drivers/md/md.c

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