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