err.no Git - linux-2.6/blob - drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45
  46 #include <linux/init.h>
  47
  48 #include <linux/file.h>
  49
  50 #ifdef CONFIG_KMOD
  51 #include <linux/kmod.h>
  52 #endif
  53
  54 #include <asm/unaligned.h>
  55
  56 #define MAJOR_NR MD_MAJOR
  57 #define MD_DRIVER
  58
  59 /* 63 partitions with the alternate major number (mdp) */
  60 #define MdpMinorShift 6
  61
  62 #define DEBUG 0
  63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  64
  65
  66 #ifndef MODULE
  67 static void autostart_arrays (int part);
  68 #endif
  69
  70 static mdk_personality_t *pers[MAX_PERSONALITY];
  71 static DEFINE_SPINLOCK(pers_lock);
  72
  73 /*
  74  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  75  * is 1000 KB/sec, so the extra system load does not show up that much.
  76  * Increase it if you want to have more _guaranteed_ speed. Note that
  77  * the RAID driver will use the maximum available bandwidth if the IO
  78  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  79  * speed limit - in case reconstruction slows down your system despite
  80  * idle IO detection.
  81  *
  82  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  83  */
  84
  85 static int sysctl_speed_limit_min = 1000;
  86 static int sysctl_speed_limit_max = 200000;
  87
  88 static struct ctl_table_header *raid_table_header;
  89
  90 static ctl_table raid_table[] = {
  91         {
  92                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  93                 .procname       = "speed_limit_min",
  94                 .data           = &sysctl_speed_limit_min,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec,
  98         },
  99         {
 100                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 101                 .procname       = "speed_limit_max",
 102                 .data           = &sysctl_speed_limit_max,
 103                 .maxlen         = sizeof(int),
 104                 .mode           = 0644,
 105                 .proc_handler   = &proc_dointvec,
 106         },
 107         { .ctl_name = 0 }
 108 };
 109
 110 static ctl_table raid_dir_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID,
 113                 .procname       = "raid",
 114                 .maxlen         = 0,
 115                 .mode           = 0555,
 116                 .child          = raid_table,
 117         },
 118         { .ctl_name = 0 }
 119 };
 120
 121 static ctl_table raid_root_table[] = {
 122         {
 123                 .ctl_name       = CTL_DEV,
 124                 .procname       = "dev",
 125                 .maxlen         = 0,
 126                 .mode           = 0555,
 127                 .child          = raid_dir_table,
 128         },
 129         { .ctl_name = 0 }
 130 };
 131
 132 static struct block_device_operations md_fops;
 133
 134 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 = 1024;
1032
1033                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1034                 memcpy(mddev->uuid, sb->set_uuid, 16);
1035
1036                 mddev->max_disks =  (4096-256)/2;
1037
1038                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1039                     mddev->bitmap_file == NULL ) {
1040                         if (mddev->level != 1) {
1041                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1042                                 return -EINVAL;
1043                         }
1044                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1045                 }
1046         } else if (mddev->pers == NULL) {
1047                 /* Insist of good event counter while assembling */
1048                 __u64 ev1 = le64_to_cpu(sb->events);
1049                 ++ev1;
1050                 if (ev1 < mddev->events)
1051                         return -EINVAL;
1052         } else if (mddev->bitmap) {
1053                 /* If adding to array with a bitmap, then we can accept an
1054                  * older device, but not too old.
1055                  */
1056                 __u64 ev1 = le64_to_cpu(sb->events);
1057                 if (ev1 < mddev->bitmap->events_cleared)
1058                         return 0;
1059         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1060                 return 0;
1061
1062         if (mddev->level != LEVEL_MULTIPATH) {
1063                 int role;
1064                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1065                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1066                 switch(role) {
1067                 case 0xffff: /* spare */
1068                         break;
1069                 case 0xfffe: /* faulty */
1070                         set_bit(Faulty, &rdev->flags);
1071                         break;
1072                 default:
1073                         set_bit(In_sync, &rdev->flags);
1074                         rdev->raid_disk = role;
1075                         break;
1076                 }
1077                 if (sb->devflags & WriteMostly1)
1078                         set_bit(WriteMostly, &rdev->flags);
1079         } else /* MULTIPATH are always insync */
1080                 set_bit(In_sync, &rdev->flags);
1081
1082         return 0;
1083 }
1084
1085 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1086 {
1087         struct mdp_superblock_1 *sb;
1088         struct list_head *tmp;
1089         mdk_rdev_t *rdev2;
1090         int max_dev, i;
1091         /* make rdev->sb match mddev and rdev data. */
1092
1093         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1094
1095         sb->feature_map = 0;
1096         sb->pad0 = 0;
1097         memset(sb->pad1, 0, sizeof(sb->pad1));
1098         memset(sb->pad2, 0, sizeof(sb->pad2));
1099         memset(sb->pad3, 0, sizeof(sb->pad3));
1100
1101         sb->utime = cpu_to_le64((__u64)mddev->utime);
1102         sb->events = cpu_to_le64(mddev->events);
1103         if (mddev->in_sync)
1104                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1105         else
1106                 sb->resync_offset = cpu_to_le64(0);
1107
1108         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1109                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1110                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1111         }
1112
1113         max_dev = 0;
1114         ITERATE_RDEV(mddev,rdev2,tmp)
1115                 if (rdev2->desc_nr+1 > max_dev)
1116                         max_dev = rdev2->desc_nr+1;
1117
1118         sb->max_dev = cpu_to_le32(max_dev);
1119         for (i=0; i<max_dev;i++)
1120                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1121
1122         ITERATE_RDEV(mddev,rdev2,tmp) {
1123                 i = rdev2->desc_nr;
1124                 if (test_bit(Faulty, &rdev2->flags))
1125                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1126                 else if (test_bit(In_sync, &rdev2->flags))
1127                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1128                 else
1129                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1130         }
1131
1132         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1133         sb->sb_csum = calc_sb_1_csum(sb);
1134 }
1135
1136
1137 static struct super_type super_types[] = {
1138         [0] = {
1139                 .name   = "0.90.0",
1140                 .owner  = THIS_MODULE,
1141                 .load_super     = super_90_load,
1142                 .validate_super = super_90_validate,
1143                 .sync_super     = super_90_sync,
1144         },
1145         [1] = {
1146                 .name   = "md-1",
1147                 .owner  = THIS_MODULE,
1148                 .load_super     = super_1_load,
1149                 .validate_super = super_1_validate,
1150                 .sync_super     = super_1_sync,
1151         },
1152 };
1153
1154 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1155 {
1156         struct list_head *tmp;
1157         mdk_rdev_t *rdev;
1158
1159         ITERATE_RDEV(mddev,rdev,tmp)
1160                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1161                         return rdev;
1162
1163         return NULL;
1164 }
1165
1166 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1167 {
1168         struct list_head *tmp;
1169         mdk_rdev_t *rdev;
1170
1171         ITERATE_RDEV(mddev1,rdev,tmp)
1172                 if (match_dev_unit(mddev2, rdev))
1173                         return 1;
1174
1175         return 0;
1176 }
1177
1178 static LIST_HEAD(pending_raid_disks);
1179
1180 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1181 {
1182         mdk_rdev_t *same_pdev;
1183         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1184         struct kobject *ko;
1185
1186         if (rdev->mddev) {
1187                 MD_BUG();
1188                 return -EINVAL;
1189         }
1190         same_pdev = match_dev_unit(mddev, rdev);
1191         if (same_pdev)
1192                 printk(KERN_WARNING
1193                         "%s: WARNING: %s appears to be on the same physical"
1194                         " disk as %s. True\n     protection against single-disk"
1195                         " failure might be compromised.\n",
1196                         mdname(mddev), bdevname(rdev->bdev,b),
1197                         bdevname(same_pdev->bdev,b2));
1198
1199         /* Verify rdev->desc_nr is unique.
1200          * If it is -1, assign a free number, else
1201          * check number is not in use
1202          */
1203         if (rdev->desc_nr < 0) {
1204                 int choice = 0;
1205                 if (mddev->pers) choice = mddev->raid_disks;
1206                 while (find_rdev_nr(mddev, choice))
1207                         choice++;
1208                 rdev->desc_nr = choice;
1209         } else {
1210                 if (find_rdev_nr(mddev, rdev->desc_nr))
1211                         return -EBUSY;
1212         }
1213         bdevname(rdev->bdev,b);
1214         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1215                 return -ENOMEM;
1216
1217         list_add(&rdev->same_set, &mddev->disks);
1218         rdev->mddev = mddev;
1219         printk(KERN_INFO "md: bind<%s>\n", b);
1220
1221         rdev->kobj.parent = &mddev->kobj;
1222         kobject_add(&rdev->kobj);
1223
1224         if (rdev->bdev->bd_part)
1225                 ko = &rdev->bdev->bd_part->kobj;
1226         else
1227                 ko = &rdev->bdev->bd_disk->kobj;
1228         sysfs_create_link(&rdev->kobj, ko, "block");
1229         return 0;
1230 }
1231
1232 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1233 {
1234         char b[BDEVNAME_SIZE];
1235         if (!rdev->mddev) {
1236                 MD_BUG();
1237                 return;
1238         }
1239         list_del_init(&rdev->same_set);
1240         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1241         rdev->mddev = NULL;
1242         sysfs_remove_link(&rdev->kobj, "block");
1243         kobject_del(&rdev->kobj);
1244 }
1245
1246 /*
1247  * prevent the device from being mounted, repartitioned or
1248  * otherwise reused by a RAID array (or any other kernel
1249  * subsystem), by bd_claiming the device.
1250  */
1251 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1252 {
1253         int err = 0;
1254         struct block_device *bdev;
1255         char b[BDEVNAME_SIZE];
1256
1257         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1258         if (IS_ERR(bdev)) {
1259                 printk(KERN_ERR "md: could not open %s.\n",
1260                         __bdevname(dev, b));
1261                 return PTR_ERR(bdev);
1262         }
1263         err = bd_claim(bdev, rdev);
1264         if (err) {
1265                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1266                         bdevname(bdev, b));
1267                 blkdev_put(bdev);
1268                 return err;
1269         }
1270         rdev->bdev = bdev;
1271         return err;
1272 }
1273
1274 static void unlock_rdev(mdk_rdev_t *rdev)
1275 {
1276         struct block_device *bdev = rdev->bdev;
1277         rdev->bdev = NULL;
1278         if (!bdev)
1279                 MD_BUG();
1280         bd_release(bdev);
1281         blkdev_put(bdev);
1282 }
1283
1284 void md_autodetect_dev(dev_t dev);
1285
1286 static void export_rdev(mdk_rdev_t * rdev)
1287 {
1288         char b[BDEVNAME_SIZE];
1289         printk(KERN_INFO "md: export_rdev(%s)\n",
1290                 bdevname(rdev->bdev,b));
1291         if (rdev->mddev)
1292                 MD_BUG();
1293         free_disk_sb(rdev);
1294         list_del_init(&rdev->same_set);
1295 #ifndef MODULE
1296         md_autodetect_dev(rdev->bdev->bd_dev);
1297 #endif
1298         unlock_rdev(rdev);
1299         kobject_put(&rdev->kobj);
1300 }
1301
1302 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1303 {
1304         unbind_rdev_from_array(rdev);
1305         export_rdev(rdev);
1306 }
1307
1308 static void export_array(mddev_t *mddev)
1309 {
1310         struct list_head *tmp;
1311         mdk_rdev_t *rdev;
1312
1313         ITERATE_RDEV(mddev,rdev,tmp) {
1314                 if (!rdev->mddev) {
1315                         MD_BUG();
1316                         continue;
1317                 }
1318                 kick_rdev_from_array(rdev);
1319         }
1320         if (!list_empty(&mddev->disks))
1321                 MD_BUG();
1322         mddev->raid_disks = 0;
1323         mddev->major_version = 0;
1324 }
1325
1326 static void print_desc(mdp_disk_t *desc)
1327 {
1328         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1329                 desc->major,desc->minor,desc->raid_disk,desc->state);
1330 }
1331
1332 static void print_sb(mdp_super_t *sb)
1333 {
1334         int i;
1335
1336         printk(KERN_INFO
1337                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1338                 sb->major_version, sb->minor_version, sb->patch_version,
1339                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1340                 sb->ctime);
1341         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1342                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1343                 sb->md_minor, sb->layout, sb->chunk_size);
1344         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1345                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1346                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1347                 sb->failed_disks, sb->spare_disks,
1348                 sb->sb_csum, (unsigned long)sb->events_lo);
1349
1350         printk(KERN_INFO);
1351         for (i = 0; i < MD_SB_DISKS; i++) {
1352                 mdp_disk_t *desc;
1353
1354                 desc = sb->disks + i;
1355                 if (desc->number || desc->major || desc->minor ||
1356                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1357                         printk("     D %2d: ", i);
1358                         print_desc(desc);
1359                 }
1360         }
1361         printk(KERN_INFO "md:     THIS: ");
1362         print_desc(&sb->this_disk);
1363
1364 }
1365
1366 static void print_rdev(mdk_rdev_t *rdev)
1367 {
1368         char b[BDEVNAME_SIZE];
1369         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1370                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1371                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1372                 rdev->desc_nr);
1373         if (rdev->sb_loaded) {
1374                 printk(KERN_INFO "md: rdev superblock:\n");
1375                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1376         } else
1377                 printk(KERN_INFO "md: no rdev superblock!\n");
1378 }
1379
1380 void md_print_devices(void)
1381 {
1382         struct list_head *tmp, *tmp2;
1383         mdk_rdev_t *rdev;
1384         mddev_t *mddev;
1385         char b[BDEVNAME_SIZE];
1386
1387         printk("\n");
1388         printk("md:     **********************************\n");
1389         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1390         printk("md:     **********************************\n");
1391         ITERATE_MDDEV(mddev,tmp) {
1392
1393                 if (mddev->bitmap)
1394                         bitmap_print_sb(mddev->bitmap);
1395                 else
1396                         printk("%s: ", mdname(mddev));
1397                 ITERATE_RDEV(mddev,rdev,tmp2)
1398                         printk("<%s>", bdevname(rdev->bdev,b));
1399                 printk("\n");
1400
1401                 ITERATE_RDEV(mddev,rdev,tmp2)
1402                         print_rdev(rdev);
1403         }
1404         printk("md:     **********************************\n");
1405         printk("\n");
1406 }
1407
1408
1409 static void sync_sbs(mddev_t * mddev)
1410 {
1411         mdk_rdev_t *rdev;
1412         struct list_head *tmp;
1413
1414         ITERATE_RDEV(mddev,rdev,tmp) {
1415                 super_types[mddev->major_version].
1416                         sync_super(mddev, rdev);
1417                 rdev->sb_loaded = 1;
1418         }
1419 }
1420
1421 static void md_update_sb(mddev_t * mddev)
1422 {
1423         int err;
1424         struct list_head *tmp;
1425         mdk_rdev_t *rdev;
1426         int sync_req;
1427
1428 repeat:
1429         spin_lock_irq(&mddev->write_lock);
1430         sync_req = mddev->in_sync;
1431         mddev->utime = get_seconds();
1432         mddev->events ++;
1433
1434         if (!mddev->events) {
1435                 /*
1436                  * oops, this 64-bit counter should never wrap.
1437                  * Either we are in around ~1 trillion A.C., assuming
1438                  * 1 reboot per second, or we have a bug:
1439                  */
1440                 MD_BUG();
1441                 mddev->events --;
1442         }
1443         mddev->sb_dirty = 2;
1444         sync_sbs(mddev);
1445
1446         /*
1447          * do not write anything to disk if using
1448          * nonpersistent superblocks
1449          */
1450         if (!mddev->persistent) {
1451                 mddev->sb_dirty = 0;
1452                 spin_unlock_irq(&mddev->write_lock);
1453                 wake_up(&mddev->sb_wait);
1454                 return;
1455         }
1456         spin_unlock_irq(&mddev->write_lock);
1457
1458         dprintk(KERN_INFO
1459                 "md: updating %s RAID superblock on device (in sync %d)\n",
1460                 mdname(mddev),mddev->in_sync);
1461
1462         err = bitmap_update_sb(mddev->bitmap);
1463         ITERATE_RDEV(mddev,rdev,tmp) {
1464                 char b[BDEVNAME_SIZE];
1465                 dprintk(KERN_INFO "md: ");
1466                 if (test_bit(Faulty, &rdev->flags))
1467                         dprintk("(skipping faulty ");
1468
1469                 dprintk("%s ", bdevname(rdev->bdev,b));
1470                 if (!test_bit(Faulty, &rdev->flags)) {
1471                         md_super_write(mddev,rdev,
1472                                        rdev->sb_offset<<1, rdev->sb_size,
1473                                        rdev->sb_page);
1474                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1475                                 bdevname(rdev->bdev,b),
1476                                 (unsigned long long)rdev->sb_offset);
1477
1478                 } else
1479                         dprintk(")\n");
1480                 if (mddev->level == LEVEL_MULTIPATH)
1481                         /* only need to write one superblock... */
1482                         break;
1483         }
1484         md_super_wait(mddev);
1485         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1486
1487         spin_lock_irq(&mddev->write_lock);
1488         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1489                 /* have to write it out again */
1490                 spin_unlock_irq(&mddev->write_lock);
1491                 goto repeat;
1492         }
1493         mddev->sb_dirty = 0;
1494         spin_unlock_irq(&mddev->write_lock);
1495         wake_up(&mddev->sb_wait);
1496
1497 }
1498
1499 struct rdev_sysfs_entry {
1500         struct attribute attr;
1501         ssize_t (*show)(mdk_rdev_t *, char *);
1502         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1503 };
1504
1505 static ssize_t
1506 state_show(mdk_rdev_t *rdev, char *page)
1507 {
1508         char *sep = "";
1509         int len=0;
1510
1511         if (test_bit(Faulty, &rdev->flags)) {
1512                 len+= sprintf(page+len, "%sfaulty",sep);
1513                 sep = ",";
1514         }
1515         if (test_bit(In_sync, &rdev->flags)) {
1516                 len += sprintf(page+len, "%sin_sync",sep);
1517                 sep = ",";
1518         }
1519         if (!test_bit(Faulty, &rdev->flags) &&
1520             !test_bit(In_sync, &rdev->flags)) {
1521                 len += sprintf(page+len, "%sspare", sep);
1522                 sep = ",";
1523         }
1524         return len+sprintf(page+len, "\n");
1525 }
1526
1527 static struct rdev_sysfs_entry
1528 rdev_state = __ATTR_RO(state);
1529
1530 static ssize_t
1531 super_show(mdk_rdev_t *rdev, char *page)
1532 {
1533         if (rdev->sb_loaded && rdev->sb_size) {
1534                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1535                 return rdev->sb_size;
1536         } else
1537                 return 0;
1538 }
1539 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1540
1541 static struct attribute *rdev_default_attrs[] = {
1542         &rdev_state.attr,
1543         &rdev_super.attr,
1544         NULL,
1545 };
1546 static ssize_t
1547 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1548 {
1549         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1550         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1551
1552         if (!entry->show)
1553                 return -EIO;
1554         return entry->show(rdev, page);
1555 }
1556
1557 static ssize_t
1558 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1559               const char *page, size_t length)
1560 {
1561         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1562         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1563
1564         if (!entry->store)
1565                 return -EIO;
1566         return entry->store(rdev, page, length);
1567 }
1568
1569 static void rdev_free(struct kobject *ko)
1570 {
1571         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1572         kfree(rdev);
1573 }
1574 static struct sysfs_ops rdev_sysfs_ops = {
1575         .show           = rdev_attr_show,
1576         .store          = rdev_attr_store,
1577 };
1578 static struct kobj_type rdev_ktype = {
1579         .release        = rdev_free,
1580         .sysfs_ops      = &rdev_sysfs_ops,
1581         .default_attrs  = rdev_default_attrs,
1582 };
1583
1584 /*
1585  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1586  *
1587  * mark the device faulty if:
1588  *
1589  *   - the device is nonexistent (zero size)
1590  *   - the device has no valid superblock
1591  *
1592  * a faulty rdev _never_ has rdev->sb set.
1593  */
1594 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1595 {
1596         char b[BDEVNAME_SIZE];
1597         int err;
1598         mdk_rdev_t *rdev;
1599         sector_t size;
1600
1601         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1602         if (!rdev) {
1603                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1604                 return ERR_PTR(-ENOMEM);
1605         }
1606         memset(rdev, 0, sizeof(*rdev));
1607
1608         if ((err = alloc_disk_sb(rdev)))
1609                 goto abort_free;
1610
1611         err = lock_rdev(rdev, newdev);
1612         if (err)
1613                 goto abort_free;
1614
1615         rdev->kobj.parent = NULL;
1616         rdev->kobj.ktype = &rdev_ktype;
1617         kobject_init(&rdev->kobj);
1618
1619         rdev->desc_nr = -1;
1620         rdev->flags = 0;
1621         rdev->data_offset = 0;
1622         atomic_set(&rdev->nr_pending, 0);
1623         atomic_set(&rdev->read_errors, 0);
1624
1625         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1626         if (!size) {
1627                 printk(KERN_WARNING
1628                         "md: %s has zero or unknown size, marking faulty!\n",
1629                         bdevname(rdev->bdev,b));
1630                 err = -EINVAL;
1631                 goto abort_free;
1632         }
1633
1634         if (super_format >= 0) {
1635                 err = super_types[super_format].
1636                         load_super(rdev, NULL, super_minor);
1637                 if (err == -EINVAL) {
1638                         printk(KERN_WARNING
1639                                 "md: %s has invalid sb, not importing!\n",
1640                                 bdevname(rdev->bdev,b));
1641                         goto abort_free;
1642                 }
1643                 if (err < 0) {
1644                         printk(KERN_WARNING
1645                                 "md: could not read %s's sb, not importing!\n",
1646                                 bdevname(rdev->bdev,b));
1647                         goto abort_free;
1648                 }
1649         }
1650         INIT_LIST_HEAD(&rdev->same_set);
1651
1652         return rdev;
1653
1654 abort_free:
1655         if (rdev->sb_page) {
1656                 if (rdev->bdev)
1657                         unlock_rdev(rdev);
1658                 free_disk_sb(rdev);
1659         }
1660         kfree(rdev);
1661         return ERR_PTR(err);
1662 }
1663
1664 /*
1665  * Check a full RAID array for plausibility
1666  */
1667
1668
1669 static void analyze_sbs(mddev_t * mddev)
1670 {
1671         int i;
1672         struct list_head *tmp;
1673         mdk_rdev_t *rdev, *freshest;
1674         char b[BDEVNAME_SIZE];
1675
1676         freshest = NULL;
1677         ITERATE_RDEV(mddev,rdev,tmp)
1678                 switch (super_types[mddev->major_version].
1679                         load_super(rdev, freshest, mddev->minor_version)) {
1680                 case 1:
1681                         freshest = rdev;
1682                         break;
1683                 case 0:
1684                         break;
1685                 default:
1686                         printk( KERN_ERR \
1687                                 "md: fatal superblock inconsistency in %s"
1688                                 " -- removing from array\n",
1689                                 bdevname(rdev->bdev,b));
1690                         kick_rdev_from_array(rdev);
1691                 }
1692
1693
1694         super_types[mddev->major_version].
1695                 validate_super(mddev, freshest);
1696
1697         i = 0;
1698         ITERATE_RDEV(mddev,rdev,tmp) {
1699                 if (rdev != freshest)
1700                         if (super_types[mddev->major_version].
1701                             validate_super(mddev, rdev)) {
1702                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1703                                         " from array!\n",
1704                                         bdevname(rdev->bdev,b));
1705                                 kick_rdev_from_array(rdev);
1706                                 continue;
1707                         }
1708                 if (mddev->level == LEVEL_MULTIPATH) {
1709                         rdev->desc_nr = i++;
1710                         rdev->raid_disk = rdev->desc_nr;
1711                         set_bit(In_sync, &rdev->flags);
1712                 }
1713         }
1714
1715
1716
1717         if (mddev->recovery_cp != MaxSector &&
1718             mddev->level >= 1)
1719                 printk(KERN_ERR "md: %s: raid array is not clean"
1720                        " -- starting background reconstruction\n",
1721                        mdname(mddev));
1722
1723 }
1724
1725 static ssize_t
1726 level_show(mddev_t *mddev, char *page)
1727 {
1728         mdk_personality_t *p = mddev->pers;
1729         if (p == NULL && mddev->raid_disks == 0)
1730                 return 0;
1731         if (mddev->level >= 0)
1732                 return sprintf(page, "raid%d\n", mddev->level);
1733         else
1734                 return sprintf(page, "%s\n", p->name);
1735 }
1736
1737 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1738
1739 static ssize_t
1740 raid_disks_show(mddev_t *mddev, char *page)
1741 {
1742         if (mddev->raid_disks == 0)
1743                 return 0;
1744         return sprintf(page, "%d\n", mddev->raid_disks);
1745 }
1746
1747 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1748
1749 static ssize_t
1750 action_show(mddev_t *mddev, char *page)
1751 {
1752         char *type = "idle";
1753         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1754             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1755                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1756                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1757                                 type = "resync";
1758                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1759                                 type = "check";
1760                         else
1761                                 type = "repair";
1762                 } else
1763                         type = "recover";
1764         }
1765         return sprintf(page, "%s\n", type);
1766 }
1767
1768 static ssize_t
1769 action_store(mddev_t *mddev, const char *page, size_t len)
1770 {
1771         if (!mddev->pers || !mddev->pers->sync_request)
1772                 return -EINVAL;
1773
1774         if (strcmp(page, "idle")==0 || strcmp(page, "idle\n")==0) {
1775                 if (mddev->sync_thread) {
1776                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1777                         md_unregister_thread(mddev->sync_thread);
1778                         mddev->sync_thread = NULL;
1779                         mddev->recovery = 0;
1780                 }
1781                 return len;
1782         }
1783
1784         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1785             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1786                 return -EBUSY;
1787         if (strcmp(page, "resync")==0 || strcmp(page, "resync\n")==0 ||
1788             strcmp(page, "recover")==0 || strcmp(page, "recover\n")==0)
1789                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1790         else {
1791                 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1792                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1793                 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1794                         return -EINVAL;
1795                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1796                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1797                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1798         }
1799         md_wakeup_thread(mddev->thread);
1800         return len;
1801 }
1802
1803 static ssize_t
1804 mismatch_cnt_show(mddev_t *mddev, char *page)
1805 {
1806         return sprintf(page, "%llu\n",
1807                        (unsigned long long) mddev->resync_mismatches);
1808 }
1809
1810 static struct md_sysfs_entry
1811 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1812
1813
1814 static struct md_sysfs_entry
1815 md_mismatches = __ATTR_RO(mismatch_cnt);
1816
1817 static struct attribute *md_default_attrs[] = {
1818         &md_level.attr,
1819         &md_raid_disks.attr,
1820         NULL,
1821 };
1822
1823 static struct attribute *md_redundancy_attrs[] = {
1824         &md_scan_mode.attr,
1825         &md_mismatches.attr,
1826         NULL,
1827 };
1828 static struct attribute_group md_redundancy_group = {
1829         .name = NULL,
1830         .attrs = md_redundancy_attrs,
1831 };
1832
1833
1834 static ssize_t
1835 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1836 {
1837         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1838         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1839         ssize_t rv;
1840
1841         if (!entry->show)
1842                 return -EIO;
1843         mddev_lock(mddev);
1844         rv = entry->show(mddev, page);
1845         mddev_unlock(mddev);
1846         return rv;
1847 }
1848
1849 static ssize_t
1850 md_attr_store(struct kobject *kobj, struct attribute *attr,
1851               const char *page, size_t length)
1852 {
1853         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1854         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1855         ssize_t rv;
1856
1857         if (!entry->store)
1858                 return -EIO;
1859         mddev_lock(mddev);
1860         rv = entry->store(mddev, page, length);
1861         mddev_unlock(mddev);
1862         return rv;
1863 }
1864
1865 static void md_free(struct kobject *ko)
1866 {
1867         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1868         kfree(mddev);
1869 }
1870
1871 static struct sysfs_ops md_sysfs_ops = {
1872         .show   = md_attr_show,
1873         .store  = md_attr_store,
1874 };
1875 static struct kobj_type md_ktype = {
1876         .release        = md_free,
1877         .sysfs_ops      = &md_sysfs_ops,
1878         .default_attrs  = md_default_attrs,
1879 };
1880
1881 int mdp_major = 0;
1882
1883 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1884 {
1885         static DECLARE_MUTEX(disks_sem);
1886         mddev_t *mddev = mddev_find(dev);
1887         struct gendisk *disk;
1888         int partitioned = (MAJOR(dev) != MD_MAJOR);
1889         int shift = partitioned ? MdpMinorShift : 0;
1890         int unit = MINOR(dev) >> shift;
1891
1892         if (!mddev)
1893                 return NULL;
1894
1895         down(&disks_sem);
1896         if (mddev->gendisk) {
1897                 up(&disks_sem);
1898                 mddev_put(mddev);
1899                 return NULL;
1900         }
1901         disk = alloc_disk(1 << shift);
1902         if (!disk) {
1903                 up(&disks_sem);
1904                 mddev_put(mddev);
1905                 return NULL;
1906         }
1907         disk->major = MAJOR(dev);
1908         disk->first_minor = unit << shift;
1909         if (partitioned) {
1910                 sprintf(disk->disk_name, "md_d%d", unit);
1911                 sprintf(disk->devfs_name, "md/d%d", unit);
1912         } else {
1913                 sprintf(disk->disk_name, "md%d", unit);
1914                 sprintf(disk->devfs_name, "md/%d", unit);
1915         }
1916         disk->fops = &md_fops;
1917         disk->private_data = mddev;
1918         disk->queue = mddev->queue;
1919         add_disk(disk);
1920         mddev->gendisk = disk;
1921         up(&disks_sem);
1922         mddev->kobj.parent = &disk->kobj;
1923         mddev->kobj.k_name = NULL;
1924         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1925         mddev->kobj.ktype = &md_ktype;
1926         kobject_register(&mddev->kobj);
1927         return NULL;
1928 }
1929
1930 void md_wakeup_thread(mdk_thread_t *thread);
1931
1932 static void md_safemode_timeout(unsigned long data)
1933 {
1934         mddev_t *mddev = (mddev_t *) data;
1935
1936         mddev->safemode = 1;
1937         md_wakeup_thread(mddev->thread);
1938 }
1939
1940 static int start_dirty_degraded;
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         mddev->ok_start_degraded = start_dirty_degraded;
2053
2054         if (start_readonly)
2055                 mddev->ro = 2; /* read-only, but switch on first write */
2056
2057         err = mddev->pers->run(mddev);
2058         if (!err && mddev->pers->sync_request) {
2059                 err = bitmap_create(mddev);
2060                 if (err) {
2061                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2062                                mdname(mddev), err);
2063                         mddev->pers->stop(mddev);
2064                 }
2065         }
2066         if (err) {
2067                 printk(KERN_ERR "md: pers->run() failed ...\n");
2068                 module_put(mddev->pers->owner);
2069                 mddev->pers = NULL;
2070                 bitmap_destroy(mddev);
2071                 return err;
2072         }
2073         if (mddev->pers->sync_request)
2074                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2075         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2076                 mddev->ro = 0;
2077
2078         atomic_set(&mddev->writes_pending,0);
2079         mddev->safemode = 0;
2080         mddev->safemode_timer.function = md_safemode_timeout;
2081         mddev->safemode_timer.data = (unsigned long) mddev;
2082         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2083         mddev->in_sync = 1;
2084
2085         ITERATE_RDEV(mddev,rdev,tmp)
2086                 if (rdev->raid_disk >= 0) {
2087                         char nm[20];
2088                         sprintf(nm, "rd%d", rdev->raid_disk);
2089                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2090                 }
2091
2092         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2093         md_wakeup_thread(mddev->thread);
2094
2095         if (mddev->sb_dirty)
2096                 md_update_sb(mddev);
2097
2098         set_capacity(disk, mddev->array_size<<1);
2099
2100         /* If we call blk_queue_make_request here, it will
2101          * re-initialise max_sectors etc which may have been
2102          * refined inside -> run.  So just set the bits we need to set.
2103          * Most initialisation happended when we called
2104          * blk_queue_make_request(..., md_fail_request)
2105          * earlier.
2106          */
2107         mddev->queue->queuedata = mddev;
2108         mddev->queue->make_request_fn = mddev->pers->make_request;
2109
2110         mddev->changed = 1;
2111         return 0;
2112 }
2113
2114 static int restart_array(mddev_t *mddev)
2115 {
2116         struct gendisk *disk = mddev->gendisk;
2117         int err;
2118
2119         /*
2120          * Complain if it has no devices
2121          */
2122         err = -ENXIO;
2123         if (list_empty(&mddev->disks))
2124                 goto out;
2125
2126         if (mddev->pers) {
2127                 err = -EBUSY;
2128                 if (!mddev->ro)
2129                         goto out;
2130
2131                 mddev->safemode = 0;
2132                 mddev->ro = 0;
2133                 set_disk_ro(disk, 0);
2134
2135                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2136                         mdname(mddev));
2137                 /*
2138                  * Kick recovery or resync if necessary
2139                  */
2140                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2141                 md_wakeup_thread(mddev->thread);
2142                 err = 0;
2143         } else {
2144                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2145                         mdname(mddev));
2146                 err = -EINVAL;
2147         }
2148
2149 out:
2150         return err;
2151 }
2152
2153 static int do_md_stop(mddev_t * mddev, int ro)
2154 {
2155         int err = 0;
2156         struct gendisk *disk = mddev->gendisk;
2157
2158         if (mddev->pers) {
2159                 if (atomic_read(&mddev->active)>2) {
2160                         printk("md: %s still in use.\n",mdname(mddev));
2161                         return -EBUSY;
2162                 }
2163
2164                 if (mddev->sync_thread) {
2165                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2166                         md_unregister_thread(mddev->sync_thread);
2167                         mddev->sync_thread = NULL;
2168                 }
2169
2170                 del_timer_sync(&mddev->safemode_timer);
2171
2172                 invalidate_partition(disk, 0);
2173
2174                 if (ro) {
2175                         err  = -ENXIO;
2176                         if (mddev->ro==1)
2177                                 goto out;
2178                         mddev->ro = 1;
2179                 } else {
2180                         bitmap_flush(mddev);
2181                         md_super_wait(mddev);
2182                         if (mddev->ro)
2183                                 set_disk_ro(disk, 0);
2184                         blk_queue_make_request(mddev->queue, md_fail_request);
2185                         mddev->pers->stop(mddev);
2186                         if (mddev->pers->sync_request)
2187                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2188
2189                         module_put(mddev->pers->owner);
2190                         mddev->pers = NULL;
2191                         if (mddev->ro)
2192                                 mddev->ro = 0;
2193                 }
2194                 if (!mddev->in_sync) {
2195                         /* mark array as shutdown cleanly */
2196                         mddev->in_sync = 1;
2197                         md_update_sb(mddev);
2198                 }
2199                 if (ro)
2200                         set_disk_ro(disk, 1);
2201         }
2202
2203         bitmap_destroy(mddev);
2204         if (mddev->bitmap_file) {
2205                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2206                 fput(mddev->bitmap_file);
2207                 mddev->bitmap_file = NULL;
2208         }
2209         mddev->bitmap_offset = 0;
2210
2211         /*
2212          * Free resources if final stop
2213          */
2214         if (!ro) {
2215                 mdk_rdev_t *rdev;
2216                 struct list_head *tmp;
2217                 struct gendisk *disk;
2218                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2219
2220                 ITERATE_RDEV(mddev,rdev,tmp)
2221                         if (rdev->raid_disk >= 0) {
2222                                 char nm[20];
2223                                 sprintf(nm, "rd%d", rdev->raid_disk);
2224                                 sysfs_remove_link(&mddev->kobj, nm);
2225                         }
2226
2227                 export_array(mddev);
2228
2229                 mddev->array_size = 0;
2230                 disk = mddev->gendisk;
2231                 if (disk)
2232                         set_capacity(disk, 0);
2233                 mddev->changed = 1;
2234         } else
2235                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2236                         mdname(mddev));
2237         err = 0;
2238 out:
2239         return err;
2240 }
2241
2242 static void autorun_array(mddev_t *mddev)
2243 {
2244         mdk_rdev_t *rdev;
2245         struct list_head *tmp;
2246         int err;
2247
2248         if (list_empty(&mddev->disks))
2249                 return;
2250
2251         printk(KERN_INFO "md: running: ");
2252
2253         ITERATE_RDEV(mddev,rdev,tmp) {
2254                 char b[BDEVNAME_SIZE];
2255                 printk("<%s>", bdevname(rdev->bdev,b));
2256         }
2257         printk("\n");
2258
2259         err = do_md_run (mddev);
2260         if (err) {
2261                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2262                 do_md_stop (mddev, 0);
2263         }
2264 }
2265
2266 /*
2267  * lets try to run arrays based on all disks that have arrived
2268  * until now. (those are in pending_raid_disks)
2269  *
2270  * the method: pick the first pending disk, collect all disks with
2271  * the same UUID, remove all from the pending list and put them into
2272  * the 'same_array' list. Then order this list based on superblock
2273  * update time (freshest comes first), kick out 'old' disks and
2274  * compare superblocks. If everything's fine then run it.
2275  *
2276  * If "unit" is allocated, then bump its reference count
2277  */
2278 static void autorun_devices(int part)
2279 {
2280         struct list_head candidates;
2281         struct list_head *tmp;
2282         mdk_rdev_t *rdev0, *rdev;
2283         mddev_t *mddev;
2284         char b[BDEVNAME_SIZE];
2285
2286         printk(KERN_INFO "md: autorun ...\n");
2287         while (!list_empty(&pending_raid_disks)) {
2288                 dev_t dev;
2289                 rdev0 = list_entry(pending_raid_disks.next,
2290                                          mdk_rdev_t, same_set);
2291
2292                 printk(KERN_INFO "md: considering %s ...\n",
2293                         bdevname(rdev0->bdev,b));
2294                 INIT_LIST_HEAD(&candidates);
2295                 ITERATE_RDEV_PENDING(rdev,tmp)
2296                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2297                                 printk(KERN_INFO "md:  adding %s ...\n",
2298                                         bdevname(rdev->bdev,b));
2299                                 list_move(&rdev->same_set, &candidates);
2300                         }
2301                 /*
2302                  * now we have a set of devices, with all of them having
2303                  * mostly sane superblocks. It's time to allocate the
2304                  * mddev.
2305                  */
2306                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2307                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2308                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2309                         break;
2310                 }
2311                 if (part)
2312                         dev = MKDEV(mdp_major,
2313                                     rdev0->preferred_minor << MdpMinorShift);
2314                 else
2315                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2316
2317                 md_probe(dev, NULL, NULL);
2318                 mddev = mddev_find(dev);
2319                 if (!mddev) {
2320                         printk(KERN_ERR
2321                                 "md: cannot allocate memory for md drive.\n");
2322                         break;
2323                 }
2324                 if (mddev_lock(mddev))
2325                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2326                                mdname(mddev));
2327                 else if (mddev->raid_disks || mddev->major_version
2328                          || !list_empty(&mddev->disks)) {
2329                         printk(KERN_WARNING
2330                                 "md: %s already running, cannot run %s\n",
2331                                 mdname(mddev), bdevname(rdev0->bdev,b));
2332                         mddev_unlock(mddev);
2333                 } else {
2334                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2335                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2336                                 list_del_init(&rdev->same_set);
2337                                 if (bind_rdev_to_array(rdev, mddev))
2338                                         export_rdev(rdev);
2339                         }
2340                         autorun_array(mddev);
2341                         mddev_unlock(mddev);
2342                 }
2343                 /* on success, candidates will be empty, on error
2344                  * it won't...
2345                  */
2346                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2347                         export_rdev(rdev);
2348                 mddev_put(mddev);
2349         }
2350         printk(KERN_INFO "md: ... autorun DONE.\n");
2351 }
2352
2353 /*
2354  * import RAID devices based on one partition
2355  * if possible, the array gets run as well.
2356  */
2357
2358 static int autostart_array(dev_t startdev)
2359 {
2360         char b[BDEVNAME_SIZE];
2361         int err = -EINVAL, i;
2362         mdp_super_t *sb = NULL;
2363         mdk_rdev_t *start_rdev = NULL, *rdev;
2364
2365         start_rdev = md_import_device(startdev, 0, 0);
2366         if (IS_ERR(start_rdev))
2367                 return err;
2368
2369
2370         /* NOTE: this can only work for 0.90.0 superblocks */
2371         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2372         if (sb->major_version != 0 ||
2373             sb->minor_version != 90 ) {
2374                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2375                 export_rdev(start_rdev);
2376                 return err;
2377         }
2378
2379         if (test_bit(Faulty, &start_rdev->flags)) {
2380                 printk(KERN_WARNING
2381                         "md: can not autostart based on faulty %s!\n",
2382                         bdevname(start_rdev->bdev,b));
2383                 export_rdev(start_rdev);
2384                 return err;
2385         }
2386         list_add(&start_rdev->same_set, &pending_raid_disks);
2387
2388         for (i = 0; i < MD_SB_DISKS; i++) {
2389                 mdp_disk_t *desc = sb->disks + i;
2390                 dev_t dev = MKDEV(desc->major, desc->minor);
2391
2392                 if (!dev)
2393                         continue;
2394                 if (dev == startdev)
2395                         continue;
2396                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2397                         continue;
2398                 rdev = md_import_device(dev, 0, 0);
2399                 if (IS_ERR(rdev))
2400                         continue;
2401
2402                 list_add(&rdev->same_set, &pending_raid_disks);
2403         }
2404
2405         /*
2406          * possibly return codes
2407          */
2408         autorun_devices(0);
2409         return 0;
2410
2411 }
2412
2413
2414 static int get_version(void __user * arg)
2415 {
2416         mdu_version_t ver;
2417
2418         ver.major = MD_MAJOR_VERSION;
2419         ver.minor = MD_MINOR_VERSION;
2420         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2421
2422         if (copy_to_user(arg, &ver, sizeof(ver)))
2423                 return -EFAULT;
2424
2425         return 0;
2426 }
2427
2428 static int get_array_info(mddev_t * mddev, void __user * arg)
2429 {
2430         mdu_array_info_t info;
2431         int nr,working,active,failed,spare;
2432         mdk_rdev_t *rdev;
2433         struct list_head *tmp;
2434
2435         nr=working=active=failed=spare=0;
2436         ITERATE_RDEV(mddev,rdev,tmp) {
2437                 nr++;
2438                 if (test_bit(Faulty, &rdev->flags))
2439                         failed++;
2440                 else {
2441                         working++;
2442                         if (test_bit(In_sync, &rdev->flags))
2443                                 active++;
2444                         else
2445                                 spare++;
2446                 }
2447         }
2448
2449         info.major_version = mddev->major_version;
2450         info.minor_version = mddev->minor_version;
2451         info.patch_version = MD_PATCHLEVEL_VERSION;
2452         info.ctime         = mddev->ctime;
2453         info.level         = mddev->level;
2454         info.size          = mddev->size;
2455         info.nr_disks      = nr;
2456         info.raid_disks    = mddev->raid_disks;
2457         info.md_minor      = mddev->md_minor;
2458         info.not_persistent= !mddev->persistent;
2459
2460         info.utime         = mddev->utime;
2461         info.state         = 0;
2462         if (mddev->in_sync)
2463                 info.state = (1<<MD_SB_CLEAN);
2464         if (mddev->bitmap && mddev->bitmap_offset)
2465                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2466         info.active_disks  = active;
2467         info.working_disks = working;
2468         info.failed_disks  = failed;
2469         info.spare_disks   = spare;
2470
2471         info.layout        = mddev->layout;
2472         info.chunk_size    = mddev->chunk_size;
2473
2474         if (copy_to_user(arg, &info, sizeof(info)))
2475                 return -EFAULT;
2476
2477         return 0;
2478 }
2479
2480 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2481 {
2482         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2483         char *ptr, *buf = NULL;
2484         int err = -ENOMEM;
2485
2486         file = kmalloc(sizeof(*file), GFP_KERNEL);
2487         if (!file)
2488                 goto out;
2489
2490         /* bitmap disabled, zero the first byte and copy out */
2491         if (!mddev->bitmap || !mddev->bitmap->file) {
2492                 file->pathname[0] = '\0';
2493                 goto copy_out;
2494         }
2495
2496         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2497         if (!buf)
2498                 goto out;
2499
2500         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2501         if (!ptr)
2502                 goto out;
2503
2504         strcpy(file->pathname, ptr);
2505
2506 copy_out:
2507         err = 0;
2508         if (copy_to_user(arg, file, sizeof(*file)))
2509                 err = -EFAULT;
2510 out:
2511         kfree(buf);
2512         kfree(file);
2513         return err;
2514 }
2515
2516 static int get_disk_info(mddev_t * mddev, void __user * arg)
2517 {
2518         mdu_disk_info_t info;
2519         unsigned int nr;
2520         mdk_rdev_t *rdev;
2521
2522         if (copy_from_user(&info, arg, sizeof(info)))
2523                 return -EFAULT;
2524
2525         nr = info.number;
2526
2527         rdev = find_rdev_nr(mddev, nr);
2528         if (rdev) {
2529                 info.major = MAJOR(rdev->bdev->bd_dev);
2530                 info.minor = MINOR(rdev->bdev->bd_dev);
2531                 info.raid_disk = rdev->raid_disk;
2532                 info.state = 0;
2533                 if (test_bit(Faulty, &rdev->flags))
2534                         info.state |= (1<<MD_DISK_FAULTY);
2535                 else if (test_bit(In_sync, &rdev->flags)) {
2536                         info.state |= (1<<MD_DISK_ACTIVE);
2537                         info.state |= (1<<MD_DISK_SYNC);
2538                 }
2539                 if (test_bit(WriteMostly, &rdev->flags))
2540                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2541         } else {
2542                 info.major = info.minor = 0;
2543                 info.raid_disk = -1;
2544                 info.state = (1<<MD_DISK_REMOVED);
2545         }
2546
2547         if (copy_to_user(arg, &info, sizeof(info)))
2548                 return -EFAULT;
2549
2550         return 0;
2551 }
2552
2553 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2554 {
2555         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2556         mdk_rdev_t *rdev;
2557         dev_t dev = MKDEV(info->major,info->minor);
2558
2559         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2560                 return -EOVERFLOW;
2561
2562         if (!mddev->raid_disks) {
2563                 int err;
2564                 /* expecting a device which has a superblock */
2565                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2566                 if (IS_ERR(rdev)) {
2567                         printk(KERN_WARNING
2568                                 "md: md_import_device returned %ld\n",
2569                                 PTR_ERR(rdev));
2570                         return PTR_ERR(rdev);
2571                 }
2572                 if (!list_empty(&mddev->disks)) {
2573                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2574                                                         mdk_rdev_t, same_set);
2575                         int err = super_types[mddev->major_version]
2576                                 .load_super(rdev, rdev0, mddev->minor_version);
2577                         if (err < 0) {
2578                                 printk(KERN_WARNING
2579                                         "md: %s has different UUID to %s\n",
2580                                         bdevname(rdev->bdev,b),
2581                                         bdevname(rdev0->bdev,b2));
2582                                 export_rdev(rdev);
2583                                 return -EINVAL;
2584                         }
2585                 }
2586                 err = bind_rdev_to_array(rdev, mddev);
2587                 if (err)
2588                         export_rdev(rdev);
2589                 return err;
2590         }
2591
2592         /*
2593          * add_new_disk can be used once the array is assembled
2594          * to add "hot spares".  They must already have a superblock
2595          * written
2596          */
2597         if (mddev->pers) {
2598                 int err;
2599                 if (!mddev->pers->hot_add_disk) {
2600                         printk(KERN_WARNING
2601                                 "%s: personality does not support diskops!\n",
2602                                mdname(mddev));
2603                         return -EINVAL;
2604                 }
2605                 if (mddev->persistent)
2606                         rdev = md_import_device(dev, mddev->major_version,
2607                                                 mddev->minor_version);
2608                 else
2609                         rdev = md_import_device(dev, -1, -1);
2610                 if (IS_ERR(rdev)) {
2611                         printk(KERN_WARNING
2612                                 "md: md_import_device returned %ld\n",
2613                                 PTR_ERR(rdev));
2614                         return PTR_ERR(rdev);
2615                 }
2616                 /* set save_raid_disk if appropriate */
2617                 if (!mddev->persistent) {
2618                         if (info->state & (1<<MD_DISK_SYNC)  &&
2619                             info->raid_disk < mddev->raid_disks)
2620                                 rdev->raid_disk = info->raid_disk;
2621                         else
2622                                 rdev->raid_disk = -1;
2623                 } else
2624                         super_types[mddev->major_version].
2625                                 validate_super(mddev, rdev);
2626                 rdev->saved_raid_disk = rdev->raid_disk;
2627
2628                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2629                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2630                         set_bit(WriteMostly, &rdev->flags);
2631
2632                 rdev->raid_disk = -1;
2633                 err = bind_rdev_to_array(rdev, mddev);
2634                 if (err)
2635                         export_rdev(rdev);
2636
2637                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2638                 md_wakeup_thread(mddev->thread);
2639                 return err;
2640         }
2641
2642         /* otherwise, add_new_disk is only allowed
2643          * for major_version==0 superblocks
2644          */
2645         if (mddev->major_version != 0) {
2646                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2647                        mdname(mddev));
2648                 return -EINVAL;
2649         }
2650
2651         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2652                 int err;
2653                 rdev = md_import_device (dev, -1, 0);
2654                 if (IS_ERR(rdev)) {
2655                         printk(KERN_WARNING
2656                                 "md: error, md_import_device() returned %ld\n",
2657                                 PTR_ERR(rdev));
2658                         return PTR_ERR(rdev);
2659                 }
2660                 rdev->desc_nr = info->number;
2661                 if (info->raid_disk < mddev->raid_disks)
2662                         rdev->raid_disk = info->raid_disk;
2663                 else
2664                         rdev->raid_disk = -1;
2665
2666                 rdev->flags = 0;
2667
2668                 if (rdev->raid_disk < mddev->raid_disks)
2669                         if (info->state & (1<<MD_DISK_SYNC))
2670                                 set_bit(In_sync, &rdev->flags);
2671
2672                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2673                         set_bit(WriteMostly, &rdev->flags);
2674
2675                 err = bind_rdev_to_array(rdev, mddev);
2676                 if (err) {
2677                         export_rdev(rdev);
2678                         return err;
2679                 }
2680
2681                 if (!mddev->persistent) {
2682                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2683                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2684                 } else
2685                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2686                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2687
2688                 if (!mddev->size || (mddev->size > rdev->size))
2689                         mddev->size = rdev->size;
2690         }
2691
2692         return 0;
2693 }
2694
2695 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2696 {
2697         char b[BDEVNAME_SIZE];
2698         mdk_rdev_t *rdev;
2699
2700         if (!mddev->pers)
2701                 return -ENODEV;
2702
2703         rdev = find_rdev(mddev, dev);
2704         if (!rdev)
2705                 return -ENXIO;
2706
2707         if (rdev->raid_disk >= 0)
2708                 goto busy;
2709
2710         kick_rdev_from_array(rdev);
2711         md_update_sb(mddev);
2712
2713         return 0;
2714 busy:
2715         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2716                 bdevname(rdev->bdev,b), mdname(mddev));
2717         return -EBUSY;
2718 }
2719
2720 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2721 {
2722         char b[BDEVNAME_SIZE];
2723         int err;
2724         unsigned int size;
2725         mdk_rdev_t *rdev;
2726
2727         if (!mddev->pers)
2728                 return -ENODEV;
2729
2730         if (mddev->major_version != 0) {
2731                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2732                         " version-0 superblocks.\n",
2733                         mdname(mddev));
2734                 return -EINVAL;
2735         }
2736         if (!mddev->pers->hot_add_disk) {
2737                 printk(KERN_WARNING
2738                         "%s: personality does not support diskops!\n",
2739                         mdname(mddev));
2740                 return -EINVAL;
2741         }
2742
2743         rdev = md_import_device (dev, -1, 0);
2744         if (IS_ERR(rdev)) {
2745                 printk(KERN_WARNING
2746                         "md: error, md_import_device() returned %ld\n",
2747                         PTR_ERR(rdev));
2748                 return -EINVAL;
2749         }
2750
2751         if (mddev->persistent)
2752                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2753         else
2754                 rdev->sb_offset =
2755                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2756
2757         size = calc_dev_size(rdev, mddev->chunk_size);
2758         rdev->size = size;
2759
2760         if (size < mddev->size) {
2761                 printk(KERN_WARNING
2762                         "%s: disk size %llu blocks < array size %llu\n",
2763                         mdname(mddev), (unsigned long long)size,
2764                         (unsigned long long)mddev->size);
2765                 err = -ENOSPC;
2766                 goto abort_export;
2767         }
2768
2769         if (test_bit(Faulty, &rdev->flags)) {
2770                 printk(KERN_WARNING
2771                         "md: can not hot-add faulty %s disk to %s!\n",
2772                         bdevname(rdev->bdev,b), mdname(mddev));
2773                 err = -EINVAL;
2774                 goto abort_export;
2775         }
2776         clear_bit(In_sync, &rdev->flags);
2777         rdev->desc_nr = -1;
2778         bind_rdev_to_array(rdev, mddev);
2779
2780         /*
2781          * The rest should better be atomic, we can have disk failures
2782          * noticed in interrupt contexts ...
2783          */
2784
2785         if (rdev->desc_nr == mddev->max_disks) {
2786                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2787                         mdname(mddev));
2788                 err = -EBUSY;
2789                 goto abort_unbind_export;
2790         }
2791
2792         rdev->raid_disk = -1;
2793
2794         md_update_sb(mddev);
2795
2796         /*
2797          * Kick recovery, maybe this spare has to be added to the
2798          * array immediately.
2799          */
2800         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2801         md_wakeup_thread(mddev->thread);
2802
2803         return 0;
2804
2805 abort_unbind_export:
2806         unbind_rdev_from_array(rdev);
2807
2808 abort_export:
2809         export_rdev(rdev);
2810         return err;
2811 }
2812
2813 /* similar to deny_write_access, but accounts for our holding a reference
2814  * to the file ourselves */
2815 static int deny_bitmap_write_access(struct file * file)
2816 {
2817         struct inode *inode = file->f_mapping->host;
2818
2819         spin_lock(&inode->i_lock);
2820         if (atomic_read(&inode->i_writecount) > 1) {
2821                 spin_unlock(&inode->i_lock);
2822                 return -ETXTBSY;
2823         }
2824         atomic_set(&inode->i_writecount, -1);
2825         spin_unlock(&inode->i_lock);
2826
2827         return 0;
2828 }
2829
2830 static int set_bitmap_file(mddev_t *mddev, int fd)
2831 {
2832         int err;
2833
2834         if (mddev->pers) {
2835                 if (!mddev->pers->quiesce)
2836                         return -EBUSY;
2837                 if (mddev->recovery || mddev->sync_thread)
2838                         return -EBUSY;
2839                 /* we should be able to change the bitmap.. */
2840         }
2841
2842
2843         if (fd >= 0) {
2844                 if (mddev->bitmap)
2845                         return -EEXIST; /* cannot add when bitmap is present */
2846                 mddev->bitmap_file = fget(fd);
2847
2848                 if (mddev->bitmap_file == NULL) {
2849                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2850                                mdname(mddev));
2851                         return -EBADF;
2852                 }
2853
2854                 err = deny_bitmap_write_access(mddev->bitmap_file);
2855                 if (err) {
2856                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2857                                mdname(mddev));
2858                         fput(mddev->bitmap_file);
2859                         mddev->bitmap_file = NULL;
2860                         return err;
2861                 }
2862                 mddev->bitmap_offset = 0; /* file overrides offset */
2863         } else if (mddev->bitmap == NULL)
2864                 return -ENOENT; /* cannot remove what isn't there */
2865         err = 0;
2866         if (mddev->pers) {
2867                 mddev->pers->quiesce(mddev, 1);
2868                 if (fd >= 0)
2869                         err = bitmap_create(mddev);
2870                 if (fd < 0 || err)
2871                         bitmap_destroy(mddev);
2872                 mddev->pers->quiesce(mddev, 0);
2873         } else if (fd < 0) {
2874                 if (mddev->bitmap_file)
2875                         fput(mddev->bitmap_file);
2876                 mddev->bitmap_file = NULL;
2877         }
2878
2879         return err;
2880 }
2881
2882 /*
2883  * set_array_info is used two different ways
2884  * The original usage is when creating a new array.
2885  * In this usage, raid_disks is > 0 and it together with
2886  *  level, size, not_persistent,layout,chunksize determine the
2887  *  shape of the array.
2888  *  This will always create an array with a type-0.90.0 superblock.
2889  * The newer usage is when assembling an array.
2890  *  In this case raid_disks will be 0, and the major_version field is
2891  *  use to determine which style super-blocks are to be found on the devices.
2892  *  The minor and patch _version numbers are also kept incase the
2893  *  super_block handler wishes to interpret them.
2894  */
2895 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2896 {
2897
2898         if (info->raid_disks == 0) {
2899                 /* just setting version number for superblock loading */
2900                 if (info->major_version < 0 ||
2901                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2902                     super_types[info->major_version].name == NULL) {
2903                         /* maybe try to auto-load a module? */
2904                         printk(KERN_INFO
2905                                 "md: superblock version %d not known\n",
2906                                 info->major_version);
2907                         return -EINVAL;
2908                 }
2909                 mddev->major_version = info->major_version;
2910                 mddev->minor_version = info->minor_version;
2911                 mddev->patch_version = info->patch_version;
2912                 return 0;
2913         }
2914         mddev->major_version = MD_MAJOR_VERSION;
2915         mddev->minor_version = MD_MINOR_VERSION;
2916         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2917         mddev->ctime         = get_seconds();
2918
2919         mddev->level         = info->level;
2920         mddev->size          = info->size;
2921         mddev->raid_disks    = info->raid_disks;
2922         /* don't set md_minor, it is determined by which /dev/md* was
2923          * openned
2924          */
2925         if (info->state & (1<<MD_SB_CLEAN))
2926                 mddev->recovery_cp = MaxSector;
2927         else
2928                 mddev->recovery_cp = 0;
2929         mddev->persistent    = ! info->not_persistent;
2930
2931         mddev->layout        = info->layout;
2932         mddev->chunk_size    = info->chunk_size;
2933
2934         mddev->max_disks     = MD_SB_DISKS;
2935
2936         mddev->sb_dirty      = 1;
2937
2938         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2939         mddev->bitmap_offset = 0;
2940
2941         /*
2942          * Generate a 128 bit UUID
2943          */
2944         get_random_bytes(mddev->uuid, 16);
2945
2946         return 0;
2947 }
2948
2949 /*
2950  * update_array_info is used to change the configuration of an
2951  * on-line array.
2952  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2953  * fields in the info are checked against the array.
2954  * Any differences that cannot be handled will cause an error.
2955  * Normally, only one change can be managed at a time.
2956  */
2957 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2958 {
2959         int rv = 0;
2960         int cnt = 0;
2961         int state = 0;
2962
2963         /* calculate expected state,ignoring low bits */
2964         if (mddev->bitmap && mddev->bitmap_offset)
2965                 state |= (1 << MD_SB_BITMAP_PRESENT);
2966
2967         if (mddev->major_version != info->major_version ||
2968             mddev->minor_version != info->minor_version ||
2969 /*          mddev->patch_version != info->patch_version || */
2970             mddev->ctime         != info->ctime         ||
2971             mddev->level         != info->level         ||
2972 /*          mddev->layout        != info->layout        || */
2973             !mddev->persistent   != info->not_persistent||
2974             mddev->chunk_size    != info->chunk_size    ||
2975             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2976             ((state^info->state) & 0xfffffe00)
2977                 )
2978                 return -EINVAL;
2979         /* Check there is only one change */
2980         if (mddev->size != info->size) cnt++;
2981         if (mddev->raid_disks != info->raid_disks) cnt++;
2982         if (mddev->layout != info->layout) cnt++;
2983         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2984         if (cnt == 0) return 0;
2985         if (cnt > 1) return -EINVAL;
2986
2987         if (mddev->layout != info->layout) {
2988                 /* Change layout
2989                  * we don't need to do anything at the md level, the
2990                  * personality will take care of it all.
2991                  */
2992                 if (mddev->pers->reconfig == NULL)
2993                         return -EINVAL;
2994                 else
2995                         return mddev->pers->reconfig(mddev, info->layout, -1);
2996         }
2997         if (mddev->size != info->size) {
2998                 mdk_rdev_t * rdev;
2999                 struct list_head *tmp;
3000                 if (mddev->pers->resize == NULL)
3001                         return -EINVAL;
3002                 /* The "size" is the amount of each device that is used.
3003                  * This can only make sense for arrays with redundancy.
3004                  * linear and raid0 always use whatever space is available
3005                  * We can only consider changing the size if no resync
3006                  * or reconstruction is happening, and if the new size
3007                  * is acceptable. It must fit before the sb_offset or,
3008                  * if that is <data_offset, it must fit before the
3009                  * size of each device.
3010                  * If size is zero, we find the largest size that fits.
3011                  */
3012                 if (mddev->sync_thread)
3013                         return -EBUSY;
3014                 ITERATE_RDEV(mddev,rdev,tmp) {
3015                         sector_t avail;
3016                         int fit = (info->size == 0);
3017                         if (rdev->sb_offset > rdev->data_offset)
3018                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
3019                         else
3020                                 avail = get_capacity(rdev->bdev->bd_disk)
3021                                         - rdev->data_offset;
3022                         if (fit && (info->size == 0 || info->size > avail/2))
3023                                 info->size = avail/2;
3024                         if (avail < ((sector_t)info->size << 1))
3025                                 return -ENOSPC;
3026                 }
3027                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3028                 if (!rv) {
3029                         struct block_device *bdev;
3030
3031                         bdev = bdget_disk(mddev->gendisk, 0);
3032                         if (bdev) {
3033                                 down(&bdev->bd_inode->i_sem);
3034                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3035                                 up(&bdev->bd_inode->i_sem);
3036                                 bdput(bdev);
3037                         }
3038                 }
3039         }
3040         if (mddev->raid_disks    != info->raid_disks) {
3041                 /* change the number of raid disks */
3042                 if (mddev->pers->reshape == NULL)
3043                         return -EINVAL;
3044                 if (info->raid_disks <= 0 ||
3045                     info->raid_disks >= mddev->max_disks)
3046                         return -EINVAL;
3047                 if (mddev->sync_thread)
3048                         return -EBUSY;
3049                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3050                 if (!rv) {
3051                         struct block_device *bdev;
3052
3053                         bdev = bdget_disk(mddev->gendisk, 0);
3054                         if (bdev) {
3055                                 down(&bdev->bd_inode->i_sem);
3056                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3057                                 up(&bdev->bd_inode->i_sem);
3058                                 bdput(bdev);
3059                         }
3060                 }
3061         }
3062         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3063                 if (mddev->pers->quiesce == NULL)
3064                         return -EINVAL;
3065                 if (mddev->recovery || mddev->sync_thread)
3066                         return -EBUSY;
3067                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3068                         /* add the bitmap */
3069                         if (mddev->bitmap)
3070                                 return -EEXIST;
3071                         if (mddev->default_bitmap_offset == 0)
3072                                 return -EINVAL;
3073                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3074                         mddev->pers->quiesce(mddev, 1);
3075                         rv = bitmap_create(mddev);
3076                         if (rv)
3077                                 bitmap_destroy(mddev);
3078                         mddev->pers->quiesce(mddev, 0);
3079                 } else {
3080                         /* remove the bitmap */
3081                         if (!mddev->bitmap)
3082                                 return -ENOENT;
3083                         if (mddev->bitmap->file)
3084                                 return -EINVAL;
3085                         mddev->pers->quiesce(mddev, 1);
3086                         bitmap_destroy(mddev);
3087                         mddev->pers->quiesce(mddev, 0);
3088                         mddev->bitmap_offset = 0;
3089                 }
3090         }
3091         md_update_sb(mddev);
3092         return rv;
3093 }
3094
3095 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3096 {
3097         mdk_rdev_t *rdev;
3098
3099         if (mddev->pers == NULL)
3100                 return -ENODEV;
3101
3102         rdev = find_rdev(mddev, dev);
3103         if (!rdev)
3104                 return -ENODEV;
3105
3106         md_error(mddev, rdev);
3107         return 0;
3108 }
3109
3110 static int md_ioctl(struct inode *inode, struct file *file,
3111                         unsigned int cmd, unsigned long arg)
3112 {
3113         int err = 0;
3114         void __user *argp = (void __user *)arg;
3115         struct hd_geometry __user *loc = argp;
3116         mddev_t *mddev = NULL;
3117
3118         if (!capable(CAP_SYS_ADMIN))
3119                 return -EACCES;
3120
3121         /*
3122          * Commands dealing with the RAID driver but not any
3123          * particular array:
3124          */
3125         switch (cmd)
3126         {
3127                 case RAID_VERSION:
3128                         err = get_version(argp);
3129                         goto done;
3130
3131                 case PRINT_RAID_DEBUG:
3132                         err = 0;
3133                         md_print_devices();
3134                         goto done;
3135
3136 #ifndef MODULE
3137                 case RAID_AUTORUN:
3138                         err = 0;
3139                         autostart_arrays(arg);
3140                         goto done;
3141 #endif
3142                 default:;
3143         }
3144
3145         /*
3146          * Commands creating/starting a new array:
3147          */
3148
3149         mddev = inode->i_bdev->bd_disk->private_data;
3150
3151         if (!mddev) {
3152                 BUG();
3153                 goto abort;
3154         }
3155
3156
3157         if (cmd == START_ARRAY) {
3158                 /* START_ARRAY doesn't need to lock the array as autostart_array
3159                  * does the locking, and it could even be a different array
3160                  */
3161                 static int cnt = 3;
3162                 if (cnt > 0 ) {
3163                         printk(KERN_WARNING
3164                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3165                                "This will not be supported beyond July 2006\n",
3166                                current->comm, current->pid);
3167                         cnt--;
3168                 }
3169                 err = autostart_array(new_decode_dev(arg));
3170                 if (err) {
3171                         printk(KERN_WARNING "md: autostart failed!\n");
3172                         goto abort;
3173                 }
3174                 goto done;
3175         }
3176
3177         err = mddev_lock(mddev);
3178         if (err) {
3179                 printk(KERN_INFO
3180                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3181                         err, cmd);
3182                 goto abort;
3183         }
3184
3185         switch (cmd)
3186         {
3187                 case SET_ARRAY_INFO:
3188                         {
3189                                 mdu_array_info_t info;
3190                                 if (!arg)
3191                                         memset(&info, 0, sizeof(info));
3192                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3193                                         err = -EFAULT;
3194                                         goto abort_unlock;
3195                                 }
3196                                 if (mddev->pers) {
3197                                         err = update_array_info(mddev, &info);
3198                                         if (err) {
3199                                                 printk(KERN_WARNING "md: couldn't update"
3200                                                        " array info. %d\n", err);
3201                                                 goto abort_unlock;
3202                                         }
3203                                         goto done_unlock;
3204                                 }
3205                                 if (!list_empty(&mddev->disks)) {
3206                                         printk(KERN_WARNING
3207                                                "md: array %s already has disks!\n",
3208                                                mdname(mddev));
3209                                         err = -EBUSY;
3210                                         goto abort_unlock;
3211                                 }
3212                                 if (mddev->raid_disks) {
3213                                         printk(KERN_WARNING
3214                                                "md: array %s already initialised!\n",
3215                                                mdname(mddev));
3216                                         err = -EBUSY;
3217                                         goto abort_unlock;
3218                                 }
3219                                 err = set_array_info(mddev, &info);
3220                                 if (err) {
3221                                         printk(KERN_WARNING "md: couldn't set"
3222                                                " array info. %d\n", err);
3223                                         goto abort_unlock;
3224                                 }
3225                         }
3226                         goto done_unlock;
3227
3228                 default:;
3229         }
3230
3231         /*
3232          * Commands querying/configuring an existing array:
3233          */
3234         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3235          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3236         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3237                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3238                 err = -ENODEV;
3239                 goto abort_unlock;
3240         }
3241
3242         /*
3243          * Commands even a read-only array can execute:
3244          */
3245         switch (cmd)
3246         {
3247                 case GET_ARRAY_INFO:
3248                         err = get_array_info(mddev, argp);
3249                         goto done_unlock;
3250
3251                 case GET_BITMAP_FILE:
3252                         err = get_bitmap_file(mddev, argp);
3253                         goto done_unlock;
3254
3255                 case GET_DISK_INFO:
3256                         err = get_disk_info(mddev, argp);
3257                         goto done_unlock;
3258
3259                 case RESTART_ARRAY_RW:
3260                         err = restart_array(mddev);
3261                         goto done_unlock;
3262
3263                 case STOP_ARRAY:
3264                         err = do_md_stop (mddev, 0);
3265                         goto done_unlock;
3266
3267                 case STOP_ARRAY_RO:
3268                         err = do_md_stop (mddev, 1);
3269                         goto done_unlock;
3270
3271         /*
3272          * We have a problem here : there is no easy way to give a CHS
3273          * virtual geometry. We currently pretend that we have a 2 heads
3274          * 4 sectors (with a BIG number of cylinders...). This drives
3275          * dosfs just mad... ;-)
3276          */
3277                 case HDIO_GETGEO:
3278                         if (!loc) {
3279                                 err = -EINVAL;
3280                                 goto abort_unlock;
3281                         }
3282                         err = put_user (2, (char __user *) &loc->heads);
3283                         if (err)
3284                                 goto abort_unlock;
3285                         err = put_user (4, (char __user *) &loc->sectors);
3286                         if (err)
3287                                 goto abort_unlock;
3288                         err = put_user(get_capacity(mddev->gendisk)/8,
3289                                         (short __user *) &loc->cylinders);
3290                         if (err)
3291                                 goto abort_unlock;
3292                         err = put_user (get_start_sect(inode->i_bdev),
3293                                                 (long __user *) &loc->start);
3294                         goto done_unlock;
3295         }
3296
3297         /*
3298          * The remaining ioctls are changing the state of the
3299          * superblock, so we do not allow them on read-only arrays.
3300          * However non-MD ioctls (e.g. get-size) will still come through
3301          * here and hit the 'default' below, so only disallow
3302          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3303          */
3304         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3305             mddev->ro && mddev->pers) {
3306                 if (mddev->ro == 2) {
3307                         mddev->ro = 0;
3308                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3309                 md_wakeup_thread(mddev->thread);
3310
3311                 } else {
3312                         err = -EROFS;
3313                         goto abort_unlock;
3314                 }
3315         }
3316
3317         switch (cmd)
3318         {
3319                 case ADD_NEW_DISK:
3320                 {
3321                         mdu_disk_info_t info;
3322                         if (copy_from_user(&info, argp, sizeof(info)))
3323                                 err = -EFAULT;
3324                         else
3325                                 err = add_new_disk(mddev, &info);
3326                         goto done_unlock;
3327                 }
3328
3329                 case HOT_REMOVE_DISK:
3330                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3331                         goto done_unlock;
3332
3333                 case HOT_ADD_DISK:
3334                         err = hot_add_disk(mddev, new_decode_dev(arg));
3335                         goto done_unlock;
3336
3337                 case SET_DISK_FAULTY:
3338                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3339                         goto done_unlock;
3340
3341                 case RUN_ARRAY:
3342                         err = do_md_run (mddev);
3343                         goto done_unlock;
3344
3345                 case SET_BITMAP_FILE:
3346                         err = set_bitmap_file(mddev, (int)arg);
3347                         goto done_unlock;
3348
3349                 default:
3350                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3351                                 printk(KERN_WARNING "md: %s(pid %d) used"
3352                                         " obsolete MD ioctl, upgrade your"
3353                                         " software to use new ictls.\n",
3354                                         current->comm, current->pid);
3355                         err = -EINVAL;
3356                         goto abort_unlock;
3357         }
3358
3359 done_unlock:
3360 abort_unlock:
3361         mddev_unlock(mddev);
3362
3363         return err;
3364 done:
3365         if (err)
3366                 MD_BUG();
3367 abort:
3368         return err;
3369 }
3370
3371 static int md_open(struct inode *inode, struct file *file)
3372 {
3373         /*
3374          * Succeed if we can lock the mddev, which confirms that
3375          * it isn't being stopped right now.
3376          */
3377         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3378         int err;
3379
3380         if ((err = mddev_lock(mddev)))
3381                 goto out;
3382
3383         err = 0;
3384         mddev_get(mddev);
3385         mddev_unlock(mddev);
3386
3387         check_disk_change(inode->i_bdev);
3388  out:
3389         return err;
3390 }
3391
3392 static int md_release(struct inode *inode, struct file * file)
3393 {
3394         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3395
3396         if (!mddev)
3397                 BUG();
3398         mddev_put(mddev);
3399
3400         return 0;
3401 }
3402
3403 static int md_media_changed(struct gendisk *disk)
3404 {
3405         mddev_t *mddev = disk->private_data;
3406
3407         return mddev->changed;
3408 }
3409
3410 static int md_revalidate(struct gendisk *disk)
3411 {
3412         mddev_t *mddev = disk->private_data;
3413
3414         mddev->changed = 0;
3415         return 0;
3416 }
3417 static struct block_device_operations md_fops =
3418 {
3419         .owner          = THIS_MODULE,
3420         .open           = md_open,
3421         .release        = md_release,
3422         .ioctl          = md_ioctl,
3423         .media_changed  = md_media_changed,
3424         .revalidate_disk= md_revalidate,
3425 };
3426
3427 static int md_thread(void * arg)
3428 {
3429         mdk_thread_t *thread = arg;
3430
3431         /*
3432          * md_thread is a 'system-thread', it's priority should be very
3433          * high. We avoid resource deadlocks individually in each
3434          * raid personality. (RAID5 does preallocation) We also use RR and
3435          * the very same RT priority as kswapd, thus we will never get
3436          * into a priority inversion deadlock.
3437          *
3438          * we definitely have to have equal or higher priority than
3439          * bdflush, otherwise bdflush will deadlock if there are too
3440          * many dirty RAID5 blocks.
3441          */
3442
3443         allow_signal(SIGKILL);
3444         while (!kthread_should_stop()) {
3445
3446                 /* We need to wait INTERRUPTIBLE so that
3447                  * we don't add to the load-average.
3448                  * That means we need to be sure no signals are
3449                  * pending
3450                  */
3451                 if (signal_pending(current))
3452                         flush_signals(current);
3453
3454                 wait_event_interruptible_timeout
3455                         (thread->wqueue,
3456                          test_bit(THREAD_WAKEUP, &thread->flags)
3457                          || kthread_should_stop(),
3458                          thread->timeout);
3459                 try_to_freeze();
3460
3461                 clear_bit(THREAD_WAKEUP, &thread->flags);
3462
3463                 thread->run(thread->mddev);
3464         }
3465
3466         return 0;
3467 }
3468
3469 void md_wakeup_thread(mdk_thread_t *thread)
3470 {
3471         if (thread) {
3472                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3473                 set_bit(THREAD_WAKEUP, &thread->flags);
3474                 wake_up(&thread->wqueue);
3475         }
3476 }
3477
3478 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3479                                  const char *name)
3480 {
3481         mdk_thread_t *thread;
3482
3483         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3484         if (!thread)
3485                 return NULL;
3486
3487         memset(thread, 0, sizeof(mdk_thread_t));
3488         init_waitqueue_head(&thread->wqueue);
3489
3490         thread->run = run;
3491         thread->mddev = mddev;
3492         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3493         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3494         if (IS_ERR(thread->tsk)) {
3495                 kfree(thread);
3496                 return NULL;
3497         }
3498         return thread;
3499 }
3500
3501 void md_unregister_thread(mdk_thread_t *thread)
3502 {
3503         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3504
3505         kthread_stop(thread->tsk);
3506         kfree(thread);
3507 }
3508
3509 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3510 {
3511         if (!mddev) {
3512                 MD_BUG();
3513                 return;
3514         }
3515
3516         if (!rdev || test_bit(Faulty, &rdev->flags))
3517                 return;
3518 /*
3519         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3520                 mdname(mddev),
3521                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3522                 __builtin_return_address(0),__builtin_return_address(1),
3523                 __builtin_return_address(2),__builtin_return_address(3));
3524 */
3525         if (!mddev->pers->error_handler)
3526                 return;
3527         mddev->pers->error_handler(mddev,rdev);
3528         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3529         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3530         md_wakeup_thread(mddev->thread);
3531 }
3532
3533 /* seq_file implementation /proc/mdstat */
3534
3535 static void status_unused(struct seq_file *seq)
3536 {
3537         int i = 0;
3538         mdk_rdev_t *rdev;
3539         struct list_head *tmp;
3540
3541         seq_printf(seq, "unused devices: ");
3542
3543         ITERATE_RDEV_PENDING(rdev,tmp) {
3544                 char b[BDEVNAME_SIZE];
3545                 i++;
3546                 seq_printf(seq, "%s ",
3547                               bdevname(rdev->bdev,b));
3548         }
3549         if (!i)
3550                 seq_printf(seq, "<none>");
3551
3552         seq_printf(seq, "\n");
3553 }
3554
3555
3556 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3557 {
3558         unsigned long max_blocks, resync, res, dt, db, rt;
3559
3560         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3561
3562         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3563                 max_blocks = mddev->resync_max_sectors >> 1;
3564         else
3565                 max_blocks = mddev->size;
3566
3567         /*
3568          * Should not happen.
3569          */
3570         if (!max_blocks) {
3571                 MD_BUG();
3572                 return;
3573         }
3574         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3575         {
3576                 int i, x = res/50, y = 20-x;
3577                 seq_printf(seq, "[");
3578                 for (i = 0; i < x; i++)
3579                         seq_printf(seq, "=");
3580                 seq_printf(seq, ">");
3581                 for (i = 0; i < y; i++)
3582                         seq_printf(seq, ".");
3583                 seq_printf(seq, "] ");
3584         }
3585         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3586                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3587                        "resync" : "recovery"),
3588                       res/10, res % 10, resync, max_blocks);
3589
3590         /*
3591          * We do not want to overflow, so the order of operands and
3592          * the * 100 / 100 trick are important. We do a +1 to be
3593          * safe against division by zero. We only estimate anyway.
3594          *
3595          * dt: time from mark until now
3596          * db: blocks written from mark until now
3597          * rt: remaining time
3598          */
3599         dt = ((jiffies - mddev->resync_mark) / HZ);
3600         if (!dt) dt++;
3601         db = resync - (mddev->resync_mark_cnt/2);
3602         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3603
3604         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3605
3606         seq_printf(seq, " speed=%ldK/sec", db/dt);
3607 }
3608
3609 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3610 {
3611         struct list_head *tmp;
3612         loff_t l = *pos;
3613         mddev_t *mddev;
3614
3615         if (l >= 0x10000)
3616                 return NULL;
3617         if (!l--)
3618                 /* header */
3619                 return (void*)1;
3620
3621         spin_lock(&all_mddevs_lock);
3622         list_for_each(tmp,&all_mddevs)
3623                 if (!l--) {
3624                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3625                         mddev_get(mddev);
3626                         spin_unlock(&all_mddevs_lock);
3627                         return mddev;
3628                 }
3629         spin_unlock(&all_mddevs_lock);
3630         if (!l--)
3631                 return (void*)2;/* tail */
3632         return NULL;
3633 }
3634
3635 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3636 {
3637         struct list_head *tmp;
3638         mddev_t *next_mddev, *mddev = v;
3639
3640         ++*pos;
3641         if (v == (void*)2)
3642                 return NULL;
3643
3644         spin_lock(&all_mddevs_lock);
3645         if (v == (void*)1)
3646                 tmp = all_mddevs.next;
3647         else
3648                 tmp = mddev->all_mddevs.next;
3649         if (tmp != &all_mddevs)
3650                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3651         else {
3652                 next_mddev = (void*)2;
3653                 *pos = 0x10000;
3654         }
3655         spin_unlock(&all_mddevs_lock);
3656
3657         if (v != (void*)1)
3658                 mddev_put(mddev);
3659         return next_mddev;
3660
3661 }
3662
3663 static void md_seq_stop(struct seq_file *seq, void *v)
3664 {
3665         mddev_t *mddev = v;
3666
3667         if (mddev && v != (void*)1 && v != (void*)2)
3668                 mddev_put(mddev);
3669 }
3670
3671 static int md_seq_show(struct seq_file *seq, void *v)
3672 {
3673         mddev_t *mddev = v;
3674         sector_t size;
3675         struct list_head *tmp2;
3676         mdk_rdev_t *rdev;
3677         int i;
3678         struct bitmap *bitmap;
3679
3680         if (v == (void*)1) {
3681                 seq_printf(seq, "Personalities : ");
3682                 spin_lock(&pers_lock);
3683                 for (i = 0; i < MAX_PERSONALITY; i++)
3684                         if (pers[i])
3685                                 seq_printf(seq, "[%s] ", pers[i]->name);
3686
3687                 spin_unlock(&pers_lock);
3688                 seq_printf(seq, "\n");
3689                 return 0;
3690         }
3691         if (v == (void*)2) {
3692                 status_unused(seq);
3693                 return 0;
3694         }
3695
3696         if (mddev_lock(mddev)!=0)
3697                 return -EINTR;
3698         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3699                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3700                                                 mddev->pers ? "" : "in");
3701                 if (mddev->pers) {
3702                         if (mddev->ro==1)
3703                                 seq_printf(seq, " (read-only)");
3704                         if (mddev->ro==2)
3705                                 seq_printf(seq, "(auto-read-only)");
3706                         seq_printf(seq, " %s", mddev->pers->name);
3707                 }
3708
3709                 size = 0;
3710                 ITERATE_RDEV(mddev,rdev,tmp2) {
3711                         char b[BDEVNAME_SIZE];
3712                         seq_printf(seq, " %s[%d]",
3713                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3714                         if (test_bit(WriteMostly, &rdev->flags))
3715                                 seq_printf(seq, "(W)");
3716                         if (test_bit(Faulty, &rdev->flags)) {
3717                                 seq_printf(seq, "(F)");
3718                                 continue;
3719                         } else if (rdev->raid_disk < 0)
3720                                 seq_printf(seq, "(S)"); /* spare */
3721                         size += rdev->size;
3722                 }
3723
3724                 if (!list_empty(&mddev->disks)) {
3725                         if (mddev->pers)
3726                                 seq_printf(seq, "\n      %llu blocks",
3727                                         (unsigned long long)mddev->array_size);
3728                         else
3729                                 seq_printf(seq, "\n      %llu blocks",
3730                                         (unsigned long long)size);
3731                 }
3732                 if (mddev->persistent) {
3733                         if (mddev->major_version != 0 ||
3734                             mddev->minor_version != 90) {
3735                                 seq_printf(seq," super %d.%d",
3736                                            mddev->major_version,
3737                                            mddev->minor_version);
3738                         }
3739                 } else
3740                         seq_printf(seq, " super non-persistent");
3741
3742                 if (mddev->pers) {
3743                         mddev->pers->status (seq, mddev);
3744                         seq_printf(seq, "\n      ");
3745                         if (mddev->pers->sync_request) {
3746                                 if (mddev->curr_resync > 2) {
3747                                         status_resync (seq, mddev);
3748                                         seq_printf(seq, "\n      ");
3749                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3750                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3751                                 else if (mddev->recovery_cp < MaxSector)
3752                                         seq_printf(seq, "\tresync=PENDING\n      ");
3753                         }
3754                 } else
3755                         seq_printf(seq, "\n       ");
3756
3757                 if ((bitmap = mddev->bitmap)) {
3758                         unsigned long chunk_kb;
3759                         unsigned long flags;
3760                         spin_lock_irqsave(&bitmap->lock, flags);
3761                         chunk_kb = bitmap->chunksize >> 10;
3762                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3763                                 "%lu%s chunk",
3764                                 bitmap->pages - bitmap->missing_pages,
3765                                 bitmap->pages,
3766                                 (bitmap->pages - bitmap->missing_pages)
3767                                         << (PAGE_SHIFT - 10),
3768                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3769                                 chunk_kb ? "KB" : "B");
3770                         if (bitmap->file) {
3771                                 seq_printf(seq, ", file: ");
3772                                 seq_path(seq, bitmap->file->f_vfsmnt,
3773                                          bitmap->file->f_dentry," \t\n");
3774                         }
3775
3776                         seq_printf(seq, "\n");
3777                         spin_unlock_irqrestore(&bitmap->lock, flags);
3778                 }
3779
3780                 seq_printf(seq, "\n");
3781         }
3782         mddev_unlock(mddev);
3783
3784         return 0;
3785 }
3786
3787 static struct seq_operations md_seq_ops = {
3788         .start  = md_seq_start,
3789         .next   = md_seq_next,
3790         .stop   = md_seq_stop,
3791         .show   = md_seq_show,
3792 };
3793
3794 static int md_seq_open(struct inode *inode, struct file *file)
3795 {
3796         int error;
3797
3798         error = seq_open(file, &md_seq_ops);
3799         return error;
3800 }
3801
3802 static struct file_operations md_seq_fops = {
3803         .open           = md_seq_open,
3804         .read           = seq_read,
3805         .llseek         = seq_lseek,
3806         .release        = seq_release,
3807 };
3808
3809 int register_md_personality(int pnum, mdk_personality_t *p)
3810 {
3811         if (pnum >= MAX_PERSONALITY) {
3812                 printk(KERN_ERR
3813                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3814                        p->name, pnum, MAX_PERSONALITY-1);
3815                 return -EINVAL;
3816         }
3817
3818         spin_lock(&pers_lock);
3819         if (pers[pnum]) {
3820                 spin_unlock(&pers_lock);
3821                 return -EBUSY;
3822         }
3823
3824         pers[pnum] = p;
3825         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3826         spin_unlock(&pers_lock);
3827         return 0;
3828 }
3829
3830 int unregister_md_personality(int pnum)
3831 {
3832         if (pnum >= MAX_PERSONALITY)
3833                 return -EINVAL;
3834
3835         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3836         spin_lock(&pers_lock);
3837         pers[pnum] = NULL;
3838         spin_unlock(&pers_lock);
3839         return 0;
3840 }
3841
3842 static int is_mddev_idle(mddev_t *mddev)
3843 {
3844         mdk_rdev_t * rdev;
3845         struct list_head *tmp;
3846         int idle;
3847         unsigned long curr_events;
3848
3849         idle = 1;
3850         ITERATE_RDEV(mddev,rdev,tmp) {
3851                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3852                 curr_events = disk_stat_read(disk, sectors[0]) +
3853                                 disk_stat_read(disk, sectors[1]) -
3854                                 atomic_read(&disk->sync_io);
3855                 /* The difference between curr_events and last_events
3856                  * will be affected by any new non-sync IO (making
3857                  * curr_events bigger) and any difference in the amount of
3858                  * in-flight syncio (making current_events bigger or smaller)
3859                  * The amount in-flight is currently limited to
3860                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3861                  * which is at most 4096 sectors.
3862                  * These numbers are fairly fragile and should be made
3863                  * more robust, probably by enforcing the
3864                  * 'window size' that md_do_sync sort-of uses.
3865                  *
3866                  * Note: the following is an unsigned comparison.
3867                  */
3868                 if ((curr_events - rdev->last_events + 4096) > 8192) {
3869                         rdev->last_events = curr_events;
3870                         idle = 0;
3871                 }
3872         }
3873         return idle;
3874 }
3875
3876 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3877 {
3878         /* another "blocks" (512byte) blocks have been synced */
3879         atomic_sub(blocks, &mddev->recovery_active);
3880         wake_up(&mddev->recovery_wait);
3881         if (!ok) {
3882                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3883                 md_wakeup_thread(mddev->thread);
3884                 // stop recovery, signal do_sync ....
3885         }
3886 }
3887
3888
3889 /* md_write_start(mddev, bi)
3890  * If we need to update some array metadata (e.g. 'active' flag
3891  * in superblock) before writing, schedule a superblock update
3892  * and wait for it to complete.
3893  */
3894 void md_write_start(mddev_t *mddev, struct bio *bi)
3895 {
3896         if (bio_data_dir(bi) != WRITE)
3897                 return;
3898
3899         BUG_ON(mddev->ro == 1);
3900         if (mddev->ro == 2) {
3901                 /* need to switch to read/write */
3902                 mddev->ro = 0;
3903                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3904                 md_wakeup_thread(mddev->thread);
3905         }
3906         atomic_inc(&mddev->writes_pending);
3907         if (mddev->in_sync) {
3908                 spin_lock_irq(&mddev->write_lock);
3909                 if (mddev->in_sync) {
3910                         mddev->in_sync = 0;
3911                         mddev->sb_dirty = 1;
3912                         md_wakeup_thread(mddev->thread);
3913                 }
3914                 spin_unlock_irq(&mddev->write_lock);
3915         }
3916         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3917 }
3918
3919 void md_write_end(mddev_t *mddev)
3920 {
3921         if (atomic_dec_and_test(&mddev->writes_pending)) {
3922                 if (mddev->safemode == 2)
3923                         md_wakeup_thread(mddev->thread);
3924                 else
3925                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3926         }
3927 }
3928
3929 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3930
3931 #define SYNC_MARKS      10
3932 #define SYNC_MARK_STEP  (3*HZ)
3933 static void md_do_sync(mddev_t *mddev)
3934 {
3935         mddev_t *mddev2;
3936         unsigned int currspeed = 0,
3937                  window;
3938         sector_t max_sectors,j, io_sectors;
3939         unsigned long mark[SYNC_MARKS];
3940         sector_t mark_cnt[SYNC_MARKS];
3941         int last_mark,m;
3942         struct list_head *tmp;
3943         sector_t last_check;
3944         int skipped = 0;
3945
3946         /* just incase thread restarts... */
3947         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3948                 return;
3949
3950         /* we overload curr_resync somewhat here.
3951          * 0 == not engaged in resync at all
3952          * 2 == checking that there is no conflict with another sync
3953          * 1 == like 2, but have yielded to allow conflicting resync to
3954          *              commense
3955          * other == active in resync - this many blocks
3956          *
3957          * Before starting a resync we must have set curr_resync to
3958          * 2, and then checked that every "conflicting" array has curr_resync
3959          * less than ours.  When we find one that is the same or higher
3960          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3961          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3962          * This will mean we have to start checking from the beginning again.
3963          *
3964          */
3965
3966         do {
3967                 mddev->curr_resync = 2;
3968
3969         try_again:
3970                 if (kthread_should_stop()) {
3971                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3972                         goto skip;
3973                 }
3974                 ITERATE_MDDEV(mddev2,tmp) {
3975                         if (mddev2 == mddev)
3976                                 continue;
3977                         if (mddev2->curr_resync &&
3978                             match_mddev_units(mddev,mddev2)) {
3979                                 DEFINE_WAIT(wq);
3980                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3981                                         /* arbitrarily yield */
3982                                         mddev->curr_resync = 1;
3983                                         wake_up(&resync_wait);
3984                                 }
3985                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3986                                         /* no need to wait here, we can wait the next
3987                                          * time 'round when curr_resync == 2
3988                                          */
3989                                         continue;
3990                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
3991                                 if (!kthread_should_stop() &&
3992                                     mddev2->curr_resync >= mddev->curr_resync) {
3993                                         printk(KERN_INFO "md: delaying resync of %s"
3994                                                " until %s has finished resync (they"
3995                                                " share one or more physical units)\n",
3996                                                mdname(mddev), mdname(mddev2));
3997                                         mddev_put(mddev2);
3998                                         schedule();
3999                                         finish_wait(&resync_wait, &wq);
4000                                         goto try_again;
4001                                 }
4002                                 finish_wait(&resync_wait, &wq);
4003                         }
4004                 }
4005         } while (mddev->curr_resync < 2);
4006
4007         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4008                 /* resync follows the size requested by the personality,
4009                  * which defaults to physical size, but can be virtual size
4010                  */
4011                 max_sectors = mddev->resync_max_sectors;
4012                 mddev->resync_mismatches = 0;
4013         } else
4014                 /* recovery follows the physical size of devices */
4015                 max_sectors = mddev->size << 1;
4016
4017         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4018         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4019                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4020         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4021                "(but not more than %d KB/sec) for reconstruction.\n",
4022                sysctl_speed_limit_max);
4023
4024         is_mddev_idle(mddev); /* this also initializes IO event counters */
4025         /* we don't use the checkpoint if there's a bitmap */
4026         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4027             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4028                 j = mddev->recovery_cp;
4029         else
4030                 j = 0;
4031         io_sectors = 0;
4032         for (m = 0; m < SYNC_MARKS; m++) {
4033                 mark[m] = jiffies;
4034                 mark_cnt[m] = io_sectors;
4035         }
4036         last_mark = 0;
4037         mddev->resync_mark = mark[last_mark];
4038         mddev->resync_mark_cnt = mark_cnt[last_mark];
4039
4040         /*
4041          * Tune reconstruction:
4042          */
4043         window = 32*(PAGE_SIZE/512);
4044         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4045                 window/2,(unsigned long long) max_sectors/2);
4046
4047         atomic_set(&mddev->recovery_active, 0);
4048         init_waitqueue_head(&mddev->recovery_wait);
4049         last_check = 0;
4050
4051         if (j>2) {
4052                 printk(KERN_INFO
4053                         "md: resuming recovery of %s from checkpoint.\n",
4054                         mdname(mddev));
4055                 mddev->curr_resync = j;
4056         }
4057
4058         while (j < max_sectors) {
4059                 sector_t sectors;
4060
4061                 skipped = 0;
4062                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4063                                             currspeed < sysctl_speed_limit_min);
4064                 if (sectors == 0) {
4065                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4066                         goto out;
4067                 }
4068
4069                 if (!skipped) { /* actual IO requested */
4070                         io_sectors += sectors;
4071                         atomic_add(sectors, &mddev->recovery_active);
4072                 }
4073
4074                 j += sectors;
4075                 if (j>1) mddev->curr_resync = j;
4076
4077
4078                 if (last_check + window > io_sectors || j == max_sectors)
4079                         continue;
4080
4081                 last_check = io_sectors;
4082
4083                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4084                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4085                         break;
4086
4087         repeat:
4088                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4089                         /* step marks */
4090                         int next = (last_mark+1) % SYNC_MARKS;
4091
4092                         mddev->resync_mark = mark[next];
4093                         mddev->resync_mark_cnt = mark_cnt[next];
4094                         mark[next] = jiffies;
4095                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4096                         last_mark = next;
4097                 }
4098
4099
4100                 if (kthread_should_stop()) {
4101                         /*
4102                          * got a signal, exit.
4103                          */
4104                         printk(KERN_INFO
4105                                 "md: md_do_sync() got signal ... exiting\n");
4106                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4107                         goto out;
4108                 }
4109
4110                 /*
4111                  * this loop exits only if either when we are slower than
4112                  * the 'hard' speed limit, or the system was IO-idle for
4113                  * a jiffy.
4114                  * the system might be non-idle CPU-wise, but we only care
4115                  * about not overloading the IO subsystem. (things like an
4116                  * e2fsck being done on the RAID array should execute fast)
4117                  */
4118                 mddev->queue->unplug_fn(mddev->queue);
4119                 cond_resched();
4120
4121                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4122                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4123
4124                 if (currspeed > sysctl_speed_limit_min) {
4125                         if ((currspeed > sysctl_speed_limit_max) ||
4126                                         !is_mddev_idle(mddev)) {
4127                                 msleep(500);
4128                                 goto repeat;
4129                         }
4130                 }
4131         }
4132         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4133         /*
4134          * this also signals 'finished resyncing' to md_stop
4135          */
4136  out:
4137         mddev->queue->unplug_fn(mddev->queue);
4138
4139         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4140
4141         /* tell personality that we are finished */
4142         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4143
4144         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4145             mddev->curr_resync > 2 &&
4146             mddev->curr_resync >= mddev->recovery_cp) {
4147                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4148                         printk(KERN_INFO
4149                                 "md: checkpointing recovery of %s.\n",
4150                                 mdname(mddev));
4151                         mddev->recovery_cp = mddev->curr_resync;
4152                 } else
4153                         mddev->recovery_cp = MaxSector;
4154         }
4155
4156  skip:
4157         mddev->curr_resync = 0;
4158         wake_up(&resync_wait);
4159         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4160         md_wakeup_thread(mddev->thread);
4161 }
4162
4163
4164 /*
4165  * This routine is regularly called by all per-raid-array threads to
4166  * deal with generic issues like resync and super-block update.
4167  * Raid personalities that don't have a thread (linear/raid0) do not
4168  * need this as they never do any recovery or update the superblock.
4169  *
4170  * It does not do any resync itself, but rather "forks" off other threads
4171  * to do that as needed.
4172  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4173  * "->recovery" and create a thread at ->sync_thread.
4174  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4175  * and wakeups up this thread which will reap the thread and finish up.
4176  * This thread also removes any faulty devices (with nr_pending == 0).
4177  *
4178  * The overall approach is:
4179  *  1/ if the superblock needs updating, update it.
4180  *  2/ If a recovery thread is running, don't do anything else.
4181  *  3/ If recovery has finished, clean up, possibly marking spares active.
4182  *  4/ If there are any faulty devices, remove them.
4183  *  5/ If array is degraded, try to add spares devices
4184  *  6/ If array has spares or is not in-sync, start a resync thread.
4185  */
4186 void md_check_recovery(mddev_t *mddev)
4187 {
4188         mdk_rdev_t *rdev;
4189         struct list_head *rtmp;
4190
4191
4192         if (mddev->bitmap)
4193                 bitmap_daemon_work(mddev->bitmap);
4194
4195         if (mddev->ro)
4196                 return;
4197
4198         if (signal_pending(current)) {
4199                 if (mddev->pers->sync_request) {
4200                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4201                                mdname(mddev));
4202                         mddev->safemode = 2;
4203                 }
4204                 flush_signals(current);
4205         }
4206
4207         if ( ! (
4208                 mddev->sb_dirty ||
4209                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4210                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4211                 (mddev->safemode == 1) ||
4212                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4213                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4214                 ))
4215                 return;
4216
4217         if (mddev_trylock(mddev)==0) {
4218                 int spares =0;
4219
4220                 spin_lock_irq(&mddev->write_lock);
4221                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4222                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4223                         mddev->in_sync = 1;
4224                         mddev->sb_dirty = 1;
4225                 }
4226                 if (mddev->safemode == 1)
4227                         mddev->safemode = 0;
4228                 spin_unlock_irq(&mddev->write_lock);
4229
4230                 if (mddev->sb_dirty)
4231                         md_update_sb(mddev);
4232
4233
4234                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4235                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4236                         /* resync/recovery still happening */
4237                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4238                         goto unlock;
4239                 }
4240                 if (mddev->sync_thread) {
4241                         /* resync has finished, collect result */
4242                         md_unregister_thread(mddev->sync_thread);
4243                         mddev->sync_thread = NULL;
4244                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4245                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4246                                 /* success...*/
4247                                 /* activate any spares */
4248                                 mddev->pers->spare_active(mddev);
4249                         }
4250                         md_update_sb(mddev);
4251
4252                         /* if array is no-longer degraded, then any saved_raid_disk
4253                          * information must be scrapped
4254                          */
4255                         if (!mddev->degraded)
4256                                 ITERATE_RDEV(mddev,rdev,rtmp)
4257                                         rdev->saved_raid_disk = -1;
4258
4259                         mddev->recovery = 0;
4260                         /* flag recovery needed just to double check */
4261                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4262                         goto unlock;
4263                 }
4264                 /* Clear some bits that don't mean anything, but
4265                  * might be left set
4266                  */
4267                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4268                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4269                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4270                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4271
4272                 /* no recovery is running.
4273                  * remove any failed drives, then
4274                  * add spares if possible.
4275                  * Spare are also removed and re-added, to allow
4276                  * the personality to fail the re-add.
4277                  */
4278                 ITERATE_RDEV(mddev,rdev,rtmp)
4279                         if (rdev->raid_disk >= 0 &&
4280                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4281                             atomic_read(&rdev->nr_pending)==0) {
4282                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4283                                         char nm[20];
4284                                         sprintf(nm,"rd%d", rdev->raid_disk);
4285                                         sysfs_remove_link(&mddev->kobj, nm);
4286                                         rdev->raid_disk = -1;
4287                                 }
4288                         }
4289
4290                 if (mddev->degraded) {
4291                         ITERATE_RDEV(mddev,rdev,rtmp)
4292                                 if (rdev->raid_disk < 0
4293                                     && !test_bit(Faulty, &rdev->flags)) {
4294                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4295                                                 char nm[20];
4296                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4297                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4298                                                 spares++;
4299                                         } else
4300                                                 break;
4301                                 }
4302                 }
4303
4304                 if (spares) {
4305                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4306                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4307                 } else if (mddev->recovery_cp < MaxSector) {
4308                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4309                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4310                         /* nothing to be done ... */
4311                         goto unlock;
4312
4313                 if (mddev->pers->sync_request) {
4314                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4315                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4316                                 /* We are adding a device or devices to an array
4317                                  * which has the bitmap stored on all devices.
4318                                  * So make sure all bitmap pages get written
4319                                  */
4320                                 bitmap_write_all(mddev->bitmap);
4321                         }
4322                         mddev->sync_thread = md_register_thread(md_do_sync,
4323                                                                 mddev,
4324                                                                 "%s_resync");
4325                         if (!mddev->sync_thread) {
4326                                 printk(KERN_ERR "%s: could not start resync"
4327                                         " thread...\n",
4328                                         mdname(mddev));
4329                                 /* leave the spares where they are, it shouldn't hurt */
4330                                 mddev->recovery = 0;
4331                         } else {
4332                                 md_wakeup_thread(mddev->sync_thread);
4333                         }
4334                 }
4335         unlock:
4336                 mddev_unlock(mddev);
4337         }
4338 }
4339
4340 static int md_notify_reboot(struct notifier_block *this,
4341                             unsigned long code, void *x)
4342 {
4343         struct list_head *tmp;
4344         mddev_t *mddev;
4345
4346         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4347
4348                 printk(KERN_INFO "md: stopping all md devices.\n");
4349
4350                 ITERATE_MDDEV(mddev,tmp)
4351                         if (mddev_trylock(mddev)==0)
4352                                 do_md_stop (mddev, 1);
4353                 /*
4354                  * certain more exotic SCSI devices are known to be
4355                  * volatile wrt too early system reboots. While the
4356                  * right place to handle this issue is the given
4357                  * driver, we do want to have a safe RAID driver ...
4358                  */
4359                 mdelay(1000*1);
4360         }
4361         return NOTIFY_DONE;
4362 }
4363
4364 static struct notifier_block md_notifier = {
4365         .notifier_call  = md_notify_reboot,
4366         .next           = NULL,
4367         .priority       = INT_MAX, /* before any real devices */
4368 };
4369
4370 static void md_geninit(void)
4371 {
4372         struct proc_dir_entry *p;
4373
4374         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4375
4376         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4377         if (p)
4378                 p->proc_fops = &md_seq_fops;
4379 }
4380
4381 static int __init md_init(void)
4382 {
4383         int minor;
4384
4385         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4386                         " MD_SB_DISKS=%d\n",
4387                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4388                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4389         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4390                         BITMAP_MINOR);
4391
4392         if (register_blkdev(MAJOR_NR, "md"))
4393                 return -1;
4394         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4395                 unregister_blkdev(MAJOR_NR, "md");
4396                 return -1;
4397         }
4398         devfs_mk_dir("md");
4399         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4400                                 md_probe, NULL, NULL);
4401         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4402                             md_probe, NULL, NULL);
4403
4404         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4405                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4406                                 S_IFBLK|S_IRUSR|S_IWUSR,
4407                                 "md/%d", minor);
4408
4409         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4410                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4411                               S_IFBLK|S_IRUSR|S_IWUSR,
4412                               "md/mdp%d", minor);
4413
4414
4415         register_reboot_notifier(&md_notifier);
4416         raid_table_header = register_sysctl_table(raid_root_table, 1);
4417
4418         md_geninit();
4419         return (0);
4420 }
4421
4422
4423 #ifndef MODULE
4424
4425 /*
4426  * Searches all registered partitions for autorun RAID arrays
4427  * at boot time.
4428  */
4429 static dev_t detected_devices[128];
4430 static int dev_cnt;
4431
4432 void md_autodetect_dev(dev_t dev)
4433 {
4434         if (dev_cnt >= 0 && dev_cnt < 127)
4435                 detected_devices[dev_cnt++] = dev;
4436 }
4437
4438
4439 static void autostart_arrays(int part)
4440 {
4441         mdk_rdev_t *rdev;
4442         int i;
4443
4444         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4445
4446         for (i = 0; i < dev_cnt; i++) {
4447                 dev_t dev = detected_devices[i];
4448
4449                 rdev = md_import_device(dev,0, 0);
4450                 if (IS_ERR(rdev))
4451                         continue;
4452
4453                 if (test_bit(Faulty, &rdev->flags)) {
4454                         MD_BUG();
4455                         continue;
4456                 }
4457                 list_add(&rdev->same_set, &pending_raid_disks);
4458         }
4459         dev_cnt = 0;
4460
4461         autorun_devices(part);
4462 }
4463
4464 #endif
4465
4466 static __exit void md_exit(void)
4467 {
4468         mddev_t *mddev;
4469         struct list_head *tmp;
4470         int i;
4471         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4472         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4473         for (i=0; i < MAX_MD_DEVS; i++)
4474                 devfs_remove("md/%d", i);
4475         for (i=0; i < MAX_MD_DEVS; i++)
4476                 devfs_remove("md/d%d", i);
4477
4478         devfs_remove("md");
4479
4480         unregister_blkdev(MAJOR_NR,"md");
4481         unregister_blkdev(mdp_major, "mdp");
4482         unregister_reboot_notifier(&md_notifier);
4483         unregister_sysctl_table(raid_table_header);
4484         remove_proc_entry("mdstat", NULL);
4485         ITERATE_MDDEV(mddev,tmp) {
4486                 struct gendisk *disk = mddev->gendisk;
4487                 if (!disk)
4488                         continue;
4489                 export_array(mddev);
4490                 del_gendisk(disk);
4491                 put_disk(disk);
4492                 mddev->gendisk = NULL;
4493                 mddev_put(mddev);
4494         }
4495 }
4496
4497 module_init(md_init)
4498 module_exit(md_exit)
4499
4500 static int get_ro(char *buffer, struct kernel_param *kp)
4501 {
4502         return sprintf(buffer, "%d", start_readonly);
4503 }
4504 static int set_ro(const char *val, struct kernel_param *kp)
4505 {
4506         char *e;
4507         int num = simple_strtoul(val, &e, 10);
4508         if (*val && (*e == '\0' || *e == '\n')) {
4509                 start_readonly = num;
4510                 return 0;;
4511         }
4512         return -EINVAL;
4513 }
4514
4515 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4516 module_param(start_dirty_degraded, int, 0644);
4517
4518
4519 EXPORT_SYMBOL(register_md_personality);
4520 EXPORT_SYMBOL(unregister_md_personality);
4521 EXPORT_SYMBOL(md_error);
4522 EXPORT_SYMBOL(md_done_sync);
4523 EXPORT_SYMBOL(md_write_start);
4524 EXPORT_SYMBOL(md_write_end);
4525 EXPORT_SYMBOL(md_register_thread);
4526 EXPORT_SYMBOL(md_unregister_thread);
4527 EXPORT_SYMBOL(md_wakeup_thread);
4528 EXPORT_SYMBOL(md_print_devices);
4529 EXPORT_SYMBOL(md_check_recovery);
4530 MODULE_LICENSE("GPL");
4531 MODULE_ALIAS("md");
4532 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);