drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45
  46 #include <linux/init.h>
  47
  48 #include <linux/file.h>
  49
  50 #ifdef CONFIG_KMOD
  51 #include <linux/kmod.h>
  52 #endif
  53
  54 #include <asm/unaligned.h>
  55
  56 #define MAJOR_NR MD_MAJOR
  57 #define MD_DRIVER
  58
  59 /* 63 partitions with the alternate major number (mdp) */
  60 #define MdpMinorShift 6
  61
  62 #define DEBUG 0
  63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  64
  65
  66 #ifndef MODULE
  67 static void autostart_arrays (int part);
  68 #endif
  69
  70 static mdk_personality_t *pers[MAX_PERSONALITY];
  71 static DEFINE_SPINLOCK(pers_lock);
  72
  73 /*
  74  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  75  * is 1000 KB/sec, so the extra system load does not show up that much.
  76  * Increase it if you want to have more _guaranteed_ speed. Note that
  77  * the RAID driver will use the maximum available bandwidth if the IO
  78  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  79  * speed limit - in case reconstruction slows down your system despite
  80  * idle IO detection.
  81  *
  82  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  83  */
  84
  85 static int sysctl_speed_limit_min = 1000;
  86 static int sysctl_speed_limit_max = 200000;
  87
  88 static struct ctl_table_header *raid_table_header;
  89
  90 static ctl_table raid_table[] = {
  91         {
  92                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  93                 .procname       = "speed_limit_min",
  94                 .data           = &sysctl_speed_limit_min,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec,
  98         },
  99         {
 100                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 101                 .procname       = "speed_limit_max",
 102                 .data           = &sysctl_speed_limit_max,
 103                 .maxlen         = sizeof(int),
 104                 .mode           = 0644,
 105                 .proc_handler   = &proc_dointvec,
 106         },
 107         { .ctl_name = 0 }
 108 };
 109
 110 static ctl_table raid_dir_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID,
 113                 .procname       = "raid",
 114                 .maxlen         = 0,
 115                 .mode           = 0555,
 116                 .child          = raid_table,
 117         },
 118         { .ctl_name = 0 }
 119 };
 120
 121 static ctl_table raid_root_table[] = {
 122         {
 123                 .ctl_name       = CTL_DEV,
 124                 .procname       = "dev",
 125                 .maxlen         = 0,
 126                 .mode           = 0555,
 127                 .child          = raid_dir_table,
 128         },
 129         { .ctl_name = 0 }
 130 };
 131
 132 static struct block_device_operations md_fops;
 133
 134 static int start_readonly;
 135
 136 /*
 137  * Enables to iterate over all existing md arrays
 138  * all_mddevs_lock protects this list.
 139  */
 140 static LIST_HEAD(all_mddevs);
 141 static DEFINE_SPINLOCK(all_mddevs_lock);
 142
 143
 144 /*
 145  * iterates through all used mddevs in the system.
 146  * We take care to grab the all_mddevs_lock whenever navigating
 147  * the list, and to always hold a refcount when unlocked.
 148  * Any code which breaks out of this loop while own
 149  * a reference to the current mddev and must mddev_put it.
 150  */
 151 #define ITERATE_MDDEV(mddev,tmp)                                        \
 152                                                                         \
 153         for (({ spin_lock(&all_mddevs_lock);                            \
 154                 tmp = all_mddevs.next;                                  \
 155                 mddev = NULL;});                                        \
 156              ({ if (tmp != &all_mddevs)                                 \
 157                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 158                 spin_unlock(&all_mddevs_lock);                          \
 159                 if (mddev) mddev_put(mddev);                            \
 160                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 161                 tmp != &all_mddevs;});                                  \
 162              ({ spin_lock(&all_mddevs_lock);                            \
 163                 tmp = tmp->next;})                                      \
 164                 )
 165
 166
 167 static int md_fail_request (request_queue_t *q, struct bio *bio)
 168 {
 169         bio_io_error(bio, bio->bi_size);
 170         return 0;
 171 }
 172
 173 static inline mddev_t *mddev_get(mddev_t *mddev)
 174 {
 175         atomic_inc(&mddev->active);
 176         return mddev;
 177 }
 178
 179 static void mddev_put(mddev_t *mddev)
 180 {
 181         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 182                 return;
 183         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 184                 list_del(&mddev->all_mddevs);
 185                 blk_put_queue(mddev->queue);
 186                 kobject_unregister(&mddev->kobj);
 187         }
 188         spin_unlock(&all_mddevs_lock);
 189 }
 190
 191 static mddev_t * mddev_find(dev_t unit)
 192 {
 193         mddev_t *mddev, *new = NULL;
 194
 195  retry:
 196         spin_lock(&all_mddevs_lock);
 197         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 198                 if (mddev->unit == unit) {
 199                         mddev_get(mddev);
 200                         spin_unlock(&all_mddevs_lock);
 201                         kfree(new);
 202                         return mddev;
 203                 }
 204
 205         if (new) {
 206                 list_add(&new->all_mddevs, &all_mddevs);
 207                 spin_unlock(&all_mddevs_lock);
 208                 return new;
 209         }
 210         spin_unlock(&all_mddevs_lock);
 211
 212         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 213         if (!new)
 214                 return NULL;
 215
 216         memset(new, 0, sizeof(*new));
 217
 218         new->unit = unit;
 219         if (MAJOR(unit) == MD_MAJOR)
 220                 new->md_minor = MINOR(unit);
 221         else
 222                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 223
 224         init_MUTEX(&new->reconfig_sem);
 225         INIT_LIST_HEAD(&new->disks);
 226         INIT_LIST_HEAD(&new->all_mddevs);
 227         init_timer(&new->safemode_timer);
 228         atomic_set(&new->active, 1);
 229         spin_lock_init(&new->write_lock);
 230         init_waitqueue_head(&new->sb_wait);
 231
 232         new->queue = blk_alloc_queue(GFP_KERNEL);
 233         if (!new->queue) {
 234                 kfree(new);
 235                 return NULL;
 236         }
 237
 238         blk_queue_make_request(new->queue, md_fail_request);
 239
 240         goto retry;
 241 }
 242
 243 static inline int mddev_lock(mddev_t * mddev)
 244 {
 245         return down_interruptible(&mddev->reconfig_sem);
 246 }
 247
 248 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 249 {
 250         down(&mddev->reconfig_sem);
 251 }
 252
 253 static inline int mddev_trylock(mddev_t * mddev)
 254 {
 255         return down_trylock(&mddev->reconfig_sem);
 256 }
 257
 258 static inline void mddev_unlock(mddev_t * mddev)
 259 {
 260         up(&mddev->reconfig_sem);
 261
 262         md_wakeup_thread(mddev->thread);
 263 }
 264
 265 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 266 {
 267         mdk_rdev_t * rdev;
 268         struct list_head *tmp;
 269
 270         ITERATE_RDEV(mddev,rdev,tmp) {
 271                 if (rdev->desc_nr == nr)
 272                         return rdev;
 273         }
 274         return NULL;
 275 }
 276
 277 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 278 {
 279         struct list_head *tmp;
 280         mdk_rdev_t *rdev;
 281
 282         ITERATE_RDEV(mddev,rdev,tmp) {
 283                 if (rdev->bdev->bd_dev == dev)
 284                         return rdev;
 285         }
 286         return NULL;
 287 }
 288
 289 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 290 {
 291         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 292         return MD_NEW_SIZE_BLOCKS(size);
 293 }
 294
 295 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 296 {
 297         sector_t size;
 298
 299         size = rdev->sb_offset;
 300
 301         if (chunk_size)
 302                 size &= ~((sector_t)chunk_size/1024 - 1);
 303         return size;
 304 }
 305
 306 static int alloc_disk_sb(mdk_rdev_t * rdev)
 307 {
 308         if (rdev->sb_page)
 309                 MD_BUG();
 310
 311         rdev->sb_page = alloc_page(GFP_KERNEL);
 312         if (!rdev->sb_page) {
 313                 printk(KERN_ALERT "md: out of memory.\n");
 314                 return -EINVAL;
 315         }
 316
 317         return 0;
 318 }
 319
 320 static void free_disk_sb(mdk_rdev_t * rdev)
 321 {
 322         if (rdev->sb_page) {
 323                 page_cache_release(rdev->sb_page);
 324                 rdev->sb_loaded = 0;
 325                 rdev->sb_page = NULL;
 326                 rdev->sb_offset = 0;
 327                 rdev->size = 0;
 328         }
 329 }
 330
 331
 332 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 333 {
 334         mdk_rdev_t *rdev = bio->bi_private;
 335         mddev_t *mddev = rdev->mddev;
 336         if (bio->bi_size)
 337                 return 1;
 338
 339         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 340                 md_error(mddev, rdev);
 341
 342         if (atomic_dec_and_test(&mddev->pending_writes))
 343                 wake_up(&mddev->sb_wait);
 344         bio_put(bio);
 345         return 0;
 346 }
 347
 348 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 349 {
 350         struct bio *bio2 = bio->bi_private;
 351         mdk_rdev_t *rdev = bio2->bi_private;
 352         mddev_t *mddev = rdev->mddev;
 353         if (bio->bi_size)
 354                 return 1;
 355
 356         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 357             error == -EOPNOTSUPP) {
 358                 unsigned long flags;
 359                 /* barriers don't appear to be supported :-( */
 360                 set_bit(BarriersNotsupp, &rdev->flags);
 361                 mddev->barriers_work = 0;
 362                 spin_lock_irqsave(&mddev->write_lock, flags);
 363                 bio2->bi_next = mddev->biolist;
 364                 mddev->biolist = bio2;
 365                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 366                 wake_up(&mddev->sb_wait);
 367                 bio_put(bio);
 368                 return 0;
 369         }
 370         bio_put(bio2);
 371         bio->bi_private = rdev;
 372         return super_written(bio, bytes_done, error);
 373 }
 374
 375 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 376                    sector_t sector, int size, struct page *page)
 377 {
 378         /* write first size bytes of page to sector of rdev
 379          * Increment mddev->pending_writes before returning
 380          * and decrement it on completion, waking up sb_wait
 381          * if zero is reached.
 382          * If an error occurred, call md_error
 383          *
 384          * As we might need to resubmit the request if BIO_RW_BARRIER
 385          * causes ENOTSUPP, we allocate a spare bio...
 386          */
 387         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 388         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 389
 390         bio->bi_bdev = rdev->bdev;
 391         bio->bi_sector = sector;
 392         bio_add_page(bio, page, size, 0);
 393         bio->bi_private = rdev;
 394         bio->bi_end_io = super_written;
 395         bio->bi_rw = rw;
 396
 397         atomic_inc(&mddev->pending_writes);
 398         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 399                 struct bio *rbio;
 400                 rw |= (1<<BIO_RW_BARRIER);
 401                 rbio = bio_clone(bio, GFP_NOIO);
 402                 rbio->bi_private = bio;
 403                 rbio->bi_end_io = super_written_barrier;
 404                 submit_bio(rw, rbio);
 405         } else
 406                 submit_bio(rw, bio);
 407 }
 408
 409 void md_super_wait(mddev_t *mddev)
 410 {
 411         /* wait for all superblock writes that were scheduled to complete.
 412          * if any had to be retried (due to BARRIER problems), retry them
 413          */
 414         DEFINE_WAIT(wq);
 415         for(;;) {
 416                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 417                 if (atomic_read(&mddev->pending_writes)==0)
 418                         break;
 419                 while (mddev->biolist) {
 420                         struct bio *bio;
 421                         spin_lock_irq(&mddev->write_lock);
 422                         bio = mddev->biolist;
 423                         mddev->biolist = bio->bi_next ;
 424                         bio->bi_next = NULL;
 425                         spin_unlock_irq(&mddev->write_lock);
 426                         submit_bio(bio->bi_rw, bio);
 427                 }
 428                 schedule();
 429         }
 430         finish_wait(&mddev->sb_wait, &wq);
 431 }
 432
 433 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 434 {
 435         if (bio->bi_size)
 436                 return 1;
 437
 438         complete((struct completion*)bio->bi_private);
 439         return 0;
 440 }
 441
 442 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 443                    struct page *page, int rw)
 444 {
 445         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 446         struct completion event;
 447         int ret;
 448
 449         rw |= (1 << BIO_RW_SYNC);
 450
 451         bio->bi_bdev = bdev;
 452         bio->bi_sector = sector;
 453         bio_add_page(bio, page, size, 0);
 454         init_completion(&event);
 455         bio->bi_private = &event;
 456         bio->bi_end_io = bi_complete;
 457         submit_bio(rw, bio);
 458         wait_for_completion(&event);
 459
 460         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 461         bio_put(bio);
 462         return ret;
 463 }
 464
 465 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 466 {
 467         char b[BDEVNAME_SIZE];
 468         if (!rdev->sb_page) {
 469                 MD_BUG();
 470                 return -EINVAL;
 471         }
 472         if (rdev->sb_loaded)
 473                 return 0;
 474
 475
 476         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 477                 goto fail;
 478         rdev->sb_loaded = 1;
 479         return 0;
 480
 481 fail:
 482         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 483                 bdevname(rdev->bdev,b));
 484         return -EINVAL;
 485 }
 486
 487 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 488 {
 489         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 490                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 491                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 492                 (sb1->set_uuid3 == sb2->set_uuid3))
 493
 494                 return 1;
 495
 496         return 0;
 497 }
 498
 499
 500 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 501 {
 502         int ret;
 503         mdp_super_t *tmp1, *tmp2;
 504
 505         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 506         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 507
 508         if (!tmp1 || !tmp2) {
 509                 ret = 0;
 510                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 511                 goto abort;
 512         }
 513
 514         *tmp1 = *sb1;
 515         *tmp2 = *sb2;
 516
 517         /*
 518          * nr_disks is not constant
 519          */
 520         tmp1->nr_disks = 0;
 521         tmp2->nr_disks = 0;
 522
 523         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 524                 ret = 0;
 525         else
 526                 ret = 1;
 527
 528 abort:
 529         kfree(tmp1);
 530         kfree(tmp2);
 531         return ret;
 532 }
 533
 534 static unsigned int calc_sb_csum(mdp_super_t * sb)
 535 {
 536         unsigned int disk_csum, csum;
 537
 538         disk_csum = sb->sb_csum;
 539         sb->sb_csum = 0;
 540         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 541         sb->sb_csum = disk_csum;
 542         return csum;
 543 }
 544
 545
 546 /*
 547  * Handle superblock details.
 548  * We want to be able to handle multiple superblock formats
 549  * so we have a common interface to them all, and an array of
 550  * different handlers.
 551  * We rely on user-space to write the initial superblock, and support
 552  * reading and updating of superblocks.
 553  * Interface methods are:
 554  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 555  *      loads and validates a superblock on dev.
 556  *      if refdev != NULL, compare superblocks on both devices
 557  *    Return:
 558  *      0 - dev has a superblock that is compatible with refdev
 559  *      1 - dev has a superblock that is compatible and newer than refdev
 560  *          so dev should be used as the refdev in future
 561  *     -EINVAL superblock incompatible or invalid
 562  *     -othererror e.g. -EIO
 563  *
 564  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 565  *      Verify that dev is acceptable into mddev.
 566  *       The first time, mddev->raid_disks will be 0, and data from
 567  *       dev should be merged in.  Subsequent calls check that dev
 568  *       is new enough.  Return 0 or -EINVAL
 569  *
 570  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 571  *     Update the superblock for rdev with data in mddev
 572  *     This does not write to disc.
 573  *
 574  */
 575
 576 struct super_type  {
 577         char            *name;
 578         struct module   *owner;
 579         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 580         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 581         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 582 };
 583
 584 /*
 585  * load_super for 0.90.0
 586  */
 587 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 588 {
 589         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 590         mdp_super_t *sb;
 591         int ret;
 592         sector_t sb_offset;
 593
 594         /*
 595          * Calculate the position of the superblock,
 596          * it's at the end of the disk.
 597          *
 598          * It also happens to be a multiple of 4Kb.
 599          */
 600         sb_offset = calc_dev_sboffset(rdev->bdev);
 601         rdev->sb_offset = sb_offset;
 602
 603         ret = read_disk_sb(rdev, MD_SB_BYTES);
 604         if (ret) return ret;
 605
 606         ret = -EINVAL;
 607
 608         bdevname(rdev->bdev, b);
 609         sb = (mdp_super_t*)page_address(rdev->sb_page);
 610
 611         if (sb->md_magic != MD_SB_MAGIC) {
 612                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 613                        b);
 614                 goto abort;
 615         }
 616
 617         if (sb->major_version != 0 ||
 618             sb->minor_version != 90) {
 619                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 620                         sb->major_version, sb->minor_version,
 621                         b);
 622                 goto abort;
 623         }
 624
 625         if (sb->raid_disks <= 0)
 626                 goto abort;
 627
 628         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 629                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 630                         b);
 631                 goto abort;
 632         }
 633
 634         rdev->preferred_minor = sb->md_minor;
 635         rdev->data_offset = 0;
 636         rdev->sb_size = MD_SB_BYTES;
 637
 638         if (sb->level == LEVEL_MULTIPATH)
 639                 rdev->desc_nr = -1;
 640         else
 641                 rdev->desc_nr = sb->this_disk.number;
 642
 643         if (refdev == 0)
 644                 ret = 1;
 645         else {
 646                 __u64 ev1, ev2;
 647                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 648                 if (!uuid_equal(refsb, sb)) {
 649                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 650                                 b, bdevname(refdev->bdev,b2));
 651                         goto abort;
 652                 }
 653                 if (!sb_equal(refsb, sb)) {
 654                         printk(KERN_WARNING "md: %s has same UUID"
 655                                " but different superblock to %s\n",
 656                                b, bdevname(refdev->bdev, b2));
 657                         goto abort;
 658                 }
 659                 ev1 = md_event(sb);
 660                 ev2 = md_event(refsb);
 661                 if (ev1 > ev2)
 662                         ret = 1;
 663                 else
 664                         ret = 0;
 665         }
 666         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 667
 668  abort:
 669         return ret;
 670 }
 671
 672 /*
 673  * validate_super for 0.90.0
 674  */
 675 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 676 {
 677         mdp_disk_t *desc;
 678         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 679
 680         rdev->raid_disk = -1;
 681         rdev->flags = 0;
 682         if (mddev->raid_disks == 0) {
 683                 mddev->major_version = 0;
 684                 mddev->minor_version = sb->minor_version;
 685                 mddev->patch_version = sb->patch_version;
 686                 mddev->persistent = ! sb->not_persistent;
 687                 mddev->chunk_size = sb->chunk_size;
 688                 mddev->ctime = sb->ctime;
 689                 mddev->utime = sb->utime;
 690                 mddev->level = sb->level;
 691                 mddev->layout = sb->layout;
 692                 mddev->raid_disks = sb->raid_disks;
 693                 mddev->size = sb->size;
 694                 mddev->events = md_event(sb);
 695                 mddev->bitmap_offset = 0;
 696                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 697
 698                 if (sb->state & (1<<MD_SB_CLEAN))
 699                         mddev->recovery_cp = MaxSector;
 700                 else {
 701                         if (sb->events_hi == sb->cp_events_hi &&
 702                                 sb->events_lo == sb->cp_events_lo) {
 703                                 mddev->recovery_cp = sb->recovery_cp;
 704                         } else
 705                                 mddev->recovery_cp = 0;
 706                 }
 707
 708                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 709                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 710                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 711                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 712
 713                 mddev->max_disks = MD_SB_DISKS;
 714
 715                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 716                     mddev->bitmap_file == NULL) {
 717                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
 718                                 /* FIXME use a better test */
 719                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
 720                                 return -EINVAL;
 721                         }
 722                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 723                 }
 724
 725         } else if (mddev->pers == NULL) {
 726                 /* Insist on good event counter while assembling */
 727                 __u64 ev1 = md_event(sb);
 728                 ++ev1;
 729                 if (ev1 < mddev->events)
 730                         return -EINVAL;
 731         } else if (mddev->bitmap) {
 732                 /* if adding to array with a bitmap, then we can accept an
 733                  * older device ... but not too old.
 734                  */
 735                 __u64 ev1 = md_event(sb);
 736                 if (ev1 < mddev->bitmap->events_cleared)
 737                         return 0;
 738         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 739                 return 0;
 740
 741         if (mddev->level != LEVEL_MULTIPATH) {
 742                 desc = sb->disks + rdev->desc_nr;
 743
 744                 if (desc->state & (1<<MD_DISK_FAULTY))
 745                         set_bit(Faulty, &rdev->flags);
 746                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 747                          desc->raid_disk < mddev->raid_disks) {
 748                         set_bit(In_sync, &rdev->flags);
 749                         rdev->raid_disk = desc->raid_disk;
 750                 }
 751                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 752                         set_bit(WriteMostly, &rdev->flags);
 753         } else /* MULTIPATH are always insync */
 754                 set_bit(In_sync, &rdev->flags);
 755         return 0;
 756 }
 757
 758 /*
 759  * sync_super for 0.90.0
 760  */
 761 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 762 {
 763         mdp_super_t *sb;
 764         struct list_head *tmp;
 765         mdk_rdev_t *rdev2;
 766         int next_spare = mddev->raid_disks;
 767
 768
 769         /* make rdev->sb match mddev data..
 770          *
 771          * 1/ zero out disks
 772          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 773          * 3/ any empty disks < next_spare become removed
 774          *
 775          * disks[0] gets initialised to REMOVED because
 776          * we cannot be sure from other fields if it has
 777          * been initialised or not.
 778          */
 779         int i;
 780         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 781
 782         rdev->sb_size = MD_SB_BYTES;
 783
 784         sb = (mdp_super_t*)page_address(rdev->sb_page);
 785
 786         memset(sb, 0, sizeof(*sb));
 787
 788         sb->md_magic = MD_SB_MAGIC;
 789         sb->major_version = mddev->major_version;
 790         sb->minor_version = mddev->minor_version;
 791         sb->patch_version = mddev->patch_version;
 792         sb->gvalid_words  = 0; /* ignored */
 793         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 794         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 795         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 796         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 797
 798         sb->ctime = mddev->ctime;
 799         sb->level = mddev->level;
 800         sb->size  = mddev->size;
 801         sb->raid_disks = mddev->raid_disks;
 802         sb->md_minor = mddev->md_minor;
 803         sb->not_persistent = !mddev->persistent;
 804         sb->utime = mddev->utime;
 805         sb->state = 0;
 806         sb->events_hi = (mddev->events>>32);
 807         sb->events_lo = (u32)mddev->events;
 808
 809         if (mddev->in_sync)
 810         {
 811                 sb->recovery_cp = mddev->recovery_cp;
 812                 sb->cp_events_hi = (mddev->events>>32);
 813                 sb->cp_events_lo = (u32)mddev->events;
 814                 if (mddev->recovery_cp == MaxSector)
 815                         sb->state = (1<< MD_SB_CLEAN);
 816         } else
 817                 sb->recovery_cp = 0;
 818
 819         sb->layout = mddev->layout;
 820         sb->chunk_size = mddev->chunk_size;
 821
 822         if (mddev->bitmap && mddev->bitmap_file == NULL)
 823                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 824
 825         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 826         ITERATE_RDEV(mddev,rdev2,tmp) {
 827                 mdp_disk_t *d;
 828                 int desc_nr;
 829                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 830                     && !test_bit(Faulty, &rdev2->flags))
 831                         desc_nr = rdev2->raid_disk;
 832                 else
 833                         desc_nr = next_spare++;
 834                 rdev2->desc_nr = desc_nr;
 835                 d = &sb->disks[rdev2->desc_nr];
 836                 nr_disks++;
 837                 d->number = rdev2->desc_nr;
 838                 d->major = MAJOR(rdev2->bdev->bd_dev);
 839                 d->minor = MINOR(rdev2->bdev->bd_dev);
 840                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 841                     && !test_bit(Faulty, &rdev2->flags))
 842                         d->raid_disk = rdev2->raid_disk;
 843                 else
 844                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 845                 if (test_bit(Faulty, &rdev2->flags)) {
 846                         d->state = (1<<MD_DISK_FAULTY);
 847                         failed++;
 848                 } else if (test_bit(In_sync, &rdev2->flags)) {
 849                         d->state = (1<<MD_DISK_ACTIVE);
 850                         d->state |= (1<<MD_DISK_SYNC);
 851                         active++;
 852                         working++;
 853                 } else {
 854                         d->state = 0;
 855                         spare++;
 856                         working++;
 857                 }
 858                 if (test_bit(WriteMostly, &rdev2->flags))
 859                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 860         }
 861         /* now set the "removed" and "faulty" bits on any missing devices */
 862         for (i=0 ; i < mddev->raid_disks ; i++) {
 863                 mdp_disk_t *d = &sb->disks[i];
 864                 if (d->state == 0 && d->number == 0) {
 865                         d->number = i;
 866                         d->raid_disk = i;
 867                         d->state = (1<<MD_DISK_REMOVED);
 868                         d->state |= (1<<MD_DISK_FAULTY);
 869                         failed++;
 870                 }
 871         }
 872         sb->nr_disks = nr_disks;
 873         sb->active_disks = active;
 874         sb->working_disks = working;
 875         sb->failed_disks = failed;
 876         sb->spare_disks = spare;
 877
 878         sb->this_disk = sb->disks[rdev->desc_nr];
 879         sb->sb_csum = calc_sb_csum(sb);
 880 }
 881
 882 /*
 883  * version 1 superblock
 884  */
 885
 886 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 887 {
 888         unsigned int disk_csum, csum;
 889         unsigned long long newcsum;
 890         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 891         unsigned int *isuper = (unsigned int*)sb;
 892         int i;
 893
 894         disk_csum = sb->sb_csum;
 895         sb->sb_csum = 0;
 896         newcsum = 0;
 897         for (i=0; size>=4; size -= 4 )
 898                 newcsum += le32_to_cpu(*isuper++);
 899
 900         if (size == 2)
 901                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 902
 903         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 904         sb->sb_csum = disk_csum;
 905         return cpu_to_le32(csum);
 906 }
 907
 908 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 909 {
 910         struct mdp_superblock_1 *sb;
 911         int ret;
 912         sector_t sb_offset;
 913         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 914         int bmask;
 915
 916         /*
 917          * Calculate the position of the superblock.
 918          * It is always aligned to a 4K boundary and
 919          * depeding on minor_version, it can be:
 920          * 0: At least 8K, but less than 12K, from end of device
 921          * 1: At start of device
 922          * 2: 4K from start of device.
 923          */
 924         switch(minor_version) {
 925         case 0:
 926                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 927                 sb_offset -= 8*2;
 928                 sb_offset &= ~(sector_t)(4*2-1);
 929                 /* convert from sectors to K */
 930                 sb_offset /= 2;
 931                 break;
 932         case 1:
 933                 sb_offset = 0;
 934                 break;
 935         case 2:
 936                 sb_offset = 4;
 937                 break;
 938         default:
 939                 return -EINVAL;
 940         }
 941         rdev->sb_offset = sb_offset;
 942
 943         /* superblock is rarely larger than 1K, but it can be larger,
 944          * and it is safe to read 4k, so we do that
 945          */
 946         ret = read_disk_sb(rdev, 4096);
 947         if (ret) return ret;
 948
 949
 950         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 951
 952         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 953             sb->major_version != cpu_to_le32(1) ||
 954             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 955             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 956             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 957                 return -EINVAL;
 958
 959         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 960                 printk("md: invalid superblock checksum on %s\n",
 961                         bdevname(rdev->bdev,b));
 962                 return -EINVAL;
 963         }
 964         if (le64_to_cpu(sb->data_size) < 10) {
 965                 printk("md: data_size too small on %s\n",
 966                        bdevname(rdev->bdev,b));
 967                 return -EINVAL;
 968         }
 969         rdev->preferred_minor = 0xffff;
 970         rdev->data_offset = le64_to_cpu(sb->data_offset);
 971
 972         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 973         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 974         if (rdev->sb_size & bmask)
 975                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 976
 977         if (refdev == 0)
 978                 return 1;
 979         else {
 980                 __u64 ev1, ev2;
 981                 struct mdp_superblock_1 *refsb =
 982                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 983
 984                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 985                     sb->level != refsb->level ||
 986                     sb->layout != refsb->layout ||
 987                     sb->chunksize != refsb->chunksize) {
 988                         printk(KERN_WARNING "md: %s has strangely different"
 989                                 " superblock to %s\n",
 990                                 bdevname(rdev->bdev,b),
 991                                 bdevname(refdev->bdev,b2));
 992                         return -EINVAL;
 993                 }
 994                 ev1 = le64_to_cpu(sb->events);
 995                 ev2 = le64_to_cpu(refsb->events);
 996
 997                 if (ev1 > ev2)
 998                         return 1;
 999         }
1000         if (minor_version)
1001                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1002         else
1003                 rdev->size = rdev->sb_offset;
1004         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1005                 return -EINVAL;
1006         rdev->size = le64_to_cpu(sb->data_size)/2;
1007         if (le32_to_cpu(sb->chunksize))
1008                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1009         return 0;
1010 }
1011
1012 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1013 {
1014         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1015
1016         rdev->raid_disk = -1;
1017         rdev->flags = 0;
1018         if (mddev->raid_disks == 0) {
1019                 mddev->major_version = 1;
1020                 mddev->patch_version = 0;
1021                 mddev->persistent = 1;
1022                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1023                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1024                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1025                 mddev->level = le32_to_cpu(sb->level);
1026                 mddev->layout = le32_to_cpu(sb->layout);
1027                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1028                 mddev->size = le64_to_cpu(sb->size)/2;
1029                 mddev->events = le64_to_cpu(sb->events);
1030                 mddev->bitmap_offset = 0;
1031                 mddev->default_bitmap_offset = 1024;
1032
1033                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1034                 memcpy(mddev->uuid, sb->set_uuid, 16);
1035
1036                 mddev->max_disks =  (4096-256)/2;
1037
1038                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1039                     mddev->bitmap_file == NULL ) {
1040                         if (mddev->level != 1) {
1041                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1042                                 return -EINVAL;
1043                         }
1044                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1045                 }
1046         } else if (mddev->pers == NULL) {
1047                 /* Insist of good event counter while assembling */
1048                 __u64 ev1 = le64_to_cpu(sb->events);
1049                 ++ev1;
1050                 if (ev1 < mddev->events)
1051                         return -EINVAL;
1052         } else if (mddev->bitmap) {
1053                 /* If adding to array with a bitmap, then we can accept an
1054                  * older device, but not too old.
1055                  */
1056                 __u64 ev1 = le64_to_cpu(sb->events);
1057                 if (ev1 < mddev->bitmap->events_cleared)
1058                         return 0;
1059         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1060                 return 0;
1061
1062         if (mddev->level != LEVEL_MULTIPATH) {
1063                 int role;
1064                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1065                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1066                 switch(role) {
1067                 case 0xffff: /* spare */
1068                         break;
1069                 case 0xfffe: /* faulty */
1070                         set_bit(Faulty, &rdev->flags);
1071                         break;
1072                 default:
1073                         set_bit(In_sync, &rdev->flags);
1074                         rdev->raid_disk = role;
1075                         break;
1076                 }
1077                 if (sb->devflags & WriteMostly1)
1078                         set_bit(WriteMostly, &rdev->flags);
1079         } else /* MULTIPATH are always insync */
1080                 set_bit(In_sync, &rdev->flags);
1081
1082         return 0;
1083 }
1084
1085 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1086 {
1087         struct mdp_superblock_1 *sb;
1088         struct list_head *tmp;
1089         mdk_rdev_t *rdev2;
1090         int max_dev, i;
1091         /* make rdev->sb match mddev and rdev data. */
1092
1093         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1094
1095         sb->feature_map = 0;
1096         sb->pad0 = 0;
1097         memset(sb->pad1, 0, sizeof(sb->pad1));
1098         memset(sb->pad2, 0, sizeof(sb->pad2));
1099         memset(sb->pad3, 0, sizeof(sb->pad3));
1100
1101         sb->utime = cpu_to_le64((__u64)mddev->utime);
1102         sb->events = cpu_to_le64(mddev->events);
1103         if (mddev->in_sync)
1104                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1105         else
1106                 sb->resync_offset = cpu_to_le64(0);
1107
1108         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1109                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1110                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1111         }
1112
1113         max_dev = 0;
1114         ITERATE_RDEV(mddev,rdev2,tmp)
1115                 if (rdev2->desc_nr+1 > max_dev)
1116                         max_dev = rdev2->desc_nr+1;
1117
1118         sb->max_dev = cpu_to_le32(max_dev);
1119         for (i=0; i<max_dev;i++)
1120                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1121
1122         ITERATE_RDEV(mddev,rdev2,tmp) {
1123                 i = rdev2->desc_nr;
1124                 if (test_bit(Faulty, &rdev2->flags))
1125                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1126                 else if (test_bit(In_sync, &rdev2->flags))
1127                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1128                 else
1129                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1130         }
1131
1132         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1133         sb->sb_csum = calc_sb_1_csum(sb);
1134 }
1135
1136
1137 static struct super_type super_types[] = {
1138         [0] = {
1139                 .name   = "0.90.0",
1140                 .owner  = THIS_MODULE,
1141                 .load_super     = super_90_load,
1142                 .validate_super = super_90_validate,
1143                 .sync_super     = super_90_sync,
1144         },
1145         [1] = {
1146                 .name   = "md-1",
1147                 .owner  = THIS_MODULE,
1148                 .load_super     = super_1_load,
1149                 .validate_super = super_1_validate,
1150                 .sync_super     = super_1_sync,
1151         },
1152 };
1153
1154 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1155 {
1156         struct list_head *tmp;
1157         mdk_rdev_t *rdev;
1158
1159         ITERATE_RDEV(mddev,rdev,tmp)
1160                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1161                         return rdev;
1162
1163         return NULL;
1164 }
1165
1166 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1167 {
1168         struct list_head *tmp;
1169         mdk_rdev_t *rdev;
1170
1171         ITERATE_RDEV(mddev1,rdev,tmp)
1172                 if (match_dev_unit(mddev2, rdev))
1173                         return 1;
1174
1175         return 0;
1176 }
1177
1178 static LIST_HEAD(pending_raid_disks);
1179
1180 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1181 {
1182         mdk_rdev_t *same_pdev;
1183         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1184         struct kobject *ko;
1185
1186         if (rdev->mddev) {
1187                 MD_BUG();
1188                 return -EINVAL;
1189         }
1190         same_pdev = match_dev_unit(mddev, rdev);
1191         if (same_pdev)
1192                 printk(KERN_WARNING
1193                         "%s: WARNING: %s appears to be on the same physical"
1194                         " disk as %s. True\n     protection against single-disk"
1195                         " failure might be compromised.\n",
1196                         mdname(mddev), bdevname(rdev->bdev,b),
1197                         bdevname(same_pdev->bdev,b2));
1198
1199         /* Verify rdev->desc_nr is unique.
1200          * If it is -1, assign a free number, else
1201          * check number is not in use
1202          */
1203         if (rdev->desc_nr < 0) {
1204                 int choice = 0;
1205                 if (mddev->pers) choice = mddev->raid_disks;
1206                 while (find_rdev_nr(mddev, choice))
1207                         choice++;
1208                 rdev->desc_nr = choice;
1209         } else {
1210                 if (find_rdev_nr(mddev, rdev->desc_nr))
1211                         return -EBUSY;
1212         }
1213         bdevname(rdev->bdev,b);
1214         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1215                 return -ENOMEM;
1216
1217         list_add(&rdev->same_set, &mddev->disks);
1218         rdev->mddev = mddev;
1219         printk(KERN_INFO "md: bind<%s>\n", b);
1220
1221         rdev->kobj.parent = &mddev->kobj;
1222         kobject_add(&rdev->kobj);
1223
1224         if (rdev->bdev->bd_part)
1225                 ko = &rdev->bdev->bd_part->kobj;
1226         else
1227                 ko = &rdev->bdev->bd_disk->kobj;
1228         sysfs_create_link(&rdev->kobj, ko, "block");
1229         return 0;
1230 }
1231
1232 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1233 {
1234         char b[BDEVNAME_SIZE];
1235         if (!rdev->mddev) {
1236                 MD_BUG();
1237                 return;
1238         }
1239         list_del_init(&rdev->same_set);
1240         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1241         rdev->mddev = NULL;
1242         sysfs_remove_link(&rdev->kobj, "block");
1243         kobject_del(&rdev->kobj);
1244 }
1245
1246 /*
1247  * prevent the device from being mounted, repartitioned or
1248  * otherwise reused by a RAID array (or any other kernel
1249  * subsystem), by bd_claiming the device.
1250  */
1251 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1252 {
1253         int err = 0;
1254         struct block_device *bdev;
1255         char b[BDEVNAME_SIZE];
1256
1257         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1258         if (IS_ERR(bdev)) {
1259                 printk(KERN_ERR "md: could not open %s.\n",
1260                         __bdevname(dev, b));
1261                 return PTR_ERR(bdev);
1262         }
1263         err = bd_claim(bdev, rdev);
1264         if (err) {
1265                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1266                         bdevname(bdev, b));
1267                 blkdev_put(bdev);
1268                 return err;
1269         }
1270         rdev->bdev = bdev;
1271         return err;
1272 }
1273
1274 static void unlock_rdev(mdk_rdev_t *rdev)
1275 {
1276         struct block_device *bdev = rdev->bdev;
1277         rdev->bdev = NULL;
1278         if (!bdev)
1279                 MD_BUG();
1280         bd_release(bdev);
1281         blkdev_put(bdev);
1282 }
1283
1284 void md_autodetect_dev(dev_t dev);
1285
1286 static void export_rdev(mdk_rdev_t * rdev)
1287 {
1288         char b[BDEVNAME_SIZE];
1289         printk(KERN_INFO "md: export_rdev(%s)\n",
1290                 bdevname(rdev->bdev,b));
1291         if (rdev->mddev)
1292                 MD_BUG();
1293         free_disk_sb(rdev);
1294         list_del_init(&rdev->same_set);
1295 #ifndef MODULE
1296         md_autodetect_dev(rdev->bdev->bd_dev);
1297 #endif
1298         unlock_rdev(rdev);
1299         kobject_put(&rdev->kobj);
1300 }
1301
1302 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1303 {
1304         unbind_rdev_from_array(rdev);
1305         export_rdev(rdev);
1306 }
1307
1308 static void export_array(mddev_t *mddev)
1309 {
1310         struct list_head *tmp;
1311         mdk_rdev_t *rdev;
1312
1313         ITERATE_RDEV(mddev,rdev,tmp) {
1314                 if (!rdev->mddev) {
1315                         MD_BUG();
1316                         continue;
1317                 }
1318                 kick_rdev_from_array(rdev);
1319         }
1320         if (!list_empty(&mddev->disks))
1321                 MD_BUG();
1322         mddev->raid_disks = 0;
1323         mddev->major_version = 0;
1324 }
1325
1326 static void print_desc(mdp_disk_t *desc)
1327 {
1328         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1329                 desc->major,desc->minor,desc->raid_disk,desc->state);
1330 }
1331
1332 static void print_sb(mdp_super_t *sb)
1333 {
1334         int i;
1335
1336         printk(KERN_INFO
1337                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1338                 sb->major_version, sb->minor_version, sb->patch_version,
1339                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1340                 sb->ctime);
1341         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1342                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1343                 sb->md_minor, sb->layout, sb->chunk_size);
1344         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1345                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1346                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1347                 sb->failed_disks, sb->spare_disks,
1348                 sb->sb_csum, (unsigned long)sb->events_lo);
1349
1350         printk(KERN_INFO);
1351         for (i = 0; i < MD_SB_DISKS; i++) {
1352                 mdp_disk_t *desc;
1353
1354                 desc = sb->disks + i;
1355                 if (desc->number || desc->major || desc->minor ||
1356                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1357                         printk("     D %2d: ", i);
1358                         print_desc(desc);
1359                 }
1360         }
1361         printk(KERN_INFO "md:     THIS: ");
1362         print_desc(&sb->this_disk);
1363
1364 }
1365
1366 static void print_rdev(mdk_rdev_t *rdev)
1367 {
1368         char b[BDEVNAME_SIZE];
1369         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1370                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1371                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1372                 rdev->desc_nr);
1373         if (rdev->sb_loaded) {
1374                 printk(KERN_INFO "md: rdev superblock:\n");
1375                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1376         } else
1377                 printk(KERN_INFO "md: no rdev superblock!\n");
1378 }
1379
1380 void md_print_devices(void)
1381 {
1382         struct list_head *tmp, *tmp2;
1383         mdk_rdev_t *rdev;
1384         mddev_t *mddev;
1385         char b[BDEVNAME_SIZE];
1386
1387         printk("\n");
1388         printk("md:     **********************************\n");
1389         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1390         printk("md:     **********************************\n");
1391         ITERATE_MDDEV(mddev,tmp) {
1392
1393                 if (mddev->bitmap)
1394                         bitmap_print_sb(mddev->bitmap);
1395                 else
1396                         printk("%s: ", mdname(mddev));
1397                 ITERATE_RDEV(mddev,rdev,tmp2)
1398                         printk("<%s>", bdevname(rdev->bdev,b));
1399                 printk("\n");
1400
1401                 ITERATE_RDEV(mddev,rdev,tmp2)
1402                         print_rdev(rdev);
1403         }
1404         printk("md:     **********************************\n");
1405         printk("\n");
1406 }
1407
1408
1409 static void sync_sbs(mddev_t * mddev)
1410 {
1411         mdk_rdev_t *rdev;
1412         struct list_head *tmp;
1413
1414         ITERATE_RDEV(mddev,rdev,tmp) {
1415                 super_types[mddev->major_version].
1416                         sync_super(mddev, rdev);
1417                 rdev->sb_loaded = 1;
1418         }
1419 }
1420
1421 static void md_update_sb(mddev_t * mddev)
1422 {
1423         int err;
1424         struct list_head *tmp;
1425         mdk_rdev_t *rdev;
1426         int sync_req;
1427
1428 repeat:
1429         spin_lock_irq(&mddev->write_lock);
1430         sync_req = mddev->in_sync;
1431         mddev->utime = get_seconds();
1432         mddev->events ++;
1433
1434         if (!mddev->events) {
1435                 /*
1436                  * oops, this 64-bit counter should never wrap.
1437                  * Either we are in around ~1 trillion A.C., assuming
1438                  * 1 reboot per second, or we have a bug:
1439                  */
1440                 MD_BUG();
1441                 mddev->events --;
1442         }
1443         mddev->sb_dirty = 2;
1444         sync_sbs(mddev);
1445
1446         /*
1447          * do not write anything to disk if using
1448          * nonpersistent superblocks
1449          */
1450         if (!mddev->persistent) {
1451                 mddev->sb_dirty = 0;
1452                 spin_unlock_irq(&mddev->write_lock);
1453                 wake_up(&mddev->sb_wait);
1454                 return;
1455         }
1456         spin_unlock_irq(&mddev->write_lock);
1457
1458         dprintk(KERN_INFO
1459                 "md: updating %s RAID superblock on device (in sync %d)\n",
1460                 mdname(mddev),mddev->in_sync);
1461
1462         err = bitmap_update_sb(mddev->bitmap);
1463         ITERATE_RDEV(mddev,rdev,tmp) {
1464                 char b[BDEVNAME_SIZE];
1465                 dprintk(KERN_INFO "md: ");
1466                 if (test_bit(Faulty, &rdev->flags))
1467                         dprintk("(skipping faulty ");
1468
1469                 dprintk("%s ", bdevname(rdev->bdev,b));
1470                 if (!test_bit(Faulty, &rdev->flags)) {
1471                         md_super_write(mddev,rdev,
1472                                        rdev->sb_offset<<1, rdev->sb_size,
1473                                        rdev->sb_page);
1474                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1475                                 bdevname(rdev->bdev,b),
1476                                 (unsigned long long)rdev->sb_offset);
1477
1478                 } else
1479                         dprintk(")\n");
1480                 if (mddev->level == LEVEL_MULTIPATH)
1481                         /* only need to write one superblock... */
1482                         break;
1483         }
1484         md_super_wait(mddev);
1485         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1486
1487         spin_lock_irq(&mddev->write_lock);
1488         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1489                 /* have to write it out again */
1490                 spin_unlock_irq(&mddev->write_lock);
1491                 goto repeat;
1492         }
1493         mddev->sb_dirty = 0;
1494         spin_unlock_irq(&mddev->write_lock);
1495         wake_up(&mddev->sb_wait);
1496
1497 }
1498
1499 struct rdev_sysfs_entry {
1500         struct attribute attr;
1501         ssize_t (*show)(mdk_rdev_t *, char *);
1502         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1503 };
1504
1505 static ssize_t
1506 state_show(mdk_rdev_t *rdev, char *page)
1507 {
1508         char *sep = "";
1509         int len=0;
1510
1511         if (test_bit(Faulty, &rdev->flags)) {
1512                 len+= sprintf(page+len, "%sfaulty",sep);
1513                 sep = ",";
1514         }
1515         if (test_bit(In_sync, &rdev->flags)) {
1516                 len += sprintf(page+len, "%sin_sync",sep);
1517                 sep = ",";
1518         }
1519         if (!test_bit(Faulty, &rdev->flags) &&
1520             !test_bit(In_sync, &rdev->flags)) {
1521                 len += sprintf(page+len, "%sspare", sep);
1522                 sep = ",";
1523         }
1524         return len+sprintf(page+len, "\n");
1525 }
1526
1527 static struct rdev_sysfs_entry
1528 rdev_state = __ATTR_RO(state);
1529
1530 static ssize_t
1531 super_show(mdk_rdev_t *rdev, char *page)
1532 {
1533         if (rdev->sb_loaded && rdev->sb_size) {
1534                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1535                 return rdev->sb_size;
1536         } else
1537                 return 0;
1538 }
1539 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1540
1541 static struct attribute *rdev_default_attrs[] = {
1542         &rdev_state.attr,
1543         &rdev_super.attr,
1544         NULL,
1545 };
1546 static ssize_t
1547 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1548 {
1549         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1550         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1551
1552         if (!entry->show)
1553                 return -EIO;
1554         return entry->show(rdev, page);
1555 }
1556
1557 static ssize_t
1558 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1559               const char *page, size_t length)
1560 {
1561         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1562         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1563
1564         if (!entry->store)
1565                 return -EIO;
1566         return entry->store(rdev, page, length);
1567 }
1568
1569 static void rdev_free(struct kobject *ko)
1570 {
1571         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1572         kfree(rdev);
1573 }
1574 static struct sysfs_ops rdev_sysfs_ops = {
1575         .show           = rdev_attr_show,
1576         .store          = rdev_attr_store,
1577 };
1578 static struct kobj_type rdev_ktype = {
1579         .release        = rdev_free,
1580         .sysfs_ops      = &rdev_sysfs_ops,
1581         .default_attrs  = rdev_default_attrs,
1582 };
1583
1584 /*
1585  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1586  *
1587  * mark the device faulty if:
1588  *
1589  *   - the device is nonexistent (zero size)
1590  *   - the device has no valid superblock
1591  *
1592  * a faulty rdev _never_ has rdev->sb set.
1593  */
1594 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1595 {
1596         char b[BDEVNAME_SIZE];
1597         int err;
1598         mdk_rdev_t *rdev;
1599         sector_t size;
1600
1601         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1602         if (!rdev) {
1603                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1604                 return ERR_PTR(-ENOMEM);
1605         }
1606         memset(rdev, 0, sizeof(*rdev));
1607
1608         if ((err = alloc_disk_sb(rdev)))
1609                 goto abort_free;
1610
1611         err = lock_rdev(rdev, newdev);
1612         if (err)
1613                 goto abort_free;
1614
1615         rdev->kobj.parent = NULL;
1616         rdev->kobj.ktype = &rdev_ktype;
1617         kobject_init(&rdev->kobj);
1618
1619         rdev->desc_nr = -1;
1620         rdev->flags = 0;
1621         rdev->data_offset = 0;
1622         atomic_set(&rdev->nr_pending, 0);
1623         atomic_set(&rdev->read_errors, 0);
1624
1625         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1626         if (!size) {
1627                 printk(KERN_WARNING
1628                         "md: %s has zero or unknown size, marking faulty!\n",
1629                         bdevname(rdev->bdev,b));
1630                 err = -EINVAL;
1631                 goto abort_free;
1632         }
1633
1634         if (super_format >= 0) {
1635                 err = super_types[super_format].
1636                         load_super(rdev, NULL, super_minor);
1637                 if (err == -EINVAL) {
1638                         printk(KERN_WARNING
1639                                 "md: %s has invalid sb, not importing!\n",
1640                                 bdevname(rdev->bdev,b));
1641                         goto abort_free;
1642                 }
1643                 if (err < 0) {
1644                         printk(KERN_WARNING
1645                                 "md: could not read %s's sb, not importing!\n",
1646                                 bdevname(rdev->bdev,b));
1647                         goto abort_free;
1648                 }
1649         }
1650         INIT_LIST_HEAD(&rdev->same_set);
1651
1652         return rdev;
1653
1654 abort_free:
1655         if (rdev->sb_page) {
1656                 if (rdev->bdev)
1657                         unlock_rdev(rdev);
1658                 free_disk_sb(rdev);
1659         }
1660         kfree(rdev);
1661         return ERR_PTR(err);
1662 }
1663
1664 /*
1665  * Check a full RAID array for plausibility
1666  */
1667
1668
1669 static void analyze_sbs(mddev_t * mddev)
1670 {
1671         int i;
1672         struct list_head *tmp;
1673         mdk_rdev_t *rdev, *freshest;
1674         char b[BDEVNAME_SIZE];
1675
1676         freshest = NULL;
1677         ITERATE_RDEV(mddev,rdev,tmp)
1678                 switch (super_types[mddev->major_version].
1679                         load_super(rdev, freshest, mddev->minor_version)) {
1680                 case 1:
1681                         freshest = rdev;
1682                         break;
1683                 case 0:
1684                         break;
1685                 default:
1686                         printk( KERN_ERR \
1687                                 "md: fatal superblock inconsistency in %s"
1688                                 " -- removing from array\n",
1689                                 bdevname(rdev->bdev,b));
1690                         kick_rdev_from_array(rdev);
1691                 }
1692
1693
1694         super_types[mddev->major_version].
1695                 validate_super(mddev, freshest);
1696
1697         i = 0;
1698         ITERATE_RDEV(mddev,rdev,tmp) {
1699                 if (rdev != freshest)
1700                         if (super_types[mddev->major_version].
1701                             validate_super(mddev, rdev)) {
1702                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1703                                         " from array!\n",
1704                                         bdevname(rdev->bdev,b));
1705                                 kick_rdev_from_array(rdev);
1706                                 continue;
1707                         }
1708                 if (mddev->level == LEVEL_MULTIPATH) {
1709                         rdev->desc_nr = i++;
1710                         rdev->raid_disk = rdev->desc_nr;
1711                         set_bit(In_sync, &rdev->flags);
1712                 }
1713         }
1714
1715
1716
1717         if (mddev->recovery_cp != MaxSector &&
1718             mddev->level >= 1)
1719                 printk(KERN_ERR "md: %s: raid array is not clean"
1720                        " -- starting background reconstruction\n",
1721                        mdname(mddev));
1722
1723 }
1724
1725 static ssize_t
1726 level_show(mddev_t *mddev, char *page)
1727 {
1728         mdk_personality_t *p = mddev->pers;
1729         if (p == NULL && mddev->raid_disks == 0)
1730                 return 0;
1731         if (mddev->level >= 0)
1732                 return sprintf(page, "raid%d\n", mddev->level);
1733         else
1734                 return sprintf(page, "%s\n", p->name);
1735 }
1736
1737 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1738
1739 static ssize_t
1740 raid_disks_show(mddev_t *mddev, char *page)
1741 {
1742         if (mddev->raid_disks == 0)
1743                 return 0;
1744         return sprintf(page, "%d\n", mddev->raid_disks);
1745 }
1746
1747 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1748
1749 static ssize_t
1750 action_show(mddev_t *mddev, char *page)
1751 {
1752         char *type = "idle";
1753         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1754             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1755                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1756                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1757                                 type = "resync";
1758                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1759                                 type = "check";
1760                         else
1761                                 type = "repair";
1762                 } else
1763                         type = "recover";
1764         }
1765         return sprintf(page, "%s\n", type);
1766 }
1767
1768 static ssize_t
1769 action_store(mddev_t *mddev, const char *page, size_t len)
1770 {
1771         if (!mddev->pers || !mddev->pers->sync_request)
1772                 return -EINVAL;
1773
1774         if (strcmp(page, "idle")==0 || strcmp(page, "idle\n")==0) {
1775                 if (mddev->sync_thread) {
1776                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1777                         md_unregister_thread(mddev->sync_thread);
1778                         mddev->sync_thread = NULL;
1779                         mddev->recovery = 0;
1780                 }
1781                 return len;
1782         }
1783
1784         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1785             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1786                 return -EBUSY;
1787         if (strcmp(page, "resync")==0 || strcmp(page, "resync\n")==0 ||
1788             strcmp(page, "recover")==0 || strcmp(page, "recover\n")==0)
1789                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1790         else {
1791                 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1792                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1793                 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1794                         return -EINVAL;
1795                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1796                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1797                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1798         }
1799         md_wakeup_thread(mddev->thread);
1800         return len;
1801 }
1802
1803 static ssize_t
1804 mismatch_cnt_show(mddev_t *mddev, char *page)
1805 {
1806         return sprintf(page, "%llu\n",
1807                        (unsigned long long) mddev->resync_mismatches);
1808 }
1809
1810 static struct md_sysfs_entry
1811 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1812
1813
1814 static struct md_sysfs_entry
1815 md_mismatches = __ATTR_RO(mismatch_cnt);
1816
1817 static struct attribute *md_default_attrs[] = {
1818         &md_level.attr,
1819         &md_raid_disks.attr,
1820         NULL,
1821 };
1822
1823 static struct attribute *md_redundancy_attrs[] = {
1824         &md_scan_mode.attr,
1825         &md_mismatches.attr,
1826         NULL,
1827 };
1828 static struct attribute_group md_redundancy_group = {
1829         .name = NULL,
1830         .attrs = md_redundancy_attrs,
1831 };
1832
1833
1834 static ssize_t
1835 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1836 {
1837         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1838         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1839         ssize_t rv;
1840
1841         if (!entry->show)
1842                 return -EIO;
1843         mddev_lock(mddev);
1844         rv = entry->show(mddev, page);
1845         mddev_unlock(mddev);
1846         return rv;
1847 }
1848
1849 static ssize_t
1850 md_attr_store(struct kobject *kobj, struct attribute *attr,
1851               const char *page, size_t length)
1852 {
1853         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1854         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1855         ssize_t rv;
1856
1857         if (!entry->store)
1858                 return -EIO;
1859         mddev_lock(mddev);
1860         rv = entry->store(mddev, page, length);
1861         mddev_unlock(mddev);
1862         return rv;
1863 }
1864
1865 static void md_free(struct kobject *ko)
1866 {
1867         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1868         kfree(mddev);
1869 }
1870
1871 static struct sysfs_ops md_sysfs_ops = {
1872         .show   = md_attr_show,
1873         .store  = md_attr_store,
1874 };
1875 static struct kobj_type md_ktype = {
1876         .release        = md_free,
1877         .sysfs_ops      = &md_sysfs_ops,
1878         .default_attrs  = md_default_attrs,
1879 };
1880
1881 int mdp_major = 0;
1882
1883 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1884 {
1885         static DECLARE_MUTEX(disks_sem);
1886         mddev_t *mddev = mddev_find(dev);
1887         struct gendisk *disk;
1888         int partitioned = (MAJOR(dev) != MD_MAJOR);
1889         int shift = partitioned ? MdpMinorShift : 0;
1890         int unit = MINOR(dev) >> shift;
1891
1892         if (!mddev)
1893                 return NULL;
1894
1895         down(&disks_sem);
1896         if (mddev->gendisk) {
1897                 up(&disks_sem);
1898                 mddev_put(mddev);
1899                 return NULL;
1900         }
1901         disk = alloc_disk(1 << shift);
1902         if (!disk) {
1903                 up(&disks_sem);
1904                 mddev_put(mddev);
1905                 return NULL;
1906         }
1907         disk->major = MAJOR(dev);
1908         disk->first_minor = unit << shift;
1909         if (partitioned) {
1910                 sprintf(disk->disk_name, "md_d%d", unit);
1911                 sprintf(disk->devfs_name, "md/d%d", unit);
1912         } else {
1913                 sprintf(disk->disk_name, "md%d", unit);
1914                 sprintf(disk->devfs_name, "md/%d", unit);
1915         }
1916         disk->fops = &md_fops;
1917         disk->private_data = mddev;
1918         disk->queue = mddev->queue;
1919         add_disk(disk);
1920         mddev->gendisk = disk;
1921         up(&disks_sem);
1922         mddev->kobj.parent = &disk->kobj;
1923         mddev->kobj.k_name = NULL;
1924         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1925         mddev->kobj.ktype = &md_ktype;
1926         kobject_register(&mddev->kobj);
1927         return NULL;
1928 }
1929
1930 void md_wakeup_thread(mdk_thread_t *thread);
1931
1932 static void md_safemode_timeout(unsigned long data)
1933 {
1934         mddev_t *mddev = (mddev_t *) data;
1935
1936         mddev->safemode = 1;
1937         md_wakeup_thread(mddev->thread);
1938 }
1939
1940 static int start_dirty_degraded;
1941
1942 static int do_md_run(mddev_t * mddev)
1943 {
1944         int pnum, err;
1945         int chunk_size;
1946         struct list_head *tmp;
1947         mdk_rdev_t *rdev;
1948         struct gendisk *disk;
1949         char b[BDEVNAME_SIZE];
1950
1951         if (list_empty(&mddev->disks))
1952                 /* cannot run an array with no devices.. */
1953                 return -EINVAL;
1954
1955         if (mddev->pers)
1956                 return -EBUSY;
1957
1958         /*
1959          * Analyze all RAID superblock(s)
1960          */
1961         if (!mddev->raid_disks)
1962                 analyze_sbs(mddev);
1963
1964         chunk_size = mddev->chunk_size;
1965         pnum = level_to_pers(mddev->level);
1966
1967         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1968                 if (!chunk_size) {
1969                         /*
1970                          * 'default chunksize' in the old md code used to
1971                          * be PAGE_SIZE, baaad.
1972                          * we abort here to be on the safe side. We don't
1973                          * want to continue the bad practice.
1974                          */
1975                         printk(KERN_ERR
1976                                 "no chunksize specified, see 'man raidtab'\n");
1977                         return -EINVAL;
1978                 }
1979                 if (chunk_size > MAX_CHUNK_SIZE) {
1980                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1981                                 chunk_size, MAX_CHUNK_SIZE);
1982                         return -EINVAL;
1983                 }
1984                 /*
1985                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1986                  */
1987                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1988                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1989                         return -EINVAL;
1990                 }
1991                 if (chunk_size < PAGE_SIZE) {
1992                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1993                                 chunk_size, PAGE_SIZE);
1994                         return -EINVAL;
1995                 }
1996
1997                 /* devices must have minimum size of one chunk */
1998                 ITERATE_RDEV(mddev,rdev,tmp) {
1999                         if (test_bit(Faulty, &rdev->flags))
2000                                 continue;
2001                         if (rdev->size < chunk_size / 1024) {
2002                                 printk(KERN_WARNING
2003                                         "md: Dev %s smaller than chunk_size:"
2004                                         " %lluk < %dk\n",
2005                                         bdevname(rdev->bdev,b),
2006                                         (unsigned long long)rdev->size,
2007                                         chunk_size / 1024);
2008                                 return -EINVAL;
2009                         }
2010                 }
2011         }
2012
2013 #ifdef CONFIG_KMOD
2014         if (!pers[pnum])
2015         {
2016                 request_module("md-personality-%d", pnum);
2017         }
2018 #endif
2019
2020         /*
2021          * Drop all container device buffers, from now on
2022          * the only valid external interface is through the md
2023          * device.
2024          * Also find largest hardsector size
2025          */
2026         ITERATE_RDEV(mddev,rdev,tmp) {
2027                 if (test_bit(Faulty, &rdev->flags))
2028                         continue;
2029                 sync_blockdev(rdev->bdev);
2030                 invalidate_bdev(rdev->bdev, 0);
2031         }
2032
2033         md_probe(mddev->unit, NULL, NULL);
2034         disk = mddev->gendisk;
2035         if (!disk)
2036                 return -ENOMEM;
2037
2038         spin_lock(&pers_lock);
2039         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2040                 spin_unlock(&pers_lock);
2041                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2042                        pnum);
2043                 return -EINVAL;
2044         }
2045
2046         mddev->pers = pers[pnum];
2047         spin_unlock(&pers_lock);
2048
2049         mddev->recovery = 0;
2050         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2051         mddev->barriers_work = 1;
2052         mddev->ok_start_degraded = start_dirty_degraded;
2053
2054         if (start_readonly)
2055                 mddev->ro = 2; /* read-only, but switch on first write */
2056
2057         /* before we start the array running, initialise the bitmap */
2058         err = bitmap_create(mddev);
2059         if (err)
2060                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2061                         mdname(mddev), err);
2062         else
2063                 err = mddev->pers->run(mddev);
2064         if (err) {
2065                 printk(KERN_ERR "md: pers->run() failed ...\n");
2066                 module_put(mddev->pers->owner);
2067                 mddev->pers = NULL;
2068                 bitmap_destroy(mddev);
2069                 return err;
2070         }
2071         if (mddev->pers->sync_request)
2072                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2073         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2074                 mddev->ro = 0;
2075
2076         atomic_set(&mddev->writes_pending,0);
2077         mddev->safemode = 0;
2078         mddev->safemode_timer.function = md_safemode_timeout;
2079         mddev->safemode_timer.data = (unsigned long) mddev;
2080         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2081         mddev->in_sync = 1;
2082
2083         ITERATE_RDEV(mddev,rdev,tmp)
2084                 if (rdev->raid_disk >= 0) {
2085                         char nm[20];
2086                         sprintf(nm, "rd%d", rdev->raid_disk);
2087                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2088                 }
2089
2090         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2091         md_wakeup_thread(mddev->thread);
2092
2093         if (mddev->sb_dirty)
2094                 md_update_sb(mddev);
2095
2096         set_capacity(disk, mddev->array_size<<1);
2097
2098         /* If we call blk_queue_make_request here, it will
2099          * re-initialise max_sectors etc which may have been
2100          * refined inside -> run.  So just set the bits we need to set.
2101          * Most initialisation happended when we called
2102          * blk_queue_make_request(..., md_fail_request)
2103          * earlier.
2104          */
2105         mddev->queue->queuedata = mddev;
2106         mddev->queue->make_request_fn = mddev->pers->make_request;
2107
2108         mddev->changed = 1;
2109         return 0;
2110 }
2111
2112 static int restart_array(mddev_t *mddev)
2113 {
2114         struct gendisk *disk = mddev->gendisk;
2115         int err;
2116
2117         /*
2118          * Complain if it has no devices
2119          */
2120         err = -ENXIO;
2121         if (list_empty(&mddev->disks))
2122                 goto out;
2123
2124         if (mddev->pers) {
2125                 err = -EBUSY;
2126                 if (!mddev->ro)
2127                         goto out;
2128
2129                 mddev->safemode = 0;
2130                 mddev->ro = 0;
2131                 set_disk_ro(disk, 0);
2132
2133                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2134                         mdname(mddev));
2135                 /*
2136                  * Kick recovery or resync if necessary
2137                  */
2138                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2139                 md_wakeup_thread(mddev->thread);
2140                 err = 0;
2141         } else {
2142                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2143                         mdname(mddev));
2144                 err = -EINVAL;
2145         }
2146
2147 out:
2148         return err;
2149 }
2150
2151 static int do_md_stop(mddev_t * mddev, int ro)
2152 {
2153         int err = 0;
2154         struct gendisk *disk = mddev->gendisk;
2155
2156         if (mddev->pers) {
2157                 if (atomic_read(&mddev->active)>2) {
2158                         printk("md: %s still in use.\n",mdname(mddev));
2159                         return -EBUSY;
2160                 }
2161
2162                 if (mddev->sync_thread) {
2163                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2164                         md_unregister_thread(mddev->sync_thread);
2165                         mddev->sync_thread = NULL;
2166                 }
2167
2168                 del_timer_sync(&mddev->safemode_timer);
2169
2170                 invalidate_partition(disk, 0);
2171
2172                 if (ro) {
2173                         err  = -ENXIO;
2174                         if (mddev->ro==1)
2175                                 goto out;
2176                         mddev->ro = 1;
2177                 } else {
2178                         bitmap_flush(mddev);
2179                         md_super_wait(mddev);
2180                         if (mddev->ro)
2181                                 set_disk_ro(disk, 0);
2182                         blk_queue_make_request(mddev->queue, md_fail_request);
2183                         mddev->pers->stop(mddev);
2184                         if (mddev->pers->sync_request)
2185                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2186
2187                         module_put(mddev->pers->owner);
2188                         mddev->pers = NULL;
2189                         if (mddev->ro)
2190                                 mddev->ro = 0;
2191                 }
2192                 if (!mddev->in_sync) {
2193                         /* mark array as shutdown cleanly */
2194                         mddev->in_sync = 1;
2195                         md_update_sb(mddev);
2196                 }
2197                 if (ro)
2198                         set_disk_ro(disk, 1);
2199         }
2200
2201         bitmap_destroy(mddev);
2202         if (mddev->bitmap_file) {
2203                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2204                 fput(mddev->bitmap_file);
2205                 mddev->bitmap_file = NULL;
2206         }
2207         mddev->bitmap_offset = 0;
2208
2209         /*
2210          * Free resources if final stop
2211          */
2212         if (!ro) {
2213                 mdk_rdev_t *rdev;
2214                 struct list_head *tmp;
2215                 struct gendisk *disk;
2216                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2217
2218                 ITERATE_RDEV(mddev,rdev,tmp)
2219                         if (rdev->raid_disk >= 0) {
2220                                 char nm[20];
2221                                 sprintf(nm, "rd%d", rdev->raid_disk);
2222                                 sysfs_remove_link(&mddev->kobj, nm);
2223                         }
2224
2225                 export_array(mddev);
2226
2227                 mddev->array_size = 0;
2228                 disk = mddev->gendisk;
2229                 if (disk)
2230                         set_capacity(disk, 0);
2231                 mddev->changed = 1;
2232         } else
2233                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2234                         mdname(mddev));
2235         err = 0;
2236 out:
2237         return err;
2238 }
2239
2240 static void autorun_array(mddev_t *mddev)
2241 {
2242         mdk_rdev_t *rdev;
2243         struct list_head *tmp;
2244         int err;
2245
2246         if (list_empty(&mddev->disks))
2247                 return;
2248
2249         printk(KERN_INFO "md: running: ");
2250
2251         ITERATE_RDEV(mddev,rdev,tmp) {
2252                 char b[BDEVNAME_SIZE];
2253                 printk("<%s>", bdevname(rdev->bdev,b));
2254         }
2255         printk("\n");
2256
2257         err = do_md_run (mddev);
2258         if (err) {
2259                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2260                 do_md_stop (mddev, 0);
2261         }
2262 }
2263
2264 /*
2265  * lets try to run arrays based on all disks that have arrived
2266  * until now. (those are in pending_raid_disks)
2267  *
2268  * the method: pick the first pending disk, collect all disks with
2269  * the same UUID, remove all from the pending list and put them into
2270  * the 'same_array' list. Then order this list based on superblock
2271  * update time (freshest comes first), kick out 'old' disks and
2272  * compare superblocks. If everything's fine then run it.
2273  *
2274  * If "unit" is allocated, then bump its reference count
2275  */
2276 static void autorun_devices(int part)
2277 {
2278         struct list_head candidates;
2279         struct list_head *tmp;
2280         mdk_rdev_t *rdev0, *rdev;
2281         mddev_t *mddev;
2282         char b[BDEVNAME_SIZE];
2283
2284         printk(KERN_INFO "md: autorun ...\n");
2285         while (!list_empty(&pending_raid_disks)) {
2286                 dev_t dev;
2287                 rdev0 = list_entry(pending_raid_disks.next,
2288                                          mdk_rdev_t, same_set);
2289
2290                 printk(KERN_INFO "md: considering %s ...\n",
2291                         bdevname(rdev0->bdev,b));
2292                 INIT_LIST_HEAD(&candidates);
2293                 ITERATE_RDEV_PENDING(rdev,tmp)
2294                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2295                                 printk(KERN_INFO "md:  adding %s ...\n",
2296                                         bdevname(rdev->bdev,b));
2297                                 list_move(&rdev->same_set, &candidates);
2298                         }
2299                 /*
2300                  * now we have a set of devices, with all of them having
2301                  * mostly sane superblocks. It's time to allocate the
2302                  * mddev.
2303                  */
2304                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2305                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2306                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2307                         break;
2308                 }
2309                 if (part)
2310                         dev = MKDEV(mdp_major,
2311                                     rdev0->preferred_minor << MdpMinorShift);
2312                 else
2313                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2314
2315                 md_probe(dev, NULL, NULL);
2316                 mddev = mddev_find(dev);
2317                 if (!mddev) {
2318                         printk(KERN_ERR
2319                                 "md: cannot allocate memory for md drive.\n");
2320                         break;
2321                 }
2322                 if (mddev_lock(mddev))
2323                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2324                                mdname(mddev));
2325                 else if (mddev->raid_disks || mddev->major_version
2326                          || !list_empty(&mddev->disks)) {
2327                         printk(KERN_WARNING
2328                                 "md: %s already running, cannot run %s\n",
2329                                 mdname(mddev), bdevname(rdev0->bdev,b));
2330                         mddev_unlock(mddev);
2331                 } else {
2332                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2333                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2334                                 list_del_init(&rdev->same_set);
2335                                 if (bind_rdev_to_array(rdev, mddev))
2336                                         export_rdev(rdev);
2337                         }
2338                         autorun_array(mddev);
2339                         mddev_unlock(mddev);
2340                 }
2341                 /* on success, candidates will be empty, on error
2342                  * it won't...
2343                  */
2344                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2345                         export_rdev(rdev);
2346                 mddev_put(mddev);
2347         }
2348         printk(KERN_INFO "md: ... autorun DONE.\n");
2349 }
2350
2351 /*
2352  * import RAID devices based on one partition
2353  * if possible, the array gets run as well.
2354  */
2355
2356 static int autostart_array(dev_t startdev)
2357 {
2358         char b[BDEVNAME_SIZE];
2359         int err = -EINVAL, i;
2360         mdp_super_t *sb = NULL;
2361         mdk_rdev_t *start_rdev = NULL, *rdev;
2362
2363         start_rdev = md_import_device(startdev, 0, 0);
2364         if (IS_ERR(start_rdev))
2365                 return err;
2366
2367
2368         /* NOTE: this can only work for 0.90.0 superblocks */
2369         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2370         if (sb->major_version != 0 ||
2371             sb->minor_version != 90 ) {
2372                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2373                 export_rdev(start_rdev);
2374                 return err;
2375         }
2376
2377         if (test_bit(Faulty, &start_rdev->flags)) {
2378                 printk(KERN_WARNING
2379                         "md: can not autostart based on faulty %s!\n",
2380                         bdevname(start_rdev->bdev,b));
2381                 export_rdev(start_rdev);
2382                 return err;
2383         }
2384         list_add(&start_rdev->same_set, &pending_raid_disks);
2385
2386         for (i = 0; i < MD_SB_DISKS; i++) {
2387                 mdp_disk_t *desc = sb->disks + i;
2388                 dev_t dev = MKDEV(desc->major, desc->minor);
2389
2390                 if (!dev)
2391                         continue;
2392                 if (dev == startdev)
2393                         continue;
2394                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2395                         continue;
2396                 rdev = md_import_device(dev, 0, 0);
2397                 if (IS_ERR(rdev))
2398                         continue;
2399
2400                 list_add(&rdev->same_set, &pending_raid_disks);
2401         }
2402
2403         /*
2404          * possibly return codes
2405          */
2406         autorun_devices(0);
2407         return 0;
2408
2409 }
2410
2411
2412 static int get_version(void __user * arg)
2413 {
2414         mdu_version_t ver;
2415
2416         ver.major = MD_MAJOR_VERSION;
2417         ver.minor = MD_MINOR_VERSION;
2418         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2419
2420         if (copy_to_user(arg, &ver, sizeof(ver)))
2421                 return -EFAULT;
2422
2423         return 0;
2424 }
2425
2426 static int get_array_info(mddev_t * mddev, void __user * arg)
2427 {
2428         mdu_array_info_t info;
2429         int nr,working,active,failed,spare;
2430         mdk_rdev_t *rdev;
2431         struct list_head *tmp;
2432
2433         nr=working=active=failed=spare=0;
2434         ITERATE_RDEV(mddev,rdev,tmp) {
2435                 nr++;
2436                 if (test_bit(Faulty, &rdev->flags))
2437                         failed++;
2438                 else {
2439                         working++;
2440                         if (test_bit(In_sync, &rdev->flags))
2441                                 active++;
2442                         else
2443                                 spare++;
2444                 }
2445         }
2446
2447         info.major_version = mddev->major_version;
2448         info.minor_version = mddev->minor_version;
2449         info.patch_version = MD_PATCHLEVEL_VERSION;
2450         info.ctime         = mddev->ctime;
2451         info.level         = mddev->level;
2452         info.size          = mddev->size;
2453         info.nr_disks      = nr;
2454         info.raid_disks    = mddev->raid_disks;
2455         info.md_minor      = mddev->md_minor;
2456         info.not_persistent= !mddev->persistent;
2457
2458         info.utime         = mddev->utime;
2459         info.state         = 0;
2460         if (mddev->in_sync)
2461                 info.state = (1<<MD_SB_CLEAN);
2462         if (mddev->bitmap && mddev->bitmap_offset)
2463                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2464         info.active_disks  = active;
2465         info.working_disks = working;
2466         info.failed_disks  = failed;
2467         info.spare_disks   = spare;
2468
2469         info.layout        = mddev->layout;
2470         info.chunk_size    = mddev->chunk_size;
2471
2472         if (copy_to_user(arg, &info, sizeof(info)))
2473                 return -EFAULT;
2474
2475         return 0;
2476 }
2477
2478 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2479 {
2480         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2481         char *ptr, *buf = NULL;
2482         int err = -ENOMEM;
2483
2484         file = kmalloc(sizeof(*file), GFP_KERNEL);
2485         if (!file)
2486                 goto out;
2487
2488         /* bitmap disabled, zero the first byte and copy out */
2489         if (!mddev->bitmap || !mddev->bitmap->file) {
2490                 file->pathname[0] = '\0';
2491                 goto copy_out;
2492         }
2493
2494         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2495         if (!buf)
2496                 goto out;
2497
2498         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2499         if (!ptr)
2500                 goto out;
2501
2502         strcpy(file->pathname, ptr);
2503
2504 copy_out:
2505         err = 0;
2506         if (copy_to_user(arg, file, sizeof(*file)))
2507                 err = -EFAULT;
2508 out:
2509         kfree(buf);
2510         kfree(file);
2511         return err;
2512 }
2513
2514 static int get_disk_info(mddev_t * mddev, void __user * arg)
2515 {
2516         mdu_disk_info_t info;
2517         unsigned int nr;
2518         mdk_rdev_t *rdev;
2519
2520         if (copy_from_user(&info, arg, sizeof(info)))
2521                 return -EFAULT;
2522
2523         nr = info.number;
2524
2525         rdev = find_rdev_nr(mddev, nr);
2526         if (rdev) {
2527                 info.major = MAJOR(rdev->bdev->bd_dev);
2528                 info.minor = MINOR(rdev->bdev->bd_dev);
2529                 info.raid_disk = rdev->raid_disk;
2530                 info.state = 0;
2531                 if (test_bit(Faulty, &rdev->flags))
2532                         info.state |= (1<<MD_DISK_FAULTY);
2533                 else if (test_bit(In_sync, &rdev->flags)) {
2534                         info.state |= (1<<MD_DISK_ACTIVE);
2535                         info.state |= (1<<MD_DISK_SYNC);
2536                 }
2537                 if (test_bit(WriteMostly, &rdev->flags))
2538                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2539         } else {
2540                 info.major = info.minor = 0;
2541                 info.raid_disk = -1;
2542                 info.state = (1<<MD_DISK_REMOVED);
2543         }
2544
2545         if (copy_to_user(arg, &info, sizeof(info)))
2546                 return -EFAULT;
2547
2548         return 0;
2549 }
2550
2551 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2552 {
2553         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2554         mdk_rdev_t *rdev;
2555         dev_t dev = MKDEV(info->major,info->minor);
2556
2557         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2558                 return -EOVERFLOW;
2559
2560         if (!mddev->raid_disks) {
2561                 int err;
2562                 /* expecting a device which has a superblock */
2563                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2564                 if (IS_ERR(rdev)) {
2565                         printk(KERN_WARNING
2566                                 "md: md_import_device returned %ld\n",
2567                                 PTR_ERR(rdev));
2568                         return PTR_ERR(rdev);
2569                 }
2570                 if (!list_empty(&mddev->disks)) {
2571                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2572                                                         mdk_rdev_t, same_set);
2573                         int err = super_types[mddev->major_version]
2574                                 .load_super(rdev, rdev0, mddev->minor_version);
2575                         if (err < 0) {
2576                                 printk(KERN_WARNING
2577                                         "md: %s has different UUID to %s\n",
2578                                         bdevname(rdev->bdev,b),
2579                                         bdevname(rdev0->bdev,b2));
2580                                 export_rdev(rdev);
2581                                 return -EINVAL;
2582                         }
2583                 }
2584                 err = bind_rdev_to_array(rdev, mddev);
2585                 if (err)
2586                         export_rdev(rdev);
2587                 return err;
2588         }
2589
2590         /*
2591          * add_new_disk can be used once the array is assembled
2592          * to add "hot spares".  They must already have a superblock
2593          * written
2594          */
2595         if (mddev->pers) {
2596                 int err;
2597                 if (!mddev->pers->hot_add_disk) {
2598                         printk(KERN_WARNING
2599                                 "%s: personality does not support diskops!\n",
2600                                mdname(mddev));
2601                         return -EINVAL;
2602                 }
2603                 if (mddev->persistent)
2604                         rdev = md_import_device(dev, mddev->major_version,
2605                                                 mddev->minor_version);
2606                 else
2607                         rdev = md_import_device(dev, -1, -1);
2608                 if (IS_ERR(rdev)) {
2609                         printk(KERN_WARNING
2610                                 "md: md_import_device returned %ld\n",
2611                                 PTR_ERR(rdev));
2612                         return PTR_ERR(rdev);
2613                 }
2614                 /* set save_raid_disk if appropriate */
2615                 if (!mddev->persistent) {
2616                         if (info->state & (1<<MD_DISK_SYNC)  &&
2617                             info->raid_disk < mddev->raid_disks)
2618                                 rdev->raid_disk = info->raid_disk;
2619                         else
2620                                 rdev->raid_disk = -1;
2621                 } else
2622                         super_types[mddev->major_version].
2623                                 validate_super(mddev, rdev);
2624                 rdev->saved_raid_disk = rdev->raid_disk;
2625
2626                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2627                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2628                         set_bit(WriteMostly, &rdev->flags);
2629
2630                 rdev->raid_disk = -1;
2631                 err = bind_rdev_to_array(rdev, mddev);
2632                 if (err)
2633                         export_rdev(rdev);
2634
2635                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2636                 md_wakeup_thread(mddev->thread);
2637                 return err;
2638         }
2639
2640         /* otherwise, add_new_disk is only allowed
2641          * for major_version==0 superblocks
2642          */
2643         if (mddev->major_version != 0) {
2644                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2645                        mdname(mddev));
2646                 return -EINVAL;
2647         }
2648
2649         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2650                 int err;
2651                 rdev = md_import_device (dev, -1, 0);
2652                 if (IS_ERR(rdev)) {
2653                         printk(KERN_WARNING
2654                                 "md: error, md_import_device() returned %ld\n",
2655                                 PTR_ERR(rdev));
2656                         return PTR_ERR(rdev);
2657                 }
2658                 rdev->desc_nr = info->number;
2659                 if (info->raid_disk < mddev->raid_disks)
2660                         rdev->raid_disk = info->raid_disk;
2661                 else
2662                         rdev->raid_disk = -1;
2663
2664                 rdev->flags = 0;
2665
2666                 if (rdev->raid_disk < mddev->raid_disks)
2667                         if (info->state & (1<<MD_DISK_SYNC))
2668                                 set_bit(In_sync, &rdev->flags);
2669
2670                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2671                         set_bit(WriteMostly, &rdev->flags);
2672
2673                 err = bind_rdev_to_array(rdev, mddev);
2674                 if (err) {
2675                         export_rdev(rdev);
2676                         return err;
2677                 }
2678
2679                 if (!mddev->persistent) {
2680                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2681                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2682                 } else
2683                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2684                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2685
2686                 if (!mddev->size || (mddev->size > rdev->size))
2687                         mddev->size = rdev->size;
2688         }
2689
2690         return 0;
2691 }
2692
2693 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2694 {
2695         char b[BDEVNAME_SIZE];
2696         mdk_rdev_t *rdev;
2697
2698         if (!mddev->pers)
2699                 return -ENODEV;
2700
2701         rdev = find_rdev(mddev, dev);
2702         if (!rdev)
2703                 return -ENXIO;
2704
2705         if (rdev->raid_disk >= 0)
2706                 goto busy;
2707
2708         kick_rdev_from_array(rdev);
2709         md_update_sb(mddev);
2710
2711         return 0;
2712 busy:
2713         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2714                 bdevname(rdev->bdev,b), mdname(mddev));
2715         return -EBUSY;
2716 }
2717
2718 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2719 {
2720         char b[BDEVNAME_SIZE];
2721         int err;
2722         unsigned int size;
2723         mdk_rdev_t *rdev;
2724
2725         if (!mddev->pers)
2726                 return -ENODEV;
2727
2728         if (mddev->major_version != 0) {
2729                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2730                         " version-0 superblocks.\n",
2731                         mdname(mddev));
2732                 return -EINVAL;
2733         }
2734         if (!mddev->pers->hot_add_disk) {
2735                 printk(KERN_WARNING
2736                         "%s: personality does not support diskops!\n",
2737                         mdname(mddev));
2738                 return -EINVAL;
2739         }
2740
2741         rdev = md_import_device (dev, -1, 0);
2742         if (IS_ERR(rdev)) {
2743                 printk(KERN_WARNING
2744                         "md: error, md_import_device() returned %ld\n",
2745                         PTR_ERR(rdev));
2746                 return -EINVAL;
2747         }
2748
2749         if (mddev->persistent)
2750                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2751         else
2752                 rdev->sb_offset =
2753                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2754
2755         size = calc_dev_size(rdev, mddev->chunk_size);
2756         rdev->size = size;
2757
2758         if (size < mddev->size) {
2759                 printk(KERN_WARNING
2760                         "%s: disk size %llu blocks < array size %llu\n",
2761                         mdname(mddev), (unsigned long long)size,
2762                         (unsigned long long)mddev->size);
2763                 err = -ENOSPC;
2764                 goto abort_export;
2765         }
2766
2767         if (test_bit(Faulty, &rdev->flags)) {
2768                 printk(KERN_WARNING
2769                         "md: can not hot-add faulty %s disk to %s!\n",
2770                         bdevname(rdev->bdev,b), mdname(mddev));
2771                 err = -EINVAL;
2772                 goto abort_export;
2773         }
2774         clear_bit(In_sync, &rdev->flags);
2775         rdev->desc_nr = -1;
2776         bind_rdev_to_array(rdev, mddev);
2777
2778         /*
2779          * The rest should better be atomic, we can have disk failures
2780          * noticed in interrupt contexts ...
2781          */
2782
2783         if (rdev->desc_nr == mddev->max_disks) {
2784                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2785                         mdname(mddev));
2786                 err = -EBUSY;
2787                 goto abort_unbind_export;
2788         }
2789
2790         rdev->raid_disk = -1;
2791
2792         md_update_sb(mddev);
2793
2794         /*
2795          * Kick recovery, maybe this spare has to be added to the
2796          * array immediately.
2797          */
2798         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2799         md_wakeup_thread(mddev->thread);
2800
2801         return 0;
2802
2803 abort_unbind_export:
2804         unbind_rdev_from_array(rdev);
2805
2806 abort_export:
2807         export_rdev(rdev);
2808         return err;
2809 }
2810
2811 /* similar to deny_write_access, but accounts for our holding a reference
2812  * to the file ourselves */
2813 static int deny_bitmap_write_access(struct file * file)
2814 {
2815         struct inode *inode = file->f_mapping->host;
2816
2817         spin_lock(&inode->i_lock);
2818         if (atomic_read(&inode->i_writecount) > 1) {
2819                 spin_unlock(&inode->i_lock);
2820                 return -ETXTBSY;
2821         }
2822         atomic_set(&inode->i_writecount, -1);
2823         spin_unlock(&inode->i_lock);
2824
2825         return 0;
2826 }
2827
2828 static int set_bitmap_file(mddev_t *mddev, int fd)
2829 {
2830         int err;
2831
2832         if (mddev->pers) {
2833                 if (!mddev->pers->quiesce)
2834                         return -EBUSY;
2835                 if (mddev->recovery || mddev->sync_thread)
2836                         return -EBUSY;
2837                 /* we should be able to change the bitmap.. */
2838         }
2839
2840
2841         if (fd >= 0) {
2842                 if (mddev->bitmap)
2843                         return -EEXIST; /* cannot add when bitmap is present */
2844                 mddev->bitmap_file = fget(fd);
2845
2846                 if (mddev->bitmap_file == NULL) {
2847                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2848                                mdname(mddev));
2849                         return -EBADF;
2850                 }
2851
2852                 err = deny_bitmap_write_access(mddev->bitmap_file);
2853                 if (err) {
2854                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2855                                mdname(mddev));
2856                         fput(mddev->bitmap_file);
2857                         mddev->bitmap_file = NULL;
2858                         return err;
2859                 }
2860                 mddev->bitmap_offset = 0; /* file overrides offset */
2861         } else if (mddev->bitmap == NULL)
2862                 return -ENOENT; /* cannot remove what isn't there */
2863         err = 0;
2864         if (mddev->pers) {
2865                 mddev->pers->quiesce(mddev, 1);
2866                 if (fd >= 0)
2867                         err = bitmap_create(mddev);
2868                 if (fd < 0 || err)
2869                         bitmap_destroy(mddev);
2870                 mddev->pers->quiesce(mddev, 0);
2871         } else if (fd < 0) {
2872                 if (mddev->bitmap_file)
2873                         fput(mddev->bitmap_file);
2874                 mddev->bitmap_file = NULL;
2875         }
2876
2877         return err;
2878 }
2879
2880 /*
2881  * set_array_info is used two different ways
2882  * The original usage is when creating a new array.
2883  * In this usage, raid_disks is > 0 and it together with
2884  *  level, size, not_persistent,layout,chunksize determine the
2885  *  shape of the array.
2886  *  This will always create an array with a type-0.90.0 superblock.
2887  * The newer usage is when assembling an array.
2888  *  In this case raid_disks will be 0, and the major_version field is
2889  *  use to determine which style super-blocks are to be found on the devices.
2890  *  The minor and patch _version numbers are also kept incase the
2891  *  super_block handler wishes to interpret them.
2892  */
2893 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2894 {
2895
2896         if (info->raid_disks == 0) {
2897                 /* just setting version number for superblock loading */
2898                 if (info->major_version < 0 ||
2899                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2900                     super_types[info->major_version].name == NULL) {
2901                         /* maybe try to auto-load a module? */
2902                         printk(KERN_INFO
2903                                 "md: superblock version %d not known\n",
2904                                 info->major_version);
2905                         return -EINVAL;
2906                 }
2907                 mddev->major_version = info->major_version;
2908                 mddev->minor_version = info->minor_version;
2909                 mddev->patch_version = info->patch_version;
2910                 return 0;
2911         }
2912         mddev->major_version = MD_MAJOR_VERSION;
2913         mddev->minor_version = MD_MINOR_VERSION;
2914         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2915         mddev->ctime         = get_seconds();
2916
2917         mddev->level         = info->level;
2918         mddev->size          = info->size;
2919         mddev->raid_disks    = info->raid_disks;
2920         /* don't set md_minor, it is determined by which /dev/md* was
2921          * openned
2922          */
2923         if (info->state & (1<<MD_SB_CLEAN))
2924                 mddev->recovery_cp = MaxSector;
2925         else
2926                 mddev->recovery_cp = 0;
2927         mddev->persistent    = ! info->not_persistent;
2928
2929         mddev->layout        = info->layout;
2930         mddev->chunk_size    = info->chunk_size;
2931
2932         mddev->max_disks     = MD_SB_DISKS;
2933
2934         mddev->sb_dirty      = 1;
2935
2936         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2937         mddev->bitmap_offset = 0;
2938
2939         /*
2940          * Generate a 128 bit UUID
2941          */
2942         get_random_bytes(mddev->uuid, 16);
2943
2944         return 0;
2945 }
2946
2947 /*
2948  * update_array_info is used to change the configuration of an
2949  * on-line array.
2950  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2951  * fields in the info are checked against the array.
2952  * Any differences that cannot be handled will cause an error.
2953  * Normally, only one change can be managed at a time.
2954  */
2955 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2956 {
2957         int rv = 0;
2958         int cnt = 0;
2959         int state = 0;
2960
2961         /* calculate expected state,ignoring low bits */
2962         if (mddev->bitmap && mddev->bitmap_offset)
2963                 state |= (1 << MD_SB_BITMAP_PRESENT);
2964
2965         if (mddev->major_version != info->major_version ||
2966             mddev->minor_version != info->minor_version ||
2967 /*          mddev->patch_version != info->patch_version || */
2968             mddev->ctime         != info->ctime         ||
2969             mddev->level         != info->level         ||
2970 /*          mddev->layout        != info->layout        || */
2971             !mddev->persistent   != info->not_persistent||
2972             mddev->chunk_size    != info->chunk_size    ||
2973             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2974             ((state^info->state) & 0xfffffe00)
2975                 )
2976                 return -EINVAL;
2977         /* Check there is only one change */
2978         if (mddev->size != info->size) cnt++;
2979         if (mddev->raid_disks != info->raid_disks) cnt++;
2980         if (mddev->layout != info->layout) cnt++;
2981         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2982         if (cnt == 0) return 0;
2983         if (cnt > 1) return -EINVAL;
2984
2985         if (mddev->layout != info->layout) {
2986                 /* Change layout
2987                  * we don't need to do anything at the md level, the
2988                  * personality will take care of it all.
2989                  */
2990                 if (mddev->pers->reconfig == NULL)
2991                         return -EINVAL;
2992                 else
2993                         return mddev->pers->reconfig(mddev, info->layout, -1);
2994         }
2995         if (mddev->size != info->size) {
2996                 mdk_rdev_t * rdev;
2997                 struct list_head *tmp;
2998                 if (mddev->pers->resize == NULL)
2999                         return -EINVAL;
3000                 /* The "size" is the amount of each device that is used.
3001                  * This can only make sense for arrays with redundancy.
3002                  * linear and raid0 always use whatever space is available
3003                  * We can only consider changing the size if no resync
3004                  * or reconstruction is happening, and if the new size
3005                  * is acceptable. It must fit before the sb_offset or,
3006                  * if that is <data_offset, it must fit before the
3007                  * size of each device.
3008                  * If size is zero, we find the largest size that fits.
3009                  */
3010                 if (mddev->sync_thread)
3011                         return -EBUSY;
3012                 ITERATE_RDEV(mddev,rdev,tmp) {
3013                         sector_t avail;
3014                         int fit = (info->size == 0);
3015                         if (rdev->sb_offset > rdev->data_offset)
3016                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
3017                         else
3018                                 avail = get_capacity(rdev->bdev->bd_disk)
3019                                         - rdev->data_offset;
3020                         if (fit && (info->size == 0 || info->size > avail/2))
3021                                 info->size = avail/2;
3022                         if (avail < ((sector_t)info->size << 1))
3023                                 return -ENOSPC;
3024                 }
3025                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3026                 if (!rv) {
3027                         struct block_device *bdev;
3028
3029                         bdev = bdget_disk(mddev->gendisk, 0);
3030                         if (bdev) {
3031                                 down(&bdev->bd_inode->i_sem);
3032                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3033                                 up(&bdev->bd_inode->i_sem);
3034                                 bdput(bdev);
3035                         }
3036                 }
3037         }
3038         if (mddev->raid_disks    != info->raid_disks) {
3039                 /* change the number of raid disks */
3040                 if (mddev->pers->reshape == NULL)
3041                         return -EINVAL;
3042                 if (info->raid_disks <= 0 ||
3043                     info->raid_disks >= mddev->max_disks)
3044                         return -EINVAL;
3045                 if (mddev->sync_thread)
3046                         return -EBUSY;
3047                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3048                 if (!rv) {
3049                         struct block_device *bdev;
3050
3051                         bdev = bdget_disk(mddev->gendisk, 0);
3052                         if (bdev) {
3053                                 down(&bdev->bd_inode->i_sem);
3054                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3055                                 up(&bdev->bd_inode->i_sem);
3056                                 bdput(bdev);
3057                         }
3058                 }
3059         }
3060         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3061                 if (mddev->pers->quiesce == NULL)
3062                         return -EINVAL;
3063                 if (mddev->recovery || mddev->sync_thread)
3064                         return -EBUSY;
3065                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3066                         /* add the bitmap */
3067                         if (mddev->bitmap)
3068                                 return -EEXIST;
3069                         if (mddev->default_bitmap_offset == 0)
3070                                 return -EINVAL;
3071                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3072                         mddev->pers->quiesce(mddev, 1);
3073                         rv = bitmap_create(mddev);
3074                         if (rv)
3075                                 bitmap_destroy(mddev);
3076                         mddev->pers->quiesce(mddev, 0);
3077                 } else {
3078                         /* remove the bitmap */
3079                         if (!mddev->bitmap)
3080                                 return -ENOENT;
3081                         if (mddev->bitmap->file)
3082                                 return -EINVAL;
3083                         mddev->pers->quiesce(mddev, 1);
3084                         bitmap_destroy(mddev);
3085                         mddev->pers->quiesce(mddev, 0);
3086                         mddev->bitmap_offset = 0;
3087                 }
3088         }
3089         md_update_sb(mddev);
3090         return rv;
3091 }
3092
3093 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3094 {
3095         mdk_rdev_t *rdev;
3096
3097         if (mddev->pers == NULL)
3098                 return -ENODEV;
3099
3100         rdev = find_rdev(mddev, dev);
3101         if (!rdev)
3102                 return -ENODEV;
3103
3104         md_error(mddev, rdev);
3105         return 0;
3106 }
3107
3108 static int md_ioctl(struct inode *inode, struct file *file,
3109                         unsigned int cmd, unsigned long arg)
3110 {
3111         int err = 0;
3112         void __user *argp = (void __user *)arg;
3113         struct hd_geometry __user *loc = argp;
3114         mddev_t *mddev = NULL;
3115
3116         if (!capable(CAP_SYS_ADMIN))
3117                 return -EACCES;
3118
3119         /*
3120          * Commands dealing with the RAID driver but not any
3121          * particular array:
3122          */
3123         switch (cmd)
3124         {
3125                 case RAID_VERSION:
3126                         err = get_version(argp);
3127                         goto done;
3128
3129                 case PRINT_RAID_DEBUG:
3130                         err = 0;
3131                         md_print_devices();
3132                         goto done;
3133
3134 #ifndef MODULE
3135                 case RAID_AUTORUN:
3136                         err = 0;
3137                         autostart_arrays(arg);
3138                         goto done;
3139 #endif
3140                 default:;
3141         }
3142
3143         /*
3144          * Commands creating/starting a new array:
3145          */
3146
3147         mddev = inode->i_bdev->bd_disk->private_data;
3148
3149         if (!mddev) {
3150                 BUG();
3151                 goto abort;
3152         }
3153
3154
3155         if (cmd == START_ARRAY) {
3156                 /* START_ARRAY doesn't need to lock the array as autostart_array
3157                  * does the locking, and it could even be a different array
3158                  */
3159                 static int cnt = 3;
3160                 if (cnt > 0 ) {
3161                         printk(KERN_WARNING
3162                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3163                                "This will not be supported beyond July 2006\n",
3164                                current->comm, current->pid);
3165                         cnt--;
3166                 }
3167                 err = autostart_array(new_decode_dev(arg));
3168                 if (err) {
3169                         printk(KERN_WARNING "md: autostart failed!\n");
3170                         goto abort;
3171                 }
3172                 goto done;
3173         }
3174
3175         err = mddev_lock(mddev);
3176         if (err) {
3177                 printk(KERN_INFO
3178                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3179                         err, cmd);
3180                 goto abort;
3181         }
3182
3183         switch (cmd)
3184         {
3185                 case SET_ARRAY_INFO:
3186                         {
3187                                 mdu_array_info_t info;
3188                                 if (!arg)
3189                                         memset(&info, 0, sizeof(info));
3190                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3191                                         err = -EFAULT;
3192                                         goto abort_unlock;
3193                                 }
3194                                 if (mddev->pers) {
3195                                         err = update_array_info(mddev, &info);
3196                                         if (err) {
3197                                                 printk(KERN_WARNING "md: couldn't update"
3198                                                        " array info. %d\n", err);
3199                                                 goto abort_unlock;
3200                                         }
3201                                         goto done_unlock;
3202                                 }
3203                                 if (!list_empty(&mddev->disks)) {
3204                                         printk(KERN_WARNING
3205                                                "md: array %s already has disks!\n",
3206                                                mdname(mddev));
3207                                         err = -EBUSY;
3208                                         goto abort_unlock;
3209                                 }
3210                                 if (mddev->raid_disks) {
3211                                         printk(KERN_WARNING
3212                                                "md: array %s already initialised!\n",
3213                                                mdname(mddev));
3214                                         err = -EBUSY;
3215                                         goto abort_unlock;
3216                                 }
3217                                 err = set_array_info(mddev, &info);
3218                                 if (err) {
3219                                         printk(KERN_WARNING "md: couldn't set"
3220                                                " array info. %d\n", err);
3221                                         goto abort_unlock;
3222                                 }
3223                         }
3224                         goto done_unlock;
3225
3226                 default:;
3227         }
3228
3229         /*
3230          * Commands querying/configuring an existing array:
3231          */
3232         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3233          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3234         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3235                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3236                 err = -ENODEV;
3237                 goto abort_unlock;
3238         }
3239
3240         /*
3241          * Commands even a read-only array can execute:
3242          */
3243         switch (cmd)
3244         {
3245                 case GET_ARRAY_INFO:
3246                         err = get_array_info(mddev, argp);
3247                         goto done_unlock;
3248
3249                 case GET_BITMAP_FILE:
3250                         err = get_bitmap_file(mddev, argp);
3251                         goto done_unlock;
3252
3253                 case GET_DISK_INFO:
3254                         err = get_disk_info(mddev, argp);
3255                         goto done_unlock;
3256
3257                 case RESTART_ARRAY_RW:
3258                         err = restart_array(mddev);
3259                         goto done_unlock;
3260
3261                 case STOP_ARRAY:
3262                         err = do_md_stop (mddev, 0);
3263                         goto done_unlock;
3264
3265                 case STOP_ARRAY_RO:
3266                         err = do_md_stop (mddev, 1);
3267                         goto done_unlock;
3268
3269         /*
3270          * We have a problem here : there is no easy way to give a CHS
3271          * virtual geometry. We currently pretend that we have a 2 heads
3272          * 4 sectors (with a BIG number of cylinders...). This drives
3273          * dosfs just mad... ;-)
3274          */
3275                 case HDIO_GETGEO:
3276                         if (!loc) {
3277                                 err = -EINVAL;
3278                                 goto abort_unlock;
3279                         }
3280                         err = put_user (2, (char __user *) &loc->heads);
3281                         if (err)
3282                                 goto abort_unlock;
3283                         err = put_user (4, (char __user *) &loc->sectors);
3284                         if (err)
3285                                 goto abort_unlock;
3286                         err = put_user(get_capacity(mddev->gendisk)/8,
3287                                         (short __user *) &loc->cylinders);
3288                         if (err)
3289                                 goto abort_unlock;
3290                         err = put_user (get_start_sect(inode->i_bdev),
3291                                                 (long __user *) &loc->start);
3292                         goto done_unlock;
3293         }
3294
3295         /*
3296          * The remaining ioctls are changing the state of the
3297          * superblock, so we do not allow them on read-only arrays.
3298          * However non-MD ioctls (e.g. get-size) will still come through
3299          * here and hit the 'default' below, so only disallow
3300          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3301          */
3302         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3303             mddev->ro && mddev->pers) {
3304                 if (mddev->ro == 2) {
3305                         mddev->ro = 0;
3306                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3307                 md_wakeup_thread(mddev->thread);
3308
3309                 } else {
3310                         err = -EROFS;
3311                         goto abort_unlock;
3312                 }
3313         }
3314
3315         switch (cmd)
3316         {
3317                 case ADD_NEW_DISK:
3318                 {
3319                         mdu_disk_info_t info;
3320                         if (copy_from_user(&info, argp, sizeof(info)))
3321                                 err = -EFAULT;
3322                         else
3323                                 err = add_new_disk(mddev, &info);
3324                         goto done_unlock;
3325                 }
3326
3327                 case HOT_REMOVE_DISK:
3328                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3329                         goto done_unlock;
3330
3331                 case HOT_ADD_DISK:
3332                         err = hot_add_disk(mddev, new_decode_dev(arg));
3333                         goto done_unlock;
3334
3335                 case SET_DISK_FAULTY:
3336                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3337                         goto done_unlock;
3338
3339                 case RUN_ARRAY:
3340                         err = do_md_run (mddev);
3341                         goto done_unlock;
3342
3343                 case SET_BITMAP_FILE:
3344                         err = set_bitmap_file(mddev, (int)arg);
3345                         goto done_unlock;
3346
3347                 default:
3348                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3349                                 printk(KERN_WARNING "md: %s(pid %d) used"
3350                                         " obsolete MD ioctl, upgrade your"
3351                                         " software to use new ictls.\n",
3352                                         current->comm, current->pid);
3353                         err = -EINVAL;
3354                         goto abort_unlock;
3355         }
3356
3357 done_unlock:
3358 abort_unlock:
3359         mddev_unlock(mddev);
3360
3361         return err;
3362 done:
3363         if (err)
3364                 MD_BUG();
3365 abort:
3366         return err;
3367 }
3368
3369 static int md_open(struct inode *inode, struct file *file)
3370 {
3371         /*
3372          * Succeed if we can lock the mddev, which confirms that
3373          * it isn't being stopped right now.
3374          */
3375         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3376         int err;
3377
3378         if ((err = mddev_lock(mddev)))
3379                 goto out;
3380
3381         err = 0;
3382         mddev_get(mddev);
3383         mddev_unlock(mddev);
3384
3385         check_disk_change(inode->i_bdev);
3386  out:
3387         return err;
3388 }
3389
3390 static int md_release(struct inode *inode, struct file * file)
3391 {
3392         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3393
3394         if (!mddev)
3395                 BUG();
3396         mddev_put(mddev);
3397
3398         return 0;
3399 }
3400
3401 static int md_media_changed(struct gendisk *disk)
3402 {
3403         mddev_t *mddev = disk->private_data;
3404
3405         return mddev->changed;
3406 }
3407
3408 static int md_revalidate(struct gendisk *disk)
3409 {
3410         mddev_t *mddev = disk->private_data;
3411
3412         mddev->changed = 0;
3413         return 0;
3414 }
3415 static struct block_device_operations md_fops =
3416 {
3417         .owner          = THIS_MODULE,
3418         .open           = md_open,
3419         .release        = md_release,
3420         .ioctl          = md_ioctl,
3421         .media_changed  = md_media_changed,
3422         .revalidate_disk= md_revalidate,
3423 };
3424
3425 static int md_thread(void * arg)
3426 {
3427         mdk_thread_t *thread = arg;
3428
3429         /*
3430          * md_thread is a 'system-thread', it's priority should be very
3431          * high. We avoid resource deadlocks individually in each
3432          * raid personality. (RAID5 does preallocation) We also use RR and
3433          * the very same RT priority as kswapd, thus we will never get
3434          * into a priority inversion deadlock.
3435          *
3436          * we definitely have to have equal or higher priority than
3437          * bdflush, otherwise bdflush will deadlock if there are too
3438          * many dirty RAID5 blocks.
3439          */
3440
3441         allow_signal(SIGKILL);
3442         while (!kthread_should_stop()) {
3443
3444                 /* We need to wait INTERRUPTIBLE so that
3445                  * we don't add to the load-average.
3446                  * That means we need to be sure no signals are
3447                  * pending
3448                  */
3449                 if (signal_pending(current))
3450                         flush_signals(current);
3451
3452                 wait_event_interruptible_timeout
3453                         (thread->wqueue,
3454                          test_bit(THREAD_WAKEUP, &thread->flags)
3455                          || kthread_should_stop(),
3456                          thread->timeout);
3457                 try_to_freeze();
3458
3459                 clear_bit(THREAD_WAKEUP, &thread->flags);
3460
3461                 thread->run(thread->mddev);
3462         }
3463
3464         return 0;
3465 }
3466
3467 void md_wakeup_thread(mdk_thread_t *thread)
3468 {
3469         if (thread) {
3470                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3471                 set_bit(THREAD_WAKEUP, &thread->flags);
3472                 wake_up(&thread->wqueue);
3473         }
3474 }
3475
3476 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3477                                  const char *name)
3478 {
3479         mdk_thread_t *thread;
3480
3481         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3482         if (!thread)
3483                 return NULL;
3484
3485         memset(thread, 0, sizeof(mdk_thread_t));
3486         init_waitqueue_head(&thread->wqueue);
3487
3488         thread->run = run;
3489         thread->mddev = mddev;
3490         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3491         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3492         if (IS_ERR(thread->tsk)) {
3493                 kfree(thread);
3494                 return NULL;
3495         }
3496         return thread;
3497 }
3498
3499 void md_unregister_thread(mdk_thread_t *thread)
3500 {
3501         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3502
3503         kthread_stop(thread->tsk);
3504         kfree(thread);
3505 }
3506
3507 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3508 {
3509         if (!mddev) {
3510                 MD_BUG();
3511                 return;
3512         }
3513
3514         if (!rdev || test_bit(Faulty, &rdev->flags))
3515                 return;
3516 /*
3517         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3518                 mdname(mddev),
3519                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3520                 __builtin_return_address(0),__builtin_return_address(1),
3521                 __builtin_return_address(2),__builtin_return_address(3));
3522 */
3523         if (!mddev->pers->error_handler)
3524                 return;
3525         mddev->pers->error_handler(mddev,rdev);
3526         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3527         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3528         md_wakeup_thread(mddev->thread);
3529 }
3530
3531 /* seq_file implementation /proc/mdstat */
3532
3533 static void status_unused(struct seq_file *seq)
3534 {
3535         int i = 0;
3536         mdk_rdev_t *rdev;
3537         struct list_head *tmp;
3538
3539         seq_printf(seq, "unused devices: ");
3540
3541         ITERATE_RDEV_PENDING(rdev,tmp) {
3542                 char b[BDEVNAME_SIZE];
3543                 i++;
3544                 seq_printf(seq, "%s ",
3545                               bdevname(rdev->bdev,b));
3546         }
3547         if (!i)
3548                 seq_printf(seq, "<none>");
3549
3550         seq_printf(seq, "\n");
3551 }
3552
3553
3554 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3555 {
3556         unsigned long max_blocks, resync, res, dt, db, rt;
3557
3558         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3559
3560         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3561                 max_blocks = mddev->resync_max_sectors >> 1;
3562         else
3563                 max_blocks = mddev->size;
3564
3565         /*
3566          * Should not happen.
3567          */
3568         if (!max_blocks) {
3569                 MD_BUG();
3570                 return;
3571         }
3572         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3573         {
3574                 int i, x = res/50, y = 20-x;
3575                 seq_printf(seq, "[");
3576                 for (i = 0; i < x; i++)
3577                         seq_printf(seq, "=");
3578                 seq_printf(seq, ">");
3579                 for (i = 0; i < y; i++)
3580                         seq_printf(seq, ".");
3581                 seq_printf(seq, "] ");
3582         }
3583         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3584                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3585                        "resync" : "recovery"),
3586                       res/10, res % 10, resync, max_blocks);
3587
3588         /*
3589          * We do not want to overflow, so the order of operands and
3590          * the * 100 / 100 trick are important. We do a +1 to be
3591          * safe against division by zero. We only estimate anyway.
3592          *
3593          * dt: time from mark until now
3594          * db: blocks written from mark until now
3595          * rt: remaining time
3596          */
3597         dt = ((jiffies - mddev->resync_mark) / HZ);
3598         if (!dt) dt++;
3599         db = resync - (mddev->resync_mark_cnt/2);
3600         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3601
3602         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3603
3604         seq_printf(seq, " speed=%ldK/sec", db/dt);
3605 }
3606
3607 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3608 {
3609         struct list_head *tmp;
3610         loff_t l = *pos;
3611         mddev_t *mddev;
3612
3613         if (l >= 0x10000)
3614                 return NULL;
3615         if (!l--)
3616                 /* header */
3617                 return (void*)1;
3618
3619         spin_lock(&all_mddevs_lock);
3620         list_for_each(tmp,&all_mddevs)
3621                 if (!l--) {
3622                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3623                         mddev_get(mddev);
3624                         spin_unlock(&all_mddevs_lock);
3625                         return mddev;
3626                 }
3627         spin_unlock(&all_mddevs_lock);
3628         if (!l--)
3629                 return (void*)2;/* tail */
3630         return NULL;
3631 }
3632
3633 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3634 {
3635         struct list_head *tmp;
3636         mddev_t *next_mddev, *mddev = v;
3637
3638         ++*pos;
3639         if (v == (void*)2)
3640                 return NULL;
3641
3642         spin_lock(&all_mddevs_lock);
3643         if (v == (void*)1)
3644                 tmp = all_mddevs.next;
3645         else
3646                 tmp = mddev->all_mddevs.next;
3647         if (tmp != &all_mddevs)
3648                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3649         else {
3650                 next_mddev = (void*)2;
3651                 *pos = 0x10000;
3652         }
3653         spin_unlock(&all_mddevs_lock);
3654
3655         if (v != (void*)1)
3656                 mddev_put(mddev);
3657         return next_mddev;
3658
3659 }
3660
3661 static void md_seq_stop(struct seq_file *seq, void *v)
3662 {
3663         mddev_t *mddev = v;
3664
3665         if (mddev && v != (void*)1 && v != (void*)2)
3666                 mddev_put(mddev);
3667 }
3668
3669 static int md_seq_show(struct seq_file *seq, void *v)
3670 {
3671         mddev_t *mddev = v;
3672         sector_t size;
3673         struct list_head *tmp2;
3674         mdk_rdev_t *rdev;
3675         int i;
3676         struct bitmap *bitmap;
3677
3678         if (v == (void*)1) {
3679                 seq_printf(seq, "Personalities : ");
3680                 spin_lock(&pers_lock);
3681                 for (i = 0; i < MAX_PERSONALITY; i++)
3682                         if (pers[i])
3683                                 seq_printf(seq, "[%s] ", pers[i]->name);
3684
3685                 spin_unlock(&pers_lock);
3686                 seq_printf(seq, "\n");
3687                 return 0;
3688         }
3689         if (v == (void*)2) {
3690                 status_unused(seq);
3691                 return 0;
3692         }
3693
3694         if (mddev_lock(mddev)!=0)
3695                 return -EINTR;
3696         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3697                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3698                                                 mddev->pers ? "" : "in");
3699                 if (mddev->pers) {
3700                         if (mddev->ro==1)
3701                                 seq_printf(seq, " (read-only)");
3702                         if (mddev->ro==2)
3703                                 seq_printf(seq, "(auto-read-only)");
3704                         seq_printf(seq, " %s", mddev->pers->name);
3705                 }
3706
3707                 size = 0;
3708                 ITERATE_RDEV(mddev,rdev,tmp2) {
3709                         char b[BDEVNAME_SIZE];
3710                         seq_printf(seq, " %s[%d]",
3711                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3712                         if (test_bit(WriteMostly, &rdev->flags))
3713                                 seq_printf(seq, "(W)");
3714                         if (test_bit(Faulty, &rdev->flags)) {
3715                                 seq_printf(seq, "(F)");
3716                                 continue;
3717                         } else if (rdev->raid_disk < 0)
3718                                 seq_printf(seq, "(S)"); /* spare */
3719                         size += rdev->size;
3720                 }
3721
3722                 if (!list_empty(&mddev->disks)) {
3723                         if (mddev->pers)
3724                                 seq_printf(seq, "\n      %llu blocks",
3725                                         (unsigned long long)mddev->array_size);
3726                         else
3727                                 seq_printf(seq, "\n      %llu blocks",
3728                                         (unsigned long long)size);
3729                 }
3730                 if (mddev->persistent) {
3731                         if (mddev->major_version != 0 ||
3732                             mddev->minor_version != 90) {
3733                                 seq_printf(seq," super %d.%d",
3734                                            mddev->major_version,
3735                                            mddev->minor_version);
3736                         }
3737                 } else
3738                         seq_printf(seq, " super non-persistent");
3739
3740                 if (mddev->pers) {
3741                         mddev->pers->status (seq, mddev);
3742                         seq_printf(seq, "\n      ");
3743                         if (mddev->pers->sync_request) {
3744                                 if (mddev->curr_resync > 2) {
3745                                         status_resync (seq, mddev);
3746                                         seq_printf(seq, "\n      ");
3747                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3748                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3749                                 else if (mddev->recovery_cp < MaxSector)
3750                                         seq_printf(seq, "\tresync=PENDING\n      ");
3751                         }
3752                 } else
3753                         seq_printf(seq, "\n       ");
3754
3755                 if ((bitmap = mddev->bitmap)) {
3756                         unsigned long chunk_kb;
3757                         unsigned long flags;
3758                         spin_lock_irqsave(&bitmap->lock, flags);
3759                         chunk_kb = bitmap->chunksize >> 10;
3760                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3761                                 "%lu%s chunk",
3762                                 bitmap->pages - bitmap->missing_pages,
3763                                 bitmap->pages,
3764                                 (bitmap->pages - bitmap->missing_pages)
3765                                         << (PAGE_SHIFT - 10),
3766                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3767                                 chunk_kb ? "KB" : "B");
3768                         if (bitmap->file) {
3769                                 seq_printf(seq, ", file: ");
3770                                 seq_path(seq, bitmap->file->f_vfsmnt,
3771                                          bitmap->file->f_dentry," \t\n");
3772                         }
3773
3774                         seq_printf(seq, "\n");
3775                         spin_unlock_irqrestore(&bitmap->lock, flags);
3776                 }
3777
3778                 seq_printf(seq, "\n");
3779         }
3780         mddev_unlock(mddev);
3781
3782         return 0;
3783 }
3784
3785 static struct seq_operations md_seq_ops = {
3786         .start  = md_seq_start,
3787         .next   = md_seq_next,
3788         .stop   = md_seq_stop,
3789         .show   = md_seq_show,
3790 };
3791
3792 static int md_seq_open(struct inode *inode, struct file *file)
3793 {
3794         int error;
3795
3796         error = seq_open(file, &md_seq_ops);
3797         return error;
3798 }
3799
3800 static struct file_operations md_seq_fops = {
3801         .open           = md_seq_open,
3802         .read           = seq_read,
3803         .llseek         = seq_lseek,
3804         .release        = seq_release,
3805 };
3806
3807 int register_md_personality(int pnum, mdk_personality_t *p)
3808 {
3809         if (pnum >= MAX_PERSONALITY) {
3810                 printk(KERN_ERR
3811                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3812                        p->name, pnum, MAX_PERSONALITY-1);
3813                 return -EINVAL;
3814         }
3815
3816         spin_lock(&pers_lock);
3817         if (pers[pnum]) {
3818                 spin_unlock(&pers_lock);
3819                 return -EBUSY;
3820         }
3821
3822         pers[pnum] = p;
3823         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3824         spin_unlock(&pers_lock);
3825         return 0;
3826 }
3827
3828 int unregister_md_personality(int pnum)
3829 {
3830         if (pnum >= MAX_PERSONALITY)
3831                 return -EINVAL;
3832
3833         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3834         spin_lock(&pers_lock);
3835         pers[pnum] = NULL;
3836         spin_unlock(&pers_lock);
3837         return 0;
3838 }
3839
3840 static int is_mddev_idle(mddev_t *mddev)
3841 {
3842         mdk_rdev_t * rdev;
3843         struct list_head *tmp;
3844         int idle;
3845         unsigned long curr_events;
3846
3847         idle = 1;
3848         ITERATE_RDEV(mddev,rdev,tmp) {
3849                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3850                 curr_events = disk_stat_read(disk, sectors[0]) +
3851                                 disk_stat_read(disk, sectors[1]) -
3852                                 atomic_read(&disk->sync_io);
3853                 /* The difference between curr_events and last_events
3854                  * will be affected by any new non-sync IO (making
3855                  * curr_events bigger) and any difference in the amount of
3856                  * in-flight syncio (making current_events bigger or smaller)
3857                  * The amount in-flight is currently limited to
3858                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3859                  * which is at most 4096 sectors.
3860                  * These numbers are fairly fragile and should be made
3861                  * more robust, probably by enforcing the
3862                  * 'window size' that md_do_sync sort-of uses.
3863                  *
3864                  * Note: the following is an unsigned comparison.
3865                  */
3866                 if ((curr_events - rdev->last_events + 4096) > 8192) {
3867                         rdev->last_events = curr_events;
3868                         idle = 0;
3869                 }
3870         }
3871         return idle;
3872 }
3873
3874 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3875 {
3876         /* another "blocks" (512byte) blocks have been synced */
3877         atomic_sub(blocks, &mddev->recovery_active);
3878         wake_up(&mddev->recovery_wait);
3879         if (!ok) {
3880                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3881                 md_wakeup_thread(mddev->thread);
3882                 // stop recovery, signal do_sync ....
3883         }
3884 }
3885
3886
3887 /* md_write_start(mddev, bi)
3888  * If we need to update some array metadata (e.g. 'active' flag
3889  * in superblock) before writing, schedule a superblock update
3890  * and wait for it to complete.
3891  */
3892 void md_write_start(mddev_t *mddev, struct bio *bi)
3893 {
3894         if (bio_data_dir(bi) != WRITE)
3895                 return;
3896
3897         BUG_ON(mddev->ro == 1);
3898         if (mddev->ro == 2) {
3899                 /* need to switch to read/write */
3900                 mddev->ro = 0;
3901                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3902                 md_wakeup_thread(mddev->thread);
3903         }
3904         atomic_inc(&mddev->writes_pending);
3905         if (mddev->in_sync) {
3906                 spin_lock_irq(&mddev->write_lock);
3907                 if (mddev->in_sync) {
3908                         mddev->in_sync = 0;
3909                         mddev->sb_dirty = 1;
3910                         md_wakeup_thread(mddev->thread);
3911                 }
3912                 spin_unlock_irq(&mddev->write_lock);
3913         }
3914         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3915 }
3916
3917 void md_write_end(mddev_t *mddev)
3918 {
3919         if (atomic_dec_and_test(&mddev->writes_pending)) {
3920                 if (mddev->safemode == 2)
3921                         md_wakeup_thread(mddev->thread);
3922                 else
3923                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3924         }
3925 }
3926
3927 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3928
3929 #define SYNC_MARKS      10
3930 #define SYNC_MARK_STEP  (3*HZ)
3931 static void md_do_sync(mddev_t *mddev)
3932 {
3933         mddev_t *mddev2;
3934         unsigned int currspeed = 0,
3935                  window;
3936         sector_t max_sectors,j, io_sectors;
3937         unsigned long mark[SYNC_MARKS];
3938         sector_t mark_cnt[SYNC_MARKS];
3939         int last_mark,m;
3940         struct list_head *tmp;
3941         sector_t last_check;
3942         int skipped = 0;
3943
3944         /* just incase thread restarts... */
3945         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3946                 return;
3947
3948         /* we overload curr_resync somewhat here.
3949          * 0 == not engaged in resync at all
3950          * 2 == checking that there is no conflict with another sync
3951          * 1 == like 2, but have yielded to allow conflicting resync to
3952          *              commense
3953          * other == active in resync - this many blocks
3954          *
3955          * Before starting a resync we must have set curr_resync to
3956          * 2, and then checked that every "conflicting" array has curr_resync
3957          * less than ours.  When we find one that is the same or higher
3958          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3959          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3960          * This will mean we have to start checking from the beginning again.
3961          *
3962          */
3963
3964         do {
3965                 mddev->curr_resync = 2;
3966
3967         try_again:
3968                 if (kthread_should_stop()) {
3969                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3970                         goto skip;
3971                 }
3972                 ITERATE_MDDEV(mddev2,tmp) {
3973                         if (mddev2 == mddev)
3974                                 continue;
3975                         if (mddev2->curr_resync &&
3976                             match_mddev_units(mddev,mddev2)) {
3977                                 DEFINE_WAIT(wq);
3978                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3979                                         /* arbitrarily yield */
3980                                         mddev->curr_resync = 1;
3981                                         wake_up(&resync_wait);
3982                                 }
3983                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3984                                         /* no need to wait here, we can wait the next
3985                                          * time 'round when curr_resync == 2
3986                                          */
3987                                         continue;
3988                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
3989                                 if (!kthread_should_stop() &&
3990                                     mddev2->curr_resync >= mddev->curr_resync) {
3991                                         printk(KERN_INFO "md: delaying resync of %s"
3992                                                " until %s has finished resync (they"
3993                                                " share one or more physical units)\n",
3994                                                mdname(mddev), mdname(mddev2));
3995                                         mddev_put(mddev2);
3996                                         schedule();
3997                                         finish_wait(&resync_wait, &wq);
3998                                         goto try_again;
3999                                 }
4000                                 finish_wait(&resync_wait, &wq);
4001                         }
4002                 }
4003         } while (mddev->curr_resync < 2);
4004
4005         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4006                 /* resync follows the size requested by the personality,
4007                  * which defaults to physical size, but can be virtual size
4008                  */
4009                 max_sectors = mddev->resync_max_sectors;
4010                 mddev->resync_mismatches = 0;
4011         } else
4012                 /* recovery follows the physical size of devices */
4013                 max_sectors = mddev->size << 1;
4014
4015         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4016         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4017                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4018         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4019                "(but not more than %d KB/sec) for reconstruction.\n",
4020                sysctl_speed_limit_max);
4021
4022         is_mddev_idle(mddev); /* this also initializes IO event counters */
4023         /* we don't use the checkpoint if there's a bitmap */
4024         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4025             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4026                 j = mddev->recovery_cp;
4027         else
4028                 j = 0;
4029         io_sectors = 0;
4030         for (m = 0; m < SYNC_MARKS; m++) {
4031                 mark[m] = jiffies;
4032                 mark_cnt[m] = io_sectors;
4033         }
4034         last_mark = 0;
4035         mddev->resync_mark = mark[last_mark];
4036         mddev->resync_mark_cnt = mark_cnt[last_mark];
4037
4038         /*
4039          * Tune reconstruction:
4040          */
4041         window = 32*(PAGE_SIZE/512);
4042         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4043                 window/2,(unsigned long long) max_sectors/2);
4044
4045         atomic_set(&mddev->recovery_active, 0);
4046         init_waitqueue_head(&mddev->recovery_wait);
4047         last_check = 0;
4048
4049         if (j>2) {
4050                 printk(KERN_INFO
4051                         "md: resuming recovery of %s from checkpoint.\n",
4052                         mdname(mddev));
4053                 mddev->curr_resync = j;
4054         }
4055
4056         while (j < max_sectors) {
4057                 sector_t sectors;
4058
4059                 skipped = 0;
4060                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4061                                             currspeed < sysctl_speed_limit_min);
4062                 if (sectors == 0) {
4063                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4064                         goto out;
4065                 }
4066
4067                 if (!skipped) { /* actual IO requested */
4068                         io_sectors += sectors;
4069                         atomic_add(sectors, &mddev->recovery_active);
4070                 }
4071
4072                 j += sectors;
4073                 if (j>1) mddev->curr_resync = j;
4074
4075
4076                 if (last_check + window > io_sectors || j == max_sectors)
4077                         continue;
4078
4079                 last_check = io_sectors;
4080
4081                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4082                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4083                         break;
4084
4085         repeat:
4086                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4087                         /* step marks */
4088                         int next = (last_mark+1) % SYNC_MARKS;
4089
4090                         mddev->resync_mark = mark[next];
4091                         mddev->resync_mark_cnt = mark_cnt[next];
4092                         mark[next] = jiffies;
4093                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4094                         last_mark = next;
4095                 }
4096
4097
4098                 if (kthread_should_stop()) {
4099                         /*
4100                          * got a signal, exit.
4101                          */
4102                         printk(KERN_INFO
4103                                 "md: md_do_sync() got signal ... exiting\n");
4104                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4105                         goto out;
4106                 }
4107
4108                 /*
4109                  * this loop exits only if either when we are slower than
4110                  * the 'hard' speed limit, or the system was IO-idle for
4111                  * a jiffy.
4112                  * the system might be non-idle CPU-wise, but we only care
4113                  * about not overloading the IO subsystem. (things like an
4114                  * e2fsck being done on the RAID array should execute fast)
4115                  */
4116                 mddev->queue->unplug_fn(mddev->queue);
4117                 cond_resched();
4118
4119                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4120                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4121
4122                 if (currspeed > sysctl_speed_limit_min) {
4123                         if ((currspeed > sysctl_speed_limit_max) ||
4124                                         !is_mddev_idle(mddev)) {
4125                                 msleep(500);
4126                                 goto repeat;
4127                         }
4128                 }
4129         }
4130         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4131         /*
4132          * this also signals 'finished resyncing' to md_stop
4133          */
4134  out:
4135         mddev->queue->unplug_fn(mddev->queue);
4136
4137         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4138
4139         /* tell personality that we are finished */
4140         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4141
4142         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4143             mddev->curr_resync > 2 &&
4144             mddev->curr_resync >= mddev->recovery_cp) {
4145                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4146                         printk(KERN_INFO
4147                                 "md: checkpointing recovery of %s.\n",
4148                                 mdname(mddev));
4149                         mddev->recovery_cp = mddev->curr_resync;
4150                 } else
4151                         mddev->recovery_cp = MaxSector;
4152         }
4153
4154  skip:
4155         mddev->curr_resync = 0;
4156         wake_up(&resync_wait);
4157         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4158         md_wakeup_thread(mddev->thread);
4159 }
4160
4161
4162 /*
4163  * This routine is regularly called by all per-raid-array threads to
4164  * deal with generic issues like resync and super-block update.
4165  * Raid personalities that don't have a thread (linear/raid0) do not
4166  * need this as they never do any recovery or update the superblock.
4167  *
4168  * It does not do any resync itself, but rather "forks" off other threads
4169  * to do that as needed.
4170  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4171  * "->recovery" and create a thread at ->sync_thread.
4172  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4173  * and wakeups up this thread which will reap the thread and finish up.
4174  * This thread also removes any faulty devices (with nr_pending == 0).
4175  *
4176  * The overall approach is:
4177  *  1/ if the superblock needs updating, update it.
4178  *  2/ If a recovery thread is running, don't do anything else.
4179  *  3/ If recovery has finished, clean up, possibly marking spares active.
4180  *  4/ If there are any faulty devices, remove them.
4181  *  5/ If array is degraded, try to add spares devices
4182  *  6/ If array has spares or is not in-sync, start a resync thread.
4183  */
4184 void md_check_recovery(mddev_t *mddev)
4185 {
4186         mdk_rdev_t *rdev;
4187         struct list_head *rtmp;
4188
4189
4190         if (mddev->bitmap)
4191                 bitmap_daemon_work(mddev->bitmap);
4192
4193         if (mddev->ro)
4194                 return;
4195
4196         if (signal_pending(current)) {
4197                 if (mddev->pers->sync_request) {
4198                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4199                                mdname(mddev));
4200                         mddev->safemode = 2;
4201                 }
4202                 flush_signals(current);
4203         }
4204
4205         if ( ! (
4206                 mddev->sb_dirty ||
4207                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4208                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4209                 (mddev->safemode == 1) ||
4210                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4211                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4212                 ))
4213                 return;
4214
4215         if (mddev_trylock(mddev)==0) {
4216                 int spares =0;
4217
4218                 spin_lock_irq(&mddev->write_lock);
4219                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4220                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4221                         mddev->in_sync = 1;
4222                         mddev->sb_dirty = 1;
4223                 }
4224                 if (mddev->safemode == 1)
4225                         mddev->safemode = 0;
4226                 spin_unlock_irq(&mddev->write_lock);
4227
4228                 if (mddev->sb_dirty)
4229                         md_update_sb(mddev);
4230
4231
4232                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4233                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4234                         /* resync/recovery still happening */
4235                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4236                         goto unlock;
4237                 }
4238                 if (mddev->sync_thread) {
4239                         /* resync has finished, collect result */
4240                         md_unregister_thread(mddev->sync_thread);
4241                         mddev->sync_thread = NULL;
4242                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4243                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4244                                 /* success...*/
4245                                 /* activate any spares */
4246                                 mddev->pers->spare_active(mddev);
4247                         }
4248                         md_update_sb(mddev);
4249
4250                         /* if array is no-longer degraded, then any saved_raid_disk
4251                          * information must be scrapped
4252                          */
4253                         if (!mddev->degraded)
4254                                 ITERATE_RDEV(mddev,rdev,rtmp)
4255                                         rdev->saved_raid_disk = -1;
4256
4257                         mddev->recovery = 0;
4258                         /* flag recovery needed just to double check */
4259                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4260                         goto unlock;
4261                 }
4262                 /* Clear some bits that don't mean anything, but
4263                  * might be left set
4264                  */
4265                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4266                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4267                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4268                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4269
4270                 /* no recovery is running.
4271                  * remove any failed drives, then
4272                  * add spares if possible.
4273                  * Spare are also removed and re-added, to allow
4274                  * the personality to fail the re-add.
4275                  */
4276                 ITERATE_RDEV(mddev,rdev,rtmp)
4277                         if (rdev->raid_disk >= 0 &&
4278                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4279                             atomic_read(&rdev->nr_pending)==0) {
4280                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4281                                         char nm[20];
4282                                         sprintf(nm,"rd%d", rdev->raid_disk);
4283                                         sysfs_remove_link(&mddev->kobj, nm);
4284                                         rdev->raid_disk = -1;
4285                                 }
4286                         }
4287
4288                 if (mddev->degraded) {
4289                         ITERATE_RDEV(mddev,rdev,rtmp)
4290                                 if (rdev->raid_disk < 0
4291                                     && !test_bit(Faulty, &rdev->flags)) {
4292                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4293                                                 char nm[20];
4294                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4295                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4296                                                 spares++;
4297                                         } else
4298                                                 break;
4299                                 }
4300                 }
4301
4302                 if (spares) {
4303                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4304                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4305                 } else if (mddev->recovery_cp < MaxSector) {
4306                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4307                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4308                         /* nothing to be done ... */
4309                         goto unlock;
4310
4311                 if (mddev->pers->sync_request) {
4312                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4313                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4314                                 /* We are adding a device or devices to an array
4315                                  * which has the bitmap stored on all devices.
4316                                  * So make sure all bitmap pages get written
4317                                  */
4318                                 bitmap_write_all(mddev->bitmap);
4319                         }
4320                         mddev->sync_thread = md_register_thread(md_do_sync,
4321                                                                 mddev,
4322                                                                 "%s_resync");
4323                         if (!mddev->sync_thread) {
4324                                 printk(KERN_ERR "%s: could not start resync"
4325                                         " thread...\n",
4326                                         mdname(mddev));
4327                                 /* leave the spares where they are, it shouldn't hurt */
4328                                 mddev->recovery = 0;
4329                         } else {
4330                                 md_wakeup_thread(mddev->sync_thread);
4331                         }
4332                 }
4333         unlock:
4334                 mddev_unlock(mddev);
4335         }
4336 }
4337
4338 static int md_notify_reboot(struct notifier_block *this,
4339                             unsigned long code, void *x)
4340 {
4341         struct list_head *tmp;
4342         mddev_t *mddev;
4343
4344         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4345
4346                 printk(KERN_INFO "md: stopping all md devices.\n");
4347
4348                 ITERATE_MDDEV(mddev,tmp)
4349                         if (mddev_trylock(mddev)==0)
4350                                 do_md_stop (mddev, 1);
4351                 /*
4352                  * certain more exotic SCSI devices are known to be
4353                  * volatile wrt too early system reboots. While the
4354                  * right place to handle this issue is the given
4355                  * driver, we do want to have a safe RAID driver ...
4356                  */
4357                 mdelay(1000*1);
4358         }
4359         return NOTIFY_DONE;
4360 }
4361
4362 static struct notifier_block md_notifier = {
4363         .notifier_call  = md_notify_reboot,
4364         .next           = NULL,
4365         .priority       = INT_MAX, /* before any real devices */
4366 };
4367
4368 static void md_geninit(void)
4369 {
4370         struct proc_dir_entry *p;
4371
4372         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4373
4374         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4375         if (p)
4376                 p->proc_fops = &md_seq_fops;
4377 }
4378
4379 static int __init md_init(void)
4380 {
4381         int minor;
4382
4383         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4384                         " MD_SB_DISKS=%d\n",
4385                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4386                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4387         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4388                         BITMAP_MINOR);
4389
4390         if (register_blkdev(MAJOR_NR, "md"))
4391                 return -1;
4392         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4393                 unregister_blkdev(MAJOR_NR, "md");
4394                 return -1;
4395         }
4396         devfs_mk_dir("md");
4397         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4398                                 md_probe, NULL, NULL);
4399         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4400                             md_probe, NULL, NULL);
4401
4402         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4403                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4404                                 S_IFBLK|S_IRUSR|S_IWUSR,
4405                                 "md/%d", minor);
4406
4407         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4408                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4409                               S_IFBLK|S_IRUSR|S_IWUSR,
4410                               "md/mdp%d", minor);
4411
4412
4413         register_reboot_notifier(&md_notifier);
4414         raid_table_header = register_sysctl_table(raid_root_table, 1);
4415
4416         md_geninit();
4417         return (0);
4418 }
4419
4420
4421 #ifndef MODULE
4422
4423 /*
4424  * Searches all registered partitions for autorun RAID arrays
4425  * at boot time.
4426  */
4427 static dev_t detected_devices[128];
4428 static int dev_cnt;
4429
4430 void md_autodetect_dev(dev_t dev)
4431 {
4432         if (dev_cnt >= 0 && dev_cnt < 127)
4433                 detected_devices[dev_cnt++] = dev;
4434 }
4435
4436
4437 static void autostart_arrays(int part)
4438 {
4439         mdk_rdev_t *rdev;
4440         int i;
4441
4442         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4443
4444         for (i = 0; i < dev_cnt; i++) {
4445                 dev_t dev = detected_devices[i];
4446
4447                 rdev = md_import_device(dev,0, 0);
4448                 if (IS_ERR(rdev))
4449                         continue;
4450
4451                 if (test_bit(Faulty, &rdev->flags)) {
4452                         MD_BUG();
4453                         continue;
4454                 }
4455                 list_add(&rdev->same_set, &pending_raid_disks);
4456         }
4457         dev_cnt = 0;
4458
4459         autorun_devices(part);
4460 }
4461
4462 #endif
4463
4464 static __exit void md_exit(void)
4465 {
4466         mddev_t *mddev;
4467         struct list_head *tmp;
4468         int i;
4469         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4470         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4471         for (i=0; i < MAX_MD_DEVS; i++)
4472                 devfs_remove("md/%d", i);
4473         for (i=0; i < MAX_MD_DEVS; i++)
4474                 devfs_remove("md/d%d", i);
4475
4476         devfs_remove("md");
4477
4478         unregister_blkdev(MAJOR_NR,"md");
4479         unregister_blkdev(mdp_major, "mdp");
4480         unregister_reboot_notifier(&md_notifier);
4481         unregister_sysctl_table(raid_table_header);
4482         remove_proc_entry("mdstat", NULL);
4483         ITERATE_MDDEV(mddev,tmp) {
4484                 struct gendisk *disk = mddev->gendisk;
4485                 if (!disk)
4486                         continue;
4487                 export_array(mddev);
4488                 del_gendisk(disk);
4489                 put_disk(disk);
4490                 mddev->gendisk = NULL;
4491                 mddev_put(mddev);
4492         }
4493 }
4494
4495 module_init(md_init)
4496 module_exit(md_exit)
4497
4498 static int get_ro(char *buffer, struct kernel_param *kp)
4499 {
4500         return sprintf(buffer, "%d", start_readonly);
4501 }
4502 static int set_ro(const char *val, struct kernel_param *kp)
4503 {
4504         char *e;
4505         int num = simple_strtoul(val, &e, 10);
4506         if (*val && (*e == '\0' || *e == '\n')) {
4507                 start_readonly = num;
4508                 return 0;;
4509         }
4510         return -EINVAL;
4511 }
4512
4513 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4514 module_param(start_dirty_degraded, int, 0644);
4515
4516
4517 EXPORT_SYMBOL(register_md_personality);
4518 EXPORT_SYMBOL(unregister_md_personality);
4519 EXPORT_SYMBOL(md_error);
4520 EXPORT_SYMBOL(md_done_sync);
4521 EXPORT_SYMBOL(md_write_start);
4522 EXPORT_SYMBOL(md_write_end);
4523 EXPORT_SYMBOL(md_register_thread);
4524 EXPORT_SYMBOL(md_unregister_thread);
4525 EXPORT_SYMBOL(md_wakeup_thread);
4526 EXPORT_SYMBOL(md_print_devices);
4527 EXPORT_SYMBOL(md_check_recovery);
4528 MODULE_LICENSE("GPL");
4529 MODULE_ALIAS("md");
4530 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);