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/kthread.h>
  37 #include <linux/linkage.h>
  38 #include <linux/raid/md.h>
  39 #include <linux/raid/bitmap.h>
  40 #include <linux/sysctl.h>
  41 #include <linux/buffer_head.h> /* for invalidate_bdev */
  42 #include <linux/poll.h>
  43 #include <linux/mutex.h>
  44 #include <linux/ctype.h>
  45 #include <linux/freezer.h>
  46
  47 #include <linux/init.h>
  48
  49 #include <linux/file.h>
  50
  51 #ifdef CONFIG_KMOD
  52 #include <linux/kmod.h>
  53 #endif
  54
  55 #include <asm/unaligned.h>
  56
  57 #define MAJOR_NR MD_MAJOR
  58 #define MD_DRIVER
  59
  60 /* 63 partitions with the alternate major number (mdp) */
  61 #define MdpMinorShift 6
  62
  63 #define DEBUG 0
  64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  65
  66
  67 #ifndef MODULE
  68 static void autostart_arrays (int part);
  69 #endif
  70
  71 static LIST_HEAD(pers_list);
  72 static DEFINE_SPINLOCK(pers_lock);
  73
  74 static void md_print_devices(void);
  75
  76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  77
  78 /*
  79  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  80  * is 1000 KB/sec, so the extra system load does not show up that much.
  81  * Increase it if you want to have more _guaranteed_ speed. Note that
  82  * the RAID driver will use the maximum available bandwidth if the IO
  83  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  84  * speed limit - in case reconstruction slows down your system despite
  85  * idle IO detection.
  86  *
  87  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  88  * or /sys/block/mdX/md/sync_speed_{min,max}
  89  */
  90
  91 static int sysctl_speed_limit_min = 1000;
  92 static int sysctl_speed_limit_max = 200000;
  93 static inline int speed_min(mddev_t *mddev)
  94 {
  95         return mddev->sync_speed_min ?
  96                 mddev->sync_speed_min : sysctl_speed_limit_min;
  97 }
  98
  99 static inline int speed_max(mddev_t *mddev)
 100 {
 101         return mddev->sync_speed_max ?
 102                 mddev->sync_speed_max : sysctl_speed_limit_max;
 103 }
 104
 105 static struct ctl_table_header *raid_table_header;
 106
 107 static ctl_table raid_table[] = {
 108         {
 109                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 110                 .procname       = "speed_limit_min",
 111                 .data           = &sysctl_speed_limit_min,
 112                 .maxlen         = sizeof(int),
 113                 .mode           = S_IRUGO|S_IWUSR,
 114                 .proc_handler   = &proc_dointvec,
 115         },
 116         {
 117                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 118                 .procname       = "speed_limit_max",
 119                 .data           = &sysctl_speed_limit_max,
 120                 .maxlen         = sizeof(int),
 121                 .mode           = S_IRUGO|S_IWUSR,
 122                 .proc_handler   = &proc_dointvec,
 123         },
 124         { .ctl_name = 0 }
 125 };
 126
 127 static ctl_table raid_dir_table[] = {
 128         {
 129                 .ctl_name       = DEV_RAID,
 130                 .procname       = "raid",
 131                 .maxlen         = 0,
 132                 .mode           = S_IRUGO|S_IXUGO,
 133                 .child          = raid_table,
 134         },
 135         { .ctl_name = 0 }
 136 };
 137
 138 static ctl_table raid_root_table[] = {
 139         {
 140                 .ctl_name       = CTL_DEV,
 141                 .procname       = "dev",
 142                 .maxlen         = 0,
 143                 .mode           = 0555,
 144                 .child          = raid_dir_table,
 145         },
 146         { .ctl_name = 0 }
 147 };
 148
 149 static struct block_device_operations md_fops;
 150
 151 static int start_readonly;
 152
 153 /*
 154  * We have a system wide 'event count' that is incremented
 155  * on any 'interesting' event, and readers of /proc/mdstat
 156  * can use 'poll' or 'select' to find out when the event
 157  * count increases.
 158  *
 159  * Events are:
 160  *  start array, stop array, error, add device, remove device,
 161  *  start build, activate spare
 162  */
 163 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 164 static atomic_t md_event_count;
 165 void md_new_event(mddev_t *mddev)
 166 {
 167         atomic_inc(&md_event_count);
 168         wake_up(&md_event_waiters);
 169         sysfs_notify(&mddev->kobj, NULL, "sync_action");
 170 }
 171 EXPORT_SYMBOL_GPL(md_new_event);
 172
 173 /* Alternate version that can be called from interrupts
 174  * when calling sysfs_notify isn't needed.
 175  */
 176 static void md_new_event_inintr(mddev_t *mddev)
 177 {
 178         atomic_inc(&md_event_count);
 179         wake_up(&md_event_waiters);
 180 }
 181
 182 /*
 183  * Enables to iterate over all existing md arrays
 184  * all_mddevs_lock protects this list.
 185  */
 186 static LIST_HEAD(all_mddevs);
 187 static DEFINE_SPINLOCK(all_mddevs_lock);
 188
 189
 190 /*
 191  * iterates through all used mddevs in the system.
 192  * We take care to grab the all_mddevs_lock whenever navigating
 193  * the list, and to always hold a refcount when unlocked.
 194  * Any code which breaks out of this loop while own
 195  * a reference to the current mddev and must mddev_put it.
 196  */
 197 #define ITERATE_MDDEV(mddev,tmp)                                        \
 198                                                                         \
 199         for (({ spin_lock(&all_mddevs_lock);                            \
 200                 tmp = all_mddevs.next;                                  \
 201                 mddev = NULL;});                                        \
 202              ({ if (tmp != &all_mddevs)                                 \
 203                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 204                 spin_unlock(&all_mddevs_lock);                          \
 205                 if (mddev) mddev_put(mddev);                            \
 206                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 207                 tmp != &all_mddevs;});                                  \
 208              ({ spin_lock(&all_mddevs_lock);                            \
 209                 tmp = tmp->next;})                                      \
 210                 )
 211
 212
 213 static int md_fail_request (request_queue_t *q, struct bio *bio)
 214 {
 215         bio_io_error(bio, bio->bi_size);
 216         return 0;
 217 }
 218
 219 static inline mddev_t *mddev_get(mddev_t *mddev)
 220 {
 221         atomic_inc(&mddev->active);
 222         return mddev;
 223 }
 224
 225 static void mddev_put(mddev_t *mddev)
 226 {
 227         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 228                 return;
 229         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 230                 list_del(&mddev->all_mddevs);
 231                 spin_unlock(&all_mddevs_lock);
 232                 blk_cleanup_queue(mddev->queue);
 233                 kobject_unregister(&mddev->kobj);
 234         } else
 235                 spin_unlock(&all_mddevs_lock);
 236 }
 237
 238 static mddev_t * mddev_find(dev_t unit)
 239 {
 240         mddev_t *mddev, *new = NULL;
 241
 242  retry:
 243         spin_lock(&all_mddevs_lock);
 244         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 245                 if (mddev->unit == unit) {
 246                         mddev_get(mddev);
 247                         spin_unlock(&all_mddevs_lock);
 248                         kfree(new);
 249                         return mddev;
 250                 }
 251
 252         if (new) {
 253                 list_add(&new->all_mddevs, &all_mddevs);
 254                 spin_unlock(&all_mddevs_lock);
 255                 return new;
 256         }
 257         spin_unlock(&all_mddevs_lock);
 258
 259         new = kzalloc(sizeof(*new), GFP_KERNEL);
 260         if (!new)
 261                 return NULL;
 262
 263         new->unit = unit;
 264         if (MAJOR(unit) == MD_MAJOR)
 265                 new->md_minor = MINOR(unit);
 266         else
 267                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 268
 269         mutex_init(&new->reconfig_mutex);
 270         INIT_LIST_HEAD(&new->disks);
 271         INIT_LIST_HEAD(&new->all_mddevs);
 272         init_timer(&new->safemode_timer);
 273         atomic_set(&new->active, 1);
 274         spin_lock_init(&new->write_lock);
 275         init_waitqueue_head(&new->sb_wait);
 276
 277         new->queue = blk_alloc_queue(GFP_KERNEL);
 278         if (!new->queue) {
 279                 kfree(new);
 280                 return NULL;
 281         }
 282         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
 283
 284         blk_queue_make_request(new->queue, md_fail_request);
 285
 286         goto retry;
 287 }
 288
 289 static inline int mddev_lock(mddev_t * mddev)
 290 {
 291         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 292 }
 293
 294 static inline int mddev_trylock(mddev_t * mddev)
 295 {
 296         return mutex_trylock(&mddev->reconfig_mutex);
 297 }
 298
 299 static inline void mddev_unlock(mddev_t * mddev)
 300 {
 301         mutex_unlock(&mddev->reconfig_mutex);
 302
 303         md_wakeup_thread(mddev->thread);
 304 }
 305
 306 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 307 {
 308         mdk_rdev_t * rdev;
 309         struct list_head *tmp;
 310
 311         ITERATE_RDEV(mddev,rdev,tmp) {
 312                 if (rdev->desc_nr == nr)
 313                         return rdev;
 314         }
 315         return NULL;
 316 }
 317
 318 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 319 {
 320         struct list_head *tmp;
 321         mdk_rdev_t *rdev;
 322
 323         ITERATE_RDEV(mddev,rdev,tmp) {
 324                 if (rdev->bdev->bd_dev == dev)
 325                         return rdev;
 326         }
 327         return NULL;
 328 }
 329
 330 static struct mdk_personality *find_pers(int level, char *clevel)
 331 {
 332         struct mdk_personality *pers;
 333         list_for_each_entry(pers, &pers_list, list) {
 334                 if (level != LEVEL_NONE && pers->level == level)
 335                         return pers;
 336                 if (strcmp(pers->name, clevel)==0)
 337                         return pers;
 338         }
 339         return NULL;
 340 }
 341
 342 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 343 {
 344         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 345         return MD_NEW_SIZE_BLOCKS(size);
 346 }
 347
 348 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 349 {
 350         sector_t size;
 351
 352         size = rdev->sb_offset;
 353
 354         if (chunk_size)
 355                 size &= ~((sector_t)chunk_size/1024 - 1);
 356         return size;
 357 }
 358
 359 static int alloc_disk_sb(mdk_rdev_t * rdev)
 360 {
 361         if (rdev->sb_page)
 362                 MD_BUG();
 363
 364         rdev->sb_page = alloc_page(GFP_KERNEL);
 365         if (!rdev->sb_page) {
 366                 printk(KERN_ALERT "md: out of memory.\n");
 367                 return -EINVAL;
 368         }
 369
 370         return 0;
 371 }
 372
 373 static void free_disk_sb(mdk_rdev_t * rdev)
 374 {
 375         if (rdev->sb_page) {
 376                 put_page(rdev->sb_page);
 377                 rdev->sb_loaded = 0;
 378                 rdev->sb_page = NULL;
 379                 rdev->sb_offset = 0;
 380                 rdev->size = 0;
 381         }
 382 }
 383
 384
 385 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 386 {
 387         mdk_rdev_t *rdev = bio->bi_private;
 388         mddev_t *mddev = rdev->mddev;
 389         if (bio->bi_size)
 390                 return 1;
 391
 392         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 393                 printk("md: super_written gets error=%d, uptodate=%d\n",
 394                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 395                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 396                 md_error(mddev, rdev);
 397         }
 398
 399         if (atomic_dec_and_test(&mddev->pending_writes))
 400                 wake_up(&mddev->sb_wait);
 401         bio_put(bio);
 402         return 0;
 403 }
 404
 405 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 406 {
 407         struct bio *bio2 = bio->bi_private;
 408         mdk_rdev_t *rdev = bio2->bi_private;
 409         mddev_t *mddev = rdev->mddev;
 410         if (bio->bi_size)
 411                 return 1;
 412
 413         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 414             error == -EOPNOTSUPP) {
 415                 unsigned long flags;
 416                 /* barriers don't appear to be supported :-( */
 417                 set_bit(BarriersNotsupp, &rdev->flags);
 418                 mddev->barriers_work = 0;
 419                 spin_lock_irqsave(&mddev->write_lock, flags);
 420                 bio2->bi_next = mddev->biolist;
 421                 mddev->biolist = bio2;
 422                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 423                 wake_up(&mddev->sb_wait);
 424                 bio_put(bio);
 425                 return 0;
 426         }
 427         bio_put(bio2);
 428         bio->bi_private = rdev;
 429         return super_written(bio, bytes_done, error);
 430 }
 431
 432 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 433                    sector_t sector, int size, struct page *page)
 434 {
 435         /* write first size bytes of page to sector of rdev
 436          * Increment mddev->pending_writes before returning
 437          * and decrement it on completion, waking up sb_wait
 438          * if zero is reached.
 439          * If an error occurred, call md_error
 440          *
 441          * As we might need to resubmit the request if BIO_RW_BARRIER
 442          * causes ENOTSUPP, we allocate a spare bio...
 443          */
 444         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 445         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 446
 447         bio->bi_bdev = rdev->bdev;
 448         bio->bi_sector = sector;
 449         bio_add_page(bio, page, size, 0);
 450         bio->bi_private = rdev;
 451         bio->bi_end_io = super_written;
 452         bio->bi_rw = rw;
 453
 454         atomic_inc(&mddev->pending_writes);
 455         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 456                 struct bio *rbio;
 457                 rw |= (1<<BIO_RW_BARRIER);
 458                 rbio = bio_clone(bio, GFP_NOIO);
 459                 rbio->bi_private = bio;
 460                 rbio->bi_end_io = super_written_barrier;
 461                 submit_bio(rw, rbio);
 462         } else
 463                 submit_bio(rw, bio);
 464 }
 465
 466 void md_super_wait(mddev_t *mddev)
 467 {
 468         /* wait for all superblock writes that were scheduled to complete.
 469          * if any had to be retried (due to BARRIER problems), retry them
 470          */
 471         DEFINE_WAIT(wq);
 472         for(;;) {
 473                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 474                 if (atomic_read(&mddev->pending_writes)==0)
 475                         break;
 476                 while (mddev->biolist) {
 477                         struct bio *bio;
 478                         spin_lock_irq(&mddev->write_lock);
 479                         bio = mddev->biolist;
 480                         mddev->biolist = bio->bi_next ;
 481                         bio->bi_next = NULL;
 482                         spin_unlock_irq(&mddev->write_lock);
 483                         submit_bio(bio->bi_rw, bio);
 484                 }
 485                 schedule();
 486         }
 487         finish_wait(&mddev->sb_wait, &wq);
 488 }
 489
 490 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 491 {
 492         if (bio->bi_size)
 493                 return 1;
 494
 495         complete((struct completion*)bio->bi_private);
 496         return 0;
 497 }
 498
 499 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 500                    struct page *page, int rw)
 501 {
 502         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 503         struct completion event;
 504         int ret;
 505
 506         rw |= (1 << BIO_RW_SYNC);
 507
 508         bio->bi_bdev = bdev;
 509         bio->bi_sector = sector;
 510         bio_add_page(bio, page, size, 0);
 511         init_completion(&event);
 512         bio->bi_private = &event;
 513         bio->bi_end_io = bi_complete;
 514         submit_bio(rw, bio);
 515         wait_for_completion(&event);
 516
 517         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 518         bio_put(bio);
 519         return ret;
 520 }
 521 EXPORT_SYMBOL_GPL(sync_page_io);
 522
 523 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 524 {
 525         char b[BDEVNAME_SIZE];
 526         if (!rdev->sb_page) {
 527                 MD_BUG();
 528                 return -EINVAL;
 529         }
 530         if (rdev->sb_loaded)
 531                 return 0;
 532
 533
 534         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 535                 goto fail;
 536         rdev->sb_loaded = 1;
 537         return 0;
 538
 539 fail:
 540         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 541                 bdevname(rdev->bdev,b));
 542         return -EINVAL;
 543 }
 544
 545 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 546 {
 547         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 548                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 549                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 550                 (sb1->set_uuid3 == sb2->set_uuid3))
 551
 552                 return 1;
 553
 554         return 0;
 555 }
 556
 557
 558 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 559 {
 560         int ret;
 561         mdp_super_t *tmp1, *tmp2;
 562
 563         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 564         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 565
 566         if (!tmp1 || !tmp2) {
 567                 ret = 0;
 568                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 569                 goto abort;
 570         }
 571
 572         *tmp1 = *sb1;
 573         *tmp2 = *sb2;
 574
 575         /*
 576          * nr_disks is not constant
 577          */
 578         tmp1->nr_disks = 0;
 579         tmp2->nr_disks = 0;
 580
 581         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 582                 ret = 0;
 583         else
 584                 ret = 1;
 585
 586 abort:
 587         kfree(tmp1);
 588         kfree(tmp2);
 589         return ret;
 590 }
 591
 592 static unsigned int calc_sb_csum(mdp_super_t * sb)
 593 {
 594         unsigned int disk_csum, csum;
 595
 596         disk_csum = sb->sb_csum;
 597         sb->sb_csum = 0;
 598         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 599         sb->sb_csum = disk_csum;
 600         return csum;
 601 }
 602
 603
 604 /*
 605  * Handle superblock details.
 606  * We want to be able to handle multiple superblock formats
 607  * so we have a common interface to them all, and an array of
 608  * different handlers.
 609  * We rely on user-space to write the initial superblock, and support
 610  * reading and updating of superblocks.
 611  * Interface methods are:
 612  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 613  *      loads and validates a superblock on dev.
 614  *      if refdev != NULL, compare superblocks on both devices
 615  *    Return:
 616  *      0 - dev has a superblock that is compatible with refdev
 617  *      1 - dev has a superblock that is compatible and newer than refdev
 618  *          so dev should be used as the refdev in future
 619  *     -EINVAL superblock incompatible or invalid
 620  *     -othererror e.g. -EIO
 621  *
 622  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 623  *      Verify that dev is acceptable into mddev.
 624  *       The first time, mddev->raid_disks will be 0, and data from
 625  *       dev should be merged in.  Subsequent calls check that dev
 626  *       is new enough.  Return 0 or -EINVAL
 627  *
 628  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 629  *     Update the superblock for rdev with data in mddev
 630  *     This does not write to disc.
 631  *
 632  */
 633
 634 struct super_type  {
 635         char            *name;
 636         struct module   *owner;
 637         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 638         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 639         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 640 };
 641
 642 /*
 643  * load_super for 0.90.0
 644  */
 645 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 646 {
 647         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 648         mdp_super_t *sb;
 649         int ret;
 650         sector_t sb_offset;
 651
 652         /*
 653          * Calculate the position of the superblock,
 654          * it's at the end of the disk.
 655          *
 656          * It also happens to be a multiple of 4Kb.
 657          */
 658         sb_offset = calc_dev_sboffset(rdev->bdev);
 659         rdev->sb_offset = sb_offset;
 660
 661         ret = read_disk_sb(rdev, MD_SB_BYTES);
 662         if (ret) return ret;
 663
 664         ret = -EINVAL;
 665
 666         bdevname(rdev->bdev, b);
 667         sb = (mdp_super_t*)page_address(rdev->sb_page);
 668
 669         if (sb->md_magic != MD_SB_MAGIC) {
 670                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 671                        b);
 672                 goto abort;
 673         }
 674
 675         if (sb->major_version != 0 ||
 676             sb->minor_version < 90 ||
 677             sb->minor_version > 91) {
 678                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 679                         sb->major_version, sb->minor_version,
 680                         b);
 681                 goto abort;
 682         }
 683
 684         if (sb->raid_disks <= 0)
 685                 goto abort;
 686
 687         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 688                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 689                         b);
 690                 goto abort;
 691         }
 692
 693         rdev->preferred_minor = sb->md_minor;
 694         rdev->data_offset = 0;
 695         rdev->sb_size = MD_SB_BYTES;
 696
 697         if (sb->level == LEVEL_MULTIPATH)
 698                 rdev->desc_nr = -1;
 699         else
 700                 rdev->desc_nr = sb->this_disk.number;
 701
 702         if (refdev == 0)
 703                 ret = 1;
 704         else {
 705                 __u64 ev1, ev2;
 706                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 707                 if (!uuid_equal(refsb, sb)) {
 708                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 709                                 b, bdevname(refdev->bdev,b2));
 710                         goto abort;
 711                 }
 712                 if (!sb_equal(refsb, sb)) {
 713                         printk(KERN_WARNING "md: %s has same UUID"
 714                                " but different superblock to %s\n",
 715                                b, bdevname(refdev->bdev, b2));
 716                         goto abort;
 717                 }
 718                 ev1 = md_event(sb);
 719                 ev2 = md_event(refsb);
 720                 if (ev1 > ev2)
 721                         ret = 1;
 722                 else
 723                         ret = 0;
 724         }
 725         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 726
 727         if (rdev->size < sb->size && sb->level > 1)
 728                 /* "this cannot possibly happen" ... */
 729                 ret = -EINVAL;
 730
 731  abort:
 732         return ret;
 733 }
 734
 735 /*
 736  * validate_super for 0.90.0
 737  */
 738 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 739 {
 740         mdp_disk_t *desc;
 741         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 742         __u64 ev1 = md_event(sb);
 743
 744         rdev->raid_disk = -1;
 745         rdev->flags = 0;
 746         if (mddev->raid_disks == 0) {
 747                 mddev->major_version = 0;
 748                 mddev->minor_version = sb->minor_version;
 749                 mddev->patch_version = sb->patch_version;
 750                 mddev->persistent = ! sb->not_persistent;
 751                 mddev->chunk_size = sb->chunk_size;
 752                 mddev->ctime = sb->ctime;
 753                 mddev->utime = sb->utime;
 754                 mddev->level = sb->level;
 755                 mddev->clevel[0] = 0;
 756                 mddev->layout = sb->layout;
 757                 mddev->raid_disks = sb->raid_disks;
 758                 mddev->size = sb->size;
 759                 mddev->events = ev1;
 760                 mddev->bitmap_offset = 0;
 761                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 762
 763                 if (mddev->minor_version >= 91) {
 764                         mddev->reshape_position = sb->reshape_position;
 765                         mddev->delta_disks = sb->delta_disks;
 766                         mddev->new_level = sb->new_level;
 767                         mddev->new_layout = sb->new_layout;
 768                         mddev->new_chunk = sb->new_chunk;
 769                 } else {
 770                         mddev->reshape_position = MaxSector;
 771                         mddev->delta_disks = 0;
 772                         mddev->new_level = mddev->level;
 773                         mddev->new_layout = mddev->layout;
 774                         mddev->new_chunk = mddev->chunk_size;
 775                 }
 776
 777                 if (sb->state & (1<<MD_SB_CLEAN))
 778                         mddev->recovery_cp = MaxSector;
 779                 else {
 780                         if (sb->events_hi == sb->cp_events_hi &&
 781                                 sb->events_lo == sb->cp_events_lo) {
 782                                 mddev->recovery_cp = sb->recovery_cp;
 783                         } else
 784                                 mddev->recovery_cp = 0;
 785                 }
 786
 787                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 788                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 789                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 790                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 791
 792                 mddev->max_disks = MD_SB_DISKS;
 793
 794                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 795                     mddev->bitmap_file == NULL) {
 796                         if (mddev->level != 1 && mddev->level != 4
 797                             && mddev->level != 5 && mddev->level != 6
 798                             && mddev->level != 10) {
 799                                 /* FIXME use a better test */
 800                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
 801                                 return -EINVAL;
 802                         }
 803                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 804                 }
 805
 806         } else if (mddev->pers == NULL) {
 807                 /* Insist on good event counter while assembling */
 808                 ++ev1;
 809                 if (ev1 < mddev->events)
 810                         return -EINVAL;
 811         } else if (mddev->bitmap) {
 812                 /* if adding to array with a bitmap, then we can accept an
 813                  * older device ... but not too old.
 814                  */
 815                 if (ev1 < mddev->bitmap->events_cleared)
 816                         return 0;
 817         } else {
 818                 if (ev1 < mddev->events)
 819                         /* just a hot-add of a new device, leave raid_disk at -1 */
 820                         return 0;
 821         }
 822
 823         if (mddev->level != LEVEL_MULTIPATH) {
 824                 desc = sb->disks + rdev->desc_nr;
 825
 826                 if (desc->state & (1<<MD_DISK_FAULTY))
 827                         set_bit(Faulty, &rdev->flags);
 828                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
 829                             desc->raid_disk < mddev->raid_disks */) {
 830                         set_bit(In_sync, &rdev->flags);
 831                         rdev->raid_disk = desc->raid_disk;
 832                 }
 833                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 834                         set_bit(WriteMostly, &rdev->flags);
 835         } else /* MULTIPATH are always insync */
 836                 set_bit(In_sync, &rdev->flags);
 837         return 0;
 838 }
 839
 840 /*
 841  * sync_super for 0.90.0
 842  */
 843 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 844 {
 845         mdp_super_t *sb;
 846         struct list_head *tmp;
 847         mdk_rdev_t *rdev2;
 848         int next_spare = mddev->raid_disks;
 849
 850
 851         /* make rdev->sb match mddev data..
 852          *
 853          * 1/ zero out disks
 854          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 855          * 3/ any empty disks < next_spare become removed
 856          *
 857          * disks[0] gets initialised to REMOVED because
 858          * we cannot be sure from other fields if it has
 859          * been initialised or not.
 860          */
 861         int i;
 862         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 863
 864         rdev->sb_size = MD_SB_BYTES;
 865
 866         sb = (mdp_super_t*)page_address(rdev->sb_page);
 867
 868         memset(sb, 0, sizeof(*sb));
 869
 870         sb->md_magic = MD_SB_MAGIC;
 871         sb->major_version = mddev->major_version;
 872         sb->patch_version = mddev->patch_version;
 873         sb->gvalid_words  = 0; /* ignored */
 874         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 875         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 876         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 877         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 878
 879         sb->ctime = mddev->ctime;
 880         sb->level = mddev->level;
 881         sb->size  = mddev->size;
 882         sb->raid_disks = mddev->raid_disks;
 883         sb->md_minor = mddev->md_minor;
 884         sb->not_persistent = !mddev->persistent;
 885         sb->utime = mddev->utime;
 886         sb->state = 0;
 887         sb->events_hi = (mddev->events>>32);
 888         sb->events_lo = (u32)mddev->events;
 889
 890         if (mddev->reshape_position == MaxSector)
 891                 sb->minor_version = 90;
 892         else {
 893                 sb->minor_version = 91;
 894                 sb->reshape_position = mddev->reshape_position;
 895                 sb->new_level = mddev->new_level;
 896                 sb->delta_disks = mddev->delta_disks;
 897                 sb->new_layout = mddev->new_layout;
 898                 sb->new_chunk = mddev->new_chunk;
 899         }
 900         mddev->minor_version = sb->minor_version;
 901         if (mddev->in_sync)
 902         {
 903                 sb->recovery_cp = mddev->recovery_cp;
 904                 sb->cp_events_hi = (mddev->events>>32);
 905                 sb->cp_events_lo = (u32)mddev->events;
 906                 if (mddev->recovery_cp == MaxSector)
 907                         sb->state = (1<< MD_SB_CLEAN);
 908         } else
 909                 sb->recovery_cp = 0;
 910
 911         sb->layout = mddev->layout;
 912         sb->chunk_size = mddev->chunk_size;
 913
 914         if (mddev->bitmap && mddev->bitmap_file == NULL)
 915                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 916
 917         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 918         ITERATE_RDEV(mddev,rdev2,tmp) {
 919                 mdp_disk_t *d;
 920                 int desc_nr;
 921                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 922                     && !test_bit(Faulty, &rdev2->flags))
 923                         desc_nr = rdev2->raid_disk;
 924                 else
 925                         desc_nr = next_spare++;
 926                 rdev2->desc_nr = desc_nr;
 927                 d = &sb->disks[rdev2->desc_nr];
 928                 nr_disks++;
 929                 d->number = rdev2->desc_nr;
 930                 d->major = MAJOR(rdev2->bdev->bd_dev);
 931                 d->minor = MINOR(rdev2->bdev->bd_dev);
 932                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 933                     && !test_bit(Faulty, &rdev2->flags))
 934                         d->raid_disk = rdev2->raid_disk;
 935                 else
 936                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 937                 if (test_bit(Faulty, &rdev2->flags))
 938                         d->state = (1<<MD_DISK_FAULTY);
 939                 else if (test_bit(In_sync, &rdev2->flags)) {
 940                         d->state = (1<<MD_DISK_ACTIVE);
 941                         d->state |= (1<<MD_DISK_SYNC);
 942                         active++;
 943                         working++;
 944                 } else {
 945                         d->state = 0;
 946                         spare++;
 947                         working++;
 948                 }
 949                 if (test_bit(WriteMostly, &rdev2->flags))
 950                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 951         }
 952         /* now set the "removed" and "faulty" bits on any missing devices */
 953         for (i=0 ; i < mddev->raid_disks ; i++) {
 954                 mdp_disk_t *d = &sb->disks[i];
 955                 if (d->state == 0 && d->number == 0) {
 956                         d->number = i;
 957                         d->raid_disk = i;
 958                         d->state = (1<<MD_DISK_REMOVED);
 959                         d->state |= (1<<MD_DISK_FAULTY);
 960                         failed++;
 961                 }
 962         }
 963         sb->nr_disks = nr_disks;
 964         sb->active_disks = active;
 965         sb->working_disks = working;
 966         sb->failed_disks = failed;
 967         sb->spare_disks = spare;
 968
 969         sb->this_disk = sb->disks[rdev->desc_nr];
 970         sb->sb_csum = calc_sb_csum(sb);
 971 }
 972
 973 /*
 974  * version 1 superblock
 975  */
 976
 977 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
 978 {
 979         __le32 disk_csum;
 980         u32 csum;
 981         unsigned long long newcsum;
 982         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 983         __le32 *isuper = (__le32*)sb;
 984         int i;
 985
 986         disk_csum = sb->sb_csum;
 987         sb->sb_csum = 0;
 988         newcsum = 0;
 989         for (i=0; size>=4; size -= 4 )
 990                 newcsum += le32_to_cpu(*isuper++);
 991
 992         if (size == 2)
 993                 newcsum += le16_to_cpu(*(__le16*) isuper);
 994
 995         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 996         sb->sb_csum = disk_csum;
 997         return cpu_to_le32(csum);
 998 }
 999
1000 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1001 {
1002         struct mdp_superblock_1 *sb;
1003         int ret;
1004         sector_t sb_offset;
1005         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1006         int bmask;
1007
1008         /*
1009          * Calculate the position of the superblock.
1010          * It is always aligned to a 4K boundary and
1011          * depeding on minor_version, it can be:
1012          * 0: At least 8K, but less than 12K, from end of device
1013          * 1: At start of device
1014          * 2: 4K from start of device.
1015          */
1016         switch(minor_version) {
1017         case 0:
1018                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1019                 sb_offset -= 8*2;
1020                 sb_offset &= ~(sector_t)(4*2-1);
1021                 /* convert from sectors to K */
1022                 sb_offset /= 2;
1023                 break;
1024         case 1:
1025                 sb_offset = 0;
1026                 break;
1027         case 2:
1028                 sb_offset = 4;
1029                 break;
1030         default:
1031                 return -EINVAL;
1032         }
1033         rdev->sb_offset = sb_offset;
1034
1035         /* superblock is rarely larger than 1K, but it can be larger,
1036          * and it is safe to read 4k, so we do that
1037          */
1038         ret = read_disk_sb(rdev, 4096);
1039         if (ret) return ret;
1040
1041
1042         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1043
1044         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1045             sb->major_version != cpu_to_le32(1) ||
1046             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1047             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1048             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1049                 return -EINVAL;
1050
1051         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1052                 printk("md: invalid superblock checksum on %s\n",
1053                         bdevname(rdev->bdev,b));
1054                 return -EINVAL;
1055         }
1056         if (le64_to_cpu(sb->data_size) < 10) {
1057                 printk("md: data_size too small on %s\n",
1058                        bdevname(rdev->bdev,b));
1059                 return -EINVAL;
1060         }
1061         rdev->preferred_minor = 0xffff;
1062         rdev->data_offset = le64_to_cpu(sb->data_offset);
1063         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1064
1065         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1066         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1067         if (rdev->sb_size & bmask)
1068                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1069
1070         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1071                 rdev->desc_nr = -1;
1072         else
1073                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1074
1075         if (refdev == 0)
1076                 ret = 1;
1077         else {
1078                 __u64 ev1, ev2;
1079                 struct mdp_superblock_1 *refsb =
1080                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1081
1082                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1083                     sb->level != refsb->level ||
1084                     sb->layout != refsb->layout ||
1085                     sb->chunksize != refsb->chunksize) {
1086                         printk(KERN_WARNING "md: %s has strangely different"
1087                                 " superblock to %s\n",
1088                                 bdevname(rdev->bdev,b),
1089                                 bdevname(refdev->bdev,b2));
1090                         return -EINVAL;
1091                 }
1092                 ev1 = le64_to_cpu(sb->events);
1093                 ev2 = le64_to_cpu(refsb->events);
1094
1095                 if (ev1 > ev2)
1096                         ret = 1;
1097                 else
1098                         ret = 0;
1099         }
1100         if (minor_version)
1101                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1102         else
1103                 rdev->size = rdev->sb_offset;
1104         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1105                 return -EINVAL;
1106         rdev->size = le64_to_cpu(sb->data_size)/2;
1107         if (le32_to_cpu(sb->chunksize))
1108                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1109
1110         if (le64_to_cpu(sb->size) > rdev->size*2)
1111                 return -EINVAL;
1112         return ret;
1113 }
1114
1115 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1116 {
1117         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1118         __u64 ev1 = le64_to_cpu(sb->events);
1119
1120         rdev->raid_disk = -1;
1121         rdev->flags = 0;
1122         if (mddev->raid_disks == 0) {
1123                 mddev->major_version = 1;
1124                 mddev->patch_version = 0;
1125                 mddev->persistent = 1;
1126                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1127                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1128                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1129                 mddev->level = le32_to_cpu(sb->level);
1130                 mddev->clevel[0] = 0;
1131                 mddev->layout = le32_to_cpu(sb->layout);
1132                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1133                 mddev->size = le64_to_cpu(sb->size)/2;
1134                 mddev->events = ev1;
1135                 mddev->bitmap_offset = 0;
1136                 mddev->default_bitmap_offset = 1024 >> 9;
1137
1138                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1139                 memcpy(mddev->uuid, sb->set_uuid, 16);
1140
1141                 mddev->max_disks =  (4096-256)/2;
1142
1143                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1144                     mddev->bitmap_file == NULL ) {
1145                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1146                             && mddev->level != 10) {
1147                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1148                                 return -EINVAL;
1149                         }
1150                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1151                 }
1152                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1153                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1154                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1155                         mddev->new_level = le32_to_cpu(sb->new_level);
1156                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1157                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1158                 } else {
1159                         mddev->reshape_position = MaxSector;
1160                         mddev->delta_disks = 0;
1161                         mddev->new_level = mddev->level;
1162                         mddev->new_layout = mddev->layout;
1163                         mddev->new_chunk = mddev->chunk_size;
1164                 }
1165
1166         } else if (mddev->pers == NULL) {
1167                 /* Insist of good event counter while assembling */
1168                 ++ev1;
1169                 if (ev1 < mddev->events)
1170                         return -EINVAL;
1171         } else if (mddev->bitmap) {
1172                 /* If adding to array with a bitmap, then we can accept an
1173                  * older device, but not too old.
1174                  */
1175                 if (ev1 < mddev->bitmap->events_cleared)
1176                         return 0;
1177         } else {
1178                 if (ev1 < mddev->events)
1179                         /* just a hot-add of a new device, leave raid_disk at -1 */
1180                         return 0;
1181         }
1182         if (mddev->level != LEVEL_MULTIPATH) {
1183                 int role;
1184                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1185                 switch(role) {
1186                 case 0xffff: /* spare */
1187                         break;
1188                 case 0xfffe: /* faulty */
1189                         set_bit(Faulty, &rdev->flags);
1190                         break;
1191                 default:
1192                         if ((le32_to_cpu(sb->feature_map) &
1193                              MD_FEATURE_RECOVERY_OFFSET))
1194                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1195                         else
1196                                 set_bit(In_sync, &rdev->flags);
1197                         rdev->raid_disk = role;
1198                         break;
1199                 }
1200                 if (sb->devflags & WriteMostly1)
1201                         set_bit(WriteMostly, &rdev->flags);
1202         } else /* MULTIPATH are always insync */
1203                 set_bit(In_sync, &rdev->flags);
1204
1205         return 0;
1206 }
1207
1208 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1209 {
1210         struct mdp_superblock_1 *sb;
1211         struct list_head *tmp;
1212         mdk_rdev_t *rdev2;
1213         int max_dev, i;
1214         /* make rdev->sb match mddev and rdev data. */
1215
1216         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1217
1218         sb->feature_map = 0;
1219         sb->pad0 = 0;
1220         sb->recovery_offset = cpu_to_le64(0);
1221         memset(sb->pad1, 0, sizeof(sb->pad1));
1222         memset(sb->pad2, 0, sizeof(sb->pad2));
1223         memset(sb->pad3, 0, sizeof(sb->pad3));
1224
1225         sb->utime = cpu_to_le64((__u64)mddev->utime);
1226         sb->events = cpu_to_le64(mddev->events);
1227         if (mddev->in_sync)
1228                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1229         else
1230                 sb->resync_offset = cpu_to_le64(0);
1231
1232         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1233
1234         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1235         sb->size = cpu_to_le64(mddev->size<<1);
1236
1237         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1238                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1239                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1240         }
1241
1242         if (rdev->raid_disk >= 0 &&
1243             !test_bit(In_sync, &rdev->flags) &&
1244             rdev->recovery_offset > 0) {
1245                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1246                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1247         }
1248
1249         if (mddev->reshape_position != MaxSector) {
1250                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1251                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1252                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1253                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1254                 sb->new_level = cpu_to_le32(mddev->new_level);
1255                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1256         }
1257
1258         max_dev = 0;
1259         ITERATE_RDEV(mddev,rdev2,tmp)
1260                 if (rdev2->desc_nr+1 > max_dev)
1261                         max_dev = rdev2->desc_nr+1;
1262
1263         sb->max_dev = cpu_to_le32(max_dev);
1264         for (i=0; i<max_dev;i++)
1265                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1266
1267         ITERATE_RDEV(mddev,rdev2,tmp) {
1268                 i = rdev2->desc_nr;
1269                 if (test_bit(Faulty, &rdev2->flags))
1270                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1271                 else if (test_bit(In_sync, &rdev2->flags))
1272                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1273                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1274                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1275                 else
1276                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1277         }
1278
1279         sb->sb_csum = calc_sb_1_csum(sb);
1280 }
1281
1282
1283 static struct super_type super_types[] = {
1284         [0] = {
1285                 .name   = "0.90.0",
1286                 .owner  = THIS_MODULE,
1287                 .load_super     = super_90_load,
1288                 .validate_super = super_90_validate,
1289                 .sync_super     = super_90_sync,
1290         },
1291         [1] = {
1292                 .name   = "md-1",
1293                 .owner  = THIS_MODULE,
1294                 .load_super     = super_1_load,
1295                 .validate_super = super_1_validate,
1296                 .sync_super     = super_1_sync,
1297         },
1298 };
1299
1300 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1301 {
1302         struct list_head *tmp, *tmp2;
1303         mdk_rdev_t *rdev, *rdev2;
1304
1305         ITERATE_RDEV(mddev1,rdev,tmp)
1306                 ITERATE_RDEV(mddev2, rdev2, tmp2)
1307                         if (rdev->bdev->bd_contains ==
1308                             rdev2->bdev->bd_contains)
1309                                 return 1;
1310
1311         return 0;
1312 }
1313
1314 static LIST_HEAD(pending_raid_disks);
1315
1316 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1317 {
1318         char b[BDEVNAME_SIZE];
1319         struct kobject *ko;
1320         char *s;
1321         int err;
1322
1323         if (rdev->mddev) {
1324                 MD_BUG();
1325                 return -EINVAL;
1326         }
1327         /* make sure rdev->size exceeds mddev->size */
1328         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1329                 if (mddev->pers)
1330                         /* Cannot change size, so fail */
1331                         return -ENOSPC;
1332                 else
1333                         mddev->size = rdev->size;
1334         }
1335
1336         /* Verify rdev->desc_nr is unique.
1337          * If it is -1, assign a free number, else
1338          * check number is not in use
1339          */
1340         if (rdev->desc_nr < 0) {
1341                 int choice = 0;
1342                 if (mddev->pers) choice = mddev->raid_disks;
1343                 while (find_rdev_nr(mddev, choice))
1344                         choice++;
1345                 rdev->desc_nr = choice;
1346         } else {
1347                 if (find_rdev_nr(mddev, rdev->desc_nr))
1348                         return -EBUSY;
1349         }
1350         bdevname(rdev->bdev,b);
1351         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1352                 return -ENOMEM;
1353         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1354                 *s = '!';
1355
1356         rdev->mddev = mddev;
1357         printk(KERN_INFO "md: bind<%s>\n", b);
1358
1359         rdev->kobj.parent = &mddev->kobj;
1360         if ((err = kobject_add(&rdev->kobj)))
1361                 goto fail;
1362
1363         if (rdev->bdev->bd_part)
1364                 ko = &rdev->bdev->bd_part->kobj;
1365         else
1366                 ko = &rdev->bdev->bd_disk->kobj;
1367         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1368                 kobject_del(&rdev->kobj);
1369                 goto fail;
1370         }
1371         list_add(&rdev->same_set, &mddev->disks);
1372         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1373         return 0;
1374
1375  fail:
1376         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1377                b, mdname(mddev));
1378         return err;
1379 }
1380
1381 static void delayed_delete(struct work_struct *ws)
1382 {
1383         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1384         kobject_del(&rdev->kobj);
1385 }
1386
1387 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1388 {
1389         char b[BDEVNAME_SIZE];
1390         if (!rdev->mddev) {
1391                 MD_BUG();
1392                 return;
1393         }
1394         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1395         list_del_init(&rdev->same_set);
1396         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1397         rdev->mddev = NULL;
1398         sysfs_remove_link(&rdev->kobj, "block");
1399
1400         /* We need to delay this, otherwise we can deadlock when
1401          * writing to 'remove' to "dev/state"
1402          */
1403         INIT_WORK(&rdev->del_work, delayed_delete);
1404         schedule_work(&rdev->del_work);
1405 }
1406
1407 /*
1408  * prevent the device from being mounted, repartitioned or
1409  * otherwise reused by a RAID array (or any other kernel
1410  * subsystem), by bd_claiming the device.
1411  */
1412 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1413 {
1414         int err = 0;
1415         struct block_device *bdev;
1416         char b[BDEVNAME_SIZE];
1417
1418         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1419         if (IS_ERR(bdev)) {
1420                 printk(KERN_ERR "md: could not open %s.\n",
1421                         __bdevname(dev, b));
1422                 return PTR_ERR(bdev);
1423         }
1424         err = bd_claim(bdev, rdev);
1425         if (err) {
1426                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1427                         bdevname(bdev, b));
1428                 blkdev_put(bdev);
1429                 return err;
1430         }
1431         rdev->bdev = bdev;
1432         return err;
1433 }
1434
1435 static void unlock_rdev(mdk_rdev_t *rdev)
1436 {
1437         struct block_device *bdev = rdev->bdev;
1438         rdev->bdev = NULL;
1439         if (!bdev)
1440                 MD_BUG();
1441         bd_release(bdev);
1442         blkdev_put(bdev);
1443 }
1444
1445 void md_autodetect_dev(dev_t dev);
1446
1447 static void export_rdev(mdk_rdev_t * rdev)
1448 {
1449         char b[BDEVNAME_SIZE];
1450         printk(KERN_INFO "md: export_rdev(%s)\n",
1451                 bdevname(rdev->bdev,b));
1452         if (rdev->mddev)
1453                 MD_BUG();
1454         free_disk_sb(rdev);
1455         list_del_init(&rdev->same_set);
1456 #ifndef MODULE
1457         md_autodetect_dev(rdev->bdev->bd_dev);
1458 #endif
1459         unlock_rdev(rdev);
1460         kobject_put(&rdev->kobj);
1461 }
1462
1463 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1464 {
1465         unbind_rdev_from_array(rdev);
1466         export_rdev(rdev);
1467 }
1468
1469 static void export_array(mddev_t *mddev)
1470 {
1471         struct list_head *tmp;
1472         mdk_rdev_t *rdev;
1473
1474         ITERATE_RDEV(mddev,rdev,tmp) {
1475                 if (!rdev->mddev) {
1476                         MD_BUG();
1477                         continue;
1478                 }
1479                 kick_rdev_from_array(rdev);
1480         }
1481         if (!list_empty(&mddev->disks))
1482                 MD_BUG();
1483         mddev->raid_disks = 0;
1484         mddev->major_version = 0;
1485 }
1486
1487 static void print_desc(mdp_disk_t *desc)
1488 {
1489         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1490                 desc->major,desc->minor,desc->raid_disk,desc->state);
1491 }
1492
1493 static void print_sb(mdp_super_t *sb)
1494 {
1495         int i;
1496
1497         printk(KERN_INFO
1498                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1499                 sb->major_version, sb->minor_version, sb->patch_version,
1500                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1501                 sb->ctime);
1502         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1503                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1504                 sb->md_minor, sb->layout, sb->chunk_size);
1505         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1506                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1507                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1508                 sb->failed_disks, sb->spare_disks,
1509                 sb->sb_csum, (unsigned long)sb->events_lo);
1510
1511         printk(KERN_INFO);
1512         for (i = 0; i < MD_SB_DISKS; i++) {
1513                 mdp_disk_t *desc;
1514
1515                 desc = sb->disks + i;
1516                 if (desc->number || desc->major || desc->minor ||
1517                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1518                         printk("     D %2d: ", i);
1519                         print_desc(desc);
1520                 }
1521         }
1522         printk(KERN_INFO "md:     THIS: ");
1523         print_desc(&sb->this_disk);
1524
1525 }
1526
1527 static void print_rdev(mdk_rdev_t *rdev)
1528 {
1529         char b[BDEVNAME_SIZE];
1530         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1531                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1532                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1533                 rdev->desc_nr);
1534         if (rdev->sb_loaded) {
1535                 printk(KERN_INFO "md: rdev superblock:\n");
1536                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1537         } else
1538                 printk(KERN_INFO "md: no rdev superblock!\n");
1539 }
1540
1541 static void md_print_devices(void)
1542 {
1543         struct list_head *tmp, *tmp2;
1544         mdk_rdev_t *rdev;
1545         mddev_t *mddev;
1546         char b[BDEVNAME_SIZE];
1547
1548         printk("\n");
1549         printk("md:     **********************************\n");
1550         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1551         printk("md:     **********************************\n");
1552         ITERATE_MDDEV(mddev,tmp) {
1553
1554                 if (mddev->bitmap)
1555                         bitmap_print_sb(mddev->bitmap);
1556                 else
1557                         printk("%s: ", mdname(mddev));
1558                 ITERATE_RDEV(mddev,rdev,tmp2)
1559                         printk("<%s>", bdevname(rdev->bdev,b));
1560                 printk("\n");
1561
1562                 ITERATE_RDEV(mddev,rdev,tmp2)
1563                         print_rdev(rdev);
1564         }
1565         printk("md:     **********************************\n");
1566         printk("\n");
1567 }
1568
1569
1570 static void sync_sbs(mddev_t * mddev, int nospares)
1571 {
1572         /* Update each superblock (in-memory image), but
1573          * if we are allowed to, skip spares which already
1574          * have the right event counter, or have one earlier
1575          * (which would mean they aren't being marked as dirty
1576          * with the rest of the array)
1577          */
1578         mdk_rdev_t *rdev;
1579         struct list_head *tmp;
1580
1581         ITERATE_RDEV(mddev,rdev,tmp) {
1582                 if (rdev->sb_events == mddev->events ||
1583                     (nospares &&
1584                      rdev->raid_disk < 0 &&
1585                      (rdev->sb_events&1)==0 &&
1586                      rdev->sb_events+1 == mddev->events)) {
1587                         /* Don't update this superblock */
1588                         rdev->sb_loaded = 2;
1589                 } else {
1590                         super_types[mddev->major_version].
1591                                 sync_super(mddev, rdev);
1592                         rdev->sb_loaded = 1;
1593                 }
1594         }
1595 }
1596
1597 static void md_update_sb(mddev_t * mddev, int force_change)
1598 {
1599         int err;
1600         struct list_head *tmp;
1601         mdk_rdev_t *rdev;
1602         int sync_req;
1603         int nospares = 0;
1604
1605 repeat:
1606         spin_lock_irq(&mddev->write_lock);
1607
1608         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1609         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1610                 force_change = 1;
1611         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1612                 /* just a clean<-> dirty transition, possibly leave spares alone,
1613                  * though if events isn't the right even/odd, we will have to do
1614                  * spares after all
1615                  */
1616                 nospares = 1;
1617         if (force_change)
1618                 nospares = 0;
1619         if (mddev->degraded)
1620                 /* If the array is degraded, then skipping spares is both
1621                  * dangerous and fairly pointless.
1622                  * Dangerous because a device that was removed from the array
1623                  * might have a event_count that still looks up-to-date,
1624                  * so it can be re-added without a resync.
1625                  * Pointless because if there are any spares to skip,
1626                  * then a recovery will happen and soon that array won't
1627                  * be degraded any more and the spare can go back to sleep then.
1628                  */
1629                 nospares = 0;
1630
1631         sync_req = mddev->in_sync;
1632         mddev->utime = get_seconds();
1633
1634         /* If this is just a dirty<->clean transition, and the array is clean
1635          * and 'events' is odd, we can roll back to the previous clean state */
1636         if (nospares
1637             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1638             && (mddev->events & 1)
1639             && mddev->events != 1)
1640                 mddev->events--;
1641         else {
1642                 /* otherwise we have to go forward and ... */
1643                 mddev->events ++;
1644                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1645                         /* .. if the array isn't clean, insist on an odd 'events' */
1646                         if ((mddev->events&1)==0) {
1647                                 mddev->events++;
1648                                 nospares = 0;
1649                         }
1650                 } else {
1651                         /* otherwise insist on an even 'events' (for clean states) */
1652                         if ((mddev->events&1)) {
1653                                 mddev->events++;
1654                                 nospares = 0;
1655                         }
1656                 }
1657         }
1658
1659         if (!mddev->events) {
1660                 /*
1661                  * oops, this 64-bit counter should never wrap.
1662                  * Either we are in around ~1 trillion A.C., assuming
1663                  * 1 reboot per second, or we have a bug:
1664                  */
1665                 MD_BUG();
1666                 mddev->events --;
1667         }
1668         sync_sbs(mddev, nospares);
1669
1670         /*
1671          * do not write anything to disk if using
1672          * nonpersistent superblocks
1673          */
1674         if (!mddev->persistent) {
1675                 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1676                 spin_unlock_irq(&mddev->write_lock);
1677                 wake_up(&mddev->sb_wait);
1678                 return;
1679         }
1680         spin_unlock_irq(&mddev->write_lock);
1681
1682         dprintk(KERN_INFO
1683                 "md: updating %s RAID superblock on device (in sync %d)\n",
1684                 mdname(mddev),mddev->in_sync);
1685
1686         err = bitmap_update_sb(mddev->bitmap);
1687         ITERATE_RDEV(mddev,rdev,tmp) {
1688                 char b[BDEVNAME_SIZE];
1689                 dprintk(KERN_INFO "md: ");
1690                 if (rdev->sb_loaded != 1)
1691                         continue; /* no noise on spare devices */
1692                 if (test_bit(Faulty, &rdev->flags))
1693                         dprintk("(skipping faulty ");
1694
1695                 dprintk("%s ", bdevname(rdev->bdev,b));
1696                 if (!test_bit(Faulty, &rdev->flags)) {
1697                         md_super_write(mddev,rdev,
1698                                        rdev->sb_offset<<1, rdev->sb_size,
1699                                        rdev->sb_page);
1700                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1701                                 bdevname(rdev->bdev,b),
1702                                 (unsigned long long)rdev->sb_offset);
1703                         rdev->sb_events = mddev->events;
1704
1705                 } else
1706                         dprintk(")\n");
1707                 if (mddev->level == LEVEL_MULTIPATH)
1708                         /* only need to write one superblock... */
1709                         break;
1710         }
1711         md_super_wait(mddev);
1712         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1713
1714         spin_lock_irq(&mddev->write_lock);
1715         if (mddev->in_sync != sync_req ||
1716             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1717                 /* have to write it out again */
1718                 spin_unlock_irq(&mddev->write_lock);
1719                 goto repeat;
1720         }
1721         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1722         spin_unlock_irq(&mddev->write_lock);
1723         wake_up(&mddev->sb_wait);
1724
1725 }
1726
1727 /* words written to sysfs files may, or my not, be \n terminated.
1728  * We want to accept with case. For this we use cmd_match.
1729  */
1730 static int cmd_match(const char *cmd, const char *str)
1731 {
1732         /* See if cmd, written into a sysfs file, matches
1733          * str.  They must either be the same, or cmd can
1734          * have a trailing newline
1735          */
1736         while (*cmd && *str && *cmd == *str) {
1737                 cmd++;
1738                 str++;
1739         }
1740         if (*cmd == '\n')
1741                 cmd++;
1742         if (*str || *cmd)
1743                 return 0;
1744         return 1;
1745 }
1746
1747 struct rdev_sysfs_entry {
1748         struct attribute attr;
1749         ssize_t (*show)(mdk_rdev_t *, char *);
1750         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1751 };
1752
1753 static ssize_t
1754 state_show(mdk_rdev_t *rdev, char *page)
1755 {
1756         char *sep = "";
1757         int len=0;
1758
1759         if (test_bit(Faulty, &rdev->flags)) {
1760                 len+= sprintf(page+len, "%sfaulty",sep);
1761                 sep = ",";
1762         }
1763         if (test_bit(In_sync, &rdev->flags)) {
1764                 len += sprintf(page+len, "%sin_sync",sep);
1765                 sep = ",";
1766         }
1767         if (test_bit(WriteMostly, &rdev->flags)) {
1768                 len += sprintf(page+len, "%swrite_mostly",sep);
1769                 sep = ",";
1770         }
1771         if (!test_bit(Faulty, &rdev->flags) &&
1772             !test_bit(In_sync, &rdev->flags)) {
1773                 len += sprintf(page+len, "%sspare", sep);
1774                 sep = ",";
1775         }
1776         return len+sprintf(page+len, "\n");
1777 }
1778
1779 static ssize_t
1780 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1781 {
1782         /* can write
1783          *  faulty  - simulates and error
1784          *  remove  - disconnects the device
1785          *  writemostly - sets write_mostly
1786          *  -writemostly - clears write_mostly
1787          */
1788         int err = -EINVAL;
1789         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1790                 md_error(rdev->mddev, rdev);
1791                 err = 0;
1792         } else if (cmd_match(buf, "remove")) {
1793                 if (rdev->raid_disk >= 0)
1794                         err = -EBUSY;
1795                 else {
1796                         mddev_t *mddev = rdev->mddev;
1797                         kick_rdev_from_array(rdev);
1798                         if (mddev->pers)
1799                                 md_update_sb(mddev, 1);
1800                         md_new_event(mddev);
1801                         err = 0;
1802                 }
1803         } else if (cmd_match(buf, "writemostly")) {
1804                 set_bit(WriteMostly, &rdev->flags);
1805                 err = 0;
1806         } else if (cmd_match(buf, "-writemostly")) {
1807                 clear_bit(WriteMostly, &rdev->flags);
1808                 err = 0;
1809         }
1810         return err ? err : len;
1811 }
1812 static struct rdev_sysfs_entry rdev_state =
1813 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1814
1815 static ssize_t
1816 super_show(mdk_rdev_t *rdev, char *page)
1817 {
1818         if (rdev->sb_loaded && rdev->sb_size) {
1819                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1820                 return rdev->sb_size;
1821         } else
1822                 return 0;
1823 }
1824 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1825
1826 static ssize_t
1827 errors_show(mdk_rdev_t *rdev, char *page)
1828 {
1829         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1830 }
1831
1832 static ssize_t
1833 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1834 {
1835         char *e;
1836         unsigned long n = simple_strtoul(buf, &e, 10);
1837         if (*buf && (*e == 0 || *e == '\n')) {
1838                 atomic_set(&rdev->corrected_errors, n);
1839                 return len;
1840         }
1841         return -EINVAL;
1842 }
1843 static struct rdev_sysfs_entry rdev_errors =
1844 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1845
1846 static ssize_t
1847 slot_show(mdk_rdev_t *rdev, char *page)
1848 {
1849         if (rdev->raid_disk < 0)
1850                 return sprintf(page, "none\n");
1851         else
1852                 return sprintf(page, "%d\n", rdev->raid_disk);
1853 }
1854
1855 static ssize_t
1856 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1857 {
1858         char *e;
1859         int slot = simple_strtoul(buf, &e, 10);
1860         if (strncmp(buf, "none", 4)==0)
1861                 slot = -1;
1862         else if (e==buf || (*e && *e!= '\n'))
1863                 return -EINVAL;
1864         if (rdev->mddev->pers)
1865                 /* Cannot set slot in active array (yet) */
1866                 return -EBUSY;
1867         if (slot >= rdev->mddev->raid_disks)
1868                 return -ENOSPC;
1869         rdev->raid_disk = slot;
1870         /* assume it is working */
1871         rdev->flags = 0;
1872         set_bit(In_sync, &rdev->flags);
1873         return len;
1874 }
1875
1876
1877 static struct rdev_sysfs_entry rdev_slot =
1878 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1879
1880 static ssize_t
1881 offset_show(mdk_rdev_t *rdev, char *page)
1882 {
1883         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1884 }
1885
1886 static ssize_t
1887 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1888 {
1889         char *e;
1890         unsigned long long offset = simple_strtoull(buf, &e, 10);
1891         if (e==buf || (*e && *e != '\n'))
1892                 return -EINVAL;
1893         if (rdev->mddev->pers)
1894                 return -EBUSY;
1895         rdev->data_offset = offset;
1896         return len;
1897 }
1898
1899 static struct rdev_sysfs_entry rdev_offset =
1900 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1901
1902 static ssize_t
1903 rdev_size_show(mdk_rdev_t *rdev, char *page)
1904 {
1905         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1906 }
1907
1908 static ssize_t
1909 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1910 {
1911         char *e;
1912         unsigned long long size = simple_strtoull(buf, &e, 10);
1913         if (e==buf || (*e && *e != '\n'))
1914                 return -EINVAL;
1915         if (rdev->mddev->pers)
1916                 return -EBUSY;
1917         rdev->size = size;
1918         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1919                 rdev->mddev->size = size;
1920         return len;
1921 }
1922
1923 static struct rdev_sysfs_entry rdev_size =
1924 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1925
1926 static struct attribute *rdev_default_attrs[] = {
1927         &rdev_state.attr,
1928         &rdev_super.attr,
1929         &rdev_errors.attr,
1930         &rdev_slot.attr,
1931         &rdev_offset.attr,
1932         &rdev_size.attr,
1933         NULL,
1934 };
1935 static ssize_t
1936 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1937 {
1938         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1939         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1940
1941         if (!entry->show)
1942                 return -EIO;
1943         return entry->show(rdev, page);
1944 }
1945
1946 static ssize_t
1947 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1948               const char *page, size_t length)
1949 {
1950         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1951         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1952
1953         if (!entry->store)
1954                 return -EIO;
1955         if (!capable(CAP_SYS_ADMIN))
1956                 return -EACCES;
1957         return entry->store(rdev, page, length);
1958 }
1959
1960 static void rdev_free(struct kobject *ko)
1961 {
1962         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1963         kfree(rdev);
1964 }
1965 static struct sysfs_ops rdev_sysfs_ops = {
1966         .show           = rdev_attr_show,
1967         .store          = rdev_attr_store,
1968 };
1969 static struct kobj_type rdev_ktype = {
1970         .release        = rdev_free,
1971         .sysfs_ops      = &rdev_sysfs_ops,
1972         .default_attrs  = rdev_default_attrs,
1973 };
1974
1975 /*
1976  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1977  *
1978  * mark the device faulty if:
1979  *
1980  *   - the device is nonexistent (zero size)
1981  *   - the device has no valid superblock
1982  *
1983  * a faulty rdev _never_ has rdev->sb set.
1984  */
1985 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1986 {
1987         char b[BDEVNAME_SIZE];
1988         int err;
1989         mdk_rdev_t *rdev;
1990         sector_t size;
1991
1992         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1993         if (!rdev) {
1994                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1995                 return ERR_PTR(-ENOMEM);
1996         }
1997
1998         if ((err = alloc_disk_sb(rdev)))
1999                 goto abort_free;
2000
2001         err = lock_rdev(rdev, newdev);
2002         if (err)
2003                 goto abort_free;
2004
2005         rdev->kobj.parent = NULL;
2006         rdev->kobj.ktype = &rdev_ktype;
2007         kobject_init(&rdev->kobj);
2008
2009         rdev->desc_nr = -1;
2010         rdev->saved_raid_disk = -1;
2011         rdev->raid_disk = -1;
2012         rdev->flags = 0;
2013         rdev->data_offset = 0;
2014         rdev->sb_events = 0;
2015         atomic_set(&rdev->nr_pending, 0);
2016         atomic_set(&rdev->read_errors, 0);
2017         atomic_set(&rdev->corrected_errors, 0);
2018
2019         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2020         if (!size) {
2021                 printk(KERN_WARNING
2022                         "md: %s has zero or unknown size, marking faulty!\n",
2023                         bdevname(rdev->bdev,b));
2024                 err = -EINVAL;
2025                 goto abort_free;
2026         }
2027
2028         if (super_format >= 0) {
2029                 err = super_types[super_format].
2030                         load_super(rdev, NULL, super_minor);
2031                 if (err == -EINVAL) {
2032                         printk(KERN_WARNING
2033                                 "md: %s has invalid sb, not importing!\n",
2034                                 bdevname(rdev->bdev,b));
2035                         goto abort_free;
2036                 }
2037                 if (err < 0) {
2038                         printk(KERN_WARNING
2039                                 "md: could not read %s's sb, not importing!\n",
2040                                 bdevname(rdev->bdev,b));
2041                         goto abort_free;
2042                 }
2043         }
2044         INIT_LIST_HEAD(&rdev->same_set);
2045
2046         return rdev;
2047
2048 abort_free:
2049         if (rdev->sb_page) {
2050                 if (rdev->bdev)
2051                         unlock_rdev(rdev);
2052                 free_disk_sb(rdev);
2053         }
2054         kfree(rdev);
2055         return ERR_PTR(err);
2056 }
2057
2058 /*
2059  * Check a full RAID array for plausibility
2060  */
2061
2062
2063 static void analyze_sbs(mddev_t * mddev)
2064 {
2065         int i;
2066         struct list_head *tmp;
2067         mdk_rdev_t *rdev, *freshest;
2068         char b[BDEVNAME_SIZE];
2069
2070         freshest = NULL;
2071         ITERATE_RDEV(mddev,rdev,tmp)
2072                 switch (super_types[mddev->major_version].
2073                         load_super(rdev, freshest, mddev->minor_version)) {
2074                 case 1:
2075                         freshest = rdev;
2076                         break;
2077                 case 0:
2078                         break;
2079                 default:
2080                         printk( KERN_ERR \
2081                                 "md: fatal superblock inconsistency in %s"
2082                                 " -- removing from array\n",
2083                                 bdevname(rdev->bdev,b));
2084                         kick_rdev_from_array(rdev);
2085                 }
2086
2087
2088         super_types[mddev->major_version].
2089                 validate_super(mddev, freshest);
2090
2091         i = 0;
2092         ITERATE_RDEV(mddev,rdev,tmp) {
2093                 if (rdev != freshest)
2094                         if (super_types[mddev->major_version].
2095                             validate_super(mddev, rdev)) {
2096                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2097                                         " from array!\n",
2098                                         bdevname(rdev->bdev,b));
2099                                 kick_rdev_from_array(rdev);
2100                                 continue;
2101                         }
2102                 if (mddev->level == LEVEL_MULTIPATH) {
2103                         rdev->desc_nr = i++;
2104                         rdev->raid_disk = rdev->desc_nr;
2105                         set_bit(In_sync, &rdev->flags);
2106                 }
2107         }
2108
2109
2110
2111         if (mddev->recovery_cp != MaxSector &&
2112             mddev->level >= 1)
2113                 printk(KERN_ERR "md: %s: raid array is not clean"
2114                        " -- starting background reconstruction\n",
2115                        mdname(mddev));
2116
2117 }
2118
2119 static ssize_t
2120 safe_delay_show(mddev_t *mddev, char *page)
2121 {
2122         int msec = (mddev->safemode_delay*1000)/HZ;
2123         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2124 }
2125 static ssize_t
2126 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2127 {
2128         int scale=1;
2129         int dot=0;
2130         int i;
2131         unsigned long msec;
2132         char buf[30];
2133         char *e;
2134         /* remove a period, and count digits after it */
2135         if (len >= sizeof(buf))
2136                 return -EINVAL;
2137         strlcpy(buf, cbuf, len);
2138         buf[len] = 0;
2139         for (i=0; i<len; i++) {
2140                 if (dot) {
2141                         if (isdigit(buf[i])) {
2142                                 buf[i-1] = buf[i];
2143                                 scale *= 10;
2144                         }
2145                         buf[i] = 0;
2146                 } else if (buf[i] == '.') {
2147                         dot=1;
2148                         buf[i] = 0;
2149                 }
2150         }
2151         msec = simple_strtoul(buf, &e, 10);
2152         if (e == buf || (*e && *e != '\n'))
2153                 return -EINVAL;
2154         msec = (msec * 1000) / scale;
2155         if (msec == 0)
2156                 mddev->safemode_delay = 0;
2157         else {
2158                 mddev->safemode_delay = (msec*HZ)/1000;
2159                 if (mddev->safemode_delay == 0)
2160                         mddev->safemode_delay = 1;
2161         }
2162         return len;
2163 }
2164 static struct md_sysfs_entry md_safe_delay =
2165 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2166
2167 static ssize_t
2168 level_show(mddev_t *mddev, char *page)
2169 {
2170         struct mdk_personality *p = mddev->pers;
2171         if (p)
2172                 return sprintf(page, "%s\n", p->name);
2173         else if (mddev->clevel[0])
2174                 return sprintf(page, "%s\n", mddev->clevel);
2175         else if (mddev->level != LEVEL_NONE)
2176                 return sprintf(page, "%d\n", mddev->level);
2177         else
2178                 return 0;
2179 }
2180
2181 static ssize_t
2182 level_store(mddev_t *mddev, const char *buf, size_t len)
2183 {
2184         int rv = len;
2185         if (mddev->pers)
2186                 return -EBUSY;
2187         if (len == 0)
2188                 return 0;
2189         if (len >= sizeof(mddev->clevel))
2190                 return -ENOSPC;
2191         strncpy(mddev->clevel, buf, len);
2192         if (mddev->clevel[len-1] == '\n')
2193                 len--;
2194         mddev->clevel[len] = 0;
2195         mddev->level = LEVEL_NONE;
2196         return rv;
2197 }
2198
2199 static struct md_sysfs_entry md_level =
2200 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2201
2202
2203 static ssize_t
2204 layout_show(mddev_t *mddev, char *page)
2205 {
2206         /* just a number, not meaningful for all levels */
2207         return sprintf(page, "%d\n", mddev->layout);
2208 }
2209
2210 static ssize_t
2211 layout_store(mddev_t *mddev, const char *buf, size_t len)
2212 {
2213         char *e;
2214         unsigned long n = simple_strtoul(buf, &e, 10);
2215         if (mddev->pers)
2216                 return -EBUSY;
2217
2218         if (!*buf || (*e && *e != '\n'))
2219                 return -EINVAL;
2220
2221         mddev->layout = n;
2222         return len;
2223 }
2224 static struct md_sysfs_entry md_layout =
2225 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2226
2227
2228 static ssize_t
2229 raid_disks_show(mddev_t *mddev, char *page)
2230 {
2231         if (mddev->raid_disks == 0)
2232                 return 0;
2233         return sprintf(page, "%d\n", mddev->raid_disks);
2234 }
2235
2236 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2237
2238 static ssize_t
2239 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2240 {
2241         char *e;
2242         int rv = 0;
2243         unsigned long n = simple_strtoul(buf, &e, 10);
2244
2245         if (!*buf || (*e && *e != '\n'))
2246                 return -EINVAL;
2247
2248         if (mddev->pers)
2249                 rv = update_raid_disks(mddev, n);
2250         else
2251                 mddev->raid_disks = n;
2252         return rv ? rv : len;
2253 }
2254 static struct md_sysfs_entry md_raid_disks =
2255 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2256
2257 static ssize_t
2258 chunk_size_show(mddev_t *mddev, char *page)
2259 {
2260         return sprintf(page, "%d\n", mddev->chunk_size);
2261 }
2262
2263 static ssize_t
2264 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2265 {
2266         /* can only set chunk_size if array is not yet active */
2267         char *e;
2268         unsigned long n = simple_strtoul(buf, &e, 10);
2269
2270         if (mddev->pers)
2271                 return -EBUSY;
2272         if (!*buf || (*e && *e != '\n'))
2273                 return -EINVAL;
2274
2275         mddev->chunk_size = n;
2276         return len;
2277 }
2278 static struct md_sysfs_entry md_chunk_size =
2279 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2280
2281 static ssize_t
2282 resync_start_show(mddev_t *mddev, char *page)
2283 {
2284         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2285 }
2286
2287 static ssize_t
2288 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2289 {
2290         /* can only set chunk_size if array is not yet active */
2291         char *e;
2292         unsigned long long n = simple_strtoull(buf, &e, 10);
2293
2294         if (mddev->pers)
2295                 return -EBUSY;
2296         if (!*buf || (*e && *e != '\n'))
2297                 return -EINVAL;
2298
2299         mddev->recovery_cp = n;
2300         return len;
2301 }
2302 static struct md_sysfs_entry md_resync_start =
2303 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2304
2305 /*
2306  * The array state can be:
2307  *
2308  * clear
2309  *     No devices, no size, no level
2310  *     Equivalent to STOP_ARRAY ioctl
2311  * inactive
2312  *     May have some settings, but array is not active
2313  *        all IO results in error
2314  *     When written, doesn't tear down array, but just stops it
2315  * suspended (not supported yet)
2316  *     All IO requests will block. The array can be reconfigured.
2317  *     Writing this, if accepted, will block until array is quiessent
2318  * readonly
2319  *     no resync can happen.  no superblocks get written.
2320  *     write requests fail
2321  * read-auto
2322  *     like readonly, but behaves like 'clean' on a write request.
2323  *
2324  * clean - no pending writes, but otherwise active.
2325  *     When written to inactive array, starts without resync
2326  *     If a write request arrives then
2327  *       if metadata is known, mark 'dirty' and switch to 'active'.
2328  *       if not known, block and switch to write-pending
2329  *     If written to an active array that has pending writes, then fails.
2330  * active
2331  *     fully active: IO and resync can be happening.
2332  *     When written to inactive array, starts with resync
2333  *
2334  * write-pending
2335  *     clean, but writes are blocked waiting for 'active' to be written.
2336  *
2337  * active-idle
2338  *     like active, but no writes have been seen for a while (100msec).
2339  *
2340  */
2341 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2342                    write_pending, active_idle, bad_word};
2343 static char *array_states[] = {
2344         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2345         "write-pending", "active-idle", NULL };
2346
2347 static int match_word(const char *word, char **list)
2348 {
2349         int n;
2350         for (n=0; list[n]; n++)
2351                 if (cmd_match(word, list[n]))
2352                         break;
2353         return n;
2354 }
2355
2356 static ssize_t
2357 array_state_show(mddev_t *mddev, char *page)
2358 {
2359         enum array_state st = inactive;
2360
2361         if (mddev->pers)
2362                 switch(mddev->ro) {
2363                 case 1:
2364                         st = readonly;
2365                         break;
2366                 case 2:
2367                         st = read_auto;
2368                         break;
2369                 case 0:
2370                         if (mddev->in_sync)
2371                                 st = clean;
2372                         else if (mddev->safemode)
2373                                 st = active_idle;
2374                         else
2375                                 st = active;
2376                 }
2377         else {
2378                 if (list_empty(&mddev->disks) &&
2379                     mddev->raid_disks == 0 &&
2380                     mddev->size == 0)
2381                         st = clear;
2382                 else
2383                         st = inactive;
2384         }
2385         return sprintf(page, "%s\n", array_states[st]);
2386 }
2387
2388 static int do_md_stop(mddev_t * mddev, int ro);
2389 static int do_md_run(mddev_t * mddev);
2390 static int restart_array(mddev_t *mddev);
2391
2392 static ssize_t
2393 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2394 {
2395         int err = -EINVAL;
2396         enum array_state st = match_word(buf, array_states);
2397         switch(st) {
2398         case bad_word:
2399                 break;
2400         case clear:
2401                 /* stopping an active array */
2402                 if (mddev->pers) {
2403                         if (atomic_read(&mddev->active) > 1)
2404                                 return -EBUSY;
2405                         err = do_md_stop(mddev, 0);
2406                 }
2407                 break;
2408         case inactive:
2409                 /* stopping an active array */
2410                 if (mddev->pers) {
2411                         if (atomic_read(&mddev->active) > 1)
2412                                 return -EBUSY;
2413                         err = do_md_stop(mddev, 2);
2414                 }
2415                 break;
2416         case suspended:
2417                 break; /* not supported yet */
2418         case readonly:
2419                 if (mddev->pers)
2420                         err = do_md_stop(mddev, 1);
2421                 else {
2422                         mddev->ro = 1;
2423                         err = do_md_run(mddev);
2424                 }
2425                 break;
2426         case read_auto:
2427                 /* stopping an active array */
2428                 if (mddev->pers) {
2429                         err = do_md_stop(mddev, 1);
2430                         if (err == 0)
2431                                 mddev->ro = 2; /* FIXME mark devices writable */
2432                 } else {
2433                         mddev->ro = 2;
2434                         err = do_md_run(mddev);
2435                 }
2436                 break;
2437         case clean:
2438                 if (mddev->pers) {
2439                         restart_array(mddev);
2440                         spin_lock_irq(&mddev->write_lock);
2441                         if (atomic_read(&mddev->writes_pending) == 0) {
2442                                 mddev->in_sync = 1;
2443                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2444                         }
2445                         spin_unlock_irq(&mddev->write_lock);
2446                 } else {
2447                         mddev->ro = 0;
2448                         mddev->recovery_cp = MaxSector;
2449                         err = do_md_run(mddev);
2450                 }
2451                 break;
2452         case active:
2453                 if (mddev->pers) {
2454                         restart_array(mddev);
2455                         clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2456                         wake_up(&mddev->sb_wait);
2457                         err = 0;
2458                 } else {
2459                         mddev->ro = 0;
2460                         err = do_md_run(mddev);
2461                 }
2462                 break;
2463         case write_pending:
2464         case active_idle:
2465                 /* these cannot be set */
2466                 break;
2467         }
2468         if (err)
2469                 return err;
2470         else
2471                 return len;
2472 }
2473 static struct md_sysfs_entry md_array_state =
2474 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2475
2476 static ssize_t
2477 null_show(mddev_t *mddev, char *page)
2478 {
2479         return -EINVAL;
2480 }
2481
2482 static ssize_t
2483 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2484 {
2485         /* buf must be %d:%d\n? giving major and minor numbers */
2486         /* The new device is added to the array.
2487          * If the array has a persistent superblock, we read the
2488          * superblock to initialise info and check validity.
2489          * Otherwise, only checking done is that in bind_rdev_to_array,
2490          * which mainly checks size.
2491          */
2492         char *e;
2493         int major = simple_strtoul(buf, &e, 10);
2494         int minor;
2495         dev_t dev;
2496         mdk_rdev_t *rdev;
2497         int err;
2498
2499         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2500                 return -EINVAL;
2501         minor = simple_strtoul(e+1, &e, 10);
2502         if (*e && *e != '\n')
2503                 return -EINVAL;
2504         dev = MKDEV(major, minor);
2505         if (major != MAJOR(dev) ||
2506             minor != MINOR(dev))
2507                 return -EOVERFLOW;
2508
2509
2510         if (mddev->persistent) {
2511                 rdev = md_import_device(dev, mddev->major_version,
2512                                         mddev->minor_version);
2513                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2514                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2515                                                        mdk_rdev_t, same_set);
2516                         err = super_types[mddev->major_version]
2517                                 .load_super(rdev, rdev0, mddev->minor_version);
2518                         if (err < 0)
2519                                 goto out;
2520                 }
2521         } else
2522                 rdev = md_import_device(dev, -1, -1);
2523
2524         if (IS_ERR(rdev))
2525                 return PTR_ERR(rdev);
2526         err = bind_rdev_to_array(rdev, mddev);
2527  out:
2528         if (err)
2529                 export_rdev(rdev);
2530         return err ? err : len;
2531 }
2532
2533 static struct md_sysfs_entry md_new_device =
2534 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2535
2536 static ssize_t
2537 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2538 {
2539         char *end;
2540         unsigned long chunk, end_chunk;
2541
2542         if (!mddev->bitmap)
2543                 goto out;
2544         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2545         while (*buf) {
2546                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2547                 if (buf == end) break;
2548                 if (*end == '-') { /* range */
2549                         buf = end + 1;
2550                         end_chunk = simple_strtoul(buf, &end, 0);
2551                         if (buf == end) break;
2552                 }
2553                 if (*end && !isspace(*end)) break;
2554                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2555                 buf = end;
2556                 while (isspace(*buf)) buf++;
2557         }
2558         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2559 out:
2560         return len;
2561 }
2562
2563 static struct md_sysfs_entry md_bitmap =
2564 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2565
2566 static ssize_t
2567 size_show(mddev_t *mddev, char *page)
2568 {
2569         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2570 }
2571
2572 static int update_size(mddev_t *mddev, unsigned long size);
2573
2574 static ssize_t
2575 size_store(mddev_t *mddev, const char *buf, size_t len)
2576 {
2577         /* If array is inactive, we can reduce the component size, but
2578          * not increase it (except from 0).
2579          * If array is active, we can try an on-line resize
2580          */
2581         char *e;
2582         int err = 0;
2583         unsigned long long size = simple_strtoull(buf, &e, 10);
2584         if (!*buf || *buf == '\n' ||
2585             (*e && *e != '\n'))
2586                 return -EINVAL;
2587
2588         if (mddev->pers) {
2589                 err = update_size(mddev, size);
2590                 md_update_sb(mddev, 1);
2591         } else {
2592                 if (mddev->size == 0 ||
2593                     mddev->size > size)
2594                         mddev->size = size;
2595                 else
2596                         err = -ENOSPC;
2597         }
2598         return err ? err : len;
2599 }
2600
2601 static struct md_sysfs_entry md_size =
2602 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2603
2604
2605 /* Metdata version.
2606  * This is either 'none' for arrays with externally managed metadata,
2607  * or N.M for internally known formats
2608  */
2609 static ssize_t
2610 metadata_show(mddev_t *mddev, char *page)
2611 {
2612         if (mddev->persistent)
2613                 return sprintf(page, "%d.%d\n",
2614                                mddev->major_version, mddev->minor_version);
2615         else
2616                 return sprintf(page, "none\n");
2617 }
2618
2619 static ssize_t
2620 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2621 {
2622         int major, minor;
2623         char *e;
2624         if (!list_empty(&mddev->disks))
2625                 return -EBUSY;
2626
2627         if (cmd_match(buf, "none")) {
2628                 mddev->persistent = 0;
2629                 mddev->major_version = 0;
2630                 mddev->minor_version = 90;
2631                 return len;
2632         }
2633         major = simple_strtoul(buf, &e, 10);
2634         if (e==buf || *e != '.')
2635                 return -EINVAL;
2636         buf = e+1;
2637         minor = simple_strtoul(buf, &e, 10);
2638         if (e==buf || (*e && *e != '\n') )
2639                 return -EINVAL;
2640         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2641             super_types[major].name == NULL)
2642                 return -ENOENT;
2643         mddev->major_version = major;
2644         mddev->minor_version = minor;
2645         mddev->persistent = 1;
2646         return len;
2647 }
2648
2649 static struct md_sysfs_entry md_metadata =
2650 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2651
2652 static ssize_t
2653 action_show(mddev_t *mddev, char *page)
2654 {
2655         char *type = "idle";
2656         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2657             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2658                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2659                         type = "reshape";
2660                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2661                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2662                                 type = "resync";
2663                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2664                                 type = "check";
2665                         else
2666                                 type = "repair";
2667                 } else
2668                         type = "recover";
2669         }
2670         return sprintf(page, "%s\n", type);
2671 }
2672
2673 static ssize_t
2674 action_store(mddev_t *mddev, const char *page, size_t len)
2675 {
2676         if (!mddev->pers || !mddev->pers->sync_request)
2677                 return -EINVAL;
2678
2679         if (cmd_match(page, "idle")) {
2680                 if (mddev->sync_thread) {
2681                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2682                         md_unregister_thread(mddev->sync_thread);
2683                         mddev->sync_thread = NULL;
2684                         mddev->recovery = 0;
2685                 }
2686         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2687                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2688                 return -EBUSY;
2689         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2690                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2691         else if (cmd_match(page, "reshape")) {
2692                 int err;
2693                 if (mddev->pers->start_reshape == NULL)
2694                         return -EINVAL;
2695                 err = mddev->pers->start_reshape(mddev);
2696                 if (err)
2697                         return err;
2698         } else {
2699                 if (cmd_match(page, "check"))
2700                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2701                 else if (!cmd_match(page, "repair"))
2702                         return -EINVAL;
2703                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2704                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2705         }
2706         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2707         md_wakeup_thread(mddev->thread);
2708         return len;
2709 }
2710
2711 static ssize_t
2712 mismatch_cnt_show(mddev_t *mddev, char *page)
2713 {
2714         return sprintf(page, "%llu\n",
2715                        (unsigned long long) mddev->resync_mismatches);
2716 }
2717
2718 static struct md_sysfs_entry md_scan_mode =
2719 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2720
2721
2722 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2723
2724 static ssize_t
2725 sync_min_show(mddev_t *mddev, char *page)
2726 {
2727         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2728                        mddev->sync_speed_min ? "local": "system");
2729 }
2730
2731 static ssize_t
2732 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2733 {
2734         int min;
2735         char *e;
2736         if (strncmp(buf, "system", 6)==0) {
2737                 mddev->sync_speed_min = 0;
2738                 return len;
2739         }
2740         min = simple_strtoul(buf, &e, 10);
2741         if (buf == e || (*e && *e != '\n') || min <= 0)
2742                 return -EINVAL;
2743         mddev->sync_speed_min = min;
2744         return len;
2745 }
2746
2747 static struct md_sysfs_entry md_sync_min =
2748 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2749
2750 static ssize_t
2751 sync_max_show(mddev_t *mddev, char *page)
2752 {
2753         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2754                        mddev->sync_speed_max ? "local": "system");
2755 }
2756
2757 static ssize_t
2758 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2759 {
2760         int max;
2761         char *e;
2762         if (strncmp(buf, "system", 6)==0) {
2763                 mddev->sync_speed_max = 0;
2764                 return len;
2765         }
2766         max = simple_strtoul(buf, &e, 10);
2767         if (buf == e || (*e && *e != '\n') || max <= 0)
2768                 return -EINVAL;
2769         mddev->sync_speed_max = max;
2770         return len;
2771 }
2772
2773 static struct md_sysfs_entry md_sync_max =
2774 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2775
2776
2777 static ssize_t
2778 sync_speed_show(mddev_t *mddev, char *page)
2779 {
2780         unsigned long resync, dt, db;
2781         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2782         dt = ((jiffies - mddev->resync_mark) / HZ);
2783         if (!dt) dt++;
2784         db = resync - (mddev->resync_mark_cnt);
2785         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2786 }
2787
2788 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2789
2790 static ssize_t
2791 sync_completed_show(mddev_t *mddev, char *page)
2792 {
2793         unsigned long max_blocks, resync;
2794
2795         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2796                 max_blocks = mddev->resync_max_sectors;
2797         else
2798                 max_blocks = mddev->size << 1;
2799
2800         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2801         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2802 }
2803
2804 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2805
2806 static ssize_t
2807 suspend_lo_show(mddev_t *mddev, char *page)
2808 {
2809         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2810 }
2811
2812 static ssize_t
2813 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2814 {
2815         char *e;
2816         unsigned long long new = simple_strtoull(buf, &e, 10);
2817
2818         if (mddev->pers->quiesce == NULL)
2819                 return -EINVAL;
2820         if (buf == e || (*e && *e != '\n'))
2821                 return -EINVAL;
2822         if (new >= mddev->suspend_hi ||
2823             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2824                 mddev->suspend_lo = new;
2825                 mddev->pers->quiesce(mddev, 2);
2826                 return len;
2827         } else
2828                 return -EINVAL;
2829 }
2830 static struct md_sysfs_entry md_suspend_lo =
2831 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2832
2833
2834 static ssize_t
2835 suspend_hi_show(mddev_t *mddev, char *page)
2836 {
2837         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2838 }
2839
2840 static ssize_t
2841 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2842 {
2843         char *e;
2844         unsigned long long new = simple_strtoull(buf, &e, 10);
2845
2846         if (mddev->pers->quiesce == NULL)
2847                 return -EINVAL;
2848         if (buf == e || (*e && *e != '\n'))
2849                 return -EINVAL;
2850         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2851             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2852                 mddev->suspend_hi = new;
2853                 mddev->pers->quiesce(mddev, 1);
2854                 mddev->pers->quiesce(mddev, 0);
2855                 return len;
2856         } else
2857                 return -EINVAL;
2858 }
2859 static struct md_sysfs_entry md_suspend_hi =
2860 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2861
2862
2863 static struct attribute *md_default_attrs[] = {
2864         &md_level.attr,
2865         &md_layout.attr,
2866         &md_raid_disks.attr,
2867         &md_chunk_size.attr,
2868         &md_size.attr,
2869         &md_resync_start.attr,
2870         &md_metadata.attr,
2871         &md_new_device.attr,
2872         &md_safe_delay.attr,
2873         &md_array_state.attr,
2874         NULL,
2875 };
2876
2877 static struct attribute *md_redundancy_attrs[] = {
2878         &md_scan_mode.attr,
2879         &md_mismatches.attr,
2880         &md_sync_min.attr,
2881         &md_sync_max.attr,
2882         &md_sync_speed.attr,
2883         &md_sync_completed.attr,
2884         &md_suspend_lo.attr,
2885         &md_suspend_hi.attr,
2886         &md_bitmap.attr,
2887         NULL,
2888 };
2889 static struct attribute_group md_redundancy_group = {
2890         .name = NULL,
2891         .attrs = md_redundancy_attrs,
2892 };
2893
2894
2895 static ssize_t
2896 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2897 {
2898         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2899         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2900         ssize_t rv;
2901
2902         if (!entry->show)
2903                 return -EIO;
2904         rv = mddev_lock(mddev);
2905         if (!rv) {
2906                 rv = entry->show(mddev, page);
2907                 mddev_unlock(mddev);
2908         }
2909         return rv;
2910 }
2911
2912 static ssize_t
2913 md_attr_store(struct kobject *kobj, struct attribute *attr,
2914               const char *page, size_t length)
2915 {
2916         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2917         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2918         ssize_t rv;
2919
2920         if (!entry->store)
2921                 return -EIO;
2922         if (!capable(CAP_SYS_ADMIN))
2923                 return -EACCES;
2924         rv = mddev_lock(mddev);
2925         if (!rv) {
2926                 rv = entry->store(mddev, page, length);
2927                 mddev_unlock(mddev);
2928         }
2929         return rv;
2930 }
2931
2932 static void md_free(struct kobject *ko)
2933 {
2934         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2935         kfree(mddev);
2936 }
2937
2938 static struct sysfs_ops md_sysfs_ops = {
2939         .show   = md_attr_show,
2940         .store  = md_attr_store,
2941 };
2942 static struct kobj_type md_ktype = {
2943         .release        = md_free,
2944         .sysfs_ops      = &md_sysfs_ops,
2945         .default_attrs  = md_default_attrs,
2946 };
2947
2948 int mdp_major = 0;
2949
2950 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2951 {
2952         static DEFINE_MUTEX(disks_mutex);
2953         mddev_t *mddev = mddev_find(dev);
2954         struct gendisk *disk;
2955         int partitioned = (MAJOR(dev) != MD_MAJOR);
2956         int shift = partitioned ? MdpMinorShift : 0;
2957         int unit = MINOR(dev) >> shift;
2958
2959         if (!mddev)
2960                 return NULL;
2961
2962         mutex_lock(&disks_mutex);
2963         if (mddev->gendisk) {
2964                 mutex_unlock(&disks_mutex);
2965                 mddev_put(mddev);
2966                 return NULL;
2967         }
2968         disk = alloc_disk(1 << shift);
2969         if (!disk) {
2970                 mutex_unlock(&disks_mutex);
2971                 mddev_put(mddev);
2972                 return NULL;
2973         }
2974         disk->major = MAJOR(dev);
2975         disk->first_minor = unit << shift;
2976         if (partitioned)
2977                 sprintf(disk->disk_name, "md_d%d", unit);
2978         else
2979                 sprintf(disk->disk_name, "md%d", unit);
2980         disk->fops = &md_fops;
2981         disk->private_data = mddev;
2982         disk->queue = mddev->queue;
2983         add_disk(disk);
2984         mddev->gendisk = disk;
2985         mutex_unlock(&disks_mutex);
2986         mddev->kobj.parent = &disk->kobj;
2987         mddev->kobj.k_name = NULL;
2988         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2989         mddev->kobj.ktype = &md_ktype;
2990         if (kobject_register(&mddev->kobj))
2991                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
2992                        disk->disk_name);
2993         return NULL;
2994 }
2995
2996 static void md_safemode_timeout(unsigned long data)
2997 {
2998         mddev_t *mddev = (mddev_t *) data;
2999
3000         mddev->safemode = 1;
3001         md_wakeup_thread(mddev->thread);
3002 }
3003
3004 static int start_dirty_degraded;
3005
3006 static int do_md_run(mddev_t * mddev)
3007 {
3008         int err;
3009         int chunk_size;
3010         struct list_head *tmp;
3011         mdk_rdev_t *rdev;
3012         struct gendisk *disk;
3013         struct mdk_personality *pers;
3014         char b[BDEVNAME_SIZE];
3015
3016         if (list_empty(&mddev->disks))
3017                 /* cannot run an array with no devices.. */
3018                 return -EINVAL;
3019
3020         if (mddev->pers)
3021                 return -EBUSY;
3022
3023         /*
3024          * Analyze all RAID superblock(s)
3025          */
3026         if (!mddev->raid_disks)
3027                 analyze_sbs(mddev);
3028
3029         chunk_size = mddev->chunk_size;
3030
3031         if (chunk_size) {
3032                 if (chunk_size > MAX_CHUNK_SIZE) {
3033                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3034                                 chunk_size, MAX_CHUNK_SIZE);
3035                         return -EINVAL;
3036                 }
3037                 /*
3038                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3039                  */
3040                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3041                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3042                         return -EINVAL;
3043                 }
3044                 if (chunk_size < PAGE_SIZE) {
3045                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3046                                 chunk_size, PAGE_SIZE);
3047                         return -EINVAL;
3048                 }
3049
3050                 /* devices must have minimum size of one chunk */
3051                 ITERATE_RDEV(mddev,rdev,tmp) {
3052                         if (test_bit(Faulty, &rdev->flags))
3053                                 continue;
3054                         if (rdev->size < chunk_size / 1024) {
3055                                 printk(KERN_WARNING
3056                                         "md: Dev %s smaller than chunk_size:"
3057                                         " %lluk < %dk\n",
3058                                         bdevname(rdev->bdev,b),
3059                                         (unsigned long long)rdev->size,
3060                                         chunk_size / 1024);
3061                                 return -EINVAL;
3062                         }
3063                 }
3064         }
3065
3066 #ifdef CONFIG_KMOD
3067         if (mddev->level != LEVEL_NONE)
3068                 request_module("md-level-%d", mddev->level);
3069         else if (mddev->clevel[0])
3070                 request_module("md-%s", mddev->clevel);
3071 #endif
3072
3073         /*
3074          * Drop all container device buffers, from now on
3075          * the only valid external interface is through the md
3076          * device.
3077          * Also find largest hardsector size
3078          */
3079         ITERATE_RDEV(mddev,rdev,tmp) {
3080                 if (test_bit(Faulty, &rdev->flags))
3081                         continue;
3082                 sync_blockdev(rdev->bdev);
3083                 invalidate_bdev(rdev->bdev);
3084         }
3085
3086         md_probe(mddev->unit, NULL, NULL);
3087         disk = mddev->gendisk;
3088         if (!disk)
3089                 return -ENOMEM;
3090
3091         spin_lock(&pers_lock);
3092         pers = find_pers(mddev->level, mddev->clevel);
3093         if (!pers || !try_module_get(pers->owner)) {
3094                 spin_unlock(&pers_lock);
3095                 if (mddev->level != LEVEL_NONE)
3096                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3097                                mddev->level);
3098                 else
3099                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3100                                mddev->clevel);
3101                 return -EINVAL;
3102         }
3103         mddev->pers = pers;
3104         spin_unlock(&pers_lock);
3105         mddev->level = pers->level;
3106         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3107
3108         if (mddev->reshape_position != MaxSector &&
3109             pers->start_reshape == NULL) {
3110                 /* This personality cannot handle reshaping... */
3111                 mddev->pers = NULL;
3112                 module_put(pers->owner);
3113                 return -EINVAL;
3114         }
3115
3116         if (pers->sync_request) {
3117                 /* Warn if this is a potentially silly
3118                  * configuration.
3119                  */
3120                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3121                 mdk_rdev_t *rdev2;
3122                 struct list_head *tmp2;
3123                 int warned = 0;
3124                 ITERATE_RDEV(mddev, rdev, tmp) {
3125                         ITERATE_RDEV(mddev, rdev2, tmp2) {
3126                                 if (rdev < rdev2 &&
3127                                     rdev->bdev->bd_contains ==
3128                                     rdev2->bdev->bd_contains) {
3129                                         printk(KERN_WARNING
3130                                                "%s: WARNING: %s appears to be"
3131                                                " on the same physical disk as"
3132                                                " %s.\n",
3133                                                mdname(mddev),
3134                                                bdevname(rdev->bdev,b),
3135                                                bdevname(rdev2->bdev,b2));
3136                                         warned = 1;
3137                                 }
3138                         }
3139                 }
3140                 if (warned)
3141                         printk(KERN_WARNING
3142                                "True protection against single-disk"
3143                                " failure might be compromised.\n");
3144         }
3145
3146         mddev->recovery = 0;
3147         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3148         mddev->barriers_work = 1;
3149         mddev->ok_start_degraded = start_dirty_degraded;
3150
3151         if (start_readonly)
3152                 mddev->ro = 2; /* read-only, but switch on first write */
3153
3154         err = mddev->pers->run(mddev);
3155         if (!err && mddev->pers->sync_request) {
3156                 err = bitmap_create(mddev);
3157                 if (err) {
3158                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3159                                mdname(mddev), err);
3160                         mddev->pers->stop(mddev);
3161                 }
3162         }
3163         if (err) {
3164                 printk(KERN_ERR "md: pers->run() failed ...\n");
3165                 module_put(mddev->pers->owner);
3166                 mddev->pers = NULL;
3167                 bitmap_destroy(mddev);
3168                 return err;
3169         }
3170         if (mddev->pers->sync_request) {
3171                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3172                         printk(KERN_WARNING
3173                                "md: cannot register extra attributes for %s\n",
3174                                mdname(mddev));
3175         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3176                 mddev->ro = 0;
3177
3178         atomic_set(&mddev->writes_pending,0);
3179         mddev->safemode = 0;
3180         mddev->safemode_timer.function = md_safemode_timeout;
3181         mddev->safemode_timer.data = (unsigned long) mddev;
3182         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3183         mddev->in_sync = 1;
3184
3185         ITERATE_RDEV(mddev,rdev,tmp)
3186                 if (rdev->raid_disk >= 0) {
3187                         char nm[20];
3188                         sprintf(nm, "rd%d", rdev->raid_disk);
3189                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3190                                 printk("md: cannot register %s for %s\n",
3191                                        nm, mdname(mddev));
3192                 }
3193
3194         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3195
3196         if (mddev->flags)
3197                 md_update_sb(mddev, 0);
3198
3199         set_capacity(disk, mddev->array_size<<1);
3200
3201         /* If we call blk_queue_make_request here, it will
3202          * re-initialise max_sectors etc which may have been
3203          * refined inside -> run.  So just set the bits we need to set.
3204          * Most initialisation happended when we called
3205          * blk_queue_make_request(..., md_fail_request)
3206          * earlier.
3207          */
3208         mddev->queue->queuedata = mddev;
3209         mddev->queue->make_request_fn = mddev->pers->make_request;
3210
3211         /* If there is a partially-recovered drive we need to
3212          * start recovery here.  If we leave it to md_check_recovery,
3213          * it will remove the drives and not do the right thing
3214          */
3215         if (mddev->degraded && !mddev->sync_thread) {
3216                 struct list_head *rtmp;
3217                 int spares = 0;
3218                 ITERATE_RDEV(mddev,rdev,rtmp)
3219                         if (rdev->raid_disk >= 0 &&
3220                             !test_bit(In_sync, &rdev->flags) &&
3221                             !test_bit(Faulty, &rdev->flags))
3222                                 /* complete an interrupted recovery */
3223                                 spares++;
3224                 if (spares && mddev->pers->sync_request) {
3225                         mddev->recovery = 0;
3226                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3227                         mddev->sync_thread = md_register_thread(md_do_sync,
3228                                                                 mddev,
3229                                                                 "%s_resync");
3230                         if (!mddev->sync_thread) {
3231                                 printk(KERN_ERR "%s: could not start resync"
3232                                        " thread...\n",
3233                                        mdname(mddev));
3234                                 /* leave the spares where they are, it shouldn't hurt */
3235                                 mddev->recovery = 0;
3236                         }
3237                 }
3238         }
3239         md_wakeup_thread(mddev->thread);
3240         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3241
3242         mddev->changed = 1;
3243         md_new_event(mddev);
3244         kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3245         return 0;
3246 }
3247
3248 static int restart_array(mddev_t *mddev)
3249 {
3250         struct gendisk *disk = mddev->gendisk;
3251         int err;
3252
3253         /*
3254          * Complain if it has no devices
3255          */
3256         err = -ENXIO;
3257         if (list_empty(&mddev->disks))
3258                 goto out;
3259
3260         if (mddev->pers) {
3261                 err = -EBUSY;
3262                 if (!mddev->ro)
3263                         goto out;
3264
3265                 mddev->safemode = 0;
3266                 mddev->ro = 0;
3267                 set_disk_ro(disk, 0);
3268
3269                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3270                         mdname(mddev));
3271                 /*
3272                  * Kick recovery or resync if necessary
3273                  */
3274                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3275                 md_wakeup_thread(mddev->thread);
3276                 md_wakeup_thread(mddev->sync_thread);
3277                 err = 0;
3278         } else
3279                 err = -EINVAL;
3280
3281 out:
3282         return err;
3283 }
3284
3285 /* similar to deny_write_access, but accounts for our holding a reference
3286  * to the file ourselves */
3287 static int deny_bitmap_write_access(struct file * file)
3288 {
3289         struct inode *inode = file->f_mapping->host;
3290
3291         spin_lock(&inode->i_lock);
3292         if (atomic_read(&inode->i_writecount) > 1) {
3293                 spin_unlock(&inode->i_lock);
3294                 return -ETXTBSY;
3295         }
3296         atomic_set(&inode->i_writecount, -1);
3297         spin_unlock(&inode->i_lock);
3298
3299         return 0;
3300 }
3301
3302 static void restore_bitmap_write_access(struct file *file)
3303 {
3304         struct inode *inode = file->f_mapping->host;
3305
3306         spin_lock(&inode->i_lock);
3307         atomic_set(&inode->i_writecount, 1);
3308         spin_unlock(&inode->i_lock);
3309 }
3310
3311 /* mode:
3312  *   0 - completely stop and dis-assemble array
3313  *   1 - switch to readonly
3314  *   2 - stop but do not disassemble array
3315  */
3316 static int do_md_stop(mddev_t * mddev, int mode)
3317 {
3318         int err = 0;
3319         struct gendisk *disk = mddev->gendisk;
3320
3321         if (mddev->pers) {
3322                 if (atomic_read(&mddev->active)>2) {
3323                         printk("md: %s still in use.\n",mdname(mddev));
3324                         return -EBUSY;
3325                 }
3326
3327                 if (mddev->sync_thread) {
3328                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3329                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3330                         md_unregister_thread(mddev->sync_thread);
3331                         mddev->sync_thread = NULL;
3332                 }
3333
3334                 del_timer_sync(&mddev->safemode_timer);
3335
3336                 invalidate_partition(disk, 0);
3337
3338                 switch(mode) {
3339                 case 1: /* readonly */
3340                         err  = -ENXIO;
3341                         if (mddev->ro==1)
3342                                 goto out;
3343                         mddev->ro = 1;
3344                         break;
3345                 case 0: /* disassemble */
3346                 case 2: /* stop */
3347                         bitmap_flush(mddev);
3348                         md_super_wait(mddev);
3349                         if (mddev->ro)
3350                                 set_disk_ro(disk, 0);
3351                         blk_queue_make_request(mddev->queue, md_fail_request);
3352                         mddev->pers->stop(mddev);
3353                         mddev->queue->merge_bvec_fn = NULL;
3354                         mddev->queue->unplug_fn = NULL;
3355                         mddev->queue->issue_flush_fn = NULL;
3356                         mddev->queue->backing_dev_info.congested_fn = NULL;
3357                         if (mddev->pers->sync_request)
3358                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3359
3360                         module_put(mddev->pers->owner);
3361                         mddev->pers = NULL;
3362
3363                         set_capacity(disk, 0);
3364                         mddev->changed = 1;
3365
3366                         if (mddev->ro)
3367                                 mddev->ro = 0;
3368                 }
3369                 if (!mddev->in_sync || mddev->flags) {
3370                         /* mark array as shutdown cleanly */
3371                         mddev->in_sync = 1;
3372                         md_update_sb(mddev, 1);
3373                 }
3374                 if (mode == 1)
3375                         set_disk_ro(disk, 1);
3376                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3377         }
3378
3379         /*
3380          * Free resources if final stop
3381          */
3382         if (mode == 0) {
3383                 mdk_rdev_t *rdev;
3384                 struct list_head *tmp;
3385
3386                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3387
3388                 bitmap_destroy(mddev);
3389                 if (mddev->bitmap_file) {
3390                         restore_bitmap_write_access(mddev->bitmap_file);
3391                         fput(mddev->bitmap_file);
3392                         mddev->bitmap_file = NULL;
3393                 }
3394                 mddev->bitmap_offset = 0;
3395
3396                 ITERATE_RDEV(mddev,rdev,tmp)
3397                         if (rdev->raid_disk >= 0) {
3398                                 char nm[20];
3399                                 sprintf(nm, "rd%d", rdev->raid_disk);
3400                                 sysfs_remove_link(&mddev->kobj, nm);
3401                         }
3402
3403                 /* make sure all delayed_delete calls have finished */
3404                 flush_scheduled_work();
3405
3406                 export_array(mddev);
3407
3408                 mddev->array_size = 0;
3409                 mddev->size = 0;
3410                 mddev->raid_disks = 0;
3411                 mddev->recovery_cp = 0;
3412
3413         } else if (mddev->pers)
3414                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3415                         mdname(mddev));
3416         err = 0;
3417         md_new_event(mddev);
3418 out:
3419         return err;
3420 }
3421
3422 #ifndef MODULE
3423 static void autorun_array(mddev_t *mddev)
3424 {
3425         mdk_rdev_t *rdev;
3426         struct list_head *tmp;
3427         int err;
3428
3429         if (list_empty(&mddev->disks))
3430                 return;
3431
3432         printk(KERN_INFO "md: running: ");
3433
3434         ITERATE_RDEV(mddev,rdev,tmp) {
3435                 char b[BDEVNAME_SIZE];
3436                 printk("<%s>", bdevname(rdev->bdev,b));
3437         }
3438         printk("\n");
3439
3440         err = do_md_run (mddev);
3441         if (err) {
3442                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3443                 do_md_stop (mddev, 0);
3444         }
3445 }
3446
3447 /*
3448  * lets try to run arrays based on all disks that have arrived
3449  * until now. (those are in pending_raid_disks)
3450  *
3451  * the method: pick the first pending disk, collect all disks with
3452  * the same UUID, remove all from the pending list and put them into
3453  * the 'same_array' list. Then order this list based on superblock
3454  * update time (freshest comes first), kick out 'old' disks and
3455  * compare superblocks. If everything's fine then run it.
3456  *
3457  * If "unit" is allocated, then bump its reference count
3458  */
3459 static void autorun_devices(int part)
3460 {
3461         struct list_head *tmp;
3462         mdk_rdev_t *rdev0, *rdev;
3463         mddev_t *mddev;
3464         char b[BDEVNAME_SIZE];
3465
3466         printk(KERN_INFO "md: autorun ...\n");
3467         while (!list_empty(&pending_raid_disks)) {
3468                 int unit;
3469                 dev_t dev;
3470                 LIST_HEAD(candidates);
3471                 rdev0 = list_entry(pending_raid_disks.next,
3472                                          mdk_rdev_t, same_set);
3473
3474                 printk(KERN_INFO "md: considering %s ...\n",
3475                         bdevname(rdev0->bdev,b));
3476                 INIT_LIST_HEAD(&candidates);
3477                 ITERATE_RDEV_PENDING(rdev,tmp)
3478                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3479                                 printk(KERN_INFO "md:  adding %s ...\n",
3480                                         bdevname(rdev->bdev,b));
3481                                 list_move(&rdev->same_set, &candidates);
3482                         }
3483                 /*
3484                  * now we have a set of devices, with all of them having
3485                  * mostly sane superblocks. It's time to allocate the
3486                  * mddev.
3487                  */
3488                 if (part) {
3489                         dev = MKDEV(mdp_major,
3490                                     rdev0->preferred_minor << MdpMinorShift);
3491                         unit = MINOR(dev) >> MdpMinorShift;
3492                 } else {
3493                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3494                         unit = MINOR(dev);
3495                 }
3496                 if (rdev0->preferred_minor != unit) {
3497                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3498                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3499                         break;
3500                 }
3501
3502                 md_probe(dev, NULL, NULL);
3503                 mddev = mddev_find(dev);
3504                 if (!mddev) {
3505                         printk(KERN_ERR
3506                                 "md: cannot allocate memory for md drive.\n");
3507                         break;
3508                 }
3509                 if (mddev_lock(mddev))
3510                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3511                                mdname(mddev));
3512                 else if (mddev->raid_disks || mddev->major_version
3513                          || !list_empty(&mddev->disks)) {
3514                         printk(KERN_WARNING
3515                                 "md: %s already running, cannot run %s\n",
3516                                 mdname(mddev), bdevname(rdev0->bdev,b));
3517                         mddev_unlock(mddev);
3518                 } else {
3519                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3520                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3521                                 list_del_init(&rdev->same_set);
3522                                 if (bind_rdev_to_array(rdev, mddev))
3523                                         export_rdev(rdev);
3524                         }
3525                         autorun_array(mddev);
3526                         mddev_unlock(mddev);
3527                 }
3528                 /* on success, candidates will be empty, on error
3529                  * it won't...
3530                  */
3531                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3532                         export_rdev(rdev);
3533                 mddev_put(mddev);
3534         }
3535         printk(KERN_INFO "md: ... autorun DONE.\n");
3536 }
3537 #endif /* !MODULE */
3538
3539 static int get_version(void __user * arg)
3540 {
3541         mdu_version_t ver;
3542
3543         ver.major = MD_MAJOR_VERSION;
3544         ver.minor = MD_MINOR_VERSION;
3545         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3546
3547         if (copy_to_user(arg, &ver, sizeof(ver)))
3548                 return -EFAULT;
3549
3550         return 0;
3551 }
3552
3553 static int get_array_info(mddev_t * mddev, void __user * arg)
3554 {
3555         mdu_array_info_t info;
3556         int nr,working,active,failed,spare;
3557         mdk_rdev_t *rdev;
3558         struct list_head *tmp;
3559
3560         nr=working=active=failed=spare=0;
3561         ITERATE_RDEV(mddev,rdev,tmp) {
3562                 nr++;
3563                 if (test_bit(Faulty, &rdev->flags))
3564                         failed++;
3565                 else {
3566                         working++;
3567                         if (test_bit(In_sync, &rdev->flags))
3568                                 active++;
3569                         else
3570                                 spare++;
3571                 }
3572         }
3573
3574         info.major_version = mddev->major_version;
3575         info.minor_version = mddev->minor_version;
3576         info.patch_version = MD_PATCHLEVEL_VERSION;
3577         info.ctime         = mddev->ctime;
3578         info.level         = mddev->level;
3579         info.size          = mddev->size;
3580         if (info.size != mddev->size) /* overflow */
3581                 info.size = -1;
3582         info.nr_disks      = nr;
3583         info.raid_disks    = mddev->raid_disks;
3584         info.md_minor      = mddev->md_minor;
3585         info.not_persistent= !mddev->persistent;
3586
3587         info.utime         = mddev->utime;
3588         info.state         = 0;
3589         if (mddev->in_sync)
3590                 info.state = (1<<MD_SB_CLEAN);
3591         if (mddev->bitmap && mddev->bitmap_offset)
3592                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3593         info.active_disks  = active;
3594         info.working_disks = working;
3595         info.failed_disks  = failed;
3596         info.spare_disks   = spare;
3597
3598         info.layout        = mddev->layout;
3599         info.chunk_size    = mddev->chunk_size;
3600
3601         if (copy_to_user(arg, &info, sizeof(info)))
3602                 return -EFAULT;
3603
3604         return 0;
3605 }
3606
3607 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3608 {
3609         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3610         char *ptr, *buf = NULL;
3611         int err = -ENOMEM;
3612
3613         md_allow_write(mddev);
3614
3615         file = kmalloc(sizeof(*file), GFP_KERNEL);
3616         if (!file)
3617                 goto out;
3618
3619         /* bitmap disabled, zero the first byte and copy out */
3620         if (!mddev->bitmap || !mddev->bitmap->file) {
3621                 file->pathname[0] = '\0';
3622                 goto copy_out;
3623         }
3624
3625         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3626         if (!buf)
3627                 goto out;
3628
3629         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3630         if (!ptr)
3631                 goto out;
3632
3633         strcpy(file->pathname, ptr);
3634
3635 copy_out:
3636         err = 0;
3637         if (copy_to_user(arg, file, sizeof(*file)))
3638                 err = -EFAULT;
3639 out:
3640         kfree(buf);
3641         kfree(file);
3642         return err;
3643 }
3644
3645 static int get_disk_info(mddev_t * mddev, void __user * arg)
3646 {
3647         mdu_disk_info_t info;
3648         unsigned int nr;
3649         mdk_rdev_t *rdev;
3650
3651         if (copy_from_user(&info, arg, sizeof(info)))
3652                 return -EFAULT;
3653
3654         nr = info.number;
3655
3656         rdev = find_rdev_nr(mddev, nr);
3657         if (rdev) {
3658                 info.major = MAJOR(rdev->bdev->bd_dev);
3659                 info.minor = MINOR(rdev->bdev->bd_dev);
3660                 info.raid_disk = rdev->raid_disk;
3661                 info.state = 0;
3662                 if (test_bit(Faulty, &rdev->flags))
3663                         info.state |= (1<<MD_DISK_FAULTY);
3664                 else if (test_bit(In_sync, &rdev->flags)) {
3665                         info.state |= (1<<MD_DISK_ACTIVE);
3666                         info.state |= (1<<MD_DISK_SYNC);
3667                 }
3668                 if (test_bit(WriteMostly, &rdev->flags))
3669                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3670         } else {
3671                 info.major = info.minor = 0;
3672                 info.raid_disk = -1;
3673                 info.state = (1<<MD_DISK_REMOVED);
3674         }
3675
3676         if (copy_to_user(arg, &info, sizeof(info)))
3677                 return -EFAULT;
3678
3679         return 0;
3680 }
3681
3682 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3683 {
3684         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3685         mdk_rdev_t *rdev;
3686         dev_t dev = MKDEV(info->major,info->minor);
3687
3688         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3689                 return -EOVERFLOW;
3690
3691         if (!mddev->raid_disks) {
3692                 int err;
3693                 /* expecting a device which has a superblock */
3694                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3695                 if (IS_ERR(rdev)) {
3696                         printk(KERN_WARNING
3697                                 "md: md_import_device returned %ld\n",
3698                                 PTR_ERR(rdev));
3699                         return PTR_ERR(rdev);
3700                 }
3701                 if (!list_empty(&mddev->disks)) {
3702                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3703                                                         mdk_rdev_t, same_set);
3704                         int err = super_types[mddev->major_version]
3705                                 .load_super(rdev, rdev0, mddev->minor_version);
3706                         if (err < 0) {
3707                                 printk(KERN_WARNING
3708                                         "md: %s has different UUID to %s\n",
3709                                         bdevname(rdev->bdev,b),
3710                                         bdevname(rdev0->bdev,b2));
3711                                 export_rdev(rdev);
3712                                 return -EINVAL;
3713                         }
3714                 }
3715                 err = bind_rdev_to_array(rdev, mddev);
3716                 if (err)
3717                         export_rdev(rdev);
3718                 return err;
3719         }
3720
3721         /*
3722          * add_new_disk can be used once the array is assembled
3723          * to add "hot spares".  They must already have a superblock
3724          * written
3725          */
3726         if (mddev->pers) {
3727                 int err;
3728                 if (!mddev->pers->hot_add_disk) {
3729                         printk(KERN_WARNING
3730                                 "%s: personality does not support diskops!\n",
3731                                mdname(mddev));
3732                         return -EINVAL;
3733                 }
3734                 if (mddev->persistent)
3735                         rdev = md_import_device(dev, mddev->major_version,
3736                                                 mddev->minor_version);
3737                 else
3738                         rdev = md_import_device(dev, -1, -1);
3739                 if (IS_ERR(rdev)) {
3740                         printk(KERN_WARNING
3741                                 "md: md_import_device returned %ld\n",
3742                                 PTR_ERR(rdev));
3743                         return PTR_ERR(rdev);
3744                 }
3745                 /* set save_raid_disk if appropriate */
3746                 if (!mddev->persistent) {
3747                         if (info->state & (1<<MD_DISK_SYNC)  &&
3748                             info->raid_disk < mddev->raid_disks)
3749                                 rdev->raid_disk = info->raid_disk;
3750                         else
3751                                 rdev->raid_disk = -1;
3752                 } else
3753                         super_types[mddev->major_version].
3754                                 validate_super(mddev, rdev);
3755                 rdev->saved_raid_disk = rdev->raid_disk;
3756
3757                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3758                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3759                         set_bit(WriteMostly, &rdev->flags);
3760
3761                 rdev->raid_disk = -1;
3762                 err = bind_rdev_to_array(rdev, mddev);
3763                 if (!err && !mddev->pers->hot_remove_disk) {
3764                         /* If there is hot_add_disk but no hot_remove_disk
3765                          * then added disks for geometry changes,
3766                          * and should be added immediately.
3767                          */
3768                         super_types[mddev->major_version].
3769                                 validate_super(mddev, rdev);
3770                         err = mddev->pers->hot_add_disk(mddev, rdev);
3771                         if (err)
3772                                 unbind_rdev_from_array(rdev);
3773                 }
3774                 if (err)
3775                         export_rdev(rdev);
3776
3777                 md_update_sb(mddev, 1);
3778                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3779                 md_wakeup_thread(mddev->thread);
3780                 return err;
3781         }
3782
3783         /* otherwise, add_new_disk is only allowed
3784          * for major_version==0 superblocks
3785          */
3786         if (mddev->major_version != 0) {
3787                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3788                        mdname(mddev));
3789                 return -EINVAL;
3790         }
3791
3792         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3793                 int err;
3794                 rdev = md_import_device (dev, -1, 0);
3795                 if (IS_ERR(rdev)) {
3796                         printk(KERN_WARNING
3797                                 "md: error, md_import_device() returned %ld\n",
3798                                 PTR_ERR(rdev));
3799                         return PTR_ERR(rdev);
3800                 }
3801                 rdev->desc_nr = info->number;
3802                 if (info->raid_disk < mddev->raid_disks)
3803                         rdev->raid_disk = info->raid_disk;
3804                 else
3805                         rdev->raid_disk = -1;
3806
3807                 rdev->flags = 0;
3808
3809                 if (rdev->raid_disk < mddev->raid_disks)
3810                         if (info->state & (1<<MD_DISK_SYNC))
3811                                 set_bit(In_sync, &rdev->flags);
3812
3813                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3814                         set_bit(WriteMostly, &rdev->flags);
3815
3816                 if (!mddev->persistent) {
3817                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3818                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3819                 } else
3820                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3821                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3822
3823                 err = bind_rdev_to_array(rdev, mddev);
3824                 if (err) {
3825                         export_rdev(rdev);
3826                         return err;
3827                 }
3828         }
3829
3830         return 0;
3831 }
3832
3833 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3834 {
3835         char b[BDEVNAME_SIZE];
3836         mdk_rdev_t *rdev;
3837
3838         if (!mddev->pers)
3839                 return -ENODEV;
3840
3841         rdev = find_rdev(mddev, dev);
3842         if (!rdev)
3843                 return -ENXIO;
3844
3845         if (rdev->raid_disk >= 0)
3846                 goto busy;
3847
3848         kick_rdev_from_array(rdev);
3849         md_update_sb(mddev, 1);
3850         md_new_event(mddev);
3851
3852         return 0;
3853 busy:
3854         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3855                 bdevname(rdev->bdev,b), mdname(mddev));
3856         return -EBUSY;
3857 }
3858
3859 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3860 {
3861         char b[BDEVNAME_SIZE];
3862         int err;
3863         unsigned int size;
3864         mdk_rdev_t *rdev;
3865
3866         if (!mddev->pers)
3867                 return -ENODEV;
3868
3869         if (mddev->major_version != 0) {
3870                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3871                         " version-0 superblocks.\n",
3872                         mdname(mddev));
3873                 return -EINVAL;
3874         }
3875         if (!mddev->pers->hot_add_disk) {
3876                 printk(KERN_WARNING
3877                         "%s: personality does not support diskops!\n",
3878                         mdname(mddev));
3879                 return -EINVAL;
3880         }
3881
3882         rdev = md_import_device (dev, -1, 0);
3883         if (IS_ERR(rdev)) {
3884                 printk(KERN_WARNING
3885                         "md: error, md_import_device() returned %ld\n",
3886                         PTR_ERR(rdev));
3887                 return -EINVAL;
3888         }
3889
3890         if (mddev->persistent)
3891                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3892         else
3893                 rdev->sb_offset =
3894                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3895
3896         size = calc_dev_size(rdev, mddev->chunk_size);
3897         rdev->size = size;
3898
3899         if (test_bit(Faulty, &rdev->flags)) {
3900                 printk(KERN_WARNING
3901                         "md: can not hot-add faulty %s disk to %s!\n",
3902                         bdevname(rdev->bdev,b), mdname(mddev));
3903                 err = -EINVAL;
3904                 goto abort_export;
3905         }
3906         clear_bit(In_sync, &rdev->flags);
3907         rdev->desc_nr = -1;
3908         rdev->saved_raid_disk = -1;
3909         err = bind_rdev_to_array(rdev, mddev);
3910         if (err)
3911                 goto abort_export;
3912
3913         /*
3914          * The rest should better be atomic, we can have disk failures
3915          * noticed in interrupt contexts ...
3916          */
3917
3918         if (rdev->desc_nr == mddev->max_disks) {
3919                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3920                         mdname(mddev));
3921                 err = -EBUSY;
3922                 goto abort_unbind_export;
3923         }
3924
3925         rdev->raid_disk = -1;
3926
3927         md_update_sb(mddev, 1);
3928
3929         /*
3930          * Kick recovery, maybe this spare has to be added to the
3931          * array immediately.
3932          */
3933         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3934         md_wakeup_thread(mddev->thread);
3935         md_new_event(mddev);
3936         return 0;
3937
3938 abort_unbind_export:
3939         unbind_rdev_from_array(rdev);
3940
3941 abort_export:
3942         export_rdev(rdev);
3943         return err;
3944 }
3945
3946 static int set_bitmap_file(mddev_t *mddev, int fd)
3947 {
3948         int err;
3949
3950         if (mddev->pers) {
3951                 if (!mddev->pers->quiesce)
3952                         return -EBUSY;
3953                 if (mddev->recovery || mddev->sync_thread)
3954                         return -EBUSY;
3955                 /* we should be able to change the bitmap.. */
3956         }
3957
3958
3959         if (fd >= 0) {
3960                 if (mddev->bitmap)
3961                         return -EEXIST; /* cannot add when bitmap is present */
3962                 mddev->bitmap_file = fget(fd);
3963
3964                 if (mddev->bitmap_file == NULL) {
3965                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3966                                mdname(mddev));
3967                         return -EBADF;
3968                 }
3969
3970                 err = deny_bitmap_write_access(mddev->bitmap_file);
3971                 if (err) {
3972                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3973                                mdname(mddev));
3974                         fput(mddev->bitmap_file);
3975                         mddev->bitmap_file = NULL;
3976                         return err;
3977                 }
3978                 mddev->bitmap_offset = 0; /* file overrides offset */
3979         } else if (mddev->bitmap == NULL)
3980                 return -ENOENT; /* cannot remove what isn't there */
3981         err = 0;
3982         if (mddev->pers) {
3983                 mddev->pers->quiesce(mddev, 1);
3984                 if (fd >= 0)
3985                         err = bitmap_create(mddev);
3986                 if (fd < 0 || err) {
3987                         bitmap_destroy(mddev);
3988                         fd = -1; /* make sure to put the file */
3989                 }
3990                 mddev->pers->quiesce(mddev, 0);
3991         }
3992         if (fd < 0) {
3993                 if (mddev->bitmap_file) {
3994                         restore_bitmap_write_access(mddev->bitmap_file);
3995                         fput(mddev->bitmap_file);
3996                 }
3997                 mddev->bitmap_file = NULL;
3998         }
3999
4000         return err;
4001 }
4002
4003 /*
4004  * set_array_info is used two different ways
4005  * The original usage is when creating a new array.
4006  * In this usage, raid_disks is > 0 and it together with
4007  *  level, size, not_persistent,layout,chunksize determine the
4008  *  shape of the array.
4009  *  This will always create an array with a type-0.90.0 superblock.
4010  * The newer usage is when assembling an array.
4011  *  In this case raid_disks will be 0, and the major_version field is
4012  *  use to determine which style super-blocks are to be found on the devices.
4013  *  The minor and patch _version numbers are also kept incase the
4014  *  super_block handler wishes to interpret them.
4015  */
4016 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4017 {
4018
4019         if (info->raid_disks == 0) {
4020                 /* just setting version number for superblock loading */
4021                 if (info->major_version < 0 ||
4022                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
4023                     super_types[info->major_version].name == NULL) {
4024                         /* maybe try to auto-load a module? */
4025                         printk(KERN_INFO
4026                                 "md: superblock version %d not known\n",
4027                                 info->major_version);
4028                         return -EINVAL;
4029                 }
4030                 mddev->major_version = info->major_version;
4031                 mddev->minor_version = info->minor_version;
4032                 mddev->patch_version = info->patch_version;
4033                 mddev->persistent = !info->not_persistent;
4034                 return 0;
4035         }
4036         mddev->major_version = MD_MAJOR_VERSION;
4037         mddev->minor_version = MD_MINOR_VERSION;
4038         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4039         mddev->ctime         = get_seconds();
4040
4041         mddev->level         = info->level;
4042         mddev->clevel[0]     = 0;
4043         mddev->size          = info->size;
4044         mddev->raid_disks    = info->raid_disks;
4045         /* don't set md_minor, it is determined by which /dev/md* was
4046          * openned
4047          */
4048         if (info->state & (1<<MD_SB_CLEAN))
4049                 mddev->recovery_cp = MaxSector;
4050         else
4051                 mddev->recovery_cp = 0;
4052         mddev->persistent    = ! info->not_persistent;
4053
4054         mddev->layout        = info->layout;
4055         mddev->chunk_size    = info->chunk_size;
4056
4057         mddev->max_disks     = MD_SB_DISKS;
4058
4059         mddev->flags         = 0;
4060         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4061
4062         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4063         mddev->bitmap_offset = 0;
4064
4065         mddev->reshape_position = MaxSector;
4066
4067         /*
4068          * Generate a 128 bit UUID
4069          */
4070         get_random_bytes(mddev->uuid, 16);
4071
4072         mddev->new_level = mddev->level;
4073         mddev->new_chunk = mddev->chunk_size;
4074         mddev->new_layout = mddev->layout;
4075         mddev->delta_disks = 0;
4076
4077         return 0;
4078 }
4079
4080 static int update_size(mddev_t *mddev, unsigned long size)
4081 {
4082         mdk_rdev_t * rdev;
4083         int rv;
4084         struct list_head *tmp;
4085         int fit = (size == 0);
4086
4087         if (mddev->pers->resize == NULL)
4088                 return -EINVAL;
4089         /* The "size" is the amount of each device that is used.
4090          * This can only make sense for arrays with redundancy.
4091          * linear and raid0 always use whatever space is available
4092          * We can only consider changing the size if no resync
4093          * or reconstruction is happening, and if the new size
4094          * is acceptable. It must fit before the sb_offset or,
4095          * if that is <data_offset, it must fit before the
4096          * size of each device.
4097          * If size is zero, we find the largest size that fits.
4098          */
4099         if (mddev->sync_thread)
4100                 return -EBUSY;
4101         ITERATE_RDEV(mddev,rdev,tmp) {
4102                 sector_t avail;
4103                 avail = rdev->size * 2;
4104
4105                 if (fit && (size == 0 || size > avail/2))
4106                         size = avail/2;
4107                 if (avail < ((sector_t)size << 1))
4108                         return -ENOSPC;
4109         }
4110         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4111         if (!rv) {
4112                 struct block_device *bdev;
4113
4114                 bdev = bdget_disk(mddev->gendisk, 0);
4115                 if (bdev) {
4116                         mutex_lock(&bdev->bd_inode->i_mutex);
4117                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4118                         mutex_unlock(&bdev->bd_inode->i_mutex);
4119                         bdput(bdev);
4120                 }
4121         }
4122         return rv;
4123 }
4124
4125 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4126 {
4127         int rv;
4128         /* change the number of raid disks */
4129         if (mddev->pers->check_reshape == NULL)
4130                 return -EINVAL;
4131         if (raid_disks <= 0 ||
4132             raid_disks >= mddev->max_disks)
4133                 return -EINVAL;
4134         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4135                 return -EBUSY;
4136         mddev->delta_disks = raid_disks - mddev->raid_disks;
4137
4138         rv = mddev->pers->check_reshape(mddev);
4139         return rv;
4140 }
4141
4142
4143 /*
4144  * update_array_info is used to change the configuration of an
4145  * on-line array.
4146  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4147  * fields in the info are checked against the array.
4148  * Any differences that cannot be handled will cause an error.
4149  * Normally, only one change can be managed at a time.
4150  */
4151 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4152 {
4153         int rv = 0;
4154         int cnt = 0;
4155         int state = 0;
4156
4157         /* calculate expected state,ignoring low bits */
4158         if (mddev->bitmap && mddev->bitmap_offset)
4159                 state |= (1 << MD_SB_BITMAP_PRESENT);
4160
4161         if (mddev->major_version != info->major_version ||
4162             mddev->minor_version != info->minor_version ||
4163 /*          mddev->patch_version != info->patch_version || */
4164             mddev->ctime         != info->ctime         ||
4165             mddev->level         != info->level         ||
4166 /*          mddev->layout        != info->layout        || */
4167             !mddev->persistent   != info->not_persistent||
4168             mddev->chunk_size    != info->chunk_size    ||
4169             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4170             ((state^info->state) & 0xfffffe00)
4171                 )
4172                 return -EINVAL;
4173         /* Check there is only one change */
4174         if (info->size >= 0 && mddev->size != info->size) cnt++;
4175         if (mddev->raid_disks != info->raid_disks) cnt++;
4176         if (mddev->layout != info->layout) cnt++;
4177         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4178         if (cnt == 0) return 0;
4179         if (cnt > 1) return -EINVAL;
4180
4181         if (mddev->layout != info->layout) {
4182                 /* Change layout
4183                  * we don't need to do anything at the md level, the
4184                  * personality will take care of it all.
4185                  */
4186                 if (mddev->pers->reconfig == NULL)
4187                         return -EINVAL;
4188                 else
4189                         return mddev->pers->reconfig(mddev, info->layout, -1);
4190         }
4191         if (info->size >= 0 && mddev->size != info->size)
4192                 rv = update_size(mddev, info->size);
4193
4194         if (mddev->raid_disks    != info->raid_disks)
4195                 rv = update_raid_disks(mddev, info->raid_disks);
4196
4197         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4198                 if (mddev->pers->quiesce == NULL)
4199                         return -EINVAL;
4200                 if (mddev->recovery || mddev->sync_thread)
4201                         return -EBUSY;
4202                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4203                         /* add the bitmap */
4204                         if (mddev->bitmap)
4205                                 return -EEXIST;
4206                         if (mddev->default_bitmap_offset == 0)
4207                                 return -EINVAL;
4208                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4209                         mddev->pers->quiesce(mddev, 1);
4210                         rv = bitmap_create(mddev);
4211                         if (rv)
4212                                 bitmap_destroy(mddev);
4213                         mddev->pers->quiesce(mddev, 0);
4214                 } else {
4215                         /* remove the bitmap */
4216                         if (!mddev->bitmap)
4217                                 return -ENOENT;
4218                         if (mddev->bitmap->file)
4219                                 return -EINVAL;
4220                         mddev->pers->quiesce(mddev, 1);
4221                         bitmap_destroy(mddev);
4222                         mddev->pers->quiesce(mddev, 0);
4223                         mddev->bitmap_offset = 0;
4224                 }
4225         }
4226         md_update_sb(mddev, 1);
4227         return rv;
4228 }
4229
4230 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4231 {
4232         mdk_rdev_t *rdev;
4233
4234         if (mddev->pers == NULL)
4235                 return -ENODEV;
4236
4237         rdev = find_rdev(mddev, dev);
4238         if (!rdev)
4239                 return -ENODEV;
4240
4241         md_error(mddev, rdev);
4242         return 0;
4243 }
4244
4245 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4246 {
4247         mddev_t *mddev = bdev->bd_disk->private_data;
4248
4249         geo->heads = 2;
4250         geo->sectors = 4;
4251         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4252         return 0;
4253 }
4254
4255 static int md_ioctl(struct inode *inode, struct file *file,
4256                         unsigned int cmd, unsigned long arg)
4257 {
4258         int err = 0;
4259         void __user *argp = (void __user *)arg;
4260         mddev_t *mddev = NULL;
4261
4262         if (!capable(CAP_SYS_ADMIN))
4263                 return -EACCES;
4264
4265         /*
4266          * Commands dealing with the RAID driver but not any
4267          * particular array:
4268          */
4269         switch (cmd)
4270         {
4271                 case RAID_VERSION:
4272                         err = get_version(argp);
4273                         goto done;
4274
4275                 case PRINT_RAID_DEBUG:
4276                         err = 0;
4277                         md_print_devices();
4278                         goto done;
4279
4280 #ifndef MODULE
4281                 case RAID_AUTORUN:
4282                         err = 0;
4283                         autostart_arrays(arg);
4284                         goto done;
4285 #endif
4286                 default:;
4287         }
4288
4289         /*
4290          * Commands creating/starting a new array:
4291          */
4292
4293         mddev = inode->i_bdev->bd_disk->private_data;
4294
4295         if (!mddev) {
4296                 BUG();
4297                 goto abort;
4298         }
4299
4300         err = mddev_lock(mddev);
4301         if (err) {
4302                 printk(KERN_INFO
4303                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4304                         err, cmd);
4305                 goto abort;
4306         }
4307
4308         switch (cmd)
4309         {
4310                 case SET_ARRAY_INFO:
4311                         {
4312                                 mdu_array_info_t info;
4313                                 if (!arg)
4314                                         memset(&info, 0, sizeof(info));
4315                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4316                                         err = -EFAULT;
4317                                         goto abort_unlock;
4318                                 }
4319                                 if (mddev->pers) {
4320                                         err = update_array_info(mddev, &info);
4321                                         if (err) {
4322                                                 printk(KERN_WARNING "md: couldn't update"
4323                                                        " array info. %d\n", err);
4324                                                 goto abort_unlock;
4325                                         }
4326                                         goto done_unlock;
4327                                 }
4328                                 if (!list_empty(&mddev->disks)) {
4329                                         printk(KERN_WARNING
4330                                                "md: array %s already has disks!\n",
4331                                                mdname(mddev));
4332                                         err = -EBUSY;
4333                                         goto abort_unlock;
4334                                 }
4335                                 if (mddev->raid_disks) {
4336                                         printk(KERN_WARNING
4337                                                "md: array %s already initialised!\n",
4338                                                mdname(mddev));
4339                                         err = -EBUSY;
4340                                         goto abort_unlock;
4341                                 }
4342                                 err = set_array_info(mddev, &info);
4343                                 if (err) {
4344                                         printk(KERN_WARNING "md: couldn't set"
4345                                                " array info. %d\n", err);
4346                                         goto abort_unlock;
4347                                 }
4348                         }
4349                         goto done_unlock;
4350
4351                 default:;
4352         }
4353
4354         /*
4355          * Commands querying/configuring an existing array:
4356          */
4357         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4358          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4359         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4360                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4361                         && cmd != GET_BITMAP_FILE) {
4362                 err = -ENODEV;
4363                 goto abort_unlock;
4364         }
4365
4366         /*
4367          * Commands even a read-only array can execute:
4368          */
4369         switch (cmd)
4370         {
4371                 case GET_ARRAY_INFO:
4372                         err = get_array_info(mddev, argp);
4373                         goto done_unlock;
4374
4375                 case GET_BITMAP_FILE:
4376                         err = get_bitmap_file(mddev, argp);
4377                         goto done_unlock;
4378
4379                 case GET_DISK_INFO:
4380                         err = get_disk_info(mddev, argp);
4381                         goto done_unlock;
4382
4383                 case RESTART_ARRAY_RW:
4384                         err = restart_array(mddev);
4385                         goto done_unlock;
4386
4387                 case STOP_ARRAY:
4388                         err = do_md_stop (mddev, 0);
4389                         goto done_unlock;
4390
4391                 case STOP_ARRAY_RO:
4392                         err = do_md_stop (mddev, 1);
4393                         goto done_unlock;
4394
4395         /*
4396          * We have a problem here : there is no easy way to give a CHS
4397          * virtual geometry. We currently pretend that we have a 2 heads
4398          * 4 sectors (with a BIG number of cylinders...). This drives
4399          * dosfs just mad... ;-)
4400          */
4401         }
4402
4403         /*
4404          * The remaining ioctls are changing the state of the
4405          * superblock, so we do not allow them on read-only arrays.
4406          * However non-MD ioctls (e.g. get-size) will still come through
4407          * here and hit the 'default' below, so only disallow
4408          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4409          */
4410         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4411             mddev->ro && mddev->pers) {
4412                 if (mddev->ro == 2) {
4413                         mddev->ro = 0;
4414                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4415                 md_wakeup_thread(mddev->thread);
4416
4417                 } else {
4418                         err = -EROFS;
4419                         goto abort_unlock;
4420                 }
4421         }
4422
4423         switch (cmd)
4424         {
4425                 case ADD_NEW_DISK:
4426                 {
4427                         mdu_disk_info_t info;
4428                         if (copy_from_user(&info, argp, sizeof(info)))
4429                                 err = -EFAULT;
4430                         else
4431                                 err = add_new_disk(mddev, &info);
4432                         goto done_unlock;
4433                 }
4434
4435                 case HOT_REMOVE_DISK:
4436                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4437                         goto done_unlock;
4438
4439                 case HOT_ADD_DISK:
4440                         err = hot_add_disk(mddev, new_decode_dev(arg));
4441                         goto done_unlock;
4442
4443                 case SET_DISK_FAULTY:
4444                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4445                         goto done_unlock;
4446
4447                 case RUN_ARRAY:
4448                         err = do_md_run (mddev);
4449                         goto done_unlock;
4450
4451                 case SET_BITMAP_FILE:
4452                         err = set_bitmap_file(mddev, (int)arg);
4453                         goto done_unlock;
4454
4455                 default:
4456                         err = -EINVAL;
4457                         goto abort_unlock;
4458         }
4459
4460 done_unlock:
4461 abort_unlock:
4462         mddev_unlock(mddev);
4463
4464         return err;
4465 done:
4466         if (err)
4467                 MD_BUG();
4468 abort:
4469         return err;
4470 }
4471
4472 static int md_open(struct inode *inode, struct file *file)
4473 {
4474         /*
4475          * Succeed if we can lock the mddev, which confirms that
4476          * it isn't being stopped right now.
4477          */
4478         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4479         int err;
4480
4481         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4482                 goto out;
4483
4484         err = 0;
4485         mddev_get(mddev);
4486         mddev_unlock(mddev);
4487
4488         check_disk_change(inode->i_bdev);
4489  out:
4490         return err;
4491 }
4492
4493 static int md_release(struct inode *inode, struct file * file)
4494 {
4495         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4496
4497         BUG_ON(!mddev);
4498         mddev_put(mddev);
4499
4500         return 0;
4501 }
4502
4503 static int md_media_changed(struct gendisk *disk)
4504 {
4505         mddev_t *mddev = disk->private_data;
4506
4507         return mddev->changed;
4508 }
4509
4510 static int md_revalidate(struct gendisk *disk)
4511 {
4512         mddev_t *mddev = disk->private_data;
4513
4514         mddev->changed = 0;
4515         return 0;
4516 }
4517 static struct block_device_operations md_fops =
4518 {
4519         .owner          = THIS_MODULE,
4520         .open           = md_open,
4521         .release        = md_release,
4522         .ioctl          = md_ioctl,
4523         .getgeo         = md_getgeo,
4524         .media_changed  = md_media_changed,
4525         .revalidate_disk= md_revalidate,
4526 };
4527
4528 static int md_thread(void * arg)
4529 {
4530         mdk_thread_t *thread = arg;
4531
4532         /*
4533          * md_thread is a 'system-thread', it's priority should be very
4534          * high. We avoid resource deadlocks individually in each
4535          * raid personality. (RAID5 does preallocation) We also use RR and
4536          * the very same RT priority as kswapd, thus we will never get
4537          * into a priority inversion deadlock.
4538          *
4539          * we definitely have to have equal or higher priority than
4540          * bdflush, otherwise bdflush will deadlock if there are too
4541          * many dirty RAID5 blocks.
4542          */
4543
4544         current->flags |= PF_NOFREEZE;
4545         allow_signal(SIGKILL);
4546         while (!kthread_should_stop()) {
4547
4548                 /* We need to wait INTERRUPTIBLE so that
4549                  * we don't add to the load-average.
4550                  * That means we need to be sure no signals are
4551                  * pending
4552                  */
4553                 if (signal_pending(current))
4554                         flush_signals(current);
4555
4556                 wait_event_interruptible_timeout
4557                         (thread->wqueue,
4558                          test_bit(THREAD_WAKEUP, &thread->flags)
4559                          || kthread_should_stop(),
4560                          thread->timeout);
4561
4562                 clear_bit(THREAD_WAKEUP, &thread->flags);
4563
4564                 thread->run(thread->mddev);
4565         }
4566
4567         return 0;
4568 }
4569
4570 void md_wakeup_thread(mdk_thread_t *thread)
4571 {
4572         if (thread) {
4573                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4574                 set_bit(THREAD_WAKEUP, &thread->flags);
4575                 wake_up(&thread->wqueue);
4576         }
4577 }
4578
4579 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4580                                  const char *name)
4581 {
4582         mdk_thread_t *thread;
4583
4584         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4585         if (!thread)
4586                 return NULL;
4587
4588         init_waitqueue_head(&thread->wqueue);
4589
4590         thread->run = run;
4591         thread->mddev = mddev;
4592         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4593         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4594         if (IS_ERR(thread->tsk)) {
4595                 kfree(thread);
4596                 return NULL;
4597         }
4598         return thread;
4599 }
4600
4601 void md_unregister_thread(mdk_thread_t *thread)
4602 {
4603         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4604
4605         kthread_stop(thread->tsk);
4606         kfree(thread);
4607 }
4608
4609 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4610 {
4611         if (!mddev) {
4612                 MD_BUG();
4613                 return;
4614         }
4615
4616         if (!rdev || test_bit(Faulty, &rdev->flags))
4617                 return;
4618 /*
4619         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4620                 mdname(mddev),
4621                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4622                 __builtin_return_address(0),__builtin_return_address(1),
4623                 __builtin_return_address(2),__builtin_return_address(3));
4624 */
4625         if (!mddev->pers)
4626                 return;
4627         if (!mddev->pers->error_handler)
4628                 return;
4629         mddev->pers->error_handler(mddev,rdev);
4630         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4631         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4632         md_wakeup_thread(mddev->thread);
4633         md_new_event_inintr(mddev);
4634 }
4635
4636 /* seq_file implementation /proc/mdstat */
4637
4638 static void status_unused(struct seq_file *seq)
4639 {
4640         int i = 0;
4641         mdk_rdev_t *rdev;
4642         struct list_head *tmp;
4643
4644         seq_printf(seq, "unused devices: ");
4645
4646         ITERATE_RDEV_PENDING(rdev,tmp) {
4647                 char b[BDEVNAME_SIZE];
4648                 i++;
4649                 seq_printf(seq, "%s ",
4650                               bdevname(rdev->bdev,b));
4651         }
4652         if (!i)
4653                 seq_printf(seq, "<none>");
4654
4655         seq_printf(seq, "\n");
4656 }
4657
4658
4659 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4660 {
4661         sector_t max_blocks, resync, res;
4662         unsigned long dt, db, rt;
4663         int scale;
4664         unsigned int per_milli;
4665
4666         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4667
4668         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4669                 max_blocks = mddev->resync_max_sectors >> 1;
4670         else
4671                 max_blocks = mddev->size;
4672
4673         /*
4674          * Should not happen.
4675          */
4676         if (!max_blocks) {
4677                 MD_BUG();
4678                 return;
4679         }
4680         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4681          * in a sector_t, and (max_blocks>>scale) will fit in a
4682          * u32, as those are the requirements for sector_div.
4683          * Thus 'scale' must be at least 10
4684          */
4685         scale = 10;
4686         if (sizeof(sector_t) > sizeof(unsigned long)) {
4687                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4688                         scale++;
4689         }
4690         res = (resync>>scale)*1000;
4691         sector_div(res, (u32)((max_blocks>>scale)+1));
4692
4693         per_milli = res;
4694         {
4695                 int i, x = per_milli/50, y = 20-x;
4696                 seq_printf(seq, "[");
4697                 for (i = 0; i < x; i++)
4698                         seq_printf(seq, "=");
4699                 seq_printf(seq, ">");
4700                 for (i = 0; i < y; i++)
4701                         seq_printf(seq, ".");
4702                 seq_printf(seq, "] ");
4703         }
4704         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4705                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4706                     "reshape" :
4707                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4708                      "check" :
4709                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4710                       "resync" : "recovery"))),
4711                    per_milli/10, per_milli % 10,
4712                    (unsigned long long) resync,
4713                    (unsigned long long) max_blocks);
4714
4715         /*
4716          * We do not want to overflow, so the order of operands and
4717          * the * 100 / 100 trick are important. We do a +1 to be
4718          * safe against division by zero. We only estimate anyway.
4719          *
4720          * dt: time from mark until now
4721          * db: blocks written from mark until now
4722          * rt: remaining time
4723          */
4724         dt = ((jiffies - mddev->resync_mark) / HZ);
4725         if (!dt) dt++;
4726         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4727                 - mddev->resync_mark_cnt;
4728         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4729
4730         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4731
4732         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4733 }
4734
4735 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4736 {
4737         struct list_head *tmp;
4738         loff_t l = *pos;
4739         mddev_t *mddev;
4740
4741         if (l >= 0x10000)
4742                 return NULL;
4743         if (!l--)
4744                 /* header */
4745                 return (void*)1;
4746
4747         spin_lock(&all_mddevs_lock);
4748         list_for_each(tmp,&all_mddevs)
4749                 if (!l--) {
4750                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4751                         mddev_get(mddev);
4752                         spin_unlock(&all_mddevs_lock);
4753                         return mddev;
4754                 }
4755         spin_unlock(&all_mddevs_lock);
4756         if (!l--)
4757                 return (void*)2;/* tail */
4758         return NULL;
4759 }
4760
4761 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4762 {
4763         struct list_head *tmp;
4764         mddev_t *next_mddev, *mddev = v;
4765
4766         ++*pos;
4767         if (v == (void*)2)
4768                 return NULL;
4769
4770         spin_lock(&all_mddevs_lock);
4771         if (v == (void*)1)
4772                 tmp = all_mddevs.next;
4773         else
4774                 tmp = mddev->all_mddevs.next;
4775         if (tmp != &all_mddevs)
4776                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4777         else {
4778                 next_mddev = (void*)2;
4779                 *pos = 0x10000;
4780         }
4781         spin_unlock(&all_mddevs_lock);
4782
4783         if (v != (void*)1)
4784                 mddev_put(mddev);
4785         return next_mddev;
4786
4787 }
4788
4789 static void md_seq_stop(struct seq_file *seq, void *v)
4790 {
4791         mddev_t *mddev = v;
4792
4793         if (mddev && v != (void*)1 && v != (void*)2)
4794                 mddev_put(mddev);
4795 }
4796
4797 struct mdstat_info {
4798         int event;
4799 };
4800
4801 static int md_seq_show(struct seq_file *seq, void *v)
4802 {
4803         mddev_t *mddev = v;
4804         sector_t size;
4805         struct list_head *tmp2;
4806         mdk_rdev_t *rdev;
4807         struct mdstat_info *mi = seq->private;
4808         struct bitmap *bitmap;
4809
4810         if (v == (void*)1) {
4811                 struct mdk_personality *pers;
4812                 seq_printf(seq, "Personalities : ");
4813                 spin_lock(&pers_lock);
4814                 list_for_each_entry(pers, &pers_list, list)
4815                         seq_printf(seq, "[%s] ", pers->name);
4816
4817                 spin_unlock(&pers_lock);
4818                 seq_printf(seq, "\n");
4819                 mi->event = atomic_read(&md_event_count);
4820                 return 0;
4821         }
4822         if (v == (void*)2) {
4823                 status_unused(seq);
4824                 return 0;
4825         }
4826
4827         if (mddev_lock(mddev) < 0)
4828                 return -EINTR;
4829
4830         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4831                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4832                                                 mddev->pers ? "" : "in");
4833                 if (mddev->pers) {
4834                         if (mddev->ro==1)
4835                                 seq_printf(seq, " (read-only)");
4836                         if (mddev->ro==2)
4837                                 seq_printf(seq, "(auto-read-only)");
4838                         seq_printf(seq, " %s", mddev->pers->name);
4839                 }
4840
4841                 size = 0;
4842                 ITERATE_RDEV(mddev,rdev,tmp2) {
4843                         char b[BDEVNAME_SIZE];
4844                         seq_printf(seq, " %s[%d]",
4845                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4846                         if (test_bit(WriteMostly, &rdev->flags))
4847                                 seq_printf(seq, "(W)");
4848                         if (test_bit(Faulty, &rdev->flags)) {
4849                                 seq_printf(seq, "(F)");
4850                                 continue;
4851                         } else if (rdev->raid_disk < 0)
4852                                 seq_printf(seq, "(S)"); /* spare */
4853                         size += rdev->size;
4854                 }
4855
4856                 if (!list_empty(&mddev->disks)) {
4857                         if (mddev->pers)
4858                                 seq_printf(seq, "\n      %llu blocks",
4859                                         (unsigned long long)mddev->array_size);
4860                         else
4861                                 seq_printf(seq, "\n      %llu blocks",
4862                                         (unsigned long long)size);
4863                 }
4864                 if (mddev->persistent) {
4865                         if (mddev->major_version != 0 ||
4866                             mddev->minor_version != 90) {
4867                                 seq_printf(seq," super %d.%d",
4868                                            mddev->major_version,
4869                                            mddev->minor_version);
4870                         }
4871                 } else
4872                         seq_printf(seq, " super non-persistent");
4873
4874                 if (mddev->pers) {
4875                         mddev->pers->status (seq, mddev);
4876                         seq_printf(seq, "\n      ");
4877                         if (mddev->pers->sync_request) {
4878                                 if (mddev->curr_resync > 2) {
4879                                         status_resync (seq, mddev);
4880                                         seq_printf(seq, "\n      ");
4881                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4882                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4883                                 else if (mddev->recovery_cp < MaxSector)
4884                                         seq_printf(seq, "\tresync=PENDING\n      ");
4885                         }
4886                 } else
4887                         seq_printf(seq, "\n       ");
4888
4889                 if ((bitmap = mddev->bitmap)) {
4890                         unsigned long chunk_kb;
4891                         unsigned long flags;
4892                         spin_lock_irqsave(&bitmap->lock, flags);
4893                         chunk_kb = bitmap->chunksize >> 10;
4894                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4895                                 "%lu%s chunk",
4896                                 bitmap->pages - bitmap->missing_pages,
4897                                 bitmap->pages,
4898                                 (bitmap->pages - bitmap->missing_pages)
4899                                         << (PAGE_SHIFT - 10),
4900                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4901                                 chunk_kb ? "KB" : "B");
4902                         if (bitmap->file) {
4903                                 seq_printf(seq, ", file: ");
4904                                 seq_path(seq, bitmap->file->f_path.mnt,
4905                                          bitmap->file->f_path.dentry," \t\n");
4906                         }
4907
4908                         seq_printf(seq, "\n");
4909                         spin_unlock_irqrestore(&bitmap->lock, flags);
4910                 }
4911
4912                 seq_printf(seq, "\n");
4913         }
4914         mddev_unlock(mddev);
4915
4916         return 0;
4917 }
4918
4919 static struct seq_operations md_seq_ops = {
4920         .start  = md_seq_start,
4921         .next   = md_seq_next,
4922         .stop   = md_seq_stop,
4923         .show   = md_seq_show,
4924 };
4925
4926 static int md_seq_open(struct inode *inode, struct file *file)
4927 {
4928         int error;
4929         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4930         if (mi == NULL)
4931                 return -ENOMEM;
4932
4933         error = seq_open(file, &md_seq_ops);
4934         if (error)
4935                 kfree(mi);
4936         else {
4937                 struct seq_file *p = file->private_data;
4938                 p->private = mi;
4939                 mi->event = atomic_read(&md_event_count);
4940         }
4941         return error;
4942 }
4943
4944 static int md_seq_release(struct inode *inode, struct file *file)
4945 {
4946         struct seq_file *m = file->private_data;
4947         struct mdstat_info *mi = m->private;
4948         m->private = NULL;
4949         kfree(mi);
4950         return seq_release(inode, file);
4951 }
4952
4953 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4954 {
4955         struct seq_file *m = filp->private_data;
4956         struct mdstat_info *mi = m->private;
4957         int mask;
4958
4959         poll_wait(filp, &md_event_waiters, wait);
4960
4961         /* always allow read */
4962         mask = POLLIN | POLLRDNORM;
4963
4964         if (mi->event != atomic_read(&md_event_count))
4965                 mask |= POLLERR | POLLPRI;
4966         return mask;
4967 }
4968
4969 static const struct file_operations md_seq_fops = {
4970         .owner          = THIS_MODULE,
4971         .open           = md_seq_open,
4972         .read           = seq_read,
4973         .llseek         = seq_lseek,
4974         .release        = md_seq_release,
4975         .poll           = mdstat_poll,
4976 };
4977
4978 int register_md_personality(struct mdk_personality *p)
4979 {
4980         spin_lock(&pers_lock);
4981         list_add_tail(&p->list, &pers_list);
4982         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4983         spin_unlock(&pers_lock);
4984         return 0;
4985 }
4986
4987 int unregister_md_personality(struct mdk_personality *p)
4988 {
4989         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4990         spin_lock(&pers_lock);
4991         list_del_init(&p->list);
4992         spin_unlock(&pers_lock);
4993         return 0;
4994 }
4995
4996 static int is_mddev_idle(mddev_t *mddev)
4997 {
4998         mdk_rdev_t * rdev;
4999         struct list_head *tmp;
5000         int idle;
5001         unsigned long curr_events;
5002
5003         idle = 1;
5004         ITERATE_RDEV(mddev,rdev,tmp) {
5005                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5006                 curr_events = disk_stat_read(disk, sectors[0]) +
5007                                 disk_stat_read(disk, sectors[1]) -
5008                                 atomic_read(&disk->sync_io);
5009                 /* The difference between curr_events and last_events
5010                  * will be affected by any new non-sync IO (making
5011                  * curr_events bigger) and any difference in the amount of
5012                  * in-flight syncio (making current_events bigger or smaller)
5013                  * The amount in-flight is currently limited to
5014                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
5015                  * which is at most 4096 sectors.
5016                  * These numbers are fairly fragile and should be made
5017                  * more robust, probably by enforcing the
5018                  * 'window size' that md_do_sync sort-of uses.
5019                  *
5020                  * Note: the following is an unsigned comparison.
5021                  */
5022                 if ((curr_events - rdev->last_events + 4096) > 8192) {
5023                         rdev->last_events = curr_events;
5024                         idle = 0;
5025                 }
5026         }
5027         return idle;
5028 }
5029
5030 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5031 {
5032         /* another "blocks" (512byte) blocks have been synced */
5033         atomic_sub(blocks, &mddev->recovery_active);
5034         wake_up(&mddev->recovery_wait);
5035         if (!ok) {
5036                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5037                 md_wakeup_thread(mddev->thread);
5038                 // stop recovery, signal do_sync ....
5039         }
5040 }
5041
5042
5043 /* md_write_start(mddev, bi)
5044  * If we need to update some array metadata (e.g. 'active' flag
5045  * in superblock) before writing, schedule a superblock update
5046  * and wait for it to complete.
5047  */
5048 void md_write_start(mddev_t *mddev, struct bio *bi)
5049 {
5050         if (bio_data_dir(bi) != WRITE)
5051                 return;
5052
5053         BUG_ON(mddev->ro == 1);
5054         if (mddev->ro == 2) {
5055                 /* need to switch to read/write */
5056                 mddev->ro = 0;
5057                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5058                 md_wakeup_thread(mddev->thread);
5059         }
5060         atomic_inc(&mddev->writes_pending);
5061         if (mddev->in_sync) {
5062                 spin_lock_irq(&mddev->write_lock);
5063                 if (mddev->in_sync) {
5064                         mddev->in_sync = 0;
5065                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5066                         md_wakeup_thread(mddev->thread);
5067                 }
5068                 spin_unlock_irq(&mddev->write_lock);
5069         }
5070         wait_event(mddev->sb_wait, mddev->flags==0);
5071 }
5072
5073 void md_write_end(mddev_t *mddev)
5074 {
5075         if (atomic_dec_and_test(&mddev->writes_pending)) {
5076                 if (mddev->safemode == 2)
5077                         md_wakeup_thread(mddev->thread);
5078                 else if (mddev->safemode_delay)
5079                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5080         }
5081 }
5082
5083 /* md_allow_write(mddev)
5084  * Calling this ensures that the array is marked 'active' so that writes
5085  * may proceed without blocking.  It is important to call this before
5086  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5087  * Must be called with mddev_lock held.
5088  */
5089 void md_allow_write(mddev_t *mddev)
5090 {
5091         if (!mddev->pers)
5092                 return;
5093         if (mddev->ro)
5094                 return;
5095
5096         spin_lock_irq(&mddev->write_lock);
5097         if (mddev->in_sync) {
5098                 mddev->in_sync = 0;
5099                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5100                 if (mddev->safemode_delay &&
5101                     mddev->safemode == 0)
5102                         mddev->safemode = 1;
5103                 spin_unlock_irq(&mddev->write_lock);
5104                 md_update_sb(mddev, 0);
5105         } else
5106                 spin_unlock_irq(&mddev->write_lock);
5107 }
5108 EXPORT_SYMBOL_GPL(md_allow_write);
5109
5110 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5111
5112 #define SYNC_MARKS      10
5113 #define SYNC_MARK_STEP  (3*HZ)
5114 void md_do_sync(mddev_t *mddev)
5115 {
5116         mddev_t *mddev2;
5117         unsigned int currspeed = 0,
5118                  window;
5119         sector_t max_sectors,j, io_sectors;
5120         unsigned long mark[SYNC_MARKS];
5121         sector_t mark_cnt[SYNC_MARKS];
5122         int last_mark,m;
5123         struct list_head *tmp;
5124         sector_t last_check;
5125         int skipped = 0;
5126         struct list_head *rtmp;
5127         mdk_rdev_t *rdev;
5128         char *desc;
5129
5130         /* just incase thread restarts... */
5131         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5132                 return;
5133         if (mddev->ro) /* never try to sync a read-only array */
5134                 return;
5135
5136         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5137                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5138                         desc = "data-check";
5139                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5140                         desc = "requested-resync";
5141                 else
5142                         desc = "resync";
5143         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5144                 desc = "reshape";
5145         else
5146                 desc = "recovery";
5147
5148         /* we overload curr_resync somewhat here.
5149          * 0 == not engaged in resync at all
5150          * 2 == checking that there is no conflict with another sync
5151          * 1 == like 2, but have yielded to allow conflicting resync to
5152          *              commense
5153          * other == active in resync - this many blocks
5154          *
5155          * Before starting a resync we must have set curr_resync to
5156          * 2, and then checked that every "conflicting" array has curr_resync
5157          * less than ours.  When we find one that is the same or higher
5158          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5159          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5160          * This will mean we have to start checking from the beginning again.
5161          *
5162          */
5163
5164         do {
5165                 mddev->curr_resync = 2;
5166
5167         try_again:
5168                 if (kthread_should_stop()) {
5169                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5170                         goto skip;
5171                 }
5172                 ITERATE_MDDEV(mddev2,tmp) {
5173                         if (mddev2 == mddev)
5174                                 continue;
5175                         if (mddev2->curr_resync &&
5176                             match_mddev_units(mddev,mddev2)) {
5177                                 DEFINE_WAIT(wq);
5178                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5179                                         /* arbitrarily yield */
5180                                         mddev->curr_resync = 1;
5181                                         wake_up(&resync_wait);
5182                                 }
5183                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5184                                         /* no need to wait here, we can wait the next
5185                                          * time 'round when curr_resync == 2
5186                                          */
5187                                         continue;
5188                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5189                                 if (!kthread_should_stop() &&
5190                                     mddev2->curr_resync >= mddev->curr_resync) {
5191                                         printk(KERN_INFO "md: delaying %s of %s"
5192                                                " until %s has finished (they"
5193                                                " share one or more physical units)\n",
5194                                                desc, mdname(mddev), mdname(mddev2));
5195                                         mddev_put(mddev2);
5196                                         schedule();
5197                                         finish_wait(&resync_wait, &wq);
5198                                         goto try_again;
5199                                 }
5200                                 finish_wait(&resync_wait, &wq);
5201                         }
5202                 }
5203         } while (mddev->curr_resync < 2);
5204
5205         j = 0;
5206         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5207                 /* resync follows the size requested by the personality,
5208                  * which defaults to physical size, but can be virtual size
5209                  */
5210                 max_sectors = mddev->resync_max_sectors;
5211                 mddev->resync_mismatches = 0;
5212                 /* we don't use the checkpoint if there's a bitmap */
5213                 if (!mddev->bitmap &&
5214                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5215                         j = mddev->recovery_cp;
5216         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5217                 max_sectors = mddev->size << 1;
5218         else {
5219                 /* recovery follows the physical size of devices */
5220                 max_sectors = mddev->size << 1;
5221                 j = MaxSector;
5222                 ITERATE_RDEV(mddev,rdev,rtmp)
5223                         if (rdev->raid_disk >= 0 &&
5224                             !test_bit(Faulty, &rdev->flags) &&
5225                             !test_bit(In_sync, &rdev->flags) &&
5226                             rdev->recovery_offset < j)
5227                                 j = rdev->recovery_offset;
5228         }
5229
5230         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5231         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5232                 " %d KB/sec/disk.\n", speed_min(mddev));
5233         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5234                "(but not more than %d KB/sec) for %s.\n",
5235                speed_max(mddev), desc);
5236
5237         is_mddev_idle(mddev); /* this also initializes IO event counters */
5238
5239         io_sectors = 0;
5240         for (m = 0; m < SYNC_MARKS; m++) {
5241                 mark[m] = jiffies;
5242                 mark_cnt[m] = io_sectors;
5243         }
5244         last_mark = 0;
5245         mddev->resync_mark = mark[last_mark];
5246         mddev->resync_mark_cnt = mark_cnt[last_mark];
5247
5248         /*
5249          * Tune reconstruction:
5250          */
5251         window = 32*(PAGE_SIZE/512);
5252         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5253                 window/2,(unsigned long long) max_sectors/2);
5254
5255         atomic_set(&mddev->recovery_active, 0);
5256         init_waitqueue_head(&mddev->recovery_wait);
5257         last_check = 0;
5258
5259         if (j>2) {
5260                 printk(KERN_INFO
5261                        "md: resuming %s of %s from checkpoint.\n",
5262                        desc, mdname(mddev));
5263                 mddev->curr_resync = j;
5264         }
5265
5266         while (j < max_sectors) {
5267                 sector_t sectors;
5268
5269                 skipped = 0;
5270                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5271                                             currspeed < speed_min(mddev));
5272                 if (sectors == 0) {
5273                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5274                         goto out;
5275                 }
5276
5277                 if (!skipped) { /* actual IO requested */
5278                         io_sectors += sectors;
5279                         atomic_add(sectors, &mddev->recovery_active);
5280                 }
5281
5282                 j += sectors;
5283                 if (j>1) mddev->curr_resync = j;
5284                 mddev->curr_mark_cnt = io_sectors;
5285                 if (last_check == 0)
5286                         /* this is the earliers that rebuilt will be
5287                          * visible in /proc/mdstat
5288                          */
5289                         md_new_event(mddev);
5290
5291                 if (last_check + window > io_sectors || j == max_sectors)
5292                         continue;
5293
5294                 last_check = io_sectors;
5295
5296                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5297                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5298                         break;
5299
5300         repeat:
5301                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5302                         /* step marks */
5303                         int next = (last_mark+1) % SYNC_MARKS;
5304
5305                         mddev->resync_mark = mark[next];
5306                         mddev->resync_mark_cnt = mark_cnt[next];
5307                         mark[next] = jiffies;
5308                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5309                         last_mark = next;
5310                 }
5311
5312
5313                 if (kthread_should_stop()) {
5314                         /*
5315                          * got a signal, exit.
5316                          */
5317                         printk(KERN_INFO
5318                                 "md: md_do_sync() got signal ... exiting\n");
5319                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5320                         goto out;
5321                 }
5322
5323                 /*
5324                  * this loop exits only if either when we are slower than
5325                  * the 'hard' speed limit, or the system was IO-idle for
5326                  * a jiffy.
5327                  * the system might be non-idle CPU-wise, but we only care
5328                  * about not overloading the IO subsystem. (things like an
5329                  * e2fsck being done on the RAID array should execute fast)
5330                  */
5331                 mddev->queue->unplug_fn(mddev->queue);
5332                 cond_resched();
5333
5334                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5335                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5336
5337                 if (currspeed > speed_min(mddev)) {
5338                         if ((currspeed > speed_max(mddev)) ||
5339                                         !is_mddev_idle(mddev)) {
5340                                 msleep(500);
5341                                 goto repeat;
5342                         }
5343                 }
5344         }
5345         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5346         /*
5347          * this also signals 'finished resyncing' to md_stop
5348          */
5349  out:
5350         mddev->queue->unplug_fn(mddev->queue);
5351
5352         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5353
5354         /* tell personality that we are finished */
5355         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5356
5357         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5358             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5359             mddev->curr_resync > 2) {
5360                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5361                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5362                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5363                                         printk(KERN_INFO
5364                                                "md: checkpointing %s of %s.\n",
5365                                                desc, mdname(mddev));
5366                                         mddev->recovery_cp = mddev->curr_resync;
5367                                 }
5368                         } else
5369                                 mddev->recovery_cp = MaxSector;
5370                 } else {
5371                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5372                                 mddev->curr_resync = MaxSector;
5373                         ITERATE_RDEV(mddev,rdev,rtmp)
5374                                 if (rdev->raid_disk >= 0 &&
5375                                     !test_bit(Faulty, &rdev->flags) &&
5376                                     !test_bit(In_sync, &rdev->flags) &&
5377                                     rdev->recovery_offset < mddev->curr_resync)
5378                                         rdev->recovery_offset = mddev->curr_resync;
5379                 }
5380         }
5381         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5382
5383  skip:
5384         mddev->curr_resync = 0;
5385         wake_up(&resync_wait);
5386         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5387         md_wakeup_thread(mddev->thread);
5388 }
5389 EXPORT_SYMBOL_GPL(md_do_sync);
5390
5391
5392 static int remove_and_add_spares(mddev_t *mddev)
5393 {
5394         mdk_rdev_t *rdev;
5395         struct list_head *rtmp;
5396         int spares = 0;
5397
5398         ITERATE_RDEV(mddev,rdev,rtmp)
5399                 if (rdev->raid_disk >= 0 &&
5400                     (test_bit(Faulty, &rdev->flags) ||
5401                      ! test_bit(In_sync, &rdev->flags)) &&
5402                     atomic_read(&rdev->nr_pending)==0) {
5403                         if (mddev->pers->hot_remove_disk(
5404                                     mddev, rdev->raid_disk)==0) {
5405                                 char nm[20];
5406                                 sprintf(nm,"rd%d", rdev->raid_disk);
5407                                 sysfs_remove_link(&mddev->kobj, nm);
5408                                 rdev->raid_disk = -1;
5409                         }
5410                 }
5411
5412         if (mddev->degraded) {
5413                 ITERATE_RDEV(mddev,rdev,rtmp)
5414                         if (rdev->raid_disk < 0
5415                             && !test_bit(Faulty, &rdev->flags)) {
5416                                 rdev->recovery_offset = 0;
5417                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5418                                         char nm[20];
5419                                         sprintf(nm, "rd%d", rdev->raid_disk);
5420                                         if (sysfs_create_link(&mddev->kobj,
5421                                                               &rdev->kobj, nm))
5422                                                 printk(KERN_WARNING
5423                                                        "md: cannot register "
5424                                                        "%s for %s\n",
5425                                                        nm, mdname(mddev));
5426                                         spares++;
5427                                         md_new_event(mddev);
5428                                 } else
5429                                         break;
5430                         }
5431         }
5432         return spares;
5433 }
5434 /*
5435  * This routine is regularly called by all per-raid-array threads to
5436  * deal with generic issues like resync and super-block update.
5437  * Raid personalities that don't have a thread (linear/raid0) do not
5438  * need this as they never do any recovery or update the superblock.
5439  *
5440  * It does not do any resync itself, but rather "forks" off other threads
5441  * to do that as needed.
5442  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5443  * "->recovery" and create a thread at ->sync_thread.
5444  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5445  * and wakeups up this thread which will reap the thread and finish up.
5446  * This thread also removes any faulty devices (with nr_pending == 0).
5447  *
5448  * The overall approach is:
5449  *  1/ if the superblock needs updating, update it.
5450  *  2/ If a recovery thread is running, don't do anything else.
5451  *  3/ If recovery has finished, clean up, possibly marking spares active.
5452  *  4/ If there are any faulty devices, remove them.
5453  *  5/ If array is degraded, try to add spares devices
5454  *  6/ If array has spares or is not in-sync, start a resync thread.
5455  */
5456 void md_check_recovery(mddev_t *mddev)
5457 {
5458         mdk_rdev_t *rdev;
5459         struct list_head *rtmp;
5460
5461
5462         if (mddev->bitmap)
5463                 bitmap_daemon_work(mddev->bitmap);
5464
5465         if (mddev->ro)
5466                 return;
5467
5468         if (signal_pending(current)) {
5469                 if (mddev->pers->sync_request) {
5470                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5471                                mdname(mddev));
5472                         mddev->safemode = 2;
5473                 }
5474                 flush_signals(current);
5475         }
5476
5477         if ( ! (
5478                 mddev->flags ||
5479                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5480                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5481                 (mddev->safemode == 1) ||
5482                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5483                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5484                 ))
5485                 return;
5486
5487         if (mddev_trylock(mddev)) {
5488                 int spares = 0;
5489
5490                 spin_lock_irq(&mddev->write_lock);
5491                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5492                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5493                         mddev->in_sync = 1;
5494                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5495                 }
5496                 if (mddev->safemode == 1)
5497                         mddev->safemode = 0;
5498                 spin_unlock_irq(&mddev->write_lock);
5499
5500                 if (mddev->flags)
5501                         md_update_sb(mddev, 0);
5502
5503
5504                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5505                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5506                         /* resync/recovery still happening */
5507                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5508                         goto unlock;
5509                 }
5510                 if (mddev->sync_thread) {
5511                         /* resync has finished, collect result */
5512                         md_unregister_thread(mddev->sync_thread);
5513                         mddev->sync_thread = NULL;
5514                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5515                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5516                                 /* success...*/
5517                                 /* activate any spares */
5518                                 mddev->pers->spare_active(mddev);
5519                         }
5520                         md_update_sb(mddev, 1);
5521
5522                         /* if array is no-longer degraded, then any saved_raid_disk
5523                          * information must be scrapped
5524                          */
5525                         if (!mddev->degraded)
5526                                 ITERATE_RDEV(mddev,rdev,rtmp)
5527                                         rdev->saved_raid_disk = -1;
5528
5529                         mddev->recovery = 0;
5530                         /* flag recovery needed just to double check */
5531                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5532                         md_new_event(mddev);
5533                         goto unlock;
5534                 }
5535                 /* Clear some bits that don't mean anything, but
5536                  * might be left set
5537                  */
5538                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5539                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5540                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5541                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5542
5543                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5544                         goto unlock;
5545                 /* no recovery is running.
5546                  * remove any failed drives, then
5547                  * add spares if possible.
5548                  * Spare are also removed and re-added, to allow
5549                  * the personality to fail the re-add.
5550                  */
5551
5552                 if (mddev->reshape_position != MaxSector) {
5553                         if (mddev->pers->check_reshape(mddev) != 0)
5554                                 /* Cannot proceed */
5555                                 goto unlock;
5556                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5557                 } else if ((spares = remove_and_add_spares(mddev))) {
5558                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5559                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5560                 } else if (mddev->recovery_cp < MaxSector) {
5561                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5562                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5563                         /* nothing to be done ... */
5564                         goto unlock;
5565
5566                 if (mddev->pers->sync_request) {
5567                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5568                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5569                                 /* We are adding a device or devices to an array
5570                                  * which has the bitmap stored on all devices.
5571                                  * So make sure all bitmap pages get written
5572                                  */
5573                                 bitmap_write_all(mddev->bitmap);
5574                         }
5575                         mddev->sync_thread = md_register_thread(md_do_sync,
5576                                                                 mddev,
5577                                                                 "%s_resync");
5578                         if (!mddev->sync_thread) {
5579                                 printk(KERN_ERR "%s: could not start resync"
5580                                         " thread...\n",
5581                                         mdname(mddev));
5582                                 /* leave the spares where they are, it shouldn't hurt */
5583                                 mddev->recovery = 0;
5584                         } else
5585                                 md_wakeup_thread(mddev->sync_thread);
5586                         md_new_event(mddev);
5587                 }
5588         unlock:
5589                 mddev_unlock(mddev);
5590         }
5591 }
5592
5593 static int md_notify_reboot(struct notifier_block *this,
5594                             unsigned long code, void *x)
5595 {
5596         struct list_head *tmp;
5597         mddev_t *mddev;
5598
5599         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5600
5601                 printk(KERN_INFO "md: stopping all md devices.\n");
5602
5603                 ITERATE_MDDEV(mddev,tmp)
5604                         if (mddev_trylock(mddev)) {
5605                                 do_md_stop (mddev, 1);
5606                                 mddev_unlock(mddev);
5607                         }
5608                 /*
5609                  * certain more exotic SCSI devices are known to be
5610                  * volatile wrt too early system reboots. While the
5611                  * right place to handle this issue is the given
5612                  * driver, we do want to have a safe RAID driver ...
5613                  */
5614                 mdelay(1000*1);
5615         }
5616         return NOTIFY_DONE;
5617 }
5618
5619 static struct notifier_block md_notifier = {
5620         .notifier_call  = md_notify_reboot,
5621         .next           = NULL,
5622         .priority       = INT_MAX, /* before any real devices */
5623 };
5624
5625 static void md_geninit(void)
5626 {
5627         struct proc_dir_entry *p;
5628
5629         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5630
5631         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5632         if (p)
5633                 p->proc_fops = &md_seq_fops;
5634 }
5635
5636 static int __init md_init(void)
5637 {
5638         if (register_blkdev(MAJOR_NR, "md"))
5639                 return -1;
5640         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5641                 unregister_blkdev(MAJOR_NR, "md");
5642                 return -1;
5643         }
5644         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5645                             md_probe, NULL, NULL);
5646         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5647                             md_probe, NULL, NULL);
5648
5649         register_reboot_notifier(&md_notifier);
5650         raid_table_header = register_sysctl_table(raid_root_table);
5651
5652         md_geninit();
5653         return (0);
5654 }
5655
5656
5657 #ifndef MODULE
5658
5659 /*
5660  * Searches all registered partitions for autorun RAID arrays
5661  * at boot time.
5662  */
5663 static dev_t detected_devices[128];
5664 static int dev_cnt;
5665
5666 void md_autodetect_dev(dev_t dev)
5667 {
5668         if (dev_cnt >= 0 && dev_cnt < 127)
5669                 detected_devices[dev_cnt++] = dev;
5670 }
5671
5672
5673 static void autostart_arrays(int part)
5674 {
5675         mdk_rdev_t *rdev;
5676         int i;
5677
5678         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5679
5680         for (i = 0; i < dev_cnt; i++) {
5681                 dev_t dev = detected_devices[i];
5682
5683                 rdev = md_import_device(dev,0, 0);
5684                 if (IS_ERR(rdev))
5685                         continue;
5686
5687                 if (test_bit(Faulty, &rdev->flags)) {
5688                         MD_BUG();
5689                         continue;
5690                 }
5691                 list_add(&rdev->same_set, &pending_raid_disks);
5692         }
5693         dev_cnt = 0;
5694
5695         autorun_devices(part);
5696 }
5697
5698 #endif /* !MODULE */
5699
5700 static __exit void md_exit(void)
5701 {
5702         mddev_t *mddev;
5703         struct list_head *tmp;
5704
5705         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5706         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5707
5708         unregister_blkdev(MAJOR_NR,"md");
5709         unregister_blkdev(mdp_major, "mdp");
5710         unregister_reboot_notifier(&md_notifier);
5711         unregister_sysctl_table(raid_table_header);
5712         remove_proc_entry("mdstat", NULL);
5713         ITERATE_MDDEV(mddev,tmp) {
5714                 struct gendisk *disk = mddev->gendisk;
5715                 if (!disk)
5716                         continue;
5717                 export_array(mddev);
5718                 del_gendisk(disk);
5719                 put_disk(disk);
5720                 mddev->gendisk = NULL;
5721                 mddev_put(mddev);
5722         }
5723 }
5724
5725 module_init(md_init)
5726 module_exit(md_exit)
5727
5728 static int get_ro(char *buffer, struct kernel_param *kp)
5729 {
5730         return sprintf(buffer, "%d", start_readonly);
5731 }
5732 static int set_ro(const char *val, struct kernel_param *kp)
5733 {
5734         char *e;
5735         int num = simple_strtoul(val, &e, 10);
5736         if (*val && (*e == '\0' || *e == '\n')) {
5737                 start_readonly = num;
5738                 return 0;
5739         }
5740         return -EINVAL;
5741 }
5742
5743 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5744 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5745
5746
5747 EXPORT_SYMBOL(register_md_personality);
5748 EXPORT_SYMBOL(unregister_md_personality);
5749 EXPORT_SYMBOL(md_error);
5750 EXPORT_SYMBOL(md_done_sync);
5751 EXPORT_SYMBOL(md_write_start);
5752 EXPORT_SYMBOL(md_write_end);
5753 EXPORT_SYMBOL(md_register_thread);
5754 EXPORT_SYMBOL(md_unregister_thread);
5755 EXPORT_SYMBOL(md_wakeup_thread);
5756 EXPORT_SYMBOL(md_check_recovery);
5757 MODULE_LICENSE("GPL");
5758 MODULE_ALIAS("md");
5759 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);