drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45
  46 #include <linux/init.h>
  47
  48 #include <linux/file.h>
  49
  50 #ifdef CONFIG_KMOD
  51 #include <linux/kmod.h>
  52 #endif
  53
  54 #include <asm/unaligned.h>
  55
  56 #define MAJOR_NR MD_MAJOR
  57 #define MD_DRIVER
  58
  59 /* 63 partitions with the alternate major number (mdp) */
  60 #define MdpMinorShift 6
  61
  62 #define DEBUG 0
  63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  64
  65
  66 #ifndef MODULE
  67 static void autostart_arrays (int part);
  68 #endif
  69
  70 static mdk_personality_t *pers[MAX_PERSONALITY];
  71 static DEFINE_SPINLOCK(pers_lock);
  72
  73 /*
  74  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  75  * is 1000 KB/sec, so the extra system load does not show up that much.
  76  * Increase it if you want to have more _guaranteed_ speed. Note that
  77  * the RAID driver will use the maximum available bandwidth if the IO
  78  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  79  * speed limit - in case reconstruction slows down your system despite
  80  * idle IO detection.
  81  *
  82  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  83  */
  84
  85 static int sysctl_speed_limit_min = 1000;
  86 static int sysctl_speed_limit_max = 200000;
  87
  88 static struct ctl_table_header *raid_table_header;
  89
  90 static ctl_table raid_table[] = {
  91         {
  92                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  93                 .procname       = "speed_limit_min",
  94                 .data           = &sysctl_speed_limit_min,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec,
  98         },
  99         {
 100                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 101                 .procname       = "speed_limit_max",
 102                 .data           = &sysctl_speed_limit_max,
 103                 .maxlen         = sizeof(int),
 104                 .mode           = 0644,
 105                 .proc_handler   = &proc_dointvec,
 106         },
 107         { .ctl_name = 0 }
 108 };
 109
 110 static ctl_table raid_dir_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID,
 113                 .procname       = "raid",
 114                 .maxlen         = 0,
 115                 .mode           = 0555,
 116                 .child          = raid_table,
 117         },
 118         { .ctl_name = 0 }
 119 };
 120
 121 static ctl_table raid_root_table[] = {
 122         {
 123                 .ctl_name       = CTL_DEV,
 124                 .procname       = "dev",
 125                 .maxlen         = 0,
 126                 .mode           = 0555,
 127                 .child          = raid_dir_table,
 128         },
 129         { .ctl_name = 0 }
 130 };
 131
 132 static struct block_device_operations md_fops;
 133
 134 static int start_readonly;
 135
 136 /*
 137  * Enables to iterate over all existing md arrays
 138  * all_mddevs_lock protects this list.
 139  */
 140 static LIST_HEAD(all_mddevs);
 141 static DEFINE_SPINLOCK(all_mddevs_lock);
 142
 143
 144 /*
 145  * iterates through all used mddevs in the system.
 146  * We take care to grab the all_mddevs_lock whenever navigating
 147  * the list, and to always hold a refcount when unlocked.
 148  * Any code which breaks out of this loop while own
 149  * a reference to the current mddev and must mddev_put it.
 150  */
 151 #define ITERATE_MDDEV(mddev,tmp)                                        \
 152                                                                         \
 153         for (({ spin_lock(&all_mddevs_lock);                            \
 154                 tmp = all_mddevs.next;                                  \
 155                 mddev = NULL;});                                        \
 156              ({ if (tmp != &all_mddevs)                                 \
 157                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 158                 spin_unlock(&all_mddevs_lock);                          \
 159                 if (mddev) mddev_put(mddev);                            \
 160                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 161                 tmp != &all_mddevs;});                                  \
 162              ({ spin_lock(&all_mddevs_lock);                            \
 163                 tmp = tmp->next;})                                      \
 164                 )
 165
 166
 167 static int md_fail_request (request_queue_t *q, struct bio *bio)
 168 {
 169         bio_io_error(bio, bio->bi_size);
 170         return 0;
 171 }
 172
 173 static inline mddev_t *mddev_get(mddev_t *mddev)
 174 {
 175         atomic_inc(&mddev->active);
 176         return mddev;
 177 }
 178
 179 static void mddev_put(mddev_t *mddev)
 180 {
 181         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 182                 return;
 183         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 184                 list_del(&mddev->all_mddevs);
 185                 blk_put_queue(mddev->queue);
 186                 kobject_unregister(&mddev->kobj);
 187         }
 188         spin_unlock(&all_mddevs_lock);
 189 }
 190
 191 static mddev_t * mddev_find(dev_t unit)
 192 {
 193         mddev_t *mddev, *new = NULL;
 194
 195  retry:
 196         spin_lock(&all_mddevs_lock);
 197         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 198                 if (mddev->unit == unit) {
 199                         mddev_get(mddev);
 200                         spin_unlock(&all_mddevs_lock);
 201                         kfree(new);
 202                         return mddev;
 203                 }
 204
 205         if (new) {
 206                 list_add(&new->all_mddevs, &all_mddevs);
 207                 spin_unlock(&all_mddevs_lock);
 208                 return new;
 209         }
 210         spin_unlock(&all_mddevs_lock);
 211
 212         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 213         if (!new)
 214                 return NULL;
 215
 216         memset(new, 0, sizeof(*new));
 217
 218         new->unit = unit;
 219         if (MAJOR(unit) == MD_MAJOR)
 220                 new->md_minor = MINOR(unit);
 221         else
 222                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 223
 224         init_MUTEX(&new->reconfig_sem);
 225         INIT_LIST_HEAD(&new->disks);
 226         INIT_LIST_HEAD(&new->all_mddevs);
 227         init_timer(&new->safemode_timer);
 228         atomic_set(&new->active, 1);
 229         spin_lock_init(&new->write_lock);
 230         init_waitqueue_head(&new->sb_wait);
 231
 232         new->queue = blk_alloc_queue(GFP_KERNEL);
 233         if (!new->queue) {
 234                 kfree(new);
 235                 return NULL;
 236         }
 237
 238         blk_queue_make_request(new->queue, md_fail_request);
 239
 240         goto retry;
 241 }
 242
 243 static inline int mddev_lock(mddev_t * mddev)
 244 {
 245         return down_interruptible(&mddev->reconfig_sem);
 246 }
 247
 248 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 249 {
 250         down(&mddev->reconfig_sem);
 251 }
 252
 253 static inline int mddev_trylock(mddev_t * mddev)
 254 {
 255         return down_trylock(&mddev->reconfig_sem);
 256 }
 257
 258 static inline void mddev_unlock(mddev_t * mddev)
 259 {
 260         up(&mddev->reconfig_sem);
 261
 262         md_wakeup_thread(mddev->thread);
 263 }
 264
 265 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 266 {
 267         mdk_rdev_t * rdev;
 268         struct list_head *tmp;
 269
 270         ITERATE_RDEV(mddev,rdev,tmp) {
 271                 if (rdev->desc_nr == nr)
 272                         return rdev;
 273         }
 274         return NULL;
 275 }
 276
 277 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 278 {
 279         struct list_head *tmp;
 280         mdk_rdev_t *rdev;
 281
 282         ITERATE_RDEV(mddev,rdev,tmp) {
 283                 if (rdev->bdev->bd_dev == dev)
 284                         return rdev;
 285         }
 286         return NULL;
 287 }
 288
 289 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 290 {
 291         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 292         return MD_NEW_SIZE_BLOCKS(size);
 293 }
 294
 295 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 296 {
 297         sector_t size;
 298
 299         size = rdev->sb_offset;
 300
 301         if (chunk_size)
 302                 size &= ~((sector_t)chunk_size/1024 - 1);
 303         return size;
 304 }
 305
 306 static int alloc_disk_sb(mdk_rdev_t * rdev)
 307 {
 308         if (rdev->sb_page)
 309                 MD_BUG();
 310
 311         rdev->sb_page = alloc_page(GFP_KERNEL);
 312         if (!rdev->sb_page) {
 313                 printk(KERN_ALERT "md: out of memory.\n");
 314                 return -EINVAL;
 315         }
 316
 317         return 0;
 318 }
 319
 320 static void free_disk_sb(mdk_rdev_t * rdev)
 321 {
 322         if (rdev->sb_page) {
 323                 page_cache_release(rdev->sb_page);
 324                 rdev->sb_loaded = 0;
 325                 rdev->sb_page = NULL;
 326                 rdev->sb_offset = 0;
 327                 rdev->size = 0;
 328         }
 329 }
 330
 331
 332 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 333 {
 334         mdk_rdev_t *rdev = bio->bi_private;
 335         mddev_t *mddev = rdev->mddev;
 336         if (bio->bi_size)
 337                 return 1;
 338
 339         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 340                 md_error(mddev, rdev);
 341
 342         if (atomic_dec_and_test(&mddev->pending_writes))
 343                 wake_up(&mddev->sb_wait);
 344         bio_put(bio);
 345         return 0;
 346 }
 347
 348 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 349 {
 350         struct bio *bio2 = bio->bi_private;
 351         mdk_rdev_t *rdev = bio2->bi_private;
 352         mddev_t *mddev = rdev->mddev;
 353         if (bio->bi_size)
 354                 return 1;
 355
 356         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 357             error == -EOPNOTSUPP) {
 358                 unsigned long flags;
 359                 /* barriers don't appear to be supported :-( */
 360                 set_bit(BarriersNotsupp, &rdev->flags);
 361                 mddev->barriers_work = 0;
 362                 spin_lock_irqsave(&mddev->write_lock, flags);
 363                 bio2->bi_next = mddev->biolist;
 364                 mddev->biolist = bio2;
 365                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 366                 wake_up(&mddev->sb_wait);
 367                 bio_put(bio);
 368                 return 0;
 369         }
 370         bio_put(bio2);
 371         bio->bi_private = rdev;
 372         return super_written(bio, bytes_done, error);
 373 }
 374
 375 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 376                    sector_t sector, int size, struct page *page)
 377 {
 378         /* write first size bytes of page to sector of rdev
 379          * Increment mddev->pending_writes before returning
 380          * and decrement it on completion, waking up sb_wait
 381          * if zero is reached.
 382          * If an error occurred, call md_error
 383          *
 384          * As we might need to resubmit the request if BIO_RW_BARRIER
 385          * causes ENOTSUPP, we allocate a spare bio...
 386          */
 387         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 388         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 389
 390         bio->bi_bdev = rdev->bdev;
 391         bio->bi_sector = sector;
 392         bio_add_page(bio, page, size, 0);
 393         bio->bi_private = rdev;
 394         bio->bi_end_io = super_written;
 395         bio->bi_rw = rw;
 396
 397         atomic_inc(&mddev->pending_writes);
 398         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 399                 struct bio *rbio;
 400                 rw |= (1<<BIO_RW_BARRIER);
 401                 rbio = bio_clone(bio, GFP_NOIO);
 402                 rbio->bi_private = bio;
 403                 rbio->bi_end_io = super_written_barrier;
 404                 submit_bio(rw, rbio);
 405         } else
 406                 submit_bio(rw, bio);
 407 }
 408
 409 void md_super_wait(mddev_t *mddev)
 410 {
 411         /* wait for all superblock writes that were scheduled to complete.
 412          * if any had to be retried (due to BARRIER problems), retry them
 413          */
 414         DEFINE_WAIT(wq);
 415         for(;;) {
 416                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 417                 if (atomic_read(&mddev->pending_writes)==0)
 418                         break;
 419                 while (mddev->biolist) {
 420                         struct bio *bio;
 421                         spin_lock_irq(&mddev->write_lock);
 422                         bio = mddev->biolist;
 423                         mddev->biolist = bio->bi_next ;
 424                         bio->bi_next = NULL;
 425                         spin_unlock_irq(&mddev->write_lock);
 426                         submit_bio(bio->bi_rw, bio);
 427                 }
 428                 schedule();
 429         }
 430         finish_wait(&mddev->sb_wait, &wq);
 431 }
 432
 433 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 434 {
 435         if (bio->bi_size)
 436                 return 1;
 437
 438         complete((struct completion*)bio->bi_private);
 439         return 0;
 440 }
 441
 442 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 443                    struct page *page, int rw)
 444 {
 445         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 446         struct completion event;
 447         int ret;
 448
 449         rw |= (1 << BIO_RW_SYNC);
 450
 451         bio->bi_bdev = bdev;
 452         bio->bi_sector = sector;
 453         bio_add_page(bio, page, size, 0);
 454         init_completion(&event);
 455         bio->bi_private = &event;
 456         bio->bi_end_io = bi_complete;
 457         submit_bio(rw, bio);
 458         wait_for_completion(&event);
 459
 460         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 461         bio_put(bio);
 462         return ret;
 463 }
 464
 465 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 466 {
 467         char b[BDEVNAME_SIZE];
 468         if (!rdev->sb_page) {
 469                 MD_BUG();
 470                 return -EINVAL;
 471         }
 472         if (rdev->sb_loaded)
 473                 return 0;
 474
 475
 476         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 477                 goto fail;
 478         rdev->sb_loaded = 1;
 479         return 0;
 480
 481 fail:
 482         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 483                 bdevname(rdev->bdev,b));
 484         return -EINVAL;
 485 }
 486
 487 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 488 {
 489         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 490                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 491                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 492                 (sb1->set_uuid3 == sb2->set_uuid3))
 493
 494                 return 1;
 495
 496         return 0;
 497 }
 498
 499
 500 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 501 {
 502         int ret;
 503         mdp_super_t *tmp1, *tmp2;
 504
 505         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 506         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 507
 508         if (!tmp1 || !tmp2) {
 509                 ret = 0;
 510                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 511                 goto abort;
 512         }
 513
 514         *tmp1 = *sb1;
 515         *tmp2 = *sb2;
 516
 517         /*
 518          * nr_disks is not constant
 519          */
 520         tmp1->nr_disks = 0;
 521         tmp2->nr_disks = 0;
 522
 523         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 524                 ret = 0;
 525         else
 526                 ret = 1;
 527
 528 abort:
 529         kfree(tmp1);
 530         kfree(tmp2);
 531         return ret;
 532 }
 533
 534 static unsigned int calc_sb_csum(mdp_super_t * sb)
 535 {
 536         unsigned int disk_csum, csum;
 537
 538         disk_csum = sb->sb_csum;
 539         sb->sb_csum = 0;
 540         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 541         sb->sb_csum = disk_csum;
 542         return csum;
 543 }
 544
 545
 546 /*
 547  * Handle superblock details.
 548  * We want to be able to handle multiple superblock formats
 549  * so we have a common interface to them all, and an array of
 550  * different handlers.
 551  * We rely on user-space to write the initial superblock, and support
 552  * reading and updating of superblocks.
 553  * Interface methods are:
 554  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 555  *      loads and validates a superblock on dev.
 556  *      if refdev != NULL, compare superblocks on both devices
 557  *    Return:
 558  *      0 - dev has a superblock that is compatible with refdev
 559  *      1 - dev has a superblock that is compatible and newer than refdev
 560  *          so dev should be used as the refdev in future
 561  *     -EINVAL superblock incompatible or invalid
 562  *     -othererror e.g. -EIO
 563  *
 564  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 565  *      Verify that dev is acceptable into mddev.
 566  *       The first time, mddev->raid_disks will be 0, and data from
 567  *       dev should be merged in.  Subsequent calls check that dev
 568  *       is new enough.  Return 0 or -EINVAL
 569  *
 570  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 571  *     Update the superblock for rdev with data in mddev
 572  *     This does not write to disc.
 573  *
 574  */
 575
 576 struct super_type  {
 577         char            *name;
 578         struct module   *owner;
 579         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 580         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 581         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 582 };
 583
 584 /*
 585  * load_super for 0.90.0
 586  */
 587 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 588 {
 589         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 590         mdp_super_t *sb;
 591         int ret;
 592         sector_t sb_offset;
 593
 594         /*
 595          * Calculate the position of the superblock,
 596          * it's at the end of the disk.
 597          *
 598          * It also happens to be a multiple of 4Kb.
 599          */
 600         sb_offset = calc_dev_sboffset(rdev->bdev);
 601         rdev->sb_offset = sb_offset;
 602
 603         ret = read_disk_sb(rdev, MD_SB_BYTES);
 604         if (ret) return ret;
 605
 606         ret = -EINVAL;
 607
 608         bdevname(rdev->bdev, b);
 609         sb = (mdp_super_t*)page_address(rdev->sb_page);
 610
 611         if (sb->md_magic != MD_SB_MAGIC) {
 612                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 613                        b);
 614                 goto abort;
 615         }
 616
 617         if (sb->major_version != 0 ||
 618             sb->minor_version != 90) {
 619                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 620                         sb->major_version, sb->minor_version,
 621                         b);
 622                 goto abort;
 623         }
 624
 625         if (sb->raid_disks <= 0)
 626                 goto abort;
 627
 628         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 629                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 630                         b);
 631                 goto abort;
 632         }
 633
 634         rdev->preferred_minor = sb->md_minor;
 635         rdev->data_offset = 0;
 636         rdev->sb_size = MD_SB_BYTES;
 637
 638         if (sb->level == LEVEL_MULTIPATH)
 639                 rdev->desc_nr = -1;
 640         else
 641                 rdev->desc_nr = sb->this_disk.number;
 642
 643         if (refdev == 0)
 644                 ret = 1;
 645         else {
 646                 __u64 ev1, ev2;
 647                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 648                 if (!uuid_equal(refsb, sb)) {
 649                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 650                                 b, bdevname(refdev->bdev,b2));
 651                         goto abort;
 652                 }
 653                 if (!sb_equal(refsb, sb)) {
 654                         printk(KERN_WARNING "md: %s has same UUID"
 655                                " but different superblock to %s\n",
 656                                b, bdevname(refdev->bdev, b2));
 657                         goto abort;
 658                 }
 659                 ev1 = md_event(sb);
 660                 ev2 = md_event(refsb);
 661                 if (ev1 > ev2)
 662                         ret = 1;
 663                 else
 664                         ret = 0;
 665         }
 666         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 667
 668  abort:
 669         return ret;
 670 }
 671
 672 /*
 673  * validate_super for 0.90.0
 674  */
 675 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 676 {
 677         mdp_disk_t *desc;
 678         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 679
 680         rdev->raid_disk = -1;
 681         rdev->flags = 0;
 682         if (mddev->raid_disks == 0) {
 683                 mddev->major_version = 0;
 684                 mddev->minor_version = sb->minor_version;
 685                 mddev->patch_version = sb->patch_version;
 686                 mddev->persistent = ! sb->not_persistent;
 687                 mddev->chunk_size = sb->chunk_size;
 688                 mddev->ctime = sb->ctime;
 689                 mddev->utime = sb->utime;
 690                 mddev->level = sb->level;
 691                 mddev->layout = sb->layout;
 692                 mddev->raid_disks = sb->raid_disks;
 693                 mddev->size = sb->size;
 694                 mddev->events = md_event(sb);
 695                 mddev->bitmap_offset = 0;
 696                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 697
 698                 if (sb->state & (1<<MD_SB_CLEAN))
 699                         mddev->recovery_cp = MaxSector;
 700                 else {
 701                         if (sb->events_hi == sb->cp_events_hi &&
 702                                 sb->events_lo == sb->cp_events_lo) {
 703                                 mddev->recovery_cp = sb->recovery_cp;
 704                         } else
 705                                 mddev->recovery_cp = 0;
 706                 }
 707
 708                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 709                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 710                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 711                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 712
 713                 mddev->max_disks = MD_SB_DISKS;
 714
 715                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 716                     mddev->bitmap_file == NULL) {
 717                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
 718                                 /* FIXME use a better test */
 719                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
 720                                 return -EINVAL;
 721                         }
 722                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 723                 }
 724
 725         } else if (mddev->pers == NULL) {
 726                 /* Insist on good event counter while assembling */
 727                 __u64 ev1 = md_event(sb);
 728                 ++ev1;
 729                 if (ev1 < mddev->events)
 730                         return -EINVAL;
 731         } else if (mddev->bitmap) {
 732                 /* if adding to array with a bitmap, then we can accept an
 733                  * older device ... but not too old.
 734                  */
 735                 __u64 ev1 = md_event(sb);
 736                 if (ev1 < mddev->bitmap->events_cleared)
 737                         return 0;
 738         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 739                 return 0;
 740
 741         if (mddev->level != LEVEL_MULTIPATH) {
 742                 desc = sb->disks + rdev->desc_nr;
 743
 744                 if (desc->state & (1<<MD_DISK_FAULTY))
 745                         set_bit(Faulty, &rdev->flags);
 746                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 747                          desc->raid_disk < mddev->raid_disks) {
 748                         set_bit(In_sync, &rdev->flags);
 749                         rdev->raid_disk = desc->raid_disk;
 750                 }
 751                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 752                         set_bit(WriteMostly, &rdev->flags);
 753         } else /* MULTIPATH are always insync */
 754                 set_bit(In_sync, &rdev->flags);
 755         return 0;
 756 }
 757
 758 /*
 759  * sync_super for 0.90.0
 760  */
 761 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 762 {
 763         mdp_super_t *sb;
 764         struct list_head *tmp;
 765         mdk_rdev_t *rdev2;
 766         int next_spare = mddev->raid_disks;
 767
 768
 769         /* make rdev->sb match mddev data..
 770          *
 771          * 1/ zero out disks
 772          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 773          * 3/ any empty disks < next_spare become removed
 774          *
 775          * disks[0] gets initialised to REMOVED because
 776          * we cannot be sure from other fields if it has
 777          * been initialised or not.
 778          */
 779         int i;
 780         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 781
 782         rdev->sb_size = MD_SB_BYTES;
 783
 784         sb = (mdp_super_t*)page_address(rdev->sb_page);
 785
 786         memset(sb, 0, sizeof(*sb));
 787
 788         sb->md_magic = MD_SB_MAGIC;
 789         sb->major_version = mddev->major_version;
 790         sb->minor_version = mddev->minor_version;
 791         sb->patch_version = mddev->patch_version;
 792         sb->gvalid_words  = 0; /* ignored */
 793         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 794         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 795         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 796         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 797
 798         sb->ctime = mddev->ctime;
 799         sb->level = mddev->level;
 800         sb->size  = mddev->size;
 801         sb->raid_disks = mddev->raid_disks;
 802         sb->md_minor = mddev->md_minor;
 803         sb->not_persistent = !mddev->persistent;
 804         sb->utime = mddev->utime;
 805         sb->state = 0;
 806         sb->events_hi = (mddev->events>>32);
 807         sb->events_lo = (u32)mddev->events;
 808
 809         if (mddev->in_sync)
 810         {
 811                 sb->recovery_cp = mddev->recovery_cp;
 812                 sb->cp_events_hi = (mddev->events>>32);
 813                 sb->cp_events_lo = (u32)mddev->events;
 814                 if (mddev->recovery_cp == MaxSector)
 815                         sb->state = (1<< MD_SB_CLEAN);
 816         } else
 817                 sb->recovery_cp = 0;
 818
 819         sb->layout = mddev->layout;
 820         sb->chunk_size = mddev->chunk_size;
 821
 822         if (mddev->bitmap && mddev->bitmap_file == NULL)
 823                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 824
 825         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 826         ITERATE_RDEV(mddev,rdev2,tmp) {
 827                 mdp_disk_t *d;
 828                 int desc_nr;
 829                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 830                     && !test_bit(Faulty, &rdev2->flags))
 831                         desc_nr = rdev2->raid_disk;
 832                 else
 833                         desc_nr = next_spare++;
 834                 rdev2->desc_nr = desc_nr;
 835                 d = &sb->disks[rdev2->desc_nr];
 836                 nr_disks++;
 837                 d->number = rdev2->desc_nr;
 838                 d->major = MAJOR(rdev2->bdev->bd_dev);
 839                 d->minor = MINOR(rdev2->bdev->bd_dev);
 840                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 841                     && !test_bit(Faulty, &rdev2->flags))
 842                         d->raid_disk = rdev2->raid_disk;
 843                 else
 844                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 845                 if (test_bit(Faulty, &rdev2->flags)) {
 846                         d->state = (1<<MD_DISK_FAULTY);
 847                         failed++;
 848                 } else if (test_bit(In_sync, &rdev2->flags)) {
 849                         d->state = (1<<MD_DISK_ACTIVE);
 850                         d->state |= (1<<MD_DISK_SYNC);
 851                         active++;
 852                         working++;
 853                 } else {
 854                         d->state = 0;
 855                         spare++;
 856                         working++;
 857                 }
 858                 if (test_bit(WriteMostly, &rdev2->flags))
 859                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 860         }
 861         /* now set the "removed" and "faulty" bits on any missing devices */
 862         for (i=0 ; i < mddev->raid_disks ; i++) {
 863                 mdp_disk_t *d = &sb->disks[i];
 864                 if (d->state == 0 && d->number == 0) {
 865                         d->number = i;
 866                         d->raid_disk = i;
 867                         d->state = (1<<MD_DISK_REMOVED);
 868                         d->state |= (1<<MD_DISK_FAULTY);
 869                         failed++;
 870                 }
 871         }
 872         sb->nr_disks = nr_disks;
 873         sb->active_disks = active;
 874         sb->working_disks = working;
 875         sb->failed_disks = failed;
 876         sb->spare_disks = spare;
 877
 878         sb->this_disk = sb->disks[rdev->desc_nr];
 879         sb->sb_csum = calc_sb_csum(sb);
 880 }
 881
 882 /*
 883  * version 1 superblock
 884  */
 885
 886 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 887 {
 888         unsigned int disk_csum, csum;
 889         unsigned long long newcsum;
 890         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 891         unsigned int *isuper = (unsigned int*)sb;
 892         int i;
 893
 894         disk_csum = sb->sb_csum;
 895         sb->sb_csum = 0;
 896         newcsum = 0;
 897         for (i=0; size>=4; size -= 4 )
 898                 newcsum += le32_to_cpu(*isuper++);
 899
 900         if (size == 2)
 901                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 902
 903         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 904         sb->sb_csum = disk_csum;
 905         return cpu_to_le32(csum);
 906 }
 907
 908 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 909 {
 910         struct mdp_superblock_1 *sb;
 911         int ret;
 912         sector_t sb_offset;
 913         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 914         int bmask;
 915
 916         /*
 917          * Calculate the position of the superblock.
 918          * It is always aligned to a 4K boundary and
 919          * depeding on minor_version, it can be:
 920          * 0: At least 8K, but less than 12K, from end of device
 921          * 1: At start of device
 922          * 2: 4K from start of device.
 923          */
 924         switch(minor_version) {
 925         case 0:
 926                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 927                 sb_offset -= 8*2;
 928                 sb_offset &= ~(sector_t)(4*2-1);
 929                 /* convert from sectors to K */
 930                 sb_offset /= 2;
 931                 break;
 932         case 1:
 933                 sb_offset = 0;
 934                 break;
 935         case 2:
 936                 sb_offset = 4;
 937                 break;
 938         default:
 939                 return -EINVAL;
 940         }
 941         rdev->sb_offset = sb_offset;
 942
 943         /* superblock is rarely larger than 1K, but it can be larger,
 944          * and it is safe to read 4k, so we do that
 945          */
 946         ret = read_disk_sb(rdev, 4096);
 947         if (ret) return ret;
 948
 949
 950         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 951
 952         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 953             sb->major_version != cpu_to_le32(1) ||
 954             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 955             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 956             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 957                 return -EINVAL;
 958
 959         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 960                 printk("md: invalid superblock checksum on %s\n",
 961                         bdevname(rdev->bdev,b));
 962                 return -EINVAL;
 963         }
 964         if (le64_to_cpu(sb->data_size) < 10) {
 965                 printk("md: data_size too small on %s\n",
 966                        bdevname(rdev->bdev,b));
 967                 return -EINVAL;
 968         }
 969         rdev->preferred_minor = 0xffff;
 970         rdev->data_offset = le64_to_cpu(sb->data_offset);
 971
 972         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 973         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 974         if (rdev->sb_size & bmask)
 975                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 976
 977         if (refdev == 0)
 978                 return 1;
 979         else {
 980                 __u64 ev1, ev2;
 981                 struct mdp_superblock_1 *refsb =
 982                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 983
 984                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 985                     sb->level != refsb->level ||
 986                     sb->layout != refsb->layout ||
 987                     sb->chunksize != refsb->chunksize) {
 988                         printk(KERN_WARNING "md: %s has strangely different"
 989                                 " superblock to %s\n",
 990                                 bdevname(rdev->bdev,b),
 991                                 bdevname(refdev->bdev,b2));
 992                         return -EINVAL;
 993                 }
 994                 ev1 = le64_to_cpu(sb->events);
 995                 ev2 = le64_to_cpu(refsb->events);
 996
 997                 if (ev1 > ev2)
 998                         return 1;
 999         }
1000         if (minor_version)
1001                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1002         else
1003                 rdev->size = rdev->sb_offset;
1004         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1005                 return -EINVAL;
1006         rdev->size = le64_to_cpu(sb->data_size)/2;
1007         if (le32_to_cpu(sb->chunksize))
1008                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1009         return 0;
1010 }
1011
1012 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1013 {
1014         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1015
1016         rdev->raid_disk = -1;
1017         rdev->flags = 0;
1018         if (mddev->raid_disks == 0) {
1019                 mddev->major_version = 1;
1020                 mddev->patch_version = 0;
1021                 mddev->persistent = 1;
1022                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1023                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1024                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1025                 mddev->level = le32_to_cpu(sb->level);
1026                 mddev->layout = le32_to_cpu(sb->layout);
1027                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1028                 mddev->size = le64_to_cpu(sb->size)/2;
1029                 mddev->events = le64_to_cpu(sb->events);
1030                 mddev->bitmap_offset = 0;
1031                 mddev->default_bitmap_offset = 0;
1032                 mddev->default_bitmap_offset = 1024;
1033
1034                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1035                 memcpy(mddev->uuid, sb->set_uuid, 16);
1036
1037                 mddev->max_disks =  (4096-256)/2;
1038
1039                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1040                     mddev->bitmap_file == NULL ) {
1041                         if (mddev->level != 1) {
1042                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1043                                 return -EINVAL;
1044                         }
1045                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1046                 }
1047         } else if (mddev->pers == NULL) {
1048                 /* Insist of good event counter while assembling */
1049                 __u64 ev1 = le64_to_cpu(sb->events);
1050                 ++ev1;
1051                 if (ev1 < mddev->events)
1052                         return -EINVAL;
1053         } else if (mddev->bitmap) {
1054                 /* If adding to array with a bitmap, then we can accept an
1055                  * older device, but not too old.
1056                  */
1057                 __u64 ev1 = le64_to_cpu(sb->events);
1058                 if (ev1 < mddev->bitmap->events_cleared)
1059                         return 0;
1060         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1061                 return 0;
1062
1063         if (mddev->level != LEVEL_MULTIPATH) {
1064                 int role;
1065                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1066                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1067                 switch(role) {
1068                 case 0xffff: /* spare */
1069                         break;
1070                 case 0xfffe: /* faulty */
1071                         set_bit(Faulty, &rdev->flags);
1072                         break;
1073                 default:
1074                         set_bit(In_sync, &rdev->flags);
1075                         rdev->raid_disk = role;
1076                         break;
1077                 }
1078                 if (sb->devflags & WriteMostly1)
1079                         set_bit(WriteMostly, &rdev->flags);
1080         } else /* MULTIPATH are always insync */
1081                 set_bit(In_sync, &rdev->flags);
1082
1083         return 0;
1084 }
1085
1086 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1087 {
1088         struct mdp_superblock_1 *sb;
1089         struct list_head *tmp;
1090         mdk_rdev_t *rdev2;
1091         int max_dev, i;
1092         /* make rdev->sb match mddev and rdev data. */
1093
1094         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1095
1096         sb->feature_map = 0;
1097         sb->pad0 = 0;
1098         memset(sb->pad1, 0, sizeof(sb->pad1));
1099         memset(sb->pad2, 0, sizeof(sb->pad2));
1100         memset(sb->pad3, 0, sizeof(sb->pad3));
1101
1102         sb->utime = cpu_to_le64((__u64)mddev->utime);
1103         sb->events = cpu_to_le64(mddev->events);
1104         if (mddev->in_sync)
1105                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1106         else
1107                 sb->resync_offset = cpu_to_le64(0);
1108
1109         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1110                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1111                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1112         }
1113
1114         max_dev = 0;
1115         ITERATE_RDEV(mddev,rdev2,tmp)
1116                 if (rdev2->desc_nr+1 > max_dev)
1117                         max_dev = rdev2->desc_nr+1;
1118
1119         sb->max_dev = cpu_to_le32(max_dev);
1120         for (i=0; i<max_dev;i++)
1121                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1122
1123         ITERATE_RDEV(mddev,rdev2,tmp) {
1124                 i = rdev2->desc_nr;
1125                 if (test_bit(Faulty, &rdev2->flags))
1126                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1127                 else if (test_bit(In_sync, &rdev2->flags))
1128                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1129                 else
1130                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1131         }
1132
1133         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1134         sb->sb_csum = calc_sb_1_csum(sb);
1135 }
1136
1137
1138 static struct super_type super_types[] = {
1139         [0] = {
1140                 .name   = "0.90.0",
1141                 .owner  = THIS_MODULE,
1142                 .load_super     = super_90_load,
1143                 .validate_super = super_90_validate,
1144                 .sync_super     = super_90_sync,
1145         },
1146         [1] = {
1147                 .name   = "md-1",
1148                 .owner  = THIS_MODULE,
1149                 .load_super     = super_1_load,
1150                 .validate_super = super_1_validate,
1151                 .sync_super     = super_1_sync,
1152         },
1153 };
1154
1155 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1156 {
1157         struct list_head *tmp;
1158         mdk_rdev_t *rdev;
1159
1160         ITERATE_RDEV(mddev,rdev,tmp)
1161                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1162                         return rdev;
1163
1164         return NULL;
1165 }
1166
1167 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1168 {
1169         struct list_head *tmp;
1170         mdk_rdev_t *rdev;
1171
1172         ITERATE_RDEV(mddev1,rdev,tmp)
1173                 if (match_dev_unit(mddev2, rdev))
1174                         return 1;
1175
1176         return 0;
1177 }
1178
1179 static LIST_HEAD(pending_raid_disks);
1180
1181 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1182 {
1183         mdk_rdev_t *same_pdev;
1184         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1185         struct kobject *ko;
1186
1187         if (rdev->mddev) {
1188                 MD_BUG();
1189                 return -EINVAL;
1190         }
1191         same_pdev = match_dev_unit(mddev, rdev);
1192         if (same_pdev)
1193                 printk(KERN_WARNING
1194                         "%s: WARNING: %s appears to be on the same physical"
1195                         " disk as %s. True\n     protection against single-disk"
1196                         " failure might be compromised.\n",
1197                         mdname(mddev), bdevname(rdev->bdev,b),
1198                         bdevname(same_pdev->bdev,b2));
1199
1200         /* Verify rdev->desc_nr is unique.
1201          * If it is -1, assign a free number, else
1202          * check number is not in use
1203          */
1204         if (rdev->desc_nr < 0) {
1205                 int choice = 0;
1206                 if (mddev->pers) choice = mddev->raid_disks;
1207                 while (find_rdev_nr(mddev, choice))
1208                         choice++;
1209                 rdev->desc_nr = choice;
1210         } else {
1211                 if (find_rdev_nr(mddev, rdev->desc_nr))
1212                         return -EBUSY;
1213         }
1214         bdevname(rdev->bdev,b);
1215         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1216                 return -ENOMEM;
1217
1218         list_add(&rdev->same_set, &mddev->disks);
1219         rdev->mddev = mddev;
1220         printk(KERN_INFO "md: bind<%s>\n", b);
1221
1222         rdev->kobj.parent = &mddev->kobj;
1223         kobject_add(&rdev->kobj);
1224
1225         if (rdev->bdev->bd_part)
1226                 ko = &rdev->bdev->bd_part->kobj;
1227         else
1228                 ko = &rdev->bdev->bd_disk->kobj;
1229         sysfs_create_link(&rdev->kobj, ko, "block");
1230         return 0;
1231 }
1232
1233 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1234 {
1235         char b[BDEVNAME_SIZE];
1236         if (!rdev->mddev) {
1237                 MD_BUG();
1238                 return;
1239         }
1240         list_del_init(&rdev->same_set);
1241         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1242         rdev->mddev = NULL;
1243         sysfs_remove_link(&rdev->kobj, "block");
1244         kobject_del(&rdev->kobj);
1245 }
1246
1247 /*
1248  * prevent the device from being mounted, repartitioned or
1249  * otherwise reused by a RAID array (or any other kernel
1250  * subsystem), by bd_claiming the device.
1251  */
1252 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1253 {
1254         int err = 0;
1255         struct block_device *bdev;
1256         char b[BDEVNAME_SIZE];
1257
1258         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1259         if (IS_ERR(bdev)) {
1260                 printk(KERN_ERR "md: could not open %s.\n",
1261                         __bdevname(dev, b));
1262                 return PTR_ERR(bdev);
1263         }
1264         err = bd_claim(bdev, rdev);
1265         if (err) {
1266                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1267                         bdevname(bdev, b));
1268                 blkdev_put(bdev);
1269                 return err;
1270         }
1271         rdev->bdev = bdev;
1272         return err;
1273 }
1274
1275 static void unlock_rdev(mdk_rdev_t *rdev)
1276 {
1277         struct block_device *bdev = rdev->bdev;
1278         rdev->bdev = NULL;
1279         if (!bdev)
1280                 MD_BUG();
1281         bd_release(bdev);
1282         blkdev_put(bdev);
1283 }
1284
1285 void md_autodetect_dev(dev_t dev);
1286
1287 static void export_rdev(mdk_rdev_t * rdev)
1288 {
1289         char b[BDEVNAME_SIZE];
1290         printk(KERN_INFO "md: export_rdev(%s)\n",
1291                 bdevname(rdev->bdev,b));
1292         if (rdev->mddev)
1293                 MD_BUG();
1294         free_disk_sb(rdev);
1295         list_del_init(&rdev->same_set);
1296 #ifndef MODULE
1297         md_autodetect_dev(rdev->bdev->bd_dev);
1298 #endif
1299         unlock_rdev(rdev);
1300         kobject_put(&rdev->kobj);
1301 }
1302
1303 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1304 {
1305         unbind_rdev_from_array(rdev);
1306         export_rdev(rdev);
1307 }
1308
1309 static void export_array(mddev_t *mddev)
1310 {
1311         struct list_head *tmp;
1312         mdk_rdev_t *rdev;
1313
1314         ITERATE_RDEV(mddev,rdev,tmp) {
1315                 if (!rdev->mddev) {
1316                         MD_BUG();
1317                         continue;
1318                 }
1319                 kick_rdev_from_array(rdev);
1320         }
1321         if (!list_empty(&mddev->disks))
1322                 MD_BUG();
1323         mddev->raid_disks = 0;
1324         mddev->major_version = 0;
1325 }
1326
1327 static void print_desc(mdp_disk_t *desc)
1328 {
1329         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1330                 desc->major,desc->minor,desc->raid_disk,desc->state);
1331 }
1332
1333 static void print_sb(mdp_super_t *sb)
1334 {
1335         int i;
1336
1337         printk(KERN_INFO
1338                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1339                 sb->major_version, sb->minor_version, sb->patch_version,
1340                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1341                 sb->ctime);
1342         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1343                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1344                 sb->md_minor, sb->layout, sb->chunk_size);
1345         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1346                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1347                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1348                 sb->failed_disks, sb->spare_disks,
1349                 sb->sb_csum, (unsigned long)sb->events_lo);
1350
1351         printk(KERN_INFO);
1352         for (i = 0; i < MD_SB_DISKS; i++) {
1353                 mdp_disk_t *desc;
1354
1355                 desc = sb->disks + i;
1356                 if (desc->number || desc->major || desc->minor ||
1357                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1358                         printk("     D %2d: ", i);
1359                         print_desc(desc);
1360                 }
1361         }
1362         printk(KERN_INFO "md:     THIS: ");
1363         print_desc(&sb->this_disk);
1364
1365 }
1366
1367 static void print_rdev(mdk_rdev_t *rdev)
1368 {
1369         char b[BDEVNAME_SIZE];
1370         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1371                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1372                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1373                 rdev->desc_nr);
1374         if (rdev->sb_loaded) {
1375                 printk(KERN_INFO "md: rdev superblock:\n");
1376                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1377         } else
1378                 printk(KERN_INFO "md: no rdev superblock!\n");
1379 }
1380
1381 void md_print_devices(void)
1382 {
1383         struct list_head *tmp, *tmp2;
1384         mdk_rdev_t *rdev;
1385         mddev_t *mddev;
1386         char b[BDEVNAME_SIZE];
1387
1388         printk("\n");
1389         printk("md:     **********************************\n");
1390         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1391         printk("md:     **********************************\n");
1392         ITERATE_MDDEV(mddev,tmp) {
1393
1394                 if (mddev->bitmap)
1395                         bitmap_print_sb(mddev->bitmap);
1396                 else
1397                         printk("%s: ", mdname(mddev));
1398                 ITERATE_RDEV(mddev,rdev,tmp2)
1399                         printk("<%s>", bdevname(rdev->bdev,b));
1400                 printk("\n");
1401
1402                 ITERATE_RDEV(mddev,rdev,tmp2)
1403                         print_rdev(rdev);
1404         }
1405         printk("md:     **********************************\n");
1406         printk("\n");
1407 }
1408
1409
1410 static void sync_sbs(mddev_t * mddev)
1411 {
1412         mdk_rdev_t *rdev;
1413         struct list_head *tmp;
1414
1415         ITERATE_RDEV(mddev,rdev,tmp) {
1416                 super_types[mddev->major_version].
1417                         sync_super(mddev, rdev);
1418                 rdev->sb_loaded = 1;
1419         }
1420 }
1421
1422 static void md_update_sb(mddev_t * mddev)
1423 {
1424         int err;
1425         struct list_head *tmp;
1426         mdk_rdev_t *rdev;
1427         int sync_req;
1428
1429 repeat:
1430         spin_lock_irq(&mddev->write_lock);
1431         sync_req = mddev->in_sync;
1432         mddev->utime = get_seconds();
1433         mddev->events ++;
1434
1435         if (!mddev->events) {
1436                 /*
1437                  * oops, this 64-bit counter should never wrap.
1438                  * Either we are in around ~1 trillion A.C., assuming
1439                  * 1 reboot per second, or we have a bug:
1440                  */
1441                 MD_BUG();
1442                 mddev->events --;
1443         }
1444         mddev->sb_dirty = 2;
1445         sync_sbs(mddev);
1446
1447         /*
1448          * do not write anything to disk if using
1449          * nonpersistent superblocks
1450          */
1451         if (!mddev->persistent) {
1452                 mddev->sb_dirty = 0;
1453                 spin_unlock_irq(&mddev->write_lock);
1454                 wake_up(&mddev->sb_wait);
1455                 return;
1456         }
1457         spin_unlock_irq(&mddev->write_lock);
1458
1459         dprintk(KERN_INFO
1460                 "md: updating %s RAID superblock on device (in sync %d)\n",
1461                 mdname(mddev),mddev->in_sync);
1462
1463         err = bitmap_update_sb(mddev->bitmap);
1464         ITERATE_RDEV(mddev,rdev,tmp) {
1465                 char b[BDEVNAME_SIZE];
1466                 dprintk(KERN_INFO "md: ");
1467                 if (test_bit(Faulty, &rdev->flags))
1468                         dprintk("(skipping faulty ");
1469
1470                 dprintk("%s ", bdevname(rdev->bdev,b));
1471                 if (!test_bit(Faulty, &rdev->flags)) {
1472                         md_super_write(mddev,rdev,
1473                                        rdev->sb_offset<<1, rdev->sb_size,
1474                                        rdev->sb_page);
1475                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1476                                 bdevname(rdev->bdev,b),
1477                                 (unsigned long long)rdev->sb_offset);
1478
1479                 } else
1480                         dprintk(")\n");
1481                 if (mddev->level == LEVEL_MULTIPATH)
1482                         /* only need to write one superblock... */
1483                         break;
1484         }
1485         md_super_wait(mddev);
1486         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1487
1488         spin_lock_irq(&mddev->write_lock);
1489         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1490                 /* have to write it out again */
1491                 spin_unlock_irq(&mddev->write_lock);
1492                 goto repeat;
1493         }
1494         mddev->sb_dirty = 0;
1495         spin_unlock_irq(&mddev->write_lock);
1496         wake_up(&mddev->sb_wait);
1497
1498 }
1499
1500 struct rdev_sysfs_entry {
1501         struct attribute attr;
1502         ssize_t (*show)(mdk_rdev_t *, char *);
1503         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1504 };
1505
1506 static ssize_t
1507 state_show(mdk_rdev_t *rdev, char *page)
1508 {
1509         char *sep = "";
1510         int len=0;
1511
1512         if (test_bit(Faulty, &rdev->flags)) {
1513                 len+= sprintf(page+len, "%sfaulty",sep);
1514                 sep = ",";
1515         }
1516         if (test_bit(In_sync, &rdev->flags)) {
1517                 len += sprintf(page+len, "%sin_sync",sep);
1518                 sep = ",";
1519         }
1520         if (!test_bit(Faulty, &rdev->flags) &&
1521             !test_bit(In_sync, &rdev->flags)) {
1522                 len += sprintf(page+len, "%sspare", sep);
1523                 sep = ",";
1524         }
1525         return len+sprintf(page+len, "\n");
1526 }
1527
1528 static struct rdev_sysfs_entry
1529 rdev_state = __ATTR_RO(state);
1530
1531 static ssize_t
1532 super_show(mdk_rdev_t *rdev, char *page)
1533 {
1534         if (rdev->sb_loaded && rdev->sb_size) {
1535                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1536                 return rdev->sb_size;
1537         } else
1538                 return 0;
1539 }
1540 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1541
1542 static struct attribute *rdev_default_attrs[] = {
1543         &rdev_state.attr,
1544         &rdev_super.attr,
1545         NULL,
1546 };
1547 static ssize_t
1548 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1549 {
1550         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1551         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1552
1553         if (!entry->show)
1554                 return -EIO;
1555         return entry->show(rdev, page);
1556 }
1557
1558 static ssize_t
1559 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1560               const char *page, size_t length)
1561 {
1562         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1563         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1564
1565         if (!entry->store)
1566                 return -EIO;
1567         return entry->store(rdev, page, length);
1568 }
1569
1570 static void rdev_free(struct kobject *ko)
1571 {
1572         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1573         kfree(rdev);
1574 }
1575 static struct sysfs_ops rdev_sysfs_ops = {
1576         .show           = rdev_attr_show,
1577         .store          = rdev_attr_store,
1578 };
1579 static struct kobj_type rdev_ktype = {
1580         .release        = rdev_free,
1581         .sysfs_ops      = &rdev_sysfs_ops,
1582         .default_attrs  = rdev_default_attrs,
1583 };
1584
1585 /*
1586  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1587  *
1588  * mark the device faulty if:
1589  *
1590  *   - the device is nonexistent (zero size)
1591  *   - the device has no valid superblock
1592  *
1593  * a faulty rdev _never_ has rdev->sb set.
1594  */
1595 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1596 {
1597         char b[BDEVNAME_SIZE];
1598         int err;
1599         mdk_rdev_t *rdev;
1600         sector_t size;
1601
1602         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1603         if (!rdev) {
1604                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1605                 return ERR_PTR(-ENOMEM);
1606         }
1607         memset(rdev, 0, sizeof(*rdev));
1608
1609         if ((err = alloc_disk_sb(rdev)))
1610                 goto abort_free;
1611
1612         err = lock_rdev(rdev, newdev);
1613         if (err)
1614                 goto abort_free;
1615
1616         rdev->kobj.parent = NULL;
1617         rdev->kobj.ktype = &rdev_ktype;
1618         kobject_init(&rdev->kobj);
1619
1620         rdev->desc_nr = -1;
1621         rdev->flags = 0;
1622         rdev->data_offset = 0;
1623         atomic_set(&rdev->nr_pending, 0);
1624         atomic_set(&rdev->read_errors, 0);
1625
1626         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1627         if (!size) {
1628                 printk(KERN_WARNING
1629                         "md: %s has zero or unknown size, marking faulty!\n",
1630                         bdevname(rdev->bdev,b));
1631                 err = -EINVAL;
1632                 goto abort_free;
1633         }
1634
1635         if (super_format >= 0) {
1636                 err = super_types[super_format].
1637                         load_super(rdev, NULL, super_minor);
1638                 if (err == -EINVAL) {
1639                         printk(KERN_WARNING
1640                                 "md: %s has invalid sb, not importing!\n",
1641                                 bdevname(rdev->bdev,b));
1642                         goto abort_free;
1643                 }
1644                 if (err < 0) {
1645                         printk(KERN_WARNING
1646                                 "md: could not read %s's sb, not importing!\n",
1647                                 bdevname(rdev->bdev,b));
1648                         goto abort_free;
1649                 }
1650         }
1651         INIT_LIST_HEAD(&rdev->same_set);
1652
1653         return rdev;
1654
1655 abort_free:
1656         if (rdev->sb_page) {
1657                 if (rdev->bdev)
1658                         unlock_rdev(rdev);
1659                 free_disk_sb(rdev);
1660         }
1661         kfree(rdev);
1662         return ERR_PTR(err);
1663 }
1664
1665 /*
1666  * Check a full RAID array for plausibility
1667  */
1668
1669
1670 static void analyze_sbs(mddev_t * mddev)
1671 {
1672         int i;
1673         struct list_head *tmp;
1674         mdk_rdev_t *rdev, *freshest;
1675         char b[BDEVNAME_SIZE];
1676
1677         freshest = NULL;
1678         ITERATE_RDEV(mddev,rdev,tmp)
1679                 switch (super_types[mddev->major_version].
1680                         load_super(rdev, freshest, mddev->minor_version)) {
1681                 case 1:
1682                         freshest = rdev;
1683                         break;
1684                 case 0:
1685                         break;
1686                 default:
1687                         printk( KERN_ERR \
1688                                 "md: fatal superblock inconsistency in %s"
1689                                 " -- removing from array\n",
1690                                 bdevname(rdev->bdev,b));
1691                         kick_rdev_from_array(rdev);
1692                 }
1693
1694
1695         super_types[mddev->major_version].
1696                 validate_super(mddev, freshest);
1697
1698         i = 0;
1699         ITERATE_RDEV(mddev,rdev,tmp) {
1700                 if (rdev != freshest)
1701                         if (super_types[mddev->major_version].
1702                             validate_super(mddev, rdev)) {
1703                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1704                                         " from array!\n",
1705                                         bdevname(rdev->bdev,b));
1706                                 kick_rdev_from_array(rdev);
1707                                 continue;
1708                         }
1709                 if (mddev->level == LEVEL_MULTIPATH) {
1710                         rdev->desc_nr = i++;
1711                         rdev->raid_disk = rdev->desc_nr;
1712                         set_bit(In_sync, &rdev->flags);
1713                 }
1714         }
1715
1716
1717
1718         if (mddev->recovery_cp != MaxSector &&
1719             mddev->level >= 1)
1720                 printk(KERN_ERR "md: %s: raid array is not clean"
1721                        " -- starting background reconstruction\n",
1722                        mdname(mddev));
1723
1724 }
1725
1726 static ssize_t
1727 level_show(mddev_t *mddev, char *page)
1728 {
1729         mdk_personality_t *p = mddev->pers;
1730         if (p == NULL)
1731                 return 0;
1732         if (mddev->level >= 0)
1733                 return sprintf(page, "RAID-%d\n", mddev->level);
1734         else
1735                 return sprintf(page, "%s\n", p->name);
1736 }
1737
1738 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1739
1740 static ssize_t
1741 raid_disks_show(mddev_t *mddev, char *page)
1742 {
1743         return sprintf(page, "%d\n", mddev->raid_disks);
1744 }
1745
1746 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1747
1748 static ssize_t
1749 md_show_scan(mddev_t *mddev, char *page)
1750 {
1751         char *type = "none";
1752         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1753             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1754                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1755                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1756                                 type = "resync";
1757                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1758                                 type = "check";
1759                         else
1760                                 type = "repair";
1761                 } else
1762                         type = "recover";
1763         }
1764         return sprintf(page, "%s\n", type);
1765 }
1766
1767 static ssize_t
1768 md_store_scan(mddev_t *mddev, const char *page, size_t len)
1769 {
1770         int canscan=0;
1771
1772         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1773             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1774                 return -EBUSY;
1775
1776         if (mddev->pers && mddev->pers->sync_request)
1777                 canscan=1;
1778
1779         if (!canscan)
1780                 return -EINVAL;
1781
1782         if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1783                 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1784         else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1785                 return -EINVAL;
1786         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1787         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1788         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1789         md_wakeup_thread(mddev->thread);
1790         return len;
1791 }
1792
1793 static ssize_t
1794 mismatch_cnt_show(mddev_t *mddev, char *page)
1795 {
1796         return sprintf(page, "%llu\n",
1797                        (unsigned long long) mddev->resync_mismatches);
1798 }
1799
1800 static struct md_sysfs_entry
1801 md_scan_mode = __ATTR(scan_mode, S_IRUGO|S_IWUSR, md_show_scan, md_store_scan);
1802
1803
1804 static struct md_sysfs_entry
1805 md_mismatches = __ATTR_RO(mismatch_cnt);
1806
1807 static struct attribute *md_default_attrs[] = {
1808         &md_level.attr,
1809         &md_raid_disks.attr,
1810         NULL,
1811 };
1812
1813 static struct attribute *md_redundancy_attrs[] = {
1814         &md_scan_mode.attr,
1815         &md_mismatches.attr,
1816         NULL,
1817 };
1818 static struct attribute_group md_redundancy_group = {
1819         .name = NULL,
1820         .attrs = md_redundancy_attrs,
1821 };
1822
1823
1824 static ssize_t
1825 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1826 {
1827         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1828         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1829         ssize_t rv;
1830
1831         if (!entry->show)
1832                 return -EIO;
1833         mddev_lock(mddev);
1834         rv = entry->show(mddev, page);
1835         mddev_unlock(mddev);
1836         return rv;
1837 }
1838
1839 static ssize_t
1840 md_attr_store(struct kobject *kobj, struct attribute *attr,
1841               const char *page, size_t length)
1842 {
1843         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1844         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1845         ssize_t rv;
1846
1847         if (!entry->store)
1848                 return -EIO;
1849         mddev_lock(mddev);
1850         rv = entry->store(mddev, page, length);
1851         mddev_unlock(mddev);
1852         return rv;
1853 }
1854
1855 static void md_free(struct kobject *ko)
1856 {
1857         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1858         kfree(mddev);
1859 }
1860
1861 static struct sysfs_ops md_sysfs_ops = {
1862         .show   = md_attr_show,
1863         .store  = md_attr_store,
1864 };
1865 static struct kobj_type md_ktype = {
1866         .release        = md_free,
1867         .sysfs_ops      = &md_sysfs_ops,
1868         .default_attrs  = md_default_attrs,
1869 };
1870
1871 int mdp_major = 0;
1872
1873 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1874 {
1875         static DECLARE_MUTEX(disks_sem);
1876         mddev_t *mddev = mddev_find(dev);
1877         struct gendisk *disk;
1878         int partitioned = (MAJOR(dev) != MD_MAJOR);
1879         int shift = partitioned ? MdpMinorShift : 0;
1880         int unit = MINOR(dev) >> shift;
1881
1882         if (!mddev)
1883                 return NULL;
1884
1885         down(&disks_sem);
1886         if (mddev->gendisk) {
1887                 up(&disks_sem);
1888                 mddev_put(mddev);
1889                 return NULL;
1890         }
1891         disk = alloc_disk(1 << shift);
1892         if (!disk) {
1893                 up(&disks_sem);
1894                 mddev_put(mddev);
1895                 return NULL;
1896         }
1897         disk->major = MAJOR(dev);
1898         disk->first_minor = unit << shift;
1899         if (partitioned) {
1900                 sprintf(disk->disk_name, "md_d%d", unit);
1901                 sprintf(disk->devfs_name, "md/d%d", unit);
1902         } else {
1903                 sprintf(disk->disk_name, "md%d", unit);
1904                 sprintf(disk->devfs_name, "md/%d", unit);
1905         }
1906         disk->fops = &md_fops;
1907         disk->private_data = mddev;
1908         disk->queue = mddev->queue;
1909         add_disk(disk);
1910         mddev->gendisk = disk;
1911         up(&disks_sem);
1912         mddev->kobj.parent = &disk->kobj;
1913         mddev->kobj.k_name = NULL;
1914         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1915         mddev->kobj.ktype = &md_ktype;
1916         kobject_register(&mddev->kobj);
1917         return NULL;
1918 }
1919
1920 void md_wakeup_thread(mdk_thread_t *thread);
1921
1922 static void md_safemode_timeout(unsigned long data)
1923 {
1924         mddev_t *mddev = (mddev_t *) data;
1925
1926         mddev->safemode = 1;
1927         md_wakeup_thread(mddev->thread);
1928 }
1929
1930
1931 static int do_md_run(mddev_t * mddev)
1932 {
1933         int pnum, err;
1934         int chunk_size;
1935         struct list_head *tmp;
1936         mdk_rdev_t *rdev;
1937         struct gendisk *disk;
1938         char b[BDEVNAME_SIZE];
1939
1940         if (list_empty(&mddev->disks))
1941                 /* cannot run an array with no devices.. */
1942                 return -EINVAL;
1943
1944         if (mddev->pers)
1945                 return -EBUSY;
1946
1947         /*
1948          * Analyze all RAID superblock(s)
1949          */
1950         if (!mddev->raid_disks)
1951                 analyze_sbs(mddev);
1952
1953         chunk_size = mddev->chunk_size;
1954         pnum = level_to_pers(mddev->level);
1955
1956         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1957                 if (!chunk_size) {
1958                         /*
1959                          * 'default chunksize' in the old md code used to
1960                          * be PAGE_SIZE, baaad.
1961                          * we abort here to be on the safe side. We don't
1962                          * want to continue the bad practice.
1963                          */
1964                         printk(KERN_ERR
1965                                 "no chunksize specified, see 'man raidtab'\n");
1966                         return -EINVAL;
1967                 }
1968                 if (chunk_size > MAX_CHUNK_SIZE) {
1969                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1970                                 chunk_size, MAX_CHUNK_SIZE);
1971                         return -EINVAL;
1972                 }
1973                 /*
1974                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1975                  */
1976                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1977                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1978                         return -EINVAL;
1979                 }
1980                 if (chunk_size < PAGE_SIZE) {
1981                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1982                                 chunk_size, PAGE_SIZE);
1983                         return -EINVAL;
1984                 }
1985
1986                 /* devices must have minimum size of one chunk */
1987                 ITERATE_RDEV(mddev,rdev,tmp) {
1988                         if (test_bit(Faulty, &rdev->flags))
1989                                 continue;
1990                         if (rdev->size < chunk_size / 1024) {
1991                                 printk(KERN_WARNING
1992                                         "md: Dev %s smaller than chunk_size:"
1993                                         " %lluk < %dk\n",
1994                                         bdevname(rdev->bdev,b),
1995                                         (unsigned long long)rdev->size,
1996                                         chunk_size / 1024);
1997                                 return -EINVAL;
1998                         }
1999                 }
2000         }
2001
2002 #ifdef CONFIG_KMOD
2003         if (!pers[pnum])
2004         {
2005                 request_module("md-personality-%d", pnum);
2006         }
2007 #endif
2008
2009         /*
2010          * Drop all container device buffers, from now on
2011          * the only valid external interface is through the md
2012          * device.
2013          * Also find largest hardsector size
2014          */
2015         ITERATE_RDEV(mddev,rdev,tmp) {
2016                 if (test_bit(Faulty, &rdev->flags))
2017                         continue;
2018                 sync_blockdev(rdev->bdev);
2019                 invalidate_bdev(rdev->bdev, 0);
2020         }
2021
2022         md_probe(mddev->unit, NULL, NULL);
2023         disk = mddev->gendisk;
2024         if (!disk)
2025                 return -ENOMEM;
2026
2027         spin_lock(&pers_lock);
2028         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2029                 spin_unlock(&pers_lock);
2030                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2031                        pnum);
2032                 return -EINVAL;
2033         }
2034
2035         mddev->pers = pers[pnum];
2036         spin_unlock(&pers_lock);
2037
2038         mddev->recovery = 0;
2039         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2040         mddev->barriers_work = 1;
2041
2042         if (start_readonly)
2043                 mddev->ro = 2; /* read-only, but switch on first write */
2044
2045         /* before we start the array running, initialise the bitmap */
2046         err = bitmap_create(mddev);
2047         if (err)
2048                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2049                         mdname(mddev), err);
2050         else
2051                 err = mddev->pers->run(mddev);
2052         if (err) {
2053                 printk(KERN_ERR "md: pers->run() failed ...\n");
2054                 module_put(mddev->pers->owner);
2055                 mddev->pers = NULL;
2056                 bitmap_destroy(mddev);
2057                 return err;
2058         }
2059         if (mddev->pers->sync_request)
2060                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2061         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2062                 mddev->ro = 0;
2063
2064         atomic_set(&mddev->writes_pending,0);
2065         mddev->safemode = 0;
2066         mddev->safemode_timer.function = md_safemode_timeout;
2067         mddev->safemode_timer.data = (unsigned long) mddev;
2068         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2069         mddev->in_sync = 1;
2070
2071         ITERATE_RDEV(mddev,rdev,tmp)
2072                 if (rdev->raid_disk >= 0) {
2073                         char nm[20];
2074                         sprintf(nm, "rd%d", rdev->raid_disk);
2075                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2076                 }
2077
2078         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2079         md_wakeup_thread(mddev->thread);
2080
2081         if (mddev->sb_dirty)
2082                 md_update_sb(mddev);
2083
2084         set_capacity(disk, mddev->array_size<<1);
2085
2086         /* If we call blk_queue_make_request here, it will
2087          * re-initialise max_sectors etc which may have been
2088          * refined inside -> run.  So just set the bits we need to set.
2089          * Most initialisation happended when we called
2090          * blk_queue_make_request(..., md_fail_request)
2091          * earlier.
2092          */
2093         mddev->queue->queuedata = mddev;
2094         mddev->queue->make_request_fn = mddev->pers->make_request;
2095
2096         mddev->changed = 1;
2097         return 0;
2098 }
2099
2100 static int restart_array(mddev_t *mddev)
2101 {
2102         struct gendisk *disk = mddev->gendisk;
2103         int err;
2104
2105         /*
2106          * Complain if it has no devices
2107          */
2108         err = -ENXIO;
2109         if (list_empty(&mddev->disks))
2110                 goto out;
2111
2112         if (mddev->pers) {
2113                 err = -EBUSY;
2114                 if (!mddev->ro)
2115                         goto out;
2116
2117                 mddev->safemode = 0;
2118                 mddev->ro = 0;
2119                 set_disk_ro(disk, 0);
2120
2121                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2122                         mdname(mddev));
2123                 /*
2124                  * Kick recovery or resync if necessary
2125                  */
2126                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2127                 md_wakeup_thread(mddev->thread);
2128                 err = 0;
2129         } else {
2130                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2131                         mdname(mddev));
2132                 err = -EINVAL;
2133         }
2134
2135 out:
2136         return err;
2137 }
2138
2139 static int do_md_stop(mddev_t * mddev, int ro)
2140 {
2141         int err = 0;
2142         struct gendisk *disk = mddev->gendisk;
2143
2144         if (mddev->pers) {
2145                 if (atomic_read(&mddev->active)>2) {
2146                         printk("md: %s still in use.\n",mdname(mddev));
2147                         return -EBUSY;
2148                 }
2149
2150                 if (mddev->sync_thread) {
2151                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2152                         md_unregister_thread(mddev->sync_thread);
2153                         mddev->sync_thread = NULL;
2154                 }
2155
2156                 del_timer_sync(&mddev->safemode_timer);
2157
2158                 invalidate_partition(disk, 0);
2159
2160                 if (ro) {
2161                         err  = -ENXIO;
2162                         if (mddev->ro==1)
2163                                 goto out;
2164                         mddev->ro = 1;
2165                 } else {
2166                         bitmap_flush(mddev);
2167                         md_super_wait(mddev);
2168                         if (mddev->ro)
2169                                 set_disk_ro(disk, 0);
2170                         blk_queue_make_request(mddev->queue, md_fail_request);
2171                         mddev->pers->stop(mddev);
2172                         if (mddev->pers->sync_request)
2173                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2174
2175                         module_put(mddev->pers->owner);
2176                         mddev->pers = NULL;
2177                         if (mddev->ro)
2178                                 mddev->ro = 0;
2179                 }
2180                 if (!mddev->in_sync) {
2181                         /* mark array as shutdown cleanly */
2182                         mddev->in_sync = 1;
2183                         md_update_sb(mddev);
2184                 }
2185                 if (ro)
2186                         set_disk_ro(disk, 1);
2187         }
2188
2189         bitmap_destroy(mddev);
2190         if (mddev->bitmap_file) {
2191                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2192                 fput(mddev->bitmap_file);
2193                 mddev->bitmap_file = NULL;
2194         }
2195         mddev->bitmap_offset = 0;
2196
2197         /*
2198          * Free resources if final stop
2199          */
2200         if (!ro) {
2201                 mdk_rdev_t *rdev;
2202                 struct list_head *tmp;
2203                 struct gendisk *disk;
2204                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2205
2206                 ITERATE_RDEV(mddev,rdev,tmp)
2207                         if (rdev->raid_disk >= 0) {
2208                                 char nm[20];
2209                                 sprintf(nm, "rd%d", rdev->raid_disk);
2210                                 sysfs_remove_link(&mddev->kobj, nm);
2211                         }
2212
2213                 export_array(mddev);
2214
2215                 mddev->array_size = 0;
2216                 disk = mddev->gendisk;
2217                 if (disk)
2218                         set_capacity(disk, 0);
2219                 mddev->changed = 1;
2220         } else
2221                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2222                         mdname(mddev));
2223         err = 0;
2224 out:
2225         return err;
2226 }
2227
2228 static void autorun_array(mddev_t *mddev)
2229 {
2230         mdk_rdev_t *rdev;
2231         struct list_head *tmp;
2232         int err;
2233
2234         if (list_empty(&mddev->disks))
2235                 return;
2236
2237         printk(KERN_INFO "md: running: ");
2238
2239         ITERATE_RDEV(mddev,rdev,tmp) {
2240                 char b[BDEVNAME_SIZE];
2241                 printk("<%s>", bdevname(rdev->bdev,b));
2242         }
2243         printk("\n");
2244
2245         err = do_md_run (mddev);
2246         if (err) {
2247                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2248                 do_md_stop (mddev, 0);
2249         }
2250 }
2251
2252 /*
2253  * lets try to run arrays based on all disks that have arrived
2254  * until now. (those are in pending_raid_disks)
2255  *
2256  * the method: pick the first pending disk, collect all disks with
2257  * the same UUID, remove all from the pending list and put them into
2258  * the 'same_array' list. Then order this list based on superblock
2259  * update time (freshest comes first), kick out 'old' disks and
2260  * compare superblocks. If everything's fine then run it.
2261  *
2262  * If "unit" is allocated, then bump its reference count
2263  */
2264 static void autorun_devices(int part)
2265 {
2266         struct list_head candidates;
2267         struct list_head *tmp;
2268         mdk_rdev_t *rdev0, *rdev;
2269         mddev_t *mddev;
2270         char b[BDEVNAME_SIZE];
2271
2272         printk(KERN_INFO "md: autorun ...\n");
2273         while (!list_empty(&pending_raid_disks)) {
2274                 dev_t dev;
2275                 rdev0 = list_entry(pending_raid_disks.next,
2276                                          mdk_rdev_t, same_set);
2277
2278                 printk(KERN_INFO "md: considering %s ...\n",
2279                         bdevname(rdev0->bdev,b));
2280                 INIT_LIST_HEAD(&candidates);
2281                 ITERATE_RDEV_PENDING(rdev,tmp)
2282                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2283                                 printk(KERN_INFO "md:  adding %s ...\n",
2284                                         bdevname(rdev->bdev,b));
2285                                 list_move(&rdev->same_set, &candidates);
2286                         }
2287                 /*
2288                  * now we have a set of devices, with all of them having
2289                  * mostly sane superblocks. It's time to allocate the
2290                  * mddev.
2291                  */
2292                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2293                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2294                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2295                         break;
2296                 }
2297                 if (part)
2298                         dev = MKDEV(mdp_major,
2299                                     rdev0->preferred_minor << MdpMinorShift);
2300                 else
2301                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2302
2303                 md_probe(dev, NULL, NULL);
2304                 mddev = mddev_find(dev);
2305                 if (!mddev) {
2306                         printk(KERN_ERR
2307                                 "md: cannot allocate memory for md drive.\n");
2308                         break;
2309                 }
2310                 if (mddev_lock(mddev))
2311                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2312                                mdname(mddev));
2313                 else if (mddev->raid_disks || mddev->major_version
2314                          || !list_empty(&mddev->disks)) {
2315                         printk(KERN_WARNING
2316                                 "md: %s already running, cannot run %s\n",
2317                                 mdname(mddev), bdevname(rdev0->bdev,b));
2318                         mddev_unlock(mddev);
2319                 } else {
2320                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2321                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2322                                 list_del_init(&rdev->same_set);
2323                                 if (bind_rdev_to_array(rdev, mddev))
2324                                         export_rdev(rdev);
2325                         }
2326                         autorun_array(mddev);
2327                         mddev_unlock(mddev);
2328                 }
2329                 /* on success, candidates will be empty, on error
2330                  * it won't...
2331                  */
2332                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2333                         export_rdev(rdev);
2334                 mddev_put(mddev);
2335         }
2336         printk(KERN_INFO "md: ... autorun DONE.\n");
2337 }
2338
2339 /*
2340  * import RAID devices based on one partition
2341  * if possible, the array gets run as well.
2342  */
2343
2344 static int autostart_array(dev_t startdev)
2345 {
2346         char b[BDEVNAME_SIZE];
2347         int err = -EINVAL, i;
2348         mdp_super_t *sb = NULL;
2349         mdk_rdev_t *start_rdev = NULL, *rdev;
2350
2351         start_rdev = md_import_device(startdev, 0, 0);
2352         if (IS_ERR(start_rdev))
2353                 return err;
2354
2355
2356         /* NOTE: this can only work for 0.90.0 superblocks */
2357         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2358         if (sb->major_version != 0 ||
2359             sb->minor_version != 90 ) {
2360                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2361                 export_rdev(start_rdev);
2362                 return err;
2363         }
2364
2365         if (test_bit(Faulty, &start_rdev->flags)) {
2366                 printk(KERN_WARNING
2367                         "md: can not autostart based on faulty %s!\n",
2368                         bdevname(start_rdev->bdev,b));
2369                 export_rdev(start_rdev);
2370                 return err;
2371         }
2372         list_add(&start_rdev->same_set, &pending_raid_disks);
2373
2374         for (i = 0; i < MD_SB_DISKS; i++) {
2375                 mdp_disk_t *desc = sb->disks + i;
2376                 dev_t dev = MKDEV(desc->major, desc->minor);
2377
2378                 if (!dev)
2379                         continue;
2380                 if (dev == startdev)
2381                         continue;
2382                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2383                         continue;
2384                 rdev = md_import_device(dev, 0, 0);
2385                 if (IS_ERR(rdev))
2386                         continue;
2387
2388                 list_add(&rdev->same_set, &pending_raid_disks);
2389         }
2390
2391         /*
2392          * possibly return codes
2393          */
2394         autorun_devices(0);
2395         return 0;
2396
2397 }
2398
2399
2400 static int get_version(void __user * arg)
2401 {
2402         mdu_version_t ver;
2403
2404         ver.major = MD_MAJOR_VERSION;
2405         ver.minor = MD_MINOR_VERSION;
2406         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2407
2408         if (copy_to_user(arg, &ver, sizeof(ver)))
2409                 return -EFAULT;
2410
2411         return 0;
2412 }
2413
2414 static int get_array_info(mddev_t * mddev, void __user * arg)
2415 {
2416         mdu_array_info_t info;
2417         int nr,working,active,failed,spare;
2418         mdk_rdev_t *rdev;
2419         struct list_head *tmp;
2420
2421         nr=working=active=failed=spare=0;
2422         ITERATE_RDEV(mddev,rdev,tmp) {
2423                 nr++;
2424                 if (test_bit(Faulty, &rdev->flags))
2425                         failed++;
2426                 else {
2427                         working++;
2428                         if (test_bit(In_sync, &rdev->flags))
2429                                 active++;
2430                         else
2431                                 spare++;
2432                 }
2433         }
2434
2435         info.major_version = mddev->major_version;
2436         info.minor_version = mddev->minor_version;
2437         info.patch_version = MD_PATCHLEVEL_VERSION;
2438         info.ctime         = mddev->ctime;
2439         info.level         = mddev->level;
2440         info.size          = mddev->size;
2441         info.nr_disks      = nr;
2442         info.raid_disks    = mddev->raid_disks;
2443         info.md_minor      = mddev->md_minor;
2444         info.not_persistent= !mddev->persistent;
2445
2446         info.utime         = mddev->utime;
2447         info.state         = 0;
2448         if (mddev->in_sync)
2449                 info.state = (1<<MD_SB_CLEAN);
2450         if (mddev->bitmap && mddev->bitmap_offset)
2451                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2452         info.active_disks  = active;
2453         info.working_disks = working;
2454         info.failed_disks  = failed;
2455         info.spare_disks   = spare;
2456
2457         info.layout        = mddev->layout;
2458         info.chunk_size    = mddev->chunk_size;
2459
2460         if (copy_to_user(arg, &info, sizeof(info)))
2461                 return -EFAULT;
2462
2463         return 0;
2464 }
2465
2466 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2467 {
2468         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2469         char *ptr, *buf = NULL;
2470         int err = -ENOMEM;
2471
2472         file = kmalloc(sizeof(*file), GFP_KERNEL);
2473         if (!file)
2474                 goto out;
2475
2476         /* bitmap disabled, zero the first byte and copy out */
2477         if (!mddev->bitmap || !mddev->bitmap->file) {
2478                 file->pathname[0] = '\0';
2479                 goto copy_out;
2480         }
2481
2482         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2483         if (!buf)
2484                 goto out;
2485
2486         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2487         if (!ptr)
2488                 goto out;
2489
2490         strcpy(file->pathname, ptr);
2491
2492 copy_out:
2493         err = 0;
2494         if (copy_to_user(arg, file, sizeof(*file)))
2495                 err = -EFAULT;
2496 out:
2497         kfree(buf);
2498         kfree(file);
2499         return err;
2500 }
2501
2502 static int get_disk_info(mddev_t * mddev, void __user * arg)
2503 {
2504         mdu_disk_info_t info;
2505         unsigned int nr;
2506         mdk_rdev_t *rdev;
2507
2508         if (copy_from_user(&info, arg, sizeof(info)))
2509                 return -EFAULT;
2510
2511         nr = info.number;
2512
2513         rdev = find_rdev_nr(mddev, nr);
2514         if (rdev) {
2515                 info.major = MAJOR(rdev->bdev->bd_dev);
2516                 info.minor = MINOR(rdev->bdev->bd_dev);
2517                 info.raid_disk = rdev->raid_disk;
2518                 info.state = 0;
2519                 if (test_bit(Faulty, &rdev->flags))
2520                         info.state |= (1<<MD_DISK_FAULTY);
2521                 else if (test_bit(In_sync, &rdev->flags)) {
2522                         info.state |= (1<<MD_DISK_ACTIVE);
2523                         info.state |= (1<<MD_DISK_SYNC);
2524                 }
2525                 if (test_bit(WriteMostly, &rdev->flags))
2526                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2527         } else {
2528                 info.major = info.minor = 0;
2529                 info.raid_disk = -1;
2530                 info.state = (1<<MD_DISK_REMOVED);
2531         }
2532
2533         if (copy_to_user(arg, &info, sizeof(info)))
2534                 return -EFAULT;
2535
2536         return 0;
2537 }
2538
2539 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2540 {
2541         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2542         mdk_rdev_t *rdev;
2543         dev_t dev = MKDEV(info->major,info->minor);
2544
2545         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2546                 return -EOVERFLOW;
2547
2548         if (!mddev->raid_disks) {
2549                 int err;
2550                 /* expecting a device which has a superblock */
2551                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2552                 if (IS_ERR(rdev)) {
2553                         printk(KERN_WARNING
2554                                 "md: md_import_device returned %ld\n",
2555                                 PTR_ERR(rdev));
2556                         return PTR_ERR(rdev);
2557                 }
2558                 if (!list_empty(&mddev->disks)) {
2559                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2560                                                         mdk_rdev_t, same_set);
2561                         int err = super_types[mddev->major_version]
2562                                 .load_super(rdev, rdev0, mddev->minor_version);
2563                         if (err < 0) {
2564                                 printk(KERN_WARNING
2565                                         "md: %s has different UUID to %s\n",
2566                                         bdevname(rdev->bdev,b),
2567                                         bdevname(rdev0->bdev,b2));
2568                                 export_rdev(rdev);
2569                                 return -EINVAL;
2570                         }
2571                 }
2572                 err = bind_rdev_to_array(rdev, mddev);
2573                 if (err)
2574                         export_rdev(rdev);
2575                 return err;
2576         }
2577
2578         /*
2579          * add_new_disk can be used once the array is assembled
2580          * to add "hot spares".  They must already have a superblock
2581          * written
2582          */
2583         if (mddev->pers) {
2584                 int err;
2585                 if (!mddev->pers->hot_add_disk) {
2586                         printk(KERN_WARNING
2587                                 "%s: personality does not support diskops!\n",
2588                                mdname(mddev));
2589                         return -EINVAL;
2590                 }
2591                 if (mddev->persistent)
2592                         rdev = md_import_device(dev, mddev->major_version,
2593                                                 mddev->minor_version);
2594                 else
2595                         rdev = md_import_device(dev, -1, -1);
2596                 if (IS_ERR(rdev)) {
2597                         printk(KERN_WARNING
2598                                 "md: md_import_device returned %ld\n",
2599                                 PTR_ERR(rdev));
2600                         return PTR_ERR(rdev);
2601                 }
2602                 /* set save_raid_disk if appropriate */
2603                 if (!mddev->persistent) {
2604                         if (info->state & (1<<MD_DISK_SYNC)  &&
2605                             info->raid_disk < mddev->raid_disks)
2606                                 rdev->raid_disk = info->raid_disk;
2607                         else
2608                                 rdev->raid_disk = -1;
2609                 } else
2610                         super_types[mddev->major_version].
2611                                 validate_super(mddev, rdev);
2612                 rdev->saved_raid_disk = rdev->raid_disk;
2613
2614                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2615                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2616                         set_bit(WriteMostly, &rdev->flags);
2617
2618                 rdev->raid_disk = -1;
2619                 err = bind_rdev_to_array(rdev, mddev);
2620                 if (err)
2621                         export_rdev(rdev);
2622
2623                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2624                 md_wakeup_thread(mddev->thread);
2625                 return err;
2626         }
2627
2628         /* otherwise, add_new_disk is only allowed
2629          * for major_version==0 superblocks
2630          */
2631         if (mddev->major_version != 0) {
2632                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2633                        mdname(mddev));
2634                 return -EINVAL;
2635         }
2636
2637         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2638                 int err;
2639                 rdev = md_import_device (dev, -1, 0);
2640                 if (IS_ERR(rdev)) {
2641                         printk(KERN_WARNING
2642                                 "md: error, md_import_device() returned %ld\n",
2643                                 PTR_ERR(rdev));
2644                         return PTR_ERR(rdev);
2645                 }
2646                 rdev->desc_nr = info->number;
2647                 if (info->raid_disk < mddev->raid_disks)
2648                         rdev->raid_disk = info->raid_disk;
2649                 else
2650                         rdev->raid_disk = -1;
2651
2652                 rdev->flags = 0;
2653
2654                 if (rdev->raid_disk < mddev->raid_disks)
2655                         if (info->state & (1<<MD_DISK_SYNC))
2656                                 set_bit(In_sync, &rdev->flags);
2657
2658                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2659                         set_bit(WriteMostly, &rdev->flags);
2660
2661                 err = bind_rdev_to_array(rdev, mddev);
2662                 if (err) {
2663                         export_rdev(rdev);
2664                         return err;
2665                 }
2666
2667                 if (!mddev->persistent) {
2668                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2669                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2670                 } else
2671                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2672                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2673
2674                 if (!mddev->size || (mddev->size > rdev->size))
2675                         mddev->size = rdev->size;
2676         }
2677
2678         return 0;
2679 }
2680
2681 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2682 {
2683         char b[BDEVNAME_SIZE];
2684         mdk_rdev_t *rdev;
2685
2686         if (!mddev->pers)
2687                 return -ENODEV;
2688
2689         rdev = find_rdev(mddev, dev);
2690         if (!rdev)
2691                 return -ENXIO;
2692
2693         if (rdev->raid_disk >= 0)
2694                 goto busy;
2695
2696         kick_rdev_from_array(rdev);
2697         md_update_sb(mddev);
2698
2699         return 0;
2700 busy:
2701         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2702                 bdevname(rdev->bdev,b), mdname(mddev));
2703         return -EBUSY;
2704 }
2705
2706 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2707 {
2708         char b[BDEVNAME_SIZE];
2709         int err;
2710         unsigned int size;
2711         mdk_rdev_t *rdev;
2712
2713         if (!mddev->pers)
2714                 return -ENODEV;
2715
2716         if (mddev->major_version != 0) {
2717                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2718                         " version-0 superblocks.\n",
2719                         mdname(mddev));
2720                 return -EINVAL;
2721         }
2722         if (!mddev->pers->hot_add_disk) {
2723                 printk(KERN_WARNING
2724                         "%s: personality does not support diskops!\n",
2725                         mdname(mddev));
2726                 return -EINVAL;
2727         }
2728
2729         rdev = md_import_device (dev, -1, 0);
2730         if (IS_ERR(rdev)) {
2731                 printk(KERN_WARNING
2732                         "md: error, md_import_device() returned %ld\n",
2733                         PTR_ERR(rdev));
2734                 return -EINVAL;
2735         }
2736
2737         if (mddev->persistent)
2738                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2739         else
2740                 rdev->sb_offset =
2741                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2742
2743         size = calc_dev_size(rdev, mddev->chunk_size);
2744         rdev->size = size;
2745
2746         if (size < mddev->size) {
2747                 printk(KERN_WARNING
2748                         "%s: disk size %llu blocks < array size %llu\n",
2749                         mdname(mddev), (unsigned long long)size,
2750                         (unsigned long long)mddev->size);
2751                 err = -ENOSPC;
2752                 goto abort_export;
2753         }
2754
2755         if (test_bit(Faulty, &rdev->flags)) {
2756                 printk(KERN_WARNING
2757                         "md: can not hot-add faulty %s disk to %s!\n",
2758                         bdevname(rdev->bdev,b), mdname(mddev));
2759                 err = -EINVAL;
2760                 goto abort_export;
2761         }
2762         clear_bit(In_sync, &rdev->flags);
2763         rdev->desc_nr = -1;
2764         bind_rdev_to_array(rdev, mddev);
2765
2766         /*
2767          * The rest should better be atomic, we can have disk failures
2768          * noticed in interrupt contexts ...
2769          */
2770
2771         if (rdev->desc_nr == mddev->max_disks) {
2772                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2773                         mdname(mddev));
2774                 err = -EBUSY;
2775                 goto abort_unbind_export;
2776         }
2777
2778         rdev->raid_disk = -1;
2779
2780         md_update_sb(mddev);
2781
2782         /*
2783          * Kick recovery, maybe this spare has to be added to the
2784          * array immediately.
2785          */
2786         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2787         md_wakeup_thread(mddev->thread);
2788
2789         return 0;
2790
2791 abort_unbind_export:
2792         unbind_rdev_from_array(rdev);
2793
2794 abort_export:
2795         export_rdev(rdev);
2796         return err;
2797 }
2798
2799 /* similar to deny_write_access, but accounts for our holding a reference
2800  * to the file ourselves */
2801 static int deny_bitmap_write_access(struct file * file)
2802 {
2803         struct inode *inode = file->f_mapping->host;
2804
2805         spin_lock(&inode->i_lock);
2806         if (atomic_read(&inode->i_writecount) > 1) {
2807                 spin_unlock(&inode->i_lock);
2808                 return -ETXTBSY;
2809         }
2810         atomic_set(&inode->i_writecount, -1);
2811         spin_unlock(&inode->i_lock);
2812
2813         return 0;
2814 }
2815
2816 static int set_bitmap_file(mddev_t *mddev, int fd)
2817 {
2818         int err;
2819
2820         if (mddev->pers) {
2821                 if (!mddev->pers->quiesce)
2822                         return -EBUSY;
2823                 if (mddev->recovery || mddev->sync_thread)
2824                         return -EBUSY;
2825                 /* we should be able to change the bitmap.. */
2826         }
2827
2828
2829         if (fd >= 0) {
2830                 if (mddev->bitmap)
2831                         return -EEXIST; /* cannot add when bitmap is present */
2832                 mddev->bitmap_file = fget(fd);
2833
2834                 if (mddev->bitmap_file == NULL) {
2835                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2836                                mdname(mddev));
2837                         return -EBADF;
2838                 }
2839
2840                 err = deny_bitmap_write_access(mddev->bitmap_file);
2841                 if (err) {
2842                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2843                                mdname(mddev));
2844                         fput(mddev->bitmap_file);
2845                         mddev->bitmap_file = NULL;
2846                         return err;
2847                 }
2848                 mddev->bitmap_offset = 0; /* file overrides offset */
2849         } else if (mddev->bitmap == NULL)
2850                 return -ENOENT; /* cannot remove what isn't there */
2851         err = 0;
2852         if (mddev->pers) {
2853                 mddev->pers->quiesce(mddev, 1);
2854                 if (fd >= 0)
2855                         err = bitmap_create(mddev);
2856                 if (fd < 0 || err)
2857                         bitmap_destroy(mddev);
2858                 mddev->pers->quiesce(mddev, 0);
2859         } else if (fd < 0) {
2860                 if (mddev->bitmap_file)
2861                         fput(mddev->bitmap_file);
2862                 mddev->bitmap_file = NULL;
2863         }
2864
2865         return err;
2866 }
2867
2868 /*
2869  * set_array_info is used two different ways
2870  * The original usage is when creating a new array.
2871  * In this usage, raid_disks is > 0 and it together with
2872  *  level, size, not_persistent,layout,chunksize determine the
2873  *  shape of the array.
2874  *  This will always create an array with a type-0.90.0 superblock.
2875  * The newer usage is when assembling an array.
2876  *  In this case raid_disks will be 0, and the major_version field is
2877  *  use to determine which style super-blocks are to be found on the devices.
2878  *  The minor and patch _version numbers are also kept incase the
2879  *  super_block handler wishes to interpret them.
2880  */
2881 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2882 {
2883
2884         if (info->raid_disks == 0) {
2885                 /* just setting version number for superblock loading */
2886                 if (info->major_version < 0 ||
2887                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2888                     super_types[info->major_version].name == NULL) {
2889                         /* maybe try to auto-load a module? */
2890                         printk(KERN_INFO
2891                                 "md: superblock version %d not known\n",
2892                                 info->major_version);
2893                         return -EINVAL;
2894                 }
2895                 mddev->major_version = info->major_version;
2896                 mddev->minor_version = info->minor_version;
2897                 mddev->patch_version = info->patch_version;
2898                 return 0;
2899         }
2900         mddev->major_version = MD_MAJOR_VERSION;
2901         mddev->minor_version = MD_MINOR_VERSION;
2902         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2903         mddev->ctime         = get_seconds();
2904
2905         mddev->level         = info->level;
2906         mddev->size          = info->size;
2907         mddev->raid_disks    = info->raid_disks;
2908         /* don't set md_minor, it is determined by which /dev/md* was
2909          * openned
2910          */
2911         if (info->state & (1<<MD_SB_CLEAN))
2912                 mddev->recovery_cp = MaxSector;
2913         else
2914                 mddev->recovery_cp = 0;
2915         mddev->persistent    = ! info->not_persistent;
2916
2917         mddev->layout        = info->layout;
2918         mddev->chunk_size    = info->chunk_size;
2919
2920         mddev->max_disks     = MD_SB_DISKS;
2921
2922         mddev->sb_dirty      = 1;
2923
2924         /*
2925          * Generate a 128 bit UUID
2926          */
2927         get_random_bytes(mddev->uuid, 16);
2928
2929         return 0;
2930 }
2931
2932 /*
2933  * update_array_info is used to change the configuration of an
2934  * on-line array.
2935  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2936  * fields in the info are checked against the array.
2937  * Any differences that cannot be handled will cause an error.
2938  * Normally, only one change can be managed at a time.
2939  */
2940 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2941 {
2942         int rv = 0;
2943         int cnt = 0;
2944         int state = 0;
2945
2946         /* calculate expected state,ignoring low bits */
2947         if (mddev->bitmap && mddev->bitmap_offset)
2948                 state |= (1 << MD_SB_BITMAP_PRESENT);
2949
2950         if (mddev->major_version != info->major_version ||
2951             mddev->minor_version != info->minor_version ||
2952 /*          mddev->patch_version != info->patch_version || */
2953             mddev->ctime         != info->ctime         ||
2954             mddev->level         != info->level         ||
2955 /*          mddev->layout        != info->layout        || */
2956             !mddev->persistent   != info->not_persistent||
2957             mddev->chunk_size    != info->chunk_size    ||
2958             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2959             ((state^info->state) & 0xfffffe00)
2960                 )
2961                 return -EINVAL;
2962         /* Check there is only one change */
2963         if (mddev->size != info->size) cnt++;
2964         if (mddev->raid_disks != info->raid_disks) cnt++;
2965         if (mddev->layout != info->layout) cnt++;
2966         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2967         if (cnt == 0) return 0;
2968         if (cnt > 1) return -EINVAL;
2969
2970         if (mddev->layout != info->layout) {
2971                 /* Change layout
2972                  * we don't need to do anything at the md level, the
2973                  * personality will take care of it all.
2974                  */
2975                 if (mddev->pers->reconfig == NULL)
2976                         return -EINVAL;
2977                 else
2978                         return mddev->pers->reconfig(mddev, info->layout, -1);
2979         }
2980         if (mddev->size != info->size) {
2981                 mdk_rdev_t * rdev;
2982                 struct list_head *tmp;
2983                 if (mddev->pers->resize == NULL)
2984                         return -EINVAL;
2985                 /* The "size" is the amount of each device that is used.
2986                  * This can only make sense for arrays with redundancy.
2987                  * linear and raid0 always use whatever space is available
2988                  * We can only consider changing the size if no resync
2989                  * or reconstruction is happening, and if the new size
2990                  * is acceptable. It must fit before the sb_offset or,
2991                  * if that is <data_offset, it must fit before the
2992                  * size of each device.
2993                  * If size is zero, we find the largest size that fits.
2994                  */
2995                 if (mddev->sync_thread)
2996                         return -EBUSY;
2997                 ITERATE_RDEV(mddev,rdev,tmp) {
2998                         sector_t avail;
2999                         int fit = (info->size == 0);
3000                         if (rdev->sb_offset > rdev->data_offset)
3001                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
3002                         else
3003                                 avail = get_capacity(rdev->bdev->bd_disk)
3004                                         - rdev->data_offset;
3005                         if (fit && (info->size == 0 || info->size > avail/2))
3006                                 info->size = avail/2;
3007                         if (avail < ((sector_t)info->size << 1))
3008                                 return -ENOSPC;
3009                 }
3010                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3011                 if (!rv) {
3012                         struct block_device *bdev;
3013
3014                         bdev = bdget_disk(mddev->gendisk, 0);
3015                         if (bdev) {
3016                                 down(&bdev->bd_inode->i_sem);
3017                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3018                                 up(&bdev->bd_inode->i_sem);
3019                                 bdput(bdev);
3020                         }
3021                 }
3022         }
3023         if (mddev->raid_disks    != info->raid_disks) {
3024                 /* change the number of raid disks */
3025                 if (mddev->pers->reshape == NULL)
3026                         return -EINVAL;
3027                 if (info->raid_disks <= 0 ||
3028                     info->raid_disks >= mddev->max_disks)
3029                         return -EINVAL;
3030                 if (mddev->sync_thread)
3031                         return -EBUSY;
3032                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3033                 if (!rv) {
3034                         struct block_device *bdev;
3035
3036                         bdev = bdget_disk(mddev->gendisk, 0);
3037                         if (bdev) {
3038                                 down(&bdev->bd_inode->i_sem);
3039                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3040                                 up(&bdev->bd_inode->i_sem);
3041                                 bdput(bdev);
3042                         }
3043                 }
3044         }
3045         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3046                 if (mddev->pers->quiesce == NULL)
3047                         return -EINVAL;
3048                 if (mddev->recovery || mddev->sync_thread)
3049                         return -EBUSY;
3050                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3051                         /* add the bitmap */
3052                         if (mddev->bitmap)
3053                                 return -EEXIST;
3054                         if (mddev->default_bitmap_offset == 0)
3055                                 return -EINVAL;
3056                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3057                         mddev->pers->quiesce(mddev, 1);
3058                         rv = bitmap_create(mddev);
3059                         if (rv)
3060                                 bitmap_destroy(mddev);
3061                         mddev->pers->quiesce(mddev, 0);
3062                 } else {
3063                         /* remove the bitmap */
3064                         if (!mddev->bitmap)
3065                                 return -ENOENT;
3066                         if (mddev->bitmap->file)
3067                                 return -EINVAL;
3068                         mddev->pers->quiesce(mddev, 1);
3069                         bitmap_destroy(mddev);
3070                         mddev->pers->quiesce(mddev, 0);
3071                         mddev->bitmap_offset = 0;
3072                 }
3073         }
3074         md_update_sb(mddev);
3075         return rv;
3076 }
3077
3078 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3079 {
3080         mdk_rdev_t *rdev;
3081
3082         if (mddev->pers == NULL)
3083                 return -ENODEV;
3084
3085         rdev = find_rdev(mddev, dev);
3086         if (!rdev)
3087                 return -ENODEV;
3088
3089         md_error(mddev, rdev);
3090         return 0;
3091 }
3092
3093 static int md_ioctl(struct inode *inode, struct file *file,
3094                         unsigned int cmd, unsigned long arg)
3095 {
3096         int err = 0;
3097         void __user *argp = (void __user *)arg;
3098         struct hd_geometry __user *loc = argp;
3099         mddev_t *mddev = NULL;
3100
3101         if (!capable(CAP_SYS_ADMIN))
3102                 return -EACCES;
3103
3104         /*
3105          * Commands dealing with the RAID driver but not any
3106          * particular array:
3107          */
3108         switch (cmd)
3109         {
3110                 case RAID_VERSION:
3111                         err = get_version(argp);
3112                         goto done;
3113
3114                 case PRINT_RAID_DEBUG:
3115                         err = 0;
3116                         md_print_devices();
3117                         goto done;
3118
3119 #ifndef MODULE
3120                 case RAID_AUTORUN:
3121                         err = 0;
3122                         autostart_arrays(arg);
3123                         goto done;
3124 #endif
3125                 default:;
3126         }
3127
3128         /*
3129          * Commands creating/starting a new array:
3130          */
3131
3132         mddev = inode->i_bdev->bd_disk->private_data;
3133
3134         if (!mddev) {
3135                 BUG();
3136                 goto abort;
3137         }
3138
3139
3140         if (cmd == START_ARRAY) {
3141                 /* START_ARRAY doesn't need to lock the array as autostart_array
3142                  * does the locking, and it could even be a different array
3143                  */
3144                 static int cnt = 3;
3145                 if (cnt > 0 ) {
3146                         printk(KERN_WARNING
3147                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3148                                "This will not be supported beyond 2.6\n",
3149                                current->comm, current->pid);
3150                         cnt--;
3151                 }
3152                 err = autostart_array(new_decode_dev(arg));
3153                 if (err) {
3154                         printk(KERN_WARNING "md: autostart failed!\n");
3155                         goto abort;
3156                 }
3157                 goto done;
3158         }
3159
3160         err = mddev_lock(mddev);
3161         if (err) {
3162                 printk(KERN_INFO
3163                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3164                         err, cmd);
3165                 goto abort;
3166         }
3167
3168         switch (cmd)
3169         {
3170                 case SET_ARRAY_INFO:
3171                         {
3172                                 mdu_array_info_t info;
3173                                 if (!arg)
3174                                         memset(&info, 0, sizeof(info));
3175                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3176                                         err = -EFAULT;
3177                                         goto abort_unlock;
3178                                 }
3179                                 if (mddev->pers) {
3180                                         err = update_array_info(mddev, &info);
3181                                         if (err) {
3182                                                 printk(KERN_WARNING "md: couldn't update"
3183                                                        " array info. %d\n", err);
3184                                                 goto abort_unlock;
3185                                         }
3186                                         goto done_unlock;
3187                                 }
3188                                 if (!list_empty(&mddev->disks)) {
3189                                         printk(KERN_WARNING
3190                                                "md: array %s already has disks!\n",
3191                                                mdname(mddev));
3192                                         err = -EBUSY;
3193                                         goto abort_unlock;
3194                                 }
3195                                 if (mddev->raid_disks) {
3196                                         printk(KERN_WARNING
3197                                                "md: array %s already initialised!\n",
3198                                                mdname(mddev));
3199                                         err = -EBUSY;
3200                                         goto abort_unlock;
3201                                 }
3202                                 err = set_array_info(mddev, &info);
3203                                 if (err) {
3204                                         printk(KERN_WARNING "md: couldn't set"
3205                                                " array info. %d\n", err);
3206                                         goto abort_unlock;
3207                                 }
3208                         }
3209                         goto done_unlock;
3210
3211                 default:;
3212         }
3213
3214         /*
3215          * Commands querying/configuring an existing array:
3216          */
3217         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3218          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3219         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3220                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3221                 err = -ENODEV;
3222                 goto abort_unlock;
3223         }
3224
3225         /*
3226          * Commands even a read-only array can execute:
3227          */
3228         switch (cmd)
3229         {
3230                 case GET_ARRAY_INFO:
3231                         err = get_array_info(mddev, argp);
3232                         goto done_unlock;
3233
3234                 case GET_BITMAP_FILE:
3235                         err = get_bitmap_file(mddev, argp);
3236                         goto done_unlock;
3237
3238                 case GET_DISK_INFO:
3239                         err = get_disk_info(mddev, argp);
3240                         goto done_unlock;
3241
3242                 case RESTART_ARRAY_RW:
3243                         err = restart_array(mddev);
3244                         goto done_unlock;
3245
3246                 case STOP_ARRAY:
3247                         err = do_md_stop (mddev, 0);
3248                         goto done_unlock;
3249
3250                 case STOP_ARRAY_RO:
3251                         err = do_md_stop (mddev, 1);
3252                         goto done_unlock;
3253
3254         /*
3255          * We have a problem here : there is no easy way to give a CHS
3256          * virtual geometry. We currently pretend that we have a 2 heads
3257          * 4 sectors (with a BIG number of cylinders...). This drives
3258          * dosfs just mad... ;-)
3259          */
3260                 case HDIO_GETGEO:
3261                         if (!loc) {
3262                                 err = -EINVAL;
3263                                 goto abort_unlock;
3264                         }
3265                         err = put_user (2, (char __user *) &loc->heads);
3266                         if (err)
3267                                 goto abort_unlock;
3268                         err = put_user (4, (char __user *) &loc->sectors);
3269                         if (err)
3270                                 goto abort_unlock;
3271                         err = put_user(get_capacity(mddev->gendisk)/8,
3272                                         (short __user *) &loc->cylinders);
3273                         if (err)
3274                                 goto abort_unlock;
3275                         err = put_user (get_start_sect(inode->i_bdev),
3276                                                 (long __user *) &loc->start);
3277                         goto done_unlock;
3278         }
3279
3280         /*
3281          * The remaining ioctls are changing the state of the
3282          * superblock, so we do not allow them on read-only arrays.
3283          * However non-MD ioctls (e.g. get-size) will still come through
3284          * here and hit the 'default' below, so only disallow
3285          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3286          */
3287         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3288             mddev->ro && mddev->pers) {
3289                 if (mddev->ro == 2) {
3290                         mddev->ro = 0;
3291                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3292                 md_wakeup_thread(mddev->thread);
3293
3294                 } else {
3295                         err = -EROFS;
3296                         goto abort_unlock;
3297                 }
3298         }
3299
3300         switch (cmd)
3301         {
3302                 case ADD_NEW_DISK:
3303                 {
3304                         mdu_disk_info_t info;
3305                         if (copy_from_user(&info, argp, sizeof(info)))
3306                                 err = -EFAULT;
3307                         else
3308                                 err = add_new_disk(mddev, &info);
3309                         goto done_unlock;
3310                 }
3311
3312                 case HOT_REMOVE_DISK:
3313                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3314                         goto done_unlock;
3315
3316                 case HOT_ADD_DISK:
3317                         err = hot_add_disk(mddev, new_decode_dev(arg));
3318                         goto done_unlock;
3319
3320                 case SET_DISK_FAULTY:
3321                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3322                         goto done_unlock;
3323
3324                 case RUN_ARRAY:
3325                         err = do_md_run (mddev);
3326                         goto done_unlock;
3327
3328                 case SET_BITMAP_FILE:
3329                         err = set_bitmap_file(mddev, (int)arg);
3330                         goto done_unlock;
3331
3332                 default:
3333                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3334                                 printk(KERN_WARNING "md: %s(pid %d) used"
3335                                         " obsolete MD ioctl, upgrade your"
3336                                         " software to use new ictls.\n",
3337                                         current->comm, current->pid);
3338                         err = -EINVAL;
3339                         goto abort_unlock;
3340         }
3341
3342 done_unlock:
3343 abort_unlock:
3344         mddev_unlock(mddev);
3345
3346         return err;
3347 done:
3348         if (err)
3349                 MD_BUG();
3350 abort:
3351         return err;
3352 }
3353
3354 static int md_open(struct inode *inode, struct file *file)
3355 {
3356         /*
3357          * Succeed if we can lock the mddev, which confirms that
3358          * it isn't being stopped right now.
3359          */
3360         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3361         int err;
3362
3363         if ((err = mddev_lock(mddev)))
3364                 goto out;
3365
3366         err = 0;
3367         mddev_get(mddev);
3368         mddev_unlock(mddev);
3369
3370         check_disk_change(inode->i_bdev);
3371  out:
3372         return err;
3373 }
3374
3375 static int md_release(struct inode *inode, struct file * file)
3376 {
3377         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3378
3379         if (!mddev)
3380                 BUG();
3381         mddev_put(mddev);
3382
3383         return 0;
3384 }
3385
3386 static int md_media_changed(struct gendisk *disk)
3387 {
3388         mddev_t *mddev = disk->private_data;
3389
3390         return mddev->changed;
3391 }
3392
3393 static int md_revalidate(struct gendisk *disk)
3394 {
3395         mddev_t *mddev = disk->private_data;
3396
3397         mddev->changed = 0;
3398         return 0;
3399 }
3400 static struct block_device_operations md_fops =
3401 {
3402         .owner          = THIS_MODULE,
3403         .open           = md_open,
3404         .release        = md_release,
3405         .ioctl          = md_ioctl,
3406         .media_changed  = md_media_changed,
3407         .revalidate_disk= md_revalidate,
3408 };
3409
3410 static int md_thread(void * arg)
3411 {
3412         mdk_thread_t *thread = arg;
3413
3414         /*
3415          * md_thread is a 'system-thread', it's priority should be very
3416          * high. We avoid resource deadlocks individually in each
3417          * raid personality. (RAID5 does preallocation) We also use RR and
3418          * the very same RT priority as kswapd, thus we will never get
3419          * into a priority inversion deadlock.
3420          *
3421          * we definitely have to have equal or higher priority than
3422          * bdflush, otherwise bdflush will deadlock if there are too
3423          * many dirty RAID5 blocks.
3424          */
3425
3426         allow_signal(SIGKILL);
3427         complete(thread->event);
3428         while (!kthread_should_stop()) {
3429                 void (*run)(mddev_t *);
3430
3431                 wait_event_interruptible_timeout(thread->wqueue,
3432                                                  test_bit(THREAD_WAKEUP, &thread->flags)
3433                                                  || kthread_should_stop(),
3434                                                  thread->timeout);
3435                 try_to_freeze();
3436
3437                 clear_bit(THREAD_WAKEUP, &thread->flags);
3438
3439                 run = thread->run;
3440                 if (run)
3441                         run(thread->mddev);
3442         }
3443
3444         return 0;
3445 }
3446
3447 void md_wakeup_thread(mdk_thread_t *thread)
3448 {
3449         if (thread) {
3450                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3451                 set_bit(THREAD_WAKEUP, &thread->flags);
3452                 wake_up(&thread->wqueue);
3453         }
3454 }
3455
3456 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3457                                  const char *name)
3458 {
3459         mdk_thread_t *thread;
3460         struct completion event;
3461
3462         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3463         if (!thread)
3464                 return NULL;
3465
3466         memset(thread, 0, sizeof(mdk_thread_t));
3467         init_waitqueue_head(&thread->wqueue);
3468
3469         init_completion(&event);
3470         thread->event = &event;
3471         thread->run = run;
3472         thread->mddev = mddev;
3473         thread->name = name;
3474         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3475         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3476         if (IS_ERR(thread->tsk)) {
3477                 kfree(thread);
3478                 return NULL;
3479         }
3480         wait_for_completion(&event);
3481         return thread;
3482 }
3483
3484 void md_unregister_thread(mdk_thread_t *thread)
3485 {
3486         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3487
3488         kthread_stop(thread->tsk);
3489         kfree(thread);
3490 }
3491
3492 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3493 {
3494         if (!mddev) {
3495                 MD_BUG();
3496                 return;
3497         }
3498
3499         if (!rdev || test_bit(Faulty, &rdev->flags))
3500                 return;
3501 /*
3502         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3503                 mdname(mddev),
3504                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3505                 __builtin_return_address(0),__builtin_return_address(1),
3506                 __builtin_return_address(2),__builtin_return_address(3));
3507 */
3508         if (!mddev->pers->error_handler)
3509                 return;
3510         mddev->pers->error_handler(mddev,rdev);
3511         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3512         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3513         md_wakeup_thread(mddev->thread);
3514 }
3515
3516 /* seq_file implementation /proc/mdstat */
3517
3518 static void status_unused(struct seq_file *seq)
3519 {
3520         int i = 0;
3521         mdk_rdev_t *rdev;
3522         struct list_head *tmp;
3523
3524         seq_printf(seq, "unused devices: ");
3525
3526         ITERATE_RDEV_PENDING(rdev,tmp) {
3527                 char b[BDEVNAME_SIZE];
3528                 i++;
3529                 seq_printf(seq, "%s ",
3530                               bdevname(rdev->bdev,b));
3531         }
3532         if (!i)
3533                 seq_printf(seq, "<none>");
3534
3535         seq_printf(seq, "\n");
3536 }
3537
3538
3539 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3540 {
3541         unsigned long max_blocks, resync, res, dt, db, rt;
3542
3543         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3544
3545         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3546                 max_blocks = mddev->resync_max_sectors >> 1;
3547         else
3548                 max_blocks = mddev->size;
3549
3550         /*
3551          * Should not happen.
3552          */
3553         if (!max_blocks) {
3554                 MD_BUG();
3555                 return;
3556         }
3557         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3558         {
3559                 int i, x = res/50, y = 20-x;
3560                 seq_printf(seq, "[");
3561                 for (i = 0; i < x; i++)
3562                         seq_printf(seq, "=");
3563                 seq_printf(seq, ">");
3564                 for (i = 0; i < y; i++)
3565                         seq_printf(seq, ".");
3566                 seq_printf(seq, "] ");
3567         }
3568         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3569                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3570                        "resync" : "recovery"),
3571                       res/10, res % 10, resync, max_blocks);
3572
3573         /*
3574          * We do not want to overflow, so the order of operands and
3575          * the * 100 / 100 trick are important. We do a +1 to be
3576          * safe against division by zero. We only estimate anyway.
3577          *
3578          * dt: time from mark until now
3579          * db: blocks written from mark until now
3580          * rt: remaining time
3581          */
3582         dt = ((jiffies - mddev->resync_mark) / HZ);
3583         if (!dt) dt++;
3584         db = resync - (mddev->resync_mark_cnt/2);
3585         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3586
3587         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3588
3589         seq_printf(seq, " speed=%ldK/sec", db/dt);
3590 }
3591
3592 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3593 {
3594         struct list_head *tmp;
3595         loff_t l = *pos;
3596         mddev_t *mddev;
3597
3598         if (l >= 0x10000)
3599                 return NULL;
3600         if (!l--)
3601                 /* header */
3602                 return (void*)1;
3603
3604         spin_lock(&all_mddevs_lock);
3605         list_for_each(tmp,&all_mddevs)
3606                 if (!l--) {
3607                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3608                         mddev_get(mddev);
3609                         spin_unlock(&all_mddevs_lock);
3610                         return mddev;
3611                 }
3612         spin_unlock(&all_mddevs_lock);
3613         if (!l--)
3614                 return (void*)2;/* tail */
3615         return NULL;
3616 }
3617
3618 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3619 {
3620         struct list_head *tmp;
3621         mddev_t *next_mddev, *mddev = v;
3622
3623         ++*pos;
3624         if (v == (void*)2)
3625                 return NULL;
3626
3627         spin_lock(&all_mddevs_lock);
3628         if (v == (void*)1)
3629                 tmp = all_mddevs.next;
3630         else
3631                 tmp = mddev->all_mddevs.next;
3632         if (tmp != &all_mddevs)
3633                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3634         else {
3635                 next_mddev = (void*)2;
3636                 *pos = 0x10000;
3637         }
3638         spin_unlock(&all_mddevs_lock);
3639
3640         if (v != (void*)1)
3641                 mddev_put(mddev);
3642         return next_mddev;
3643
3644 }
3645
3646 static void md_seq_stop(struct seq_file *seq, void *v)
3647 {
3648         mddev_t *mddev = v;
3649
3650         if (mddev && v != (void*)1 && v != (void*)2)
3651                 mddev_put(mddev);
3652 }
3653
3654 static int md_seq_show(struct seq_file *seq, void *v)
3655 {
3656         mddev_t *mddev = v;
3657         sector_t size;
3658         struct list_head *tmp2;
3659         mdk_rdev_t *rdev;
3660         int i;
3661         struct bitmap *bitmap;
3662
3663         if (v == (void*)1) {
3664                 seq_printf(seq, "Personalities : ");
3665                 spin_lock(&pers_lock);
3666                 for (i = 0; i < MAX_PERSONALITY; i++)
3667                         if (pers[i])
3668                                 seq_printf(seq, "[%s] ", pers[i]->name);
3669
3670                 spin_unlock(&pers_lock);
3671                 seq_printf(seq, "\n");
3672                 return 0;
3673         }
3674         if (v == (void*)2) {
3675                 status_unused(seq);
3676                 return 0;
3677         }
3678
3679         if (mddev_lock(mddev)!=0)
3680                 return -EINTR;
3681         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3682                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3683                                                 mddev->pers ? "" : "in");
3684                 if (mddev->pers) {
3685                         if (mddev->ro==1)
3686                                 seq_printf(seq, " (read-only)");
3687                         if (mddev->ro==2)
3688                                 seq_printf(seq, "(auto-read-only)");
3689                         seq_printf(seq, " %s", mddev->pers->name);
3690                 }
3691
3692                 size = 0;
3693                 ITERATE_RDEV(mddev,rdev,tmp2) {
3694                         char b[BDEVNAME_SIZE];
3695                         seq_printf(seq, " %s[%d]",
3696                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3697                         if (test_bit(WriteMostly, &rdev->flags))
3698                                 seq_printf(seq, "(W)");
3699                         if (test_bit(Faulty, &rdev->flags)) {
3700                                 seq_printf(seq, "(F)");
3701                                 continue;
3702                         } else if (rdev->raid_disk < 0)
3703                                 seq_printf(seq, "(S)"); /* spare */
3704                         size += rdev->size;
3705                 }
3706
3707                 if (!list_empty(&mddev->disks)) {
3708                         if (mddev->pers)
3709                                 seq_printf(seq, "\n      %llu blocks",
3710                                         (unsigned long long)mddev->array_size);
3711                         else
3712                                 seq_printf(seq, "\n      %llu blocks",
3713                                         (unsigned long long)size);
3714                 }
3715                 if (mddev->persistent) {
3716                         if (mddev->major_version != 0 ||
3717                             mddev->minor_version != 90) {
3718                                 seq_printf(seq," super %d.%d",
3719                                            mddev->major_version,
3720                                            mddev->minor_version);
3721                         }
3722                 } else
3723                         seq_printf(seq, " super non-persistent");
3724
3725                 if (mddev->pers) {
3726                         mddev->pers->status (seq, mddev);
3727                         seq_printf(seq, "\n      ");
3728                         if (mddev->pers->sync_request) {
3729                                 if (mddev->curr_resync > 2) {
3730                                         status_resync (seq, mddev);
3731                                         seq_printf(seq, "\n      ");
3732                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3733                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3734                                 else if (mddev->recovery_cp < MaxSector)
3735                                         seq_printf(seq, "\tresync=PENDING\n      ");
3736                         }
3737                 } else
3738                         seq_printf(seq, "\n       ");
3739
3740                 if ((bitmap = mddev->bitmap)) {
3741                         unsigned long chunk_kb;
3742                         unsigned long flags;
3743                         spin_lock_irqsave(&bitmap->lock, flags);
3744                         chunk_kb = bitmap->chunksize >> 10;
3745                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3746                                 "%lu%s chunk",
3747                                 bitmap->pages - bitmap->missing_pages,
3748                                 bitmap->pages,
3749                                 (bitmap->pages - bitmap->missing_pages)
3750                                         << (PAGE_SHIFT - 10),
3751                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3752                                 chunk_kb ? "KB" : "B");
3753                         if (bitmap->file) {
3754                                 seq_printf(seq, ", file: ");
3755                                 seq_path(seq, bitmap->file->f_vfsmnt,
3756                                          bitmap->file->f_dentry," \t\n");
3757                         }
3758
3759                         seq_printf(seq, "\n");
3760                         spin_unlock_irqrestore(&bitmap->lock, flags);
3761                 }
3762
3763                 seq_printf(seq, "\n");
3764         }
3765         mddev_unlock(mddev);
3766
3767         return 0;
3768 }
3769
3770 static struct seq_operations md_seq_ops = {
3771         .start  = md_seq_start,
3772         .next   = md_seq_next,
3773         .stop   = md_seq_stop,
3774         .show   = md_seq_show,
3775 };
3776
3777 static int md_seq_open(struct inode *inode, struct file *file)
3778 {
3779         int error;
3780
3781         error = seq_open(file, &md_seq_ops);
3782         return error;
3783 }
3784
3785 static struct file_operations md_seq_fops = {
3786         .open           = md_seq_open,
3787         .read           = seq_read,
3788         .llseek         = seq_lseek,
3789         .release        = seq_release,
3790 };
3791
3792 int register_md_personality(int pnum, mdk_personality_t *p)
3793 {
3794         if (pnum >= MAX_PERSONALITY) {
3795                 printk(KERN_ERR
3796                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3797                        p->name, pnum, MAX_PERSONALITY-1);
3798                 return -EINVAL;
3799         }
3800
3801         spin_lock(&pers_lock);
3802         if (pers[pnum]) {
3803                 spin_unlock(&pers_lock);
3804                 return -EBUSY;
3805         }
3806
3807         pers[pnum] = p;
3808         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3809         spin_unlock(&pers_lock);
3810         return 0;
3811 }
3812
3813 int unregister_md_personality(int pnum)
3814 {
3815         if (pnum >= MAX_PERSONALITY)
3816                 return -EINVAL;
3817
3818         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3819         spin_lock(&pers_lock);
3820         pers[pnum] = NULL;
3821         spin_unlock(&pers_lock);
3822         return 0;
3823 }
3824
3825 static int is_mddev_idle(mddev_t *mddev)
3826 {
3827         mdk_rdev_t * rdev;
3828         struct list_head *tmp;
3829         int idle;
3830         unsigned long curr_events;
3831
3832         idle = 1;
3833         ITERATE_RDEV(mddev,rdev,tmp) {
3834                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3835                 curr_events = disk_stat_read(disk, sectors[0]) +
3836                                 disk_stat_read(disk, sectors[1]) -
3837                                 atomic_read(&disk->sync_io);
3838                 /* Allow some slack between valud of curr_events and last_events,
3839                  * as there are some uninteresting races.
3840                  * Note: the following is an unsigned comparison.
3841                  */
3842                 if ((curr_events - rdev->last_events + 32) > 64) {
3843                         rdev->last_events = curr_events;
3844                         idle = 0;
3845                 }
3846         }
3847         return idle;
3848 }
3849
3850 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3851 {
3852         /* another "blocks" (512byte) blocks have been synced */
3853         atomic_sub(blocks, &mddev->recovery_active);
3854         wake_up(&mddev->recovery_wait);
3855         if (!ok) {
3856                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3857                 md_wakeup_thread(mddev->thread);
3858                 // stop recovery, signal do_sync ....
3859         }
3860 }
3861
3862
3863 /* md_write_start(mddev, bi)
3864  * If we need to update some array metadata (e.g. 'active' flag
3865  * in superblock) before writing, schedule a superblock update
3866  * and wait for it to complete.
3867  */
3868 void md_write_start(mddev_t *mddev, struct bio *bi)
3869 {
3870         if (bio_data_dir(bi) != WRITE)
3871                 return;
3872
3873         BUG_ON(mddev->ro == 1);
3874         if (mddev->ro == 2) {
3875                 /* need to switch to read/write */
3876                 mddev->ro = 0;
3877                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3878                 md_wakeup_thread(mddev->thread);
3879         }
3880         atomic_inc(&mddev->writes_pending);
3881         if (mddev->in_sync) {
3882                 spin_lock_irq(&mddev->write_lock);
3883                 if (mddev->in_sync) {
3884                         mddev->in_sync = 0;
3885                         mddev->sb_dirty = 1;
3886                         md_wakeup_thread(mddev->thread);
3887                 }
3888                 spin_unlock_irq(&mddev->write_lock);
3889         }
3890         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3891 }
3892
3893 void md_write_end(mddev_t *mddev)
3894 {
3895         if (atomic_dec_and_test(&mddev->writes_pending)) {
3896                 if (mddev->safemode == 2)
3897                         md_wakeup_thread(mddev->thread);
3898                 else
3899                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3900         }
3901 }
3902
3903 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3904
3905 #define SYNC_MARKS      10
3906 #define SYNC_MARK_STEP  (3*HZ)
3907 static void md_do_sync(mddev_t *mddev)
3908 {
3909         mddev_t *mddev2;
3910         unsigned int currspeed = 0,
3911                  window;
3912         sector_t max_sectors,j, io_sectors;
3913         unsigned long mark[SYNC_MARKS];
3914         sector_t mark_cnt[SYNC_MARKS];
3915         int last_mark,m;
3916         struct list_head *tmp;
3917         sector_t last_check;
3918         int skipped = 0;
3919
3920         /* just incase thread restarts... */
3921         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3922                 return;
3923
3924         /* we overload curr_resync somewhat here.
3925          * 0 == not engaged in resync at all
3926          * 2 == checking that there is no conflict with another sync
3927          * 1 == like 2, but have yielded to allow conflicting resync to
3928          *              commense
3929          * other == active in resync - this many blocks
3930          *
3931          * Before starting a resync we must have set curr_resync to
3932          * 2, and then checked that every "conflicting" array has curr_resync
3933          * less than ours.  When we find one that is the same or higher
3934          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3935          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3936          * This will mean we have to start checking from the beginning again.
3937          *
3938          */
3939
3940         do {
3941                 mddev->curr_resync = 2;
3942
3943         try_again:
3944                 if (signal_pending(current) ||
3945                     kthread_should_stop()) {
3946                         flush_signals(current);
3947                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3948                         goto skip;
3949                 }
3950                 ITERATE_MDDEV(mddev2,tmp) {
3951                         if (mddev2 == mddev)
3952                                 continue;
3953                         if (mddev2->curr_resync &&
3954                             match_mddev_units(mddev,mddev2)) {
3955                                 DEFINE_WAIT(wq);
3956                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3957                                         /* arbitrarily yield */
3958                                         mddev->curr_resync = 1;
3959                                         wake_up(&resync_wait);
3960                                 }
3961                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3962                                         /* no need to wait here, we can wait the next
3963                                          * time 'round when curr_resync == 2
3964                                          */
3965                                         continue;
3966                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3967                                 if (!signal_pending(current) &&
3968                                     !kthread_should_stop() &&
3969                                     mddev2->curr_resync >= mddev->curr_resync) {
3970                                         printk(KERN_INFO "md: delaying resync of %s"
3971                                                " until %s has finished resync (they"
3972                                                " share one or more physical units)\n",
3973                                                mdname(mddev), mdname(mddev2));
3974                                         mddev_put(mddev2);
3975                                         schedule();
3976                                         finish_wait(&resync_wait, &wq);
3977                                         goto try_again;
3978                                 }
3979                                 finish_wait(&resync_wait, &wq);
3980                         }
3981                 }
3982         } while (mddev->curr_resync < 2);
3983
3984         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3985                 /* resync follows the size requested by the personality,
3986                  * which defaults to physical size, but can be virtual size
3987                  */
3988                 max_sectors = mddev->resync_max_sectors;
3989                 mddev->resync_mismatches = 0;
3990         } else
3991                 /* recovery follows the physical size of devices */
3992                 max_sectors = mddev->size << 1;
3993
3994         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3995         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3996                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3997         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3998                "(but not more than %d KB/sec) for reconstruction.\n",
3999                sysctl_speed_limit_max);
4000
4001         is_mddev_idle(mddev); /* this also initializes IO event counters */
4002         /* we don't use the checkpoint if there's a bitmap */
4003         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4004             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4005                 j = mddev->recovery_cp;
4006         else
4007                 j = 0;
4008         io_sectors = 0;
4009         for (m = 0; m < SYNC_MARKS; m++) {
4010                 mark[m] = jiffies;
4011                 mark_cnt[m] = io_sectors;
4012         }
4013         last_mark = 0;
4014         mddev->resync_mark = mark[last_mark];
4015         mddev->resync_mark_cnt = mark_cnt[last_mark];
4016
4017         /*
4018          * Tune reconstruction:
4019          */
4020         window = 32*(PAGE_SIZE/512);
4021         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4022                 window/2,(unsigned long long) max_sectors/2);
4023
4024         atomic_set(&mddev->recovery_active, 0);
4025         init_waitqueue_head(&mddev->recovery_wait);
4026         last_check = 0;
4027
4028         if (j>2) {
4029                 printk(KERN_INFO
4030                         "md: resuming recovery of %s from checkpoint.\n",
4031                         mdname(mddev));
4032                 mddev->curr_resync = j;
4033         }
4034
4035         while (j < max_sectors) {
4036                 sector_t sectors;
4037
4038                 skipped = 0;
4039                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4040                                             currspeed < sysctl_speed_limit_min);
4041                 if (sectors == 0) {
4042                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4043                         goto out;
4044                 }
4045
4046                 if (!skipped) { /* actual IO requested */
4047                         io_sectors += sectors;
4048                         atomic_add(sectors, &mddev->recovery_active);
4049                 }
4050
4051                 j += sectors;
4052                 if (j>1) mddev->curr_resync = j;
4053
4054
4055                 if (last_check + window > io_sectors || j == max_sectors)
4056                         continue;
4057
4058                 last_check = io_sectors;
4059
4060                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4061                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4062                         break;
4063
4064         repeat:
4065                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4066                         /* step marks */
4067                         int next = (last_mark+1) % SYNC_MARKS;
4068
4069                         mddev->resync_mark = mark[next];
4070                         mddev->resync_mark_cnt = mark_cnt[next];
4071                         mark[next] = jiffies;
4072                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4073                         last_mark = next;
4074                 }
4075
4076
4077                 if (signal_pending(current) || kthread_should_stop()) {
4078                         /*
4079                          * got a signal, exit.
4080                          */
4081                         printk(KERN_INFO
4082                                 "md: md_do_sync() got signal ... exiting\n");
4083                         flush_signals(current);
4084                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4085                         goto out;
4086                 }
4087
4088                 /*
4089                  * this loop exits only if either when we are slower than
4090                  * the 'hard' speed limit, or the system was IO-idle for
4091                  * a jiffy.
4092                  * the system might be non-idle CPU-wise, but we only care
4093                  * about not overloading the IO subsystem. (things like an
4094                  * e2fsck being done on the RAID array should execute fast)
4095                  */
4096                 mddev->queue->unplug_fn(mddev->queue);
4097                 cond_resched();
4098
4099                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4100                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4101
4102                 if (currspeed > sysctl_speed_limit_min) {
4103                         if ((currspeed > sysctl_speed_limit_max) ||
4104                                         !is_mddev_idle(mddev)) {
4105                                 msleep_interruptible(250);
4106                                 goto repeat;
4107                         }
4108                 }
4109         }
4110         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4111         /*
4112          * this also signals 'finished resyncing' to md_stop
4113          */
4114  out:
4115         mddev->queue->unplug_fn(mddev->queue);
4116
4117         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4118
4119         /* tell personality that we are finished */
4120         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4121
4122         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4123             mddev->curr_resync > 2 &&
4124             mddev->curr_resync >= mddev->recovery_cp) {
4125                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4126                         printk(KERN_INFO
4127                                 "md: checkpointing recovery of %s.\n",
4128                                 mdname(mddev));
4129                         mddev->recovery_cp = mddev->curr_resync;
4130                 } else
4131                         mddev->recovery_cp = MaxSector;
4132         }
4133
4134  skip:
4135         mddev->curr_resync = 0;
4136         wake_up(&resync_wait);
4137         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4138         md_wakeup_thread(mddev->thread);
4139 }
4140
4141
4142 /*
4143  * This routine is regularly called by all per-raid-array threads to
4144  * deal with generic issues like resync and super-block update.
4145  * Raid personalities that don't have a thread (linear/raid0) do not
4146  * need this as they never do any recovery or update the superblock.
4147  *
4148  * It does not do any resync itself, but rather "forks" off other threads
4149  * to do that as needed.
4150  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4151  * "->recovery" and create a thread at ->sync_thread.
4152  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4153  * and wakeups up this thread which will reap the thread and finish up.
4154  * This thread also removes any faulty devices (with nr_pending == 0).
4155  *
4156  * The overall approach is:
4157  *  1/ if the superblock needs updating, update it.
4158  *  2/ If a recovery thread is running, don't do anything else.
4159  *  3/ If recovery has finished, clean up, possibly marking spares active.
4160  *  4/ If there are any faulty devices, remove them.
4161  *  5/ If array is degraded, try to add spares devices
4162  *  6/ If array has spares or is not in-sync, start a resync thread.
4163  */
4164 void md_check_recovery(mddev_t *mddev)
4165 {
4166         mdk_rdev_t *rdev;
4167         struct list_head *rtmp;
4168
4169
4170         if (mddev->bitmap)
4171                 bitmap_daemon_work(mddev->bitmap);
4172
4173         if (mddev->ro)
4174                 return;
4175
4176         if (signal_pending(current)) {
4177                 if (mddev->pers->sync_request) {
4178                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4179                                mdname(mddev));
4180                         mddev->safemode = 2;
4181                 }
4182                 flush_signals(current);
4183         }
4184
4185         if ( ! (
4186                 mddev->sb_dirty ||
4187                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4188                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4189                 (mddev->safemode == 1) ||
4190                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4191                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4192                 ))
4193                 return;
4194
4195         if (mddev_trylock(mddev)==0) {
4196                 int spares =0;
4197
4198                 spin_lock_irq(&mddev->write_lock);
4199                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4200                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4201                         mddev->in_sync = 1;
4202                         mddev->sb_dirty = 1;
4203                 }
4204                 if (mddev->safemode == 1)
4205                         mddev->safemode = 0;
4206                 spin_unlock_irq(&mddev->write_lock);
4207
4208                 if (mddev->sb_dirty)
4209                         md_update_sb(mddev);
4210
4211
4212                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4213                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4214                         /* resync/recovery still happening */
4215                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4216                         goto unlock;
4217                 }
4218                 if (mddev->sync_thread) {
4219                         /* resync has finished, collect result */
4220                         md_unregister_thread(mddev->sync_thread);
4221                         mddev->sync_thread = NULL;
4222                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4223                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4224                                 /* success...*/
4225                                 /* activate any spares */
4226                                 mddev->pers->spare_active(mddev);
4227                         }
4228                         md_update_sb(mddev);
4229
4230                         /* if array is no-longer degraded, then any saved_raid_disk
4231                          * information must be scrapped
4232                          */
4233                         if (!mddev->degraded)
4234                                 ITERATE_RDEV(mddev,rdev,rtmp)
4235                                         rdev->saved_raid_disk = -1;
4236
4237                         mddev->recovery = 0;
4238                         /* flag recovery needed just to double check */
4239                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4240                         goto unlock;
4241                 }
4242                 /* Clear some bits that don't mean anything, but
4243                  * might be left set
4244                  */
4245                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4246                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4247                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4248                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4249
4250                 /* no recovery is running.
4251                  * remove any failed drives, then
4252                  * add spares if possible.
4253                  * Spare are also removed and re-added, to allow
4254                  * the personality to fail the re-add.
4255                  */
4256                 ITERATE_RDEV(mddev,rdev,rtmp)
4257                         if (rdev->raid_disk >= 0 &&
4258                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4259                             atomic_read(&rdev->nr_pending)==0) {
4260                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4261                                         char nm[20];
4262                                         sprintf(nm,"rd%d", rdev->raid_disk);
4263                                         sysfs_remove_link(&mddev->kobj, nm);
4264                                         rdev->raid_disk = -1;
4265                                 }
4266                         }
4267
4268                 if (mddev->degraded) {
4269                         ITERATE_RDEV(mddev,rdev,rtmp)
4270                                 if (rdev->raid_disk < 0
4271                                     && !test_bit(Faulty, &rdev->flags)) {
4272                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4273                                                 char nm[20];
4274                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4275                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4276                                                 spares++;
4277                                         } else
4278                                                 break;
4279                                 }
4280                 }
4281
4282                 if (spares) {
4283                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4284                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4285                 } else if (mddev->recovery_cp < MaxSector) {
4286                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4287                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4288                         /* nothing to be done ... */
4289                         goto unlock;
4290
4291                 if (mddev->pers->sync_request) {
4292                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4293                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4294                                 /* We are adding a device or devices to an array
4295                                  * which has the bitmap stored on all devices.
4296                                  * So make sure all bitmap pages get written
4297                                  */
4298                                 bitmap_write_all(mddev->bitmap);
4299                         }
4300                         mddev->sync_thread = md_register_thread(md_do_sync,
4301                                                                 mddev,
4302                                                                 "%s_resync");
4303                         if (!mddev->sync_thread) {
4304                                 printk(KERN_ERR "%s: could not start resync"
4305                                         " thread...\n",
4306                                         mdname(mddev));
4307                                 /* leave the spares where they are, it shouldn't hurt */
4308                                 mddev->recovery = 0;
4309                         } else {
4310                                 md_wakeup_thread(mddev->sync_thread);
4311                         }
4312                 }
4313         unlock:
4314                 mddev_unlock(mddev);
4315         }
4316 }
4317
4318 static int md_notify_reboot(struct notifier_block *this,
4319                             unsigned long code, void *x)
4320 {
4321         struct list_head *tmp;
4322         mddev_t *mddev;
4323
4324         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4325
4326                 printk(KERN_INFO "md: stopping all md devices.\n");
4327
4328                 ITERATE_MDDEV(mddev,tmp)
4329                         if (mddev_trylock(mddev)==0)
4330                                 do_md_stop (mddev, 1);
4331                 /*
4332                  * certain more exotic SCSI devices are known to be
4333                  * volatile wrt too early system reboots. While the
4334                  * right place to handle this issue is the given
4335                  * driver, we do want to have a safe RAID driver ...
4336                  */
4337                 mdelay(1000*1);
4338         }
4339         return NOTIFY_DONE;
4340 }
4341
4342 static struct notifier_block md_notifier = {
4343         .notifier_call  = md_notify_reboot,
4344         .next           = NULL,
4345         .priority       = INT_MAX, /* before any real devices */
4346 };
4347
4348 static void md_geninit(void)
4349 {
4350         struct proc_dir_entry *p;
4351
4352         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4353
4354         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4355         if (p)
4356                 p->proc_fops = &md_seq_fops;
4357 }
4358
4359 static int __init md_init(void)
4360 {
4361         int minor;
4362
4363         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4364                         " MD_SB_DISKS=%d\n",
4365                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4366                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4367         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4368                         BITMAP_MINOR);
4369
4370         if (register_blkdev(MAJOR_NR, "md"))
4371                 return -1;
4372         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4373                 unregister_blkdev(MAJOR_NR, "md");
4374                 return -1;
4375         }
4376         devfs_mk_dir("md");
4377         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4378                                 md_probe, NULL, NULL);
4379         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4380                             md_probe, NULL, NULL);
4381
4382         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4383                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4384                                 S_IFBLK|S_IRUSR|S_IWUSR,
4385                                 "md/%d", minor);
4386
4387         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4388                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4389                               S_IFBLK|S_IRUSR|S_IWUSR,
4390                               "md/mdp%d", minor);
4391
4392
4393         register_reboot_notifier(&md_notifier);
4394         raid_table_header = register_sysctl_table(raid_root_table, 1);
4395
4396         md_geninit();
4397         return (0);
4398 }
4399
4400
4401 #ifndef MODULE
4402
4403 /*
4404  * Searches all registered partitions for autorun RAID arrays
4405  * at boot time.
4406  */
4407 static dev_t detected_devices[128];
4408 static int dev_cnt;
4409
4410 void md_autodetect_dev(dev_t dev)
4411 {
4412         if (dev_cnt >= 0 && dev_cnt < 127)
4413                 detected_devices[dev_cnt++] = dev;
4414 }
4415
4416
4417 static void autostart_arrays(int part)
4418 {
4419         mdk_rdev_t *rdev;
4420         int i;
4421
4422         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4423
4424         for (i = 0; i < dev_cnt; i++) {
4425                 dev_t dev = detected_devices[i];
4426
4427                 rdev = md_import_device(dev,0, 0);
4428                 if (IS_ERR(rdev))
4429                         continue;
4430
4431                 if (test_bit(Faulty, &rdev->flags)) {
4432                         MD_BUG();
4433                         continue;
4434                 }
4435                 list_add(&rdev->same_set, &pending_raid_disks);
4436         }
4437         dev_cnt = 0;
4438
4439         autorun_devices(part);
4440 }
4441
4442 #endif
4443
4444 static __exit void md_exit(void)
4445 {
4446         mddev_t *mddev;
4447         struct list_head *tmp;
4448         int i;
4449         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4450         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4451         for (i=0; i < MAX_MD_DEVS; i++)
4452                 devfs_remove("md/%d", i);
4453         for (i=0; i < MAX_MD_DEVS; i++)
4454                 devfs_remove("md/d%d", i);
4455
4456         devfs_remove("md");
4457
4458         unregister_blkdev(MAJOR_NR,"md");
4459         unregister_blkdev(mdp_major, "mdp");
4460         unregister_reboot_notifier(&md_notifier);
4461         unregister_sysctl_table(raid_table_header);
4462         remove_proc_entry("mdstat", NULL);
4463         ITERATE_MDDEV(mddev,tmp) {
4464                 struct gendisk *disk = mddev->gendisk;
4465                 if (!disk)
4466                         continue;
4467                 export_array(mddev);
4468                 del_gendisk(disk);
4469                 put_disk(disk);
4470                 mddev->gendisk = NULL;
4471                 mddev_put(mddev);
4472         }
4473 }
4474
4475 module_init(md_init)
4476 module_exit(md_exit)
4477
4478 static int get_ro(char *buffer, struct kernel_param *kp)
4479 {
4480         return sprintf(buffer, "%d", start_readonly);
4481 }
4482 static int set_ro(const char *val, struct kernel_param *kp)
4483 {
4484         char *e;
4485         int num = simple_strtoul(val, &e, 10);
4486         if (*val && (*e == '\0' || *e == '\n')) {
4487                 start_readonly = num;
4488                 return 0;;
4489         }
4490         return -EINVAL;
4491 }
4492
4493 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4494
4495 EXPORT_SYMBOL(register_md_personality);
4496 EXPORT_SYMBOL(unregister_md_personality);
4497 EXPORT_SYMBOL(md_error);
4498 EXPORT_SYMBOL(md_done_sync);
4499 EXPORT_SYMBOL(md_write_start);
4500 EXPORT_SYMBOL(md_write_end);
4501 EXPORT_SYMBOL(md_register_thread);
4502 EXPORT_SYMBOL(md_unregister_thread);
4503 EXPORT_SYMBOL(md_wakeup_thread);
4504 EXPORT_SYMBOL(md_print_devices);
4505 EXPORT_SYMBOL(md_check_recovery);
4506 MODULE_LICENSE("GPL");
4507 MODULE_ALIAS("md");
4508 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);