drivers/md/md.c

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