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