2565                                 "md: md_import_device returned %ld\n",
2566                                 PTR_ERR(rdev));
2567                         return PTR_ERR(rdev);
2568                 }
2569                 if (!list_empty(&mddev->disks)) {
2570                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2571                                                         mdk_rdev_t, same_set);
2572                         int err = super_types[mddev->major_version]
2573                                 .load_super(rdev, rdev0, mddev->minor_version);
2574                         if (err < 0) {
2575                                 printk(KERN_WARNING
2576                                         "md: %s has different UUID to %s\n",
2577                                         bdevname(rdev->bdev,b),
2578                                         bdevname(rdev0->bdev,b2));
2579                                 export_rdev(rdev);
2580                                 return -EINVAL;
2581                         }
2582                 }
2583                 err = bind_rdev_to_array(rdev, mddev);
2584                 if (err)
2585                         export_rdev(rdev);
2586                 return err;
2587         }
2588
2589         /*
2590          * add_new_disk can be used once the array is assembled
2591          * to add "hot spares".  They must already have a superblock
2592          * written
2593          */
2594         if (mddev->pers) {
2595                 int err;
2596                 if (!mddev->pers->hot_add_disk) {
2597                         printk(KERN_WARNING
2598                                 "%s: personality does not support diskops!\n",
2599                                mdname(mddev));
2600                         return -EINVAL;
2601                 }
2602                 if (mddev->persistent)
2603                         rdev = md_import_device(dev, mddev->major_version,
2604                                                 mddev->minor_version);
2605                 else
2606                         rdev = md_import_device(dev, -1, -1);
2607                 if (IS_ERR(rdev)) {
2608                         printk(KERN_WARNING
2609                                 "md: md_import_device returned %ld\n",
2610                                 PTR_ERR(rdev));
2611                         return PTR_ERR(rdev);
2612                 }
2613                 /* set save_raid_disk if appropriate */
2614                 if (!mddev->persistent) {
2615                         if (info->state & (1<<MD_DISK_SYNC)  &&
2616                             info->raid_disk < mddev->raid_disks)
2617                                 rdev->raid_disk = info->raid_disk;
2618                         else
2619                                 rdev->raid_disk = -1;
2620                 } else
2621                         super_types[mddev->major_version].
2622                                 validate_super(mddev, rdev);
2623                 rdev->saved_raid_disk = rdev->raid_disk;
2624
2625                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2626                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2627                         set_bit(WriteMostly, &rdev->flags);
2628
2629                 rdev->raid_disk = -1;
2630                 err = bind_rdev_to_array(rdev, mddev);
2631                 if (err)
2632                         export_rdev(rdev);
2633
2634                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2635                 md_wakeup_thread(mddev->thread);
2636                 return err;
2637         }
2638
2639         /* otherwise, add_new_disk is only allowed
2640          * for major_version==0 superblocks
2641          */
2642         if (mddev->major_version != 0) {
2643                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2644                        mdname(mddev));
2645                 return -EINVAL;
2646         }
2647
2648         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2649                 int err;
2650                 rdev = md_import_device (dev, -1, 0);
2651                 if (IS_ERR(rdev)) {
2652                         printk(KERN_WARNING
2653                                 "md: error, md_import_device() returned %ld\n",
2654                                 PTR_ERR(rdev));
2655                         return PTR_ERR(rdev);
2656                 }
2657                 rdev->desc_nr = info->number;
2658                 if (info->raid_disk < mddev->raid_disks)
2659                         rdev->raid_disk = info->raid_disk;
2660                 else
2661                         rdev->raid_disk = -1;
2662
2663                 rdev->flags = 0;
2664
2665                 if (rdev->raid_disk < mddev->raid_disks)
2666                         if (info->state & (1<<MD_DISK_SYNC))
2667                                 set_bit(In_sync, &rdev->flags);
2668
2669                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2670                         set_bit(WriteMostly, &rdev->flags);
2671
2672                 err = bind_rdev_to_array(rdev, mddev);
2673                 if (err) {
2674                         export_rdev(rdev);
2675                         return err;
2676                 }
2677
2678                 if (!mddev->persistent) {
2679                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2680                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2681                 } else
2682                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2683                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2684
2685                 if (!mddev->size || (mddev->size > rdev->size))
2686                         mddev->size = rdev->size;
2687         }
2688
2689         return 0;
2690 }
2691
2692 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2693 {
2694         char b[BDEVNAME_SIZE];
2695         mdk_rdev_t *rdev;
2696
2697         if (!mddev->pers)
2698                 return -ENODEV;
2699
2700         rdev = find_rdev(mddev, dev);
2701         if (!rdev)
2702                 return -ENXIO;
2703
2704         if (rdev->raid_disk >= 0)
2705                 goto busy;
2706
2707         kick_rdev_from_array(rdev);
2708         md_update_sb(mddev);
2709
2710         return 0;
2711 busy:
2712         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2713                 bdevname(rdev->bdev,b), mdname(mddev));
2714         return -EBUSY;
2715 }
2716
2717 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2718 {
2719         char b[BDEVNAME_SIZE];
2720         int err;
2721         unsigned int size;
2722         mdk_rdev_t *rdev;
2723
2724         if (!mddev->pers)
2725                 return -ENODEV;
2726
2727         if (mddev->major_version != 0) {
2728                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2729                         " version-0 superblocks.\n",
2730                         mdname(mddev));
2731                 return -EINVAL;
2732         }
2733         if (!mddev->pers->hot_add_disk) {
2734                 printk(KERN_WARNING
2735                         "%s: personality does not support diskops!\n",
2736                         mdname(mddev));
2737                 return -EINVAL;
2738         }
2739
2740         rdev = md_import_device (dev, -1, 0);
2741         if (IS_ERR(rdev)) {
2742                 printk(KERN_WARNING
2743                         "md: error, md_import_device() returned %ld\n",
2744                         PTR_ERR(rdev));
2745                 return -EINVAL;
2746         }
2747
2748         if (mddev->persistent)
2749                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2750         else
2751                 rdev->sb_offset =
2752                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2753
2754         size = calc_dev_size(rdev, mddev->chunk_size);
2755         rdev->size = size;
2756
2757         if (size < mddev->size) {
2758                 printk(KERN_WARNING
2759                         "%s: disk size %llu blocks < array size %llu\n",
2760                         mdname(mddev), (unsigned long long)size,
2761                         (unsigned long long)mddev->size);
2762                 err = -ENOSPC;
2763                 goto abort_export;
2764         }
2765
2766         if (test_bit(Faulty, &rdev->flags)) {
2767                 printk(KERN_WARNING
2768                         "md: can not hot-add faulty %s disk to %s!\n",
2769                         bdevname(rdev->bdev,b), mdname(mddev));
2770                 err = -EINVAL;
2771                 goto abort_export;
2772         }
2773         clear_bit(In_sync, &rdev->flags);
2774         rdev->desc_nr = -1;
2775         bind_rdev_to_array(rdev, mddev);
2776
2777         /*
2778          * The rest should better be atomic, we can have disk failures
2779          * noticed in interrupt contexts ...
2780          */
2781
2782         if (rdev->desc_nr == mddev->max_disks) {
2783                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2784                         mdname(mddev));
2785                 err = -EBUSY;
2786                 goto abort_unbind_export;
2787         }
2788
2789         rdev->raid_disk = -1;
2790
2791         md_update_sb(mddev);
2792
2793         /*
2794          * Kick recovery, maybe this spare has to be added to the
2795          * array immediately.
2796          */
2797         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2798         md_wakeup_thread(mddev->thread);
2799
2800         return 0;
2801
2802 abort_unbind_export:
2803         unbind_rdev_from_array(rdev);
2804
2805 abort_export:
2806         export_rdev(rdev);
2807         return err;
2808 }
2809
2810 /* similar to deny_write_access, but accounts for our holding a reference
2811  * to the file ourselves */
2812 static int deny_bitmap_write_access(struct file * file)
2813 {
2814         struct inode *inode = file->f_mapping->host;
2815
2816         spin_lock(&inode->i_lock);
2817         if (atomic_read(&inode->i_writecount) > 1) {
2818                 spin_unlock(&inode->i_lock);
2819                 return -ETXTBSY;
2820         }
2821         atomic_set(&inode->i_writecount, -1);
2822         spin_unlock(&inode->i_lock);
2823
2824         return 0;
2825 }
2826
2827 static int set_bitmap_file(mddev_t *mddev, int fd)
2828 {
2829         int err;
2830
2831         if (mddev->pers) {
2832                 if (!mddev->pers->quiesce)
2833                         return -EBUSY;
2834                 if (mddev->recovery || mddev->sync_thread)
2835                         return -EBUSY;
2836                 /* we should be able to change the bitmap.. */
2837         }
2838
2839
2840         if (fd >= 0) {
2841                 if (mddev->bitmap)
2842                         return -EEXIST; /* cannot add when bitmap is present */
2843                 mddev->bitmap_file = fget(fd);
2844
2845                 if (mddev->bitmap_file == NULL) {
2846                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2847                                mdname(mddev));
2848                         return -EBADF;
2849                 }
2850
2851                 err = deny_bitmap_write_access(mddev->bitmap_file);
2852                 if (err) {
2853                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2854                                mdname(mddev));
2855                         fput(mddev->bitmap_file);
2856                         mddev->bitmap_file = NULL;
2857                         return err;
2858                 }
2859                 mddev->bitmap_offset = 0; /* file overrides offset */
2860         } else if (mddev->bitmap == NULL)
2861                 return -ENOENT; /* cannot remove what isn't there */
2862         err = 0;
2863         if (mddev->pers) {
2864                 mddev->pers->quiesce(mddev, 1);
2865                 if (fd >= 0)
2866                         err = bitmap_create(mddev);
2867                 if (fd < 0 || err)
2868                         bitmap_destroy(mddev);
2869                 mddev->pers->quiesce(mddev, 0);
2870         } else if (fd < 0) {
2871                 if (mddev->bitmap_file)
2872                         fput(mddev->bitmap_file);
2873                 mddev->bitmap_file = NULL;
2874         }
2875
2876         return err;
2877 }
2878
2879 /*
2880  * set_array_info is used two different ways
2881  * The original usage is when creating a new array.
2882  * In this usage, raid_disks is > 0 and it together with
2883  *  level, size, not_persistent,layout,chunksize determine the
2884  *  shape of the array.
2885  *  This will always create an array with a type-0.90.0 superblock.
2886  * The newer usage is when assembling an array.
2887  *  In this case raid_disks will be 0, and the major_version field is
2888  *  use to determine which style super-blocks are to be found on the devices.
2889  *  The minor and patch _version numbers are also kept incase the
2890  *  super_block handler wishes to interpret them.
2891  */
2892 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2893 {
2894
2895         if (info->raid_disks == 0) {
2896                 /* just setting version number for superblock loading */
2897                 if (info->major_version < 0 ||
2898                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2899                     super_types[info->major_version].name == NULL) {
2900                         /* maybe try to auto-load a module? */
2901                         printk(KERN_INFO
2902                                 "md: superblock version %d not known\n",
2903                                 info->major_version);
2904                         return -EINVAL;
2905                 }
2906                 mddev->major_version = info->major_version;
2907                 mddev->minor_version = info->minor_version;
2908                 mddev->patch_version = info->patch_version;
2909                 return 0;
2910         }
2911         mddev->major_version = MD_MAJOR_VERSION;
2912         mddev->minor_version = MD_MINOR_VERSION;
2913         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2914         mddev->ctime         = get_seconds();
2915
2916         mddev->level         = info->level;
2917         mddev->size          = info->size;
2918         mddev->raid_disks    = info->raid_disks;
2919         /* don't set md_minor, it is determined by which /dev/md* was
2920          * openned
2921          */
2922         if (info->state & (1<<MD_SB_CLEAN))
2923                 mddev->recovery_cp = MaxSector;
2924         else
2925                 mddev->recovery_cp = 0;
2926         mddev->persistent    = ! info->not_persistent;
2927
2928         mddev->layout        = info->layout;
2929         mddev->chunk_size    = info->chunk_size;
2930
2931         mddev->max_disks     = MD_SB_DISKS;
2932
2933         mddev->sb_dirty      = 1;
2934
2935         /*
2936          * Generate a 128 bit UUID
2937          */
2938         get_random_bytes(mddev->uuid, 16);
2939
2940         return 0;
2941 }
2942
2943 /*
2944  * update_array_info is used to change the configuration of an
2945  * on-line array.
2946  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2947  * fields in the info are checked against the array.
2948  * Any differences that cannot be handled will cause an error.
2949  * Normally, only one change can be managed at a time.
2950  */
2951 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2952 {
2953         int rv = 0;
2954         int cnt = 0;
2955         int state = 0;
2956
2957         /* calculate expected state,ignoring low bits */
2958         if (mddev->bitmap && mddev->bitmap_offset)
2959                 state |= (1 << MD_SB_BITMAP_PRESENT);
2960
2961         if (mddev->major_version != info->major_version ||
2962             mddev->minor_version != info->minor_version ||
2963 /*          mddev->patch_version != info->patch_version || */
2964             mddev->ctime         != info->ctime         ||
2965             mddev->level         != info->level         ||
2966 /*          mddev->layout        != info->layout        || */
2967             !mddev->persistent   != info->not_persistent||
2968             mddev->chunk_size    != info->chunk_size    ||
2969             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2970             ((state^info->state) & 0xfffffe00)
2971                 )
2972                 return -EINVAL;
2973         /* Check there is only one change */
2974         if (mddev->size != info->size) cnt++;
2975         if (mddev->raid_disks != info->raid_disks) cnt++;
2976         if (mddev->layout != info->layout) cnt++;
2977         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2978         if (cnt == 0) return 0;
2979         if (cnt > 1) return -EINVAL;
2980
2981         if (mddev->layout != info->layout) {
2982                 /* Change layout
2983                  * we don't need to do anything at the md level, the
2984                  * personality will take care of it all.
2985                  */
2986                 if (mddev->pers->reconfig == NULL)
2987                         return -EINVAL;
2988                 else
2989                         return mddev->pers->reconfig(mddev, info->layout, -1);
2990         }
2991         if (mddev->size != info->size) {
2992                 mdk_rdev_t * rdev;
2993                 struct list_head *tmp;
2994                 if (mddev->pers->resize == NULL)
2995                         return -EINVAL;
2996                 /* The "size" is the amount of each device that is used.
2997                  * This can only make sense for arrays with redundancy.
2998                  * linear and raid0 always use whatever space is available
2999                  * We can only consider changing the size if no resync
3000                  * or reconstruction is happening, and if the new size
3001                  * is acceptable. It must fit before the sb_offset or,
3002                  * if that is <data_offset, it must fit before the
3003                  * size of each device.
3004                  * If size is zero, we find the largest size that fits.
3005                  */
3006                 if (mddev->sync_thread)
3007                         return -EBUSY;
3008                 ITERATE_RDEV(mddev,rdev,tmp) {
3009                         sector_t avail;
3010                         int fit = (info->size == 0);
3011                         if (rdev->sb_offset > rdev->data_offset)
3012                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
3013                         else
3014                                 avail = get_capacity(rdev->bdev->bd_disk)
3015                                         - rdev->data_offset;
3016                         if (fit && (info->size == 0 || info->size > avail/2))
3017                                 info->size = avail/2;
3018                         if (avail < ((sector_t)info->size << 1))
3019                                 return -ENOSPC;
3020                 }
3021                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3022                 if (!rv) {
3023                         struct block_device *bdev;
3024
3025                         bdev = bdget_disk(mddev->gendisk, 0);
3026                         if (bdev) {
3027                                 down(&bdev->bd_inode->i_sem);
3028                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3029                                 up(&bdev->bd_inode->i_sem);
3030                                 bdput(bdev);
3031                         }
3032                 }
3033         }
3034         if (mddev->raid_disks    != info->raid_disks) {
3035                 /* change the number of raid disks */
3036                 if (mddev->pers->reshape == NULL)
3037                         return -EINVAL;
3038                 if (info->raid_disks <= 0 ||
3039                     info->raid_disks >= mddev->max_disks)
3040                         return -EINVAL;
3041                 if (mddev->sync_thread)
3042                         return -EBUSY;
3043                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3044                 if (!rv) {
3045                         struct block_device *bdev;
3046
3047                         bdev = bdget_disk(mddev->gendisk, 0);
3048                         if (bdev) {
3049                                 down(&bdev->bd_inode->i_sem);
3050                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3051                                 up(&bdev->bd_inode->i_sem);
3052                                 bdput(bdev);
3053                         }
3054                 }
3055         }
3056         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3057                 if (mddev->pers->quiesce == NULL)
3058                         return -EINVAL;
3059                 if (mddev->recovery || mddev->sync_thread)
3060                         return -EBUSY;
3061                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3062                         /* add the bitmap */
3063                         if (mddev->bitmap)
3064                                 return -EEXIST;
3065                         if (mddev->default_bitmap_offset == 0)
3066                                 return -EINVAL;
3067                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3068                         mddev->pers->quiesce(mddev, 1);
3069                         rv = bitmap_create(mddev);
3070                         if (rv)
3071                                 bitmap_destroy(mddev);
3072                         mddev->pers->quiesce(mddev, 0);
3073                 } else {
3074                         /* remove the bitmap */
3075                         if (!mddev->bitmap)
3076                                 return -ENOENT;
3077                         if (mddev->bitmap->file)
3078                                 return -EINVAL;
3079                         mddev->pers->quiesce(mddev, 1);
3080                         bitmap_destroy(mddev);
3081                         mddev->pers->quiesce(mddev, 0);
3082                         mddev->bitmap_offset = 0;
3083                 }
3084         }
3085         md_update_sb(mddev);
3086         return rv;
3087 }
3088
3089 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3090 {
3091         mdk_rdev_t *rdev;
3092
3093         if (mddev->pers == NULL)
3094                 return -ENODEV;
3095
3096         rdev = find_rdev(mddev, dev);
3097         if (!rdev)
3098                 return -ENODEV;
3099
3100         md_error(mddev, rdev);
3101         return 0;
3102 }
3103
3104 static int md_ioctl(struct inode *inode, struct file *file,
3105                         unsigned int cmd, unsigned long arg)
3106 {
3107         int err = 0;
3108         void __user *argp = (void __user *)arg;
3109         struct hd_geometry __user *loc = argp;
3110         mddev_t *mddev = NULL;
3111
3112         if (!capable(CAP_SYS_ADMIN))
3113                 return -EACCES;
3114
3115         /*
3116          * Commands dealing with the RAID driver but not any
3117          * particular array:
3118          */
3119         switch (cmd)
3120         {
3121                 case RAID_VERSION:
3122                         err = get_version(argp);
3123                         goto done;
3124
3125                 case PRINT_RAID_DEBUG:
3126                         err = 0;
3127                         md_print_devices();
3128                         goto done;
3129
3130 #ifndef MODULE
3131                 case RAID_AUTORUN:
3132                         err = 0;
3133                         autostart_arrays(arg);
3134                         goto done;
3135 #endif
3136                 default:;
3137         }
3138
3139         /*
3140          * Commands creating/starting a new array:
3141          */
3142
3143         mddev = inode->i_bdev->bd_disk->private_data;
3144
3145         if (!mddev) {
3146                 BUG();
3147                 goto abort;
3148         }
3149
3150
3151         if (cmd == START_ARRAY) {
3152                 /* START_ARRAY doesn't need to lock the array as autostart_array
3153                  * does the locking, and it could even be a different array
3154                  */
3155                 static int cnt = 3;
3156                 if (cnt > 0 ) {
3157                         printk(KERN_WARNING
3158                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3159                                "This will not be supported beyond July 2006\n",
3160                                current->comm, current->pid);
3161                         cnt--;
3162                 }
3163                 err = autostart_array(new_decode_dev(arg));
3164                 if (err) {
3165                         printk(KERN_WARNING "md: autostart failed!\n");
3166                         goto abort;
3167                 }
3168                 goto done;
3169         }
3170
3171         err = mddev_lock(mddev);
3172         if (err) {
3173                 printk(KERN_INFO
3174                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3175                         err, cmd);
3176                 goto abort;
3177         }
3178
3179         switch (cmd)
3180         {
3181                 case SET_ARRAY_INFO:
3182                         {
3183                                 mdu_array_info_t info;
3184                                 if (!arg)
3185                                         memset(&info, 0, sizeof(info));
3186                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3187                                         err = -EFAULT;
3188                                         goto abort_unlock;
3189                                 }
3190                                 if (mddev->pers) {
3191                                         err = update_array_info(mddev, &info);
3192                                         if (err) {
3193                                                 printk(KERN_WARNING "md: couldn't update"
3194                                                        " array info. %d\n", err);
3195                                                 goto abort_unlock;
3196                                         }
3197                                         goto done_unlock;
3198                                 }
3199                                 if (!list_empty(&mddev->disks)) {
3200                                         printk(KERN_WARNING
3201                                                "md: array %s already has disks!\n",
3202                                                mdname(mddev));
3203                                         err = -EBUSY;
3204                                         goto abort_unlock;
3205                                 }
3206                                 if (mddev->raid_disks) {
3207                                         printk(KERN_WARNING
3208                                                "md: array %s already initialised!\n",
3209                                                mdname(mddev));
3210                                         err = -EBUSY;
3211                                         goto abort_unlock;
3212                                 }
3213                                 err = set_array_info(mddev, &info);
3214                                 if (err) {
3215                                         printk(KERN_WARNING "md: couldn't set"
3216                                                " array info. %d\n", err);
3217                                         goto abort_unlock;
3218                                 }
3219                         }
3220                         goto done_unlock;
3221
3222                 default:;
3223         }
3224
3225         /*
3226          * Commands querying/configuring an existing array:
3227          */
3228         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3229          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3230         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3231                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3232                 err = -ENODEV;
3233                 goto abort_unlock;
3234         }
3235
3236         /*
3237          * Commands even a read-only array can execute:
3238          */
3239         switch (cmd)
3240         {
3241                 case GET_ARRAY_INFO:
3242                         err = get_array_info(mddev, argp);
3243                         goto done_unlock;
3244
3245                 case GET_BITMAP_FILE:
3246                         err = get_bitmap_file(mddev, argp);
3247                         goto done_unlock;
3248
3249                 case GET_DISK_INFO:
3250                         err = get_disk_info(mddev, argp);
3251                         goto done_unlock;
3252
3253                 case RESTART_ARRAY_RW:
3254                         err = restart_array(mddev);
3255                         goto done_unlock;
3256
3257                 case STOP_ARRAY:
3258                         err = do_md_stop (mddev, 0);
3259                         goto done_unlock;
3260
3261                 case STOP_ARRAY_RO:
3262                         err = do_md_stop (mddev, 1);
3263                         goto done_unlock;
3264
3265         /*
3266          * We have a problem here : there is no easy way to give a CHS
3267          * virtual geometry. We currently pretend that we have a 2 heads
3268          * 4 sectors (with a BIG number of cylinders...). This drives
3269          * dosfs just mad... ;-)
3270          */
3271                 case HDIO_GETGEO:
3272                         if (!loc) {
3273                                 err = -EINVAL;
3274                                 goto abort_unlock;
3275                         }
3276                         err = put_user (2, (char __user *) &loc->heads);
3277                         if (err)
3278                                 goto abort_unlock;
3279                         err = put_user (4, (char __user *) &loc->sectors);
3280                         if (err)
3281                                 goto abort_unlock;
3282                         err = put_user(get_capacity(mddev->gendisk)/8,
3283                                         (short __user *) &loc->cylinders);
3284                         if (err)
3285                                 goto abort_unlock;
3286                         err = put_user (get_start_sect(inode->i_bdev),
3287                                                 (long __user *) &loc->start);
3288                         goto done_unlock;
3289         }
3290
3291         /*
3292          * The remaining ioctls are changing the state of the
3293          * superblock, so we do not allow them on read-only arrays.
3294          * However non-MD ioctls (e.g. get-size) will still come through
3295          * here and hit the 'default' below, so only disallow
3296          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3297          */
3298         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3299             mddev->ro && mddev->pers) {
3300                 if (mddev->ro == 2) {
3301                         mddev->ro = 0;
3302                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3303                 md_wakeup_thread(mddev->thread);
3304
3305                 } else {
3306                         err = -EROFS;
3307                         goto abort_unlock;
3308                 }
3309         }
3310
3311         switch (cmd)
3312         {
3313                 case ADD_NEW_DISK:
3314                 {
3315                         mdu_disk_info_t info;
3316                         if (copy_from_user(&info, argp, sizeof(info)))
3317                                 err = -EFAULT;
3318                         else
3319                                 err = add_new_disk(mddev, &info);
3320                         goto done_unlock;
3321                 }
3322
3323                 case HOT_REMOVE_DISK:
3324                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3325                         goto done_unlock;
3326
3327                 case HOT_ADD_DISK:
3328                         err = hot_add_disk(mddev, new_decode_dev(arg));
3329                         goto done_unlock;
3330
3331                 case SET_DISK_FAULTY:
3332                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3333                         goto done_unlock;
3334
3335                 case RUN_ARRAY:
3336                         err = do_md_run (mddev);
3337                         goto done_unlock;
3338
3339                 case SET_BITMAP_FILE:
3340                         err = set_bitmap_file(mddev, (int)arg);
3341                         goto done_unlock;
3342
3343                 default:
3344                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3345                                 printk(KERN_WARNING "md: %s(pid %d) used"
3346                                         " obsolete MD ioctl, upgrade your"
3347                                         " software to use new ictls.\n",
3348                                         current->comm, current->pid);
3349                         err = -EINVAL;
3350                         goto abort_unlock;
3351         }
3352
3353 done_unlock:
3354 abort_unlock:
3355         mddev_unlock(mddev);
3356
3357         return err;
3358 done:
3359         if (err)
3360                 MD_BUG();
3361 abort:
3362         return err;
3363 }
3364
3365 static int md_open(struct inode *inode, struct file *file)
3366 {
3367         /*
3368          * Succeed if we can lock the mddev, which confirms that
3369          * it isn't being stopped right now.
3370          */
3371         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3372         int err;
3373
3374         if ((err = mddev_lock(mddev)))
3375                 goto out;
3376
3377         err = 0;
3378         mddev_get(mddev);
3379         mddev_unlock(mddev);
3380
3381         check_disk_change(inode->i_bdev);
3382  out:
3383         return err;
3384 }
3385
3386 static int md_release(struct inode *inode, struct file * file)
3387 {
3388         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3389
3390         if (!mddev)
3391                 BUG();
3392         mddev_put(mddev);
3393
3394         return 0;
3395 }
3396
3397 static int md_media_changed(struct gendisk *disk)
3398 {
3399         mddev_t *mddev = disk->private_data;
3400
3401         return mddev->changed;
3402 }
3403
3404 static int md_revalidate(struct gendisk *disk)
3405 {
3406         mddev_t *mddev = disk->private_data;
3407
3408         mddev->changed = 0;
3409         return 0;
3410 }
3411 static struct block_device_operations md_fops =
3412 {
3413         .owner          = THIS_MODULE,
3414         .open           = md_open,
3415         .release        = md_release,
3416         .ioctl          = md_ioctl,
3417         .media_changed  = md_media_changed,
3418         .revalidate_disk= md_revalidate,
3419 };
3420
3421 static int md_thread(void * arg)
3422 {
3423         mdk_thread_t *thread = arg;
3424
3425         /*
3426          * md_thread is a 'system-thread', it's priority should be very
3427          * high. We avoid resource deadlocks individually in each
3428          * raid personality. (RAID5 does preallocation) We also use RR and
3429          * the very same RT priority as kswapd, thus we will never get
3430          * into a priority inversion deadlock.
3431          *
3432          * we definitely have to have equal or higher priority than
3433          * bdflush, otherwise bdflush will deadlock if there are too
3434          * many dirty RAID5 blocks.
3435          */
3436
3437         allow_signal(SIGKILL);
3438         while (!kthread_should_stop()) {
3439
3440                 /* We need to wait INTERRUPTIBLE so that
3441                  * we don't add to the load-average.
3442                  * That means we need to be sure no signals are
3443                  * pending
3444                  */
3445                 if (signal_pending(current))
3446                         flush_signals(current);
3447
3448                 wait_event_interruptible_timeout
3449                         (thread->wqueue,
3450                          test_bit(THREAD_WAKEUP, &thread->flags)
3451                          || kthread_should_stop(),
3452                          thread->timeout);
3453                 try_to_freeze();
3454
3455                 clear_bit(THREAD_WAKEUP, &thread->flags);
3456
3457                 thread->run(thread->mddev);
3458         }
3459
3460         return 0;
3461 }
3462
3463 void md_wakeup_thread(mdk_thread_t *thread)
3464 {
3465         if (thread) {
3466                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3467                 set_bit(THREAD_WAKEUP, &thread->flags);
3468                 wake_up(&thread->wqueue);
3469         }
3470 }
3471
3472 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3473                                  const char *name)
3474 {
3475         mdk_thread_t *thread;
3476
3477         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3478         if (!thread)
3479                 return NULL;
3480
3481         memset(thread, 0, sizeof(mdk_thread_t));
3482         init_waitqueue_head(&thread->wqueue);
3483
3484         thread->run = run;
3485         thread->mddev = mddev;
3486         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3487         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3488         if (IS_ERR(thread->tsk)) {
3489                 kfree(thread);
3490                 return NULL;
3491         }
3492         return thread;
3493 }
3494
3495 void md_unregister_thread(mdk_thread_t *thread)
3496 {
3497         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3498
3499         kthread_stop(thread->tsk);
3500         kfree(thread);
3501 }
3502
3503 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3504 {
3505         if (!mddev) {
3506                 MD_BUG();
3507                 return;
3508         }
3509
3510         if (!rdev || test_bit(Faulty, &rdev->flags))
3511                 return;
3512 /*
3513         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3514                 mdname(mddev),
3515                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3516                 __builtin_return_address(0),__builtin_return_address(1),
3517                 __builtin_return_address(2),__builtin_return_address(3));
3518 */
3519         if (!mddev->pers->error_handler)
3520                 return;
3521         mddev->pers->error_handler(mddev,rdev);
3522         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3523         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3524         md_wakeup_thread(mddev->thread);
3525 }
3526
3527 /* seq_file implementation /proc/mdstat */
3528
3529 static void status_unused(struct seq_file *seq)
3530 {
3531         int i = 0;
3532         mdk_rdev_t *rdev;
3533         struct list_head *tmp;
3534
3535         seq_printf(seq, "unused devices: ");
3536
3537         ITERATE_RDEV_PENDING(rdev,tmp) {
3538                 char b[BDEVNAME_SIZE];
3539                 i++;
3540                 seq_printf(seq, "%s ",
3541                               bdevname(rdev->bdev,b));
3542         }
3543         if (!i)
3544                 seq_printf(seq, "<none>");
3545
3546         seq_printf(seq, "\n");
3547 }
3548
3549
3550 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3551 {
3552         unsigned long max_blocks, resync, res, dt, db, rt;
3553
3554         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3555
3556         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3557                 max_blocks = mddev->resync_max_sectors >> 1;
3558         else
3559                 max_blocks = mddev->size;
3560
3561         /*
3562          * Should not happen.
3563          */
3564         if (!max_blocks) {
3565                 MD_BUG();
3566                 return;
3567         }
3568         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3569         {
3570                 int i, x = res/50, y = 20-x;
3571                 seq_printf(seq, "[");
3572                 for (i = 0; i < x; i++)
3573                         seq_printf(seq, "=");
3574                 seq_printf(seq, ">");
3575                 for (i = 0; i < y; i++)
3576                         seq_printf(seq, ".");
3577                 seq_printf(seq, "] ");
3578         }
3579         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3580                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3581                        "resync" : "recovery"),
3582                       res/10, res % 10, resync, max_blocks);
3583
3584         /*
3585          * We do not want to overflow, so the order of operands and
3586          * the * 100 / 100 trick are important. We do a +1 to be
3587          * safe against division by zero. We only estimate anyway.
3588          *
3589          * dt: time from mark until now
3590          * db: blocks written from mark until now
3591          * rt: remaining time
3592          */
3593         dt = ((jiffies - mddev->resync_mark) / HZ);
3594         if (!dt) dt++;
3595         db = resync - (mddev->resync_mark_cnt/2);
3596         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3597
3598         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3599
3600         seq_printf(seq, " speed=%ldK/sec", db/dt);
3601 }
3602
3603 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3604 {
3605         struct list_head *tmp;
3606         loff_t l = *pos;
3607         mddev_t *mddev;
3608
3609         if (l >= 0x10000)
3610                 return NULL;
3611         if (!l--)
3612                 /* header */
3613                 return (void*)1;
3614
3615         spin_lock(&all_mddevs_lock);
3616         list_for_each(tmp,&all_mddevs)
3617                 if (!l--) {
3618                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3619                         mddev_get(mddev);
3620                         spin_unlock(&all_mddevs_lock);
3621                         return mddev;
3622                 }
3623         spin_unlock(&all_mddevs_lock);
3624         if (!l--)
3625                 return (void*)2;/* tail */
3626         return NULL;
3627 }
3628
3629 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3630 {
3631         struct list_head *tmp;
3632         mddev_t *next_mddev, *mddev = v;
3633
3634         ++*pos;
3635         if (v == (void*)2)
3636                 return NULL;
3637
3638         spin_lock(&all_mddevs_lock);
3639         if (v == (void*)1)
3640                 tmp = all_mddevs.next;
3641         else
3642                 tmp = mddev->all_mddevs.next;
3643         if (tmp != &all_mddevs)
3644                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3645         else {
3646                 next_mddev = (void*)2;
3647                 *pos = 0x10000;
3648         }
3649         spin_unlock(&all_mddevs_lock);
3650
3651         if (v != (void*)1)
3652                 mddev_put(mddev);
3653         return next_mddev;
3654
3655 }
3656
3657 static void md_seq_stop(struct seq_file *seq, void *v)
3658 {
3659         mddev_t *mddev = v;
3660
3661         if (mddev && v != (void*)1 && v != (void*)2)
3662                 mddev_put(mddev);
3663 }
3664
3665 static int md_seq_show(struct seq_file *seq, void *v)
3666 {
3667         mddev_t *mddev = v;
3668         sector_t size;
3669         struct list_head *tmp2;
3670         mdk_rdev_t *rdev;
3671         int i;
3672         struct bitmap *bitmap;
3673
3674         if (v == (void*)1) {
3675                 seq_printf(seq, "Personalities : ");
3676                 spin_lock(&pers_lock);
3677                 for (i = 0; i < MAX_PERSONALITY; i++)
3678                         if (pers[i])
3679                                 seq_printf(seq, "[%s] ", pers[i]->name);
3680
3681                 spin_unlock(&pers_lock);
3682                 seq_printf(seq, "\n");
3683                 return 0;
3684         }
3685         if (v == (void*)2) {
3686                 status_unused(seq);
3687                 return 0;
3688         }
3689
3690         if (mddev_lock(mddev)!=0)
3691                 return -EINTR;
3692         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3693                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3694                                                 mddev->pers ? "" : "in");
3695                 if (mddev->pers) {
3696                         if (mddev->ro==1)
3697                                 seq_printf(seq, " (read-only)");
3698                         if (mddev->ro==2)
3699                                 seq_printf(seq, "(auto-read-only)");
3700                         seq_printf(seq, " %s", mddev->pers->name);
3701                 }
3702
3703                 size = 0;
3704                 ITERATE_RDEV(mddev,rdev,tmp2) {
3705                         char b[BDEVNAME_SIZE];
3706                         seq_printf(seq, " %s[%d]",
3707                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3708                         if (test_bit(WriteMostly, &rdev->flags))
3709                                 seq_printf(seq, "(W)");
3710                         if (test_bit(Faulty, &rdev->flags)) {
3711                                 seq_printf(seq, "(F)");
3712                                 continue;
3713                         } else if (rdev->raid_disk < 0)
3714                                 seq_printf(seq, "(S)"); /* spare */
3715                         size += rdev->size;
3716                 }
3717
3718                 if (!list_empty(&mddev->disks)) {
3719                         if (mddev->pers)
3720                                 seq_printf(seq, "\n      %llu blocks",
3721                                         (unsigned long long)mddev->array_size);
3722                         else
3723                                 seq_printf(seq, "\n      %llu blocks",
3724                                         (unsigned long long)size);
3725                 }
3726                 if (mddev->persistent) {
3727                         if (mddev->major_version != 0 ||
3728                             mddev->minor_version != 90) {
3729                                 seq_printf(seq," super %d.%d",
3730                                            mddev->major_version,
3731                                            mddev->minor_version);
3732                         }
3733                 } else
3734                         seq_printf(seq, " super non-persistent");
3735
3736                 if (mddev->pers) {
3737                         mddev->pers->status (seq, mddev);
3738                         seq_printf(seq, "\n      ");
3739                         if (mddev->pers->sync_request) {
3740                                 if (mddev->curr_resync > 2) {
3741                                         status_resync (seq, mddev);
3742                                         seq_printf(seq, "\n      ");
3743                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3744                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3745                                 else if (mddev->recovery_cp < MaxSector)
3746                                         seq_printf(seq, "\tresync=PENDING\n      ");
3747                         }
3748                 } else
3749                         seq_printf(seq, "\n       ");
3750
3751                 if ((bitmap = mddev->bitmap)) {
3752                         unsigned long chunk_kb;
3753                         unsigned long flags;
3754                         spin_lock_irqsave(&bitmap->lock, flags);
3755                         chunk_kb = bitmap->chunksize >> 10;
3756                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3757                                 "%lu%s chunk",
3758                                 bitmap->pages - bitmap->missing_pages,
3759                                 bitmap->pages,
3760                                 (bitmap->pages - bitmap->missing_pages)
3761                                         << (PAGE_SHIFT - 10),
3762                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3763                                 chunk_kb ? "KB" : "B");
3764                         if (bitmap->file) {
3765                                 seq_printf(seq, ", file: ");
3766                                 seq_path(seq, bitmap->file->f_vfsmnt,
3767                                          bitmap->file->f_dentry," \t\n");
3768                         }
3769
3770                         seq_printf(seq, "\n");
3771                         spin_unlock_irqrestore(&bitmap->lock, flags);
3772                 }
3773
3774                 seq_printf(seq, "\n");
3775         }
3776         mddev_unlock(mddev);
3777
3778         return 0;
3779 }
3780
3781 static struct seq_operations md_seq_ops = {
3782         .start  = md_seq_start,
3783         .next   = md_seq_next,
3784         .stop   = md_seq_stop,
3785         .show   = md_seq_show,
3786 };
3787
3788 static int md_seq_open(struct inode *inode, struct file *file)
3789 {
3790         int error;
3791
3792         error = seq_open(file, &md_seq_ops);
3793         return error;
3794 }
3795
3796 static struct file_operations md_seq_fops = {
3797         .open           = md_seq_open,
3798         .read           = seq_read,
3799         .llseek         = seq_lseek,
3800         .release        = seq_release,
3801 };
3802
3803 int register_md_personality(int pnum, mdk_personality_t *p)
3804 {
3805         if (pnum >= MAX_PERSONALITY) {
3806                 printk(KERN_ERR
3807                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3808                        p->name, pnum, MAX_PERSONALITY-1);
3809                 return -EINVAL;
3810         }
3811
3812         spin_lock(&pers_lock);
3813         if (pers[pnum]) {
3814                 spin_unlock(&pers_lock);
3815                 return -EBUSY;
3816         }
3817
3818         pers[pnum] = p;
3819         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3820         spin_unlock(&pers_lock);
3821         return 0;
3822 }
3823
3824 int unregister_md_personality(int pnum)
3825 {
3826         if (pnum >= MAX_PERSONALITY)
3827                 return -EINVAL;
3828
3829         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3830         spin_lock(&pers_lock);
3831         pers[pnum] = NULL;
3832         spin_unlock(&pers_lock);
3833         return 0;
3834 }
3835
3836 static int is_mddev_idle(mddev_t *mddev)
3837 {
3838         mdk_rdev_t * rdev;
3839         struct list_head *tmp;
3840         int idle;
3841         unsigned long curr_events;
3842
3843         idle = 1;
3844         ITERATE_RDEV(mddev,rdev,tmp) {
3845                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3846                 curr_events = disk_stat_read(disk, sectors[0]) +
3847                                 disk_stat_read(disk, sectors[1]) -
3848                                 atomic_read(&disk->sync_io);
3849                 /* The difference between curr_events and last_events
3850                  * will be affected by any new non-sync IO (making
3851                  * curr_events bigger) and any difference in the amount of
3852                  * in-flight syncio (making current_events bigger or smaller)
3853                  * The amount in-flight is currently limited to
3854                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3855                  * which is at most 4096 sectors.
3856                  * These numbers are fairly fragile and should be made
3857                  * more robust, probably by enforcing the
3858                  * 'window size' that md_do_sync sort-of uses.
3859                  *
3860                  * Note: the following is an unsigned comparison.
3861                  */
3862                 if ((curr_events - rdev->last_events + 4096) > 8192) {
3863                         rdev->last_events = curr_events;
3864                         idle = 0;
3865                 }
3866         }
3867         return idle;
3868 }
3869
3870 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3871 {
3872         /* another "blocks" (512byte) blocks have been synced */
3873         atomic_sub(blocks, &mddev->recovery_active);
3874         wake_up(&mddev->recovery_wait);
3875         if (!ok) {
3876                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3877                 md_wakeup_thread(mddev->thread);
3878                 // stop recovery, signal do_sync ....
3879         }
3880 }
3881
3882
3883 /* md_write_start(mddev, bi)
3884  * If we need to update some array metadata (e.g. 'active' flag
3885  * in superblock) before writing, schedule a superblock update
3886  * and wait for it to complete.
3887  */
3888 void md_write_start(mddev_t *mddev, struct bio *bi)
3889 {
3890         if (bio_data_dir(bi) != WRITE)
3891                 return;
3892
3893         BUG_ON(mddev->ro == 1);
3894         if (mddev->ro == 2) {
3895                 /* need to switch to read/write */
3896                 mddev->ro = 0;
3897                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3898                 md_wakeup_thread(mddev->thread);
3899         }
3900         atomic_inc(&mddev->writes_pending);
3901         if (mddev->in_sync) {
3902                 spin_lock_irq(&mddev->write_lock);
3903                 if (mddev->in_sync) {
3904                         mddev->in_sync = 0;
3905                         mddev->sb_dirty = 1;
3906                         md_wakeup_thread(mddev->thread);
3907                 }
3908                 spin_unlock_irq(&mddev->write_lock);
3909         }
3910         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3911 }
3912
3913 void md_write_end(mddev_t *mddev)
3914 {
3915         if (atomic_dec_and_test(&mddev->writes_pending)) {
3916                 if (mddev->safemode == 2)
3917                         md_wakeup_thread(mddev->thread);
3918                 else
3919                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3920         }
3921 }
3922
3923 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3924
3925 #define SYNC_MARKS      10
3926 #define SYNC_MARK_STEP  (3*HZ)
3927 static void md_do_sync(mddev_t *mddev)
3928 {
3929         mddev_t *mddev2;
3930         unsigned int currspeed = 0,
3931                  window;
3932         sector_t max_sectors,j, io_sectors;
3933         unsigned long mark[SYNC_MARKS];
3934         sector_t mark_cnt[SYNC_MARKS];
3935         int last_mark,m;
3936         struct list_head *tmp;
3937         sector_t last_check;
3938         int skipped = 0;
3939
3940         /* just incase thread restarts... */
3941         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3942                 return;
3943
3944         /* we overload curr_resync somewhat here.
3945          * 0 == not engaged in resync at all
3946          * 2 == checking that there is no conflict with another sync
3947          * 1 == like 2, but have yielded to allow conflicting resync to
3948          *              commense
3949          * other == active in resync - this many blocks
3950          *
3951          * Before starting a resync we must have set curr_resync to
3952          * 2, and then checked that every "conflicting" array has curr_resync
3953          * less than ours.  When we find one that is the same or higher
3954          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3955          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3956          * This will mean we have to start checking from the beginning again.
3957          *
3958          */
3959
3960         do {
3961                 mddev->curr_resync = 2;
3962
3963         try_again:
3964                 if (kthread_should_stop()) {
3965                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3966                         goto skip;
3967                 }
3968                 ITERATE_MDDEV(mddev2,tmp) {
3969                         if (mddev2 == mddev)
3970                                 continue;
3971                         if (mddev2->curr_resync &&
3972                             match_mddev_units(mddev,mddev2)) {
3973                                 DEFINE_WAIT(wq);
3974                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3975                                         /* arbitrarily yield */
3976                                         mddev->curr_resync = 1;
3977                                         wake_up(&resync_wait);
3978                                 }
3979                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3980                                         /* no need to wait here, we can wait the next
3981                                          * time 'round when curr_resync == 2
3982                                          */
3983                                         continue;
3984                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
3985                                 if (!kthread_should_stop() &&
3986                                     mddev2->curr_resync >= mddev->curr_resync) {
3987                                         printk(KERN_INFO "md: delaying resync of %s"
3988                                                " until %s has finished resync (they"
3989                                                " share one or more physical units)\n",
3990                                                mdname(mddev), mdname(mddev2));
3991                                         mddev_put(mddev2);
3992                                         schedule();
3993                                         finish_wait(&resync_wait, &wq);
3994                                         goto try_again;
3995                                 }
3996                                 finish_wait(&resync_wait, &wq);
3997                         }
3998                 }
3999         } while (mddev->curr_resync < 2);
4000
4001         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4002                 /* resync follows the size requested by the personality,
4003                  * which defaults to physical size, but can be virtual size
4004                  */
4005                 max_sectors = mddev->resync_max_sectors;
4006                 mddev->resync_mismatches = 0;
4007         } else
4008                 /* recovery follows the physical size of devices */
4009                 max_sectors = mddev->size << 1;
4010
4011         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4012         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4013                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4014         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4015                "(but not more than %d KB/sec) for reconstruction.\n",
4016                sysctl_speed_limit_max);
4017
4018         is_mddev_idle(mddev); /* this also initializes IO event counters */
4019         /* we don't use the checkpoint if there's a bitmap */
4020         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4021             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4022                 j = mddev->recovery_cp;
4023         else
4024                 j = 0;
4025         io_sectors = 0;
4026         for (m = 0; m < SYNC_MARKS; m++) {
4027                 mark[m] = jiffies;
4028                 mark_cnt[m] = io_sectors;
4029         }
4030         last_mark = 0;
4031         mddev->resync_mark = mark[last_mark];
4032         mddev->resync_mark_cnt = mark_cnt[last_mark];
4033
4034         /*
4035          * Tune reconstruction:
4036          */
4037         window = 32*(PAGE_SIZE/512);
4038         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4039                 window/2,(unsigned long long) max_sectors/2);
4040
4041         atomic_set(&mddev->recovery_active, 0);
4042         init_waitqueue_head(&mddev->recovery_wait);
4043         last_check = 0;
4044
4045         if (j>2) {
4046                 printk(KERN_INFO
4047                         "md: resuming recovery of %s from checkpoint.\n",
4048                         mdname(mddev));
4049                 mddev->curr_resync = j;
4050         }
4051
4052         while (j < max_sectors) {
4053                 sector_t sectors;
4054
4055                 skipped = 0;
4056                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4057                                             currspeed < sysctl_speed_limit_min);
4058                 if (sectors == 0) {
4059                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4060                         goto out;
4061                 }
4062
4063                 if (!skipped) { /* actual IO requested */
4064                         io_sectors += sectors;
4065                         atomic_add(sectors, &mddev->recovery_active);
4066                 }
4067
4068                 j += sectors;
4069                 if (j>1) mddev->curr_resync = j;
4070
4071
4072                 if (last_check + window > io_sectors || j == max_sectors)
4073                         continue;
4074
4075                 last_check = io_sectors;
4076
4077                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4078                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4079                         break;
4080
4081         repeat:
4082                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4083                         /* step marks */
4084                         int next = (last_mark+1) % SYNC_MARKS;
4085
4086                         mddev->resync_mark = mark[next];
4087                         mddev->resync_mark_cnt = mark_cnt[next];
4088                         mark[next] = jiffies;
4089                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4090                         last_mark = next;
4091                 }
4092
4093
4094                 if (kthread_should_stop()) {
4095                         /*
4096                          * got a signal, exit.
4097                          */
4098                         printk(KERN_INFO
4099                                 "md: md_do_sync() got signal ... exiting\n");
4100                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4101                         goto out;
4102                 }
4103
4104                 /*
4105                  * this loop exits only if either when we are slower than
4106                  * the 'hard' speed limit, or the system was IO-idle for
4107                  * a jiffy.
4108                  * the system might be non-idle CPU-wise, but we only care
4109                  * about not overloading the IO subsystem. (things like an
4110                  * e2fsck being done on the RAID array should execute fast)
4111                  */
4112                 mddev->queue->unplug_fn(mddev->queue);
4113                 cond_resched();
4114
4115                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4116                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4117
4118                 if (currspeed > sysctl_speed_limit_min) {
4119                         if ((currspeed > sysctl_speed_limit_max) ||
4120                                         !is_mddev_idle(mddev)) {
4121                                 msleep(500);
4122                                 goto repeat;
4123                         }
4124                 }
4125         }
4126         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4127         /*
4128          * this also signals 'finished resyncing' to md_stop
4129          */
4130  out:
4131         mddev->queue->unplug_fn(mddev->queue);
4132
4133         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4134
4135         /* tell personality that we are finished */
4136         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4137
4138         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4139             mddev->curr_resync > 2 &&
4140             mddev->curr_resync >= mddev->recovery_cp) {
4141                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4142                         printk(KERN_INFO
4143                                 "md: checkpointing recovery of %s.\n",
4144                                 mdname(mddev));
4145                         mddev->recovery_cp = mddev->curr_resync;
4146                 } else
4147                         mddev->recovery_cp = MaxSector;
4148         }
4149
4150  skip:
4151         mddev->curr_resync = 0;
4152         wake_up(&resync_wait);
4153         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4154         md_wakeup_thread(mddev->thread);
4155 }
4156
4157
4158 /*
4159  * This routine is regularly called by all per-raid-array threads to
4160  * deal with generic issues like resync and super-block update.
4161  * Raid personalities that don't have a thread (linear/raid0) do not
4162  * need this as they never do any recovery or update the superblock.
4163  *
4164  * It does not do any resync itself, but rather "forks" off other threads
4165  * to do that as needed.
4166  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4167  * "->recovery" and create a thread at ->sync_thread.
4168  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4169  * and wakeups up this thread which will reap the thread and finish up.
4170  * This thread also removes any faulty devices (with nr_pending == 0).
4171  *
4172  * The overall approach is:
4173  *  1/ if the superblock needs updating, update it.
4174  *  2/ If a recovery thread is running, don't do anything else.
4175  *  3/ If recovery has finished, clean up, possibly marking spares active.
4176  *  4/ If there are any faulty devices, remove them.
4177  *  5/ If array is degraded, try to add spares devices
4178  *  6/ If array has spares or is not in-sync, start a resync thread.
4179  */
4180 void md_check_recovery(mddev_t *mddev)
4181 {
4182         mdk_rdev_t *rdev;
4183         struct list_head *rtmp;
4184
4185
4186         if (mddev->bitmap)
4187                 bitmap_daemon_work(mddev->bitmap);
4188
4189         if (mddev->ro)
4190                 return;
4191
4192         if (signal_pending(current)) {
4193                 if (mddev->pers->sync_request) {
4194                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4195                                mdname(mddev));
4196                         mddev->safemode = 2;
4197                 }
4198                 flush_signals(current);
4199         }
4200
4201         if ( ! (
4202                 mddev->sb_dirty ||
4203                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4204                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4205                 (mddev->safemode == 1) ||
4206                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4207                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4208                 ))
4209                 return;
4210
4211         if (mddev_trylock(mddev)==0) {
4212                 int spares =0;
4213
4214                 spin_lock_irq(&mddev->write_lock);
4215                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4216                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4217                         mddev->in_sync = 1;
4218                         mddev->sb_dirty = 1;
4219                 }
4220                 if (mddev->safemode == 1)
4221                         mddev->safemode = 0;
4222                 spin_unlock_irq(&mddev->write_lock);
4223
4224                 if (mddev->sb_dirty)
4225                         md_update_sb(mddev);
4226
4227
4228                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4229                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4230                         /* resync/recovery still happening */
4231                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4232                         goto unlock;
4233                 }
4234                 if (mddev->sync_thread) {
4235                         /* resync has finished, collect result */
4236                         md_unregister_thread(mddev->sync_thread);
4237                         mddev->sync_thread = NULL;
4238                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4239                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4240                                 /* success...*/
4241                                 /* activate any spares */
4242                                 mddev->pers->spare_active(mddev);
4243                         }
4244                         md_update_sb(mddev);
4245
4246                         /* if array is no-longer degraded, then any saved_raid_disk
4247                          * information must be scrapped
4248                          */
4249                         if (!mddev->degraded)
4250                                 ITERATE_RDEV(mddev,rdev,rtmp)
4251                                         rdev->saved_raid_disk = -1;
4252
4253                         mddev->recovery = 0;
4254                         /* flag recovery needed just to double check */
4255                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4256                         goto unlock;
4257                 }
4258                 /* Clear some bits that don't mean anything, but
4259                  * might be left set
4260                  */
4261                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4262                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4263                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4264                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4265
4266                 /* no recovery is running.
4267                  * remove any failed drives, then
4268                  * add spares if possible.
4269                  * Spare are also removed and re-added, to allow
4270                  * the personality to fail the re-add.
4271                  */
4272                 ITERATE_RDEV(mddev,rdev,rtmp)
4273                         if (rdev->raid_disk >= 0 &&
4274                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4275                             atomic_read(&rdev->nr_pending)==0) {
4276                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4277                                         char nm[20];
4278                                         sprintf(nm,"rd%d", rdev->raid_disk);
4279                                         sysfs_remove_link(&mddev->kobj, nm);
4280                                         rdev->raid_disk = -1;
4281                                 }
4282                         }
4283
4284                 if (mddev->degraded) {
4285                         ITERATE_RDEV(mddev,rdev,rtmp)
4286                                 if (rdev->raid_disk < 0
4287                                     && !test_bit(Faulty, &rdev->flags)) {
4288                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4289                                                 char nm[20];
4290                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4291                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4292                                                 spares++;
4293                                         } else
4294                                                 break;
4295                                 }
4296                 }
4297
4298                 if (spares) {
4299                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4300                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4301                 } else if (mddev->recovery_cp < MaxSector) {
4302                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4303                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4304                         /* nothing to be done ... */
4305                         goto unlock;
4306
4307                 if (mddev->pers->sync_request) {
4308                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4309                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4310                                 /* We are adding a device or devices to an array
4311                                  * which has the bitmap stored on all devices.
4312                                  * So make sure all bitmap pages get written
4313                                  */
4314                                 bitmap_write_all(mddev->bitmap);
4315                         }
4316                         mddev->sync_thread = md_register_thread(md_do_sync,
4317                                                                 mddev,
4318                                                                 "%s_resync");
4319                         if (!mddev->sync_thread) {
4320                                 printk(KERN_ERR "%s: could not start resync"
4321                                         " thread...\n",
4322                                         mdname(mddev));
4323                                 /* leave the spares where they are, it shouldn't hurt */
4324                                 mddev->recovery = 0;
4325                         } else {
4326                                 md_wakeup_thread(mddev->sync_thread);
4327                         }
4328                 }
4329         unlock:
4330                 mddev_unlock(mddev);
4331         }
4332 }
4333
4334 static int md_notify_reboot(struct notifier_block *this,
4335                             unsigned long code, void *x)
4336 {
4337         struct list_head *tmp;
4338         mddev_t *mddev;
4339
4340         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4341
4342                 printk(KERN_INFO "md: stopping all md devices.\n");
4343
4344                 ITERATE_MDDEV(mddev,tmp)
4345                         if (mddev_trylock(mddev)==0)
4346                                 do_md_stop (mddev, 1);
4347                 /*
4348                  * certain more exotic SCSI devices are known to be
4349                  * volatile wrt too early system reboots. While the
4350                  * right place to handle this issue is the given
4351                  * driver, we do want to have a safe RAID driver ...
4352                  */
4353                 mdelay(1000*1);
4354         }
4355         return NOTIFY_DONE;
4356 }
4357
4358 static struct notifier_block md_notifier = {
4359         .notifier_call  = md_notify_reboot,
4360         .next           = NULL,
4361         .priority       = INT_MAX, /* before any real devices */
4362 };
4363
4364 static void md_geninit(void)
4365 {
4366         struct proc_dir_entry *p;
4367
4368         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4369
4370         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4371         if (p)
4372                 p->proc_fops = &md_seq_fops;
4373 }
4374
4375 static int __init md_init(void)
4376 {
4377         int minor;
4378
4379         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4380                         " MD_SB_DISKS=%d\n",
4381                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4382                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4383         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4384                         BITMAP_MINOR);
4385
4386         if (register_blkdev(MAJOR_NR, "md"))
4387                 return -1;
4388         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4389                 unregister_blkdev(MAJOR_NR, "md");
4390                 return -1;
4391         }
4392         devfs_mk_dir("md");
4393         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4394                                 md_probe, NULL, NULL);
4395         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4396                             md_probe, NULL, NULL);
4397
4398         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4399                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4400                                 S_IFBLK|S_IRUSR|S_IWUSR,
4401                                 "md/%d", minor);
4402
4403         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4404                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4405                               S_IFBLK|S_IRUSR|S_IWUSR,
4406                               "md/mdp%d", minor);
4407
4408
4409         register_reboot_notifier(&md_notifier);
4410         raid_table_header = register_sysctl_table(raid_root_table, 1);
4411
4412         md_geninit();
4413         return (0);
4414 }
4415
4416
4417 #ifndef MODULE
4418
4419 /*
4420  * Searches all registered partitions for autorun RAID arrays
4421  * at boot time.
4422  */
4423 static dev_t detected_devices[128];
4424 static int dev_cnt;
4425
4426 void md_autodetect_dev(dev_t dev)
4427 {
4428         if (dev_cnt >= 0 && dev_cnt < 127)
4429                 detected_devices[dev_cnt++] = dev;
4430 }
4431
4432
4433 static void autostart_arrays(int part)
4434 {
4435         mdk_rdev_t *rdev;
4436         int i;
4437
4438         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4439
4440         for (i = 0; i < dev_cnt; i++) {
4441                 dev_t dev = detected_devices[i];
4442
4443                 rdev = md_import_device(dev,0, 0);
4444                 if (IS_ERR(rdev))
4445                         continue;
4446
4447                 if (test_bit(Faulty, &rdev->flags)) {
4448                         MD_BUG();
4449                         continue;
4450                 }
4451                 list_add(&rdev->same_set, &pending_raid_disks);
4452         }
4453         dev_cnt = 0;
4454
4455         autorun_devices(part);
4456 }
4457
4458 #endif
4459
4460 static __exit void md_exit(void)
4461 {
4462         mddev_t *mddev;
4463         struct list_head *tmp;
4464         int i;
4465         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4466         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4467         for (i=0; i < MAX_MD_DEVS; i++)
4468                 devfs_remove("md/%d", i);
4469         for (i=0; i < MAX_MD_DEVS; i++)
4470                 devfs_remove("md/d%d", i);
4471
4472         devfs_remove("md");
4473
4474         unregister_blkdev(MAJOR_NR,"md");
4475         unregister_blkdev(mdp_major, "mdp");
4476         unregister_reboot_notifier(&md_notifier);
4477         unregister_sysctl_table(raid_table_header);
4478         remove_proc_entry("mdstat", NULL);
4479         ITERATE_MDDEV(mddev,tmp) {
4480                 struct gendisk *disk = mddev->gendisk;
4481                 if (!disk)
4482                         continue;
4483                 export_array(mddev);
4484                 del_gendisk(disk);
4485                 put_disk(disk);
4486                 mddev->gendisk = NULL;
4487                 mddev_put(mddev);
4488         }
4489 }
4490
4491 module_init(md_init)
4492 module_exit(md_exit)
4493
4494 static int get_ro(char *buffer, struct kernel_param *kp)
4495 {
4496         return sprintf(buffer, "%d", start_readonly);
4497 }
4498 static int set_ro(const char *val, struct kernel_param *kp)
4499 {
4500         char *e;
4501         int num = simple_strtoul(val, &e, 10);
4502         if (*val && (*e == '\0' || *e == '\n')) {
4503                 start_readonly = num;
4504                 return 0;;
4505         }
4506         return -EINVAL;
4507 }
4508
4509 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4510
4511 EXPORT_SYMBOL(register_md_personality);
4512 EXPORT_SYMBOL(unregister_md_personality);
4513 EXPORT_SYMBOL(md_error);
4514 EXPORT_SYMBOL(md_done_sync);
4515 EXPORT_SYMBOL(md_write_start);
4516 EXPORT_SYMBOL(md_write_end);
4517 EXPORT_SYMBOL(md_register_thread);
4518 EXPORT_SYMBOL(md_unregister_thread);
4519 EXPORT_SYMBOL(md_wakeup_thread);
4520 EXPORT_SYMBOL(md_print_devices);
4521 EXPORT_SYMBOL(md_check_recovery);
4522 MODULE_LICENSE("GPL");
4523 MODULE_ALIAS("md");
4524 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);