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 EXPORT_SYMBOL(sync_page_io);
 465
 466 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 467 {
 468         char b[BDEVNAME_SIZE];
 469         if (!rdev->sb_page) {
 470                 MD_BUG();
 471                 return -EINVAL;
 472         }
 473         if (rdev->sb_loaded)
 474                 return 0;
 475
 476
 477         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 478                 goto fail;
 479         rdev->sb_loaded = 1;
 480         return 0;
 481
 482 fail:
 483         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 484                 bdevname(rdev->bdev,b));
 485         return -EINVAL;
 486 }
 487
 488 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 489 {
 490         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 491                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 492                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 493                 (sb1->set_uuid3 == sb2->set_uuid3))
 494
 495                 return 1;
 496
 497         return 0;
 498 }
 499
 500
 501 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 502 {
 503         int ret;
 504         mdp_super_t *tmp1, *tmp2;
 505
 506         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 507         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 508
 509         if (!tmp1 || !tmp2) {
 510                 ret = 0;
 511                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 512                 goto abort;
 513         }
 514
 515         *tmp1 = *sb1;
 516         *tmp2 = *sb2;
 517
 518         /*
 519          * nr_disks is not constant
 520          */
 521         tmp1->nr_disks = 0;
 522         tmp2->nr_disks = 0;
 523
 524         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 525                 ret = 0;
 526         else
 527                 ret = 1;
 528
 529 abort:
 530         kfree(tmp1);
 531         kfree(tmp2);
 532         return ret;
 533 }
 534
 535 static unsigned int calc_sb_csum(mdp_super_t * sb)
 536 {
 537         unsigned int disk_csum, csum;
 538
 539         disk_csum = sb->sb_csum;
 540         sb->sb_csum = 0;
 541         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 542         sb->sb_csum = disk_csum;
 543         return csum;
 544 }
 545
 546
 547 /*
 548  * Handle superblock details.
 549  * We want to be able to handle multiple superblock formats
 550  * so we have a common interface to them all, and an array of
 551  * different handlers.
 552  * We rely on user-space to write the initial superblock, and support
 553  * reading and updating of superblocks.
 554  * Interface methods are:
 555  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 556  *      loads and validates a superblock on dev.
 557  *      if refdev != NULL, compare superblocks on both devices
 558  *    Return:
 559  *      0 - dev has a superblock that is compatible with refdev
 560  *      1 - dev has a superblock that is compatible and newer than refdev
 561  *          so dev should be used as the refdev in future
 562  *     -EINVAL superblock incompatible or invalid
 563  *     -othererror e.g. -EIO
 564  *
 565  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 566  *      Verify that dev is acceptable into mddev.
 567  *       The first time, mddev->raid_disks will be 0, and data from
 568  *       dev should be merged in.  Subsequent calls check that dev
 569  *       is new enough.  Return 0 or -EINVAL
 570  *
 571  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 572  *     Update the superblock for rdev with data in mddev
 573  *     This does not write to disc.
 574  *
 575  */
 576
 577 struct super_type  {
 578         char            *name;
 579         struct module   *owner;
 580         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 581         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 582         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 583 };
 584
 585 /*
 586  * load_super for 0.90.0
 587  */
 588 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 589 {
 590         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 591         mdp_super_t *sb;
 592         int ret;
 593         sector_t sb_offset;
 594
 595         /*
 596          * Calculate the position of the superblock,
 597          * it's at the end of the disk.
 598          *
 599          * It also happens to be a multiple of 4Kb.
 600          */
 601         sb_offset = calc_dev_sboffset(rdev->bdev);
 602         rdev->sb_offset = sb_offset;
 603
 604         ret = read_disk_sb(rdev, MD_SB_BYTES);
 605         if (ret) return ret;
 606
 607         ret = -EINVAL;
 608
 609         bdevname(rdev->bdev, b);
 610         sb = (mdp_super_t*)page_address(rdev->sb_page);
 611
 612         if (sb->md_magic != MD_SB_MAGIC) {
 613                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 614                        b);
 615                 goto abort;
 616         }
 617
 618         if (sb->major_version != 0 ||
 619             sb->minor_version != 90) {
 620                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 621                         sb->major_version, sb->minor_version,
 622                         b);
 623                 goto abort;
 624         }
 625
 626         if (sb->raid_disks <= 0)
 627                 goto abort;
 628
 629         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 630                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 631                         b);
 632                 goto abort;
 633         }
 634
 635         rdev->preferred_minor = sb->md_minor;
 636         rdev->data_offset = 0;
 637         rdev->sb_size = MD_SB_BYTES;
 638
 639         if (sb->level == LEVEL_MULTIPATH)
 640                 rdev->desc_nr = -1;
 641         else
 642                 rdev->desc_nr = sb->this_disk.number;
 643
 644         if (refdev == 0)
 645                 ret = 1;
 646         else {
 647                 __u64 ev1, ev2;
 648                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 649                 if (!uuid_equal(refsb, sb)) {
 650                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 651                                 b, bdevname(refdev->bdev,b2));
 652                         goto abort;
 653                 }
 654                 if (!sb_equal(refsb, sb)) {
 655                         printk(KERN_WARNING "md: %s has same UUID"
 656                                " but different superblock to %s\n",
 657                                b, bdevname(refdev->bdev, b2));
 658                         goto abort;
 659                 }
 660                 ev1 = md_event(sb);
 661                 ev2 = md_event(refsb);
 662                 if (ev1 > ev2)
 663                         ret = 1;
 664                 else
 665                         ret = 0;
 666         }
 667         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 668
 669  abort:
 670         return ret;
 671 }
 672
 673 /*
 674  * validate_super for 0.90.0
 675  */
 676 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 677 {
 678         mdp_disk_t *desc;
 679         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 680
 681         rdev->raid_disk = -1;
 682         rdev->flags = 0;
 683         if (mddev->raid_disks == 0) {
 684                 mddev->major_version = 0;
 685                 mddev->minor_version = sb->minor_version;
 686                 mddev->patch_version = sb->patch_version;
 687                 mddev->persistent = ! sb->not_persistent;
 688                 mddev->chunk_size = sb->chunk_size;
 689                 mddev->ctime = sb->ctime;
 690                 mddev->utime = sb->utime;
 691                 mddev->level = sb->level;
 692                 mddev->layout = sb->layout;
 693                 mddev->raid_disks = sb->raid_disks;
 694                 mddev->size = sb->size;
 695                 mddev->events = md_event(sb);
 696                 mddev->bitmap_offset = 0;
 697                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 698
 699                 if (sb->state & (1<<MD_SB_CLEAN))
 700                         mddev->recovery_cp = MaxSector;
 701                 else {
 702                         if (sb->events_hi == sb->cp_events_hi &&
 703                                 sb->events_lo == sb->cp_events_lo) {
 704                                 mddev->recovery_cp = sb->recovery_cp;
 705                         } else
 706                                 mddev->recovery_cp = 0;
 707                 }
 708
 709                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 710                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 711                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 712                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 713
 714                 mddev->max_disks = MD_SB_DISKS;
 715
 716                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 717                     mddev->bitmap_file == NULL) {
 718                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
 719                             && mddev->level != 10) {
 720                                 /* FIXME use a better test */
 721                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
 722                                 return -EINVAL;
 723                         }
 724                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 725                 }
 726
 727         } else if (mddev->pers == NULL) {
 728                 /* Insist on good event counter while assembling */
 729                 __u64 ev1 = md_event(sb);
 730                 ++ev1;
 731                 if (ev1 < mddev->events)
 732                         return -EINVAL;
 733         } else if (mddev->bitmap) {
 734                 /* if adding to array with a bitmap, then we can accept an
 735                  * older device ... but not too old.
 736                  */
 737                 __u64 ev1 = md_event(sb);
 738                 if (ev1 < mddev->bitmap->events_cleared)
 739                         return 0;
 740         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 741                 return 0;
 742
 743         if (mddev->level != LEVEL_MULTIPATH) {
 744                 desc = sb->disks + rdev->desc_nr;
 745
 746                 if (desc->state & (1<<MD_DISK_FAULTY))
 747                         set_bit(Faulty, &rdev->flags);
 748                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 749                          desc->raid_disk < mddev->raid_disks) {
 750                         set_bit(In_sync, &rdev->flags);
 751                         rdev->raid_disk = desc->raid_disk;
 752                 }
 753                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 754                         set_bit(WriteMostly, &rdev->flags);
 755         } else /* MULTIPATH are always insync */
 756                 set_bit(In_sync, &rdev->flags);
 757         return 0;
 758 }
 759
 760 /*
 761  * sync_super for 0.90.0
 762  */
 763 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 764 {
 765         mdp_super_t *sb;
 766         struct list_head *tmp;
 767         mdk_rdev_t *rdev2;
 768         int next_spare = mddev->raid_disks;
 769
 770
 771         /* make rdev->sb match mddev data..
 772          *
 773          * 1/ zero out disks
 774          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 775          * 3/ any empty disks < next_spare become removed
 776          *
 777          * disks[0] gets initialised to REMOVED because
 778          * we cannot be sure from other fields if it has
 779          * been initialised or not.
 780          */
 781         int i;
 782         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 783
 784         rdev->sb_size = MD_SB_BYTES;
 785
 786         sb = (mdp_super_t*)page_address(rdev->sb_page);
 787
 788         memset(sb, 0, sizeof(*sb));
 789
 790         sb->md_magic = MD_SB_MAGIC;
 791         sb->major_version = mddev->major_version;
 792         sb->minor_version = mddev->minor_version;
 793         sb->patch_version = mddev->patch_version;
 794         sb->gvalid_words  = 0; /* ignored */
 795         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 796         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 797         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 798         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 799
 800         sb->ctime = mddev->ctime;
 801         sb->level = mddev->level;
 802         sb->size  = mddev->size;
 803         sb->raid_disks = mddev->raid_disks;
 804         sb->md_minor = mddev->md_minor;
 805         sb->not_persistent = !mddev->persistent;
 806         sb->utime = mddev->utime;
 807         sb->state = 0;
 808         sb->events_hi = (mddev->events>>32);
 809         sb->events_lo = (u32)mddev->events;
 810
 811         if (mddev->in_sync)
 812         {
 813                 sb->recovery_cp = mddev->recovery_cp;
 814                 sb->cp_events_hi = (mddev->events>>32);
 815                 sb->cp_events_lo = (u32)mddev->events;
 816                 if (mddev->recovery_cp == MaxSector)
 817                         sb->state = (1<< MD_SB_CLEAN);
 818         } else
 819                 sb->recovery_cp = 0;
 820
 821         sb->layout = mddev->layout;
 822         sb->chunk_size = mddev->chunk_size;
 823
 824         if (mddev->bitmap && mddev->bitmap_file == NULL)
 825                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 826
 827         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 828         ITERATE_RDEV(mddev,rdev2,tmp) {
 829                 mdp_disk_t *d;
 830                 int desc_nr;
 831                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 832                     && !test_bit(Faulty, &rdev2->flags))
 833                         desc_nr = rdev2->raid_disk;
 834                 else
 835                         desc_nr = next_spare++;
 836                 rdev2->desc_nr = desc_nr;
 837                 d = &sb->disks[rdev2->desc_nr];
 838                 nr_disks++;
 839                 d->number = rdev2->desc_nr;
 840                 d->major = MAJOR(rdev2->bdev->bd_dev);
 841                 d->minor = MINOR(rdev2->bdev->bd_dev);
 842                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 843                     && !test_bit(Faulty, &rdev2->flags))
 844                         d->raid_disk = rdev2->raid_disk;
 845                 else
 846                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 847                 if (test_bit(Faulty, &rdev2->flags)) {
 848                         d->state = (1<<MD_DISK_FAULTY);
 849                         failed++;
 850                 } else if (test_bit(In_sync, &rdev2->flags)) {
 851                         d->state = (1<<MD_DISK_ACTIVE);
 852                         d->state |= (1<<MD_DISK_SYNC);
 853                         active++;
 854                         working++;
 855                 } else {
 856                         d->state = 0;
 857                         spare++;
 858                         working++;
 859                 }
 860                 if (test_bit(WriteMostly, &rdev2->flags))
 861                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 862         }
 863         /* now set the "removed" and "faulty" bits on any missing devices */
 864         for (i=0 ; i < mddev->raid_disks ; i++) {
 865                 mdp_disk_t *d = &sb->disks[i];
 866                 if (d->state == 0 && d->number == 0) {
 867                         d->number = i;
 868                         d->raid_disk = i;
 869                         d->state = (1<<MD_DISK_REMOVED);
 870                         d->state |= (1<<MD_DISK_FAULTY);
 871                         failed++;
 872                 }
 873         }
 874         sb->nr_disks = nr_disks;
 875         sb->active_disks = active;
 876         sb->working_disks = working;
 877         sb->failed_disks = failed;
 878         sb->spare_disks = spare;
 879
 880         sb->this_disk = sb->disks[rdev->desc_nr];
 881         sb->sb_csum = calc_sb_csum(sb);
 882 }
 883
 884 /*
 885  * version 1 superblock
 886  */
 887
 888 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 889 {
 890         unsigned int disk_csum, csum;
 891         unsigned long long newcsum;
 892         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 893         unsigned int *isuper = (unsigned int*)sb;
 894         int i;
 895
 896         disk_csum = sb->sb_csum;
 897         sb->sb_csum = 0;
 898         newcsum = 0;
 899         for (i=0; size>=4; size -= 4 )
 900                 newcsum += le32_to_cpu(*isuper++);
 901
 902         if (size == 2)
 903                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 904
 905         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 906         sb->sb_csum = disk_csum;
 907         return cpu_to_le32(csum);
 908 }
 909
 910 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 911 {
 912         struct mdp_superblock_1 *sb;
 913         int ret;
 914         sector_t sb_offset;
 915         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 916         int bmask;
 917
 918         /*
 919          * Calculate the position of the superblock.
 920          * It is always aligned to a 4K boundary and
 921          * depeding on minor_version, it can be:
 922          * 0: At least 8K, but less than 12K, from end of device
 923          * 1: At start of device
 924          * 2: 4K from start of device.
 925          */
 926         switch(minor_version) {
 927         case 0:
 928                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 929                 sb_offset -= 8*2;
 930                 sb_offset &= ~(sector_t)(4*2-1);
 931                 /* convert from sectors to K */
 932                 sb_offset /= 2;
 933                 break;
 934         case 1:
 935                 sb_offset = 0;
 936                 break;
 937         case 2:
 938                 sb_offset = 4;
 939                 break;
 940         default:
 941                 return -EINVAL;
 942         }
 943         rdev->sb_offset = sb_offset;
 944
 945         /* superblock is rarely larger than 1K, but it can be larger,
 946          * and it is safe to read 4k, so we do that
 947          */
 948         ret = read_disk_sb(rdev, 4096);
 949         if (ret) return ret;
 950
 951
 952         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 953
 954         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 955             sb->major_version != cpu_to_le32(1) ||
 956             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 957             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 958             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 959                 return -EINVAL;
 960
 961         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 962                 printk("md: invalid superblock checksum on %s\n",
 963                         bdevname(rdev->bdev,b));
 964                 return -EINVAL;
 965         }
 966         if (le64_to_cpu(sb->data_size) < 10) {
 967                 printk("md: data_size too small on %s\n",
 968                        bdevname(rdev->bdev,b));
 969                 return -EINVAL;
 970         }
 971         rdev->preferred_minor = 0xffff;
 972         rdev->data_offset = le64_to_cpu(sb->data_offset);
 973
 974         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 975         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
 976         if (rdev->sb_size & bmask)
 977                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 978
 979         if (refdev == 0)
 980                 return 1;
 981         else {
 982                 __u64 ev1, ev2;
 983                 struct mdp_superblock_1 *refsb =
 984                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 985
 986                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 987                     sb->level != refsb->level ||
 988                     sb->layout != refsb->layout ||
 989                     sb->chunksize != refsb->chunksize) {
 990                         printk(KERN_WARNING "md: %s has strangely different"
 991                                 " superblock to %s\n",
 992                                 bdevname(rdev->bdev,b),
 993                                 bdevname(refdev->bdev,b2));
 994                         return -EINVAL;
 995                 }
 996                 ev1 = le64_to_cpu(sb->events);
 997                 ev2 = le64_to_cpu(refsb->events);
 998
 999                 if (ev1 > ev2)
1000                         return 1;
1001         }
1002         if (minor_version)
1003                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1004         else
1005                 rdev->size = rdev->sb_offset;
1006         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1007                 return -EINVAL;
1008         rdev->size = le64_to_cpu(sb->data_size)/2;
1009         if (le32_to_cpu(sb->chunksize))
1010                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1011         return 0;
1012 }
1013
1014 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1015 {
1016         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1017
1018         rdev->raid_disk = -1;
1019         rdev->flags = 0;
1020         if (mddev->raid_disks == 0) {
1021                 mddev->major_version = 1;
1022                 mddev->patch_version = 0;
1023                 mddev->persistent = 1;
1024                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1025                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1026                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1027                 mddev->level = le32_to_cpu(sb->level);
1028                 mddev->layout = le32_to_cpu(sb->layout);
1029                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1030                 mddev->size = le64_to_cpu(sb->size)/2;
1031                 mddev->events = le64_to_cpu(sb->events);
1032                 mddev->bitmap_offset = 0;
1033                 mddev->default_bitmap_offset = 1024;
1034
1035                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1036                 memcpy(mddev->uuid, sb->set_uuid, 16);
1037
1038                 mddev->max_disks =  (4096-256)/2;
1039
1040                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1041                     mddev->bitmap_file == NULL ) {
1042                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1043                             && mddev->level != 10) {
1044                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1045                                 return -EINVAL;
1046                         }
1047                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1048                 }
1049         } else if (mddev->pers == NULL) {
1050                 /* Insist of good event counter while assembling */
1051                 __u64 ev1 = le64_to_cpu(sb->events);
1052                 ++ev1;
1053                 if (ev1 < mddev->events)
1054                         return -EINVAL;
1055         } else if (mddev->bitmap) {
1056                 /* If adding to array with a bitmap, then we can accept an
1057                  * older device, but not too old.
1058                  */
1059                 __u64 ev1 = le64_to_cpu(sb->events);
1060                 if (ev1 < mddev->bitmap->events_cleared)
1061                         return 0;
1062         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1063                 return 0;
1064
1065         if (mddev->level != LEVEL_MULTIPATH) {
1066                 int role;
1067                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1068                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1069                 switch(role) {
1070                 case 0xffff: /* spare */
1071                         break;
1072                 case 0xfffe: /* faulty */
1073                         set_bit(Faulty, &rdev->flags);
1074                         break;
1075                 default:
1076                         set_bit(In_sync, &rdev->flags);
1077                         rdev->raid_disk = role;
1078                         break;
1079                 }
1080                 if (sb->devflags & WriteMostly1)
1081                         set_bit(WriteMostly, &rdev->flags);
1082         } else /* MULTIPATH are always insync */
1083                 set_bit(In_sync, &rdev->flags);
1084
1085         return 0;
1086 }
1087
1088 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1089 {
1090         struct mdp_superblock_1 *sb;
1091         struct list_head *tmp;
1092         mdk_rdev_t *rdev2;
1093         int max_dev, i;
1094         /* make rdev->sb match mddev and rdev data. */
1095
1096         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1097
1098         sb->feature_map = 0;
1099         sb->pad0 = 0;
1100         memset(sb->pad1, 0, sizeof(sb->pad1));
1101         memset(sb->pad2, 0, sizeof(sb->pad2));
1102         memset(sb->pad3, 0, sizeof(sb->pad3));
1103
1104         sb->utime = cpu_to_le64((__u64)mddev->utime);
1105         sb->events = cpu_to_le64(mddev->events);
1106         if (mddev->in_sync)
1107                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1108         else
1109                 sb->resync_offset = cpu_to_le64(0);
1110
1111         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1112                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1113                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1114         }
1115
1116         max_dev = 0;
1117         ITERATE_RDEV(mddev,rdev2,tmp)
1118                 if (rdev2->desc_nr+1 > max_dev)
1119                         max_dev = rdev2->desc_nr+1;
1120
1121         sb->max_dev = cpu_to_le32(max_dev);
1122         for (i=0; i<max_dev;i++)
1123                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1124
1125         ITERATE_RDEV(mddev,rdev2,tmp) {
1126                 i = rdev2->desc_nr;
1127                 if (test_bit(Faulty, &rdev2->flags))
1128                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1129                 else if (test_bit(In_sync, &rdev2->flags))
1130                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1131                 else
1132                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1133         }
1134
1135         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1136         sb->sb_csum = calc_sb_1_csum(sb);
1137 }
1138
1139
1140 static struct super_type super_types[] = {
1141         [0] = {
1142                 .name   = "0.90.0",
1143                 .owner  = THIS_MODULE,
1144                 .load_super     = super_90_load,
1145                 .validate_super = super_90_validate,
1146                 .sync_super     = super_90_sync,
1147         },
1148         [1] = {
1149                 .name   = "md-1",
1150                 .owner  = THIS_MODULE,
1151                 .load_super     = super_1_load,
1152                 .validate_super = super_1_validate,
1153                 .sync_super     = super_1_sync,
1154         },
1155 };
1156
1157 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1158 {
1159         struct list_head *tmp;
1160         mdk_rdev_t *rdev;
1161
1162         ITERATE_RDEV(mddev,rdev,tmp)
1163                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1164                         return rdev;
1165
1166         return NULL;
1167 }
1168
1169 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1170 {
1171         struct list_head *tmp;
1172         mdk_rdev_t *rdev;
1173
1174         ITERATE_RDEV(mddev1,rdev,tmp)
1175                 if (match_dev_unit(mddev2, rdev))
1176                         return 1;
1177
1178         return 0;
1179 }
1180
1181 static LIST_HEAD(pending_raid_disks);
1182
1183 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1184 {
1185         mdk_rdev_t *same_pdev;
1186         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1187         struct kobject *ko;
1188
1189         if (rdev->mddev) {
1190                 MD_BUG();
1191                 return -EINVAL;
1192         }
1193         same_pdev = match_dev_unit(mddev, rdev);
1194         if (same_pdev)
1195                 printk(KERN_WARNING
1196                         "%s: WARNING: %s appears to be on the same physical"
1197                         " disk as %s. True\n     protection against single-disk"
1198                         " failure might be compromised.\n",
1199                         mdname(mddev), bdevname(rdev->bdev,b),
1200                         bdevname(same_pdev->bdev,b2));
1201
1202         /* Verify rdev->desc_nr is unique.
1203          * If it is -1, assign a free number, else
1204          * check number is not in use
1205          */
1206         if (rdev->desc_nr < 0) {
1207                 int choice = 0;
1208                 if (mddev->pers) choice = mddev->raid_disks;
1209                 while (find_rdev_nr(mddev, choice))
1210                         choice++;
1211                 rdev->desc_nr = choice;
1212         } else {
1213                 if (find_rdev_nr(mddev, rdev->desc_nr))
1214                         return -EBUSY;
1215         }
1216         bdevname(rdev->bdev,b);
1217         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1218                 return -ENOMEM;
1219
1220         list_add(&rdev->same_set, &mddev->disks);
1221         rdev->mddev = mddev;
1222         printk(KERN_INFO "md: bind<%s>\n", b);
1223
1224         rdev->kobj.parent = &mddev->kobj;
1225         kobject_add(&rdev->kobj);
1226
1227         if (rdev->bdev->bd_part)
1228                 ko = &rdev->bdev->bd_part->kobj;
1229         else
1230                 ko = &rdev->bdev->bd_disk->kobj;
1231         sysfs_create_link(&rdev->kobj, ko, "block");
1232         return 0;
1233 }
1234
1235 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1236 {
1237         char b[BDEVNAME_SIZE];
1238         if (!rdev->mddev) {
1239                 MD_BUG();
1240                 return;
1241         }
1242         list_del_init(&rdev->same_set);
1243         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1244         rdev->mddev = NULL;
1245         sysfs_remove_link(&rdev->kobj, "block");
1246         kobject_del(&rdev->kobj);
1247 }
1248
1249 /*
1250  * prevent the device from being mounted, repartitioned or
1251  * otherwise reused by a RAID array (or any other kernel
1252  * subsystem), by bd_claiming the device.
1253  */
1254 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1255 {
1256         int err = 0;
1257         struct block_device *bdev;
1258         char b[BDEVNAME_SIZE];
1259
1260         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1261         if (IS_ERR(bdev)) {
1262                 printk(KERN_ERR "md: could not open %s.\n",
1263                         __bdevname(dev, b));
1264                 return PTR_ERR(bdev);
1265         }
1266         err = bd_claim(bdev, rdev);
1267         if (err) {
1268                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1269                         bdevname(bdev, b));
1270                 blkdev_put(bdev);
1271                 return err;
1272         }
1273         rdev->bdev = bdev;
1274         return err;
1275 }
1276
1277 static void unlock_rdev(mdk_rdev_t *rdev)
1278 {
1279         struct block_device *bdev = rdev->bdev;
1280         rdev->bdev = NULL;
1281         if (!bdev)
1282                 MD_BUG();
1283         bd_release(bdev);
1284         blkdev_put(bdev);
1285 }
1286
1287 void md_autodetect_dev(dev_t dev);
1288
1289 static void export_rdev(mdk_rdev_t * rdev)
1290 {
1291         char b[BDEVNAME_SIZE];
1292         printk(KERN_INFO "md: export_rdev(%s)\n",
1293                 bdevname(rdev->bdev,b));
1294         if (rdev->mddev)
1295                 MD_BUG();
1296         free_disk_sb(rdev);
1297         list_del_init(&rdev->same_set);
1298 #ifndef MODULE
1299         md_autodetect_dev(rdev->bdev->bd_dev);
1300 #endif
1301         unlock_rdev(rdev);
1302         kobject_put(&rdev->kobj);
1303 }
1304
1305 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1306 {
1307         unbind_rdev_from_array(rdev);
1308         export_rdev(rdev);
1309 }
1310
1311 static void export_array(mddev_t *mddev)
1312 {
1313         struct list_head *tmp;
1314         mdk_rdev_t *rdev;
1315
1316         ITERATE_RDEV(mddev,rdev,tmp) {
1317                 if (!rdev->mddev) {
1318                         MD_BUG();
1319                         continue;
1320                 }
1321                 kick_rdev_from_array(rdev);
1322         }
1323         if (!list_empty(&mddev->disks))
1324                 MD_BUG();
1325         mddev->raid_disks = 0;
1326         mddev->major_version = 0;
1327 }
1328
1329 static void print_desc(mdp_disk_t *desc)
1330 {
1331         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1332                 desc->major,desc->minor,desc->raid_disk,desc->state);
1333 }
1334
1335 static void print_sb(mdp_super_t *sb)
1336 {
1337         int i;
1338
1339         printk(KERN_INFO
1340                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1341                 sb->major_version, sb->minor_version, sb->patch_version,
1342                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1343                 sb->ctime);
1344         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1345                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1346                 sb->md_minor, sb->layout, sb->chunk_size);
1347         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1348                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1349                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1350                 sb->failed_disks, sb->spare_disks,
1351                 sb->sb_csum, (unsigned long)sb->events_lo);
1352
1353         printk(KERN_INFO);
1354         for (i = 0; i < MD_SB_DISKS; i++) {
1355                 mdp_disk_t *desc;
1356
1357                 desc = sb->disks + i;
1358                 if (desc->number || desc->major || desc->minor ||
1359                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1360                         printk("     D %2d: ", i);
1361                         print_desc(desc);
1362                 }
1363         }
1364         printk(KERN_INFO "md:     THIS: ");
1365         print_desc(&sb->this_disk);
1366
1367 }
1368
1369 static void print_rdev(mdk_rdev_t *rdev)
1370 {
1371         char b[BDEVNAME_SIZE];
1372         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1373                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1374                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1375                 rdev->desc_nr);
1376         if (rdev->sb_loaded) {
1377                 printk(KERN_INFO "md: rdev superblock:\n");
1378                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1379         } else
1380                 printk(KERN_INFO "md: no rdev superblock!\n");
1381 }
1382
1383 void md_print_devices(void)
1384 {
1385         struct list_head *tmp, *tmp2;
1386         mdk_rdev_t *rdev;
1387         mddev_t *mddev;
1388         char b[BDEVNAME_SIZE];
1389
1390         printk("\n");
1391         printk("md:     **********************************\n");
1392         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1393         printk("md:     **********************************\n");
1394         ITERATE_MDDEV(mddev,tmp) {
1395
1396                 if (mddev->bitmap)
1397                         bitmap_print_sb(mddev->bitmap);
1398                 else
1399                         printk("%s: ", mdname(mddev));
1400                 ITERATE_RDEV(mddev,rdev,tmp2)
1401                         printk("<%s>", bdevname(rdev->bdev,b));
1402                 printk("\n");
1403
1404                 ITERATE_RDEV(mddev,rdev,tmp2)
1405                         print_rdev(rdev);
1406         }
1407         printk("md:     **********************************\n");
1408         printk("\n");
1409 }
1410
1411
1412 static void sync_sbs(mddev_t * mddev)
1413 {
1414         mdk_rdev_t *rdev;
1415         struct list_head *tmp;
1416
1417         ITERATE_RDEV(mddev,rdev,tmp) {
1418                 super_types[mddev->major_version].
1419                         sync_super(mddev, rdev);
1420                 rdev->sb_loaded = 1;
1421         }
1422 }
1423
1424 static void md_update_sb(mddev_t * mddev)
1425 {
1426         int err;
1427         struct list_head *tmp;
1428         mdk_rdev_t *rdev;
1429         int sync_req;
1430
1431 repeat:
1432         spin_lock_irq(&mddev->write_lock);
1433         sync_req = mddev->in_sync;
1434         mddev->utime = get_seconds();
1435         mddev->events ++;
1436
1437         if (!mddev->events) {
1438                 /*
1439                  * oops, this 64-bit counter should never wrap.
1440                  * Either we are in around ~1 trillion A.C., assuming
1441                  * 1 reboot per second, or we have a bug:
1442                  */
1443                 MD_BUG();
1444                 mddev->events --;
1445         }
1446         mddev->sb_dirty = 2;
1447         sync_sbs(mddev);
1448
1449         /*
1450          * do not write anything to disk if using
1451          * nonpersistent superblocks
1452          */
1453         if (!mddev->persistent) {
1454                 mddev->sb_dirty = 0;
1455                 spin_unlock_irq(&mddev->write_lock);
1456                 wake_up(&mddev->sb_wait);
1457                 return;
1458         }
1459         spin_unlock_irq(&mddev->write_lock);
1460
1461         dprintk(KERN_INFO
1462                 "md: updating %s RAID superblock on device (in sync %d)\n",
1463                 mdname(mddev),mddev->in_sync);
1464
1465         err = bitmap_update_sb(mddev->bitmap);
1466         ITERATE_RDEV(mddev,rdev,tmp) {
1467                 char b[BDEVNAME_SIZE];
1468                 dprintk(KERN_INFO "md: ");
1469                 if (test_bit(Faulty, &rdev->flags))
1470                         dprintk("(skipping faulty ");
1471
1472                 dprintk("%s ", bdevname(rdev->bdev,b));
1473                 if (!test_bit(Faulty, &rdev->flags)) {
1474                         md_super_write(mddev,rdev,
1475                                        rdev->sb_offset<<1, rdev->sb_size,
1476                                        rdev->sb_page);
1477                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1478                                 bdevname(rdev->bdev,b),
1479                                 (unsigned long long)rdev->sb_offset);
1480
1481                 } else
1482                         dprintk(")\n");
1483                 if (mddev->level == LEVEL_MULTIPATH)
1484                         /* only need to write one superblock... */
1485                         break;
1486         }
1487         md_super_wait(mddev);
1488         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1489
1490         spin_lock_irq(&mddev->write_lock);
1491         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1492                 /* have to write it out again */
1493                 spin_unlock_irq(&mddev->write_lock);
1494                 goto repeat;
1495         }
1496         mddev->sb_dirty = 0;
1497         spin_unlock_irq(&mddev->write_lock);
1498         wake_up(&mddev->sb_wait);
1499
1500 }
1501
1502 struct rdev_sysfs_entry {
1503         struct attribute attr;
1504         ssize_t (*show)(mdk_rdev_t *, char *);
1505         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1506 };
1507
1508 static ssize_t
1509 state_show(mdk_rdev_t *rdev, char *page)
1510 {
1511         char *sep = "";
1512         int len=0;
1513
1514         if (test_bit(Faulty, &rdev->flags)) {
1515                 len+= sprintf(page+len, "%sfaulty",sep);
1516                 sep = ",";
1517         }
1518         if (test_bit(In_sync, &rdev->flags)) {
1519                 len += sprintf(page+len, "%sin_sync",sep);
1520                 sep = ",";
1521         }
1522         if (!test_bit(Faulty, &rdev->flags) &&
1523             !test_bit(In_sync, &rdev->flags)) {
1524                 len += sprintf(page+len, "%sspare", sep);
1525                 sep = ",";
1526         }
1527         return len+sprintf(page+len, "\n");
1528 }
1529
1530 static struct rdev_sysfs_entry
1531 rdev_state = __ATTR_RO(state);
1532
1533 static ssize_t
1534 super_show(mdk_rdev_t *rdev, char *page)
1535 {
1536         if (rdev->sb_loaded && rdev->sb_size) {
1537                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1538                 return rdev->sb_size;
1539         } else
1540                 return 0;
1541 }
1542 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1543
1544 static struct attribute *rdev_default_attrs[] = {
1545         &rdev_state.attr,
1546         &rdev_super.attr,
1547         NULL,
1548 };
1549 static ssize_t
1550 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1551 {
1552         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1553         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1554
1555         if (!entry->show)
1556                 return -EIO;
1557         return entry->show(rdev, page);
1558 }
1559
1560 static ssize_t
1561 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1562               const char *page, size_t length)
1563 {
1564         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1565         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1566
1567         if (!entry->store)
1568                 return -EIO;
1569         return entry->store(rdev, page, length);
1570 }
1571
1572 static void rdev_free(struct kobject *ko)
1573 {
1574         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1575         kfree(rdev);
1576 }
1577 static struct sysfs_ops rdev_sysfs_ops = {
1578         .show           = rdev_attr_show,
1579         .store          = rdev_attr_store,
1580 };
1581 static struct kobj_type rdev_ktype = {
1582         .release        = rdev_free,
1583         .sysfs_ops      = &rdev_sysfs_ops,
1584         .default_attrs  = rdev_default_attrs,
1585 };
1586
1587 /*
1588  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1589  *
1590  * mark the device faulty if:
1591  *
1592  *   - the device is nonexistent (zero size)
1593  *   - the device has no valid superblock
1594  *
1595  * a faulty rdev _never_ has rdev->sb set.
1596  */
1597 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1598 {
1599         char b[BDEVNAME_SIZE];
1600         int err;
1601         mdk_rdev_t *rdev;
1602         sector_t size;
1603
1604         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1605         if (!rdev) {
1606                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1607                 return ERR_PTR(-ENOMEM);
1608         }
1609         memset(rdev, 0, sizeof(*rdev));
1610
1611         if ((err = alloc_disk_sb(rdev)))
1612                 goto abort_free;
1613
1614         err = lock_rdev(rdev, newdev);
1615         if (err)
1616                 goto abort_free;
1617
1618         rdev->kobj.parent = NULL;
1619         rdev->kobj.ktype = &rdev_ktype;
1620         kobject_init(&rdev->kobj);
1621
1622         rdev->desc_nr = -1;
1623         rdev->flags = 0;
1624         rdev->data_offset = 0;
1625         atomic_set(&rdev->nr_pending, 0);
1626         atomic_set(&rdev->read_errors, 0);
1627
1628         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1629         if (!size) {
1630                 printk(KERN_WARNING
1631                         "md: %s has zero or unknown size, marking faulty!\n",
1632                         bdevname(rdev->bdev,b));
1633                 err = -EINVAL;
1634                 goto abort_free;
1635         }
1636
1637         if (super_format >= 0) {
1638                 err = super_types[super_format].
1639                         load_super(rdev, NULL, super_minor);
1640                 if (err == -EINVAL) {
1641                         printk(KERN_WARNING
1642                                 "md: %s has invalid sb, not importing!\n",
1643                                 bdevname(rdev->bdev,b));
1644                         goto abort_free;
1645                 }
1646                 if (err < 0) {
1647                         printk(KERN_WARNING
1648                                 "md: could not read %s's sb, not importing!\n",
1649                                 bdevname(rdev->bdev,b));
1650                         goto abort_free;
1651                 }
1652         }
1653         INIT_LIST_HEAD(&rdev->same_set);
1654
1655         return rdev;
1656
1657 abort_free:
1658         if (rdev->sb_page) {
1659                 if (rdev->bdev)
1660                         unlock_rdev(rdev);
1661                 free_disk_sb(rdev);
1662         }
1663         kfree(rdev);
1664         return ERR_PTR(err);
1665 }
1666
1667 /*
1668  * Check a full RAID array for plausibility
1669  */
1670
1671
1672 static void analyze_sbs(mddev_t * mddev)
1673 {
1674         int i;
1675         struct list_head *tmp;
1676         mdk_rdev_t *rdev, *freshest;
1677         char b[BDEVNAME_SIZE];
1678
1679         freshest = NULL;
1680         ITERATE_RDEV(mddev,rdev,tmp)
1681                 switch (super_types[mddev->major_version].
1682                         load_super(rdev, freshest, mddev->minor_version)) {
1683                 case 1:
1684                         freshest = rdev;
1685                         break;
1686                 case 0:
1687                         break;
1688                 default:
1689                         printk( KERN_ERR \
1690                                 "md: fatal superblock inconsistency in %s"
1691                                 " -- removing from array\n",
1692                                 bdevname(rdev->bdev,b));
1693                         kick_rdev_from_array(rdev);
1694                 }
1695
1696
1697         super_types[mddev->major_version].
1698                 validate_super(mddev, freshest);
1699
1700         i = 0;
1701         ITERATE_RDEV(mddev,rdev,tmp) {
1702                 if (rdev != freshest)
1703                         if (super_types[mddev->major_version].
1704                             validate_super(mddev, rdev)) {
1705                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1706                                         " from array!\n",
1707                                         bdevname(rdev->bdev,b));
1708                                 kick_rdev_from_array(rdev);
1709                                 continue;
1710                         }
1711                 if (mddev->level == LEVEL_MULTIPATH) {
1712                         rdev->desc_nr = i++;
1713                         rdev->raid_disk = rdev->desc_nr;
1714                         set_bit(In_sync, &rdev->flags);
1715                 }
1716         }
1717
1718
1719
1720         if (mddev->recovery_cp != MaxSector &&
1721             mddev->level >= 1)
1722                 printk(KERN_ERR "md: %s: raid array is not clean"
1723                        " -- starting background reconstruction\n",
1724                        mdname(mddev));
1725
1726 }
1727
1728 static ssize_t
1729 level_show(mddev_t *mddev, char *page)
1730 {
1731         mdk_personality_t *p = mddev->pers;
1732         if (p == NULL && mddev->raid_disks == 0)
1733                 return 0;
1734         if (mddev->level >= 0)
1735                 return sprintf(page, "raid%d\n", mddev->level);
1736         else
1737                 return sprintf(page, "%s\n", p->name);
1738 }
1739
1740 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1741
1742 static ssize_t
1743 raid_disks_show(mddev_t *mddev, char *page)
1744 {
1745         if (mddev->raid_disks == 0)
1746                 return 0;
1747         return sprintf(page, "%d\n", mddev->raid_disks);
1748 }
1749
1750 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1751
1752 static ssize_t
1753 action_show(mddev_t *mddev, char *page)
1754 {
1755         char *type = "idle";
1756         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1757             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1758                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1759                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1760                                 type = "resync";
1761                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1762                                 type = "check";
1763                         else
1764                                 type = "repair";
1765                 } else
1766                         type = "recover";
1767         }
1768         return sprintf(page, "%s\n", type);
1769 }
1770
1771 static ssize_t
1772 action_store(mddev_t *mddev, const char *page, size_t len)
1773 {
1774         if (!mddev->pers || !mddev->pers->sync_request)
1775                 return -EINVAL;
1776
1777         if (strcmp(page, "idle")==0 || strcmp(page, "idle\n")==0) {
1778                 if (mddev->sync_thread) {
1779                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1780                         md_unregister_thread(mddev->sync_thread);
1781                         mddev->sync_thread = NULL;
1782                         mddev->recovery = 0;
1783                 }
1784                 return len;
1785         }
1786
1787         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1788             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1789                 return -EBUSY;
1790         if (strcmp(page, "resync")==0 || strcmp(page, "resync\n")==0 ||
1791             strcmp(page, "recover")==0 || strcmp(page, "recover\n")==0)
1792                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1793         else {
1794                 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1795                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1796                 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1797                         return -EINVAL;
1798                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1799                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1800                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1801         }
1802         md_wakeup_thread(mddev->thread);
1803         return len;
1804 }
1805
1806 static ssize_t
1807 mismatch_cnt_show(mddev_t *mddev, char *page)
1808 {
1809         return sprintf(page, "%llu\n",
1810                        (unsigned long long) mddev->resync_mismatches);
1811 }
1812
1813 static struct md_sysfs_entry
1814 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1815
1816
1817 static struct md_sysfs_entry
1818 md_mismatches = __ATTR_RO(mismatch_cnt);
1819
1820 static struct attribute *md_default_attrs[] = {
1821         &md_level.attr,
1822         &md_raid_disks.attr,
1823         NULL,
1824 };
1825
1826 static struct attribute *md_redundancy_attrs[] = {
1827         &md_scan_mode.attr,
1828         &md_mismatches.attr,
1829         NULL,
1830 };
1831 static struct attribute_group md_redundancy_group = {
1832         .name = NULL,
1833         .attrs = md_redundancy_attrs,
1834 };
1835
1836
1837 static ssize_t
1838 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1839 {
1840         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1841         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1842         ssize_t rv;
1843
1844         if (!entry->show)
1845                 return -EIO;
1846         mddev_lock(mddev);
1847         rv = entry->show(mddev, page);
1848         mddev_unlock(mddev);
1849         return rv;
1850 }
1851
1852 static ssize_t
1853 md_attr_store(struct kobject *kobj, struct attribute *attr,
1854               const char *page, size_t length)
1855 {
1856         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1857         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1858         ssize_t rv;
1859
1860         if (!entry->store)
1861                 return -EIO;
1862         mddev_lock(mddev);
1863         rv = entry->store(mddev, page, length);
1864         mddev_unlock(mddev);
1865         return rv;
1866 }
1867
1868 static void md_free(struct kobject *ko)
1869 {
1870         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1871         kfree(mddev);
1872 }
1873
1874 static struct sysfs_ops md_sysfs_ops = {
1875         .show   = md_attr_show,
1876         .store  = md_attr_store,
1877 };
1878 static struct kobj_type md_ktype = {
1879         .release        = md_free,
1880         .sysfs_ops      = &md_sysfs_ops,
1881         .default_attrs  = md_default_attrs,
1882 };
1883
1884 int mdp_major = 0;
1885
1886 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1887 {
1888         static DECLARE_MUTEX(disks_sem);
1889         mddev_t *mddev = mddev_find(dev);
1890         struct gendisk *disk;
1891         int partitioned = (MAJOR(dev) != MD_MAJOR);
1892         int shift = partitioned ? MdpMinorShift : 0;
1893         int unit = MINOR(dev) >> shift;
1894
1895         if (!mddev)
1896                 return NULL;
1897
1898         down(&disks_sem);
1899         if (mddev->gendisk) {
1900                 up(&disks_sem);
1901                 mddev_put(mddev);
1902                 return NULL;
1903         }
1904         disk = alloc_disk(1 << shift);
1905         if (!disk) {
1906                 up(&disks_sem);
1907                 mddev_put(mddev);
1908                 return NULL;
1909         }
1910         disk->major = MAJOR(dev);
1911         disk->first_minor = unit << shift;
1912         if (partitioned) {
1913                 sprintf(disk->disk_name, "md_d%d", unit);
1914                 sprintf(disk->devfs_name, "md/d%d", unit);
1915         } else {
1916                 sprintf(disk->disk_name, "md%d", unit);
1917                 sprintf(disk->devfs_name, "md/%d", unit);
1918         }
1919         disk->fops = &md_fops;
1920         disk->private_data = mddev;
1921         disk->queue = mddev->queue;
1922         add_disk(disk);
1923         mddev->gendisk = disk;
1924         up(&disks_sem);
1925         mddev->kobj.parent = &disk->kobj;
1926         mddev->kobj.k_name = NULL;
1927         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1928         mddev->kobj.ktype = &md_ktype;
1929         kobject_register(&mddev->kobj);
1930         return NULL;
1931 }
1932
1933 void md_wakeup_thread(mdk_thread_t *thread);
1934
1935 static void md_safemode_timeout(unsigned long data)
1936 {
1937         mddev_t *mddev = (mddev_t *) data;
1938
1939         mddev->safemode = 1;
1940         md_wakeup_thread(mddev->thread);
1941 }
1942
1943 static int start_dirty_degraded;
1944
1945 static int do_md_run(mddev_t * mddev)
1946 {
1947         int pnum, err;
1948         int chunk_size;
1949         struct list_head *tmp;
1950         mdk_rdev_t *rdev;
1951         struct gendisk *disk;
1952         char b[BDEVNAME_SIZE];
1953
1954         if (list_empty(&mddev->disks))
1955                 /* cannot run an array with no devices.. */
1956                 return -EINVAL;
1957
1958         if (mddev->pers)
1959                 return -EBUSY;
1960
1961         /*
1962          * Analyze all RAID superblock(s)
1963          */
1964         if (!mddev->raid_disks)
1965                 analyze_sbs(mddev);
1966
1967         chunk_size = mddev->chunk_size;
1968         pnum = level_to_pers(mddev->level);
1969
1970         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1971                 if (!chunk_size) {
1972                         /*
1973                          * 'default chunksize' in the old md code used to
1974                          * be PAGE_SIZE, baaad.
1975                          * we abort here to be on the safe side. We don't
1976                          * want to continue the bad practice.
1977                          */
1978                         printk(KERN_ERR
1979                                 "no chunksize specified, see 'man raidtab'\n");
1980                         return -EINVAL;
1981                 }
1982                 if (chunk_size > MAX_CHUNK_SIZE) {
1983                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1984                                 chunk_size, MAX_CHUNK_SIZE);
1985                         return -EINVAL;
1986                 }
1987                 /*
1988                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1989                  */
1990                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1991                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1992                         return -EINVAL;
1993                 }
1994                 if (chunk_size < PAGE_SIZE) {
1995                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1996                                 chunk_size, PAGE_SIZE);
1997                         return -EINVAL;
1998                 }
1999
2000                 /* devices must have minimum size of one chunk */
2001                 ITERATE_RDEV(mddev,rdev,tmp) {
2002                         if (test_bit(Faulty, &rdev->flags))
2003                                 continue;
2004                         if (rdev->size < chunk_size / 1024) {
2005                                 printk(KERN_WARNING
2006                                         "md: Dev %s smaller than chunk_size:"
2007                                         " %lluk < %dk\n",
2008                                         bdevname(rdev->bdev,b),
2009                                         (unsigned long long)rdev->size,
2010                                         chunk_size / 1024);
2011                                 return -EINVAL;
2012                         }
2013                 }
2014         }
2015
2016 #ifdef CONFIG_KMOD
2017         if (!pers[pnum])
2018         {
2019                 request_module("md-personality-%d", pnum);
2020         }
2021 #endif
2022
2023         /*
2024          * Drop all container device buffers, from now on
2025          * the only valid external interface is through the md
2026          * device.
2027          * Also find largest hardsector size
2028          */
2029         ITERATE_RDEV(mddev,rdev,tmp) {
2030                 if (test_bit(Faulty, &rdev->flags))
2031                         continue;
2032                 sync_blockdev(rdev->bdev);
2033                 invalidate_bdev(rdev->bdev, 0);
2034         }
2035
2036         md_probe(mddev->unit, NULL, NULL);
2037         disk = mddev->gendisk;
2038         if (!disk)
2039                 return -ENOMEM;
2040
2041         spin_lock(&pers_lock);
2042         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2043                 spin_unlock(&pers_lock);
2044                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2045                        pnum);
2046                 return -EINVAL;
2047         }
2048
2049         mddev->pers = pers[pnum];
2050         spin_unlock(&pers_lock);
2051
2052         mddev->recovery = 0;
2053         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2054         mddev->barriers_work = 1;
2055         mddev->ok_start_degraded = start_dirty_degraded;
2056
2057         if (start_readonly)
2058                 mddev->ro = 2; /* read-only, but switch on first write */
2059
2060         err = mddev->pers->run(mddev);
2061         if (!err && mddev->pers->sync_request) {
2062                 err = bitmap_create(mddev);
2063                 if (err) {
2064                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2065                                mdname(mddev), err);
2066                         mddev->pers->stop(mddev);
2067                 }
2068         }
2069         if (err) {
2070                 printk(KERN_ERR "md: pers->run() failed ...\n");
2071                 module_put(mddev->pers->owner);
2072                 mddev->pers = NULL;
2073                 bitmap_destroy(mddev);
2074                 return err;
2075         }
2076         if (mddev->pers->sync_request)
2077                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2078         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2079                 mddev->ro = 0;
2080
2081         atomic_set(&mddev->writes_pending,0);
2082         mddev->safemode = 0;
2083         mddev->safemode_timer.function = md_safemode_timeout;
2084         mddev->safemode_timer.data = (unsigned long) mddev;
2085         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2086         mddev->in_sync = 1;
2087
2088         ITERATE_RDEV(mddev,rdev,tmp)
2089                 if (rdev->raid_disk >= 0) {
2090                         char nm[20];
2091                         sprintf(nm, "rd%d", rdev->raid_disk);
2092                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2093                 }
2094
2095         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2096         md_wakeup_thread(mddev->thread);
2097
2098         if (mddev->sb_dirty)
2099                 md_update_sb(mddev);
2100
2101         set_capacity(disk, mddev->array_size<<1);
2102
2103         /* If we call blk_queue_make_request here, it will
2104          * re-initialise max_sectors etc which may have been
2105          * refined inside -> run.  So just set the bits we need to set.
2106          * Most initialisation happended when we called
2107          * blk_queue_make_request(..., md_fail_request)
2108          * earlier.
2109          */
2110         mddev->queue->queuedata = mddev;
2111         mddev->queue->make_request_fn = mddev->pers->make_request;
2112
2113         mddev->changed = 1;
2114         return 0;
2115 }
2116
2117 static int restart_array(mddev_t *mddev)
2118 {
2119         struct gendisk *disk = mddev->gendisk;
2120         int err;
2121
2122         /*
2123          * Complain if it has no devices
2124          */
2125         err = -ENXIO;
2126         if (list_empty(&mddev->disks))
2127                 goto out;
2128
2129         if (mddev->pers) {
2130                 err = -EBUSY;
2131                 if (!mddev->ro)
2132                         goto out;
2133
2134                 mddev->safemode = 0;
2135                 mddev->ro = 0;
2136                 set_disk_ro(disk, 0);
2137
2138                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2139                         mdname(mddev));
2140                 /*
2141                  * Kick recovery or resync if necessary
2142                  */
2143                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2144                 md_wakeup_thread(mddev->thread);
2145                 err = 0;
2146         } else {
2147                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2148                         mdname(mddev));
2149                 err = -EINVAL;
2150         }
2151
2152 out:
2153         return err;
2154 }
2155
2156 static int do_md_stop(mddev_t * mddev, int ro)
2157 {
2158         int err = 0;
2159         struct gendisk *disk = mddev->gendisk;
2160
2161         if (mddev->pers) {
2162                 if (atomic_read(&mddev->active)>2) {
2163                         printk("md: %s still in use.\n",mdname(mddev));
2164                         return -EBUSY;
2165                 }
2166
2167                 if (mddev->sync_thread) {
2168                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2169                         md_unregister_thread(mddev->sync_thread);
2170                         mddev->sync_thread = NULL;
2171                 }
2172
2173                 del_timer_sync(&mddev->safemode_timer);
2174
2175                 invalidate_partition(disk, 0);
2176
2177                 if (ro) {
2178                         err  = -ENXIO;
2179                         if (mddev->ro==1)
2180                                 goto out;
2181                         mddev->ro = 1;
2182                 } else {
2183                         bitmap_flush(mddev);
2184                         md_super_wait(mddev);
2185                         if (mddev->ro)
2186                                 set_disk_ro(disk, 0);
2187                         blk_queue_make_request(mddev->queue, md_fail_request);
2188                         mddev->pers->stop(mddev);
2189                         if (mddev->pers->sync_request)
2190                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2191
2192                         module_put(mddev->pers->owner);
2193                         mddev->pers = NULL;
2194                         if (mddev->ro)
2195                                 mddev->ro = 0;
2196                 }
2197                 if (!mddev->in_sync) {
2198                         /* mark array as shutdown cleanly */
2199                         mddev->in_sync = 1;
2200                         md_update_sb(mddev);
2201                 }
2202                 if (ro)
2203                         set_disk_ro(disk, 1);
2204         }
2205
2206         bitmap_destroy(mddev);
2207         if (mddev->bitmap_file) {
2208                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2209                 fput(mddev->bitmap_file);
2210                 mddev->bitmap_file = NULL;
2211         }
2212         mddev->bitmap_offset = 0;
2213
2214         /*
2215          * Free resources if final stop
2216          */
2217         if (!ro) {
2218                 mdk_rdev_t *rdev;
2219                 struct list_head *tmp;
2220                 struct gendisk *disk;
2221                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2222
2223                 ITERATE_RDEV(mddev,rdev,tmp)
2224                         if (rdev->raid_disk >= 0) {
2225                                 char nm[20];
2226                                 sprintf(nm, "rd%d", rdev->raid_disk);
2227                                 sysfs_remove_link(&mddev->kobj, nm);
2228                         }
2229
2230                 export_array(mddev);
2231
2232                 mddev->array_size = 0;
2233                 disk = mddev->gendisk;
2234                 if (disk)
2235                         set_capacity(disk, 0);
2236                 mddev->changed = 1;
2237         } else
2238                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2239                         mdname(mddev));
2240         err = 0;
2241 out:
2242         return err;
2243 }
2244
2245 static void autorun_array(mddev_t *mddev)
2246 {
2247         mdk_rdev_t *rdev;
2248         struct list_head *tmp;
2249         int err;
2250
2251         if (list_empty(&mddev->disks))
2252                 return;
2253
2254         printk(KERN_INFO "md: running: ");
2255
2256         ITERATE_RDEV(mddev,rdev,tmp) {
2257                 char b[BDEVNAME_SIZE];
2258                 printk("<%s>", bdevname(rdev->bdev,b));
2259         }
2260         printk("\n");
2261
2262         err = do_md_run (mddev);
2263         if (err) {
2264                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2265                 do_md_stop (mddev, 0);
2266         }
2267 }
2268
2269 /*
2270  * lets try to run arrays based on all disks that have arrived
2271  * until now. (those are in pending_raid_disks)
2272  *
2273  * the method: pick the first pending disk, collect all disks with
2274  * the same UUID, remove all from the pending list and put them into
2275  * the 'same_array' list. Then order this list based on superblock
2276  * update time (freshest comes first), kick out 'old' disks and
2277  * compare superblocks. If everything's fine then run it.
2278  *
2279  * If "unit" is allocated, then bump its reference count
2280  */
2281 static void autorun_devices(int part)
2282 {
2283         struct list_head candidates;
2284         struct list_head *tmp;
2285         mdk_rdev_t *rdev0, *rdev;
2286         mddev_t *mddev;
2287         char b[BDEVNAME_SIZE];
2288
2289         printk(KERN_INFO "md: autorun ...\n");
2290         while (!list_empty(&pending_raid_disks)) {
2291                 dev_t dev;
2292                 rdev0 = list_entry(pending_raid_disks.next,
2293                                          mdk_rdev_t, same_set);
2294
2295                 printk(KERN_INFO "md: considering %s ...\n",
2296                         bdevname(rdev0->bdev,b));
2297                 INIT_LIST_HEAD(&candidates);
2298                 ITERATE_RDEV_PENDING(rdev,tmp)
2299                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2300                                 printk(KERN_INFO "md:  adding %s ...\n",
2301                                         bdevname(rdev->bdev,b));
2302                                 list_move(&rdev->same_set, &candidates);
2303                         }
2304                 /*
2305                  * now we have a set of devices, with all of them having
2306                  * mostly sane superblocks. It's time to allocate the
2307                  * mddev.
2308                  */
2309                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2310                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2311                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2312                         break;
2313                 }
2314                 if (part)
2315                         dev = MKDEV(mdp_major,
2316                                     rdev0->preferred_minor << MdpMinorShift);
2317                 else
2318                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2319
2320                 md_probe(dev, NULL, NULL);
2321                 mddev = mddev_find(dev);
2322                 if (!mddev) {
2323                         printk(KERN_ERR
2324                                 "md: cannot allocate memory for md drive.\n");
2325                         break;
2326                 }
2327                 if (mddev_lock(mddev))
2328                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2329                                mdname(mddev));
2330                 else if (mddev->raid_disks || mddev->major_version
2331                          || !list_empty(&mddev->disks)) {
2332                         printk(KERN_WARNING
2333                                 "md: %s already running, cannot run %s\n",
2334                                 mdname(mddev), bdevname(rdev0->bdev,b));
2335                         mddev_unlock(mddev);
2336                 } else {
2337                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2338                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2339                                 list_del_init(&rdev->same_set);
2340                                 if (bind_rdev_to_array(rdev, mddev))
2341                                         export_rdev(rdev);
2342                         }
2343                         autorun_array(mddev);
2344                         mddev_unlock(mddev);
2345                 }
2346                 /* on success, candidates will be empty, on error
2347                  * it won't...
2348                  */
2349                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2350                         export_rdev(rdev);
2351                 mddev_put(mddev);
2352         }
2353         printk(KERN_INFO "md: ... autorun DONE.\n");
2354 }
2355
2356 /*
2357  * import RAID devices based on one partition
2358  * if possible, the array gets run as well.
2359  */
2360
2361 static int autostart_array(dev_t startdev)
2362 {
2363         char b[BDEVNAME_SIZE];
2364         int err = -EINVAL, i;
2365         mdp_super_t *sb = NULL;
2366         mdk_rdev_t *start_rdev = NULL, *rdev;
2367
2368         start_rdev = md_import_device(startdev, 0, 0);
2369         if (IS_ERR(start_rdev))
2370                 return err;
2371
2372
2373         /* NOTE: this can only work for 0.90.0 superblocks */
2374         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2375         if (sb->major_version != 0 ||
2376             sb->minor_version != 90 ) {
2377                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2378                 export_rdev(start_rdev);
2379                 return err;
2380         }
2381
2382         if (test_bit(Faulty, &start_rdev->flags)) {
2383                 printk(KERN_WARNING
2384                         "md: can not autostart based on faulty %s!\n",
2385                         bdevname(start_rdev->bdev,b));
2386                 export_rdev(start_rdev);
2387                 return err;
2388         }
2389         list_add(&start_rdev->same_set, &pending_raid_disks);
2390
2391         for (i = 0; i < MD_SB_DISKS; i++) {
2392                 mdp_disk_t *desc = sb->disks + i;
2393                 dev_t dev = MKDEV(desc->major, desc->minor);
2394
2395                 if (!dev)
2396                         continue;
2397                 if (dev == startdev)
2398                         continue;
2399                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2400                         continue;
2401                 rdev = md_import_device(dev, 0, 0);
2402                 if (IS_ERR(rdev))
2403                         continue;
2404
2405                 list_add(&rdev->same_set, &pending_raid_disks);
2406         }
2407
2408         /*
2409          * possibly return codes
2410          */
2411         autorun_devices(0);
2412         return 0;
2413
2414 }
2415
2416
2417 static int get_version(void __user * arg)
2418 {
2419         mdu_version_t ver;
2420
2421         ver.major = MD_MAJOR_VERSION;
2422         ver.minor = MD_MINOR_VERSION;
2423         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2424
2425         if (copy_to_user(arg, &ver, sizeof(ver)))
2426                 return -EFAULT;
2427
2428         return 0;
2429 }
2430
2431 static int get_array_info(mddev_t * mddev, void __user * arg)
2432 {
2433         mdu_array_info_t info;
2434         int nr,working,active,failed,spare;
2435         mdk_rdev_t *rdev;
2436         struct list_head *tmp;
2437
2438         nr=working=active=failed=spare=0;
2439         ITERATE_RDEV(mddev,rdev,tmp) {
2440                 nr++;
2441                 if (test_bit(Faulty, &rdev->flags))
2442                         failed++;
2443                 else {
2444                         working++;
2445                         if (test_bit(In_sync, &rdev->flags))
2446                                 active++;
2447                         else
2448                                 spare++;
2449                 }
2450         }
2451
2452         info.major_version = mddev->major_version;
2453         info.minor_version = mddev->minor_version;
2454         info.patch_version = MD_PATCHLEVEL_VERSION;
2455         info.ctime         = mddev->ctime;
2456         info.level         = mddev->level;
2457         info.size          = mddev->size;
2458         info.nr_disks      = nr;
2459         info.raid_disks    = mddev->raid_disks;
2460         info.md_minor      = mddev->md_minor;
2461         info.not_persistent= !mddev->persistent;
2462
2463         info.utime         = mddev->utime;
2464         info.state         = 0;
2465         if (mddev->in_sync)
2466                 info.state = (1<<MD_SB_CLEAN);
2467         if (mddev->bitmap && mddev->bitmap_offset)
2468                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2469         info.active_disks  = active;
2470         info.working_disks = working;
2471         info.failed_disks  = failed;
2472         info.spare_disks   = spare;
2473
2474         info.layout        = mddev->layout;
2475         info.chunk_size    = mddev->chunk_size;
2476
2477         if (copy_to_user(arg, &info, sizeof(info)))
2478                 return -EFAULT;
2479
2480         return 0;
2481 }
2482
2483 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2484 {
2485         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2486         char *ptr, *buf = NULL;
2487         int err = -ENOMEM;
2488
2489         file = kmalloc(sizeof(*file), GFP_KERNEL);
2490         if (!file)
2491                 goto out;
2492
2493         /* bitmap disabled, zero the first byte and copy out */
2494         if (!mddev->bitmap || !mddev->bitmap->file) {
2495                 file->pathname[0] = '\0';
2496                 goto copy_out;
2497         }
2498
2499         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2500         if (!buf)
2501                 goto out;
2502
2503         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2504         if (!ptr)
2505                 goto out;
2506
2507         strcpy(file->pathname, ptr);
2508
2509 copy_out:
2510         err = 0;
2511         if (copy_to_user(arg, file, sizeof(*file)))
2512                 err = -EFAULT;
2513 out:
2514         kfree(buf);
2515         kfree(file);
2516         return err;
2517 }
2518
2519 static int get_disk_info(mddev_t * mddev, void __user * arg)
2520 {
2521         mdu_disk_info_t info;
2522         unsigned int nr;
2523         mdk_rdev_t *rdev;
2524
2525         if (copy_from_user(&info, arg, sizeof(info)))
2526                 return -EFAULT;
2527
2528         nr = info.number;
2529
2530         rdev = find_rdev_nr(mddev, nr);
2531         if (rdev) {
2532                 info.major = MAJOR(rdev->bdev->bd_dev);
2533                 info.minor = MINOR(rdev->bdev->bd_dev);
2534                 info.raid_disk = rdev->raid_disk;
2535                 info.state = 0;
2536                 if (test_bit(Faulty, &rdev->flags))
2537                         info.state |= (1<<MD_DISK_FAULTY);
2538                 else if (test_bit(In_sync, &rdev->flags)) {
2539                         info.state |= (1<<MD_DISK_ACTIVE);
2540                         info.state |= (1<<MD_DISK_SYNC);
2541                 }
2542                 if (test_bit(WriteMostly, &rdev->flags))
2543                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2544         } else {
2545                 info.major = info.minor = 0;
2546                 info.raid_disk = -1;
2547                 info.state = (1<<MD_DISK_REMOVED);
2548         }
2549
2550         if (copy_to_user(arg, &info, sizeof(info)))
2551                 return -EFAULT;
2552
2553         return 0;
2554 }
2555
2556 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2557 {
2558         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2559         mdk_rdev_t *rdev;
2560         dev_t dev = MKDEV(info->major,info->minor);
2561
2562         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2563                 return -EOVERFLOW;
2564
2565         if (!mddev->raid_disks) {
2566                 int err;
2567                 /* expecting a device which has a superblock */
2568                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2569                 if (IS_ERR(rdev)) {
2570                         printk(KERN_WARNING
2571                                 "md: md_import_device returned %ld\n",
2572                                 PTR_ERR(rdev));
2573                         return PTR_ERR(rdev);
2574                 }
2575                 if (!list_empty(&mddev->disks)) {
2576                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2577                                                         mdk_rdev_t, same_set);
2578                         int err = super_types[mddev->major_version]
2579                                 .load_super(rdev, rdev0, mddev->minor_version);
2580                         if (err < 0) {
2581                                 printk(KERN_WARNING
2582                                         "md: %s has different UUID to %s\n",
2583                                         bdevname(rdev->bdev,b),
2584                                         bdevname(rdev0->bdev,b2));
2585                                 export_rdev(rdev);
2586                                 return -EINVAL;
2587                         }
2588                 }
2589                 err = bind_rdev_to_array(rdev, mddev);
2590                 if (err)
2591                         export_rdev(rdev);
2592                 return err;
2593         }
2594
2595         /*
2596          * add_new_disk can be used once the array is assembled
2597          * to add "hot spares".  They must already have a superblock
2598          * written
2599          */
2600         if (mddev->pers) {
2601                 int err;
2602                 if (!mddev->pers->hot_add_disk) {
2603                         printk(KERN_WARNING
2604                                 "%s: personality does not support diskops!\n",
2605                                mdname(mddev));
2606                         return -EINVAL;
2607                 }
2608                 if (mddev->persistent)
2609                         rdev = md_import_device(dev, mddev->major_version,
2610                                                 mddev->minor_version);
2611                 else
2612                         rdev = md_import_device(dev, -1, -1);
2613                 if (IS_ERR(rdev)) {
2614                         printk(KERN_WARNING
2615                                 "md: md_import_device returned %ld\n",
2616                                 PTR_ERR(rdev));
2617                         return PTR_ERR(rdev);
2618                 }
2619                 /* set save_raid_disk if appropriate */
2620                 if (!mddev->persistent) {
2621                         if (info->state & (1<<MD_DISK_SYNC)  &&
2622                             info->raid_disk < mddev->raid_disks)
2623                                 rdev->raid_disk = info->raid_disk;
2624                         else
2625                                 rdev->raid_disk = -1;
2626                 } else
2627                         super_types[mddev->major_version].
2628                                 validate_super(mddev, rdev);
2629                 rdev->saved_raid_disk = rdev->raid_disk;
2630
2631                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2632                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2633                         set_bit(WriteMostly, &rdev->flags);
2634
2635                 rdev->raid_disk = -1;
2636                 err = bind_rdev_to_array(rdev, mddev);
2637                 if (err)
2638                         export_rdev(rdev);
2639
2640                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2641                 md_wakeup_thread(mddev->thread);
2642                 return err;
2643         }
2644
2645         /* otherwise, add_new_disk is only allowed
2646          * for major_version==0 superblocks
2647          */
2648         if (mddev->major_version != 0) {
2649                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2650                        mdname(mddev));
2651                 return -EINVAL;
2652         }
2653
2654         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2655                 int err;
2656                 rdev = md_import_device (dev, -1, 0);
2657                 if (IS_ERR(rdev)) {
2658                         printk(KERN_WARNING
2659                                 "md: error, md_import_device() returned %ld\n",
2660                                 PTR_ERR(rdev));
2661                         return PTR_ERR(rdev);
2662                 }
2663                 rdev->desc_nr = info->number;
2664                 if (info->raid_disk < mddev->raid_disks)
2665                         rdev->raid_disk = info->raid_disk;
2666                 else
2667                         rdev->raid_disk = -1;
2668
2669                 rdev->flags = 0;
2670
2671                 if (rdev->raid_disk < mddev->raid_disks)
2672                         if (info->state & (1<<MD_DISK_SYNC))
2673                                 set_bit(In_sync, &rdev->flags);
2674
2675                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2676                         set_bit(WriteMostly, &rdev->flags);
2677
2678                 err = bind_rdev_to_array(rdev, mddev);
2679                 if (err) {
2680                         export_rdev(rdev);
2681                         return err;
2682                 }
2683
2684                 if (!mddev->persistent) {
2685                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2686                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2687                 } else
2688                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2689                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2690
2691                 if (!mddev->size || (mddev->size > rdev->size))
2692                         mddev->size = rdev->size;
2693         }
2694
2695         return 0;
2696 }
2697
2698 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2699 {
2700         char b[BDEVNAME_SIZE];
2701         mdk_rdev_t *rdev;
2702
2703         if (!mddev->pers)
2704                 return -ENODEV;
2705
2706         rdev = find_rdev(mddev, dev);
2707         if (!rdev)
2708                 return -ENXIO;
2709
2710         if (rdev->raid_disk >= 0)
2711                 goto busy;
2712
2713         kick_rdev_from_array(rdev);
2714         md_update_sb(mddev);
2715
2716         return 0;
2717 busy:
2718         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2719                 bdevname(rdev->bdev,b), mdname(mddev));
2720         return -EBUSY;
2721 }
2722
2723 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2724 {
2725         char b[BDEVNAME_SIZE];
2726         int err;
2727         unsigned int size;
2728         mdk_rdev_t *rdev;
2729
2730         if (!mddev->pers)
2731                 return -ENODEV;
2732
2733         if (mddev->major_version != 0) {
2734                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2735                         " version-0 superblocks.\n",
2736                         mdname(mddev));
2737                 return -EINVAL;
2738         }
2739         if (!mddev->pers->hot_add_disk) {
2740                 printk(KERN_WARNING
2741                         "%s: personality does not support diskops!\n",
2742                         mdname(mddev));
2743                 return -EINVAL;
2744         }
2745
2746         rdev = md_import_device (dev, -1, 0);
2747         if (IS_ERR(rdev)) {
2748                 printk(KERN_WARNING
2749                         "md: error, md_import_device() returned %ld\n",
2750                         PTR_ERR(rdev));
2751                 return -EINVAL;
2752         }
2753
2754         if (mddev->persistent)
2755                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2756         else
2757                 rdev->sb_offset =
2758                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2759
2760         size = calc_dev_size(rdev, mddev->chunk_size);
2761         rdev->size = size;
2762
2763         if (size < mddev->size) {
2764                 printk(KERN_WARNING
2765                         "%s: disk size %llu blocks < array size %llu\n",
2766                         mdname(mddev), (unsigned long long)size,
2767                         (unsigned long long)mddev->size);
2768                 err = -ENOSPC;
2769                 goto abort_export;
2770         }
2771
2772         if (test_bit(Faulty, &rdev->flags)) {
2773                 printk(KERN_WARNING
2774                         "md: can not hot-add faulty %s disk to %s!\n",
2775                         bdevname(rdev->bdev,b), mdname(mddev));
2776                 err = -EINVAL;
2777                 goto abort_export;
2778         }
2779         clear_bit(In_sync, &rdev->flags);
2780         rdev->desc_nr = -1;
2781         bind_rdev_to_array(rdev, mddev);
2782
2783         /*
2784          * The rest should better be atomic, we can have disk failures
2785          * noticed in interrupt contexts ...
2786          */
2787
2788         if (rdev->desc_nr == mddev->max_disks) {
2789                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2790                         mdname(mddev));
2791                 err = -EBUSY;
2792                 goto abort_unbind_export;
2793         }
2794
2795         rdev->raid_disk = -1;
2796
2797         md_update_sb(mddev);
2798
2799         /*
2800          * Kick recovery, maybe this spare has to be added to the
2801          * array immediately.
2802          */
2803         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2804         md_wakeup_thread(mddev->thread);
2805
2806         return 0;
2807
2808 abort_unbind_export:
2809         unbind_rdev_from_array(rdev);
2810
2811 abort_export:
2812         export_rdev(rdev);
2813         return err;
2814 }
2815
2816 /* similar to deny_write_access, but accounts for our holding a reference
2817  * to the file ourselves */
2818 static int deny_bitmap_write_access(struct file * file)
2819 {
2820         struct inode *inode = file->f_mapping->host;
2821
2822         spin_lock(&inode->i_lock);
2823         if (atomic_read(&inode->i_writecount) > 1) {
2824                 spin_unlock(&inode->i_lock);
2825                 return -ETXTBSY;
2826         }
2827         atomic_set(&inode->i_writecount, -1);
2828         spin_unlock(&inode->i_lock);
2829
2830         return 0;
2831 }
2832
2833 static int set_bitmap_file(mddev_t *mddev, int fd)
2834 {
2835         int err;
2836
2837         if (mddev->pers) {
2838                 if (!mddev->pers->quiesce)
2839                         return -EBUSY;
2840                 if (mddev->recovery || mddev->sync_thread)
2841                         return -EBUSY;
2842                 /* we should be able to change the bitmap.. */
2843         }
2844
2845
2846         if (fd >= 0) {
2847                 if (mddev->bitmap)
2848                         return -EEXIST; /* cannot add when bitmap is present */
2849                 mddev->bitmap_file = fget(fd);
2850
2851                 if (mddev->bitmap_file == NULL) {
2852                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2853                                mdname(mddev));
2854                         return -EBADF;
2855                 }
2856
2857                 err = deny_bitmap_write_access(mddev->bitmap_file);
2858                 if (err) {
2859                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2860                                mdname(mddev));
2861                         fput(mddev->bitmap_file);
2862                         mddev->bitmap_file = NULL;
2863                         return err;
2864                 }
2865                 mddev->bitmap_offset = 0; /* file overrides offset */
2866         } else if (mddev->bitmap == NULL)
2867                 return -ENOENT; /* cannot remove what isn't there */
2868         err = 0;
2869         if (mddev->pers) {
2870                 mddev->pers->quiesce(mddev, 1);
2871                 if (fd >= 0)
2872                         err = bitmap_create(mddev);
2873                 if (fd < 0 || err)
2874                         bitmap_destroy(mddev);
2875                 mddev->pers->quiesce(mddev, 0);
2876         } else if (fd < 0) {
2877                 if (mddev->bitmap_file)
2878                         fput(mddev->bitmap_file);
2879                 mddev->bitmap_file = NULL;
2880         }
2881
2882         return err;
2883 }
2884
2885 /*
2886  * set_array_info is used two different ways
2887  * The original usage is when creating a new array.
2888  * In this usage, raid_disks is > 0 and it together with
2889  *  level, size, not_persistent,layout,chunksize determine the
2890  *  shape of the array.
2891  *  This will always create an array with a type-0.90.0 superblock.
2892  * The newer usage is when assembling an array.
2893  *  In this case raid_disks will be 0, and the major_version field is
2894  *  use to determine which style super-blocks are to be found on the devices.
2895  *  The minor and patch _version numbers are also kept incase the
2896  *  super_block handler wishes to interpret them.
2897  */
2898 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2899 {
2900
2901         if (info->raid_disks == 0) {
2902                 /* just setting version number for superblock loading */
2903                 if (info->major_version < 0 ||
2904                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2905                     super_types[info->major_version].name == NULL) {
2906                         /* maybe try to auto-load a module? */
2907                         printk(KERN_INFO
2908                                 "md: superblock version %d not known\n",
2909                                 info->major_version);
2910                         return -EINVAL;
2911                 }
2912                 mddev->major_version = info->major_version;
2913                 mddev->minor_version = info->minor_version;
2914                 mddev->patch_version = info->patch_version;
2915                 return 0;
2916         }
2917         mddev->major_version = MD_MAJOR_VERSION;
2918         mddev->minor_version = MD_MINOR_VERSION;
2919         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2920         mddev->ctime         = get_seconds();
2921
2922         mddev->level         = info->level;
2923         mddev->size          = info->size;
2924         mddev->raid_disks    = info->raid_disks;
2925         /* don't set md_minor, it is determined by which /dev/md* was
2926          * openned
2927          */
2928         if (info->state & (1<<MD_SB_CLEAN))
2929                 mddev->recovery_cp = MaxSector;
2930         else
2931                 mddev->recovery_cp = 0;
2932         mddev->persistent    = ! info->not_persistent;
2933
2934         mddev->layout        = info->layout;
2935         mddev->chunk_size    = info->chunk_size;
2936
2937         mddev->max_disks     = MD_SB_DISKS;
2938
2939         mddev->sb_dirty      = 1;
2940
2941         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2942         mddev->bitmap_offset = 0;
2943
2944         /*
2945          * Generate a 128 bit UUID
2946          */
2947         get_random_bytes(mddev->uuid, 16);
2948
2949         return 0;
2950 }
2951
2952 /*
2953  * update_array_info is used to change the configuration of an
2954  * on-line array.
2955  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2956  * fields in the info are checked against the array.
2957  * Any differences that cannot be handled will cause an error.
2958  * Normally, only one change can be managed at a time.
2959  */
2960 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2961 {
2962         int rv = 0;
2963         int cnt = 0;
2964         int state = 0;
2965
2966         /* calculate expected state,ignoring low bits */
2967         if (mddev->bitmap && mddev->bitmap_offset)
2968                 state |= (1 << MD_SB_BITMAP_PRESENT);
2969
2970         if (mddev->major_version != info->major_version ||
2971             mddev->minor_version != info->minor_version ||
2972 /*          mddev->patch_version != info->patch_version || */
2973             mddev->ctime         != info->ctime         ||
2974             mddev->level         != info->level         ||
2975 /*          mddev->layout        != info->layout        || */
2976             !mddev->persistent   != info->not_persistent||
2977             mddev->chunk_size    != info->chunk_size    ||
2978             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2979             ((state^info->state) & 0xfffffe00)
2980                 )
2981                 return -EINVAL;
2982         /* Check there is only one change */
2983         if (mddev->size != info->size) cnt++;
2984         if (mddev->raid_disks != info->raid_disks) cnt++;
2985         if (mddev->layout != info->layout) cnt++;
2986         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2987         if (cnt == 0) return 0;
2988         if (cnt > 1) return -EINVAL;
2989
2990         if (mddev->layout != info->layout) {
2991                 /* Change layout
2992                  * we don't need to do anything at the md level, the
2993                  * personality will take care of it all.
2994                  */
2995                 if (mddev->pers->reconfig == NULL)
2996                         return -EINVAL;
2997                 else
2998                         return mddev->pers->reconfig(mddev, info->layout, -1);
2999         }
3000         if (mddev->size != info->size) {
3001                 mdk_rdev_t * rdev;
3002                 struct list_head *tmp;
3003                 if (mddev->pers->resize == NULL)
3004                         return -EINVAL;
3005                 /* The "size" is the amount of each device that is used.
3006                  * This can only make sense for arrays with redundancy.
3007                  * linear and raid0 always use whatever space is available
3008                  * We can only consider changing the size if no resync
3009                  * or reconstruction is happening, and if the new size
3010                  * is acceptable. It must fit before the sb_offset or,
3011                  * if that is <data_offset, it must fit before the
3012                  * size of each device.
3013                  * If size is zero, we find the largest size that fits.
3014                  */
3015                 if (mddev->sync_thread)
3016                         return -EBUSY;
3017                 ITERATE_RDEV(mddev,rdev,tmp) {
3018                         sector_t avail;
3019                         int fit = (info->size == 0);
3020                         if (rdev->sb_offset > rdev->data_offset)
3021                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
3022                         else
3023                                 avail = get_capacity(rdev->bdev->bd_disk)
3024                                         - rdev->data_offset;
3025                         if (fit && (info->size == 0 || info->size > avail/2))
3026                                 info->size = avail/2;
3027                         if (avail < ((sector_t)info->size << 1))
3028                                 return -ENOSPC;
3029                 }
3030                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3031                 if (!rv) {
3032                         struct block_device *bdev;
3033
3034                         bdev = bdget_disk(mddev->gendisk, 0);
3035                         if (bdev) {
3036                                 down(&bdev->bd_inode->i_sem);
3037                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3038                                 up(&bdev->bd_inode->i_sem);
3039                                 bdput(bdev);
3040                         }
3041                 }
3042         }
3043         if (mddev->raid_disks    != info->raid_disks) {
3044                 /* change the number of raid disks */
3045                 if (mddev->pers->reshape == NULL)
3046                         return -EINVAL;
3047                 if (info->raid_disks <= 0 ||
3048                     info->raid_disks >= mddev->max_disks)
3049                         return -EINVAL;
3050                 if (mddev->sync_thread)
3051                         return -EBUSY;
3052                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3053                 if (!rv) {
3054                         struct block_device *bdev;
3055
3056                         bdev = bdget_disk(mddev->gendisk, 0);
3057                         if (bdev) {
3058                                 down(&bdev->bd_inode->i_sem);
3059                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3060                                 up(&bdev->bd_inode->i_sem);
3061                                 bdput(bdev);
3062                         }
3063                 }
3064         }
3065         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3066                 if (mddev->pers->quiesce == NULL)
3067                         return -EINVAL;
3068                 if (mddev->recovery || mddev->sync_thread)
3069                         return -EBUSY;
3070                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3071                         /* add the bitmap */
3072                         if (mddev->bitmap)
3073                                 return -EEXIST;
3074                         if (mddev->default_bitmap_offset == 0)
3075                                 return -EINVAL;
3076                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3077                         mddev->pers->quiesce(mddev, 1);
3078                         rv = bitmap_create(mddev);
3079                         if (rv)
3080                                 bitmap_destroy(mddev);
3081                         mddev->pers->quiesce(mddev, 0);
3082                 } else {
3083                         /* remove the bitmap */
3084                         if (!mddev->bitmap)
3085                                 return -ENOENT;
3086                         if (mddev->bitmap->file)
3087                                 return -EINVAL;
3088                         mddev->pers->quiesce(mddev, 1);
3089                         bitmap_destroy(mddev);
3090                         mddev->pers->quiesce(mddev, 0);
3091                         mddev->bitmap_offset = 0;
3092                 }
3093         }
3094         md_update_sb(mddev);
3095         return rv;
3096 }
3097
3098 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3099 {
3100         mdk_rdev_t *rdev;
3101
3102         if (mddev->pers == NULL)
3103                 return -ENODEV;
3104
3105         rdev = find_rdev(mddev, dev);
3106         if (!rdev)
3107                 return -ENODEV;
3108
3109         md_error(mddev, rdev);
3110         return 0;
3111 }
3112
3113 static int md_ioctl(struct inode *inode, struct file *file,
3114                         unsigned int cmd, unsigned long arg)
3115 {
3116         int err = 0;
3117         void __user *argp = (void __user *)arg;
3118         struct hd_geometry __user *loc = argp;
3119         mddev_t *mddev = NULL;
3120
3121         if (!capable(CAP_SYS_ADMIN))
3122                 return -EACCES;
3123
3124         /*
3125          * Commands dealing with the RAID driver but not any
3126          * particular array:
3127          */
3128         switch (cmd)
3129         {
3130                 case RAID_VERSION:
3131                         err = get_version(argp);
3132                         goto done;
3133
3134                 case PRINT_RAID_DEBUG:
3135                         err = 0;
3136                         md_print_devices();
3137                         goto done;
3138
3139 #ifndef MODULE
3140                 case RAID_AUTORUN:
3141                         err = 0;
3142                         autostart_arrays(arg);
3143                         goto done;
3144 #endif
3145                 default:;
3146         }
3147
3148         /*
3149          * Commands creating/starting a new array:
3150          */
3151
3152         mddev = inode->i_bdev->bd_disk->private_data;
3153
3154         if (!mddev) {
3155                 BUG();
3156                 goto abort;
3157         }
3158
3159
3160         if (cmd == START_ARRAY) {
3161                 /* START_ARRAY doesn't need to lock the array as autostart_array
3162                  * does the locking, and it could even be a different array
3163                  */
3164                 static int cnt = 3;
3165                 if (cnt > 0 ) {
3166                         printk(KERN_WARNING
3167                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3168                                "This will not be supported beyond July 2006\n",
3169                                current->comm, current->pid);
3170                         cnt--;
3171                 }
3172                 err = autostart_array(new_decode_dev(arg));
3173                 if (err) {
3174                         printk(KERN_WARNING "md: autostart failed!\n");
3175                         goto abort;
3176                 }
3177                 goto done;
3178         }
3179
3180         err = mddev_lock(mddev);
3181         if (err) {
3182                 printk(KERN_INFO
3183                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3184                         err, cmd);
3185                 goto abort;
3186         }
3187
3188         switch (cmd)
3189         {
3190                 case SET_ARRAY_INFO:
3191                         {
3192                                 mdu_array_info_t info;
3193                                 if (!arg)
3194                                         memset(&info, 0, sizeof(info));
3195                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3196                                         err = -EFAULT;
3197                                         goto abort_unlock;
3198                                 }
3199                                 if (mddev->pers) {
3200                                         err = update_array_info(mddev, &info);
3201                                         if (err) {
3202                                                 printk(KERN_WARNING "md: couldn't update"
3203                                                        " array info. %d\n", err);
3204                                                 goto abort_unlock;
3205                                         }
3206                                         goto done_unlock;
3207                                 }
3208                                 if (!list_empty(&mddev->disks)) {
3209                                         printk(KERN_WARNING
3210                                                "md: array %s already has disks!\n",
3211                                                mdname(mddev));
3212                                         err = -EBUSY;
3213                                         goto abort_unlock;
3214                                 }
3215                                 if (mddev->raid_disks) {
3216                                         printk(KERN_WARNING
3217                                                "md: array %s already initialised!\n",
3218                                                mdname(mddev));
3219                                         err = -EBUSY;
3220                                         goto abort_unlock;
3221                                 }
3222                                 err = set_array_info(mddev, &info);
3223                                 if (err) {
3224                                         printk(KERN_WARNING "md: couldn't set"
3225                                                " array info. %d\n", err);
3226                                         goto abort_unlock;
3227                                 }
3228                         }
3229                         goto done_unlock;
3230
3231                 default:;
3232         }
3233
3234         /*
3235          * Commands querying/configuring an existing array:
3236          */
3237         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3238          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3239         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3240                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3241                 err = -ENODEV;
3242                 goto abort_unlock;
3243         }
3244
3245         /*
3246          * Commands even a read-only array can execute:
3247          */
3248         switch (cmd)
3249         {
3250                 case GET_ARRAY_INFO:
3251                         err = get_array_info(mddev, argp);
3252                         goto done_unlock;
3253
3254                 case GET_BITMAP_FILE:
3255                         err = get_bitmap_file(mddev, argp);
3256                         goto done_unlock;
3257
3258                 case GET_DISK_INFO:
3259                         err = get_disk_info(mddev, argp);
3260                         goto done_unlock;
3261
3262                 case RESTART_ARRAY_RW:
3263                         err = restart_array(mddev);
3264                         goto done_unlock;
3265
3266                 case STOP_ARRAY:
3267                         err = do_md_stop (mddev, 0);
3268                         goto done_unlock;
3269
3270                 case STOP_ARRAY_RO:
3271                         err = do_md_stop (mddev, 1);
3272                         goto done_unlock;
3273
3274         /*
3275          * We have a problem here : there is no easy way to give a CHS
3276          * virtual geometry. We currently pretend that we have a 2 heads
3277          * 4 sectors (with a BIG number of cylinders...). This drives
3278          * dosfs just mad... ;-)
3279          */
3280                 case HDIO_GETGEO:
3281                         if (!loc) {
3282                                 err = -EINVAL;
3283                                 goto abort_unlock;
3284                         }
3285                         err = put_user (2, (char __user *) &loc->heads);
3286                         if (err)
3287                                 goto abort_unlock;
3288                         err = put_user (4, (char __user *) &loc->sectors);
3289                         if (err)
3290                                 goto abort_unlock;
3291                         err = put_user(get_capacity(mddev->gendisk)/8,
3292                                         (short __user *) &loc->cylinders);
3293                         if (err)
3294                                 goto abort_unlock;
3295                         err = put_user (get_start_sect(inode->i_bdev),
3296                                                 (long __user *) &loc->start);
3297                         goto done_unlock;
3298         }
3299
3300         /*
3301          * The remaining ioctls are changing the state of the
3302          * superblock, so we do not allow them on read-only arrays.
3303          * However non-MD ioctls (e.g. get-size) will still come through
3304          * here and hit the 'default' below, so only disallow
3305          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3306          */
3307         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3308             mddev->ro && mddev->pers) {
3309                 if (mddev->ro == 2) {
3310                         mddev->ro = 0;
3311                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3312                 md_wakeup_thread(mddev->thread);
3313
3314                 } else {
3315                         err = -EROFS;
3316                         goto abort_unlock;
3317                 }
3318         }
3319
3320         switch (cmd)
3321         {
3322                 case ADD_NEW_DISK:
3323                 {
3324                         mdu_disk_info_t info;
3325                         if (copy_from_user(&info, argp, sizeof(info)))
3326                                 err = -EFAULT;
3327                         else
3328                                 err = add_new_disk(mddev, &info);
3329                         goto done_unlock;
3330                 }
3331
3332                 case HOT_REMOVE_DISK:
3333                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3334                         goto done_unlock;
3335
3336                 case HOT_ADD_DISK:
3337                         err = hot_add_disk(mddev, new_decode_dev(arg));
3338                         goto done_unlock;
3339
3340                 case SET_DISK_FAULTY:
3341                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3342                         goto done_unlock;
3343
3344                 case RUN_ARRAY:
3345                         err = do_md_run (mddev);
3346                         goto done_unlock;
3347
3348                 case SET_BITMAP_FILE:
3349                         err = set_bitmap_file(mddev, (int)arg);
3350                         goto done_unlock;
3351
3352                 default:
3353                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3354                                 printk(KERN_WARNING "md: %s(pid %d) used"
3355                                         " obsolete MD ioctl, upgrade your"
3356                                         " software to use new ictls.\n",
3357                                         current->comm, current->pid);
3358                         err = -EINVAL;
3359                         goto abort_unlock;
3360         }
3361
3362 done_unlock:
3363 abort_unlock:
3364         mddev_unlock(mddev);
3365
3366         return err;
3367 done:
3368         if (err)
3369                 MD_BUG();
3370 abort:
3371         return err;
3372 }
3373
3374 static int md_open(struct inode *inode, struct file *file)
3375 {
3376         /*
3377          * Succeed if we can lock the mddev, which confirms that
3378          * it isn't being stopped right now.
3379          */
3380         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3381         int err;
3382
3383         if ((err = mddev_lock(mddev)))
3384                 goto out;
3385
3386         err = 0;
3387         mddev_get(mddev);
3388         mddev_unlock(mddev);
3389
3390         check_disk_change(inode->i_bdev);
3391  out:
3392         return err;
3393 }
3394
3395 static int md_release(struct inode *inode, struct file * file)
3396 {
3397         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3398
3399         if (!mddev)
3400                 BUG();
3401         mddev_put(mddev);
3402
3403         return 0;
3404 }
3405
3406 static int md_media_changed(struct gendisk *disk)
3407 {
3408         mddev_t *mddev = disk->private_data;
3409
3410         return mddev->changed;
3411 }
3412
3413 static int md_revalidate(struct gendisk *disk)
3414 {
3415         mddev_t *mddev = disk->private_data;
3416
3417         mddev->changed = 0;
3418         return 0;
3419 }
3420 static struct block_device_operations md_fops =
3421 {
3422         .owner          = THIS_MODULE,
3423         .open           = md_open,
3424         .release        = md_release,
3425         .ioctl          = md_ioctl,
3426         .media_changed  = md_media_changed,
3427         .revalidate_disk= md_revalidate,
3428 };
3429
3430 static int md_thread(void * arg)
3431 {
3432         mdk_thread_t *thread = arg;
3433
3434         /*
3435          * md_thread is a 'system-thread', it's priority should be very
3436          * high. We avoid resource deadlocks individually in each
3437          * raid personality. (RAID5 does preallocation) We also use RR and
3438          * the very same RT priority as kswapd, thus we will never get
3439          * into a priority inversion deadlock.
3440          *
3441          * we definitely have to have equal or higher priority than
3442          * bdflush, otherwise bdflush will deadlock if there are too
3443          * many dirty RAID5 blocks.
3444          */
3445
3446         allow_signal(SIGKILL);
3447         while (!kthread_should_stop()) {
3448
3449                 /* We need to wait INTERRUPTIBLE so that
3450                  * we don't add to the load-average.
3451                  * That means we need to be sure no signals are
3452                  * pending
3453                  */
3454                 if (signal_pending(current))
3455                         flush_signals(current);
3456
3457                 wait_event_interruptible_timeout
3458                         (thread->wqueue,
3459                          test_bit(THREAD_WAKEUP, &thread->flags)
3460                          || kthread_should_stop(),
3461                          thread->timeout);
3462                 try_to_freeze();
3463
3464                 clear_bit(THREAD_WAKEUP, &thread->flags);
3465
3466                 thread->run(thread->mddev);
3467         }
3468
3469         return 0;
3470 }
3471
3472 void md_wakeup_thread(mdk_thread_t *thread)
3473 {
3474         if (thread) {
3475                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3476                 set_bit(THREAD_WAKEUP, &thread->flags);
3477                 wake_up(&thread->wqueue);
3478         }
3479 }
3480
3481 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3482                                  const char *name)
3483 {
3484         mdk_thread_t *thread;
3485
3486         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3487         if (!thread)
3488                 return NULL;
3489
3490         memset(thread, 0, sizeof(mdk_thread_t));
3491         init_waitqueue_head(&thread->wqueue);
3492
3493         thread->run = run;
3494         thread->mddev = mddev;
3495         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3496         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3497         if (IS_ERR(thread->tsk)) {
3498                 kfree(thread);
3499                 return NULL;
3500         }
3501         return thread;
3502 }
3503
3504 void md_unregister_thread(mdk_thread_t *thread)
3505 {
3506         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3507
3508         kthread_stop(thread->tsk);
3509         kfree(thread);
3510 }
3511
3512 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3513 {
3514         if (!mddev) {
3515                 MD_BUG();
3516                 return;
3517         }
3518
3519         if (!rdev || test_bit(Faulty, &rdev->flags))
3520                 return;
3521 /*
3522         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3523                 mdname(mddev),
3524                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3525                 __builtin_return_address(0),__builtin_return_address(1),
3526                 __builtin_return_address(2),__builtin_return_address(3));
3527 */
3528         if (!mddev->pers->error_handler)
3529                 return;
3530         mddev->pers->error_handler(mddev,rdev);
3531         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3532         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3533         md_wakeup_thread(mddev->thread);
3534 }
3535
3536 /* seq_file implementation /proc/mdstat */
3537
3538 static void status_unused(struct seq_file *seq)
3539 {
3540         int i = 0;
3541         mdk_rdev_t *rdev;
3542         struct list_head *tmp;
3543
3544         seq_printf(seq, "unused devices: ");
3545
3546         ITERATE_RDEV_PENDING(rdev,tmp) {
3547                 char b[BDEVNAME_SIZE];
3548                 i++;
3549                 seq_printf(seq, "%s ",
3550                               bdevname(rdev->bdev,b));
3551         }
3552         if (!i)
3553                 seq_printf(seq, "<none>");
3554
3555         seq_printf(seq, "\n");
3556 }
3557
3558
3559 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3560 {
3561         unsigned long max_blocks, resync, res, dt, db, rt;
3562
3563         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3564
3565         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3566                 max_blocks = mddev->resync_max_sectors >> 1;
3567         else
3568                 max_blocks = mddev->size;
3569
3570         /*
3571          * Should not happen.
3572          */
3573         if (!max_blocks) {
3574                 MD_BUG();
3575                 return;
3576         }
3577         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3578         {
3579                 int i, x = res/50, y = 20-x;
3580                 seq_printf(seq, "[");
3581                 for (i = 0; i < x; i++)
3582                         seq_printf(seq, "=");
3583                 seq_printf(seq, ">");
3584                 for (i = 0; i < y; i++)
3585                         seq_printf(seq, ".");
3586                 seq_printf(seq, "] ");
3587         }
3588         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3589                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3590                        "resync" : "recovery"),
3591                       res/10, res % 10, resync, max_blocks);
3592
3593         /*
3594          * We do not want to overflow, so the order of operands and
3595          * the * 100 / 100 trick are important. We do a +1 to be
3596          * safe against division by zero. We only estimate anyway.
3597          *
3598          * dt: time from mark until now
3599          * db: blocks written from mark until now
3600          * rt: remaining time
3601          */
3602         dt = ((jiffies - mddev->resync_mark) / HZ);
3603         if (!dt) dt++;
3604         db = resync - (mddev->resync_mark_cnt/2);
3605         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3606
3607         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3608
3609         seq_printf(seq, " speed=%ldK/sec", db/dt);
3610 }
3611
3612 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3613 {
3614         struct list_head *tmp;
3615         loff_t l = *pos;
3616         mddev_t *mddev;
3617
3618         if (l >= 0x10000)
3619                 return NULL;
3620         if (!l--)
3621                 /* header */
3622                 return (void*)1;
3623
3624         spin_lock(&all_mddevs_lock);
3625         list_for_each(tmp,&all_mddevs)
3626                 if (!l--) {
3627                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3628                         mddev_get(mddev);
3629                         spin_unlock(&all_mddevs_lock);
3630                         return mddev;
3631                 }
3632         spin_unlock(&all_mddevs_lock);
3633         if (!l--)
3634                 return (void*)2;/* tail */
3635         return NULL;
3636 }
3637
3638 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3639 {
3640         struct list_head *tmp;
3641         mddev_t *next_mddev, *mddev = v;
3642
3643         ++*pos;
3644         if (v == (void*)2)
3645                 return NULL;
3646
3647         spin_lock(&all_mddevs_lock);
3648         if (v == (void*)1)
3649                 tmp = all_mddevs.next;
3650         else
3651                 tmp = mddev->all_mddevs.next;
3652         if (tmp != &all_mddevs)
3653                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3654         else {
3655                 next_mddev = (void*)2;
3656                 *pos = 0x10000;
3657         }
3658         spin_unlock(&all_mddevs_lock);
3659
3660         if (v != (void*)1)
3661                 mddev_put(mddev);
3662         return next_mddev;
3663
3664 }
3665
3666 static void md_seq_stop(struct seq_file *seq, void *v)
3667 {
3668         mddev_t *mddev = v;
3669
3670         if (mddev && v != (void*)1 && v != (void*)2)
3671                 mddev_put(mddev);
3672 }
3673
3674 static int md_seq_show(struct seq_file *seq, void *v)
3675 {
3676         mddev_t *mddev = v;
3677         sector_t size;
3678         struct list_head *tmp2;
3679         mdk_rdev_t *rdev;
3680         int i;
3681         struct bitmap *bitmap;
3682
3683         if (v == (void*)1) {
3684                 seq_printf(seq, "Personalities : ");
3685                 spin_lock(&pers_lock);
3686                 for (i = 0; i < MAX_PERSONALITY; i++)
3687                         if (pers[i])
3688                                 seq_printf(seq, "[%s] ", pers[i]->name);
3689
3690                 spin_unlock(&pers_lock);
3691                 seq_printf(seq, "\n");
3692                 return 0;
3693         }
3694         if (v == (void*)2) {
3695                 status_unused(seq);
3696                 return 0;
3697         }
3698
3699         if (mddev_lock(mddev)!=0)
3700                 return -EINTR;
3701         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3702                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3703                                                 mddev->pers ? "" : "in");
3704                 if (mddev->pers) {
3705                         if (mddev->ro==1)
3706                                 seq_printf(seq, " (read-only)");
3707                         if (mddev->ro==2)
3708                                 seq_printf(seq, "(auto-read-only)");
3709                         seq_printf(seq, " %s", mddev->pers->name);
3710                 }
3711
3712                 size = 0;
3713                 ITERATE_RDEV(mddev,rdev,tmp2) {
3714                         char b[BDEVNAME_SIZE];
3715                         seq_printf(seq, " %s[%d]",
3716                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3717                         if (test_bit(WriteMostly, &rdev->flags))
3718                                 seq_printf(seq, "(W)");
3719                         if (test_bit(Faulty, &rdev->flags)) {
3720                                 seq_printf(seq, "(F)");
3721                                 continue;
3722                         } else if (rdev->raid_disk < 0)
3723                                 seq_printf(seq, "(S)"); /* spare */
3724                         size += rdev->size;
3725                 }
3726
3727                 if (!list_empty(&mddev->disks)) {
3728                         if (mddev->pers)
3729                                 seq_printf(seq, "\n      %llu blocks",
3730                                         (unsigned long long)mddev->array_size);
3731                         else
3732                                 seq_printf(seq, "\n      %llu blocks",
3733                                         (unsigned long long)size);
3734                 }
3735                 if (mddev->persistent) {
3736                         if (mddev->major_version != 0 ||
3737                             mddev->minor_version != 90) {
3738                                 seq_printf(seq," super %d.%d",
3739                                            mddev->major_version,
3740                                            mddev->minor_version);
3741                         }
3742                 } else
3743                         seq_printf(seq, " super non-persistent");
3744
3745                 if (mddev->pers) {
3746                         mddev->pers->status (seq, mddev);
3747                         seq_printf(seq, "\n      ");
3748                         if (mddev->pers->sync_request) {
3749                                 if (mddev->curr_resync > 2) {
3750                                         status_resync (seq, mddev);
3751                                         seq_printf(seq, "\n      ");
3752                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3753                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3754                                 else if (mddev->recovery_cp < MaxSector)
3755                                         seq_printf(seq, "\tresync=PENDING\n      ");
3756                         }
3757                 } else
3758                         seq_printf(seq, "\n       ");
3759
3760                 if ((bitmap = mddev->bitmap)) {
3761                         unsigned long chunk_kb;
3762                         unsigned long flags;
3763                         spin_lock_irqsave(&bitmap->lock, flags);
3764                         chunk_kb = bitmap->chunksize >> 10;
3765                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3766                                 "%lu%s chunk",
3767                                 bitmap->pages - bitmap->missing_pages,
3768                                 bitmap->pages,
3769                                 (bitmap->pages - bitmap->missing_pages)
3770                                         << (PAGE_SHIFT - 10),
3771                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3772                                 chunk_kb ? "KB" : "B");
3773                         if (bitmap->file) {
3774                                 seq_printf(seq, ", file: ");
3775                                 seq_path(seq, bitmap->file->f_vfsmnt,
3776                                          bitmap->file->f_dentry," \t\n");
3777                         }
3778
3779                         seq_printf(seq, "\n");
3780                         spin_unlock_irqrestore(&bitmap->lock, flags);
3781                 }
3782
3783                 seq_printf(seq, "\n");
3784         }
3785         mddev_unlock(mddev);
3786
3787         return 0;
3788 }
3789
3790 static struct seq_operations md_seq_ops = {
3791         .start  = md_seq_start,
3792         .next   = md_seq_next,
3793         .stop   = md_seq_stop,
3794         .show   = md_seq_show,
3795 };
3796
3797 static int md_seq_open(struct inode *inode, struct file *file)
3798 {
3799         int error;
3800
3801         error = seq_open(file, &md_seq_ops);
3802         return error;
3803 }
3804
3805 static struct file_operations md_seq_fops = {
3806         .open           = md_seq_open,
3807         .read           = seq_read,
3808         .llseek         = seq_lseek,
3809         .release        = seq_release,
3810 };
3811
3812 int register_md_personality(int pnum, mdk_personality_t *p)
3813 {
3814         if (pnum >= MAX_PERSONALITY) {
3815                 printk(KERN_ERR
3816                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3817                        p->name, pnum, MAX_PERSONALITY-1);
3818                 return -EINVAL;
3819         }
3820
3821         spin_lock(&pers_lock);
3822         if (pers[pnum]) {
3823                 spin_unlock(&pers_lock);
3824                 return -EBUSY;
3825         }
3826
3827         pers[pnum] = p;
3828         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3829         spin_unlock(&pers_lock);
3830         return 0;
3831 }
3832
3833 int unregister_md_personality(int pnum)
3834 {
3835         if (pnum >= MAX_PERSONALITY)
3836                 return -EINVAL;
3837
3838         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3839         spin_lock(&pers_lock);
3840         pers[pnum] = NULL;
3841         spin_unlock(&pers_lock);
3842         return 0;
3843 }
3844
3845 static int is_mddev_idle(mddev_t *mddev)
3846 {
3847         mdk_rdev_t * rdev;
3848         struct list_head *tmp;
3849         int idle;
3850         unsigned long curr_events;
3851
3852         idle = 1;
3853         ITERATE_RDEV(mddev,rdev,tmp) {
3854                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3855                 curr_events = disk_stat_read(disk, sectors[0]) +
3856                                 disk_stat_read(disk, sectors[1]) -
3857                                 atomic_read(&disk->sync_io);
3858                 /* The difference between curr_events and last_events
3859                  * will be affected by any new non-sync IO (making
3860                  * curr_events bigger) and any difference in the amount of
3861                  * in-flight syncio (making current_events bigger or smaller)
3862                  * The amount in-flight is currently limited to
3863                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3864                  * which is at most 4096 sectors.
3865                  * These numbers are fairly fragile and should be made
3866                  * more robust, probably by enforcing the
3867                  * 'window size' that md_do_sync sort-of uses.
3868                  *
3869                  * Note: the following is an unsigned comparison.
3870                  */
3871                 if ((curr_events - rdev->last_events + 4096) > 8192) {
3872                         rdev->last_events = curr_events;
3873                         idle = 0;
3874                 }
3875         }
3876         return idle;
3877 }
3878
3879 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3880 {
3881         /* another "blocks" (512byte) blocks have been synced */
3882         atomic_sub(blocks, &mddev->recovery_active);
3883         wake_up(&mddev->recovery_wait);
3884         if (!ok) {
3885                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3886                 md_wakeup_thread(mddev->thread);
3887                 // stop recovery, signal do_sync ....
3888         }
3889 }
3890
3891
3892 /* md_write_start(mddev, bi)
3893  * If we need to update some array metadata (e.g. 'active' flag
3894  * in superblock) before writing, schedule a superblock update
3895  * and wait for it to complete.
3896  */
3897 void md_write_start(mddev_t *mddev, struct bio *bi)
3898 {
3899         if (bio_data_dir(bi) != WRITE)
3900                 return;
3901
3902         BUG_ON(mddev->ro == 1);
3903         if (mddev->ro == 2) {
3904                 /* need to switch to read/write */
3905                 mddev->ro = 0;
3906                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3907                 md_wakeup_thread(mddev->thread);
3908         }
3909         atomic_inc(&mddev->writes_pending);
3910         if (mddev->in_sync) {
3911                 spin_lock_irq(&mddev->write_lock);
3912                 if (mddev->in_sync) {
3913                         mddev->in_sync = 0;
3914                         mddev->sb_dirty = 1;
3915                         md_wakeup_thread(mddev->thread);
3916                 }
3917                 spin_unlock_irq(&mddev->write_lock);
3918         }
3919         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3920 }
3921
3922 void md_write_end(mddev_t *mddev)
3923 {
3924         if (atomic_dec_and_test(&mddev->writes_pending)) {
3925                 if (mddev->safemode == 2)
3926                         md_wakeup_thread(mddev->thread);
3927                 else
3928                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3929         }
3930 }
3931
3932 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3933
3934 #define SYNC_MARKS      10
3935 #define SYNC_MARK_STEP  (3*HZ)
3936 static void md_do_sync(mddev_t *mddev)
3937 {
3938         mddev_t *mddev2;
3939         unsigned int currspeed = 0,
3940                  window;
3941         sector_t max_sectors,j, io_sectors;
3942         unsigned long mark[SYNC_MARKS];
3943         sector_t mark_cnt[SYNC_MARKS];
3944         int last_mark,m;
3945         struct list_head *tmp;
3946         sector_t last_check;
3947         int skipped = 0;
3948
3949         /* just incase thread restarts... */
3950         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3951                 return;
3952
3953         /* we overload curr_resync somewhat here.
3954          * 0 == not engaged in resync at all
3955          * 2 == checking that there is no conflict with another sync
3956          * 1 == like 2, but have yielded to allow conflicting resync to
3957          *              commense
3958          * other == active in resync - this many blocks
3959          *
3960          * Before starting a resync we must have set curr_resync to
3961          * 2, and then checked that every "conflicting" array has curr_resync
3962          * less than ours.  When we find one that is the same or higher
3963          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3964          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3965          * This will mean we have to start checking from the beginning again.
3966          *
3967          */
3968
3969         do {
3970                 mddev->curr_resync = 2;
3971
3972         try_again:
3973                 if (kthread_should_stop()) {
3974                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3975                         goto skip;
3976                 }
3977                 ITERATE_MDDEV(mddev2,tmp) {
3978                         if (mddev2 == mddev)
3979                                 continue;
3980                         if (mddev2->curr_resync &&
3981                             match_mddev_units(mddev,mddev2)) {
3982                                 DEFINE_WAIT(wq);
3983                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3984                                         /* arbitrarily yield */
3985                                         mddev->curr_resync = 1;
3986                                         wake_up(&resync_wait);
3987                                 }
3988                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3989                                         /* no need to wait here, we can wait the next
3990                                          * time 'round when curr_resync == 2
3991                                          */
3992                                         continue;
3993                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
3994                                 if (!kthread_should_stop() &&
3995                                     mddev2->curr_resync >= mddev->curr_resync) {
3996                                         printk(KERN_INFO "md: delaying resync of %s"
3997                                                " until %s has finished resync (they"
3998                                                " share one or more physical units)\n",
3999                                                mdname(mddev), mdname(mddev2));
4000                                         mddev_put(mddev2);
4001                                         schedule();
4002                                         finish_wait(&resync_wait, &wq);
4003                                         goto try_again;
4004                                 }
4005                                 finish_wait(&resync_wait, &wq);
4006                         }
4007                 }
4008         } while (mddev->curr_resync < 2);
4009
4010         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4011                 /* resync follows the size requested by the personality,
4012                  * which defaults to physical size, but can be virtual size
4013                  */
4014                 max_sectors = mddev->resync_max_sectors;
4015                 mddev->resync_mismatches = 0;
4016         } else
4017                 /* recovery follows the physical size of devices */
4018                 max_sectors = mddev->size << 1;
4019
4020         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4021         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4022                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4023         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4024                "(but not more than %d KB/sec) for reconstruction.\n",
4025                sysctl_speed_limit_max);
4026
4027         is_mddev_idle(mddev); /* this also initializes IO event counters */
4028         /* we don't use the checkpoint if there's a bitmap */
4029         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4030             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4031                 j = mddev->recovery_cp;
4032         else
4033                 j = 0;
4034         io_sectors = 0;
4035         for (m = 0; m < SYNC_MARKS; m++) {
4036                 mark[m] = jiffies;
4037                 mark_cnt[m] = io_sectors;
4038         }
4039         last_mark = 0;
4040         mddev->resync_mark = mark[last_mark];
4041         mddev->resync_mark_cnt = mark_cnt[last_mark];
4042
4043         /*
4044          * Tune reconstruction:
4045          */
4046         window = 32*(PAGE_SIZE/512);
4047         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4048                 window/2,(unsigned long long) max_sectors/2);
4049
4050         atomic_set(&mddev->recovery_active, 0);
4051         init_waitqueue_head(&mddev->recovery_wait);
4052         last_check = 0;
4053
4054         if (j>2) {
4055                 printk(KERN_INFO
4056                         "md: resuming recovery of %s from checkpoint.\n",
4057                         mdname(mddev));
4058                 mddev->curr_resync = j;
4059         }
4060
4061         while (j < max_sectors) {
4062                 sector_t sectors;
4063
4064                 skipped = 0;
4065                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4066                                             currspeed < sysctl_speed_limit_min);
4067                 if (sectors == 0) {
4068                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4069                         goto out;
4070                 }
4071
4072                 if (!skipped) { /* actual IO requested */
4073                         io_sectors += sectors;
4074                         atomic_add(sectors, &mddev->recovery_active);
4075                 }
4076
4077                 j += sectors;
4078                 if (j>1) mddev->curr_resync = j;
4079
4080
4081                 if (last_check + window > io_sectors || j == max_sectors)
4082                         continue;
4083
4084                 last_check = io_sectors;
4085
4086                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4087                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4088                         break;
4089
4090         repeat:
4091                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4092                         /* step marks */
4093                         int next = (last_mark+1) % SYNC_MARKS;
4094
4095                         mddev->resync_mark = mark[next];
4096                         mddev->resync_mark_cnt = mark_cnt[next];
4097                         mark[next] = jiffies;
4098                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4099                         last_mark = next;
4100                 }
4101
4102
4103                 if (kthread_should_stop()) {
4104                         /*
4105                          * got a signal, exit.
4106                          */
4107                         printk(KERN_INFO
4108                                 "md: md_do_sync() got signal ... exiting\n");
4109                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4110                         goto out;
4111                 }
4112
4113                 /*
4114                  * this loop exits only if either when we are slower than
4115                  * the 'hard' speed limit, or the system was IO-idle for
4116                  * a jiffy.
4117                  * the system might be non-idle CPU-wise, but we only care
4118                  * about not overloading the IO subsystem. (things like an
4119                  * e2fsck being done on the RAID array should execute fast)
4120                  */
4121                 mddev->queue->unplug_fn(mddev->queue);
4122                 cond_resched();
4123
4124                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4125                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4126
4127                 if (currspeed > sysctl_speed_limit_min) {
4128                         if ((currspeed > sysctl_speed_limit_max) ||
4129                                         !is_mddev_idle(mddev)) {
4130                                 msleep(500);
4131                                 goto repeat;
4132                         }
4133                 }
4134         }
4135         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4136         /*
4137          * this also signals 'finished resyncing' to md_stop
4138          */
4139  out:
4140         mddev->queue->unplug_fn(mddev->queue);
4141
4142         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4143
4144         /* tell personality that we are finished */
4145         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4146
4147         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4148             mddev->curr_resync > 2 &&
4149             mddev->curr_resync >= mddev->recovery_cp) {
4150                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4151                         printk(KERN_INFO
4152                                 "md: checkpointing recovery of %s.\n",
4153                                 mdname(mddev));
4154                         mddev->recovery_cp = mddev->curr_resync;
4155                 } else
4156                         mddev->recovery_cp = MaxSector;
4157         }
4158
4159  skip:
4160         mddev->curr_resync = 0;
4161         wake_up(&resync_wait);
4162         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4163         md_wakeup_thread(mddev->thread);
4164 }
4165
4166
4167 /*
4168  * This routine is regularly called by all per-raid-array threads to
4169  * deal with generic issues like resync and super-block update.
4170  * Raid personalities that don't have a thread (linear/raid0) do not
4171  * need this as they never do any recovery or update the superblock.
4172  *
4173  * It does not do any resync itself, but rather "forks" off other threads
4174  * to do that as needed.
4175  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4176  * "->recovery" and create a thread at ->sync_thread.
4177  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4178  * and wakeups up this thread which will reap the thread and finish up.
4179  * This thread also removes any faulty devices (with nr_pending == 0).
4180  *
4181  * The overall approach is:
4182  *  1/ if the superblock needs updating, update it.
4183  *  2/ If a recovery thread is running, don't do anything else.
4184  *  3/ If recovery has finished, clean up, possibly marking spares active.
4185  *  4/ If there are any faulty devices, remove them.
4186  *  5/ If array is degraded, try to add spares devices
4187  *  6/ If array has spares or is not in-sync, start a resync thread.
4188  */
4189 void md_check_recovery(mddev_t *mddev)
4190 {
4191         mdk_rdev_t *rdev;
4192         struct list_head *rtmp;
4193
4194
4195         if (mddev->bitmap)
4196                 bitmap_daemon_work(mddev->bitmap);
4197
4198         if (mddev->ro)
4199                 return;
4200
4201         if (signal_pending(current)) {
4202                 if (mddev->pers->sync_request) {
4203                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4204                                mdname(mddev));
4205                         mddev->safemode = 2;
4206                 }
4207                 flush_signals(current);
4208         }
4209
4210         if ( ! (
4211                 mddev->sb_dirty ||
4212                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4213                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4214                 (mddev->safemode == 1) ||
4215                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4216                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4217                 ))
4218                 return;
4219
4220         if (mddev_trylock(mddev)==0) {
4221                 int spares =0;
4222
4223                 spin_lock_irq(&mddev->write_lock);
4224                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4225                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4226                         mddev->in_sync = 1;
4227                         mddev->sb_dirty = 1;
4228                 }
4229                 if (mddev->safemode == 1)
4230                         mddev->safemode = 0;
4231                 spin_unlock_irq(&mddev->write_lock);
4232
4233                 if (mddev->sb_dirty)
4234                         md_update_sb(mddev);
4235
4236
4237                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4238                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4239                         /* resync/recovery still happening */
4240                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4241                         goto unlock;
4242                 }
4243                 if (mddev->sync_thread) {
4244                         /* resync has finished, collect result */
4245                         md_unregister_thread(mddev->sync_thread);
4246                         mddev->sync_thread = NULL;
4247                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4248                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4249                                 /* success...*/
4250                                 /* activate any spares */
4251                                 mddev->pers->spare_active(mddev);
4252                         }
4253                         md_update_sb(mddev);
4254
4255                         /* if array is no-longer degraded, then any saved_raid_disk
4256                          * information must be scrapped
4257                          */
4258                         if (!mddev->degraded)
4259                                 ITERATE_RDEV(mddev,rdev,rtmp)
4260                                         rdev->saved_raid_disk = -1;
4261
4262                         mddev->recovery = 0;
4263                         /* flag recovery needed just to double check */
4264                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4265                         goto unlock;
4266                 }
4267                 /* Clear some bits that don't mean anything, but
4268                  * might be left set
4269                  */
4270                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4271                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4272                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4273                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4274
4275                 /* no recovery is running.
4276                  * remove any failed drives, then
4277                  * add spares if possible.
4278                  * Spare are also removed and re-added, to allow
4279                  * the personality to fail the re-add.
4280                  */
4281                 ITERATE_RDEV(mddev,rdev,rtmp)
4282                         if (rdev->raid_disk >= 0 &&
4283                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4284                             atomic_read(&rdev->nr_pending)==0) {
4285                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4286                                         char nm[20];
4287                                         sprintf(nm,"rd%d", rdev->raid_disk);
4288                                         sysfs_remove_link(&mddev->kobj, nm);
4289                                         rdev->raid_disk = -1;
4290                                 }
4291                         }
4292
4293                 if (mddev->degraded) {
4294                         ITERATE_RDEV(mddev,rdev,rtmp)
4295                                 if (rdev->raid_disk < 0
4296                                     && !test_bit(Faulty, &rdev->flags)) {
4297                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4298                                                 char nm[20];
4299                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4300                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4301                                                 spares++;
4302                                         } else
4303                                                 break;
4304                                 }
4305                 }
4306
4307                 if (spares) {
4308                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4309                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4310                 } else if (mddev->recovery_cp < MaxSector) {
4311                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4312                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4313                         /* nothing to be done ... */
4314                         goto unlock;
4315
4316                 if (mddev->pers->sync_request) {
4317                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4318                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4319                                 /* We are adding a device or devices to an array
4320                                  * which has the bitmap stored on all devices.
4321                                  * So make sure all bitmap pages get written
4322                                  */
4323                                 bitmap_write_all(mddev->bitmap);
4324                         }
4325                         mddev->sync_thread = md_register_thread(md_do_sync,
4326                                                                 mddev,
4327                                                                 "%s_resync");
4328                         if (!mddev->sync_thread) {
4329                                 printk(KERN_ERR "%s: could not start resync"
4330                                         " thread...\n",
4331                                         mdname(mddev));
4332                                 /* leave the spares where they are, it shouldn't hurt */
4333                                 mddev->recovery = 0;
4334                         } else {
4335                                 md_wakeup_thread(mddev->sync_thread);
4336                         }
4337                 }
4338         unlock:
4339                 mddev_unlock(mddev);
4340         }
4341 }
4342
4343 static int md_notify_reboot(struct notifier_block *this,
4344                             unsigned long code, void *x)
4345 {
4346         struct list_head *tmp;
4347         mddev_t *mddev;
4348
4349         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4350
4351                 printk(KERN_INFO "md: stopping all md devices.\n");
4352
4353                 ITERATE_MDDEV(mddev,tmp)
4354                         if (mddev_trylock(mddev)==0)
4355                                 do_md_stop (mddev, 1);
4356                 /*
4357                  * certain more exotic SCSI devices are known to be
4358                  * volatile wrt too early system reboots. While the
4359                  * right place to handle this issue is the given
4360                  * driver, we do want to have a safe RAID driver ...
4361                  */
4362                 mdelay(1000*1);
4363         }
4364         return NOTIFY_DONE;
4365 }
4366
4367 static struct notifier_block md_notifier = {
4368         .notifier_call  = md_notify_reboot,
4369         .next           = NULL,
4370         .priority       = INT_MAX, /* before any real devices */
4371 };
4372
4373 static void md_geninit(void)
4374 {
4375         struct proc_dir_entry *p;
4376
4377         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4378
4379         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4380         if (p)
4381                 p->proc_fops = &md_seq_fops;
4382 }
4383
4384 static int __init md_init(void)
4385 {
4386         int minor;
4387
4388         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4389                         " MD_SB_DISKS=%d\n",
4390                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4391                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4392         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4393                         BITMAP_MINOR);
4394
4395         if (register_blkdev(MAJOR_NR, "md"))
4396                 return -1;
4397         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4398                 unregister_blkdev(MAJOR_NR, "md");
4399                 return -1;
4400         }
4401         devfs_mk_dir("md");
4402         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4403                                 md_probe, NULL, NULL);
4404         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4405                             md_probe, NULL, NULL);
4406
4407         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4408                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4409                                 S_IFBLK|S_IRUSR|S_IWUSR,
4410                                 "md/%d", minor);
4411
4412         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4413                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4414                               S_IFBLK|S_IRUSR|S_IWUSR,
4415                               "md/mdp%d", minor);
4416
4417
4418         register_reboot_notifier(&md_notifier);
4419         raid_table_header = register_sysctl_table(raid_root_table, 1);
4420
4421         md_geninit();
4422         return (0);
4423 }
4424
4425
4426 #ifndef MODULE
4427
4428 /*
4429  * Searches all registered partitions for autorun RAID arrays
4430  * at boot time.
4431  */
4432 static dev_t detected_devices[128];
4433 static int dev_cnt;
4434
4435 void md_autodetect_dev(dev_t dev)
4436 {
4437         if (dev_cnt >= 0 && dev_cnt < 127)
4438                 detected_devices[dev_cnt++] = dev;
4439 }
4440
4441
4442 static void autostart_arrays(int part)
4443 {
4444         mdk_rdev_t *rdev;
4445         int i;
4446
4447         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4448
4449         for (i = 0; i < dev_cnt; i++) {
4450                 dev_t dev = detected_devices[i];
4451
4452                 rdev = md_import_device(dev,0, 0);
4453                 if (IS_ERR(rdev))
4454                         continue;
4455
4456                 if (test_bit(Faulty, &rdev->flags)) {
4457                         MD_BUG();
4458                         continue;
4459                 }
4460                 list_add(&rdev->same_set, &pending_raid_disks);
4461         }
4462         dev_cnt = 0;
4463
4464         autorun_devices(part);
4465 }
4466
4467 #endif
4468
4469 static __exit void md_exit(void)
4470 {
4471         mddev_t *mddev;
4472         struct list_head *tmp;
4473         int i;
4474         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4475         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4476         for (i=0; i < MAX_MD_DEVS; i++)
4477                 devfs_remove("md/%d", i);
4478         for (i=0; i < MAX_MD_DEVS; i++)
4479                 devfs_remove("md/d%d", i);
4480
4481         devfs_remove("md");
4482
4483         unregister_blkdev(MAJOR_NR,"md");
4484         unregister_blkdev(mdp_major, "mdp");
4485         unregister_reboot_notifier(&md_notifier);
4486         unregister_sysctl_table(raid_table_header);
4487         remove_proc_entry("mdstat", NULL);
4488         ITERATE_MDDEV(mddev,tmp) {
4489                 struct gendisk *disk = mddev->gendisk;
4490                 if (!disk)
4491                         continue;
4492                 export_array(mddev);
4493                 del_gendisk(disk);
4494                 put_disk(disk);
4495                 mddev->gendisk = NULL;
4496                 mddev_put(mddev);
4497         }
4498 }
4499
4500 module_init(md_init)
4501 module_exit(md_exit)
4502
4503 static int get_ro(char *buffer, struct kernel_param *kp)
4504 {
4505         return sprintf(buffer, "%d", start_readonly);
4506 }
4507 static int set_ro(const char *val, struct kernel_param *kp)
4508 {
4509         char *e;
4510         int num = simple_strtoul(val, &e, 10);
4511         if (*val && (*e == '\0' || *e == '\n')) {
4512                 start_readonly = num;
4513                 return 0;;
4514         }
4515         return -EINVAL;
4516 }
4517
4518 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4519 module_param(start_dirty_degraded, int, 0644);
4520
4521
4522 EXPORT_SYMBOL(register_md_personality);
4523 EXPORT_SYMBOL(unregister_md_personality);
4524 EXPORT_SYMBOL(md_error);
4525 EXPORT_SYMBOL(md_done_sync);
4526 EXPORT_SYMBOL(md_write_start);
4527 EXPORT_SYMBOL(md_write_end);
4528 EXPORT_SYMBOL(md_register_thread);
4529 EXPORT_SYMBOL(md_unregister_thread);
4530 EXPORT_SYMBOL(md_wakeup_thread);
4531 EXPORT_SYMBOL(md_print_devices);
4532 EXPORT_SYMBOL(md_check_recovery);
4533 MODULE_LICENSE("GPL");
4534 MODULE_ALIAS("md");
4535 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);