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