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/kernel.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/buffer_head.h> /* for invalidate_bdev */
  43 #include <linux/poll.h>
  44 #include <linux/mutex.h>
  45 #include <linux/ctype.h>
  46 #include <linux/freezer.h>
  47
  48 #include <linux/init.h>
  49
  50 #include <linux/file.h>
  51
  52 #ifdef CONFIG_KMOD
  53 #include <linux/kmod.h>
  54 #endif
  55
  56 #include <asm/unaligned.h>
  57
  58 #define MAJOR_NR MD_MAJOR
  59 #define MD_DRIVER
  60
  61 /* 63 partitions with the alternate major number (mdp) */
  62 #define MdpMinorShift 6
  63
  64 #define DEBUG 0
  65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  66
  67
  68 #ifndef MODULE
  69 static void autostart_arrays (int part);
  70 #endif
  71
  72 static LIST_HEAD(pers_list);
  73 static DEFINE_SPINLOCK(pers_lock);
  74
  75 static void md_print_devices(void);
  76
  77 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  78
  79 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  80
  81 /*
  82  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  83  * is 1000 KB/sec, so the extra system load does not show up that much.
  84  * Increase it if you want to have more _guaranteed_ speed. Note that
  85  * the RAID driver will use the maximum available bandwidth if the IO
  86  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  87  * speed limit - in case reconstruction slows down your system despite
  88  * idle IO detection.
  89  *
  90  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  91  * or /sys/block/mdX/md/sync_speed_{min,max}
  92  */
  93
  94 static int sysctl_speed_limit_min = 1000;
  95 static int sysctl_speed_limit_max = 200000;
  96 static inline int speed_min(mddev_t *mddev)
  97 {
  98         return mddev->sync_speed_min ?
  99                 mddev->sync_speed_min : sysctl_speed_limit_min;
 100 }
 101
 102 static inline int speed_max(mddev_t *mddev)
 103 {
 104         return mddev->sync_speed_max ?
 105                 mddev->sync_speed_max : sysctl_speed_limit_max;
 106 }
 107
 108 static struct ctl_table_header *raid_table_header;
 109
 110 static ctl_table raid_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 113                 .procname       = "speed_limit_min",
 114                 .data           = &sysctl_speed_limit_min,
 115                 .maxlen         = sizeof(int),
 116                 .mode           = S_IRUGO|S_IWUSR,
 117                 .proc_handler   = &proc_dointvec,
 118         },
 119         {
 120                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 121                 .procname       = "speed_limit_max",
 122                 .data           = &sysctl_speed_limit_max,
 123                 .maxlen         = sizeof(int),
 124                 .mode           = S_IRUGO|S_IWUSR,
 125                 .proc_handler   = &proc_dointvec,
 126         },
 127         { .ctl_name = 0 }
 128 };
 129
 130 static ctl_table raid_dir_table[] = {
 131         {
 132                 .ctl_name       = DEV_RAID,
 133                 .procname       = "raid",
 134                 .maxlen         = 0,
 135                 .mode           = S_IRUGO|S_IXUGO,
 136                 .child          = raid_table,
 137         },
 138         { .ctl_name = 0 }
 139 };
 140
 141 static ctl_table raid_root_table[] = {
 142         {
 143                 .ctl_name       = CTL_DEV,
 144                 .procname       = "dev",
 145                 .maxlen         = 0,
 146                 .mode           = 0555,
 147                 .child          = raid_dir_table,
 148         },
 149         { .ctl_name = 0 }
 150 };
 151
 152 static struct block_device_operations md_fops;
 153
 154 static int start_readonly;
 155
 156 /*
 157  * We have a system wide 'event count' that is incremented
 158  * on any 'interesting' event, and readers of /proc/mdstat
 159  * can use 'poll' or 'select' to find out when the event
 160  * count increases.
 161  *
 162  * Events are:
 163  *  start array, stop array, error, add device, remove device,
 164  *  start build, activate spare
 165  */
 166 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 167 static atomic_t md_event_count;
 168 void md_new_event(mddev_t *mddev)
 169 {
 170         atomic_inc(&md_event_count);
 171         wake_up(&md_event_waiters);
 172         sysfs_notify(&mddev->kobj, NULL, "sync_action");
 173 }
 174 EXPORT_SYMBOL_GPL(md_new_event);
 175
 176 /* Alternate version that can be called from interrupts
 177  * when calling sysfs_notify isn't needed.
 178  */
 179 static void md_new_event_inintr(mddev_t *mddev)
 180 {
 181         atomic_inc(&md_event_count);
 182         wake_up(&md_event_waiters);
 183 }
 184
 185 /*
 186  * Enables to iterate over all existing md arrays
 187  * all_mddevs_lock protects this list.
 188  */
 189 static LIST_HEAD(all_mddevs);
 190 static DEFINE_SPINLOCK(all_mddevs_lock);
 191
 192
 193 /*
 194  * iterates through all used mddevs in the system.
 195  * We take care to grab the all_mddevs_lock whenever navigating
 196  * the list, and to always hold a refcount when unlocked.
 197  * Any code which breaks out of this loop while own
 198  * a reference to the current mddev and must mddev_put it.
 199  */
 200 #define for_each_mddev(mddev,tmp)                                       \
 201                                                                         \
 202         for (({ spin_lock(&all_mddevs_lock);                            \
 203                 tmp = all_mddevs.next;                                  \
 204                 mddev = NULL;});                                        \
 205              ({ if (tmp != &all_mddevs)                                 \
 206                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 207                 spin_unlock(&all_mddevs_lock);                          \
 208                 if (mddev) mddev_put(mddev);                            \
 209                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 210                 tmp != &all_mddevs;});                                  \
 211              ({ spin_lock(&all_mddevs_lock);                            \
 212                 tmp = tmp->next;})                                      \
 213                 )
 214
 215
 216 static int md_fail_request (struct request_queue *q, struct bio *bio)
 217 {
 218         bio_io_error(bio);
 219         return 0;
 220 }
 221
 222 static inline mddev_t *mddev_get(mddev_t *mddev)
 223 {
 224         atomic_inc(&mddev->active);
 225         return mddev;
 226 }
 227
 228 static void mddev_put(mddev_t *mddev)
 229 {
 230         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 231                 return;
 232         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 233                 list_del(&mddev->all_mddevs);
 234                 spin_unlock(&all_mddevs_lock);
 235                 blk_cleanup_queue(mddev->queue);
 236                 kobject_put(&mddev->kobj);
 237         } else
 238                 spin_unlock(&all_mddevs_lock);
 239 }
 240
 241 static mddev_t * mddev_find(dev_t unit)
 242 {
 243         mddev_t *mddev, *new = NULL;
 244
 245  retry:
 246         spin_lock(&all_mddevs_lock);
 247         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 248                 if (mddev->unit == unit) {
 249                         mddev_get(mddev);
 250                         spin_unlock(&all_mddevs_lock);
 251                         kfree(new);
 252                         return mddev;
 253                 }
 254
 255         if (new) {
 256                 list_add(&new->all_mddevs, &all_mddevs);
 257                 spin_unlock(&all_mddevs_lock);
 258                 return new;
 259         }
 260         spin_unlock(&all_mddevs_lock);
 261
 262         new = kzalloc(sizeof(*new), GFP_KERNEL);
 263         if (!new)
 264                 return NULL;
 265
 266         new->unit = unit;
 267         if (MAJOR(unit) == MD_MAJOR)
 268                 new->md_minor = MINOR(unit);
 269         else
 270                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 271
 272         mutex_init(&new->reconfig_mutex);
 273         INIT_LIST_HEAD(&new->disks);
 274         INIT_LIST_HEAD(&new->all_mddevs);
 275         init_timer(&new->safemode_timer);
 276         atomic_set(&new->active, 1);
 277         spin_lock_init(&new->write_lock);
 278         init_waitqueue_head(&new->sb_wait);
 279         init_waitqueue_head(&new->recovery_wait);
 280         new->reshape_position = MaxSector;
 281         new->resync_max = MaxSector;
 282         new->level = LEVEL_NONE;
 283
 284         new->queue = blk_alloc_queue(GFP_KERNEL);
 285         if (!new->queue) {
 286                 kfree(new);
 287                 return NULL;
 288         }
 289         /* Can be unlocked because the queue is new: no concurrency */
 290         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
 291
 292         blk_queue_make_request(new->queue, md_fail_request);
 293
 294         goto retry;
 295 }
 296
 297 static inline int mddev_lock(mddev_t * mddev)
 298 {
 299         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 300 }
 301
 302 static inline int mddev_trylock(mddev_t * mddev)
 303 {
 304         return mutex_trylock(&mddev->reconfig_mutex);
 305 }
 306
 307 static inline void mddev_unlock(mddev_t * mddev)
 308 {
 309         mutex_unlock(&mddev->reconfig_mutex);
 310
 311         md_wakeup_thread(mddev->thread);
 312 }
 313
 314 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 315 {
 316         mdk_rdev_t * rdev;
 317         struct list_head *tmp;
 318
 319         rdev_for_each(rdev, tmp, mddev) {
 320                 if (rdev->desc_nr == nr)
 321                         return rdev;
 322         }
 323         return NULL;
 324 }
 325
 326 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 327 {
 328         struct list_head *tmp;
 329         mdk_rdev_t *rdev;
 330
 331         rdev_for_each(rdev, tmp, mddev) {
 332                 if (rdev->bdev->bd_dev == dev)
 333                         return rdev;
 334         }
 335         return NULL;
 336 }
 337
 338 static struct mdk_personality *find_pers(int level, char *clevel)
 339 {
 340         struct mdk_personality *pers;
 341         list_for_each_entry(pers, &pers_list, list) {
 342                 if (level != LEVEL_NONE && pers->level == level)
 343                         return pers;
 344                 if (strcmp(pers->name, clevel)==0)
 345                         return pers;
 346         }
 347         return NULL;
 348 }
 349
 350 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 351 {
 352         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 353         return MD_NEW_SIZE_BLOCKS(size);
 354 }
 355
 356 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 357 {
 358         sector_t size;
 359
 360         size = rdev->sb_offset;
 361
 362         if (chunk_size)
 363                 size &= ~((sector_t)chunk_size/1024 - 1);
 364         return size;
 365 }
 366
 367 static int alloc_disk_sb(mdk_rdev_t * rdev)
 368 {
 369         if (rdev->sb_page)
 370                 MD_BUG();
 371
 372         rdev->sb_page = alloc_page(GFP_KERNEL);
 373         if (!rdev->sb_page) {
 374                 printk(KERN_ALERT "md: out of memory.\n");
 375                 return -EINVAL;
 376         }
 377
 378         return 0;
 379 }
 380
 381 static void free_disk_sb(mdk_rdev_t * rdev)
 382 {
 383         if (rdev->sb_page) {
 384                 put_page(rdev->sb_page);
 385                 rdev->sb_loaded = 0;
 386                 rdev->sb_page = NULL;
 387                 rdev->sb_offset = 0;
 388                 rdev->size = 0;
 389         }
 390 }
 391
 392
 393 static void super_written(struct bio *bio, int error)
 394 {
 395         mdk_rdev_t *rdev = bio->bi_private;
 396         mddev_t *mddev = rdev->mddev;
 397
 398         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 399                 printk("md: super_written gets error=%d, uptodate=%d\n",
 400                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 401                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 402                 md_error(mddev, rdev);
 403         }
 404
 405         if (atomic_dec_and_test(&mddev->pending_writes))
 406                 wake_up(&mddev->sb_wait);
 407         bio_put(bio);
 408 }
 409
 410 static void super_written_barrier(struct bio *bio, int error)
 411 {
 412         struct bio *bio2 = bio->bi_private;
 413         mdk_rdev_t *rdev = bio2->bi_private;
 414         mddev_t *mddev = rdev->mddev;
 415
 416         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 417             error == -EOPNOTSUPP) {
 418                 unsigned long flags;
 419                 /* barriers don't appear to be supported :-( */
 420                 set_bit(BarriersNotsupp, &rdev->flags);
 421                 mddev->barriers_work = 0;
 422                 spin_lock_irqsave(&mddev->write_lock, flags);
 423                 bio2->bi_next = mddev->biolist;
 424                 mddev->biolist = bio2;
 425                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 426                 wake_up(&mddev->sb_wait);
 427                 bio_put(bio);
 428         } else {
 429                 bio_put(bio2);
 430                 bio->bi_private = rdev;
 431                 super_written(bio, error);
 432         }
 433 }
 434
 435 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 436                    sector_t sector, int size, struct page *page)
 437 {
 438         /* write first size bytes of page to sector of rdev
 439          * Increment mddev->pending_writes before returning
 440          * and decrement it on completion, waking up sb_wait
 441          * if zero is reached.
 442          * If an error occurred, call md_error
 443          *
 444          * As we might need to resubmit the request if BIO_RW_BARRIER
 445          * causes ENOTSUPP, we allocate a spare bio...
 446          */
 447         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 448         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 449
 450         bio->bi_bdev = rdev->bdev;
 451         bio->bi_sector = sector;
 452         bio_add_page(bio, page, size, 0);
 453         bio->bi_private = rdev;
 454         bio->bi_end_io = super_written;
 455         bio->bi_rw = rw;
 456
 457         atomic_inc(&mddev->pending_writes);
 458         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 459                 struct bio *rbio;
 460                 rw |= (1<<BIO_RW_BARRIER);
 461                 rbio = bio_clone(bio, GFP_NOIO);
 462                 rbio->bi_private = bio;
 463                 rbio->bi_end_io = super_written_barrier;
 464                 submit_bio(rw, rbio);
 465         } else
 466                 submit_bio(rw, bio);
 467 }
 468
 469 void md_super_wait(mddev_t *mddev)
 470 {
 471         /* wait for all superblock writes that were scheduled to complete.
 472          * if any had to be retried (due to BARRIER problems), retry them
 473          */
 474         DEFINE_WAIT(wq);
 475         for(;;) {
 476                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 477                 if (atomic_read(&mddev->pending_writes)==0)
 478                         break;
 479                 while (mddev->biolist) {
 480                         struct bio *bio;
 481                         spin_lock_irq(&mddev->write_lock);
 482                         bio = mddev->biolist;
 483                         mddev->biolist = bio->bi_next ;
 484                         bio->bi_next = NULL;
 485                         spin_unlock_irq(&mddev->write_lock);
 486                         submit_bio(bio->bi_rw, bio);
 487                 }
 488                 schedule();
 489         }
 490         finish_wait(&mddev->sb_wait, &wq);
 491 }
 492
 493 static void bi_complete(struct bio *bio, int error)
 494 {
 495         complete((struct completion*)bio->bi_private);
 496 }
 497
 498 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 499                    struct page *page, int rw)
 500 {
 501         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 502         struct completion event;
 503         int ret;
 504
 505         rw |= (1 << BIO_RW_SYNC);
 506
 507         bio->bi_bdev = bdev;
 508         bio->bi_sector = sector;
 509         bio_add_page(bio, page, size, 0);
 510         init_completion(&event);
 511         bio->bi_private = &event;
 512         bio->bi_end_io = bi_complete;
 513         submit_bio(rw, bio);
 514         wait_for_completion(&event);
 515
 516         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 517         bio_put(bio);
 518         return ret;
 519 }
 520 EXPORT_SYMBOL_GPL(sync_page_io);
 521
 522 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 523 {
 524         char b[BDEVNAME_SIZE];
 525         if (!rdev->sb_page) {
 526                 MD_BUG();
 527                 return -EINVAL;
 528         }
 529         if (rdev->sb_loaded)
 530                 return 0;
 531
 532
 533         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 534                 goto fail;
 535         rdev->sb_loaded = 1;
 536         return 0;
 537
 538 fail:
 539         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 540                 bdevname(rdev->bdev,b));
 541         return -EINVAL;
 542 }
 543
 544 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 545 {
 546         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 547                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 548                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 549                 (sb1->set_uuid3 == sb2->set_uuid3))
 550
 551                 return 1;
 552
 553         return 0;
 554 }
 555
 556
 557 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 558 {
 559         int ret;
 560         mdp_super_t *tmp1, *tmp2;
 561
 562         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 563         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 564
 565         if (!tmp1 || !tmp2) {
 566                 ret = 0;
 567                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 568                 goto abort;
 569         }
 570
 571         *tmp1 = *sb1;
 572         *tmp2 = *sb2;
 573
 574         /*
 575          * nr_disks is not constant
 576          */
 577         tmp1->nr_disks = 0;
 578         tmp2->nr_disks = 0;
 579
 580         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 581                 ret = 0;
 582         else
 583                 ret = 1;
 584
 585 abort:
 586         kfree(tmp1);
 587         kfree(tmp2);
 588         return ret;
 589 }
 590
 591
 592 static u32 md_csum_fold(u32 csum)
 593 {
 594         csum = (csum & 0xffff) + (csum >> 16);
 595         return (csum & 0xffff) + (csum >> 16);
 596 }
 597
 598 static unsigned int calc_sb_csum(mdp_super_t * sb)
 599 {
 600         u64 newcsum = 0;
 601         u32 *sb32 = (u32*)sb;
 602         int i;
 603         unsigned int disk_csum, csum;
 604
 605         disk_csum = sb->sb_csum;
 606         sb->sb_csum = 0;
 607
 608         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 609                 newcsum += sb32[i];
 610         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 611
 612
 613 #ifdef CONFIG_ALPHA
 614         /* This used to use csum_partial, which was wrong for several
 615          * reasons including that different results are returned on
 616          * different architectures.  It isn't critical that we get exactly
 617          * the same return value as before (we always csum_fold before
 618          * testing, and that removes any differences).  However as we
 619          * know that csum_partial always returned a 16bit value on
 620          * alphas, do a fold to maximise conformity to previous behaviour.
 621          */
 622         sb->sb_csum = md_csum_fold(disk_csum);
 623 #else
 624         sb->sb_csum = disk_csum;
 625 #endif
 626         return csum;
 627 }
 628
 629
 630 /*
 631  * Handle superblock details.
 632  * We want to be able to handle multiple superblock formats
 633  * so we have a common interface to them all, and an array of
 634  * different handlers.
 635  * We rely on user-space to write the initial superblock, and support
 636  * reading and updating of superblocks.
 637  * Interface methods are:
 638  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 639  *      loads and validates a superblock on dev.
 640  *      if refdev != NULL, compare superblocks on both devices
 641  *    Return:
 642  *      0 - dev has a superblock that is compatible with refdev
 643  *      1 - dev has a superblock that is compatible and newer than refdev
 644  *          so dev should be used as the refdev in future
 645  *     -EINVAL superblock incompatible or invalid
 646  *     -othererror e.g. -EIO
 647  *
 648  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 649  *      Verify that dev is acceptable into mddev.
 650  *       The first time, mddev->raid_disks will be 0, and data from
 651  *       dev should be merged in.  Subsequent calls check that dev
 652  *       is new enough.  Return 0 or -EINVAL
 653  *
 654  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 655  *     Update the superblock for rdev with data in mddev
 656  *     This does not write to disc.
 657  *
 658  */
 659
 660 struct super_type  {
 661         char            *name;
 662         struct module   *owner;
 663         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 664         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 665         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 666 };
 667
 668 /*
 669  * load_super for 0.90.0
 670  */
 671 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 672 {
 673         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 674         mdp_super_t *sb;
 675         int ret;
 676         sector_t sb_offset;
 677
 678         /*
 679          * Calculate the position of the superblock,
 680          * it's at the end of the disk.
 681          *
 682          * It also happens to be a multiple of 4Kb.
 683          */
 684         sb_offset = calc_dev_sboffset(rdev->bdev);
 685         rdev->sb_offset = sb_offset;
 686
 687         ret = read_disk_sb(rdev, MD_SB_BYTES);
 688         if (ret) return ret;
 689
 690         ret = -EINVAL;
 691
 692         bdevname(rdev->bdev, b);
 693         sb = (mdp_super_t*)page_address(rdev->sb_page);
 694
 695         if (sb->md_magic != MD_SB_MAGIC) {
 696                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 697                        b);
 698                 goto abort;
 699         }
 700
 701         if (sb->major_version != 0 ||
 702             sb->minor_version < 90 ||
 703             sb->minor_version > 91) {
 704                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 705                         sb->major_version, sb->minor_version,
 706                         b);
 707                 goto abort;
 708         }
 709
 710         if (sb->raid_disks <= 0)
 711                 goto abort;
 712
 713         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 714                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 715                         b);
 716                 goto abort;
 717         }
 718
 719         rdev->preferred_minor = sb->md_minor;
 720         rdev->data_offset = 0;
 721         rdev->sb_size = MD_SB_BYTES;
 722
 723         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 724                 if (sb->level != 1 && sb->level != 4
 725                     && sb->level != 5 && sb->level != 6
 726                     && sb->level != 10) {
 727                         /* FIXME use a better test */
 728                         printk(KERN_WARNING
 729                                "md: bitmaps not supported for this level.\n");
 730                         goto abort;
 731                 }
 732         }
 733
 734         if (sb->level == LEVEL_MULTIPATH)
 735                 rdev->desc_nr = -1;
 736         else
 737                 rdev->desc_nr = sb->this_disk.number;
 738
 739         if (!refdev) {
 740                 ret = 1;
 741         } else {
 742                 __u64 ev1, ev2;
 743                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 744                 if (!uuid_equal(refsb, sb)) {
 745                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 746                                 b, bdevname(refdev->bdev,b2));
 747                         goto abort;
 748                 }
 749                 if (!sb_equal(refsb, sb)) {
 750                         printk(KERN_WARNING "md: %s has same UUID"
 751                                " but different superblock to %s\n",
 752                                b, bdevname(refdev->bdev, b2));
 753                         goto abort;
 754                 }
 755                 ev1 = md_event(sb);
 756                 ev2 = md_event(refsb);
 757                 if (ev1 > ev2)
 758                         ret = 1;
 759                 else
 760                         ret = 0;
 761         }
 762         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 763
 764         if (rdev->size < sb->size && sb->level > 1)
 765                 /* "this cannot possibly happen" ... */
 766                 ret = -EINVAL;
 767
 768  abort:
 769         return ret;
 770 }
 771
 772 /*
 773  * validate_super for 0.90.0
 774  */
 775 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 776 {
 777         mdp_disk_t *desc;
 778         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 779         __u64 ev1 = md_event(sb);
 780
 781         rdev->raid_disk = -1;
 782         clear_bit(Faulty, &rdev->flags);
 783         clear_bit(In_sync, &rdev->flags);
 784         clear_bit(WriteMostly, &rdev->flags);
 785         clear_bit(BarriersNotsupp, &rdev->flags);
 786
 787         if (mddev->raid_disks == 0) {
 788                 mddev->major_version = 0;
 789                 mddev->minor_version = sb->minor_version;
 790                 mddev->patch_version = sb->patch_version;
 791                 mddev->external = 0;
 792                 mddev->chunk_size = sb->chunk_size;
 793                 mddev->ctime = sb->ctime;
 794                 mddev->utime = sb->utime;
 795                 mddev->level = sb->level;
 796                 mddev->clevel[0] = 0;
 797                 mddev->layout = sb->layout;
 798                 mddev->raid_disks = sb->raid_disks;
 799                 mddev->size = sb->size;
 800                 mddev->events = ev1;
 801                 mddev->bitmap_offset = 0;
 802                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 803
 804                 if (mddev->minor_version >= 91) {
 805                         mddev->reshape_position = sb->reshape_position;
 806                         mddev->delta_disks = sb->delta_disks;
 807                         mddev->new_level = sb->new_level;
 808                         mddev->new_layout = sb->new_layout;
 809                         mddev->new_chunk = sb->new_chunk;
 810                 } else {
 811                         mddev->reshape_position = MaxSector;
 812                         mddev->delta_disks = 0;
 813                         mddev->new_level = mddev->level;
 814                         mddev->new_layout = mddev->layout;
 815                         mddev->new_chunk = mddev->chunk_size;
 816                 }
 817
 818                 if (sb->state & (1<<MD_SB_CLEAN))
 819                         mddev->recovery_cp = MaxSector;
 820                 else {
 821                         if (sb->events_hi == sb->cp_events_hi &&
 822                                 sb->events_lo == sb->cp_events_lo) {
 823                                 mddev->recovery_cp = sb->recovery_cp;
 824                         } else
 825                                 mddev->recovery_cp = 0;
 826                 }
 827
 828                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 829                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 830                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 831                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 832
 833                 mddev->max_disks = MD_SB_DISKS;
 834
 835                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 836                     mddev->bitmap_file == NULL)
 837                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 838
 839         } else if (mddev->pers == NULL) {
 840                 /* Insist on good event counter while assembling */
 841                 ++ev1;
 842                 if (ev1 < mddev->events)
 843                         return -EINVAL;
 844         } else if (mddev->bitmap) {
 845                 /* if adding to array with a bitmap, then we can accept an
 846                  * older device ... but not too old.
 847                  */
 848                 if (ev1 < mddev->bitmap->events_cleared)
 849                         return 0;
 850         } else {
 851                 if (ev1 < mddev->events)
 852                         /* just a hot-add of a new device, leave raid_disk at -1 */
 853                         return 0;
 854         }
 855
 856         if (mddev->level != LEVEL_MULTIPATH) {
 857                 desc = sb->disks + rdev->desc_nr;
 858
 859                 if (desc->state & (1<<MD_DISK_FAULTY))
 860                         set_bit(Faulty, &rdev->flags);
 861                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
 862                             desc->raid_disk < mddev->raid_disks */) {
 863                         set_bit(In_sync, &rdev->flags);
 864                         rdev->raid_disk = desc->raid_disk;
 865                 }
 866                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 867                         set_bit(WriteMostly, &rdev->flags);
 868         } else /* MULTIPATH are always insync */
 869                 set_bit(In_sync, &rdev->flags);
 870         return 0;
 871 }
 872
 873 /*
 874  * sync_super for 0.90.0
 875  */
 876 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 877 {
 878         mdp_super_t *sb;
 879         struct list_head *tmp;
 880         mdk_rdev_t *rdev2;
 881         int next_spare = mddev->raid_disks;
 882
 883
 884         /* make rdev->sb match mddev data..
 885          *
 886          * 1/ zero out disks
 887          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 888          * 3/ any empty disks < next_spare become removed
 889          *
 890          * disks[0] gets initialised to REMOVED because
 891          * we cannot be sure from other fields if it has
 892          * been initialised or not.
 893          */
 894         int i;
 895         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 896
 897         rdev->sb_size = MD_SB_BYTES;
 898
 899         sb = (mdp_super_t*)page_address(rdev->sb_page);
 900
 901         memset(sb, 0, sizeof(*sb));
 902
 903         sb->md_magic = MD_SB_MAGIC;
 904         sb->major_version = mddev->major_version;
 905         sb->patch_version = mddev->patch_version;
 906         sb->gvalid_words  = 0; /* ignored */
 907         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 908         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 909         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 910         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 911
 912         sb->ctime = mddev->ctime;
 913         sb->level = mddev->level;
 914         sb->size  = mddev->size;
 915         sb->raid_disks = mddev->raid_disks;
 916         sb->md_minor = mddev->md_minor;
 917         sb->not_persistent = 0;
 918         sb->utime = mddev->utime;
 919         sb->state = 0;
 920         sb->events_hi = (mddev->events>>32);
 921         sb->events_lo = (u32)mddev->events;
 922
 923         if (mddev->reshape_position == MaxSector)
 924                 sb->minor_version = 90;
 925         else {
 926                 sb->minor_version = 91;
 927                 sb->reshape_position = mddev->reshape_position;
 928                 sb->new_level = mddev->new_level;
 929                 sb->delta_disks = mddev->delta_disks;
 930                 sb->new_layout = mddev->new_layout;
 931                 sb->new_chunk = mddev->new_chunk;
 932         }
 933         mddev->minor_version = sb->minor_version;
 934         if (mddev->in_sync)
 935         {
 936                 sb->recovery_cp = mddev->recovery_cp;
 937                 sb->cp_events_hi = (mddev->events>>32);
 938                 sb->cp_events_lo = (u32)mddev->events;
 939                 if (mddev->recovery_cp == MaxSector)
 940                         sb->state = (1<< MD_SB_CLEAN);
 941         } else
 942                 sb->recovery_cp = 0;
 943
 944         sb->layout = mddev->layout;
 945         sb->chunk_size = mddev->chunk_size;
 946
 947         if (mddev->bitmap && mddev->bitmap_file == NULL)
 948                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 949
 950         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 951         rdev_for_each(rdev2, tmp, mddev) {
 952                 mdp_disk_t *d;
 953                 int desc_nr;
 954                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 955                     && !test_bit(Faulty, &rdev2->flags))
 956                         desc_nr = rdev2->raid_disk;
 957                 else
 958                         desc_nr = next_spare++;
 959                 rdev2->desc_nr = desc_nr;
 960                 d = &sb->disks[rdev2->desc_nr];
 961                 nr_disks++;
 962                 d->number = rdev2->desc_nr;
 963                 d->major = MAJOR(rdev2->bdev->bd_dev);
 964                 d->minor = MINOR(rdev2->bdev->bd_dev);
 965                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 966                     && !test_bit(Faulty, &rdev2->flags))
 967                         d->raid_disk = rdev2->raid_disk;
 968                 else
 969                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 970                 if (test_bit(Faulty, &rdev2->flags))
 971                         d->state = (1<<MD_DISK_FAULTY);
 972                 else if (test_bit(In_sync, &rdev2->flags)) {
 973                         d->state = (1<<MD_DISK_ACTIVE);
 974                         d->state |= (1<<MD_DISK_SYNC);
 975                         active++;
 976                         working++;
 977                 } else {
 978                         d->state = 0;
 979                         spare++;
 980                         working++;
 981                 }
 982                 if (test_bit(WriteMostly, &rdev2->flags))
 983                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 984         }
 985         /* now set the "removed" and "faulty" bits on any missing devices */
 986         for (i=0 ; i < mddev->raid_disks ; i++) {
 987                 mdp_disk_t *d = &sb->disks[i];
 988                 if (d->state == 0 && d->number == 0) {
 989                         d->number = i;
 990                         d->raid_disk = i;
 991                         d->state = (1<<MD_DISK_REMOVED);
 992                         d->state |= (1<<MD_DISK_FAULTY);
 993                         failed++;
 994                 }
 995         }
 996         sb->nr_disks = nr_disks;
 997         sb->active_disks = active;
 998         sb->working_disks = working;
 999         sb->failed_disks = failed;
1000         sb->spare_disks = spare;
1001
1002         sb->this_disk = sb->disks[rdev->desc_nr];
1003         sb->sb_csum = calc_sb_csum(sb);
1004 }
1005
1006 /*
1007  * version 1 superblock
1008  */
1009
1010 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1011 {
1012         __le32 disk_csum;
1013         u32 csum;
1014         unsigned long long newcsum;
1015         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1016         __le32 *isuper = (__le32*)sb;
1017         int i;
1018
1019         disk_csum = sb->sb_csum;
1020         sb->sb_csum = 0;
1021         newcsum = 0;
1022         for (i=0; size>=4; size -= 4 )
1023                 newcsum += le32_to_cpu(*isuper++);
1024
1025         if (size == 2)
1026                 newcsum += le16_to_cpu(*(__le16*) isuper);
1027
1028         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1029         sb->sb_csum = disk_csum;
1030         return cpu_to_le32(csum);
1031 }
1032
1033 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1034 {
1035         struct mdp_superblock_1 *sb;
1036         int ret;
1037         sector_t sb_offset;
1038         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1039         int bmask;
1040
1041         /*
1042          * Calculate the position of the superblock.
1043          * It is always aligned to a 4K boundary and
1044          * depeding on minor_version, it can be:
1045          * 0: At least 8K, but less than 12K, from end of device
1046          * 1: At start of device
1047          * 2: 4K from start of device.
1048          */
1049         switch(minor_version) {
1050         case 0:
1051                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1052                 sb_offset -= 8*2;
1053                 sb_offset &= ~(sector_t)(4*2-1);
1054                 /* convert from sectors to K */
1055                 sb_offset /= 2;
1056                 break;
1057         case 1:
1058                 sb_offset = 0;
1059                 break;
1060         case 2:
1061                 sb_offset = 4;
1062                 break;
1063         default:
1064                 return -EINVAL;
1065         }
1066         rdev->sb_offset = sb_offset;
1067
1068         /* superblock is rarely larger than 1K, but it can be larger,
1069          * and it is safe to read 4k, so we do that
1070          */
1071         ret = read_disk_sb(rdev, 4096);
1072         if (ret) return ret;
1073
1074
1075         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1076
1077         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1078             sb->major_version != cpu_to_le32(1) ||
1079             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1080             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1081             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1082                 return -EINVAL;
1083
1084         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1085                 printk("md: invalid superblock checksum on %s\n",
1086                         bdevname(rdev->bdev,b));
1087                 return -EINVAL;
1088         }
1089         if (le64_to_cpu(sb->data_size) < 10) {
1090                 printk("md: data_size too small on %s\n",
1091                        bdevname(rdev->bdev,b));
1092                 return -EINVAL;
1093         }
1094         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1095                 if (sb->level != cpu_to_le32(1) &&
1096                     sb->level != cpu_to_le32(4) &&
1097                     sb->level != cpu_to_le32(5) &&
1098                     sb->level != cpu_to_le32(6) &&
1099                     sb->level != cpu_to_le32(10)) {
1100                         printk(KERN_WARNING
1101                                "md: bitmaps not supported for this level.\n");
1102                         return -EINVAL;
1103                 }
1104         }
1105
1106         rdev->preferred_minor = 0xffff;
1107         rdev->data_offset = le64_to_cpu(sb->data_offset);
1108         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1109
1110         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1111         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1112         if (rdev->sb_size & bmask)
1113                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1114
1115         if (minor_version
1116             && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1117                 return -EINVAL;
1118
1119         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1120                 rdev->desc_nr = -1;
1121         else
1122                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1123
1124         if (!refdev) {
1125                 ret = 1;
1126         } else {
1127                 __u64 ev1, ev2;
1128                 struct mdp_superblock_1 *refsb =
1129                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1130
1131                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1132                     sb->level != refsb->level ||
1133                     sb->layout != refsb->layout ||
1134                     sb->chunksize != refsb->chunksize) {
1135                         printk(KERN_WARNING "md: %s has strangely different"
1136                                 " superblock to %s\n",
1137                                 bdevname(rdev->bdev,b),
1138                                 bdevname(refdev->bdev,b2));
1139                         return -EINVAL;
1140                 }
1141                 ev1 = le64_to_cpu(sb->events);
1142                 ev2 = le64_to_cpu(refsb->events);
1143
1144                 if (ev1 > ev2)
1145                         ret = 1;
1146                 else
1147                         ret = 0;
1148         }
1149         if (minor_version)
1150                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1151         else
1152                 rdev->size = rdev->sb_offset;
1153         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1154                 return -EINVAL;
1155         rdev->size = le64_to_cpu(sb->data_size)/2;
1156         if (le32_to_cpu(sb->chunksize))
1157                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1158
1159         if (le64_to_cpu(sb->size) > rdev->size*2)
1160                 return -EINVAL;
1161         return ret;
1162 }
1163
1164 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1165 {
1166         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1167         __u64 ev1 = le64_to_cpu(sb->events);
1168
1169         rdev->raid_disk = -1;
1170         clear_bit(Faulty, &rdev->flags);
1171         clear_bit(In_sync, &rdev->flags);
1172         clear_bit(WriteMostly, &rdev->flags);
1173         clear_bit(BarriersNotsupp, &rdev->flags);
1174
1175         if (mddev->raid_disks == 0) {
1176                 mddev->major_version = 1;
1177                 mddev->patch_version = 0;
1178                 mddev->external = 0;
1179                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1180                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1181                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1182                 mddev->level = le32_to_cpu(sb->level);
1183                 mddev->clevel[0] = 0;
1184                 mddev->layout = le32_to_cpu(sb->layout);
1185                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1186                 mddev->size = le64_to_cpu(sb->size)/2;
1187                 mddev->events = ev1;
1188                 mddev->bitmap_offset = 0;
1189                 mddev->default_bitmap_offset = 1024 >> 9;
1190
1191                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1192                 memcpy(mddev->uuid, sb->set_uuid, 16);
1193
1194                 mddev->max_disks =  (4096-256)/2;
1195
1196                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1197                     mddev->bitmap_file == NULL )
1198                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1199
1200                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1201                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1202                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1203                         mddev->new_level = le32_to_cpu(sb->new_level);
1204                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1205                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1206                 } else {
1207                         mddev->reshape_position = MaxSector;
1208                         mddev->delta_disks = 0;
1209                         mddev->new_level = mddev->level;
1210                         mddev->new_layout = mddev->layout;
1211                         mddev->new_chunk = mddev->chunk_size;
1212                 }
1213
1214         } else if (mddev->pers == NULL) {
1215                 /* Insist of good event counter while assembling */
1216                 ++ev1;
1217                 if (ev1 < mddev->events)
1218                         return -EINVAL;
1219         } else if (mddev->bitmap) {
1220                 /* If adding to array with a bitmap, then we can accept an
1221                  * older device, but not too old.
1222                  */
1223                 if (ev1 < mddev->bitmap->events_cleared)
1224                         return 0;
1225         } else {
1226                 if (ev1 < mddev->events)
1227                         /* just a hot-add of a new device, leave raid_disk at -1 */
1228                         return 0;
1229         }
1230         if (mddev->level != LEVEL_MULTIPATH) {
1231                 int role;
1232                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1233                 switch(role) {
1234                 case 0xffff: /* spare */
1235                         break;
1236                 case 0xfffe: /* faulty */
1237                         set_bit(Faulty, &rdev->flags);
1238                         break;
1239                 default:
1240                         if ((le32_to_cpu(sb->feature_map) &
1241                              MD_FEATURE_RECOVERY_OFFSET))
1242                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1243                         else
1244                                 set_bit(In_sync, &rdev->flags);
1245                         rdev->raid_disk = role;
1246                         break;
1247                 }
1248                 if (sb->devflags & WriteMostly1)
1249                         set_bit(WriteMostly, &rdev->flags);
1250         } else /* MULTIPATH are always insync */
1251                 set_bit(In_sync, &rdev->flags);
1252
1253         return 0;
1254 }
1255
1256 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1257 {
1258         struct mdp_superblock_1 *sb;
1259         struct list_head *tmp;
1260         mdk_rdev_t *rdev2;
1261         int max_dev, i;
1262         /* make rdev->sb match mddev and rdev data. */
1263
1264         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1265
1266         sb->feature_map = 0;
1267         sb->pad0 = 0;
1268         sb->recovery_offset = cpu_to_le64(0);
1269         memset(sb->pad1, 0, sizeof(sb->pad1));
1270         memset(sb->pad2, 0, sizeof(sb->pad2));
1271         memset(sb->pad3, 0, sizeof(sb->pad3));
1272
1273         sb->utime = cpu_to_le64((__u64)mddev->utime);
1274         sb->events = cpu_to_le64(mddev->events);
1275         if (mddev->in_sync)
1276                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1277         else
1278                 sb->resync_offset = cpu_to_le64(0);
1279
1280         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1281
1282         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1283         sb->size = cpu_to_le64(mddev->size<<1);
1284
1285         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1286                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1287                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1288         }
1289
1290         if (rdev->raid_disk >= 0 &&
1291             !test_bit(In_sync, &rdev->flags) &&
1292             rdev->recovery_offset > 0) {
1293                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1294                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1295         }
1296
1297         if (mddev->reshape_position != MaxSector) {
1298                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1299                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1300                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1301                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1302                 sb->new_level = cpu_to_le32(mddev->new_level);
1303                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1304         }
1305
1306         max_dev = 0;
1307         rdev_for_each(rdev2, tmp, mddev)
1308                 if (rdev2->desc_nr+1 > max_dev)
1309                         max_dev = rdev2->desc_nr+1;
1310
1311         if (max_dev > le32_to_cpu(sb->max_dev))
1312                 sb->max_dev = cpu_to_le32(max_dev);
1313         for (i=0; i<max_dev;i++)
1314                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1315
1316         rdev_for_each(rdev2, tmp, mddev) {
1317                 i = rdev2->desc_nr;
1318                 if (test_bit(Faulty, &rdev2->flags))
1319                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1320                 else if (test_bit(In_sync, &rdev2->flags))
1321                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1322                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1323                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1324                 else
1325                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1326         }
1327
1328         sb->sb_csum = calc_sb_1_csum(sb);
1329 }
1330
1331
1332 static struct super_type super_types[] = {
1333         [0] = {
1334                 .name   = "0.90.0",
1335                 .owner  = THIS_MODULE,
1336                 .load_super     = super_90_load,
1337                 .validate_super = super_90_validate,
1338                 .sync_super     = super_90_sync,
1339         },
1340         [1] = {
1341                 .name   = "md-1",
1342                 .owner  = THIS_MODULE,
1343                 .load_super     = super_1_load,
1344                 .validate_super = super_1_validate,
1345                 .sync_super     = super_1_sync,
1346         },
1347 };
1348
1349 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1350 {
1351         struct list_head *tmp, *tmp2;
1352         mdk_rdev_t *rdev, *rdev2;
1353
1354         rdev_for_each(rdev, tmp, mddev1)
1355                 rdev_for_each(rdev2, tmp2, mddev2)
1356                         if (rdev->bdev->bd_contains ==
1357                             rdev2->bdev->bd_contains)
1358                                 return 1;
1359
1360         return 0;
1361 }
1362
1363 static LIST_HEAD(pending_raid_disks);
1364
1365 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1366 {
1367         char b[BDEVNAME_SIZE];
1368         struct kobject *ko;
1369         char *s;
1370         int err;
1371
1372         if (rdev->mddev) {
1373                 MD_BUG();
1374                 return -EINVAL;
1375         }
1376
1377         /* prevent duplicates */
1378         if (find_rdev(mddev, rdev->bdev->bd_dev))
1379                 return -EEXIST;
1380
1381         /* make sure rdev->size exceeds mddev->size */
1382         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1383                 if (mddev->pers) {
1384                         /* Cannot change size, so fail
1385                          * If mddev->level <= 0, then we don't care
1386                          * about aligning sizes (e.g. linear)
1387                          */
1388                         if (mddev->level > 0)
1389                                 return -ENOSPC;
1390                 } else
1391                         mddev->size = rdev->size;
1392         }
1393
1394         /* Verify rdev->desc_nr is unique.
1395          * If it is -1, assign a free number, else
1396          * check number is not in use
1397          */
1398         if (rdev->desc_nr < 0) {
1399                 int choice = 0;
1400                 if (mddev->pers) choice = mddev->raid_disks;
1401                 while (find_rdev_nr(mddev, choice))
1402                         choice++;
1403                 rdev->desc_nr = choice;
1404         } else {
1405                 if (find_rdev_nr(mddev, rdev->desc_nr))
1406                         return -EBUSY;
1407         }
1408         bdevname(rdev->bdev,b);
1409         while ( (s=strchr(b, '/')) != NULL)
1410                 *s = '!';
1411
1412         rdev->mddev = mddev;
1413         printk(KERN_INFO "md: bind<%s>\n", b);
1414
1415         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1416                 goto fail;
1417
1418         if (rdev->bdev->bd_part)
1419                 ko = &rdev->bdev->bd_part->dev.kobj;
1420         else
1421                 ko = &rdev->bdev->bd_disk->dev.kobj;
1422         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1423                 kobject_del(&rdev->kobj);
1424                 goto fail;
1425         }
1426         list_add(&rdev->same_set, &mddev->disks);
1427         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1428         return 0;
1429
1430  fail:
1431         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1432                b, mdname(mddev));
1433         return err;
1434 }
1435
1436 static void md_delayed_delete(struct work_struct *ws)
1437 {
1438         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1439         kobject_del(&rdev->kobj);
1440         kobject_put(&rdev->kobj);
1441 }
1442
1443 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1444 {
1445         char b[BDEVNAME_SIZE];
1446         if (!rdev->mddev) {
1447                 MD_BUG();
1448                 return;
1449         }
1450         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1451         list_del_init(&rdev->same_set);
1452         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1453         rdev->mddev = NULL;
1454         sysfs_remove_link(&rdev->kobj, "block");
1455
1456         /* We need to delay this, otherwise we can deadlock when
1457          * writing to 'remove' to "dev/state"
1458          */
1459         INIT_WORK(&rdev->del_work, md_delayed_delete);
1460         kobject_get(&rdev->kobj);
1461         schedule_work(&rdev->del_work);
1462 }
1463
1464 /*
1465  * prevent the device from being mounted, repartitioned or
1466  * otherwise reused by a RAID array (or any other kernel
1467  * subsystem), by bd_claiming the device.
1468  */
1469 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1470 {
1471         int err = 0;
1472         struct block_device *bdev;
1473         char b[BDEVNAME_SIZE];
1474
1475         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1476         if (IS_ERR(bdev)) {
1477                 printk(KERN_ERR "md: could not open %s.\n",
1478                         __bdevname(dev, b));
1479                 return PTR_ERR(bdev);
1480         }
1481         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1482         if (err) {
1483                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1484                         bdevname(bdev, b));
1485                 blkdev_put(bdev);
1486                 return err;
1487         }
1488         if (!shared)
1489                 set_bit(AllReserved, &rdev->flags);
1490         rdev->bdev = bdev;
1491         return err;
1492 }
1493
1494 static void unlock_rdev(mdk_rdev_t *rdev)
1495 {
1496         struct block_device *bdev = rdev->bdev;
1497         rdev->bdev = NULL;
1498         if (!bdev)
1499                 MD_BUG();
1500         bd_release(bdev);
1501         blkdev_put(bdev);
1502 }
1503
1504 void md_autodetect_dev(dev_t dev);
1505
1506 static void export_rdev(mdk_rdev_t * rdev)
1507 {
1508         char b[BDEVNAME_SIZE];
1509         printk(KERN_INFO "md: export_rdev(%s)\n",
1510                 bdevname(rdev->bdev,b));
1511         if (rdev->mddev)
1512                 MD_BUG();
1513         free_disk_sb(rdev);
1514         list_del_init(&rdev->same_set);
1515 #ifndef MODULE
1516         if (test_bit(AutoDetected, &rdev->flags))
1517                 md_autodetect_dev(rdev->bdev->bd_dev);
1518 #endif
1519         unlock_rdev(rdev);
1520         kobject_put(&rdev->kobj);
1521 }
1522
1523 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1524 {
1525         unbind_rdev_from_array(rdev);
1526         export_rdev(rdev);
1527 }
1528
1529 static void export_array(mddev_t *mddev)
1530 {
1531         struct list_head *tmp;
1532         mdk_rdev_t *rdev;
1533
1534         rdev_for_each(rdev, tmp, mddev) {
1535                 if (!rdev->mddev) {
1536                         MD_BUG();
1537                         continue;
1538                 }
1539                 kick_rdev_from_array(rdev);
1540         }
1541         if (!list_empty(&mddev->disks))
1542                 MD_BUG();
1543         mddev->raid_disks = 0;
1544         mddev->major_version = 0;
1545 }
1546
1547 static void print_desc(mdp_disk_t *desc)
1548 {
1549         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1550                 desc->major,desc->minor,desc->raid_disk,desc->state);
1551 }
1552
1553 static void print_sb(mdp_super_t *sb)
1554 {
1555         int i;
1556
1557         printk(KERN_INFO
1558                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1559                 sb->major_version, sb->minor_version, sb->patch_version,
1560                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1561                 sb->ctime);
1562         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1563                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1564                 sb->md_minor, sb->layout, sb->chunk_size);
1565         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1566                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1567                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1568                 sb->failed_disks, sb->spare_disks,
1569                 sb->sb_csum, (unsigned long)sb->events_lo);
1570
1571         printk(KERN_INFO);
1572         for (i = 0; i < MD_SB_DISKS; i++) {
1573                 mdp_disk_t *desc;
1574
1575                 desc = sb->disks + i;
1576                 if (desc->number || desc->major || desc->minor ||
1577                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1578                         printk("     D %2d: ", i);
1579                         print_desc(desc);
1580                 }
1581         }
1582         printk(KERN_INFO "md:     THIS: ");
1583         print_desc(&sb->this_disk);
1584
1585 }
1586
1587 static void print_rdev(mdk_rdev_t *rdev)
1588 {
1589         char b[BDEVNAME_SIZE];
1590         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1591                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1592                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1593                 rdev->desc_nr);
1594         if (rdev->sb_loaded) {
1595                 printk(KERN_INFO "md: rdev superblock:\n");
1596                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1597         } else
1598                 printk(KERN_INFO "md: no rdev superblock!\n");
1599 }
1600
1601 static void md_print_devices(void)
1602 {
1603         struct list_head *tmp, *tmp2;
1604         mdk_rdev_t *rdev;
1605         mddev_t *mddev;
1606         char b[BDEVNAME_SIZE];
1607
1608         printk("\n");
1609         printk("md:     **********************************\n");
1610         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1611         printk("md:     **********************************\n");
1612         for_each_mddev(mddev, tmp) {
1613
1614                 if (mddev->bitmap)
1615                         bitmap_print_sb(mddev->bitmap);
1616                 else
1617                         printk("%s: ", mdname(mddev));
1618                 rdev_for_each(rdev, tmp2, mddev)
1619                         printk("<%s>", bdevname(rdev->bdev,b));
1620                 printk("\n");
1621
1622                 rdev_for_each(rdev, tmp2, mddev)
1623                         print_rdev(rdev);
1624         }
1625         printk("md:     **********************************\n");
1626         printk("\n");
1627 }
1628
1629
1630 static void sync_sbs(mddev_t * mddev, int nospares)
1631 {
1632         /* Update each superblock (in-memory image), but
1633          * if we are allowed to, skip spares which already
1634          * have the right event counter, or have one earlier
1635          * (which would mean they aren't being marked as dirty
1636          * with the rest of the array)
1637          */
1638         mdk_rdev_t *rdev;
1639         struct list_head *tmp;
1640
1641         rdev_for_each(rdev, tmp, mddev) {
1642                 if (rdev->sb_events == mddev->events ||
1643                     (nospares &&
1644                      rdev->raid_disk < 0 &&
1645                      (rdev->sb_events&1)==0 &&
1646                      rdev->sb_events+1 == mddev->events)) {
1647                         /* Don't update this superblock */
1648                         rdev->sb_loaded = 2;
1649                 } else {
1650                         super_types[mddev->major_version].
1651                                 sync_super(mddev, rdev);
1652                         rdev->sb_loaded = 1;
1653                 }
1654         }
1655 }
1656
1657 static void md_update_sb(mddev_t * mddev, int force_change)
1658 {
1659         struct list_head *tmp;
1660         mdk_rdev_t *rdev;
1661         int sync_req;
1662         int nospares = 0;
1663
1664         if (mddev->external)
1665                 return;
1666 repeat:
1667         spin_lock_irq(&mddev->write_lock);
1668
1669         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1670         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1671                 force_change = 1;
1672         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1673                 /* just a clean<-> dirty transition, possibly leave spares alone,
1674                  * though if events isn't the right even/odd, we will have to do
1675                  * spares after all
1676                  */
1677                 nospares = 1;
1678         if (force_change)
1679                 nospares = 0;
1680         if (mddev->degraded)
1681                 /* If the array is degraded, then skipping spares is both
1682                  * dangerous and fairly pointless.
1683                  * Dangerous because a device that was removed from the array
1684                  * might have a event_count that still looks up-to-date,
1685                  * so it can be re-added without a resync.
1686                  * Pointless because if there are any spares to skip,
1687                  * then a recovery will happen and soon that array won't
1688                  * be degraded any more and the spare can go back to sleep then.
1689                  */
1690                 nospares = 0;
1691
1692         sync_req = mddev->in_sync;
1693         mddev->utime = get_seconds();
1694
1695         /* If this is just a dirty<->clean transition, and the array is clean
1696          * and 'events' is odd, we can roll back to the previous clean state */
1697         if (nospares
1698             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1699             && (mddev->events & 1)
1700             && mddev->events != 1)
1701                 mddev->events--;
1702         else {
1703                 /* otherwise we have to go forward and ... */
1704                 mddev->events ++;
1705                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1706                         /* .. if the array isn't clean, insist on an odd 'events' */
1707                         if ((mddev->events&1)==0) {
1708                                 mddev->events++;
1709                                 nospares = 0;
1710                         }
1711                 } else {
1712                         /* otherwise insist on an even 'events' (for clean states) */
1713                         if ((mddev->events&1)) {
1714                                 mddev->events++;
1715                                 nospares = 0;
1716                         }
1717                 }
1718         }
1719
1720         if (!mddev->events) {
1721                 /*
1722                  * oops, this 64-bit counter should never wrap.
1723                  * Either we are in around ~1 trillion A.C., assuming
1724                  * 1 reboot per second, or we have a bug:
1725                  */
1726                 MD_BUG();
1727                 mddev->events --;
1728         }
1729
1730         /*
1731          * do not write anything to disk if using
1732          * nonpersistent superblocks
1733          */
1734         if (!mddev->persistent) {
1735                 if (!mddev->external)
1736                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1737
1738                 spin_unlock_irq(&mddev->write_lock);
1739                 wake_up(&mddev->sb_wait);
1740                 return;
1741         }
1742         sync_sbs(mddev, nospares);
1743         spin_unlock_irq(&mddev->write_lock);
1744
1745         dprintk(KERN_INFO
1746                 "md: updating %s RAID superblock on device (in sync %d)\n",
1747                 mdname(mddev),mddev->in_sync);
1748
1749         bitmap_update_sb(mddev->bitmap);
1750         rdev_for_each(rdev, tmp, mddev) {
1751                 char b[BDEVNAME_SIZE];
1752                 dprintk(KERN_INFO "md: ");
1753                 if (rdev->sb_loaded != 1)
1754                         continue; /* no noise on spare devices */
1755                 if (test_bit(Faulty, &rdev->flags))
1756                         dprintk("(skipping faulty ");
1757
1758                 dprintk("%s ", bdevname(rdev->bdev,b));
1759                 if (!test_bit(Faulty, &rdev->flags)) {
1760                         md_super_write(mddev,rdev,
1761                                        rdev->sb_offset<<1, rdev->sb_size,
1762                                        rdev->sb_page);
1763                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1764                                 bdevname(rdev->bdev,b),
1765                                 (unsigned long long)rdev->sb_offset);
1766                         rdev->sb_events = mddev->events;
1767
1768                 } else
1769                         dprintk(")\n");
1770                 if (mddev->level == LEVEL_MULTIPATH)
1771                         /* only need to write one superblock... */
1772                         break;
1773         }
1774         md_super_wait(mddev);
1775         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1776
1777         spin_lock_irq(&mddev->write_lock);
1778         if (mddev->in_sync != sync_req ||
1779             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1780                 /* have to write it out again */
1781                 spin_unlock_irq(&mddev->write_lock);
1782                 goto repeat;
1783         }
1784         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1785         spin_unlock_irq(&mddev->write_lock);
1786         wake_up(&mddev->sb_wait);
1787
1788 }
1789
1790 /* words written to sysfs files may, or my not, be \n terminated.
1791  * We want to accept with case. For this we use cmd_match.
1792  */
1793 static int cmd_match(const char *cmd, const char *str)
1794 {
1795         /* See if cmd, written into a sysfs file, matches
1796          * str.  They must either be the same, or cmd can
1797          * have a trailing newline
1798          */
1799         while (*cmd && *str && *cmd == *str) {
1800                 cmd++;
1801                 str++;
1802         }
1803         if (*cmd == '\n')
1804                 cmd++;
1805         if (*str || *cmd)
1806                 return 0;
1807         return 1;
1808 }
1809
1810 struct rdev_sysfs_entry {
1811         struct attribute attr;
1812         ssize_t (*show)(mdk_rdev_t *, char *);
1813         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1814 };
1815
1816 static ssize_t
1817 state_show(mdk_rdev_t *rdev, char *page)
1818 {
1819         char *sep = "";
1820         size_t len = 0;
1821
1822         if (test_bit(Faulty, &rdev->flags)) {
1823                 len+= sprintf(page+len, "%sfaulty",sep);
1824                 sep = ",";
1825         }
1826         if (test_bit(In_sync, &rdev->flags)) {
1827                 len += sprintf(page+len, "%sin_sync",sep);
1828                 sep = ",";
1829         }
1830         if (test_bit(WriteMostly, &rdev->flags)) {
1831                 len += sprintf(page+len, "%swrite_mostly",sep);
1832                 sep = ",";
1833         }
1834         if (test_bit(Blocked, &rdev->flags)) {
1835                 len += sprintf(page+len, "%sblocked", sep);
1836                 sep = ",";
1837         }
1838         if (!test_bit(Faulty, &rdev->flags) &&
1839             !test_bit(In_sync, &rdev->flags)) {
1840                 len += sprintf(page+len, "%sspare", sep);
1841                 sep = ",";
1842         }
1843         return len+sprintf(page+len, "\n");
1844 }
1845
1846 static ssize_t
1847 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1848 {
1849         /* can write
1850          *  faulty  - simulates and error
1851          *  remove  - disconnects the device
1852          *  writemostly - sets write_mostly
1853          *  -writemostly - clears write_mostly
1854          *  blocked - sets the Blocked flag
1855          *  -blocked - clears the Blocked flag
1856          */
1857         int err = -EINVAL;
1858         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1859                 md_error(rdev->mddev, rdev);
1860                 err = 0;
1861         } else if (cmd_match(buf, "remove")) {
1862                 if (rdev->raid_disk >= 0)
1863                         err = -EBUSY;
1864                 else {
1865                         mddev_t *mddev = rdev->mddev;
1866                         kick_rdev_from_array(rdev);
1867                         if (mddev->pers)
1868                                 md_update_sb(mddev, 1);
1869                         md_new_event(mddev);
1870                         err = 0;
1871                 }
1872         } else if (cmd_match(buf, "writemostly")) {
1873                 set_bit(WriteMostly, &rdev->flags);
1874                 err = 0;
1875         } else if (cmd_match(buf, "-writemostly")) {
1876                 clear_bit(WriteMostly, &rdev->flags);
1877                 err = 0;
1878         } else if (cmd_match(buf, "blocked")) {
1879                 set_bit(Blocked, &rdev->flags);
1880                 err = 0;
1881         } else if (cmd_match(buf, "-blocked")) {
1882                 clear_bit(Blocked, &rdev->flags);
1883                 wake_up(&rdev->blocked_wait);
1884                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1885                 md_wakeup_thread(rdev->mddev->thread);
1886
1887                 err = 0;
1888         }
1889         return err ? err : len;
1890 }
1891 static struct rdev_sysfs_entry rdev_state =
1892 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1893
1894 static ssize_t
1895 errors_show(mdk_rdev_t *rdev, char *page)
1896 {
1897         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1898 }
1899
1900 static ssize_t
1901 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1902 {
1903         char *e;
1904         unsigned long n = simple_strtoul(buf, &e, 10);
1905         if (*buf && (*e == 0 || *e == '\n')) {
1906                 atomic_set(&rdev->corrected_errors, n);
1907                 return len;
1908         }
1909         return -EINVAL;
1910 }
1911 static struct rdev_sysfs_entry rdev_errors =
1912 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1913
1914 static ssize_t
1915 slot_show(mdk_rdev_t *rdev, char *page)
1916 {
1917         if (rdev->raid_disk < 0)
1918                 return sprintf(page, "none\n");
1919         else
1920                 return sprintf(page, "%d\n", rdev->raid_disk);
1921 }
1922
1923 static ssize_t
1924 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1925 {
1926         char *e;
1927         int err;
1928         char nm[20];
1929         int slot = simple_strtoul(buf, &e, 10);
1930         if (strncmp(buf, "none", 4)==0)
1931                 slot = -1;
1932         else if (e==buf || (*e && *e!= '\n'))
1933                 return -EINVAL;
1934         if (rdev->mddev->pers) {
1935                 /* Setting 'slot' on an active array requires also
1936                  * updating the 'rd%d' link, and communicating
1937                  * with the personality with ->hot_*_disk.
1938                  * For now we only support removing
1939                  * failed/spare devices.  This normally happens automatically,
1940                  * but not when the metadata is externally managed.
1941                  */
1942                 if (slot != -1)
1943                         return -EBUSY;
1944                 if (rdev->raid_disk == -1)
1945                         return -EEXIST;
1946                 /* personality does all needed checks */
1947                 if (rdev->mddev->pers->hot_add_disk == NULL)
1948                         return -EINVAL;
1949                 err = rdev->mddev->pers->
1950                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
1951                 if (err)
1952                         return err;
1953                 sprintf(nm, "rd%d", rdev->raid_disk);
1954                 sysfs_remove_link(&rdev->mddev->kobj, nm);
1955                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1956                 md_wakeup_thread(rdev->mddev->thread);
1957         } else {
1958                 if (slot >= rdev->mddev->raid_disks)
1959                         return -ENOSPC;
1960                 rdev->raid_disk = slot;
1961                 /* assume it is working */
1962                 clear_bit(Faulty, &rdev->flags);
1963                 clear_bit(WriteMostly, &rdev->flags);
1964                 set_bit(In_sync, &rdev->flags);
1965         }
1966         return len;
1967 }
1968
1969
1970 static struct rdev_sysfs_entry rdev_slot =
1971 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1972
1973 static ssize_t
1974 offset_show(mdk_rdev_t *rdev, char *page)
1975 {
1976         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1977 }
1978
1979 static ssize_t
1980 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1981 {
1982         char *e;
1983         unsigned long long offset = simple_strtoull(buf, &e, 10);
1984         if (e==buf || (*e && *e != '\n'))
1985                 return -EINVAL;
1986         if (rdev->mddev->pers)
1987                 return -EBUSY;
1988         if (rdev->size && rdev->mddev->external)
1989                 /* Must set offset before size, so overlap checks
1990                  * can be sane */
1991                 return -EBUSY;
1992         rdev->data_offset = offset;
1993         return len;
1994 }
1995
1996 static struct rdev_sysfs_entry rdev_offset =
1997 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1998
1999 static ssize_t
2000 rdev_size_show(mdk_rdev_t *rdev, char *page)
2001 {
2002         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2003 }
2004
2005 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2006 {
2007         /* check if two start/length pairs overlap */
2008         if (s1+l1 <= s2)
2009                 return 0;
2010         if (s2+l2 <= s1)
2011                 return 0;
2012         return 1;
2013 }
2014
2015 static ssize_t
2016 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2017 {
2018         char *e;
2019         unsigned long long size = simple_strtoull(buf, &e, 10);
2020         unsigned long long oldsize = rdev->size;
2021         mddev_t *my_mddev = rdev->mddev;
2022
2023         if (e==buf || (*e && *e != '\n'))
2024                 return -EINVAL;
2025         if (my_mddev->pers)
2026                 return -EBUSY;
2027         rdev->size = size;
2028         if (size > oldsize && rdev->mddev->external) {
2029                 /* need to check that all other rdevs with the same ->bdev
2030                  * do not overlap.  We need to unlock the mddev to avoid
2031                  * a deadlock.  We have already changed rdev->size, and if
2032                  * we have to change it back, we will have the lock again.
2033                  */
2034                 mddev_t *mddev;
2035                 int overlap = 0;
2036                 struct list_head *tmp, *tmp2;
2037
2038                 mddev_unlock(my_mddev);
2039                 for_each_mddev(mddev, tmp) {
2040                         mdk_rdev_t *rdev2;
2041
2042                         mddev_lock(mddev);
2043                         rdev_for_each(rdev2, tmp2, mddev)
2044                                 if (test_bit(AllReserved, &rdev2->flags) ||
2045                                     (rdev->bdev == rdev2->bdev &&
2046                                      rdev != rdev2 &&
2047                                      overlaps(rdev->data_offset, rdev->size,
2048                                             rdev2->data_offset, rdev2->size))) {
2049                                         overlap = 1;
2050                                         break;
2051                                 }
2052                         mddev_unlock(mddev);
2053                         if (overlap) {
2054                                 mddev_put(mddev);
2055                                 break;
2056                         }
2057                 }
2058                 mddev_lock(my_mddev);
2059                 if (overlap) {
2060                         /* Someone else could have slipped in a size
2061                          * change here, but doing so is just silly.
2062                          * We put oldsize back because we *know* it is
2063                          * safe, and trust userspace not to race with
2064                          * itself
2065                          */
2066                         rdev->size = oldsize;
2067                         return -EBUSY;
2068                 }
2069         }
2070         if (size < my_mddev->size || my_mddev->size == 0)
2071                 my_mddev->size = size;
2072         return len;
2073 }
2074
2075 static struct rdev_sysfs_entry rdev_size =
2076 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2077
2078 static struct attribute *rdev_default_attrs[] = {
2079         &rdev_state.attr,
2080         &rdev_errors.attr,
2081         &rdev_slot.attr,
2082         &rdev_offset.attr,
2083         &rdev_size.attr,
2084         NULL,
2085 };
2086 static ssize_t
2087 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2088 {
2089         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2090         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2091         mddev_t *mddev = rdev->mddev;
2092         ssize_t rv;
2093
2094         if (!entry->show)
2095                 return -EIO;
2096
2097         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2098         if (!rv) {
2099                 if (rdev->mddev == NULL)
2100                         rv = -EBUSY;
2101                 else
2102                         rv = entry->show(rdev, page);
2103                 mddev_unlock(mddev);
2104         }
2105         return rv;
2106 }
2107
2108 static ssize_t
2109 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2110               const char *page, size_t length)
2111 {
2112         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2113         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2114         ssize_t rv;
2115         mddev_t *mddev = rdev->mddev;
2116
2117         if (!entry->store)
2118                 return -EIO;
2119         if (!capable(CAP_SYS_ADMIN))
2120                 return -EACCES;
2121         rv = mddev ? mddev_lock(mddev): -EBUSY;
2122         if (!rv) {
2123                 if (rdev->mddev == NULL)
2124                         rv = -EBUSY;
2125                 else
2126                         rv = entry->store(rdev, page, length);
2127                 mddev_unlock(mddev);
2128         }
2129         return rv;
2130 }
2131
2132 static void rdev_free(struct kobject *ko)
2133 {
2134         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2135         kfree(rdev);
2136 }
2137 static struct sysfs_ops rdev_sysfs_ops = {
2138         .show           = rdev_attr_show,
2139         .store          = rdev_attr_store,
2140 };
2141 static struct kobj_type rdev_ktype = {
2142         .release        = rdev_free,
2143         .sysfs_ops      = &rdev_sysfs_ops,
2144         .default_attrs  = rdev_default_attrs,
2145 };
2146
2147 /*
2148  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2149  *
2150  * mark the device faulty if:
2151  *
2152  *   - the device is nonexistent (zero size)
2153  *   - the device has no valid superblock
2154  *
2155  * a faulty rdev _never_ has rdev->sb set.
2156  */
2157 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2158 {
2159         char b[BDEVNAME_SIZE];
2160         int err;
2161         mdk_rdev_t *rdev;
2162         sector_t size;
2163
2164         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2165         if (!rdev) {
2166                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2167                 return ERR_PTR(-ENOMEM);
2168         }
2169
2170         if ((err = alloc_disk_sb(rdev)))
2171                 goto abort_free;
2172
2173         err = lock_rdev(rdev, newdev, super_format == -2);
2174         if (err)
2175                 goto abort_free;
2176
2177         kobject_init(&rdev->kobj, &rdev_ktype);
2178
2179         rdev->desc_nr = -1;
2180         rdev->saved_raid_disk = -1;
2181         rdev->raid_disk = -1;
2182         rdev->flags = 0;
2183         rdev->data_offset = 0;
2184         rdev->sb_events = 0;
2185         atomic_set(&rdev->nr_pending, 0);
2186         atomic_set(&rdev->read_errors, 0);
2187         atomic_set(&rdev->corrected_errors, 0);
2188
2189         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2190         if (!size) {
2191                 printk(KERN_WARNING
2192                         "md: %s has zero or unknown size, marking faulty!\n",
2193                         bdevname(rdev->bdev,b));
2194                 err = -EINVAL;
2195                 goto abort_free;
2196         }
2197
2198         if (super_format >= 0) {
2199                 err = super_types[super_format].
2200                         load_super(rdev, NULL, super_minor);
2201                 if (err == -EINVAL) {
2202                         printk(KERN_WARNING
2203                                 "md: %s does not have a valid v%d.%d "
2204                                "superblock, not importing!\n",
2205                                 bdevname(rdev->bdev,b),
2206                                super_format, super_minor);
2207                         goto abort_free;
2208                 }
2209                 if (err < 0) {
2210                         printk(KERN_WARNING
2211                                 "md: could not read %s's sb, not importing!\n",
2212                                 bdevname(rdev->bdev,b));
2213                         goto abort_free;
2214                 }
2215         }
2216
2217         INIT_LIST_HEAD(&rdev->same_set);
2218         init_waitqueue_head(&rdev->blocked_wait);
2219
2220         return rdev;
2221
2222 abort_free:
2223         if (rdev->sb_page) {
2224                 if (rdev->bdev)
2225                         unlock_rdev(rdev);
2226                 free_disk_sb(rdev);
2227         }
2228         kfree(rdev);
2229         return ERR_PTR(err);
2230 }
2231
2232 /*
2233  * Check a full RAID array for plausibility
2234  */
2235
2236
2237 static void analyze_sbs(mddev_t * mddev)
2238 {
2239         int i;
2240         struct list_head *tmp;
2241         mdk_rdev_t *rdev, *freshest;
2242         char b[BDEVNAME_SIZE];
2243
2244         freshest = NULL;
2245         rdev_for_each(rdev, tmp, mddev)
2246                 switch (super_types[mddev->major_version].
2247                         load_super(rdev, freshest, mddev->minor_version)) {
2248                 case 1:
2249                         freshest = rdev;
2250                         break;
2251                 case 0:
2252                         break;
2253                 default:
2254                         printk( KERN_ERR \
2255                                 "md: fatal superblock inconsistency in %s"
2256                                 " -- removing from array\n",
2257                                 bdevname(rdev->bdev,b));
2258                         kick_rdev_from_array(rdev);
2259                 }
2260
2261
2262         super_types[mddev->major_version].
2263                 validate_super(mddev, freshest);
2264
2265         i = 0;
2266         rdev_for_each(rdev, tmp, mddev) {
2267                 if (rdev != freshest)
2268                         if (super_types[mddev->major_version].
2269                             validate_super(mddev, rdev)) {
2270                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2271                                         " from array!\n",
2272                                         bdevname(rdev->bdev,b));
2273                                 kick_rdev_from_array(rdev);
2274                                 continue;
2275                         }
2276                 if (mddev->level == LEVEL_MULTIPATH) {
2277                         rdev->desc_nr = i++;
2278                         rdev->raid_disk = rdev->desc_nr;
2279                         set_bit(In_sync, &rdev->flags);
2280                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2281                         rdev->raid_disk = -1;
2282                         clear_bit(In_sync, &rdev->flags);
2283                 }
2284         }
2285
2286
2287
2288         if (mddev->recovery_cp != MaxSector &&
2289             mddev->level >= 1)
2290                 printk(KERN_ERR "md: %s: raid array is not clean"
2291                        " -- starting background reconstruction\n",
2292                        mdname(mddev));
2293
2294 }
2295
2296 static ssize_t
2297 safe_delay_show(mddev_t *mddev, char *page)
2298 {
2299         int msec = (mddev->safemode_delay*1000)/HZ;
2300         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2301 }
2302 static ssize_t
2303 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2304 {
2305         int scale=1;
2306         int dot=0;
2307         int i;
2308         unsigned long msec;
2309         char buf[30];
2310         char *e;
2311         /* remove a period, and count digits after it */
2312         if (len >= sizeof(buf))
2313                 return -EINVAL;
2314         strlcpy(buf, cbuf, len);
2315         buf[len] = 0;
2316         for (i=0; i<len; i++) {
2317                 if (dot) {
2318                         if (isdigit(buf[i])) {
2319                                 buf[i-1] = buf[i];
2320                                 scale *= 10;
2321                         }
2322                         buf[i] = 0;
2323                 } else if (buf[i] == '.') {
2324                         dot=1;
2325                         buf[i] = 0;
2326                 }
2327         }
2328         msec = simple_strtoul(buf, &e, 10);
2329         if (e == buf || (*e && *e != '\n'))
2330                 return -EINVAL;
2331         msec = (msec * 1000) / scale;
2332         if (msec == 0)
2333                 mddev->safemode_delay = 0;
2334         else {
2335                 mddev->safemode_delay = (msec*HZ)/1000;
2336                 if (mddev->safemode_delay == 0)
2337                         mddev->safemode_delay = 1;
2338         }
2339         return len;
2340 }
2341 static struct md_sysfs_entry md_safe_delay =
2342 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2343
2344 static ssize_t
2345 level_show(mddev_t *mddev, char *page)
2346 {
2347         struct mdk_personality *p = mddev->pers;
2348         if (p)
2349                 return sprintf(page, "%s\n", p->name);
2350         else if (mddev->clevel[0])
2351                 return sprintf(page, "%s\n", mddev->clevel);
2352         else if (mddev->level != LEVEL_NONE)
2353                 return sprintf(page, "%d\n", mddev->level);
2354         else
2355                 return 0;
2356 }
2357
2358 static ssize_t
2359 level_store(mddev_t *mddev, const char *buf, size_t len)
2360 {
2361         ssize_t rv = len;
2362         if (mddev->pers)
2363                 return -EBUSY;
2364         if (len == 0)
2365                 return 0;
2366         if (len >= sizeof(mddev->clevel))
2367                 return -ENOSPC;
2368         strncpy(mddev->clevel, buf, len);
2369         if (mddev->clevel[len-1] == '\n')
2370                 len--;
2371         mddev->clevel[len] = 0;
2372         mddev->level = LEVEL_NONE;
2373         return rv;
2374 }
2375
2376 static struct md_sysfs_entry md_level =
2377 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2378
2379
2380 static ssize_t
2381 layout_show(mddev_t *mddev, char *page)
2382 {
2383         /* just a number, not meaningful for all levels */
2384         if (mddev->reshape_position != MaxSector &&
2385             mddev->layout != mddev->new_layout)
2386                 return sprintf(page, "%d (%d)\n",
2387                                mddev->new_layout, mddev->layout);
2388         return sprintf(page, "%d\n", mddev->layout);
2389 }
2390
2391 static ssize_t
2392 layout_store(mddev_t *mddev, const char *buf, size_t len)
2393 {
2394         char *e;
2395         unsigned long n = simple_strtoul(buf, &e, 10);
2396
2397         if (!*buf || (*e && *e != '\n'))
2398                 return -EINVAL;
2399
2400         if (mddev->pers)
2401                 return -EBUSY;
2402         if (mddev->reshape_position != MaxSector)
2403                 mddev->new_layout = n;
2404         else
2405                 mddev->layout = n;
2406         return len;
2407 }
2408 static struct md_sysfs_entry md_layout =
2409 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2410
2411
2412 static ssize_t
2413 raid_disks_show(mddev_t *mddev, char *page)
2414 {
2415         if (mddev->raid_disks == 0)
2416                 return 0;
2417         if (mddev->reshape_position != MaxSector &&
2418             mddev->delta_disks != 0)
2419                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2420                                mddev->raid_disks - mddev->delta_disks);
2421         return sprintf(page, "%d\n", mddev->raid_disks);
2422 }
2423
2424 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2425
2426 static ssize_t
2427 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2428 {
2429         char *e;
2430         int rv = 0;
2431         unsigned long n = simple_strtoul(buf, &e, 10);
2432
2433         if (!*buf || (*e && *e != '\n'))
2434                 return -EINVAL;
2435
2436         if (mddev->pers)
2437                 rv = update_raid_disks(mddev, n);
2438         else if (mddev->reshape_position != MaxSector) {
2439                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2440                 mddev->delta_disks = n - olddisks;
2441                 mddev->raid_disks = n;
2442         } else
2443                 mddev->raid_disks = n;
2444         return rv ? rv : len;
2445 }
2446 static struct md_sysfs_entry md_raid_disks =
2447 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2448
2449 static ssize_t
2450 chunk_size_show(mddev_t *mddev, char *page)
2451 {
2452         if (mddev->reshape_position != MaxSector &&
2453             mddev->chunk_size != mddev->new_chunk)
2454                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2455                                mddev->chunk_size);
2456         return sprintf(page, "%d\n", mddev->chunk_size);
2457 }
2458
2459 static ssize_t
2460 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2461 {
2462         /* can only set chunk_size if array is not yet active */
2463         char *e;
2464         unsigned long n = simple_strtoul(buf, &e, 10);
2465
2466         if (!*buf || (*e && *e != '\n'))
2467                 return -EINVAL;
2468
2469         if (mddev->pers)
2470                 return -EBUSY;
2471         else if (mddev->reshape_position != MaxSector)
2472                 mddev->new_chunk = n;
2473         else
2474                 mddev->chunk_size = n;
2475         return len;
2476 }
2477 static struct md_sysfs_entry md_chunk_size =
2478 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2479
2480 static ssize_t
2481 resync_start_show(mddev_t *mddev, char *page)
2482 {
2483         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2484 }
2485
2486 static ssize_t
2487 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2488 {
2489         char *e;
2490         unsigned long long n = simple_strtoull(buf, &e, 10);
2491
2492         if (mddev->pers)
2493                 return -EBUSY;
2494         if (!*buf || (*e && *e != '\n'))
2495                 return -EINVAL;
2496
2497         mddev->recovery_cp = n;
2498         return len;
2499 }
2500 static struct md_sysfs_entry md_resync_start =
2501 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2502
2503 /*
2504  * The array state can be:
2505  *
2506  * clear
2507  *     No devices, no size, no level
2508  *     Equivalent to STOP_ARRAY ioctl
2509  * inactive
2510  *     May have some settings, but array is not active
2511  *        all IO results in error
2512  *     When written, doesn't tear down array, but just stops it
2513  * suspended (not supported yet)
2514  *     All IO requests will block. The array can be reconfigured.
2515  *     Writing this, if accepted, will block until array is quiessent
2516  * readonly
2517  *     no resync can happen.  no superblocks get written.
2518  *     write requests fail
2519  * read-auto
2520  *     like readonly, but behaves like 'clean' on a write request.
2521  *
2522  * clean - no pending writes, but otherwise active.
2523  *     When written to inactive array, starts without resync
2524  *     If a write request arrives then
2525  *       if metadata is known, mark 'dirty' and switch to 'active'.
2526  *       if not known, block and switch to write-pending
2527  *     If written to an active array that has pending writes, then fails.
2528  * active
2529  *     fully active: IO and resync can be happening.
2530  *     When written to inactive array, starts with resync
2531  *
2532  * write-pending
2533  *     clean, but writes are blocked waiting for 'active' to be written.
2534  *
2535  * active-idle
2536  *     like active, but no writes have been seen for a while (100msec).
2537  *
2538  */
2539 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2540                    write_pending, active_idle, bad_word};
2541 static char *array_states[] = {
2542         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2543         "write-pending", "active-idle", NULL };
2544
2545 static int match_word(const char *word, char **list)
2546 {
2547         int n;
2548         for (n=0; list[n]; n++)
2549                 if (cmd_match(word, list[n]))
2550                         break;
2551         return n;
2552 }
2553
2554 static ssize_t
2555 array_state_show(mddev_t *mddev, char *page)
2556 {
2557         enum array_state st = inactive;
2558
2559         if (mddev->pers)
2560                 switch(mddev->ro) {
2561                 case 1:
2562                         st = readonly;
2563                         break;
2564                 case 2:
2565                         st = read_auto;
2566                         break;
2567                 case 0:
2568                         if (mddev->in_sync)
2569                                 st = clean;
2570                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2571                                 st = write_pending;
2572                         else if (mddev->safemode)
2573                                 st = active_idle;
2574                         else
2575                                 st = active;
2576                 }
2577         else {
2578                 if (list_empty(&mddev->disks) &&
2579                     mddev->raid_disks == 0 &&
2580                     mddev->size == 0)
2581                         st = clear;
2582                 else
2583                         st = inactive;
2584         }
2585         return sprintf(page, "%s\n", array_states[st]);
2586 }
2587
2588 static int do_md_stop(mddev_t * mddev, int ro);
2589 static int do_md_run(mddev_t * mddev);
2590 static int restart_array(mddev_t *mddev);
2591
2592 static ssize_t
2593 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2594 {
2595         int err = -EINVAL;
2596         enum array_state st = match_word(buf, array_states);
2597         switch(st) {
2598         case bad_word:
2599                 break;
2600         case clear:
2601                 /* stopping an active array */
2602                 if (atomic_read(&mddev->active) > 1)
2603                         return -EBUSY;
2604                 err = do_md_stop(mddev, 0);
2605                 break;
2606         case inactive:
2607                 /* stopping an active array */
2608                 if (mddev->pers) {
2609                         if (atomic_read(&mddev->active) > 1)
2610                                 return -EBUSY;
2611                         err = do_md_stop(mddev, 2);
2612                 } else
2613                         err = 0; /* already inactive */
2614                 break;
2615         case suspended:
2616                 break; /* not supported yet */
2617         case readonly:
2618                 if (mddev->pers)
2619                         err = do_md_stop(mddev, 1);
2620                 else {
2621                         mddev->ro = 1;
2622                         set_disk_ro(mddev->gendisk, 1);
2623                         err = do_md_run(mddev);
2624                 }
2625                 break;
2626         case read_auto:
2627                 if (mddev->pers) {
2628                         if (mddev->ro != 1)
2629                                 err = do_md_stop(mddev, 1);
2630                         else
2631                                 err = restart_array(mddev);
2632                         if (err == 0) {
2633                                 mddev->ro = 2;
2634                                 set_disk_ro(mddev->gendisk, 0);
2635                         }
2636                 } else {
2637                         mddev->ro = 2;
2638                         err = do_md_run(mddev);
2639                 }
2640                 break;
2641         case clean:
2642                 if (mddev->pers) {
2643                         restart_array(mddev);
2644                         spin_lock_irq(&mddev->write_lock);
2645                         if (atomic_read(&mddev->writes_pending) == 0) {
2646                                 if (mddev->in_sync == 0) {
2647                                         mddev->in_sync = 1;
2648                                         if (mddev->safemode == 1)
2649                                                 mddev->safemode = 0;
2650                                         if (mddev->persistent)
2651                                                 set_bit(MD_CHANGE_CLEAN,
2652                                                         &mddev->flags);
2653                                 }
2654                                 err = 0;
2655                         } else
2656                                 err = -EBUSY;
2657                         spin_unlock_irq(&mddev->write_lock);
2658                 } else {
2659                         mddev->ro = 0;
2660                         mddev->recovery_cp = MaxSector;
2661                         err = do_md_run(mddev);
2662                 }
2663                 break;
2664         case active:
2665                 if (mddev->pers) {
2666                         restart_array(mddev);
2667                         if (mddev->external)
2668                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2669                         wake_up(&mddev->sb_wait);
2670                         err = 0;
2671                 } else {
2672                         mddev->ro = 0;
2673                         set_disk_ro(mddev->gendisk, 0);
2674                         err = do_md_run(mddev);
2675                 }
2676                 break;
2677         case write_pending:
2678         case active_idle:
2679                 /* these cannot be set */
2680                 break;
2681         }
2682         if (err)
2683                 return err;
2684         else
2685                 return len;
2686 }
2687 static struct md_sysfs_entry md_array_state =
2688 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2689
2690 static ssize_t
2691 null_show(mddev_t *mddev, char *page)
2692 {
2693         return -EINVAL;
2694 }
2695
2696 static ssize_t
2697 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2698 {
2699         /* buf must be %d:%d\n? giving major and minor numbers */
2700         /* The new device is added to the array.
2701          * If the array has a persistent superblock, we read the
2702          * superblock to initialise info and check validity.
2703          * Otherwise, only checking done is that in bind_rdev_to_array,
2704          * which mainly checks size.
2705          */
2706         char *e;
2707         int major = simple_strtoul(buf, &e, 10);
2708         int minor;
2709         dev_t dev;
2710         mdk_rdev_t *rdev;
2711         int err;
2712
2713         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2714                 return -EINVAL;
2715         minor = simple_strtoul(e+1, &e, 10);
2716         if (*e && *e != '\n')
2717                 return -EINVAL;
2718         dev = MKDEV(major, minor);
2719         if (major != MAJOR(dev) ||
2720             minor != MINOR(dev))
2721                 return -EOVERFLOW;
2722
2723
2724         if (mddev->persistent) {
2725                 rdev = md_import_device(dev, mddev->major_version,
2726                                         mddev->minor_version);
2727                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2728                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2729                                                        mdk_rdev_t, same_set);
2730                         err = super_types[mddev->major_version]
2731                                 .load_super(rdev, rdev0, mddev->minor_version);
2732                         if (err < 0)
2733                                 goto out;
2734                 }
2735         } else if (mddev->external)
2736                 rdev = md_import_device(dev, -2, -1);
2737         else
2738                 rdev = md_import_device(dev, -1, -1);
2739
2740         if (IS_ERR(rdev))
2741                 return PTR_ERR(rdev);
2742         err = bind_rdev_to_array(rdev, mddev);
2743  out:
2744         if (err)
2745                 export_rdev(rdev);
2746         return err ? err : len;
2747 }
2748
2749 static struct md_sysfs_entry md_new_device =
2750 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2751
2752 static ssize_t
2753 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2754 {
2755         char *end;
2756         unsigned long chunk, end_chunk;
2757
2758         if (!mddev->bitmap)
2759                 goto out;
2760         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2761         while (*buf) {
2762                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2763                 if (buf == end) break;
2764                 if (*end == '-') { /* range */
2765                         buf = end + 1;
2766                         end_chunk = simple_strtoul(buf, &end, 0);
2767                         if (buf == end) break;
2768                 }
2769                 if (*end && !isspace(*end)) break;
2770                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2771                 buf = end;
2772                 while (isspace(*buf)) buf++;
2773         }
2774         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2775 out:
2776         return len;
2777 }
2778
2779 static struct md_sysfs_entry md_bitmap =
2780 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2781
2782 static ssize_t
2783 size_show(mddev_t *mddev, char *page)
2784 {
2785         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2786 }
2787
2788 static int update_size(mddev_t *mddev, unsigned long size);
2789
2790 static ssize_t
2791 size_store(mddev_t *mddev, const char *buf, size_t len)
2792 {
2793         /* If array is inactive, we can reduce the component size, but
2794          * not increase it (except from 0).
2795          * If array is active, we can try an on-line resize
2796          */
2797         char *e;
2798         int err = 0;
2799         unsigned long long size = simple_strtoull(buf, &e, 10);
2800         if (!*buf || *buf == '\n' ||
2801             (*e && *e != '\n'))
2802                 return -EINVAL;
2803
2804         if (mddev->pers) {
2805                 err = update_size(mddev, size);
2806                 md_update_sb(mddev, 1);
2807         } else {
2808                 if (mddev->size == 0 ||
2809                     mddev->size > size)
2810                         mddev->size = size;
2811                 else
2812                         err = -ENOSPC;
2813         }
2814         return err ? err : len;
2815 }
2816
2817 static struct md_sysfs_entry md_size =
2818 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2819
2820
2821 /* Metdata version.
2822  * This is one of
2823  *   'none' for arrays with no metadata (good luck...)
2824  *   'external' for arrays with externally managed metadata,
2825  * or N.M for internally known formats
2826  */
2827 static ssize_t
2828 metadata_show(mddev_t *mddev, char *page)
2829 {
2830         if (mddev->persistent)
2831                 return sprintf(page, "%d.%d\n",
2832                                mddev->major_version, mddev->minor_version);
2833         else if (mddev->external)
2834                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2835         else
2836                 return sprintf(page, "none\n");
2837 }
2838
2839 static ssize_t
2840 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2841 {
2842         int major, minor;
2843         char *e;
2844         if (!list_empty(&mddev->disks))
2845                 return -EBUSY;
2846
2847         if (cmd_match(buf, "none")) {
2848                 mddev->persistent = 0;
2849                 mddev->external = 0;
2850                 mddev->major_version = 0;
2851                 mddev->minor_version = 90;
2852                 return len;
2853         }
2854         if (strncmp(buf, "external:", 9) == 0) {
2855                 size_t namelen = len-9;
2856                 if (namelen >= sizeof(mddev->metadata_type))
2857                         namelen = sizeof(mddev->metadata_type)-1;
2858                 strncpy(mddev->metadata_type, buf+9, namelen);
2859                 mddev->metadata_type[namelen] = 0;
2860                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2861                         mddev->metadata_type[--namelen] = 0;
2862                 mddev->persistent = 0;
2863                 mddev->external = 1;
2864                 mddev->major_version = 0;
2865                 mddev->minor_version = 90;
2866                 return len;
2867         }
2868         major = simple_strtoul(buf, &e, 10);
2869         if (e==buf || *e != '.')
2870                 return -EINVAL;
2871         buf = e+1;
2872         minor = simple_strtoul(buf, &e, 10);
2873         if (e==buf || (*e && *e != '\n') )
2874                 return -EINVAL;
2875         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2876                 return -ENOENT;
2877         mddev->major_version = major;
2878         mddev->minor_version = minor;
2879         mddev->persistent = 1;
2880         mddev->external = 0;
2881         return len;
2882 }
2883
2884 static struct md_sysfs_entry md_metadata =
2885 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2886
2887 static ssize_t
2888 action_show(mddev_t *mddev, char *page)
2889 {
2890         char *type = "idle";
2891         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2892             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2893                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2894                         type = "reshape";
2895                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2896                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2897                                 type = "resync";
2898                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2899                                 type = "check";
2900                         else
2901                                 type = "repair";
2902                 } else
2903                         type = "recover";
2904         }
2905         return sprintf(page, "%s\n", type);
2906 }
2907
2908 static ssize_t
2909 action_store(mddev_t *mddev, const char *page, size_t len)
2910 {
2911         if (!mddev->pers || !mddev->pers->sync_request)
2912                 return -EINVAL;
2913
2914         if (cmd_match(page, "idle")) {
2915                 if (mddev->sync_thread) {
2916                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2917                         md_unregister_thread(mddev->sync_thread);
2918                         mddev->sync_thread = NULL;
2919                         mddev->recovery = 0;
2920                 }
2921         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2922                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2923                 return -EBUSY;
2924         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2925                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2926         else if (cmd_match(page, "reshape")) {
2927                 int err;
2928                 if (mddev->pers->start_reshape == NULL)
2929                         return -EINVAL;
2930                 err = mddev->pers->start_reshape(mddev);
2931                 if (err)
2932                         return err;
2933         } else {
2934                 if (cmd_match(page, "check"))
2935                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2936                 else if (!cmd_match(page, "repair"))
2937                         return -EINVAL;
2938                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2939                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2940         }
2941         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2942         md_wakeup_thread(mddev->thread);
2943         return len;
2944 }
2945
2946 static ssize_t
2947 mismatch_cnt_show(mddev_t *mddev, char *page)
2948 {
2949         return sprintf(page, "%llu\n",
2950                        (unsigned long long) mddev->resync_mismatches);
2951 }
2952
2953 static struct md_sysfs_entry md_scan_mode =
2954 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2955
2956
2957 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2958
2959 static ssize_t
2960 sync_min_show(mddev_t *mddev, char *page)
2961 {
2962         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2963                        mddev->sync_speed_min ? "local": "system");
2964 }
2965
2966 static ssize_t
2967 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2968 {
2969         int min;
2970         char *e;
2971         if (strncmp(buf, "system", 6)==0) {
2972                 mddev->sync_speed_min = 0;
2973                 return len;
2974         }
2975         min = simple_strtoul(buf, &e, 10);
2976         if (buf == e || (*e && *e != '\n') || min <= 0)
2977                 return -EINVAL;
2978         mddev->sync_speed_min = min;
2979         return len;
2980 }
2981
2982 static struct md_sysfs_entry md_sync_min =
2983 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2984
2985 static ssize_t
2986 sync_max_show(mddev_t *mddev, char *page)
2987 {
2988         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2989                        mddev->sync_speed_max ? "local": "system");
2990 }
2991
2992 static ssize_t
2993 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2994 {
2995         int max;
2996         char *e;
2997         if (strncmp(buf, "system", 6)==0) {
2998                 mddev->sync_speed_max = 0;
2999                 return len;
3000         }
3001         max = simple_strtoul(buf, &e, 10);
3002         if (buf == e || (*e && *e != '\n') || max <= 0)
3003                 return -EINVAL;
3004         mddev->sync_speed_max = max;
3005         return len;
3006 }
3007
3008 static struct md_sysfs_entry md_sync_max =
3009 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3010
3011 static ssize_t
3012 degraded_show(mddev_t *mddev, char *page)
3013 {
3014         return sprintf(page, "%d\n", mddev->degraded);
3015 }
3016 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3017
3018 static ssize_t
3019 sync_force_parallel_show(mddev_t *mddev, char *page)
3020 {
3021         return sprintf(page, "%d\n", mddev->parallel_resync);
3022 }
3023
3024 static ssize_t
3025 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3026 {
3027         long n;
3028
3029         if (strict_strtol(buf, 10, &n))
3030                 return -EINVAL;
3031
3032         if (n != 0 && n != 1)
3033                 return -EINVAL;
3034
3035         mddev->parallel_resync = n;
3036
3037         if (mddev->sync_thread)
3038                 wake_up(&resync_wait);
3039
3040         return len;
3041 }
3042
3043 /* force parallel resync, even with shared block devices */
3044 static struct md_sysfs_entry md_sync_force_parallel =
3045 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3046        sync_force_parallel_show, sync_force_parallel_store);
3047
3048 static ssize_t
3049 sync_speed_show(mddev_t *mddev, char *page)
3050 {
3051         unsigned long resync, dt, db;
3052         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
3053         dt = ((jiffies - mddev->resync_mark) / HZ);
3054         if (!dt) dt++;
3055         db = resync - (mddev->resync_mark_cnt);
3056         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
3057 }
3058
3059 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3060
3061 static ssize_t
3062 sync_completed_show(mddev_t *mddev, char *page)
3063 {
3064         unsigned long max_blocks, resync;
3065
3066         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3067                 max_blocks = mddev->resync_max_sectors;
3068         else
3069                 max_blocks = mddev->size << 1;
3070
3071         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3072         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3073 }
3074
3075 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3076
3077 static ssize_t
3078 max_sync_show(mddev_t *mddev, char *page)
3079 {
3080         if (mddev->resync_max == MaxSector)
3081                 return sprintf(page, "max\n");
3082         else
3083                 return sprintf(page, "%llu\n",
3084                                (unsigned long long)mddev->resync_max);
3085 }
3086 static ssize_t
3087 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3088 {
3089         if (strncmp(buf, "max", 3) == 0)
3090                 mddev->resync_max = MaxSector;
3091         else {
3092                 char *ep;
3093                 unsigned long long max = simple_strtoull(buf, &ep, 10);
3094                 if (ep == buf || (*ep != 0 && *ep != '\n'))
3095                         return -EINVAL;
3096                 if (max < mddev->resync_max &&
3097                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3098                         return -EBUSY;
3099
3100                 /* Must be a multiple of chunk_size */
3101                 if (mddev->chunk_size) {
3102                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3103                                 return -EINVAL;
3104                 }
3105                 mddev->resync_max = max;
3106         }
3107         wake_up(&mddev->recovery_wait);
3108         return len;
3109 }
3110
3111 static struct md_sysfs_entry md_max_sync =
3112 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3113
3114 static ssize_t
3115 suspend_lo_show(mddev_t *mddev, char *page)
3116 {
3117         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3118 }
3119
3120 static ssize_t
3121 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3122 {
3123         char *e;
3124         unsigned long long new = simple_strtoull(buf, &e, 10);
3125
3126         if (mddev->pers->quiesce == NULL)
3127                 return -EINVAL;
3128         if (buf == e || (*e && *e != '\n'))
3129                 return -EINVAL;
3130         if (new >= mddev->suspend_hi ||
3131             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3132                 mddev->suspend_lo = new;
3133                 mddev->pers->quiesce(mddev, 2);
3134                 return len;
3135         } else
3136                 return -EINVAL;
3137 }
3138 static struct md_sysfs_entry md_suspend_lo =
3139 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3140
3141
3142 static ssize_t
3143 suspend_hi_show(mddev_t *mddev, char *page)
3144 {
3145         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3146 }
3147
3148 static ssize_t
3149 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3150 {
3151         char *e;
3152         unsigned long long new = simple_strtoull(buf, &e, 10);
3153
3154         if (mddev->pers->quiesce == NULL)
3155                 return -EINVAL;
3156         if (buf == e || (*e && *e != '\n'))
3157                 return -EINVAL;
3158         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3159             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3160                 mddev->suspend_hi = new;
3161                 mddev->pers->quiesce(mddev, 1);
3162                 mddev->pers->quiesce(mddev, 0);
3163                 return len;
3164         } else
3165                 return -EINVAL;
3166 }
3167 static struct md_sysfs_entry md_suspend_hi =
3168 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3169
3170 static ssize_t
3171 reshape_position_show(mddev_t *mddev, char *page)
3172 {
3173         if (mddev->reshape_position != MaxSector)
3174                 return sprintf(page, "%llu\n",
3175                                (unsigned long long)mddev->reshape_position);
3176         strcpy(page, "none\n");
3177         return 5;
3178 }
3179
3180 static ssize_t
3181 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3182 {
3183         char *e;
3184         unsigned long long new = simple_strtoull(buf, &e, 10);
3185         if (mddev->pers)
3186                 return -EBUSY;
3187         if (buf == e || (*e && *e != '\n'))
3188                 return -EINVAL;
3189         mddev->reshape_position = new;
3190         mddev->delta_disks = 0;
3191         mddev->new_level = mddev->level;
3192         mddev->new_layout = mddev->layout;
3193         mddev->new_chunk = mddev->chunk_size;
3194         return len;
3195 }
3196
3197 static struct md_sysfs_entry md_reshape_position =
3198 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3199        reshape_position_store);
3200
3201
3202 static struct attribute *md_default_attrs[] = {
3203         &md_level.attr,
3204         &md_layout.attr,
3205         &md_raid_disks.attr,
3206         &md_chunk_size.attr,
3207         &md_size.attr,
3208         &md_resync_start.attr,
3209         &md_metadata.attr,
3210         &md_new_device.attr,
3211         &md_safe_delay.attr,
3212         &md_array_state.attr,
3213         &md_reshape_position.attr,
3214         NULL,
3215 };
3216
3217 static struct attribute *md_redundancy_attrs[] = {
3218         &md_scan_mode.attr,
3219         &md_mismatches.attr,
3220         &md_sync_min.attr,
3221         &md_sync_max.attr,
3222         &md_sync_speed.attr,
3223         &md_sync_force_parallel.attr,
3224         &md_sync_completed.attr,
3225         &md_max_sync.attr,
3226         &md_suspend_lo.attr,
3227         &md_suspend_hi.attr,
3228         &md_bitmap.attr,
3229         &md_degraded.attr,
3230         NULL,
3231 };
3232 static struct attribute_group md_redundancy_group = {
3233         .name = NULL,
3234         .attrs = md_redundancy_attrs,
3235 };
3236
3237
3238 static ssize_t
3239 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3240 {
3241         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3242         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3243         ssize_t rv;
3244
3245         if (!entry->show)
3246                 return -EIO;
3247         rv = mddev_lock(mddev);
3248         if (!rv) {
3249                 rv = entry->show(mddev, page);
3250                 mddev_unlock(mddev);
3251         }
3252         return rv;
3253 }
3254
3255 static ssize_t
3256 md_attr_store(struct kobject *kobj, struct attribute *attr,
3257               const char *page, size_t length)
3258 {
3259         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3260         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3261         ssize_t rv;
3262
3263         if (!entry->store)
3264                 return -EIO;
3265         if (!capable(CAP_SYS_ADMIN))
3266                 return -EACCES;
3267         rv = mddev_lock(mddev);
3268         if (!rv) {
3269                 rv = entry->store(mddev, page, length);
3270                 mddev_unlock(mddev);
3271         }
3272         return rv;
3273 }
3274
3275 static void md_free(struct kobject *ko)
3276 {
3277         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3278         kfree(mddev);
3279 }
3280
3281 static struct sysfs_ops md_sysfs_ops = {
3282         .show   = md_attr_show,
3283         .store  = md_attr_store,
3284 };
3285 static struct kobj_type md_ktype = {
3286         .release        = md_free,
3287         .sysfs_ops      = &md_sysfs_ops,
3288         .default_attrs  = md_default_attrs,
3289 };
3290
3291 int mdp_major = 0;
3292
3293 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3294 {
3295         static DEFINE_MUTEX(disks_mutex);
3296         mddev_t *mddev = mddev_find(dev);
3297         struct gendisk *disk;
3298         int partitioned = (MAJOR(dev) != MD_MAJOR);
3299         int shift = partitioned ? MdpMinorShift : 0;
3300         int unit = MINOR(dev) >> shift;
3301         int error;
3302
3303         if (!mddev)
3304                 return NULL;
3305
3306         mutex_lock(&disks_mutex);
3307         if (mddev->gendisk) {
3308                 mutex_unlock(&disks_mutex);
3309                 mddev_put(mddev);
3310                 return NULL;
3311         }
3312         disk = alloc_disk(1 << shift);
3313         if (!disk) {
3314                 mutex_unlock(&disks_mutex);
3315                 mddev_put(mddev);
3316                 return NULL;
3317         }
3318         disk->major = MAJOR(dev);
3319         disk->first_minor = unit << shift;
3320         if (partitioned)
3321                 sprintf(disk->disk_name, "md_d%d", unit);
3322         else
3323                 sprintf(disk->disk_name, "md%d", unit);
3324         disk->fops = &md_fops;
3325         disk->private_data = mddev;
3326         disk->queue = mddev->queue;
3327         add_disk(disk);
3328         mddev->gendisk = disk;
3329         mutex_unlock(&disks_mutex);
3330         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3331                                      "%s", "md");
3332         if (error)
3333                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3334                        disk->disk_name);
3335         else
3336                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3337         return NULL;
3338 }
3339
3340 static void md_safemode_timeout(unsigned long data)
3341 {
3342         mddev_t *mddev = (mddev_t *) data;
3343
3344         mddev->safemode = 1;
3345         md_wakeup_thread(mddev->thread);
3346 }
3347
3348 static int start_dirty_degraded;
3349
3350 static int do_md_run(mddev_t * mddev)
3351 {
3352         int err;
3353         int chunk_size;
3354         struct list_head *tmp;
3355         mdk_rdev_t *rdev;
3356         struct gendisk *disk;
3357         struct mdk_personality *pers;
3358         char b[BDEVNAME_SIZE];
3359
3360         if (list_empty(&mddev->disks))
3361                 /* cannot run an array with no devices.. */
3362                 return -EINVAL;
3363
3364         if (mddev->pers)
3365                 return -EBUSY;
3366
3367         /*
3368          * Analyze all RAID superblock(s)
3369          */
3370         if (!mddev->raid_disks) {
3371                 if (!mddev->persistent)
3372                         return -EINVAL;
3373                 analyze_sbs(mddev);
3374         }
3375
3376         chunk_size = mddev->chunk_size;
3377
3378         if (chunk_size) {
3379                 if (chunk_size > MAX_CHUNK_SIZE) {
3380                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3381                                 chunk_size, MAX_CHUNK_SIZE);
3382                         return -EINVAL;
3383                 }
3384                 /*
3385                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3386                  */
3387                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3388                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3389                         return -EINVAL;
3390                 }
3391                 if (chunk_size < PAGE_SIZE) {
3392                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3393                                 chunk_size, PAGE_SIZE);
3394                         return -EINVAL;
3395                 }
3396
3397                 /* devices must have minimum size of one chunk */
3398                 rdev_for_each(rdev, tmp, mddev) {
3399                         if (test_bit(Faulty, &rdev->flags))
3400                                 continue;
3401                         if (rdev->size < chunk_size / 1024) {
3402                                 printk(KERN_WARNING
3403                                         "md: Dev %s smaller than chunk_size:"
3404                                         " %lluk < %dk\n",
3405                                         bdevname(rdev->bdev,b),
3406                                         (unsigned long long)rdev->size,
3407                                         chunk_size / 1024);
3408                                 return -EINVAL;
3409                         }
3410                 }
3411         }
3412
3413 #ifdef CONFIG_KMOD
3414         if (mddev->level != LEVEL_NONE)
3415                 request_module("md-level-%d", mddev->level);
3416         else if (mddev->clevel[0])
3417                 request_module("md-%s", mddev->clevel);
3418 #endif
3419
3420         /*
3421          * Drop all container device buffers, from now on
3422          * the only valid external interface is through the md
3423          * device.
3424          */
3425         rdev_for_each(rdev, tmp, mddev) {
3426                 if (test_bit(Faulty, &rdev->flags))
3427                         continue;
3428                 sync_blockdev(rdev->bdev);
3429                 invalidate_bdev(rdev->bdev);
3430
3431                 /* perform some consistency tests on the device.
3432                  * We don't want the data to overlap the metadata,
3433                  * Internal Bitmap issues has handled elsewhere.
3434                  */
3435                 if (rdev->data_offset < rdev->sb_offset) {
3436                         if (mddev->size &&
3437                             rdev->data_offset + mddev->size*2
3438                             > rdev->sb_offset*2) {
3439                                 printk("md: %s: data overlaps metadata\n",
3440                                        mdname(mddev));
3441                                 return -EINVAL;
3442                         }
3443                 } else {
3444                         if (rdev->sb_offset*2 + rdev->sb_size/512
3445                             > rdev->data_offset) {
3446                                 printk("md: %s: metadata overlaps data\n",
3447                                        mdname(mddev));
3448                                 return -EINVAL;
3449                         }
3450                 }
3451         }
3452
3453         md_probe(mddev->unit, NULL, NULL);
3454         disk = mddev->gendisk;
3455         if (!disk)
3456                 return -ENOMEM;
3457
3458         spin_lock(&pers_lock);
3459         pers = find_pers(mddev->level, mddev->clevel);
3460         if (!pers || !try_module_get(pers->owner)) {
3461                 spin_unlock(&pers_lock);
3462                 if (mddev->level != LEVEL_NONE)
3463                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3464                                mddev->level);
3465                 else
3466                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3467                                mddev->clevel);
3468                 return -EINVAL;
3469         }
3470         mddev->pers = pers;
3471         spin_unlock(&pers_lock);
3472         mddev->level = pers->level;
3473         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3474
3475         if (mddev->reshape_position != MaxSector &&
3476             pers->start_reshape == NULL) {
3477                 /* This personality cannot handle reshaping... */
3478                 mddev->pers = NULL;
3479                 module_put(pers->owner);
3480                 return -EINVAL;
3481         }
3482
3483         if (pers->sync_request) {
3484                 /* Warn if this is a potentially silly
3485                  * configuration.
3486                  */
3487                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3488                 mdk_rdev_t *rdev2;
3489                 struct list_head *tmp2;
3490                 int warned = 0;
3491                 rdev_for_each(rdev, tmp, mddev) {
3492                         rdev_for_each(rdev2, tmp2, mddev) {
3493                                 if (rdev < rdev2 &&
3494                                     rdev->bdev->bd_contains ==
3495                                     rdev2->bdev->bd_contains) {
3496                                         printk(KERN_WARNING
3497                                                "%s: WARNING: %s appears to be"
3498                                                " on the same physical disk as"
3499                                                " %s.\n",
3500                                                mdname(mddev),
3501                                                bdevname(rdev->bdev,b),
3502                                                bdevname(rdev2->bdev,b2));
3503                                         warned = 1;
3504                                 }
3505                         }
3506                 }
3507                 if (warned)
3508                         printk(KERN_WARNING
3509                                "True protection against single-disk"
3510                                " failure might be compromised.\n");
3511         }
3512
3513         mddev->recovery = 0;
3514         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3515         mddev->barriers_work = 1;
3516         mddev->ok_start_degraded = start_dirty_degraded;
3517
3518         if (start_readonly)
3519                 mddev->ro = 2; /* read-only, but switch on first write */
3520
3521         err = mddev->pers->run(mddev);
3522         if (!err && mddev->pers->sync_request) {
3523                 err = bitmap_create(mddev);
3524                 if (err) {
3525                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3526                                mdname(mddev), err);
3527                         mddev->pers->stop(mddev);
3528                 }
3529         }
3530         if (err) {
3531                 printk(KERN_ERR "md: pers->run() failed ...\n");
3532                 module_put(mddev->pers->owner);
3533                 mddev->pers = NULL;
3534                 bitmap_destroy(mddev);
3535                 return err;
3536         }
3537         if (mddev->pers->sync_request) {
3538                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3539                         printk(KERN_WARNING
3540                                "md: cannot register extra attributes for %s\n",
3541                                mdname(mddev));
3542         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3543                 mddev->ro = 0;
3544
3545         atomic_set(&mddev->writes_pending,0);
3546         mddev->safemode = 0;
3547         mddev->safemode_timer.function = md_safemode_timeout;
3548         mddev->safemode_timer.data = (unsigned long) mddev;
3549         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3550         mddev->in_sync = 1;
3551
3552         rdev_for_each(rdev, tmp, mddev)
3553                 if (rdev->raid_disk >= 0) {
3554                         char nm[20];
3555                         sprintf(nm, "rd%d", rdev->raid_disk);
3556                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3557                                 printk("md: cannot register %s for %s\n",
3558                                        nm, mdname(mddev));
3559                 }
3560
3561         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3562
3563         if (mddev->flags)
3564                 md_update_sb(mddev, 0);
3565
3566         set_capacity(disk, mddev->array_size<<1);
3567
3568         /* If we call blk_queue_make_request here, it will
3569          * re-initialise max_sectors etc which may have been
3570          * refined inside -> run.  So just set the bits we need to set.
3571          * Most initialisation happended when we called
3572          * blk_queue_make_request(..., md_fail_request)
3573          * earlier.
3574          */
3575         mddev->queue->queuedata = mddev;
3576         mddev->queue->make_request_fn = mddev->pers->make_request;
3577
3578         /* If there is a partially-recovered drive we need to
3579          * start recovery here.  If we leave it to md_check_recovery,
3580          * it will remove the drives and not do the right thing
3581          */
3582         if (mddev->degraded && !mddev->sync_thread) {
3583                 struct list_head *rtmp;
3584                 int spares = 0;
3585                 rdev_for_each(rdev, rtmp, mddev)
3586                         if (rdev->raid_disk >= 0 &&
3587                             !test_bit(In_sync, &rdev->flags) &&
3588                             !test_bit(Faulty, &rdev->flags))
3589                                 /* complete an interrupted recovery */
3590                                 spares++;
3591                 if (spares && mddev->pers->sync_request) {
3592                         mddev->recovery = 0;
3593                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3594                         mddev->sync_thread = md_register_thread(md_do_sync,
3595                                                                 mddev,
3596                                                                 "%s_resync");
3597                         if (!mddev->sync_thread) {
3598                                 printk(KERN_ERR "%s: could not start resync"
3599                                        " thread...\n",
3600                                        mdname(mddev));
3601                                 /* leave the spares where they are, it shouldn't hurt */
3602                                 mddev->recovery = 0;
3603                         }
3604                 }
3605         }
3606         md_wakeup_thread(mddev->thread);
3607         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3608
3609         mddev->changed = 1;
3610         md_new_event(mddev);
3611         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3612         return 0;
3613 }
3614
3615 static int restart_array(mddev_t *mddev)
3616 {
3617         struct gendisk *disk = mddev->gendisk;
3618         int err;
3619
3620         /*
3621          * Complain if it has no devices
3622          */
3623         err = -ENXIO;
3624         if (list_empty(&mddev->disks))
3625                 goto out;
3626
3627         if (mddev->pers) {
3628                 err = -EBUSY;
3629                 if (!mddev->ro)
3630                         goto out;
3631
3632                 mddev->safemode = 0;
3633                 mddev->ro = 0;
3634                 set_disk_ro(disk, 0);
3635
3636                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3637                         mdname(mddev));
3638                 /*
3639                  * Kick recovery or resync if necessary
3640                  */
3641                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3642                 md_wakeup_thread(mddev->thread);
3643                 md_wakeup_thread(mddev->sync_thread);
3644                 err = 0;
3645         } else
3646                 err = -EINVAL;
3647
3648 out:
3649         return err;
3650 }
3651
3652 /* similar to deny_write_access, but accounts for our holding a reference
3653  * to the file ourselves */
3654 static int deny_bitmap_write_access(struct file * file)
3655 {
3656         struct inode *inode = file->f_mapping->host;
3657
3658         spin_lock(&inode->i_lock);
3659         if (atomic_read(&inode->i_writecount) > 1) {
3660                 spin_unlock(&inode->i_lock);
3661                 return -ETXTBSY;
3662         }
3663         atomic_set(&inode->i_writecount, -1);
3664         spin_unlock(&inode->i_lock);
3665
3666         return 0;
3667 }
3668
3669 static void restore_bitmap_write_access(struct file *file)
3670 {
3671         struct inode *inode = file->f_mapping->host;
3672
3673         spin_lock(&inode->i_lock);
3674         atomic_set(&inode->i_writecount, 1);
3675         spin_unlock(&inode->i_lock);
3676 }
3677
3678 /* mode:
3679  *   0 - completely stop and dis-assemble array
3680  *   1 - switch to readonly
3681  *   2 - stop but do not disassemble array
3682  */
3683 static int do_md_stop(mddev_t * mddev, int mode)
3684 {
3685         int err = 0;
3686         struct gendisk *disk = mddev->gendisk;
3687
3688         if (mddev->pers) {
3689                 if (atomic_read(&mddev->active)>2) {
3690                         printk("md: %s still in use.\n",mdname(mddev));
3691                         return -EBUSY;
3692                 }
3693
3694                 if (mddev->sync_thread) {
3695                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3696                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3697                         md_unregister_thread(mddev->sync_thread);
3698                         mddev->sync_thread = NULL;
3699                 }
3700
3701                 del_timer_sync(&mddev->safemode_timer);
3702
3703                 invalidate_partition(disk, 0);
3704
3705                 switch(mode) {
3706                 case 1: /* readonly */
3707                         err  = -ENXIO;
3708                         if (mddev->ro==1)
3709                                 goto out;
3710                         mddev->ro = 1;
3711                         break;
3712                 case 0: /* disassemble */
3713                 case 2: /* stop */
3714                         bitmap_flush(mddev);
3715                         md_super_wait(mddev);
3716                         if (mddev->ro)
3717                                 set_disk_ro(disk, 0);
3718                         blk_queue_make_request(mddev->queue, md_fail_request);
3719                         mddev->pers->stop(mddev);
3720                         mddev->queue->merge_bvec_fn = NULL;
3721                         mddev->queue->unplug_fn = NULL;
3722                         mddev->queue->backing_dev_info.congested_fn = NULL;
3723                         if (mddev->pers->sync_request)
3724                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3725
3726                         module_put(mddev->pers->owner);
3727                         mddev->pers = NULL;
3728                         /* tell userspace to handle 'inactive' */
3729                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3730
3731                         set_capacity(disk, 0);
3732                         mddev->changed = 1;
3733
3734                         if (mddev->ro)
3735                                 mddev->ro = 0;
3736                 }
3737                 if (!mddev->in_sync || mddev->flags) {
3738                         /* mark array as shutdown cleanly */
3739                         mddev->in_sync = 1;
3740                         md_update_sb(mddev, 1);
3741                 }
3742                 if (mode == 1)
3743                         set_disk_ro(disk, 1);
3744                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3745         }
3746
3747         /*
3748          * Free resources if final stop
3749          */
3750         if (mode == 0) {
3751                 mdk_rdev_t *rdev;
3752                 struct list_head *tmp;
3753
3754                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3755
3756                 bitmap_destroy(mddev);
3757                 if (mddev->bitmap_file) {
3758                         restore_bitmap_write_access(mddev->bitmap_file);
3759                         fput(mddev->bitmap_file);
3760                         mddev->bitmap_file = NULL;
3761                 }
3762                 mddev->bitmap_offset = 0;
3763
3764                 rdev_for_each(rdev, tmp, mddev)
3765                         if (rdev->raid_disk >= 0) {
3766                                 char nm[20];
3767                                 sprintf(nm, "rd%d", rdev->raid_disk);
3768                                 sysfs_remove_link(&mddev->kobj, nm);
3769                         }
3770
3771                 /* make sure all md_delayed_delete calls have finished */
3772                 flush_scheduled_work();
3773
3774                 export_array(mddev);
3775
3776                 mddev->array_size = 0;
3777                 mddev->size = 0;
3778                 mddev->raid_disks = 0;
3779                 mddev->recovery_cp = 0;
3780                 mddev->resync_max = MaxSector;
3781                 mddev->reshape_position = MaxSector;
3782                 mddev->external = 0;
3783                 mddev->persistent = 0;
3784                 mddev->level = LEVEL_NONE;
3785                 mddev->clevel[0] = 0;
3786                 mddev->flags = 0;
3787                 mddev->ro = 0;
3788                 mddev->metadata_type[0] = 0;
3789                 mddev->chunk_size = 0;
3790                 mddev->ctime = mddev->utime = 0;
3791                 mddev->layout = 0;
3792                 mddev->max_disks = 0;
3793                 mddev->events = 0;
3794                 mddev->delta_disks = 0;
3795                 mddev->new_level = LEVEL_NONE;
3796                 mddev->new_layout = 0;
3797                 mddev->new_chunk = 0;
3798                 mddev->curr_resync = 0;
3799                 mddev->resync_mismatches = 0;
3800                 mddev->suspend_lo = mddev->suspend_hi = 0;
3801                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3802                 mddev->recovery = 0;
3803                 mddev->in_sync = 0;
3804                 mddev->changed = 0;
3805                 mddev->degraded = 0;
3806                 mddev->barriers_work = 0;
3807                 mddev->safemode = 0;
3808
3809         } else if (mddev->pers)
3810                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3811                         mdname(mddev));
3812         err = 0;
3813         md_new_event(mddev);
3814 out:
3815         return err;
3816 }
3817
3818 #ifndef MODULE
3819 static void autorun_array(mddev_t *mddev)
3820 {
3821         mdk_rdev_t *rdev;
3822         struct list_head *tmp;
3823         int err;
3824
3825         if (list_empty(&mddev->disks))
3826                 return;
3827
3828         printk(KERN_INFO "md: running: ");
3829
3830         rdev_for_each(rdev, tmp, mddev) {
3831                 char b[BDEVNAME_SIZE];
3832                 printk("<%s>", bdevname(rdev->bdev,b));
3833         }
3834         printk("\n");
3835
3836         err = do_md_run (mddev);
3837         if (err) {
3838                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3839                 do_md_stop (mddev, 0);
3840         }
3841 }
3842
3843 /*
3844  * lets try to run arrays based on all disks that have arrived
3845  * until now. (those are in pending_raid_disks)
3846  *
3847  * the method: pick the first pending disk, collect all disks with
3848  * the same UUID, remove all from the pending list and put them into
3849  * the 'same_array' list. Then order this list based on superblock
3850  * update time (freshest comes first), kick out 'old' disks and
3851  * compare superblocks. If everything's fine then run it.
3852  *
3853  * If "unit" is allocated, then bump its reference count
3854  */
3855 static void autorun_devices(int part)
3856 {
3857         struct list_head *tmp;
3858         mdk_rdev_t *rdev0, *rdev;
3859         mddev_t *mddev;
3860         char b[BDEVNAME_SIZE];
3861
3862         printk(KERN_INFO "md: autorun ...\n");
3863         while (!list_empty(&pending_raid_disks)) {
3864                 int unit;
3865                 dev_t dev;
3866                 LIST_HEAD(candidates);
3867                 rdev0 = list_entry(pending_raid_disks.next,
3868                                          mdk_rdev_t, same_set);
3869
3870                 printk(KERN_INFO "md: considering %s ...\n",
3871                         bdevname(rdev0->bdev,b));
3872                 INIT_LIST_HEAD(&candidates);
3873                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3874                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3875                                 printk(KERN_INFO "md:  adding %s ...\n",
3876                                         bdevname(rdev->bdev,b));
3877                                 list_move(&rdev->same_set, &candidates);
3878                         }
3879                 /*
3880                  * now we have a set of devices, with all of them having
3881                  * mostly sane superblocks. It's time to allocate the
3882                  * mddev.
3883                  */
3884                 if (part) {
3885                         dev = MKDEV(mdp_major,
3886                                     rdev0->preferred_minor << MdpMinorShift);
3887                         unit = MINOR(dev) >> MdpMinorShift;
3888                 } else {
3889                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3890                         unit = MINOR(dev);
3891                 }
3892                 if (rdev0->preferred_minor != unit) {
3893                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3894                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3895                         break;
3896                 }
3897
3898                 md_probe(dev, NULL, NULL);
3899                 mddev = mddev_find(dev);
3900                 if (!mddev || !mddev->gendisk) {
3901                         if (mddev)
3902                                 mddev_put(mddev);
3903                         printk(KERN_ERR
3904                                 "md: cannot allocate memory for md drive.\n");
3905                         break;
3906                 }
3907                 if (mddev_lock(mddev))
3908                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3909                                mdname(mddev));
3910                 else if (mddev->raid_disks || mddev->major_version
3911                          || !list_empty(&mddev->disks)) {
3912                         printk(KERN_WARNING
3913                                 "md: %s already running, cannot run %s\n",
3914                                 mdname(mddev), bdevname(rdev0->bdev,b));
3915                         mddev_unlock(mddev);
3916                 } else {
3917                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3918                         mddev->persistent = 1;
3919                         rdev_for_each_list(rdev, tmp, candidates) {
3920                                 list_del_init(&rdev->same_set);
3921                                 if (bind_rdev_to_array(rdev, mddev))
3922                                         export_rdev(rdev);
3923                         }
3924                         autorun_array(mddev);
3925                         mddev_unlock(mddev);
3926                 }
3927                 /* on success, candidates will be empty, on error
3928                  * it won't...
3929                  */
3930                 rdev_for_each_list(rdev, tmp, candidates)
3931                         export_rdev(rdev);
3932                 mddev_put(mddev);
3933         }
3934         printk(KERN_INFO "md: ... autorun DONE.\n");
3935 }
3936 #endif /* !MODULE */
3937
3938 static int get_version(void __user * arg)
3939 {
3940         mdu_version_t ver;
3941
3942         ver.major = MD_MAJOR_VERSION;
3943         ver.minor = MD_MINOR_VERSION;
3944         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3945
3946         if (copy_to_user(arg, &ver, sizeof(ver)))
3947                 return -EFAULT;
3948
3949         return 0;
3950 }
3951
3952 static int get_array_info(mddev_t * mddev, void __user * arg)
3953 {
3954         mdu_array_info_t info;
3955         int nr,working,active,failed,spare;
3956         mdk_rdev_t *rdev;
3957         struct list_head *tmp;
3958
3959         nr=working=active=failed=spare=0;
3960         rdev_for_each(rdev, tmp, mddev) {
3961                 nr++;
3962                 if (test_bit(Faulty, &rdev->flags))
3963                         failed++;
3964                 else {
3965                         working++;
3966                         if (test_bit(In_sync, &rdev->flags))
3967                                 active++;
3968                         else
3969                                 spare++;
3970                 }
3971         }
3972
3973         info.major_version = mddev->major_version;
3974         info.minor_version = mddev->minor_version;
3975         info.patch_version = MD_PATCHLEVEL_VERSION;
3976         info.ctime         = mddev->ctime;
3977         info.level         = mddev->level;
3978         info.size          = mddev->size;
3979         if (info.size != mddev->size) /* overflow */
3980                 info.size = -1;
3981         info.nr_disks      = nr;
3982         info.raid_disks    = mddev->raid_disks;
3983         info.md_minor      = mddev->md_minor;
3984         info.not_persistent= !mddev->persistent;
3985
3986         info.utime         = mddev->utime;
3987         info.state         = 0;
3988         if (mddev->in_sync)
3989                 info.state = (1<<MD_SB_CLEAN);
3990         if (mddev->bitmap && mddev->bitmap_offset)
3991                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3992         info.active_disks  = active;
3993         info.working_disks = working;
3994         info.failed_disks  = failed;
3995         info.spare_disks   = spare;
3996
3997         info.layout        = mddev->layout;
3998         info.chunk_size    = mddev->chunk_size;
3999
4000         if (copy_to_user(arg, &info, sizeof(info)))
4001                 return -EFAULT;
4002
4003         return 0;
4004 }
4005
4006 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4007 {
4008         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4009         char *ptr, *buf = NULL;
4010         int err = -ENOMEM;
4011
4012         md_allow_write(mddev);
4013
4014         file = kmalloc(sizeof(*file), GFP_KERNEL);
4015         if (!file)
4016                 goto out;
4017
4018         /* bitmap disabled, zero the first byte and copy out */
4019         if (!mddev->bitmap || !mddev->bitmap->file) {
4020                 file->pathname[0] = '\0';
4021                 goto copy_out;
4022         }
4023
4024         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4025         if (!buf)
4026                 goto out;
4027
4028         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4029         if (IS_ERR(ptr))
4030                 goto out;
4031
4032         strcpy(file->pathname, ptr);
4033
4034 copy_out:
4035         err = 0;
4036         if (copy_to_user(arg, file, sizeof(*file)))
4037                 err = -EFAULT;
4038 out:
4039         kfree(buf);
4040         kfree(file);
4041         return err;
4042 }
4043
4044 static int get_disk_info(mddev_t * mddev, void __user * arg)
4045 {
4046         mdu_disk_info_t info;
4047         unsigned int nr;
4048         mdk_rdev_t *rdev;
4049
4050         if (copy_from_user(&info, arg, sizeof(info)))
4051                 return -EFAULT;
4052
4053         nr = info.number;
4054
4055         rdev = find_rdev_nr(mddev, nr);
4056         if (rdev) {
4057                 info.major = MAJOR(rdev->bdev->bd_dev);
4058                 info.minor = MINOR(rdev->bdev->bd_dev);
4059                 info.raid_disk = rdev->raid_disk;
4060                 info.state = 0;
4061                 if (test_bit(Faulty, &rdev->flags))
4062                         info.state |= (1<<MD_DISK_FAULTY);
4063                 else if (test_bit(In_sync, &rdev->flags)) {
4064                         info.state |= (1<<MD_DISK_ACTIVE);
4065                         info.state |= (1<<MD_DISK_SYNC);
4066                 }
4067                 if (test_bit(WriteMostly, &rdev->flags))
4068                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4069         } else {
4070                 info.major = info.minor = 0;
4071                 info.raid_disk = -1;
4072                 info.state = (1<<MD_DISK_REMOVED);
4073         }
4074
4075         if (copy_to_user(arg, &info, sizeof(info)))
4076                 return -EFAULT;
4077
4078         return 0;
4079 }
4080
4081 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4082 {
4083         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4084         mdk_rdev_t *rdev;
4085         dev_t dev = MKDEV(info->major,info->minor);
4086
4087         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4088                 return -EOVERFLOW;
4089
4090         if (!mddev->raid_disks) {
4091                 int err;
4092                 /* expecting a device which has a superblock */
4093                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4094                 if (IS_ERR(rdev)) {
4095                         printk(KERN_WARNING
4096                                 "md: md_import_device returned %ld\n",
4097                                 PTR_ERR(rdev));
4098                         return PTR_ERR(rdev);
4099                 }
4100                 if (!list_empty(&mddev->disks)) {
4101                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4102                                                         mdk_rdev_t, same_set);
4103                         int err = super_types[mddev->major_version]
4104                                 .load_super(rdev, rdev0, mddev->minor_version);
4105                         if (err < 0) {
4106                                 printk(KERN_WARNING
4107                                         "md: %s has different UUID to %s\n",
4108                                         bdevname(rdev->bdev,b),
4109                                         bdevname(rdev0->bdev,b2));
4110                                 export_rdev(rdev);
4111                                 return -EINVAL;
4112                         }
4113                 }
4114                 err = bind_rdev_to_array(rdev, mddev);
4115                 if (err)
4116                         export_rdev(rdev);
4117                 return err;
4118         }
4119
4120         /*
4121          * add_new_disk can be used once the array is assembled
4122          * to add "hot spares".  They must already have a superblock
4123          * written
4124          */
4125         if (mddev->pers) {
4126                 int err;
4127                 if (!mddev->pers->hot_add_disk) {
4128                         printk(KERN_WARNING
4129                                 "%s: personality does not support diskops!\n",
4130                                mdname(mddev));
4131                         return -EINVAL;
4132                 }
4133                 if (mddev->persistent)
4134                         rdev = md_import_device(dev, mddev->major_version,
4135                                                 mddev->minor_version);
4136                 else
4137                         rdev = md_import_device(dev, -1, -1);
4138                 if (IS_ERR(rdev)) {
4139                         printk(KERN_WARNING
4140                                 "md: md_import_device returned %ld\n",
4141                                 PTR_ERR(rdev));
4142                         return PTR_ERR(rdev);
4143                 }
4144                 /* set save_raid_disk if appropriate */
4145                 if (!mddev->persistent) {
4146                         if (info->state & (1<<MD_DISK_SYNC)  &&
4147                             info->raid_disk < mddev->raid_disks)
4148                                 rdev->raid_disk = info->raid_disk;
4149                         else
4150                                 rdev->raid_disk = -1;
4151                 } else
4152                         super_types[mddev->major_version].
4153                                 validate_super(mddev, rdev);
4154                 rdev->saved_raid_disk = rdev->raid_disk;
4155
4156                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4157                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4158                         set_bit(WriteMostly, &rdev->flags);
4159
4160                 rdev->raid_disk = -1;
4161                 err = bind_rdev_to_array(rdev, mddev);
4162                 if (!err && !mddev->pers->hot_remove_disk) {
4163                         /* If there is hot_add_disk but no hot_remove_disk
4164                          * then added disks for geometry changes,
4165                          * and should be added immediately.
4166                          */
4167                         super_types[mddev->major_version].
4168                                 validate_super(mddev, rdev);
4169                         err = mddev->pers->hot_add_disk(mddev, rdev);
4170                         if (err)
4171                                 unbind_rdev_from_array(rdev);
4172                 }
4173                 if (err)
4174                         export_rdev(rdev);
4175
4176                 md_update_sb(mddev, 1);
4177                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4178                 md_wakeup_thread(mddev->thread);
4179                 return err;
4180         }
4181
4182         /* otherwise, add_new_disk is only allowed
4183          * for major_version==0 superblocks
4184          */
4185         if (mddev->major_version != 0) {
4186                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4187                        mdname(mddev));
4188                 return -EINVAL;
4189         }
4190
4191         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4192                 int err;
4193                 rdev = md_import_device (dev, -1, 0);
4194                 if (IS_ERR(rdev)) {
4195                         printk(KERN_WARNING
4196                                 "md: error, md_import_device() returned %ld\n",
4197                                 PTR_ERR(rdev));
4198                         return PTR_ERR(rdev);
4199                 }
4200                 rdev->desc_nr = info->number;
4201                 if (info->raid_disk < mddev->raid_disks)
4202                         rdev->raid_disk = info->raid_disk;
4203                 else
4204                         rdev->raid_disk = -1;
4205
4206                 if (rdev->raid_disk < mddev->raid_disks)
4207                         if (info->state & (1<<MD_DISK_SYNC))
4208                                 set_bit(In_sync, &rdev->flags);
4209
4210                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4211                         set_bit(WriteMostly, &rdev->flags);
4212
4213                 if (!mddev->persistent) {
4214                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4215                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4216                 } else
4217                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4218                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4219
4220                 err = bind_rdev_to_array(rdev, mddev);
4221                 if (err) {
4222                         export_rdev(rdev);
4223                         return err;
4224                 }
4225         }
4226
4227         return 0;
4228 }
4229
4230 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4231 {
4232         char b[BDEVNAME_SIZE];
4233         mdk_rdev_t *rdev;
4234
4235         if (!mddev->pers)
4236                 return -ENODEV;
4237
4238         rdev = find_rdev(mddev, dev);
4239         if (!rdev)
4240                 return -ENXIO;
4241
4242         if (rdev->raid_disk >= 0)
4243                 goto busy;
4244
4245         kick_rdev_from_array(rdev);
4246         md_update_sb(mddev, 1);
4247         md_new_event(mddev);
4248
4249         return 0;
4250 busy:
4251         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4252                 bdevname(rdev->bdev,b), mdname(mddev));
4253         return -EBUSY;
4254 }
4255
4256 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4257 {
4258         char b[BDEVNAME_SIZE];
4259         int err;
4260         unsigned int size;
4261         mdk_rdev_t *rdev;
4262
4263         if (!mddev->pers)
4264                 return -ENODEV;
4265
4266         if (mddev->major_version != 0) {
4267                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4268                         " version-0 superblocks.\n",
4269                         mdname(mddev));
4270                 return -EINVAL;
4271         }
4272         if (!mddev->pers->hot_add_disk) {
4273                 printk(KERN_WARNING
4274                         "%s: personality does not support diskops!\n",
4275                         mdname(mddev));
4276                 return -EINVAL;
4277         }
4278
4279         rdev = md_import_device (dev, -1, 0);
4280         if (IS_ERR(rdev)) {
4281                 printk(KERN_WARNING
4282                         "md: error, md_import_device() returned %ld\n",
4283                         PTR_ERR(rdev));
4284                 return -EINVAL;
4285         }
4286
4287         if (mddev->persistent)
4288                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4289         else
4290                 rdev->sb_offset =
4291                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4292
4293         size = calc_dev_size(rdev, mddev->chunk_size);
4294         rdev->size = size;
4295
4296         if (test_bit(Faulty, &rdev->flags)) {
4297                 printk(KERN_WARNING
4298                         "md: can not hot-add faulty %s disk to %s!\n",
4299                         bdevname(rdev->bdev,b), mdname(mddev));
4300                 err = -EINVAL;
4301                 goto abort_export;
4302         }
4303         clear_bit(In_sync, &rdev->flags);
4304         rdev->desc_nr = -1;
4305         rdev->saved_raid_disk = -1;
4306         err = bind_rdev_to_array(rdev, mddev);
4307         if (err)
4308                 goto abort_export;
4309
4310         /*
4311          * The rest should better be atomic, we can have disk failures
4312          * noticed in interrupt contexts ...
4313          */
4314
4315         if (rdev->desc_nr == mddev->max_disks) {
4316                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4317                         mdname(mddev));
4318                 err = -EBUSY;
4319                 goto abort_unbind_export;
4320         }
4321
4322         rdev->raid_disk = -1;
4323
4324         md_update_sb(mddev, 1);
4325
4326         /*
4327          * Kick recovery, maybe this spare has to be added to the
4328          * array immediately.
4329          */
4330         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4331         md_wakeup_thread(mddev->thread);
4332         md_new_event(mddev);
4333         return 0;
4334
4335 abort_unbind_export:
4336         unbind_rdev_from_array(rdev);
4337
4338 abort_export:
4339         export_rdev(rdev);
4340         return err;
4341 }
4342
4343 static int set_bitmap_file(mddev_t *mddev, int fd)
4344 {
4345         int err;
4346
4347         if (mddev->pers) {
4348                 if (!mddev->pers->quiesce)
4349                         return -EBUSY;
4350                 if (mddev->recovery || mddev->sync_thread)
4351                         return -EBUSY;
4352                 /* we should be able to change the bitmap.. */
4353         }
4354
4355
4356         if (fd >= 0) {
4357                 if (mddev->bitmap)
4358                         return -EEXIST; /* cannot add when bitmap is present */
4359                 mddev->bitmap_file = fget(fd);
4360
4361                 if (mddev->bitmap_file == NULL) {
4362                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4363                                mdname(mddev));
4364                         return -EBADF;
4365                 }
4366
4367                 err = deny_bitmap_write_access(mddev->bitmap_file);
4368                 if (err) {
4369                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4370                                mdname(mddev));
4371                         fput(mddev->bitmap_file);
4372                         mddev->bitmap_file = NULL;
4373                         return err;
4374                 }
4375                 mddev->bitmap_offset = 0; /* file overrides offset */
4376         } else if (mddev->bitmap == NULL)
4377                 return -ENOENT; /* cannot remove what isn't there */
4378         err = 0;
4379         if (mddev->pers) {
4380                 mddev->pers->quiesce(mddev, 1);
4381                 if (fd >= 0)
4382                         err = bitmap_create(mddev);
4383                 if (fd < 0 || err) {
4384                         bitmap_destroy(mddev);
4385                         fd = -1; /* make sure to put the file */
4386                 }
4387                 mddev->pers->quiesce(mddev, 0);
4388         }
4389         if (fd < 0) {
4390                 if (mddev->bitmap_file) {
4391                         restore_bitmap_write_access(mddev->bitmap_file);
4392                         fput(mddev->bitmap_file);
4393                 }
4394                 mddev->bitmap_file = NULL;
4395         }
4396
4397         return err;
4398 }
4399
4400 /*
4401  * set_array_info is used two different ways
4402  * The original usage is when creating a new array.
4403  * In this usage, raid_disks is > 0 and it together with
4404  *  level, size, not_persistent,layout,chunksize determine the
4405  *  shape of the array.
4406  *  This will always create an array with a type-0.90.0 superblock.
4407  * The newer usage is when assembling an array.
4408  *  In this case raid_disks will be 0, and the major_version field is
4409  *  use to determine which style super-blocks are to be found on the devices.
4410  *  The minor and patch _version numbers are also kept incase the
4411  *  super_block handler wishes to interpret them.
4412  */
4413 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4414 {
4415
4416         if (info->raid_disks == 0) {
4417                 /* just setting version number for superblock loading */
4418                 if (info->major_version < 0 ||
4419                     info->major_version >= ARRAY_SIZE(super_types) ||
4420                     super_types[info->major_version].name == NULL) {
4421                         /* maybe try to auto-load a module? */
4422                         printk(KERN_INFO
4423                                 "md: superblock version %d not known\n",
4424                                 info->major_version);
4425                         return -EINVAL;
4426                 }
4427                 mddev->major_version = info->major_version;
4428                 mddev->minor_version = info->minor_version;
4429                 mddev->patch_version = info->patch_version;
4430                 mddev->persistent = !info->not_persistent;
4431                 return 0;
4432         }
4433         mddev->major_version = MD_MAJOR_VERSION;
4434         mddev->minor_version = MD_MINOR_VERSION;
4435         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4436         mddev->ctime         = get_seconds();
4437
4438         mddev->level         = info->level;
4439         mddev->clevel[0]     = 0;
4440         mddev->size          = info->size;
4441         mddev->raid_disks    = info->raid_disks;
4442         /* don't set md_minor, it is determined by which /dev/md* was
4443          * openned
4444          */
4445         if (info->state & (1<<MD_SB_CLEAN))
4446                 mddev->recovery_cp = MaxSector;
4447         else
4448                 mddev->recovery_cp = 0;
4449         mddev->persistent    = ! info->not_persistent;
4450         mddev->external      = 0;
4451
4452         mddev->layout        = info->layout;
4453         mddev->chunk_size    = info->chunk_size;
4454
4455         mddev->max_disks     = MD_SB_DISKS;
4456
4457         if (mddev->persistent)
4458                 mddev->flags         = 0;
4459         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4460
4461         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4462         mddev->bitmap_offset = 0;
4463
4464         mddev->reshape_position = MaxSector;
4465
4466         /*
4467          * Generate a 128 bit UUID
4468          */
4469         get_random_bytes(mddev->uuid, 16);
4470
4471         mddev->new_level = mddev->level;
4472         mddev->new_chunk = mddev->chunk_size;
4473         mddev->new_layout = mddev->layout;
4474         mddev->delta_disks = 0;
4475
4476         return 0;
4477 }
4478
4479 static int update_size(mddev_t *mddev, unsigned long size)
4480 {
4481         mdk_rdev_t * rdev;
4482         int rv;
4483         struct list_head *tmp;
4484         int fit = (size == 0);
4485
4486         if (mddev->pers->resize == NULL)
4487                 return -EINVAL;
4488         /* The "size" is the amount of each device that is used.
4489          * This can only make sense for arrays with redundancy.
4490          * linear and raid0 always use whatever space is available
4491          * We can only consider changing the size if no resync
4492          * or reconstruction is happening, and if the new size
4493          * is acceptable. It must fit before the sb_offset or,
4494          * if that is <data_offset, it must fit before the
4495          * size of each device.
4496          * If size is zero, we find the largest size that fits.
4497          */
4498         if (mddev->sync_thread)
4499                 return -EBUSY;
4500         rdev_for_each(rdev, tmp, mddev) {
4501                 sector_t avail;
4502                 avail = rdev->size * 2;
4503
4504                 if (fit && (size == 0 || size > avail/2))
4505                         size = avail/2;
4506                 if (avail < ((sector_t)size << 1))
4507                         return -ENOSPC;
4508         }
4509         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4510         if (!rv) {
4511                 struct block_device *bdev;
4512
4513                 bdev = bdget_disk(mddev->gendisk, 0);
4514                 if (bdev) {
4515                         mutex_lock(&bdev->bd_inode->i_mutex);
4516                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4517                         mutex_unlock(&bdev->bd_inode->i_mutex);
4518                         bdput(bdev);
4519                 }
4520         }
4521         return rv;
4522 }
4523
4524 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4525 {
4526         int rv;
4527         /* change the number of raid disks */
4528         if (mddev->pers->check_reshape == NULL)
4529                 return -EINVAL;
4530         if (raid_disks <= 0 ||
4531             raid_disks >= mddev->max_disks)
4532                 return -EINVAL;
4533         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4534                 return -EBUSY;
4535         mddev->delta_disks = raid_disks - mddev->raid_disks;
4536
4537         rv = mddev->pers->check_reshape(mddev);
4538         return rv;
4539 }
4540
4541
4542 /*
4543  * update_array_info is used to change the configuration of an
4544  * on-line array.
4545  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4546  * fields in the info are checked against the array.
4547  * Any differences that cannot be handled will cause an error.
4548  * Normally, only one change can be managed at a time.
4549  */
4550 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4551 {
4552         int rv = 0;
4553         int cnt = 0;
4554         int state = 0;
4555
4556         /* calculate expected state,ignoring low bits */
4557         if (mddev->bitmap && mddev->bitmap_offset)
4558                 state |= (1 << MD_SB_BITMAP_PRESENT);
4559
4560         if (mddev->major_version != info->major_version ||
4561             mddev->minor_version != info->minor_version ||
4562 /*          mddev->patch_version != info->patch_version || */
4563             mddev->ctime         != info->ctime         ||
4564             mddev->level         != info->level         ||
4565 /*          mddev->layout        != info->layout        || */
4566             !mddev->persistent   != info->not_persistent||
4567             mddev->chunk_size    != info->chunk_size    ||
4568             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4569             ((state^info->state) & 0xfffffe00)
4570                 )
4571                 return -EINVAL;
4572         /* Check there is only one change */
4573         if (info->size >= 0 && mddev->size != info->size) cnt++;
4574         if (mddev->raid_disks != info->raid_disks) cnt++;
4575         if (mddev->layout != info->layout) cnt++;
4576         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4577         if (cnt == 0) return 0;
4578         if (cnt > 1) return -EINVAL;
4579
4580         if (mddev->layout != info->layout) {
4581                 /* Change layout
4582                  * we don't need to do anything at the md level, the
4583                  * personality will take care of it all.
4584                  */
4585                 if (mddev->pers->reconfig == NULL)
4586                         return -EINVAL;
4587                 else
4588                         return mddev->pers->reconfig(mddev, info->layout, -1);
4589         }
4590         if (info->size >= 0 && mddev->size != info->size)
4591                 rv = update_size(mddev, info->size);
4592
4593         if (mddev->raid_disks    != info->raid_disks)
4594                 rv = update_raid_disks(mddev, info->raid_disks);
4595
4596         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4597                 if (mddev->pers->quiesce == NULL)
4598                         return -EINVAL;
4599                 if (mddev->recovery || mddev->sync_thread)
4600                         return -EBUSY;
4601                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4602                         /* add the bitmap */
4603                         if (mddev->bitmap)
4604                                 return -EEXIST;
4605                         if (mddev->default_bitmap_offset == 0)
4606                                 return -EINVAL;
4607                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4608                         mddev->pers->quiesce(mddev, 1);
4609                         rv = bitmap_create(mddev);
4610                         if (rv)
4611                                 bitmap_destroy(mddev);
4612                         mddev->pers->quiesce(mddev, 0);
4613                 } else {
4614                         /* remove the bitmap */
4615                         if (!mddev->bitmap)
4616                                 return -ENOENT;
4617                         if (mddev->bitmap->file)
4618                                 return -EINVAL;
4619                         mddev->pers->quiesce(mddev, 1);
4620                         bitmap_destroy(mddev);
4621                         mddev->pers->quiesce(mddev, 0);
4622                         mddev->bitmap_offset = 0;
4623                 }
4624         }
4625         md_update_sb(mddev, 1);
4626         return rv;
4627 }
4628
4629 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4630 {
4631         mdk_rdev_t *rdev;
4632
4633         if (mddev->pers == NULL)
4634                 return -ENODEV;
4635
4636         rdev = find_rdev(mddev, dev);
4637         if (!rdev)
4638                 return -ENODEV;
4639
4640         md_error(mddev, rdev);
4641         return 0;
4642 }
4643
4644 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4645 {
4646         mddev_t *mddev = bdev->bd_disk->private_data;
4647
4648         geo->heads = 2;
4649         geo->sectors = 4;
4650         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4651         return 0;
4652 }
4653
4654 static int md_ioctl(struct inode *inode, struct file *file,
4655                         unsigned int cmd, unsigned long arg)
4656 {
4657         int err = 0;
4658         void __user *argp = (void __user *)arg;
4659         mddev_t *mddev = NULL;
4660
4661         if (!capable(CAP_SYS_ADMIN))
4662                 return -EACCES;
4663
4664         /*
4665          * Commands dealing with the RAID driver but not any
4666          * particular array:
4667          */
4668         switch (cmd)
4669         {
4670                 case RAID_VERSION:
4671                         err = get_version(argp);
4672                         goto done;
4673
4674                 case PRINT_RAID_DEBUG:
4675                         err = 0;
4676                         md_print_devices();
4677                         goto done;
4678
4679 #ifndef MODULE
4680                 case RAID_AUTORUN:
4681                         err = 0;
4682                         autostart_arrays(arg);
4683                         goto done;
4684 #endif
4685                 default:;
4686         }
4687
4688         /*
4689          * Commands creating/starting a new array:
4690          */
4691
4692         mddev = inode->i_bdev->bd_disk->private_data;
4693
4694         if (!mddev) {
4695                 BUG();
4696                 goto abort;
4697         }
4698
4699         err = mddev_lock(mddev);
4700         if (err) {
4701                 printk(KERN_INFO
4702                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4703                         err, cmd);
4704                 goto abort;
4705         }
4706
4707         switch (cmd)
4708         {
4709                 case SET_ARRAY_INFO:
4710                         {
4711                                 mdu_array_info_t info;
4712                                 if (!arg)
4713                                         memset(&info, 0, sizeof(info));
4714                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4715                                         err = -EFAULT;
4716                                         goto abort_unlock;
4717                                 }
4718                                 if (mddev->pers) {
4719                                         err = update_array_info(mddev, &info);
4720                                         if (err) {
4721                                                 printk(KERN_WARNING "md: couldn't update"
4722                                                        " array info. %d\n", err);
4723                                                 goto abort_unlock;
4724                                         }
4725                                         goto done_unlock;
4726                                 }
4727                                 if (!list_empty(&mddev->disks)) {
4728                                         printk(KERN_WARNING
4729                                                "md: array %s already has disks!\n",
4730                                                mdname(mddev));
4731                                         err = -EBUSY;
4732                                         goto abort_unlock;
4733                                 }
4734                                 if (mddev->raid_disks) {
4735                                         printk(KERN_WARNING
4736                                                "md: array %s already initialised!\n",
4737                                                mdname(mddev));
4738                                         err = -EBUSY;
4739                                         goto abort_unlock;
4740                                 }
4741                                 err = set_array_info(mddev, &info);
4742                                 if (err) {
4743                                         printk(KERN_WARNING "md: couldn't set"
4744                                                " array info. %d\n", err);
4745                                         goto abort_unlock;
4746                                 }
4747                         }
4748                         goto done_unlock;
4749
4750                 default:;
4751         }
4752
4753         /*
4754          * Commands querying/configuring an existing array:
4755          */
4756         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4757          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4758         if ((!mddev->raid_disks && !mddev->external)
4759             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4760             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4761             && cmd != GET_BITMAP_FILE) {
4762                 err = -ENODEV;
4763                 goto abort_unlock;
4764         }
4765
4766         /*
4767          * Commands even a read-only array can execute:
4768          */
4769         switch (cmd)
4770         {
4771                 case GET_ARRAY_INFO:
4772                         err = get_array_info(mddev, argp);
4773                         goto done_unlock;
4774
4775                 case GET_BITMAP_FILE:
4776                         err = get_bitmap_file(mddev, argp);
4777                         goto done_unlock;
4778
4779                 case GET_DISK_INFO:
4780                         err = get_disk_info(mddev, argp);
4781                         goto done_unlock;
4782
4783                 case RESTART_ARRAY_RW:
4784                         err = restart_array(mddev);
4785                         goto done_unlock;
4786
4787                 case STOP_ARRAY:
4788                         err = do_md_stop (mddev, 0);
4789                         goto done_unlock;
4790
4791                 case STOP_ARRAY_RO:
4792                         err = do_md_stop (mddev, 1);
4793                         goto done_unlock;
4794
4795         /*
4796          * We have a problem here : there is no easy way to give a CHS
4797          * virtual geometry. We currently pretend that we have a 2 heads
4798          * 4 sectors (with a BIG number of cylinders...). This drives
4799          * dosfs just mad... ;-)
4800          */
4801         }
4802
4803         /*
4804          * The remaining ioctls are changing the state of the
4805          * superblock, so we do not allow them on read-only arrays.
4806          * However non-MD ioctls (e.g. get-size) will still come through
4807          * here and hit the 'default' below, so only disallow
4808          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4809          */
4810         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4811             mddev->ro && mddev->pers) {
4812                 if (mddev->ro == 2) {
4813                         mddev->ro = 0;
4814                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4815                 md_wakeup_thread(mddev->thread);
4816
4817                 } else {
4818                         err = -EROFS;
4819                         goto abort_unlock;
4820                 }
4821         }
4822
4823         switch (cmd)
4824         {
4825                 case ADD_NEW_DISK:
4826                 {
4827                         mdu_disk_info_t info;
4828                         if (copy_from_user(&info, argp, sizeof(info)))
4829                                 err = -EFAULT;
4830                         else
4831                                 err = add_new_disk(mddev, &info);
4832                         goto done_unlock;
4833                 }
4834
4835                 case HOT_REMOVE_DISK:
4836                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4837                         goto done_unlock;
4838
4839                 case HOT_ADD_DISK:
4840                         err = hot_add_disk(mddev, new_decode_dev(arg));
4841                         goto done_unlock;
4842
4843                 case SET_DISK_FAULTY:
4844                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4845                         goto done_unlock;
4846
4847                 case RUN_ARRAY:
4848                         err = do_md_run (mddev);
4849                         goto done_unlock;
4850
4851                 case SET_BITMAP_FILE:
4852                         err = set_bitmap_file(mddev, (int)arg);
4853                         goto done_unlock;
4854
4855                 default:
4856                         err = -EINVAL;
4857                         goto abort_unlock;
4858         }
4859
4860 done_unlock:
4861 abort_unlock:
4862         mddev_unlock(mddev);
4863
4864         return err;
4865 done:
4866         if (err)
4867                 MD_BUG();
4868 abort:
4869         return err;
4870 }
4871
4872 static int md_open(struct inode *inode, struct file *file)
4873 {
4874         /*
4875          * Succeed if we can lock the mddev, which confirms that
4876          * it isn't being stopped right now.
4877          */
4878         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4879         int err;
4880
4881         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4882                 goto out;
4883
4884         err = 0;
4885         mddev_get(mddev);
4886         mddev_unlock(mddev);
4887
4888         check_disk_change(inode->i_bdev);
4889  out:
4890         return err;
4891 }
4892
4893 static int md_release(struct inode *inode, struct file * file)
4894 {
4895         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4896
4897         BUG_ON(!mddev);
4898         mddev_put(mddev);
4899
4900         return 0;
4901 }
4902
4903 static int md_media_changed(struct gendisk *disk)
4904 {
4905         mddev_t *mddev = disk->private_data;
4906
4907         return mddev->changed;
4908 }
4909
4910 static int md_revalidate(struct gendisk *disk)
4911 {
4912         mddev_t *mddev = disk->private_data;
4913
4914         mddev->changed = 0;
4915         return 0;
4916 }
4917 static struct block_device_operations md_fops =
4918 {
4919         .owner          = THIS_MODULE,
4920         .open           = md_open,
4921         .release        = md_release,
4922         .ioctl          = md_ioctl,
4923         .getgeo         = md_getgeo,
4924         .media_changed  = md_media_changed,
4925         .revalidate_disk= md_revalidate,
4926 };
4927
4928 static int md_thread(void * arg)
4929 {
4930         mdk_thread_t *thread = arg;
4931
4932         /*
4933          * md_thread is a 'system-thread', it's priority should be very
4934          * high. We avoid resource deadlocks individually in each
4935          * raid personality. (RAID5 does preallocation) We also use RR and
4936          * the very same RT priority as kswapd, thus we will never get
4937          * into a priority inversion deadlock.
4938          *
4939          * we definitely have to have equal or higher priority than
4940          * bdflush, otherwise bdflush will deadlock if there are too
4941          * many dirty RAID5 blocks.
4942          */
4943
4944         allow_signal(SIGKILL);
4945         while (!kthread_should_stop()) {
4946
4947                 /* We need to wait INTERRUPTIBLE so that
4948                  * we don't add to the load-average.
4949                  * That means we need to be sure no signals are
4950                  * pending
4951                  */
4952                 if (signal_pending(current))
4953                         flush_signals(current);
4954
4955                 wait_event_interruptible_timeout
4956                         (thread->wqueue,
4957                          test_bit(THREAD_WAKEUP, &thread->flags)
4958                          || kthread_should_stop(),
4959                          thread->timeout);
4960
4961                 clear_bit(THREAD_WAKEUP, &thread->flags);
4962
4963                 thread->run(thread->mddev);
4964         }
4965
4966         return 0;
4967 }
4968
4969 void md_wakeup_thread(mdk_thread_t *thread)
4970 {
4971         if (thread) {
4972                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4973                 set_bit(THREAD_WAKEUP, &thread->flags);
4974                 wake_up(&thread->wqueue);
4975         }
4976 }
4977
4978 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4979                                  const char *name)
4980 {
4981         mdk_thread_t *thread;
4982
4983         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4984         if (!thread)
4985                 return NULL;
4986
4987         init_waitqueue_head(&thread->wqueue);
4988
4989         thread->run = run;
4990         thread->mddev = mddev;
4991         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4992         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4993         if (IS_ERR(thread->tsk)) {
4994                 kfree(thread);
4995                 return NULL;
4996         }
4997         return thread;
4998 }
4999
5000 void md_unregister_thread(mdk_thread_t *thread)
5001 {
5002         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5003
5004         kthread_stop(thread->tsk);
5005         kfree(thread);
5006 }
5007
5008 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5009 {
5010         if (!mddev) {
5011                 MD_BUG();
5012                 return;
5013         }
5014
5015         if (!rdev || test_bit(Faulty, &rdev->flags))
5016                 return;
5017
5018         if (mddev->external)
5019                 set_bit(Blocked, &rdev->flags);
5020 /*
5021         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5022                 mdname(mddev),
5023                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5024                 __builtin_return_address(0),__builtin_return_address(1),
5025                 __builtin_return_address(2),__builtin_return_address(3));
5026 */
5027         if (!mddev->pers)
5028                 return;
5029         if (!mddev->pers->error_handler)
5030                 return;
5031         mddev->pers->error_handler(mddev,rdev);
5032         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5033         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5034         md_wakeup_thread(mddev->thread);
5035         md_new_event_inintr(mddev);
5036 }
5037
5038 /* seq_file implementation /proc/mdstat */
5039
5040 static void status_unused(struct seq_file *seq)
5041 {
5042         int i = 0;
5043         mdk_rdev_t *rdev;
5044         struct list_head *tmp;
5045
5046         seq_printf(seq, "unused devices: ");
5047
5048         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5049                 char b[BDEVNAME_SIZE];
5050                 i++;
5051                 seq_printf(seq, "%s ",
5052                               bdevname(rdev->bdev,b));
5053         }
5054         if (!i)
5055                 seq_printf(seq, "<none>");
5056
5057         seq_printf(seq, "\n");
5058 }
5059
5060
5061 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5062 {
5063         sector_t max_blocks, resync, res;
5064         unsigned long dt, db, rt;
5065         int scale;
5066         unsigned int per_milli;
5067
5068         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5069
5070         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5071                 max_blocks = mddev->resync_max_sectors >> 1;
5072         else
5073                 max_blocks = mddev->size;
5074
5075         /*
5076          * Should not happen.
5077          */
5078         if (!max_blocks) {
5079                 MD_BUG();
5080                 return;
5081         }
5082         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5083          * in a sector_t, and (max_blocks>>scale) will fit in a
5084          * u32, as those are the requirements for sector_div.
5085          * Thus 'scale' must be at least 10
5086          */
5087         scale = 10;
5088         if (sizeof(sector_t) > sizeof(unsigned long)) {
5089                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5090                         scale++;
5091         }
5092         res = (resync>>scale)*1000;
5093         sector_div(res, (u32)((max_blocks>>scale)+1));
5094
5095         per_milli = res;
5096         {
5097                 int i, x = per_milli/50, y = 20-x;
5098                 seq_printf(seq, "[");
5099                 for (i = 0; i < x; i++)
5100                         seq_printf(seq, "=");
5101                 seq_printf(seq, ">");
5102                 for (i = 0; i < y; i++)
5103                         seq_printf(seq, ".");
5104                 seq_printf(seq, "] ");
5105         }
5106         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5107                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5108                     "reshape" :
5109                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5110                      "check" :
5111                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5112                       "resync" : "recovery"))),
5113                    per_milli/10, per_milli % 10,
5114                    (unsigned long long) resync,
5115                    (unsigned long long) max_blocks);
5116
5117         /*
5118          * We do not want to overflow, so the order of operands and
5119          * the * 100 / 100 trick are important. We do a +1 to be
5120          * safe against division by zero. We only estimate anyway.
5121          *
5122          * dt: time from mark until now
5123          * db: blocks written from mark until now
5124          * rt: remaining time
5125          */
5126         dt = ((jiffies - mddev->resync_mark) / HZ);
5127         if (!dt) dt++;
5128         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5129                 - mddev->resync_mark_cnt;
5130         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5131
5132         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5133
5134         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5135 }
5136
5137 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5138 {
5139         struct list_head *tmp;
5140         loff_t l = *pos;
5141         mddev_t *mddev;
5142
5143         if (l >= 0x10000)
5144                 return NULL;
5145         if (!l--)
5146                 /* header */
5147                 return (void*)1;
5148
5149         spin_lock(&all_mddevs_lock);
5150         list_for_each(tmp,&all_mddevs)
5151                 if (!l--) {
5152                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5153                         mddev_get(mddev);
5154                         spin_unlock(&all_mddevs_lock);
5155                         return mddev;
5156                 }
5157         spin_unlock(&all_mddevs_lock);
5158         if (!l--)
5159                 return (void*)2;/* tail */
5160         return NULL;
5161 }
5162
5163 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5164 {
5165         struct list_head *tmp;
5166         mddev_t *next_mddev, *mddev = v;
5167
5168         ++*pos;
5169         if (v == (void*)2)
5170                 return NULL;
5171
5172         spin_lock(&all_mddevs_lock);
5173         if (v == (void*)1)
5174                 tmp = all_mddevs.next;
5175         else
5176                 tmp = mddev->all_mddevs.next;
5177         if (tmp != &all_mddevs)
5178                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5179         else {
5180                 next_mddev = (void*)2;
5181                 *pos = 0x10000;
5182         }
5183         spin_unlock(&all_mddevs_lock);
5184
5185         if (v != (void*)1)
5186                 mddev_put(mddev);
5187         return next_mddev;
5188
5189 }
5190
5191 static void md_seq_stop(struct seq_file *seq, void *v)
5192 {
5193         mddev_t *mddev = v;
5194
5195         if (mddev && v != (void*)1 && v != (void*)2)
5196                 mddev_put(mddev);
5197 }
5198
5199 struct mdstat_info {
5200         int event;
5201 };
5202
5203 static int md_seq_show(struct seq_file *seq, void *v)
5204 {
5205         mddev_t *mddev = v;
5206         sector_t size;
5207         struct list_head *tmp2;
5208         mdk_rdev_t *rdev;
5209         struct mdstat_info *mi = seq->private;
5210         struct bitmap *bitmap;
5211
5212         if (v == (void*)1) {
5213                 struct mdk_personality *pers;
5214                 seq_printf(seq, "Personalities : ");
5215                 spin_lock(&pers_lock);
5216                 list_for_each_entry(pers, &pers_list, list)
5217                         seq_printf(seq, "[%s] ", pers->name);
5218
5219                 spin_unlock(&pers_lock);
5220                 seq_printf(seq, "\n");
5221                 mi->event = atomic_read(&md_event_count);
5222                 return 0;
5223         }
5224         if (v == (void*)2) {
5225                 status_unused(seq);
5226                 return 0;
5227         }
5228
5229         if (mddev_lock(mddev) < 0)
5230                 return -EINTR;
5231
5232         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5233                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5234                                                 mddev->pers ? "" : "in");
5235                 if (mddev->pers) {
5236                         if (mddev->ro==1)
5237                                 seq_printf(seq, " (read-only)");
5238                         if (mddev->ro==2)
5239                                 seq_printf(seq, " (auto-read-only)");
5240                         seq_printf(seq, " %s", mddev->pers->name);
5241                 }
5242
5243                 size = 0;
5244                 rdev_for_each(rdev, tmp2, mddev) {
5245                         char b[BDEVNAME_SIZE];
5246                         seq_printf(seq, " %s[%d]",
5247                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5248                         if (test_bit(WriteMostly, &rdev->flags))
5249                                 seq_printf(seq, "(W)");
5250                         if (test_bit(Faulty, &rdev->flags)) {
5251                                 seq_printf(seq, "(F)");
5252                                 continue;
5253                         } else if (rdev->raid_disk < 0)
5254                                 seq_printf(seq, "(S)"); /* spare */
5255                         size += rdev->size;
5256                 }
5257
5258                 if (!list_empty(&mddev->disks)) {
5259                         if (mddev->pers)
5260                                 seq_printf(seq, "\n      %llu blocks",
5261                                         (unsigned long long)mddev->array_size);
5262                         else
5263                                 seq_printf(seq, "\n      %llu blocks",
5264                                         (unsigned long long)size);
5265                 }
5266                 if (mddev->persistent) {
5267                         if (mddev->major_version != 0 ||
5268                             mddev->minor_version != 90) {
5269                                 seq_printf(seq," super %d.%d",
5270                                            mddev->major_version,
5271                                            mddev->minor_version);
5272                         }
5273                 } else if (mddev->external)
5274                         seq_printf(seq, " super external:%s",
5275                                    mddev->metadata_type);
5276                 else
5277                         seq_printf(seq, " super non-persistent");
5278
5279                 if (mddev->pers) {
5280                         mddev->pers->status (seq, mddev);
5281                         seq_printf(seq, "\n      ");
5282                         if (mddev->pers->sync_request) {
5283                                 if (mddev->curr_resync > 2) {
5284                                         status_resync (seq, mddev);
5285                                         seq_printf(seq, "\n      ");
5286                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5287                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5288                                 else if (mddev->recovery_cp < MaxSector)
5289                                         seq_printf(seq, "\tresync=PENDING\n      ");
5290                         }
5291                 } else
5292                         seq_printf(seq, "\n       ");
5293
5294                 if ((bitmap = mddev->bitmap)) {
5295                         unsigned long chunk_kb;
5296                         unsigned long flags;
5297                         spin_lock_irqsave(&bitmap->lock, flags);
5298                         chunk_kb = bitmap->chunksize >> 10;
5299                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5300                                 "%lu%s chunk",
5301                                 bitmap->pages - bitmap->missing_pages,
5302                                 bitmap->pages,
5303                                 (bitmap->pages - bitmap->missing_pages)
5304                                         << (PAGE_SHIFT - 10),
5305                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5306                                 chunk_kb ? "KB" : "B");
5307                         if (bitmap->file) {
5308                                 seq_printf(seq, ", file: ");
5309                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5310                         }
5311
5312                         seq_printf(seq, "\n");
5313                         spin_unlock_irqrestore(&bitmap->lock, flags);
5314                 }
5315
5316                 seq_printf(seq, "\n");
5317         }
5318         mddev_unlock(mddev);
5319
5320         return 0;
5321 }
5322
5323 static struct seq_operations md_seq_ops = {
5324         .start  = md_seq_start,
5325         .next   = md_seq_next,
5326         .stop   = md_seq_stop,
5327         .show   = md_seq_show,
5328 };
5329
5330 static int md_seq_open(struct inode *inode, struct file *file)
5331 {
5332         int error;
5333         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5334         if (mi == NULL)
5335                 return -ENOMEM;
5336
5337         error = seq_open(file, &md_seq_ops);
5338         if (error)
5339                 kfree(mi);
5340         else {
5341                 struct seq_file *p = file->private_data;
5342                 p->private = mi;
5343                 mi->event = atomic_read(&md_event_count);
5344         }
5345         return error;
5346 }
5347
5348 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5349 {
5350         struct seq_file *m = filp->private_data;
5351         struct mdstat_info *mi = m->private;
5352         int mask;
5353
5354         poll_wait(filp, &md_event_waiters, wait);
5355
5356         /* always allow read */
5357         mask = POLLIN | POLLRDNORM;
5358
5359         if (mi->event != atomic_read(&md_event_count))
5360                 mask |= POLLERR | POLLPRI;
5361         return mask;
5362 }
5363
5364 static const struct file_operations md_seq_fops = {
5365         .owner          = THIS_MODULE,
5366         .open           = md_seq_open,
5367         .read           = seq_read,
5368         .llseek         = seq_lseek,
5369         .release        = seq_release_private,
5370         .poll           = mdstat_poll,
5371 };
5372
5373 int register_md_personality(struct mdk_personality *p)
5374 {
5375         spin_lock(&pers_lock);
5376         list_add_tail(&p->list, &pers_list);
5377         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5378         spin_unlock(&pers_lock);
5379         return 0;
5380 }
5381
5382 int unregister_md_personality(struct mdk_personality *p)
5383 {
5384         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5385         spin_lock(&pers_lock);
5386         list_del_init(&p->list);
5387         spin_unlock(&pers_lock);
5388         return 0;
5389 }
5390
5391 static int is_mddev_idle(mddev_t *mddev)
5392 {
5393         mdk_rdev_t * rdev;
5394         struct list_head *tmp;
5395         int idle;
5396         long curr_events;
5397
5398         idle = 1;
5399         rdev_for_each(rdev, tmp, mddev) {
5400                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5401                 curr_events = disk_stat_read(disk, sectors[0]) +
5402                                 disk_stat_read(disk, sectors[1]) -
5403                                 atomic_read(&disk->sync_io);
5404                 /* sync IO will cause sync_io to increase before the disk_stats
5405                  * as sync_io is counted when a request starts, and
5406                  * disk_stats is counted when it completes.
5407                  * So resync activity will cause curr_events to be smaller than
5408                  * when there was no such activity.
5409                  * non-sync IO will cause disk_stat to increase without
5410                  * increasing sync_io so curr_events will (eventually)
5411                  * be larger than it was before.  Once it becomes
5412                  * substantially larger, the test below will cause
5413                  * the array to appear non-idle, and resync will slow
5414                  * down.
5415                  * If there is a lot of outstanding resync activity when
5416                  * we set last_event to curr_events, then all that activity
5417                  * completing might cause the array to appear non-idle
5418                  * and resync will be slowed down even though there might
5419                  * not have been non-resync activity.  This will only
5420                  * happen once though.  'last_events' will soon reflect
5421                  * the state where there is little or no outstanding
5422                  * resync requests, and further resync activity will
5423                  * always make curr_events less than last_events.
5424                  *
5425                  */
5426                 if (curr_events - rdev->last_events > 4096) {
5427                         rdev->last_events = curr_events;
5428                         idle = 0;
5429                 }
5430         }
5431         return idle;
5432 }
5433
5434 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5435 {
5436         /* another "blocks" (512byte) blocks have been synced */
5437         atomic_sub(blocks, &mddev->recovery_active);
5438         wake_up(&mddev->recovery_wait);
5439         if (!ok) {
5440                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5441                 md_wakeup_thread(mddev->thread);
5442                 // stop recovery, signal do_sync ....
5443         }
5444 }
5445
5446
5447 /* md_write_start(mddev, bi)
5448  * If we need to update some array metadata (e.g. 'active' flag
5449  * in superblock) before writing, schedule a superblock update
5450  * and wait for it to complete.
5451  */
5452 void md_write_start(mddev_t *mddev, struct bio *bi)
5453 {
5454         if (bio_data_dir(bi) != WRITE)
5455                 return;
5456
5457         BUG_ON(mddev->ro == 1);
5458         if (mddev->ro == 2) {
5459                 /* need to switch to read/write */
5460                 mddev->ro = 0;
5461                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5462                 md_wakeup_thread(mddev->thread);
5463                 md_wakeup_thread(mddev->sync_thread);
5464         }
5465         atomic_inc(&mddev->writes_pending);
5466         if (mddev->safemode == 1)
5467                 mddev->safemode = 0;
5468         if (mddev->in_sync) {
5469                 spin_lock_irq(&mddev->write_lock);
5470                 if (mddev->in_sync) {
5471                         mddev->in_sync = 0;
5472                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5473                         md_wakeup_thread(mddev->thread);
5474                 }
5475                 spin_unlock_irq(&mddev->write_lock);
5476                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5477         }
5478         wait_event(mddev->sb_wait,
5479                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5480                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5481 }
5482
5483 void md_write_end(mddev_t *mddev)
5484 {
5485         if (atomic_dec_and_test(&mddev->writes_pending)) {
5486                 if (mddev->safemode == 2)
5487                         md_wakeup_thread(mddev->thread);
5488                 else if (mddev->safemode_delay)
5489                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5490         }
5491 }
5492
5493 /* md_allow_write(mddev)
5494  * Calling this ensures that the array is marked 'active' so that writes
5495  * may proceed without blocking.  It is important to call this before
5496  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5497  * Must be called with mddev_lock held.
5498  */
5499 void md_allow_write(mddev_t *mddev)
5500 {
5501         if (!mddev->pers)
5502                 return;
5503         if (mddev->ro)
5504                 return;
5505
5506         spin_lock_irq(&mddev->write_lock);
5507         if (mddev->in_sync) {
5508                 mddev->in_sync = 0;
5509                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5510                 if (mddev->safemode_delay &&
5511                     mddev->safemode == 0)
5512                         mddev->safemode = 1;
5513                 spin_unlock_irq(&mddev->write_lock);
5514                 md_update_sb(mddev, 0);
5515
5516                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5517                 /* wait for the dirty state to be recorded in the metadata */
5518                 wait_event(mddev->sb_wait,
5519                            !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5520                            !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5521         } else
5522                 spin_unlock_irq(&mddev->write_lock);
5523 }
5524 EXPORT_SYMBOL_GPL(md_allow_write);
5525
5526 #define SYNC_MARKS      10
5527 #define SYNC_MARK_STEP  (3*HZ)
5528 void md_do_sync(mddev_t *mddev)
5529 {
5530         mddev_t *mddev2;
5531         unsigned int currspeed = 0,
5532                  window;
5533         sector_t max_sectors,j, io_sectors;
5534         unsigned long mark[SYNC_MARKS];
5535         sector_t mark_cnt[SYNC_MARKS];
5536         int last_mark,m;
5537         struct list_head *tmp;
5538         sector_t last_check;
5539         int skipped = 0;
5540         struct list_head *rtmp;
5541         mdk_rdev_t *rdev;
5542         char *desc;
5543
5544         /* just incase thread restarts... */
5545         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5546                 return;
5547         if (mddev->ro) /* never try to sync a read-only array */
5548                 return;
5549
5550         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5551                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5552                         desc = "data-check";
5553                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5554                         desc = "requested-resync";
5555                 else
5556                         desc = "resync";
5557         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5558                 desc = "reshape";
5559         else
5560                 desc = "recovery";
5561
5562         /* we overload curr_resync somewhat here.
5563          * 0 == not engaged in resync at all
5564          * 2 == checking that there is no conflict with another sync
5565          * 1 == like 2, but have yielded to allow conflicting resync to
5566          *              commense
5567          * other == active in resync - this many blocks
5568          *
5569          * Before starting a resync we must have set curr_resync to
5570          * 2, and then checked that every "conflicting" array has curr_resync
5571          * less than ours.  When we find one that is the same or higher
5572          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5573          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5574          * This will mean we have to start checking from the beginning again.
5575          *
5576          */
5577
5578         do {
5579                 mddev->curr_resync = 2;
5580
5581         try_again:
5582                 if (kthread_should_stop()) {
5583                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5584                         goto skip;
5585                 }
5586                 for_each_mddev(mddev2, tmp) {
5587                         if (mddev2 == mddev)
5588                                 continue;
5589                         if (!mddev->parallel_resync
5590                         &&  mddev2->curr_resync
5591                         &&  match_mddev_units(mddev, mddev2)) {
5592                                 DEFINE_WAIT(wq);
5593                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5594                                         /* arbitrarily yield */
5595                                         mddev->curr_resync = 1;
5596                                         wake_up(&resync_wait);
5597                                 }
5598                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5599                                         /* no need to wait here, we can wait the next
5600                                          * time 'round when curr_resync == 2
5601                                          */
5602                                         continue;
5603                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5604                                 if (!kthread_should_stop() &&
5605                                     mddev2->curr_resync >= mddev->curr_resync) {
5606                                         printk(KERN_INFO "md: delaying %s of %s"
5607                                                " until %s has finished (they"
5608                                                " share one or more physical units)\n",
5609                                                desc, mdname(mddev), mdname(mddev2));
5610                                         mddev_put(mddev2);
5611                                         schedule();
5612                                         finish_wait(&resync_wait, &wq);
5613                                         goto try_again;
5614                                 }
5615                                 finish_wait(&resync_wait, &wq);
5616                         }
5617                 }
5618         } while (mddev->curr_resync < 2);
5619
5620         j = 0;
5621         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5622                 /* resync follows the size requested by the personality,
5623                  * which defaults to physical size, but can be virtual size
5624                  */
5625                 max_sectors = mddev->resync_max_sectors;
5626                 mddev->resync_mismatches = 0;
5627                 /* we don't use the checkpoint if there's a bitmap */
5628                 if (!mddev->bitmap &&
5629                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5630                         j = mddev->recovery_cp;
5631         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5632                 max_sectors = mddev->size << 1;
5633         else {
5634                 /* recovery follows the physical size of devices */
5635                 max_sectors = mddev->size << 1;
5636                 j = MaxSector;
5637                 rdev_for_each(rdev, rtmp, mddev)
5638                         if (rdev->raid_disk >= 0 &&
5639                             !test_bit(Faulty, &rdev->flags) &&
5640                             !test_bit(In_sync, &rdev->flags) &&
5641                             rdev->recovery_offset < j)
5642                                 j = rdev->recovery_offset;
5643         }
5644
5645         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5646         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5647                 " %d KB/sec/disk.\n", speed_min(mddev));
5648         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5649                "(but not more than %d KB/sec) for %s.\n",
5650                speed_max(mddev), desc);
5651
5652         is_mddev_idle(mddev); /* this also initializes IO event counters */
5653
5654         io_sectors = 0;
5655         for (m = 0; m < SYNC_MARKS; m++) {
5656                 mark[m] = jiffies;
5657                 mark_cnt[m] = io_sectors;
5658         }
5659         last_mark = 0;
5660         mddev->resync_mark = mark[last_mark];
5661         mddev->resync_mark_cnt = mark_cnt[last_mark];
5662
5663         /*
5664          * Tune reconstruction:
5665          */
5666         window = 32*(PAGE_SIZE/512);
5667         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5668                 window/2,(unsigned long long) max_sectors/2);
5669
5670         atomic_set(&mddev->recovery_active, 0);
5671         last_check = 0;
5672
5673         if (j>2) {
5674                 printk(KERN_INFO
5675                        "md: resuming %s of %s from checkpoint.\n",
5676                        desc, mdname(mddev));
5677                 mddev->curr_resync = j;
5678         }
5679
5680         while (j < max_sectors) {
5681                 sector_t sectors;
5682
5683                 skipped = 0;
5684                 if (j >= mddev->resync_max) {
5685                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5686                         wait_event(mddev->recovery_wait,
5687                                    mddev->resync_max > j
5688                                    || kthread_should_stop());
5689                 }
5690                 if (kthread_should_stop())
5691                         goto interrupted;
5692                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5693                                                   currspeed < speed_min(mddev));
5694                 if (sectors == 0) {
5695                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5696                         goto out;
5697                 }
5698
5699                 if (!skipped) { /* actual IO requested */
5700                         io_sectors += sectors;
5701                         atomic_add(sectors, &mddev->recovery_active);
5702                 }
5703
5704                 j += sectors;
5705                 if (j>1) mddev->curr_resync = j;
5706                 mddev->curr_mark_cnt = io_sectors;
5707                 if (last_check == 0)
5708                         /* this is the earliers that rebuilt will be
5709                          * visible in /proc/mdstat
5710                          */
5711                         md_new_event(mddev);
5712
5713                 if (last_check + window > io_sectors || j == max_sectors)
5714                         continue;
5715
5716                 last_check = io_sectors;
5717
5718                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5719                         break;
5720
5721         repeat:
5722                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5723                         /* step marks */
5724                         int next = (last_mark+1) % SYNC_MARKS;
5725
5726                         mddev->resync_mark = mark[next];
5727                         mddev->resync_mark_cnt = mark_cnt[next];
5728                         mark[next] = jiffies;
5729                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5730                         last_mark = next;
5731                 }
5732
5733
5734                 if (kthread_should_stop())
5735                         goto interrupted;
5736
5737
5738                 /*
5739                  * this loop exits only if either when we are slower than
5740                  * the 'hard' speed limit, or the system was IO-idle for
5741                  * a jiffy.
5742                  * the system might be non-idle CPU-wise, but we only care
5743                  * about not overloading the IO subsystem. (things like an
5744                  * e2fsck being done on the RAID array should execute fast)
5745                  */
5746                 blk_unplug(mddev->queue);
5747                 cond_resched();
5748
5749                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5750                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5751
5752                 if (currspeed > speed_min(mddev)) {
5753                         if ((currspeed > speed_max(mddev)) ||
5754                                         !is_mddev_idle(mddev)) {
5755                                 msleep(500);
5756                                 goto repeat;
5757                         }
5758                 }
5759         }
5760         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5761         /*
5762          * this also signals 'finished resyncing' to md_stop
5763          */
5764  out:
5765         blk_unplug(mddev->queue);
5766
5767         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5768
5769         /* tell personality that we are finished */
5770         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5771
5772         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5773             mddev->curr_resync > 2) {
5774                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5775                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5776                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5777                                         printk(KERN_INFO
5778                                                "md: checkpointing %s of %s.\n",
5779                                                desc, mdname(mddev));
5780                                         mddev->recovery_cp = mddev->curr_resync;
5781                                 }
5782                         } else
5783                                 mddev->recovery_cp = MaxSector;
5784                 } else {
5785                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5786                                 mddev->curr_resync = MaxSector;
5787                         rdev_for_each(rdev, rtmp, mddev)
5788                                 if (rdev->raid_disk >= 0 &&
5789                                     !test_bit(Faulty, &rdev->flags) &&
5790                                     !test_bit(In_sync, &rdev->flags) &&
5791                                     rdev->recovery_offset < mddev->curr_resync)
5792                                         rdev->recovery_offset = mddev->curr_resync;
5793                 }
5794         }
5795         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5796
5797  skip:
5798         mddev->curr_resync = 0;
5799         mddev->resync_max = MaxSector;
5800         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5801         wake_up(&resync_wait);
5802         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5803         md_wakeup_thread(mddev->thread);
5804         return;
5805
5806  interrupted:
5807         /*
5808          * got a signal, exit.
5809          */
5810         printk(KERN_INFO
5811                "md: md_do_sync() got signal ... exiting\n");
5812         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5813         goto out;
5814
5815 }
5816 EXPORT_SYMBOL_GPL(md_do_sync);
5817
5818
5819 static int remove_and_add_spares(mddev_t *mddev)
5820 {
5821         mdk_rdev_t *rdev;
5822         struct list_head *rtmp;
5823         int spares = 0;
5824
5825         rdev_for_each(rdev, rtmp, mddev)
5826                 if (rdev->raid_disk >= 0 &&
5827                     !test_bit(Blocked, &rdev->flags) &&
5828                     (test_bit(Faulty, &rdev->flags) ||
5829                      ! test_bit(In_sync, &rdev->flags)) &&
5830                     atomic_read(&rdev->nr_pending)==0) {
5831                         if (mddev->pers->hot_remove_disk(
5832                                     mddev, rdev->raid_disk)==0) {
5833                                 char nm[20];
5834                                 sprintf(nm,"rd%d", rdev->raid_disk);
5835                                 sysfs_remove_link(&mddev->kobj, nm);
5836                                 rdev->raid_disk = -1;
5837                         }
5838                 }
5839
5840         if (mddev->degraded) {
5841                 rdev_for_each(rdev, rtmp, mddev) {
5842                         if (rdev->raid_disk >= 0 &&
5843                             !test_bit(In_sync, &rdev->flags))
5844                                 spares++;
5845                         if (rdev->raid_disk < 0
5846                             && !test_bit(Faulty, &rdev->flags)) {
5847                                 rdev->recovery_offset = 0;
5848                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5849                                         char nm[20];
5850                                         sprintf(nm, "rd%d", rdev->raid_disk);
5851                                         if (sysfs_create_link(&mddev->kobj,
5852                                                               &rdev->kobj, nm))
5853                                                 printk(KERN_WARNING
5854                                                        "md: cannot register "
5855                                                        "%s for %s\n",
5856                                                        nm, mdname(mddev));
5857                                         spares++;
5858                                         md_new_event(mddev);
5859                                 } else
5860                                         break;
5861                         }
5862                 }
5863         }
5864         return spares;
5865 }
5866 /*
5867  * This routine is regularly called by all per-raid-array threads to
5868  * deal with generic issues like resync and super-block update.
5869  * Raid personalities that don't have a thread (linear/raid0) do not
5870  * need this as they never do any recovery or update the superblock.
5871  *
5872  * It does not do any resync itself, but rather "forks" off other threads
5873  * to do that as needed.
5874  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5875  * "->recovery" and create a thread at ->sync_thread.
5876  * When the thread finishes it sets MD_RECOVERY_DONE
5877  * and wakeups up this thread which will reap the thread and finish up.
5878  * This thread also removes any faulty devices (with nr_pending == 0).
5879  *
5880  * The overall approach is:
5881  *  1/ if the superblock needs updating, update it.
5882  *  2/ If a recovery thread is running, don't do anything else.
5883  *  3/ If recovery has finished, clean up, possibly marking spares active.
5884  *  4/ If there are any faulty devices, remove them.
5885  *  5/ If array is degraded, try to add spares devices
5886  *  6/ If array has spares or is not in-sync, start a resync thread.
5887  */
5888 void md_check_recovery(mddev_t *mddev)
5889 {
5890         mdk_rdev_t *rdev;
5891         struct list_head *rtmp;
5892
5893
5894         if (mddev->bitmap)
5895                 bitmap_daemon_work(mddev->bitmap);
5896
5897         if (mddev->ro)
5898                 return;
5899
5900         if (signal_pending(current)) {
5901                 if (mddev->pers->sync_request && !mddev->external) {
5902                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5903                                mdname(mddev));
5904                         mddev->safemode = 2;
5905                 }
5906                 flush_signals(current);
5907         }
5908
5909         if ( ! (
5910                 (mddev->flags && !mddev->external) ||
5911                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5912                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5913                 (mddev->external == 0 && mddev->safemode == 1) ||
5914                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5915                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5916                 ))
5917                 return;
5918
5919         if (mddev_trylock(mddev)) {
5920                 int spares = 0;
5921
5922                 if (!mddev->external) {
5923                         spin_lock_irq(&mddev->write_lock);
5924                         if (mddev->safemode &&
5925                             !atomic_read(&mddev->writes_pending) &&
5926                             !mddev->in_sync &&
5927                             mddev->recovery_cp == MaxSector) {
5928                                 mddev->in_sync = 1;
5929                                 if (mddev->persistent)
5930                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5931                         }
5932                         if (mddev->safemode == 1)
5933                                 mddev->safemode = 0;
5934                         spin_unlock_irq(&mddev->write_lock);
5935                 }
5936
5937                 if (mddev->flags)
5938                         md_update_sb(mddev, 0);
5939
5940
5941                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5942                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5943                         /* resync/recovery still happening */
5944                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5945                         goto unlock;
5946                 }
5947                 if (mddev->sync_thread) {
5948                         /* resync has finished, collect result */
5949                         md_unregister_thread(mddev->sync_thread);
5950                         mddev->sync_thread = NULL;
5951                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5952                                 /* success...*/
5953                                 /* activate any spares */
5954                                 mddev->pers->spare_active(mddev);
5955                         }
5956                         md_update_sb(mddev, 1);
5957
5958                         /* if array is no-longer degraded, then any saved_raid_disk
5959                          * information must be scrapped
5960                          */
5961                         if (!mddev->degraded)
5962                                 rdev_for_each(rdev, rtmp, mddev)
5963                                         rdev->saved_raid_disk = -1;
5964
5965                         mddev->recovery = 0;
5966                         /* flag recovery needed just to double check */
5967                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5968                         md_new_event(mddev);
5969                         goto unlock;
5970                 }
5971                 /* Clear some bits that don't mean anything, but
5972                  * might be left set
5973                  */
5974                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5975                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5976                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5977
5978                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5979                         goto unlock;
5980                 /* no recovery is running.
5981                  * remove any failed drives, then
5982                  * add spares if possible.
5983                  * Spare are also removed and re-added, to allow
5984                  * the personality to fail the re-add.
5985                  */
5986
5987                 if (mddev->reshape_position != MaxSector) {
5988                         if (mddev->pers->check_reshape(mddev) != 0)
5989                                 /* Cannot proceed */
5990                                 goto unlock;
5991                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5992                 } else if ((spares = remove_and_add_spares(mddev))) {
5993                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5994                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5995                 } else if (mddev->recovery_cp < MaxSector) {
5996                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5997                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5998                         /* nothing to be done ... */
5999                         goto unlock;
6000
6001                 if (mddev->pers->sync_request) {
6002                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6003                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6004                                 /* We are adding a device or devices to an array
6005                                  * which has the bitmap stored on all devices.
6006                                  * So make sure all bitmap pages get written
6007                                  */
6008                                 bitmap_write_all(mddev->bitmap);
6009                         }
6010                         mddev->sync_thread = md_register_thread(md_do_sync,
6011                                                                 mddev,
6012                                                                 "%s_resync");
6013                         if (!mddev->sync_thread) {
6014                                 printk(KERN_ERR "%s: could not start resync"
6015                                         " thread...\n",
6016                                         mdname(mddev));
6017                                 /* leave the spares where they are, it shouldn't hurt */
6018                                 mddev->recovery = 0;
6019                         } else
6020                                 md_wakeup_thread(mddev->sync_thread);
6021                         md_new_event(mddev);
6022                 }
6023         unlock:
6024                 mddev_unlock(mddev);
6025         }
6026 }
6027
6028 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6029 {
6030         sysfs_notify(&rdev->kobj, NULL, "state");
6031         wait_event_timeout(rdev->blocked_wait,
6032                            !test_bit(Blocked, &rdev->flags),
6033                            msecs_to_jiffies(5000));
6034         rdev_dec_pending(rdev, mddev);
6035 }
6036 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6037
6038 static int md_notify_reboot(struct notifier_block *this,
6039                             unsigned long code, void *x)
6040 {
6041         struct list_head *tmp;
6042         mddev_t *mddev;
6043
6044         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6045
6046                 printk(KERN_INFO "md: stopping all md devices.\n");
6047
6048                 for_each_mddev(mddev, tmp)
6049                         if (mddev_trylock(mddev)) {
6050                                 do_md_stop (mddev, 1);
6051                                 mddev_unlock(mddev);
6052                         }
6053                 /*
6054                  * certain more exotic SCSI devices are known to be
6055                  * volatile wrt too early system reboots. While the
6056                  * right place to handle this issue is the given
6057                  * driver, we do want to have a safe RAID driver ...
6058                  */
6059                 mdelay(1000*1);
6060         }
6061         return NOTIFY_DONE;
6062 }
6063
6064 static struct notifier_block md_notifier = {
6065         .notifier_call  = md_notify_reboot,
6066         .next           = NULL,
6067         .priority       = INT_MAX, /* before any real devices */
6068 };
6069
6070 static void md_geninit(void)
6071 {
6072         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6073
6074         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6075 }
6076
6077 static int __init md_init(void)
6078 {
6079         if (register_blkdev(MAJOR_NR, "md"))
6080                 return -1;
6081         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6082                 unregister_blkdev(MAJOR_NR, "md");
6083                 return -1;
6084         }
6085         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6086                             md_probe, NULL, NULL);
6087         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6088                             md_probe, NULL, NULL);
6089
6090         register_reboot_notifier(&md_notifier);
6091         raid_table_header = register_sysctl_table(raid_root_table);
6092
6093         md_geninit();
6094         return (0);
6095 }
6096
6097
6098 #ifndef MODULE
6099
6100 /*
6101  * Searches all registered partitions for autorun RAID arrays
6102  * at boot time.
6103  */
6104
6105 static LIST_HEAD(all_detected_devices);
6106 struct detected_devices_node {
6107         struct list_head list;
6108         dev_t dev;
6109 };
6110
6111 void md_autodetect_dev(dev_t dev)
6112 {
6113         struct detected_devices_node *node_detected_dev;
6114
6115         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6116         if (node_detected_dev) {
6117                 node_detected_dev->dev = dev;
6118                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6119         } else {
6120                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6121                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6122         }
6123 }
6124
6125
6126 static void autostart_arrays(int part)
6127 {
6128         mdk_rdev_t *rdev;
6129         struct detected_devices_node *node_detected_dev;
6130         dev_t dev;
6131         int i_scanned, i_passed;
6132
6133         i_scanned = 0;
6134         i_passed = 0;
6135
6136         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6137
6138         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6139                 i_scanned++;
6140                 node_detected_dev = list_entry(all_detected_devices.next,
6141                                         struct detected_devices_node, list);
6142                 list_del(&node_detected_dev->list);
6143                 dev = node_detected_dev->dev;
6144                 kfree(node_detected_dev);
6145                 rdev = md_import_device(dev,0, 90);
6146                 if (IS_ERR(rdev))
6147                         continue;
6148
6149                 if (test_bit(Faulty, &rdev->flags)) {
6150                         MD_BUG();
6151                         continue;
6152                 }
6153                 set_bit(AutoDetected, &rdev->flags);
6154                 list_add(&rdev->same_set, &pending_raid_disks);
6155                 i_passed++;
6156         }
6157
6158         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6159                                                 i_scanned, i_passed);
6160
6161         autorun_devices(part);
6162 }
6163
6164 #endif /* !MODULE */
6165
6166 static __exit void md_exit(void)
6167 {
6168         mddev_t *mddev;
6169         struct list_head *tmp;
6170
6171         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6172         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6173
6174         unregister_blkdev(MAJOR_NR,"md");
6175         unregister_blkdev(mdp_major, "mdp");
6176         unregister_reboot_notifier(&md_notifier);
6177         unregister_sysctl_table(raid_table_header);
6178         remove_proc_entry("mdstat", NULL);
6179         for_each_mddev(mddev, tmp) {
6180                 struct gendisk *disk = mddev->gendisk;
6181                 if (!disk)
6182                         continue;
6183                 export_array(mddev);
6184                 del_gendisk(disk);
6185                 put_disk(disk);
6186                 mddev->gendisk = NULL;
6187                 mddev_put(mddev);
6188         }
6189 }
6190
6191 subsys_initcall(md_init);
6192 module_exit(md_exit)
6193
6194 static int get_ro(char *buffer, struct kernel_param *kp)
6195 {
6196         return sprintf(buffer, "%d", start_readonly);
6197 }
6198 static int set_ro(const char *val, struct kernel_param *kp)
6199 {
6200         char *e;
6201         int num = simple_strtoul(val, &e, 10);
6202         if (*val && (*e == '\0' || *e == '\n')) {
6203                 start_readonly = num;
6204                 return 0;
6205         }
6206         return -EINVAL;
6207 }
6208
6209 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6210 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6211
6212
6213 EXPORT_SYMBOL(register_md_personality);
6214 EXPORT_SYMBOL(unregister_md_personality);
6215 EXPORT_SYMBOL(md_error);
6216 EXPORT_SYMBOL(md_done_sync);
6217 EXPORT_SYMBOL(md_write_start);
6218 EXPORT_SYMBOL(md_write_end);
6219 EXPORT_SYMBOL(md_register_thread);
6220 EXPORT_SYMBOL(md_unregister_thread);
6221 EXPORT_SYMBOL(md_wakeup_thread);
6222 EXPORT_SYMBOL(md_check_recovery);
6223 MODULE_LICENSE("GPL");
6224 MODULE_ALIAS("md");
6225 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);