drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/kernel.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/buffer_head.h> /* for invalidate_bdev */
  43 #include <linux/poll.h>
  44 #include <linux/mutex.h>
  45 #include <linux/ctype.h>
  46 #include <linux/freezer.h>
  47
  48 #include <linux/init.h>
  49
  50 #include <linux/file.h>
  51
  52 #ifdef CONFIG_KMOD
  53 #include <linux/kmod.h>
  54 #endif
  55
  56 #include <asm/unaligned.h>
  57
  58 #define MAJOR_NR MD_MAJOR
  59 #define MD_DRIVER
  60
  61 /* 63 partitions with the alternate major number (mdp) */
  62 #define MdpMinorShift 6
  63
  64 #define DEBUG 0
  65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  66
  67
  68 #ifndef MODULE
  69 static void autostart_arrays (int part);
  70 #endif
  71
  72 static LIST_HEAD(pers_list);
  73 static DEFINE_SPINLOCK(pers_lock);
  74
  75 static void md_print_devices(void);
  76
  77 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  78
  79 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  80
  81 /*
  82  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  83  * is 1000 KB/sec, so the extra system load does not show up that much.
  84  * Increase it if you want to have more _guaranteed_ speed. Note that
  85  * the RAID driver will use the maximum available bandwidth if the IO
  86  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  87  * speed limit - in case reconstruction slows down your system despite
  88  * idle IO detection.
  89  *
  90  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  91  * or /sys/block/mdX/md/sync_speed_{min,max}
  92  */
  93
  94 static int sysctl_speed_limit_min = 1000;
  95 static int sysctl_speed_limit_max = 200000;
  96 static inline int speed_min(mddev_t *mddev)
  97 {
  98         return mddev->sync_speed_min ?
  99                 mddev->sync_speed_min : sysctl_speed_limit_min;
 100 }
 101
 102 static inline int speed_max(mddev_t *mddev)
 103 {
 104         return mddev->sync_speed_max ?
 105                 mddev->sync_speed_max : sysctl_speed_limit_max;
 106 }
 107
 108 static struct ctl_table_header *raid_table_header;
 109
 110 static ctl_table raid_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 113                 .procname       = "speed_limit_min",
 114                 .data           = &sysctl_speed_limit_min,
 115                 .maxlen         = sizeof(int),
 116                 .mode           = S_IRUGO|S_IWUSR,
 117                 .proc_handler   = &proc_dointvec,
 118         },
 119         {
 120                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 121                 .procname       = "speed_limit_max",
 122                 .data           = &sysctl_speed_limit_max,
 123                 .maxlen         = sizeof(int),
 124                 .mode           = S_IRUGO|S_IWUSR,
 125                 .proc_handler   = &proc_dointvec,
 126         },
 127         { .ctl_name = 0 }
 128 };
 129
 130 static ctl_table raid_dir_table[] = {
 131         {
 132                 .ctl_name       = DEV_RAID,
 133                 .procname       = "raid",
 134                 .maxlen         = 0,
 135                 .mode           = S_IRUGO|S_IXUGO,
 136                 .child          = raid_table,
 137         },
 138         { .ctl_name = 0 }
 139 };
 140
 141 static ctl_table raid_root_table[] = {
 142         {
 143                 .ctl_name       = CTL_DEV,
 144                 .procname       = "dev",
 145                 .maxlen         = 0,
 146                 .mode           = 0555,
 147                 .child          = raid_dir_table,
 148         },
 149         { .ctl_name = 0 }
 150 };
 151
 152 static struct block_device_operations md_fops;
 153
 154 static int start_readonly;
 155
 156 /*
 157  * We have a system wide 'event count' that is incremented
 158  * on any 'interesting' event, and readers of /proc/mdstat
 159  * can use 'poll' or 'select' to find out when the event
 160  * count increases.
 161  *
 162  * Events are:
 163  *  start array, stop array, error, add device, remove device,
 164  *  start build, activate spare
 165  */
 166 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 167 static atomic_t md_event_count;
 168 void md_new_event(mddev_t *mddev)
 169 {
 170         atomic_inc(&md_event_count);
 171         wake_up(&md_event_waiters);
 172         sysfs_notify(&mddev->kobj, NULL, "sync_action");
 173 }
 174 EXPORT_SYMBOL_GPL(md_new_event);
 175
 176 /* Alternate version that can be called from interrupts
 177  * when calling sysfs_notify isn't needed.
 178  */
 179 static void md_new_event_inintr(mddev_t *mddev)
 180 {
 181         atomic_inc(&md_event_count);
 182         wake_up(&md_event_waiters);
 183 }
 184
 185 /*
 186  * Enables to iterate over all existing md arrays
 187  * all_mddevs_lock protects this list.
 188  */
 189 static LIST_HEAD(all_mddevs);
 190 static DEFINE_SPINLOCK(all_mddevs_lock);
 191
 192
 193 /*
 194  * iterates through all used mddevs in the system.
 195  * We take care to grab the all_mddevs_lock whenever navigating
 196  * the list, and to always hold a refcount when unlocked.
 197  * Any code which breaks out of this loop while own
 198  * a reference to the current mddev and must mddev_put it.
 199  */
 200 #define for_each_mddev(mddev,tmp)                                       \
 201                                                                         \
 202         for (({ spin_lock(&all_mddevs_lock);                            \
 203                 tmp = all_mddevs.next;                                  \
 204                 mddev = NULL;});                                        \
 205              ({ if (tmp != &all_mddevs)                                 \
 206                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 207                 spin_unlock(&all_mddevs_lock);                          \
 208                 if (mddev) mddev_put(mddev);                            \
 209                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 210                 tmp != &all_mddevs;});                                  \
 211              ({ spin_lock(&all_mddevs_lock);                            \
 212                 tmp = tmp->next;})                                      \
 213                 )
 214
 215
 216 static int md_fail_request (struct request_queue *q, struct bio *bio)
 217 {
 218         bio_io_error(bio);
 219         return 0;
 220 }
 221
 222 static inline mddev_t *mddev_get(mddev_t *mddev)
 223 {
 224         atomic_inc(&mddev->active);
 225         return mddev;
 226 }
 227
 228 static void mddev_put(mddev_t *mddev)
 229 {
 230         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 231                 return;
 232         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 233                 list_del(&mddev->all_mddevs);
 234                 spin_unlock(&all_mddevs_lock);
 235                 blk_cleanup_queue(mddev->queue);
 236                 kobject_put(&mddev->kobj);
 237         } else
 238                 spin_unlock(&all_mddevs_lock);
 239 }
 240
 241 static mddev_t * mddev_find(dev_t unit)
 242 {
 243         mddev_t *mddev, *new = NULL;
 244
 245  retry:
 246         spin_lock(&all_mddevs_lock);
 247         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 248                 if (mddev->unit == unit) {
 249                         mddev_get(mddev);
 250                         spin_unlock(&all_mddevs_lock);
 251                         kfree(new);
 252                         return mddev;
 253                 }
 254
 255         if (new) {
 256                 list_add(&new->all_mddevs, &all_mddevs);
 257                 spin_unlock(&all_mddevs_lock);
 258                 return new;
 259         }
 260         spin_unlock(&all_mddevs_lock);
 261
 262         new = kzalloc(sizeof(*new), GFP_KERNEL);
 263         if (!new)
 264                 return NULL;
 265
 266         new->unit = unit;
 267         if (MAJOR(unit) == MD_MAJOR)
 268                 new->md_minor = MINOR(unit);
 269         else
 270                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 271
 272         mutex_init(&new->reconfig_mutex);
 273         INIT_LIST_HEAD(&new->disks);
 274         INIT_LIST_HEAD(&new->all_mddevs);
 275         init_timer(&new->safemode_timer);
 276         atomic_set(&new->active, 1);
 277         spin_lock_init(&new->write_lock);
 278         init_waitqueue_head(&new->sb_wait);
 279         init_waitqueue_head(&new->recovery_wait);
 280         new->reshape_position = MaxSector;
 281         new->resync_min = 0;
 282         new->resync_max = MaxSector;
 283         new->level = LEVEL_NONE;
 284
 285         new->queue = blk_alloc_queue(GFP_KERNEL);
 286         if (!new->queue) {
 287                 kfree(new);
 288                 return NULL;
 289         }
 290         /* Can be unlocked because the queue is new: no concurrency */
 291         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
 292
 293         blk_queue_make_request(new->queue, md_fail_request);
 294
 295         goto retry;
 296 }
 297
 298 static inline int mddev_lock(mddev_t * mddev)
 299 {
 300         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 301 }
 302
 303 static inline int mddev_trylock(mddev_t * mddev)
 304 {
 305         return mutex_trylock(&mddev->reconfig_mutex);
 306 }
 307
 308 static inline void mddev_unlock(mddev_t * mddev)
 309 {
 310         mutex_unlock(&mddev->reconfig_mutex);
 311
 312         md_wakeup_thread(mddev->thread);
 313 }
 314
 315 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 316 {
 317         mdk_rdev_t * rdev;
 318         struct list_head *tmp;
 319
 320         rdev_for_each(rdev, tmp, mddev) {
 321                 if (rdev->desc_nr == nr)
 322                         return rdev;
 323         }
 324         return NULL;
 325 }
 326
 327 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 328 {
 329         struct list_head *tmp;
 330         mdk_rdev_t *rdev;
 331
 332         rdev_for_each(rdev, tmp, mddev) {
 333                 if (rdev->bdev->bd_dev == dev)
 334                         return rdev;
 335         }
 336         return NULL;
 337 }
 338
 339 static struct mdk_personality *find_pers(int level, char *clevel)
 340 {
 341         struct mdk_personality *pers;
 342         list_for_each_entry(pers, &pers_list, list) {
 343                 if (level != LEVEL_NONE && pers->level == level)
 344                         return pers;
 345                 if (strcmp(pers->name, clevel)==0)
 346                         return pers;
 347         }
 348         return NULL;
 349 }
 350
 351 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 352 {
 353         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 354         return MD_NEW_SIZE_BLOCKS(size);
 355 }
 356
 357 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 358 {
 359         sector_t size;
 360
 361         size = rdev->sb_offset;
 362
 363         if (chunk_size)
 364                 size &= ~((sector_t)chunk_size/1024 - 1);
 365         return size;
 366 }
 367
 368 static int alloc_disk_sb(mdk_rdev_t * rdev)
 369 {
 370         if (rdev->sb_page)
 371                 MD_BUG();
 372
 373         rdev->sb_page = alloc_page(GFP_KERNEL);
 374         if (!rdev->sb_page) {
 375                 printk(KERN_ALERT "md: out of memory.\n");
 376                 return -EINVAL;
 377         }
 378
 379         return 0;
 380 }
 381
 382 static void free_disk_sb(mdk_rdev_t * rdev)
 383 {
 384         if (rdev->sb_page) {
 385                 put_page(rdev->sb_page);
 386                 rdev->sb_loaded = 0;
 387                 rdev->sb_page = NULL;
 388                 rdev->sb_offset = 0;
 389                 rdev->size = 0;
 390         }
 391 }
 392
 393
 394 static void super_written(struct bio *bio, int error)
 395 {
 396         mdk_rdev_t *rdev = bio->bi_private;
 397         mddev_t *mddev = rdev->mddev;
 398
 399         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 400                 printk("md: super_written gets error=%d, uptodate=%d\n",
 401                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 402                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 403                 md_error(mddev, rdev);
 404         }
 405
 406         if (atomic_dec_and_test(&mddev->pending_writes))
 407                 wake_up(&mddev->sb_wait);
 408         bio_put(bio);
 409 }
 410
 411 static void super_written_barrier(struct bio *bio, int error)
 412 {
 413         struct bio *bio2 = bio->bi_private;
 414         mdk_rdev_t *rdev = bio2->bi_private;
 415         mddev_t *mddev = rdev->mddev;
 416
 417         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 418             error == -EOPNOTSUPP) {
 419                 unsigned long flags;
 420                 /* barriers don't appear to be supported :-( */
 421                 set_bit(BarriersNotsupp, &rdev->flags);
 422                 mddev->barriers_work = 0;
 423                 spin_lock_irqsave(&mddev->write_lock, flags);
 424                 bio2->bi_next = mddev->biolist;
 425                 mddev->biolist = bio2;
 426                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 427                 wake_up(&mddev->sb_wait);
 428                 bio_put(bio);
 429         } else {
 430                 bio_put(bio2);
 431                 bio->bi_private = rdev;
 432                 super_written(bio, error);
 433         }
 434 }
 435
 436 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 437                    sector_t sector, int size, struct page *page)
 438 {
 439         /* write first size bytes of page to sector of rdev
 440          * Increment mddev->pending_writes before returning
 441          * and decrement it on completion, waking up sb_wait
 442          * if zero is reached.
 443          * If an error occurred, call md_error
 444          *
 445          * As we might need to resubmit the request if BIO_RW_BARRIER
 446          * causes ENOTSUPP, we allocate a spare bio...
 447          */
 448         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 449         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 450
 451         bio->bi_bdev = rdev->bdev;
 452         bio->bi_sector = sector;
 453         bio_add_page(bio, page, size, 0);
 454         bio->bi_private = rdev;
 455         bio->bi_end_io = super_written;
 456         bio->bi_rw = rw;
 457
 458         atomic_inc(&mddev->pending_writes);
 459         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 460                 struct bio *rbio;
 461                 rw |= (1<<BIO_RW_BARRIER);
 462                 rbio = bio_clone(bio, GFP_NOIO);
 463                 rbio->bi_private = bio;
 464                 rbio->bi_end_io = super_written_barrier;
 465                 submit_bio(rw, rbio);
 466         } else
 467                 submit_bio(rw, bio);
 468 }
 469
 470 void md_super_wait(mddev_t *mddev)
 471 {
 472         /* wait for all superblock writes that were scheduled to complete.
 473          * if any had to be retried (due to BARRIER problems), retry them
 474          */
 475         DEFINE_WAIT(wq);
 476         for(;;) {
 477                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 478                 if (atomic_read(&mddev->pending_writes)==0)
 479                         break;
 480                 while (mddev->biolist) {
 481                         struct bio *bio;
 482                         spin_lock_irq(&mddev->write_lock);
 483                         bio = mddev->biolist;
 484                         mddev->biolist = bio->bi_next ;
 485                         bio->bi_next = NULL;
 486                         spin_unlock_irq(&mddev->write_lock);
 487                         submit_bio(bio->bi_rw, bio);
 488                 }
 489                 schedule();
 490         }
 491         finish_wait(&mddev->sb_wait, &wq);
 492 }
 493
 494 static void bi_complete(struct bio *bio, int error)
 495 {
 496         complete((struct completion*)bio->bi_private);
 497 }
 498
 499 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 500                    struct page *page, int rw)
 501 {
 502         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 503         struct completion event;
 504         int ret;
 505
 506         rw |= (1 << BIO_RW_SYNC);
 507
 508         bio->bi_bdev = bdev;
 509         bio->bi_sector = sector;
 510         bio_add_page(bio, page, size, 0);
 511         init_completion(&event);
 512         bio->bi_private = &event;
 513         bio->bi_end_io = bi_complete;
 514         submit_bio(rw, bio);
 515         wait_for_completion(&event);
 516
 517         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 518         bio_put(bio);
 519         return ret;
 520 }
 521 EXPORT_SYMBOL_GPL(sync_page_io);
 522
 523 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 524 {
 525         char b[BDEVNAME_SIZE];
 526         if (!rdev->sb_page) {
 527                 MD_BUG();
 528                 return -EINVAL;
 529         }
 530         if (rdev->sb_loaded)
 531                 return 0;
 532
 533
 534         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 535                 goto fail;
 536         rdev->sb_loaded = 1;
 537         return 0;
 538
 539 fail:
 540         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 541                 bdevname(rdev->bdev,b));
 542         return -EINVAL;
 543 }
 544
 545 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 546 {
 547         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 548                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 549                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 550                 (sb1->set_uuid3 == sb2->set_uuid3))
 551
 552                 return 1;
 553
 554         return 0;
 555 }
 556
 557
 558 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 559 {
 560         int ret;
 561         mdp_super_t *tmp1, *tmp2;
 562
 563         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 564         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 565
 566         if (!tmp1 || !tmp2) {
 567                 ret = 0;
 568                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 569                 goto abort;
 570         }
 571
 572         *tmp1 = *sb1;
 573         *tmp2 = *sb2;
 574
 575         /*
 576          * nr_disks is not constant
 577          */
 578         tmp1->nr_disks = 0;
 579         tmp2->nr_disks = 0;
 580
 581         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 582                 ret = 0;
 583         else
 584                 ret = 1;
 585
 586 abort:
 587         kfree(tmp1);
 588         kfree(tmp2);
 589         return ret;
 590 }
 591
 592
 593 static u32 md_csum_fold(u32 csum)
 594 {
 595         csum = (csum & 0xffff) + (csum >> 16);
 596         return (csum & 0xffff) + (csum >> 16);
 597 }
 598
 599 static unsigned int calc_sb_csum(mdp_super_t * sb)
 600 {
 601         u64 newcsum = 0;
 602         u32 *sb32 = (u32*)sb;
 603         int i;
 604         unsigned int disk_csum, csum;
 605
 606         disk_csum = sb->sb_csum;
 607         sb->sb_csum = 0;
 608
 609         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 610                 newcsum += sb32[i];
 611         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 612
 613
 614 #ifdef CONFIG_ALPHA
 615         /* This used to use csum_partial, which was wrong for several
 616          * reasons including that different results are returned on
 617          * different architectures.  It isn't critical that we get exactly
 618          * the same return value as before (we always csum_fold before
 619          * testing, and that removes any differences).  However as we
 620          * know that csum_partial always returned a 16bit value on
 621          * alphas, do a fold to maximise conformity to previous behaviour.
 622          */
 623         sb->sb_csum = md_csum_fold(disk_csum);
 624 #else
 625         sb->sb_csum = disk_csum;
 626 #endif
 627         return csum;
 628 }
 629
 630
 631 /*
 632  * Handle superblock details.
 633  * We want to be able to handle multiple superblock formats
 634  * so we have a common interface to them all, and an array of
 635  * different handlers.
 636  * We rely on user-space to write the initial superblock, and support
 637  * reading and updating of superblocks.
 638  * Interface methods are:
 639  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 640  *      loads and validates a superblock on dev.
 641  *      if refdev != NULL, compare superblocks on both devices
 642  *    Return:
 643  *      0 - dev has a superblock that is compatible with refdev
 644  *      1 - dev has a superblock that is compatible and newer than refdev
 645  *          so dev should be used as the refdev in future
 646  *     -EINVAL superblock incompatible or invalid
 647  *     -othererror e.g. -EIO
 648  *
 649  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 650  *      Verify that dev is acceptable into mddev.
 651  *       The first time, mddev->raid_disks will be 0, and data from
 652  *       dev should be merged in.  Subsequent calls check that dev
 653  *       is new enough.  Return 0 or -EINVAL
 654  *
 655  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 656  *     Update the superblock for rdev with data in mddev
 657  *     This does not write to disc.
 658  *
 659  */
 660
 661 struct super_type  {
 662         char            *name;
 663         struct module   *owner;
 664         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 665         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 666         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 667 };
 668
 669 /*
 670  * load_super for 0.90.0
 671  */
 672 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 673 {
 674         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 675         mdp_super_t *sb;
 676         int ret;
 677         sector_t sb_offset;
 678
 679         /*
 680          * Calculate the position of the superblock,
 681          * it's at the end of the disk.
 682          *
 683          * It also happens to be a multiple of 4Kb.
 684          */
 685         sb_offset = calc_dev_sboffset(rdev->bdev);
 686         rdev->sb_offset = sb_offset;
 687
 688         ret = read_disk_sb(rdev, MD_SB_BYTES);
 689         if (ret) return ret;
 690
 691         ret = -EINVAL;
 692
 693         bdevname(rdev->bdev, b);
 694         sb = (mdp_super_t*)page_address(rdev->sb_page);
 695
 696         if (sb->md_magic != MD_SB_MAGIC) {
 697                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 698                        b);
 699                 goto abort;
 700         }
 701
 702         if (sb->major_version != 0 ||
 703             sb->minor_version < 90 ||
 704             sb->minor_version > 91) {
 705                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 706                         sb->major_version, sb->minor_version,
 707                         b);
 708                 goto abort;
 709         }
 710
 711         if (sb->raid_disks <= 0)
 712                 goto abort;
 713
 714         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 715                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 716                         b);
 717                 goto abort;
 718         }
 719
 720         rdev->preferred_minor = sb->md_minor;
 721         rdev->data_offset = 0;
 722         rdev->sb_size = MD_SB_BYTES;
 723
 724         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 725                 if (sb->level != 1 && sb->level != 4
 726                     && sb->level != 5 && sb->level != 6
 727                     && sb->level != 10) {
 728                         /* FIXME use a better test */
 729                         printk(KERN_WARNING
 730                                "md: bitmaps not supported for this level.\n");
 731                         goto abort;
 732                 }
 733         }
 734
 735         if (sb->level == LEVEL_MULTIPATH)
 736                 rdev->desc_nr = -1;
 737         else
 738                 rdev->desc_nr = sb->this_disk.number;
 739
 740         if (!refdev) {
 741                 ret = 1;
 742         } else {
 743                 __u64 ev1, ev2;
 744                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 745                 if (!uuid_equal(refsb, sb)) {
 746                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 747                                 b, bdevname(refdev->bdev,b2));
 748                         goto abort;
 749                 }
 750                 if (!sb_equal(refsb, sb)) {
 751                         printk(KERN_WARNING "md: %s has same UUID"
 752                                " but different superblock to %s\n",
 753                                b, bdevname(refdev->bdev, b2));
 754                         goto abort;
 755                 }
 756                 ev1 = md_event(sb);
 757                 ev2 = md_event(refsb);
 758                 if (ev1 > ev2)
 759                         ret = 1;
 760                 else
 761                         ret = 0;
 762         }
 763         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 764
 765         if (rdev->size < sb->size && sb->level > 1)
 766                 /* "this cannot possibly happen" ... */
 767                 ret = -EINVAL;
 768
 769  abort:
 770         return ret;
 771 }
 772
 773 /*
 774  * validate_super for 0.90.0
 775  */
 776 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 777 {
 778         mdp_disk_t *desc;
 779         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 780         __u64 ev1 = md_event(sb);
 781
 782         rdev->raid_disk = -1;
 783         clear_bit(Faulty, &rdev->flags);
 784         clear_bit(In_sync, &rdev->flags);
 785         clear_bit(WriteMostly, &rdev->flags);
 786         clear_bit(BarriersNotsupp, &rdev->flags);
 787
 788         if (mddev->raid_disks == 0) {
 789                 mddev->major_version = 0;
 790                 mddev->minor_version = sb->minor_version;
 791                 mddev->patch_version = sb->patch_version;
 792                 mddev->external = 0;
 793                 mddev->chunk_size = sb->chunk_size;
 794                 mddev->ctime = sb->ctime;
 795                 mddev->utime = sb->utime;
 796                 mddev->level = sb->level;
 797                 mddev->clevel[0] = 0;
 798                 mddev->layout = sb->layout;
 799                 mddev->raid_disks = sb->raid_disks;
 800                 mddev->size = sb->size;
 801                 mddev->events = ev1;
 802                 mddev->bitmap_offset = 0;
 803                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 804
 805                 if (mddev->minor_version >= 91) {
 806                         mddev->reshape_position = sb->reshape_position;
 807                         mddev->delta_disks = sb->delta_disks;
 808                         mddev->new_level = sb->new_level;
 809                         mddev->new_layout = sb->new_layout;
 810                         mddev->new_chunk = sb->new_chunk;
 811                 } else {
 812                         mddev->reshape_position = MaxSector;
 813                         mddev->delta_disks = 0;
 814                         mddev->new_level = mddev->level;
 815                         mddev->new_layout = mddev->layout;
 816                         mddev->new_chunk = mddev->chunk_size;
 817                 }
 818
 819                 if (sb->state & (1<<MD_SB_CLEAN))
 820                         mddev->recovery_cp = MaxSector;
 821                 else {
 822                         if (sb->events_hi == sb->cp_events_hi &&
 823                                 sb->events_lo == sb->cp_events_lo) {
 824                                 mddev->recovery_cp = sb->recovery_cp;
 825                         } else
 826                                 mddev->recovery_cp = 0;
 827                 }
 828
 829                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 830                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 831                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 832                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 833
 834                 mddev->max_disks = MD_SB_DISKS;
 835
 836                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 837                     mddev->bitmap_file == NULL)
 838                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 839
 840         } else if (mddev->pers == NULL) {
 841                 /* Insist on good event counter while assembling */
 842                 ++ev1;
 843                 if (ev1 < mddev->events)
 844                         return -EINVAL;
 845         } else if (mddev->bitmap) {
 846                 /* if adding to array with a bitmap, then we can accept an
 847                  * older device ... but not too old.
 848                  */
 849                 if (ev1 < mddev->bitmap->events_cleared)
 850                         return 0;
 851         } else {
 852                 if (ev1 < mddev->events)
 853                         /* just a hot-add of a new device, leave raid_disk at -1 */
 854                         return 0;
 855         }
 856
 857         if (mddev->level != LEVEL_MULTIPATH) {
 858                 desc = sb->disks + rdev->desc_nr;
 859
 860                 if (desc->state & (1<<MD_DISK_FAULTY))
 861                         set_bit(Faulty, &rdev->flags);
 862                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
 863                             desc->raid_disk < mddev->raid_disks */) {
 864                         set_bit(In_sync, &rdev->flags);
 865                         rdev->raid_disk = desc->raid_disk;
 866                 }
 867                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 868                         set_bit(WriteMostly, &rdev->flags);
 869         } else /* MULTIPATH are always insync */
 870                 set_bit(In_sync, &rdev->flags);
 871         return 0;
 872 }
 873
 874 /*
 875  * sync_super for 0.90.0
 876  */
 877 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 878 {
 879         mdp_super_t *sb;
 880         struct list_head *tmp;
 881         mdk_rdev_t *rdev2;
 882         int next_spare = mddev->raid_disks;
 883
 884
 885         /* make rdev->sb match mddev data..
 886          *
 887          * 1/ zero out disks
 888          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 889          * 3/ any empty disks < next_spare become removed
 890          *
 891          * disks[0] gets initialised to REMOVED because
 892          * we cannot be sure from other fields if it has
 893          * been initialised or not.
 894          */
 895         int i;
 896         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 897
 898         rdev->sb_size = MD_SB_BYTES;
 899
 900         sb = (mdp_super_t*)page_address(rdev->sb_page);
 901
 902         memset(sb, 0, sizeof(*sb));
 903
 904         sb->md_magic = MD_SB_MAGIC;
 905         sb->major_version = mddev->major_version;
 906         sb->patch_version = mddev->patch_version;
 907         sb->gvalid_words  = 0; /* ignored */
 908         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 909         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 910         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 911         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 912
 913         sb->ctime = mddev->ctime;
 914         sb->level = mddev->level;
 915         sb->size  = mddev->size;
 916         sb->raid_disks = mddev->raid_disks;
 917         sb->md_minor = mddev->md_minor;
 918         sb->not_persistent = 0;
 919         sb->utime = mddev->utime;
 920         sb->state = 0;
 921         sb->events_hi = (mddev->events>>32);
 922         sb->events_lo = (u32)mddev->events;
 923
 924         if (mddev->reshape_position == MaxSector)
 925                 sb->minor_version = 90;
 926         else {
 927                 sb->minor_version = 91;
 928                 sb->reshape_position = mddev->reshape_position;
 929                 sb->new_level = mddev->new_level;
 930                 sb->delta_disks = mddev->delta_disks;
 931                 sb->new_layout = mddev->new_layout;
 932                 sb->new_chunk = mddev->new_chunk;
 933         }
 934         mddev->minor_version = sb->minor_version;
 935         if (mddev->in_sync)
 936         {
 937                 sb->recovery_cp = mddev->recovery_cp;
 938                 sb->cp_events_hi = (mddev->events>>32);
 939                 sb->cp_events_lo = (u32)mddev->events;
 940                 if (mddev->recovery_cp == MaxSector)
 941                         sb->state = (1<< MD_SB_CLEAN);
 942         } else
 943                 sb->recovery_cp = 0;
 944
 945         sb->layout = mddev->layout;
 946         sb->chunk_size = mddev->chunk_size;
 947
 948         if (mddev->bitmap && mddev->bitmap_file == NULL)
 949                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 950
 951         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 952         rdev_for_each(rdev2, tmp, mddev) {
 953                 mdp_disk_t *d;
 954                 int desc_nr;
 955                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 956                     && !test_bit(Faulty, &rdev2->flags))
 957                         desc_nr = rdev2->raid_disk;
 958                 else
 959                         desc_nr = next_spare++;
 960                 rdev2->desc_nr = desc_nr;
 961                 d = &sb->disks[rdev2->desc_nr];
 962                 nr_disks++;
 963                 d->number = rdev2->desc_nr;
 964                 d->major = MAJOR(rdev2->bdev->bd_dev);
 965                 d->minor = MINOR(rdev2->bdev->bd_dev);
 966                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 967                     && !test_bit(Faulty, &rdev2->flags))
 968                         d->raid_disk = rdev2->raid_disk;
 969                 else
 970                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 971                 if (test_bit(Faulty, &rdev2->flags))
 972                         d->state = (1<<MD_DISK_FAULTY);
 973                 else if (test_bit(In_sync, &rdev2->flags)) {
 974                         d->state = (1<<MD_DISK_ACTIVE);
 975                         d->state |= (1<<MD_DISK_SYNC);
 976                         active++;
 977                         working++;
 978                 } else {
 979                         d->state = 0;
 980                         spare++;
 981                         working++;
 982                 }
 983                 if (test_bit(WriteMostly, &rdev2->flags))
 984                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 985         }
 986         /* now set the "removed" and "faulty" bits on any missing devices */
 987         for (i=0 ; i < mddev->raid_disks ; i++) {
 988                 mdp_disk_t *d = &sb->disks[i];
 989                 if (d->state == 0 && d->number == 0) {
 990                         d->number = i;
 991                         d->raid_disk = i;
 992                         d->state = (1<<MD_DISK_REMOVED);
 993                         d->state |= (1<<MD_DISK_FAULTY);
 994                         failed++;
 995                 }
 996         }
 997         sb->nr_disks = nr_disks;
 998         sb->active_disks = active;
 999         sb->working_disks = working;
1000         sb->failed_disks = failed;
1001         sb->spare_disks = spare;
1002
1003         sb->this_disk = sb->disks[rdev->desc_nr];
1004         sb->sb_csum = calc_sb_csum(sb);
1005 }
1006
1007 /*
1008  * version 1 superblock
1009  */
1010
1011 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1012 {
1013         __le32 disk_csum;
1014         u32 csum;
1015         unsigned long long newcsum;
1016         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1017         __le32 *isuper = (__le32*)sb;
1018         int i;
1019
1020         disk_csum = sb->sb_csum;
1021         sb->sb_csum = 0;
1022         newcsum = 0;
1023         for (i=0; size>=4; size -= 4 )
1024                 newcsum += le32_to_cpu(*isuper++);
1025
1026         if (size == 2)
1027                 newcsum += le16_to_cpu(*(__le16*) isuper);
1028
1029         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1030         sb->sb_csum = disk_csum;
1031         return cpu_to_le32(csum);
1032 }
1033
1034 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1035 {
1036         struct mdp_superblock_1 *sb;
1037         int ret;
1038         sector_t sb_offset;
1039         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1040         int bmask;
1041
1042         /*
1043          * Calculate the position of the superblock.
1044          * It is always aligned to a 4K boundary and
1045          * depeding on minor_version, it can be:
1046          * 0: At least 8K, but less than 12K, from end of device
1047          * 1: At start of device
1048          * 2: 4K from start of device.
1049          */
1050         switch(minor_version) {
1051         case 0:
1052                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1053                 sb_offset -= 8*2;
1054                 sb_offset &= ~(sector_t)(4*2-1);
1055                 /* convert from sectors to K */
1056                 sb_offset /= 2;
1057                 break;
1058         case 1:
1059                 sb_offset = 0;
1060                 break;
1061         case 2:
1062                 sb_offset = 4;
1063                 break;
1064         default:
1065                 return -EINVAL;
1066         }
1067         rdev->sb_offset = sb_offset;
1068
1069         /* superblock is rarely larger than 1K, but it can be larger,
1070          * and it is safe to read 4k, so we do that
1071          */
1072         ret = read_disk_sb(rdev, 4096);
1073         if (ret) return ret;
1074
1075
1076         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1077
1078         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1079             sb->major_version != cpu_to_le32(1) ||
1080             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1081             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1082             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1083                 return -EINVAL;
1084
1085         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1086                 printk("md: invalid superblock checksum on %s\n",
1087                         bdevname(rdev->bdev,b));
1088                 return -EINVAL;
1089         }
1090         if (le64_to_cpu(sb->data_size) < 10) {
1091                 printk("md: data_size too small on %s\n",
1092                        bdevname(rdev->bdev,b));
1093                 return -EINVAL;
1094         }
1095         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1096                 if (sb->level != cpu_to_le32(1) &&
1097                     sb->level != cpu_to_le32(4) &&
1098                     sb->level != cpu_to_le32(5) &&
1099                     sb->level != cpu_to_le32(6) &&
1100                     sb->level != cpu_to_le32(10)) {
1101                         printk(KERN_WARNING
1102                                "md: bitmaps not supported for this level.\n");
1103                         return -EINVAL;
1104                 }
1105         }
1106
1107         rdev->preferred_minor = 0xffff;
1108         rdev->data_offset = le64_to_cpu(sb->data_offset);
1109         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1110
1111         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1112         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1113         if (rdev->sb_size & bmask)
1114                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1115
1116         if (minor_version
1117             && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1118                 return -EINVAL;
1119
1120         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1121                 rdev->desc_nr = -1;
1122         else
1123                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1124
1125         if (!refdev) {
1126                 ret = 1;
1127         } else {
1128                 __u64 ev1, ev2;
1129                 struct mdp_superblock_1 *refsb =
1130                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1131
1132                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1133                     sb->level != refsb->level ||
1134                     sb->layout != refsb->layout ||
1135                     sb->chunksize != refsb->chunksize) {
1136                         printk(KERN_WARNING "md: %s has strangely different"
1137                                 " superblock to %s\n",
1138                                 bdevname(rdev->bdev,b),
1139                                 bdevname(refdev->bdev,b2));
1140                         return -EINVAL;
1141                 }
1142                 ev1 = le64_to_cpu(sb->events);
1143                 ev2 = le64_to_cpu(refsb->events);
1144
1145                 if (ev1 > ev2)
1146                         ret = 1;
1147                 else
1148                         ret = 0;
1149         }
1150         if (minor_version)
1151                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1152         else
1153                 rdev->size = rdev->sb_offset;
1154         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1155                 return -EINVAL;
1156         rdev->size = le64_to_cpu(sb->data_size)/2;
1157         if (le32_to_cpu(sb->chunksize))
1158                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1159
1160         if (le64_to_cpu(sb->size) > rdev->size*2)
1161                 return -EINVAL;
1162         return ret;
1163 }
1164
1165 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1166 {
1167         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1168         __u64 ev1 = le64_to_cpu(sb->events);
1169
1170         rdev->raid_disk = -1;
1171         clear_bit(Faulty, &rdev->flags);
1172         clear_bit(In_sync, &rdev->flags);
1173         clear_bit(WriteMostly, &rdev->flags);
1174         clear_bit(BarriersNotsupp, &rdev->flags);
1175
1176         if (mddev->raid_disks == 0) {
1177                 mddev->major_version = 1;
1178                 mddev->patch_version = 0;
1179                 mddev->external = 0;
1180                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1181                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1182                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1183                 mddev->level = le32_to_cpu(sb->level);
1184                 mddev->clevel[0] = 0;
1185                 mddev->layout = le32_to_cpu(sb->layout);
1186                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1187                 mddev->size = le64_to_cpu(sb->size)/2;
1188                 mddev->events = ev1;
1189                 mddev->bitmap_offset = 0;
1190                 mddev->default_bitmap_offset = 1024 >> 9;
1191
1192                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1193                 memcpy(mddev->uuid, sb->set_uuid, 16);
1194
1195                 mddev->max_disks =  (4096-256)/2;
1196
1197                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1198                     mddev->bitmap_file == NULL )
1199                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1200
1201                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1202                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1203                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1204                         mddev->new_level = le32_to_cpu(sb->new_level);
1205                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1206                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1207                 } else {
1208                         mddev->reshape_position = MaxSector;
1209                         mddev->delta_disks = 0;
1210                         mddev->new_level = mddev->level;
1211                         mddev->new_layout = mddev->layout;
1212                         mddev->new_chunk = mddev->chunk_size;
1213                 }
1214
1215         } else if (mddev->pers == NULL) {
1216                 /* Insist of good event counter while assembling */
1217                 ++ev1;
1218                 if (ev1 < mddev->events)
1219                         return -EINVAL;
1220         } else if (mddev->bitmap) {
1221                 /* If adding to array with a bitmap, then we can accept an
1222                  * older device, but not too old.
1223                  */
1224                 if (ev1 < mddev->bitmap->events_cleared)
1225                         return 0;
1226         } else {
1227                 if (ev1 < mddev->events)
1228                         /* just a hot-add of a new device, leave raid_disk at -1 */
1229                         return 0;
1230         }
1231         if (mddev->level != LEVEL_MULTIPATH) {
1232                 int role;
1233                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1234                 switch(role) {
1235                 case 0xffff: /* spare */
1236                         break;
1237                 case 0xfffe: /* faulty */
1238                         set_bit(Faulty, &rdev->flags);
1239                         break;
1240                 default:
1241                         if ((le32_to_cpu(sb->feature_map) &
1242                              MD_FEATURE_RECOVERY_OFFSET))
1243                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1244                         else
1245                                 set_bit(In_sync, &rdev->flags);
1246                         rdev->raid_disk = role;
1247                         break;
1248                 }
1249                 if (sb->devflags & WriteMostly1)
1250                         set_bit(WriteMostly, &rdev->flags);
1251         } else /* MULTIPATH are always insync */
1252                 set_bit(In_sync, &rdev->flags);
1253
1254         return 0;
1255 }
1256
1257 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1258 {
1259         struct mdp_superblock_1 *sb;
1260         struct list_head *tmp;
1261         mdk_rdev_t *rdev2;
1262         int max_dev, i;
1263         /* make rdev->sb match mddev and rdev data. */
1264
1265         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1266
1267         sb->feature_map = 0;
1268         sb->pad0 = 0;
1269         sb->recovery_offset = cpu_to_le64(0);
1270         memset(sb->pad1, 0, sizeof(sb->pad1));
1271         memset(sb->pad2, 0, sizeof(sb->pad2));
1272         memset(sb->pad3, 0, sizeof(sb->pad3));
1273
1274         sb->utime = cpu_to_le64((__u64)mddev->utime);
1275         sb->events = cpu_to_le64(mddev->events);
1276         if (mddev->in_sync)
1277                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1278         else
1279                 sb->resync_offset = cpu_to_le64(0);
1280
1281         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1282
1283         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1284         sb->size = cpu_to_le64(mddev->size<<1);
1285
1286         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1287                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1288                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1289         }
1290
1291         if (rdev->raid_disk >= 0 &&
1292             !test_bit(In_sync, &rdev->flags) &&
1293             rdev->recovery_offset > 0) {
1294                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1295                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1296         }
1297
1298         if (mddev->reshape_position != MaxSector) {
1299                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1300                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1301                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1302                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1303                 sb->new_level = cpu_to_le32(mddev->new_level);
1304                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1305         }
1306
1307         max_dev = 0;
1308         rdev_for_each(rdev2, tmp, mddev)
1309                 if (rdev2->desc_nr+1 > max_dev)
1310                         max_dev = rdev2->desc_nr+1;
1311
1312         if (max_dev > le32_to_cpu(sb->max_dev))
1313                 sb->max_dev = cpu_to_le32(max_dev);
1314         for (i=0; i<max_dev;i++)
1315                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1316
1317         rdev_for_each(rdev2, tmp, mddev) {
1318                 i = rdev2->desc_nr;
1319                 if (test_bit(Faulty, &rdev2->flags))
1320                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1321                 else if (test_bit(In_sync, &rdev2->flags))
1322                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1323                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1324                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1325                 else
1326                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1327         }
1328
1329         sb->sb_csum = calc_sb_1_csum(sb);
1330 }
1331
1332
1333 static struct super_type super_types[] = {
1334         [0] = {
1335                 .name   = "0.90.0",
1336                 .owner  = THIS_MODULE,
1337                 .load_super     = super_90_load,
1338                 .validate_super = super_90_validate,
1339                 .sync_super     = super_90_sync,
1340         },
1341         [1] = {
1342                 .name   = "md-1",
1343                 .owner  = THIS_MODULE,
1344                 .load_super     = super_1_load,
1345                 .validate_super = super_1_validate,
1346                 .sync_super     = super_1_sync,
1347         },
1348 };
1349
1350 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1351 {
1352         struct list_head *tmp, *tmp2;
1353         mdk_rdev_t *rdev, *rdev2;
1354
1355         rdev_for_each(rdev, tmp, mddev1)
1356                 rdev_for_each(rdev2, tmp2, mddev2)
1357                         if (rdev->bdev->bd_contains ==
1358                             rdev2->bdev->bd_contains)
1359                                 return 1;
1360
1361         return 0;
1362 }
1363
1364 static LIST_HEAD(pending_raid_disks);
1365
1366 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1367 {
1368         char b[BDEVNAME_SIZE];
1369         struct kobject *ko;
1370         char *s;
1371         int err;
1372
1373         if (rdev->mddev) {
1374                 MD_BUG();
1375                 return -EINVAL;
1376         }
1377
1378         /* prevent duplicates */
1379         if (find_rdev(mddev, rdev->bdev->bd_dev))
1380                 return -EEXIST;
1381
1382         /* make sure rdev->size exceeds mddev->size */
1383         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1384                 if (mddev->pers) {
1385                         /* Cannot change size, so fail
1386                          * If mddev->level <= 0, then we don't care
1387                          * about aligning sizes (e.g. linear)
1388                          */
1389                         if (mddev->level > 0)
1390                                 return -ENOSPC;
1391                 } else
1392                         mddev->size = rdev->size;
1393         }
1394
1395         /* Verify rdev->desc_nr is unique.
1396          * If it is -1, assign a free number, else
1397          * check number is not in use
1398          */
1399         if (rdev->desc_nr < 0) {
1400                 int choice = 0;
1401                 if (mddev->pers) choice = mddev->raid_disks;
1402                 while (find_rdev_nr(mddev, choice))
1403                         choice++;
1404                 rdev->desc_nr = choice;
1405         } else {
1406                 if (find_rdev_nr(mddev, rdev->desc_nr))
1407                         return -EBUSY;
1408         }
1409         bdevname(rdev->bdev,b);
1410         while ( (s=strchr(b, '/')) != NULL)
1411                 *s = '!';
1412
1413         rdev->mddev = mddev;
1414         printk(KERN_INFO "md: bind<%s>\n", b);
1415
1416         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1417                 goto fail;
1418
1419         if (rdev->bdev->bd_part)
1420                 ko = &rdev->bdev->bd_part->dev.kobj;
1421         else
1422                 ko = &rdev->bdev->bd_disk->dev.kobj;
1423         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1424                 kobject_del(&rdev->kobj);
1425                 goto fail;
1426         }
1427         list_add(&rdev->same_set, &mddev->disks);
1428         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1429         return 0;
1430
1431  fail:
1432         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1433                b, mdname(mddev));
1434         return err;
1435 }
1436
1437 static void md_delayed_delete(struct work_struct *ws)
1438 {
1439         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1440         kobject_del(&rdev->kobj);
1441         kobject_put(&rdev->kobj);
1442 }
1443
1444 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1445 {
1446         char b[BDEVNAME_SIZE];
1447         if (!rdev->mddev) {
1448                 MD_BUG();
1449                 return;
1450         }
1451         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1452         list_del_init(&rdev->same_set);
1453         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1454         rdev->mddev = NULL;
1455         sysfs_remove_link(&rdev->kobj, "block");
1456
1457         /* We need to delay this, otherwise we can deadlock when
1458          * writing to 'remove' to "dev/state"
1459          */
1460         INIT_WORK(&rdev->del_work, md_delayed_delete);
1461         kobject_get(&rdev->kobj);
1462         schedule_work(&rdev->del_work);
1463 }
1464
1465 /*
1466  * prevent the device from being mounted, repartitioned or
1467  * otherwise reused by a RAID array (or any other kernel
1468  * subsystem), by bd_claiming the device.
1469  */
1470 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1471 {
1472         int err = 0;
1473         struct block_device *bdev;
1474         char b[BDEVNAME_SIZE];
1475
1476         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1477         if (IS_ERR(bdev)) {
1478                 printk(KERN_ERR "md: could not open %s.\n",
1479                         __bdevname(dev, b));
1480                 return PTR_ERR(bdev);
1481         }
1482         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1483         if (err) {
1484                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1485                         bdevname(bdev, b));
1486                 blkdev_put(bdev);
1487                 return err;
1488         }
1489         if (!shared)
1490                 set_bit(AllReserved, &rdev->flags);
1491         rdev->bdev = bdev;
1492         return err;
1493 }
1494
1495 static void unlock_rdev(mdk_rdev_t *rdev)
1496 {
1497         struct block_device *bdev = rdev->bdev;
1498         rdev->bdev = NULL;
1499         if (!bdev)
1500                 MD_BUG();
1501         bd_release(bdev);
1502         blkdev_put(bdev);
1503 }
1504
1505 void md_autodetect_dev(dev_t dev);
1506
1507 static void export_rdev(mdk_rdev_t * rdev)
1508 {
1509         char b[BDEVNAME_SIZE];
1510         printk(KERN_INFO "md: export_rdev(%s)\n",
1511                 bdevname(rdev->bdev,b));
1512         if (rdev->mddev)
1513                 MD_BUG();
1514         free_disk_sb(rdev);
1515         list_del_init(&rdev->same_set);
1516 #ifndef MODULE
1517         if (test_bit(AutoDetected, &rdev->flags))
1518                 md_autodetect_dev(rdev->bdev->bd_dev);
1519 #endif
1520         unlock_rdev(rdev);
1521         kobject_put(&rdev->kobj);
1522 }
1523
1524 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1525 {
1526         unbind_rdev_from_array(rdev);
1527         export_rdev(rdev);
1528 }
1529
1530 static void export_array(mddev_t *mddev)
1531 {
1532         struct list_head *tmp;
1533         mdk_rdev_t *rdev;
1534
1535         rdev_for_each(rdev, tmp, mddev) {
1536                 if (!rdev->mddev) {
1537                         MD_BUG();
1538                         continue;
1539                 }
1540                 kick_rdev_from_array(rdev);
1541         }
1542         if (!list_empty(&mddev->disks))
1543                 MD_BUG();
1544         mddev->raid_disks = 0;
1545         mddev->major_version = 0;
1546 }
1547
1548 static void print_desc(mdp_disk_t *desc)
1549 {
1550         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1551                 desc->major,desc->minor,desc->raid_disk,desc->state);
1552 }
1553
1554 static void print_sb(mdp_super_t *sb)
1555 {
1556         int i;
1557
1558         printk(KERN_INFO
1559                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1560                 sb->major_version, sb->minor_version, sb->patch_version,
1561                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1562                 sb->ctime);
1563         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1564                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1565                 sb->md_minor, sb->layout, sb->chunk_size);
1566         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1567                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1568                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1569                 sb->failed_disks, sb->spare_disks,
1570                 sb->sb_csum, (unsigned long)sb->events_lo);
1571
1572         printk(KERN_INFO);
1573         for (i = 0; i < MD_SB_DISKS; i++) {
1574                 mdp_disk_t *desc;
1575
1576                 desc = sb->disks + i;
1577                 if (desc->number || desc->major || desc->minor ||
1578                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1579                         printk("     D %2d: ", i);
1580                         print_desc(desc);
1581                 }
1582         }
1583         printk(KERN_INFO "md:     THIS: ");
1584         print_desc(&sb->this_disk);
1585
1586 }
1587
1588 static void print_rdev(mdk_rdev_t *rdev)
1589 {
1590         char b[BDEVNAME_SIZE];
1591         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1592                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1593                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1594                 rdev->desc_nr);
1595         if (rdev->sb_loaded) {
1596                 printk(KERN_INFO "md: rdev superblock:\n");
1597                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1598         } else
1599                 printk(KERN_INFO "md: no rdev superblock!\n");
1600 }
1601
1602 static void md_print_devices(void)
1603 {
1604         struct list_head *tmp, *tmp2;
1605         mdk_rdev_t *rdev;
1606         mddev_t *mddev;
1607         char b[BDEVNAME_SIZE];
1608
1609         printk("\n");
1610         printk("md:     **********************************\n");
1611         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1612         printk("md:     **********************************\n");
1613         for_each_mddev(mddev, tmp) {
1614
1615                 if (mddev->bitmap)
1616                         bitmap_print_sb(mddev->bitmap);
1617                 else
1618                         printk("%s: ", mdname(mddev));
1619                 rdev_for_each(rdev, tmp2, mddev)
1620                         printk("<%s>", bdevname(rdev->bdev,b));
1621                 printk("\n");
1622
1623                 rdev_for_each(rdev, tmp2, mddev)
1624                         print_rdev(rdev);
1625         }
1626         printk("md:     **********************************\n");
1627         printk("\n");
1628 }
1629
1630
1631 static void sync_sbs(mddev_t * mddev, int nospares)
1632 {
1633         /* Update each superblock (in-memory image), but
1634          * if we are allowed to, skip spares which already
1635          * have the right event counter, or have one earlier
1636          * (which would mean they aren't being marked as dirty
1637          * with the rest of the array)
1638          */
1639         mdk_rdev_t *rdev;
1640         struct list_head *tmp;
1641
1642         rdev_for_each(rdev, tmp, mddev) {
1643                 if (rdev->sb_events == mddev->events ||
1644                     (nospares &&
1645                      rdev->raid_disk < 0 &&
1646                      (rdev->sb_events&1)==0 &&
1647                      rdev->sb_events+1 == mddev->events)) {
1648                         /* Don't update this superblock */
1649                         rdev->sb_loaded = 2;
1650                 } else {
1651                         super_types[mddev->major_version].
1652                                 sync_super(mddev, rdev);
1653                         rdev->sb_loaded = 1;
1654                 }
1655         }
1656 }
1657
1658 static void md_update_sb(mddev_t * mddev, int force_change)
1659 {
1660         struct list_head *tmp;
1661         mdk_rdev_t *rdev;
1662         int sync_req;
1663         int nospares = 0;
1664
1665         if (mddev->external)
1666                 return;
1667 repeat:
1668         spin_lock_irq(&mddev->write_lock);
1669
1670         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1671         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1672                 force_change = 1;
1673         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1674                 /* just a clean<-> dirty transition, possibly leave spares alone,
1675                  * though if events isn't the right even/odd, we will have to do
1676                  * spares after all
1677                  */
1678                 nospares = 1;
1679         if (force_change)
1680                 nospares = 0;
1681         if (mddev->degraded)
1682                 /* If the array is degraded, then skipping spares is both
1683                  * dangerous and fairly pointless.
1684                  * Dangerous because a device that was removed from the array
1685                  * might have a event_count that still looks up-to-date,
1686                  * so it can be re-added without a resync.
1687                  * Pointless because if there are any spares to skip,
1688                  * then a recovery will happen and soon that array won't
1689                  * be degraded any more and the spare can go back to sleep then.
1690                  */
1691                 nospares = 0;
1692
1693         sync_req = mddev->in_sync;
1694         mddev->utime = get_seconds();
1695
1696         /* If this is just a dirty<->clean transition, and the array is clean
1697          * and 'events' is odd, we can roll back to the previous clean state */
1698         if (nospares
1699             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1700             && (mddev->events & 1)
1701             && mddev->events != 1)
1702                 mddev->events--;
1703         else {
1704                 /* otherwise we have to go forward and ... */
1705                 mddev->events ++;
1706                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1707                         /* .. if the array isn't clean, insist on an odd 'events' */
1708                         if ((mddev->events&1)==0) {
1709                                 mddev->events++;
1710                                 nospares = 0;
1711                         }
1712                 } else {
1713                         /* otherwise insist on an even 'events' (for clean states) */
1714                         if ((mddev->events&1)) {
1715                                 mddev->events++;
1716                                 nospares = 0;
1717                         }
1718                 }
1719         }
1720
1721         if (!mddev->events) {
1722                 /*
1723                  * oops, this 64-bit counter should never wrap.
1724                  * Either we are in around ~1 trillion A.C., assuming
1725                  * 1 reboot per second, or we have a bug:
1726                  */
1727                 MD_BUG();
1728                 mddev->events --;
1729         }
1730
1731         /*
1732          * do not write anything to disk if using
1733          * nonpersistent superblocks
1734          */
1735         if (!mddev->persistent) {
1736                 if (!mddev->external)
1737                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1738
1739                 spin_unlock_irq(&mddev->write_lock);
1740                 wake_up(&mddev->sb_wait);
1741                 return;
1742         }
1743         sync_sbs(mddev, nospares);
1744         spin_unlock_irq(&mddev->write_lock);
1745
1746         dprintk(KERN_INFO
1747                 "md: updating %s RAID superblock on device (in sync %d)\n",
1748                 mdname(mddev),mddev->in_sync);
1749
1750         bitmap_update_sb(mddev->bitmap);
1751         rdev_for_each(rdev, tmp, mddev) {
1752                 char b[BDEVNAME_SIZE];
1753                 dprintk(KERN_INFO "md: ");
1754                 if (rdev->sb_loaded != 1)
1755                         continue; /* no noise on spare devices */
1756                 if (test_bit(Faulty, &rdev->flags))
1757                         dprintk("(skipping faulty ");
1758
1759                 dprintk("%s ", bdevname(rdev->bdev,b));
1760                 if (!test_bit(Faulty, &rdev->flags)) {
1761                         md_super_write(mddev,rdev,
1762                                        rdev->sb_offset<<1, rdev->sb_size,
1763                                        rdev->sb_page);
1764                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1765                                 bdevname(rdev->bdev,b),
1766                                 (unsigned long long)rdev->sb_offset);
1767                         rdev->sb_events = mddev->events;
1768
1769                 } else
1770                         dprintk(")\n");
1771                 if (mddev->level == LEVEL_MULTIPATH)
1772                         /* only need to write one superblock... */
1773                         break;
1774         }
1775         md_super_wait(mddev);
1776         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1777
1778         spin_lock_irq(&mddev->write_lock);
1779         if (mddev->in_sync != sync_req ||
1780             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1781                 /* have to write it out again */
1782                 spin_unlock_irq(&mddev->write_lock);
1783                 goto repeat;
1784         }
1785         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1786         spin_unlock_irq(&mddev->write_lock);
1787         wake_up(&mddev->sb_wait);
1788
1789 }
1790
1791 /* words written to sysfs files may, or my not, be \n terminated.
1792  * We want to accept with case. For this we use cmd_match.
1793  */
1794 static int cmd_match(const char *cmd, const char *str)
1795 {
1796         /* See if cmd, written into a sysfs file, matches
1797          * str.  They must either be the same, or cmd can
1798          * have a trailing newline
1799          */
1800         while (*cmd && *str && *cmd == *str) {
1801                 cmd++;
1802                 str++;
1803         }
1804         if (*cmd == '\n')
1805                 cmd++;
1806         if (*str || *cmd)
1807                 return 0;
1808         return 1;
1809 }
1810
1811 struct rdev_sysfs_entry {
1812         struct attribute attr;
1813         ssize_t (*show)(mdk_rdev_t *, char *);
1814         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1815 };
1816
1817 static ssize_t
1818 state_show(mdk_rdev_t *rdev, char *page)
1819 {
1820         char *sep = "";
1821         size_t len = 0;
1822
1823         if (test_bit(Faulty, &rdev->flags)) {
1824                 len+= sprintf(page+len, "%sfaulty",sep);
1825                 sep = ",";
1826         }
1827         if (test_bit(In_sync, &rdev->flags)) {
1828                 len += sprintf(page+len, "%sin_sync",sep);
1829                 sep = ",";
1830         }
1831         if (test_bit(WriteMostly, &rdev->flags)) {
1832                 len += sprintf(page+len, "%swrite_mostly",sep);
1833                 sep = ",";
1834         }
1835         if (test_bit(Blocked, &rdev->flags)) {
1836                 len += sprintf(page+len, "%sblocked", sep);
1837                 sep = ",";
1838         }
1839         if (!test_bit(Faulty, &rdev->flags) &&
1840             !test_bit(In_sync, &rdev->flags)) {
1841                 len += sprintf(page+len, "%sspare", sep);
1842                 sep = ",";
1843         }
1844         return len+sprintf(page+len, "\n");
1845 }
1846
1847 static ssize_t
1848 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1849 {
1850         /* can write
1851          *  faulty  - simulates and error
1852          *  remove  - disconnects the device
1853          *  writemostly - sets write_mostly
1854          *  -writemostly - clears write_mostly
1855          *  blocked - sets the Blocked flag
1856          *  -blocked - clears the Blocked flag
1857          */
1858         int err = -EINVAL;
1859         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1860                 md_error(rdev->mddev, rdev);
1861                 err = 0;
1862         } else if (cmd_match(buf, "remove")) {
1863                 if (rdev->raid_disk >= 0)
1864                         err = -EBUSY;
1865                 else {
1866                         mddev_t *mddev = rdev->mddev;
1867                         kick_rdev_from_array(rdev);
1868                         if (mddev->pers)
1869                                 md_update_sb(mddev, 1);
1870                         md_new_event(mddev);
1871                         err = 0;
1872                 }
1873         } else if (cmd_match(buf, "writemostly")) {
1874                 set_bit(WriteMostly, &rdev->flags);
1875                 err = 0;
1876         } else if (cmd_match(buf, "-writemostly")) {
1877                 clear_bit(WriteMostly, &rdev->flags);
1878                 err = 0;
1879         } else if (cmd_match(buf, "blocked")) {
1880                 set_bit(Blocked, &rdev->flags);
1881                 err = 0;
1882         } else if (cmd_match(buf, "-blocked")) {
1883                 clear_bit(Blocked, &rdev->flags);
1884                 wake_up(&rdev->blocked_wait);
1885                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1886                 md_wakeup_thread(rdev->mddev->thread);
1887
1888                 err = 0;
1889         }
1890         return err ? err : len;
1891 }
1892 static struct rdev_sysfs_entry rdev_state =
1893 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1894
1895 static ssize_t
1896 errors_show(mdk_rdev_t *rdev, char *page)
1897 {
1898         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1899 }
1900
1901 static ssize_t
1902 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1903 {
1904         char *e;
1905         unsigned long n = simple_strtoul(buf, &e, 10);
1906         if (*buf && (*e == 0 || *e == '\n')) {
1907                 atomic_set(&rdev->corrected_errors, n);
1908                 return len;
1909         }
1910         return -EINVAL;
1911 }
1912 static struct rdev_sysfs_entry rdev_errors =
1913 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1914
1915 static ssize_t
1916 slot_show(mdk_rdev_t *rdev, char *page)
1917 {
1918         if (rdev->raid_disk < 0)
1919                 return sprintf(page, "none\n");
1920         else
1921                 return sprintf(page, "%d\n", rdev->raid_disk);
1922 }
1923
1924 static ssize_t
1925 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1926 {
1927         char *e;
1928         int err;
1929         char nm[20];
1930         int slot = simple_strtoul(buf, &e, 10);
1931         if (strncmp(buf, "none", 4)==0)
1932                 slot = -1;
1933         else if (e==buf || (*e && *e!= '\n'))
1934                 return -EINVAL;
1935         if (rdev->mddev->pers && slot == -1) {
1936                 /* Setting 'slot' on an active array requires also
1937                  * updating the 'rd%d' link, and communicating
1938                  * with the personality with ->hot_*_disk.
1939                  * For now we only support removing
1940                  * failed/spare devices.  This normally happens automatically,
1941                  * but not when the metadata is externally managed.
1942                  */
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 if (rdev->mddev->pers) {
1957                 mdk_rdev_t *rdev2;
1958                 struct list_head *tmp;
1959                 /* Activating a spare .. or possibly reactivating
1960                  * if we every get bitmaps working here.
1961                  */
1962
1963                 if (rdev->raid_disk != -1)
1964                         return -EBUSY;
1965
1966                 if (rdev->mddev->pers->hot_add_disk == NULL)
1967                         return -EINVAL;
1968
1969                 rdev_for_each(rdev2, tmp, rdev->mddev)
1970                         if (rdev2->raid_disk == slot)
1971                                 return -EEXIST;
1972
1973                 rdev->raid_disk = slot;
1974                 if (test_bit(In_sync, &rdev->flags))
1975                         rdev->saved_raid_disk = slot;
1976                 else
1977                         rdev->saved_raid_disk = -1;
1978                 err = rdev->mddev->pers->
1979                         hot_add_disk(rdev->mddev, rdev);
1980                 if (err) {
1981                         rdev->raid_disk = -1;
1982                         return err;
1983                 }
1984                 sprintf(nm, "rd%d", rdev->raid_disk);
1985                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
1986                         printk(KERN_WARNING
1987                                "md: cannot register "
1988                                "%s for %s\n",
1989                                nm, mdname(rdev->mddev));
1990
1991                 /* don't wakeup anyone, leave that to userspace. */
1992         } else {
1993                 if (slot >= rdev->mddev->raid_disks)
1994                         return -ENOSPC;
1995                 rdev->raid_disk = slot;
1996                 /* assume it is working */
1997                 clear_bit(Faulty, &rdev->flags);
1998                 clear_bit(WriteMostly, &rdev->flags);
1999                 set_bit(In_sync, &rdev->flags);
2000         }
2001         return len;
2002 }
2003
2004
2005 static struct rdev_sysfs_entry rdev_slot =
2006 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2007
2008 static ssize_t
2009 offset_show(mdk_rdev_t *rdev, char *page)
2010 {
2011         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2012 }
2013
2014 static ssize_t
2015 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2016 {
2017         char *e;
2018         unsigned long long offset = simple_strtoull(buf, &e, 10);
2019         if (e==buf || (*e && *e != '\n'))
2020                 return -EINVAL;
2021         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2022                 return -EBUSY;
2023         if (rdev->size && rdev->mddev->external)
2024                 /* Must set offset before size, so overlap checks
2025                  * can be sane */
2026                 return -EBUSY;
2027         rdev->data_offset = offset;
2028         return len;
2029 }
2030
2031 static struct rdev_sysfs_entry rdev_offset =
2032 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2033
2034 static ssize_t
2035 rdev_size_show(mdk_rdev_t *rdev, char *page)
2036 {
2037         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2038 }
2039
2040 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2041 {
2042         /* check if two start/length pairs overlap */
2043         if (s1+l1 <= s2)
2044                 return 0;
2045         if (s2+l2 <= s1)
2046                 return 0;
2047         return 1;
2048 }
2049
2050 static ssize_t
2051 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2052 {
2053         char *e;
2054         unsigned long long size = simple_strtoull(buf, &e, 10);
2055         unsigned long long oldsize = rdev->size;
2056         mddev_t *my_mddev = rdev->mddev;
2057
2058         if (e==buf || (*e && *e != '\n'))
2059                 return -EINVAL;
2060         if (my_mddev->pers && rdev->raid_disk >= 0)
2061                 return -EBUSY;
2062         rdev->size = size;
2063         if (size > oldsize && rdev->mddev->external) {
2064                 /* need to check that all other rdevs with the same ->bdev
2065                  * do not overlap.  We need to unlock the mddev to avoid
2066                  * a deadlock.  We have already changed rdev->size, and if
2067                  * we have to change it back, we will have the lock again.
2068                  */
2069                 mddev_t *mddev;
2070                 int overlap = 0;
2071                 struct list_head *tmp, *tmp2;
2072
2073                 mddev_unlock(my_mddev);
2074                 for_each_mddev(mddev, tmp) {
2075                         mdk_rdev_t *rdev2;
2076
2077                         mddev_lock(mddev);
2078                         rdev_for_each(rdev2, tmp2, mddev)
2079                                 if (test_bit(AllReserved, &rdev2->flags) ||
2080                                     (rdev->bdev == rdev2->bdev &&
2081                                      rdev != rdev2 &&
2082                                      overlaps(rdev->data_offset, rdev->size,
2083                                             rdev2->data_offset, rdev2->size))) {
2084                                         overlap = 1;
2085                                         break;
2086                                 }
2087                         mddev_unlock(mddev);
2088                         if (overlap) {
2089                                 mddev_put(mddev);
2090                                 break;
2091                         }
2092                 }
2093                 mddev_lock(my_mddev);
2094                 if (overlap) {
2095                         /* Someone else could have slipped in a size
2096                          * change here, but doing so is just silly.
2097                          * We put oldsize back because we *know* it is
2098                          * safe, and trust userspace not to race with
2099                          * itself
2100                          */
2101                         rdev->size = oldsize;
2102                         return -EBUSY;
2103                 }
2104         }
2105         if (size < my_mddev->size || my_mddev->size == 0)
2106                 my_mddev->size = size;
2107         return len;
2108 }
2109
2110 static struct rdev_sysfs_entry rdev_size =
2111 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2112
2113 static struct attribute *rdev_default_attrs[] = {
2114         &rdev_state.attr,
2115         &rdev_errors.attr,
2116         &rdev_slot.attr,
2117         &rdev_offset.attr,
2118         &rdev_size.attr,
2119         NULL,
2120 };
2121 static ssize_t
2122 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2123 {
2124         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2125         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2126         mddev_t *mddev = rdev->mddev;
2127         ssize_t rv;
2128
2129         if (!entry->show)
2130                 return -EIO;
2131
2132         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2133         if (!rv) {
2134                 if (rdev->mddev == NULL)
2135                         rv = -EBUSY;
2136                 else
2137                         rv = entry->show(rdev, page);
2138                 mddev_unlock(mddev);
2139         }
2140         return rv;
2141 }
2142
2143 static ssize_t
2144 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2145               const char *page, size_t length)
2146 {
2147         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2148         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2149         ssize_t rv;
2150         mddev_t *mddev = rdev->mddev;
2151
2152         if (!entry->store)
2153                 return -EIO;
2154         if (!capable(CAP_SYS_ADMIN))
2155                 return -EACCES;
2156         rv = mddev ? mddev_lock(mddev): -EBUSY;
2157         if (!rv) {
2158                 if (rdev->mddev == NULL)
2159                         rv = -EBUSY;
2160                 else
2161                         rv = entry->store(rdev, page, length);
2162                 mddev_unlock(mddev);
2163         }
2164         return rv;
2165 }
2166
2167 static void rdev_free(struct kobject *ko)
2168 {
2169         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2170         kfree(rdev);
2171 }
2172 static struct sysfs_ops rdev_sysfs_ops = {
2173         .show           = rdev_attr_show,
2174         .store          = rdev_attr_store,
2175 };
2176 static struct kobj_type rdev_ktype = {
2177         .release        = rdev_free,
2178         .sysfs_ops      = &rdev_sysfs_ops,
2179         .default_attrs  = rdev_default_attrs,
2180 };
2181
2182 /*
2183  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2184  *
2185  * mark the device faulty if:
2186  *
2187  *   - the device is nonexistent (zero size)
2188  *   - the device has no valid superblock
2189  *
2190  * a faulty rdev _never_ has rdev->sb set.
2191  */
2192 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2193 {
2194         char b[BDEVNAME_SIZE];
2195         int err;
2196         mdk_rdev_t *rdev;
2197         sector_t size;
2198
2199         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2200         if (!rdev) {
2201                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2202                 return ERR_PTR(-ENOMEM);
2203         }
2204
2205         if ((err = alloc_disk_sb(rdev)))
2206                 goto abort_free;
2207
2208         err = lock_rdev(rdev, newdev, super_format == -2);
2209         if (err)
2210                 goto abort_free;
2211
2212         kobject_init(&rdev->kobj, &rdev_ktype);
2213
2214         rdev->desc_nr = -1;
2215         rdev->saved_raid_disk = -1;
2216         rdev->raid_disk = -1;
2217         rdev->flags = 0;
2218         rdev->data_offset = 0;
2219         rdev->sb_events = 0;
2220         atomic_set(&rdev->nr_pending, 0);
2221         atomic_set(&rdev->read_errors, 0);
2222         atomic_set(&rdev->corrected_errors, 0);
2223
2224         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2225         if (!size) {
2226                 printk(KERN_WARNING
2227                         "md: %s has zero or unknown size, marking faulty!\n",
2228                         bdevname(rdev->bdev,b));
2229                 err = -EINVAL;
2230                 goto abort_free;
2231         }
2232
2233         if (super_format >= 0) {
2234                 err = super_types[super_format].
2235                         load_super(rdev, NULL, super_minor);
2236                 if (err == -EINVAL) {
2237                         printk(KERN_WARNING
2238                                 "md: %s does not have a valid v%d.%d "
2239                                "superblock, not importing!\n",
2240                                 bdevname(rdev->bdev,b),
2241                                super_format, super_minor);
2242                         goto abort_free;
2243                 }
2244                 if (err < 0) {
2245                         printk(KERN_WARNING
2246                                 "md: could not read %s's sb, not importing!\n",
2247                                 bdevname(rdev->bdev,b));
2248                         goto abort_free;
2249                 }
2250         }
2251
2252         INIT_LIST_HEAD(&rdev->same_set);
2253         init_waitqueue_head(&rdev->blocked_wait);
2254
2255         return rdev;
2256
2257 abort_free:
2258         if (rdev->sb_page) {
2259                 if (rdev->bdev)
2260                         unlock_rdev(rdev);
2261                 free_disk_sb(rdev);
2262         }
2263         kfree(rdev);
2264         return ERR_PTR(err);
2265 }
2266
2267 /*
2268  * Check a full RAID array for plausibility
2269  */
2270
2271
2272 static void analyze_sbs(mddev_t * mddev)
2273 {
2274         int i;
2275         struct list_head *tmp;
2276         mdk_rdev_t *rdev, *freshest;
2277         char b[BDEVNAME_SIZE];
2278
2279         freshest = NULL;
2280         rdev_for_each(rdev, tmp, mddev)
2281                 switch (super_types[mddev->major_version].
2282                         load_super(rdev, freshest, mddev->minor_version)) {
2283                 case 1:
2284                         freshest = rdev;
2285                         break;
2286                 case 0:
2287                         break;
2288                 default:
2289                         printk( KERN_ERR \
2290                                 "md: fatal superblock inconsistency in %s"
2291                                 " -- removing from array\n",
2292                                 bdevname(rdev->bdev,b));
2293                         kick_rdev_from_array(rdev);
2294                 }
2295
2296
2297         super_types[mddev->major_version].
2298                 validate_super(mddev, freshest);
2299
2300         i = 0;
2301         rdev_for_each(rdev, tmp, mddev) {
2302                 if (rdev != freshest)
2303                         if (super_types[mddev->major_version].
2304                             validate_super(mddev, rdev)) {
2305                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2306                                         " from array!\n",
2307                                         bdevname(rdev->bdev,b));
2308                                 kick_rdev_from_array(rdev);
2309                                 continue;
2310                         }
2311                 if (mddev->level == LEVEL_MULTIPATH) {
2312                         rdev->desc_nr = i++;
2313                         rdev->raid_disk = rdev->desc_nr;
2314                         set_bit(In_sync, &rdev->flags);
2315                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2316                         rdev->raid_disk = -1;
2317                         clear_bit(In_sync, &rdev->flags);
2318                 }
2319         }
2320
2321
2322
2323         if (mddev->recovery_cp != MaxSector &&
2324             mddev->level >= 1)
2325                 printk(KERN_ERR "md: %s: raid array is not clean"
2326                        " -- starting background reconstruction\n",
2327                        mdname(mddev));
2328
2329 }
2330
2331 static ssize_t
2332 safe_delay_show(mddev_t *mddev, char *page)
2333 {
2334         int msec = (mddev->safemode_delay*1000)/HZ;
2335         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2336 }
2337 static ssize_t
2338 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2339 {
2340         int scale=1;
2341         int dot=0;
2342         int i;
2343         unsigned long msec;
2344         char buf[30];
2345         char *e;
2346         /* remove a period, and count digits after it */
2347         if (len >= sizeof(buf))
2348                 return -EINVAL;
2349         strlcpy(buf, cbuf, len);
2350         buf[len] = 0;
2351         for (i=0; i<len; i++) {
2352                 if (dot) {
2353                         if (isdigit(buf[i])) {
2354                                 buf[i-1] = buf[i];
2355                                 scale *= 10;
2356                         }
2357                         buf[i] = 0;
2358                 } else if (buf[i] == '.') {
2359                         dot=1;
2360                         buf[i] = 0;
2361                 }
2362         }
2363         msec = simple_strtoul(buf, &e, 10);
2364         if (e == buf || (*e && *e != '\n'))
2365                 return -EINVAL;
2366         msec = (msec * 1000) / scale;
2367         if (msec == 0)
2368                 mddev->safemode_delay = 0;
2369         else {
2370                 mddev->safemode_delay = (msec*HZ)/1000;
2371                 if (mddev->safemode_delay == 0)
2372                         mddev->safemode_delay = 1;
2373         }
2374         return len;
2375 }
2376 static struct md_sysfs_entry md_safe_delay =
2377 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2378
2379 static ssize_t
2380 level_show(mddev_t *mddev, char *page)
2381 {
2382         struct mdk_personality *p = mddev->pers;
2383         if (p)
2384                 return sprintf(page, "%s\n", p->name);
2385         else if (mddev->clevel[0])
2386                 return sprintf(page, "%s\n", mddev->clevel);
2387         else if (mddev->level != LEVEL_NONE)
2388                 return sprintf(page, "%d\n", mddev->level);
2389         else
2390                 return 0;
2391 }
2392
2393 static ssize_t
2394 level_store(mddev_t *mddev, const char *buf, size_t len)
2395 {
2396         ssize_t rv = len;
2397         if (mddev->pers)
2398                 return -EBUSY;
2399         if (len == 0)
2400                 return 0;
2401         if (len >= sizeof(mddev->clevel))
2402                 return -ENOSPC;
2403         strncpy(mddev->clevel, buf, len);
2404         if (mddev->clevel[len-1] == '\n')
2405                 len--;
2406         mddev->clevel[len] = 0;
2407         mddev->level = LEVEL_NONE;
2408         return rv;
2409 }
2410
2411 static struct md_sysfs_entry md_level =
2412 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2413
2414
2415 static ssize_t
2416 layout_show(mddev_t *mddev, char *page)
2417 {
2418         /* just a number, not meaningful for all levels */
2419         if (mddev->reshape_position != MaxSector &&
2420             mddev->layout != mddev->new_layout)
2421                 return sprintf(page, "%d (%d)\n",
2422                                mddev->new_layout, mddev->layout);
2423         return sprintf(page, "%d\n", mddev->layout);
2424 }
2425
2426 static ssize_t
2427 layout_store(mddev_t *mddev, const char *buf, size_t len)
2428 {
2429         char *e;
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                 return -EBUSY;
2437         if (mddev->reshape_position != MaxSector)
2438                 mddev->new_layout = n;
2439         else
2440                 mddev->layout = n;
2441         return len;
2442 }
2443 static struct md_sysfs_entry md_layout =
2444 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2445
2446
2447 static ssize_t
2448 raid_disks_show(mddev_t *mddev, char *page)
2449 {
2450         if (mddev->raid_disks == 0)
2451                 return 0;
2452         if (mddev->reshape_position != MaxSector &&
2453             mddev->delta_disks != 0)
2454                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2455                                mddev->raid_disks - mddev->delta_disks);
2456         return sprintf(page, "%d\n", mddev->raid_disks);
2457 }
2458
2459 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2460
2461 static ssize_t
2462 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2463 {
2464         char *e;
2465         int rv = 0;
2466         unsigned long n = simple_strtoul(buf, &e, 10);
2467
2468         if (!*buf || (*e && *e != '\n'))
2469                 return -EINVAL;
2470
2471         if (mddev->pers)
2472                 rv = update_raid_disks(mddev, n);
2473         else if (mddev->reshape_position != MaxSector) {
2474                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2475                 mddev->delta_disks = n - olddisks;
2476                 mddev->raid_disks = n;
2477         } else
2478                 mddev->raid_disks = n;
2479         return rv ? rv : len;
2480 }
2481 static struct md_sysfs_entry md_raid_disks =
2482 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2483
2484 static ssize_t
2485 chunk_size_show(mddev_t *mddev, char *page)
2486 {
2487         if (mddev->reshape_position != MaxSector &&
2488             mddev->chunk_size != mddev->new_chunk)
2489                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2490                                mddev->chunk_size);
2491         return sprintf(page, "%d\n", mddev->chunk_size);
2492 }
2493
2494 static ssize_t
2495 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2496 {
2497         /* can only set chunk_size if array is not yet active */
2498         char *e;
2499         unsigned long n = simple_strtoul(buf, &e, 10);
2500
2501         if (!*buf || (*e && *e != '\n'))
2502                 return -EINVAL;
2503
2504         if (mddev->pers)
2505                 return -EBUSY;
2506         else if (mddev->reshape_position != MaxSector)
2507                 mddev->new_chunk = n;
2508         else
2509                 mddev->chunk_size = n;
2510         return len;
2511 }
2512 static struct md_sysfs_entry md_chunk_size =
2513 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2514
2515 static ssize_t
2516 resync_start_show(mddev_t *mddev, char *page)
2517 {
2518         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2519 }
2520
2521 static ssize_t
2522 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2523 {
2524         char *e;
2525         unsigned long long n = simple_strtoull(buf, &e, 10);
2526
2527         if (mddev->pers)
2528                 return -EBUSY;
2529         if (!*buf || (*e && *e != '\n'))
2530                 return -EINVAL;
2531
2532         mddev->recovery_cp = n;
2533         return len;
2534 }
2535 static struct md_sysfs_entry md_resync_start =
2536 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2537
2538 /*
2539  * The array state can be:
2540  *
2541  * clear
2542  *     No devices, no size, no level
2543  *     Equivalent to STOP_ARRAY ioctl
2544  * inactive
2545  *     May have some settings, but array is not active
2546  *        all IO results in error
2547  *     When written, doesn't tear down array, but just stops it
2548  * suspended (not supported yet)
2549  *     All IO requests will block. The array can be reconfigured.
2550  *     Writing this, if accepted, will block until array is quiessent
2551  * readonly
2552  *     no resync can happen.  no superblocks get written.
2553  *     write requests fail
2554  * read-auto
2555  *     like readonly, but behaves like 'clean' on a write request.
2556  *
2557  * clean - no pending writes, but otherwise active.
2558  *     When written to inactive array, starts without resync
2559  *     If a write request arrives then
2560  *       if metadata is known, mark 'dirty' and switch to 'active'.
2561  *       if not known, block and switch to write-pending
2562  *     If written to an active array that has pending writes, then fails.
2563  * active
2564  *     fully active: IO and resync can be happening.
2565  *     When written to inactive array, starts with resync
2566  *
2567  * write-pending
2568  *     clean, but writes are blocked waiting for 'active' to be written.
2569  *
2570  * active-idle
2571  *     like active, but no writes have been seen for a while (100msec).
2572  *
2573  */
2574 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2575                    write_pending, active_idle, bad_word};
2576 static char *array_states[] = {
2577         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2578         "write-pending", "active-idle", NULL };
2579
2580 static int match_word(const char *word, char **list)
2581 {
2582         int n;
2583         for (n=0; list[n]; n++)
2584                 if (cmd_match(word, list[n]))
2585                         break;
2586         return n;
2587 }
2588
2589 static ssize_t
2590 array_state_show(mddev_t *mddev, char *page)
2591 {
2592         enum array_state st = inactive;
2593
2594         if (mddev->pers)
2595                 switch(mddev->ro) {
2596                 case 1:
2597                         st = readonly;
2598                         break;
2599                 case 2:
2600                         st = read_auto;
2601                         break;
2602                 case 0:
2603                         if (mddev->in_sync)
2604                                 st = clean;
2605                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2606                                 st = write_pending;
2607                         else if (mddev->safemode)
2608                                 st = active_idle;
2609                         else
2610                                 st = active;
2611                 }
2612         else {
2613                 if (list_empty(&mddev->disks) &&
2614                     mddev->raid_disks == 0 &&
2615                     mddev->size == 0)
2616                         st = clear;
2617                 else
2618                         st = inactive;
2619         }
2620         return sprintf(page, "%s\n", array_states[st]);
2621 }
2622
2623 static int do_md_stop(mddev_t * mddev, int ro);
2624 static int do_md_run(mddev_t * mddev);
2625 static int restart_array(mddev_t *mddev);
2626
2627 static ssize_t
2628 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2629 {
2630         int err = -EINVAL;
2631         enum array_state st = match_word(buf, array_states);
2632         switch(st) {
2633         case bad_word:
2634                 break;
2635         case clear:
2636                 /* stopping an active array */
2637                 if (atomic_read(&mddev->active) > 1)
2638                         return -EBUSY;
2639                 err = do_md_stop(mddev, 0);
2640                 break;
2641         case inactive:
2642                 /* stopping an active array */
2643                 if (mddev->pers) {
2644                         if (atomic_read(&mddev->active) > 1)
2645                                 return -EBUSY;
2646                         err = do_md_stop(mddev, 2);
2647                 } else
2648                         err = 0; /* already inactive */
2649                 break;
2650         case suspended:
2651                 break; /* not supported yet */
2652         case readonly:
2653                 if (mddev->pers)
2654                         err = do_md_stop(mddev, 1);
2655                 else {
2656                         mddev->ro = 1;
2657                         set_disk_ro(mddev->gendisk, 1);
2658                         err = do_md_run(mddev);
2659                 }
2660                 break;
2661         case read_auto:
2662                 if (mddev->pers) {
2663                         if (mddev->ro != 1)
2664                                 err = do_md_stop(mddev, 1);
2665                         else
2666                                 err = restart_array(mddev);
2667                         if (err == 0) {
2668                                 mddev->ro = 2;
2669                                 set_disk_ro(mddev->gendisk, 0);
2670                         }
2671                 } else {
2672                         mddev->ro = 2;
2673                         err = do_md_run(mddev);
2674                 }
2675                 break;
2676         case clean:
2677                 if (mddev->pers) {
2678                         restart_array(mddev);
2679                         spin_lock_irq(&mddev->write_lock);
2680                         if (atomic_read(&mddev->writes_pending) == 0) {
2681                                 if (mddev->in_sync == 0) {
2682                                         mddev->in_sync = 1;
2683                                         if (mddev->safemode == 1)
2684                                                 mddev->safemode = 0;
2685                                         if (mddev->persistent)
2686                                                 set_bit(MD_CHANGE_CLEAN,
2687                                                         &mddev->flags);
2688                                 }
2689                                 err = 0;
2690                         } else
2691                                 err = -EBUSY;
2692                         spin_unlock_irq(&mddev->write_lock);
2693                 } else {
2694                         mddev->ro = 0;
2695                         mddev->recovery_cp = MaxSector;
2696                         err = do_md_run(mddev);
2697                 }
2698                 break;
2699         case active:
2700                 if (mddev->pers) {
2701                         restart_array(mddev);
2702                         if (mddev->external)
2703                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2704                         wake_up(&mddev->sb_wait);
2705                         err = 0;
2706                 } else {
2707                         mddev->ro = 0;
2708                         set_disk_ro(mddev->gendisk, 0);
2709                         err = do_md_run(mddev);
2710                 }
2711                 break;
2712         case write_pending:
2713         case active_idle:
2714                 /* these cannot be set */
2715                 break;
2716         }
2717         if (err)
2718                 return err;
2719         else {
2720                 sysfs_notify(&mddev->kobj, NULL, "array_state");
2721                 return len;
2722         }
2723 }
2724 static struct md_sysfs_entry md_array_state =
2725 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2726
2727 static ssize_t
2728 null_show(mddev_t *mddev, char *page)
2729 {
2730         return -EINVAL;
2731 }
2732
2733 static ssize_t
2734 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2735 {
2736         /* buf must be %d:%d\n? giving major and minor numbers */
2737         /* The new device is added to the array.
2738          * If the array has a persistent superblock, we read the
2739          * superblock to initialise info and check validity.
2740          * Otherwise, only checking done is that in bind_rdev_to_array,
2741          * which mainly checks size.
2742          */
2743         char *e;
2744         int major = simple_strtoul(buf, &e, 10);
2745         int minor;
2746         dev_t dev;
2747         mdk_rdev_t *rdev;
2748         int err;
2749
2750         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2751                 return -EINVAL;
2752         minor = simple_strtoul(e+1, &e, 10);
2753         if (*e && *e != '\n')
2754                 return -EINVAL;
2755         dev = MKDEV(major, minor);
2756         if (major != MAJOR(dev) ||
2757             minor != MINOR(dev))
2758                 return -EOVERFLOW;
2759
2760
2761         if (mddev->persistent) {
2762                 rdev = md_import_device(dev, mddev->major_version,
2763                                         mddev->minor_version);
2764                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2765                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2766                                                        mdk_rdev_t, same_set);
2767                         err = super_types[mddev->major_version]
2768                                 .load_super(rdev, rdev0, mddev->minor_version);
2769                         if (err < 0)
2770                                 goto out;
2771                 }
2772         } else if (mddev->external)
2773                 rdev = md_import_device(dev, -2, -1);
2774         else
2775                 rdev = md_import_device(dev, -1, -1);
2776
2777         if (IS_ERR(rdev))
2778                 return PTR_ERR(rdev);
2779         err = bind_rdev_to_array(rdev, mddev);
2780  out:
2781         if (err)
2782                 export_rdev(rdev);
2783         return err ? err : len;
2784 }
2785
2786 static struct md_sysfs_entry md_new_device =
2787 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2788
2789 static ssize_t
2790 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2791 {
2792         char *end;
2793         unsigned long chunk, end_chunk;
2794
2795         if (!mddev->bitmap)
2796                 goto out;
2797         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2798         while (*buf) {
2799                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2800                 if (buf == end) break;
2801                 if (*end == '-') { /* range */
2802                         buf = end + 1;
2803                         end_chunk = simple_strtoul(buf, &end, 0);
2804                         if (buf == end) break;
2805                 }
2806                 if (*end && !isspace(*end)) break;
2807                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2808                 buf = end;
2809                 while (isspace(*buf)) buf++;
2810         }
2811         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2812 out:
2813         return len;
2814 }
2815
2816 static struct md_sysfs_entry md_bitmap =
2817 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2818
2819 static ssize_t
2820 size_show(mddev_t *mddev, char *page)
2821 {
2822         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2823 }
2824
2825 static int update_size(mddev_t *mddev, unsigned long size);
2826
2827 static ssize_t
2828 size_store(mddev_t *mddev, const char *buf, size_t len)
2829 {
2830         /* If array is inactive, we can reduce the component size, but
2831          * not increase it (except from 0).
2832          * If array is active, we can try an on-line resize
2833          */
2834         char *e;
2835         int err = 0;
2836         unsigned long long size = simple_strtoull(buf, &e, 10);
2837         if (!*buf || *buf == '\n' ||
2838             (*e && *e != '\n'))
2839                 return -EINVAL;
2840
2841         if (mddev->pers) {
2842                 err = update_size(mddev, size);
2843                 md_update_sb(mddev, 1);
2844         } else {
2845                 if (mddev->size == 0 ||
2846                     mddev->size > size)
2847                         mddev->size = size;
2848                 else
2849                         err = -ENOSPC;
2850         }
2851         return err ? err : len;
2852 }
2853
2854 static struct md_sysfs_entry md_size =
2855 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2856
2857
2858 /* Metdata version.
2859  * This is one of
2860  *   'none' for arrays with no metadata (good luck...)
2861  *   'external' for arrays with externally managed metadata,
2862  * or N.M for internally known formats
2863  */
2864 static ssize_t
2865 metadata_show(mddev_t *mddev, char *page)
2866 {
2867         if (mddev->persistent)
2868                 return sprintf(page, "%d.%d\n",
2869                                mddev->major_version, mddev->minor_version);
2870         else if (mddev->external)
2871                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2872         else
2873                 return sprintf(page, "none\n");
2874 }
2875
2876 static ssize_t
2877 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2878 {
2879         int major, minor;
2880         char *e;
2881         if (!list_empty(&mddev->disks))
2882                 return -EBUSY;
2883
2884         if (cmd_match(buf, "none")) {
2885                 mddev->persistent = 0;
2886                 mddev->external = 0;
2887                 mddev->major_version = 0;
2888                 mddev->minor_version = 90;
2889                 return len;
2890         }
2891         if (strncmp(buf, "external:", 9) == 0) {
2892                 size_t namelen = len-9;
2893                 if (namelen >= sizeof(mddev->metadata_type))
2894                         namelen = sizeof(mddev->metadata_type)-1;
2895                 strncpy(mddev->metadata_type, buf+9, namelen);
2896                 mddev->metadata_type[namelen] = 0;
2897                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2898                         mddev->metadata_type[--namelen] = 0;
2899                 mddev->persistent = 0;
2900                 mddev->external = 1;
2901                 mddev->major_version = 0;
2902                 mddev->minor_version = 90;
2903                 return len;
2904         }
2905         major = simple_strtoul(buf, &e, 10);
2906         if (e==buf || *e != '.')
2907                 return -EINVAL;
2908         buf = e+1;
2909         minor = simple_strtoul(buf, &e, 10);
2910         if (e==buf || (*e && *e != '\n') )
2911                 return -EINVAL;
2912         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2913                 return -ENOENT;
2914         mddev->major_version = major;
2915         mddev->minor_version = minor;
2916         mddev->persistent = 1;
2917         mddev->external = 0;
2918         return len;
2919 }
2920
2921 static struct md_sysfs_entry md_metadata =
2922 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2923
2924 static ssize_t
2925 action_show(mddev_t *mddev, char *page)
2926 {
2927         char *type = "idle";
2928         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2929             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2930                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2931                         type = "reshape";
2932                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2933                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2934                                 type = "resync";
2935                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2936                                 type = "check";
2937                         else
2938                                 type = "repair";
2939                 } else
2940                         type = "recover";
2941         }
2942         return sprintf(page, "%s\n", type);
2943 }
2944
2945 static ssize_t
2946 action_store(mddev_t *mddev, const char *page, size_t len)
2947 {
2948         if (!mddev->pers || !mddev->pers->sync_request)
2949                 return -EINVAL;
2950
2951         if (cmd_match(page, "idle")) {
2952                 if (mddev->sync_thread) {
2953                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2954                         md_unregister_thread(mddev->sync_thread);
2955                         mddev->sync_thread = NULL;
2956                         mddev->recovery = 0;
2957                 }
2958         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2959                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2960                 return -EBUSY;
2961         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2962                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2963         else if (cmd_match(page, "reshape")) {
2964                 int err;
2965                 if (mddev->pers->start_reshape == NULL)
2966                         return -EINVAL;
2967                 err = mddev->pers->start_reshape(mddev);
2968                 if (err)
2969                         return err;
2970         } else {
2971                 if (cmd_match(page, "check"))
2972                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2973                 else if (!cmd_match(page, "repair"))
2974                         return -EINVAL;
2975                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2976                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2977         }
2978         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2979         md_wakeup_thread(mddev->thread);
2980         return len;
2981 }
2982
2983 static ssize_t
2984 mismatch_cnt_show(mddev_t *mddev, char *page)
2985 {
2986         return sprintf(page, "%llu\n",
2987                        (unsigned long long) mddev->resync_mismatches);
2988 }
2989
2990 static struct md_sysfs_entry md_scan_mode =
2991 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2992
2993
2994 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2995
2996 static ssize_t
2997 sync_min_show(mddev_t *mddev, char *page)
2998 {
2999         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3000                        mddev->sync_speed_min ? "local": "system");
3001 }
3002
3003 static ssize_t
3004 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3005 {
3006         int min;
3007         char *e;
3008         if (strncmp(buf, "system", 6)==0) {
3009                 mddev->sync_speed_min = 0;
3010                 return len;
3011         }
3012         min = simple_strtoul(buf, &e, 10);
3013         if (buf == e || (*e && *e != '\n') || min <= 0)
3014                 return -EINVAL;
3015         mddev->sync_speed_min = min;
3016         return len;
3017 }
3018
3019 static struct md_sysfs_entry md_sync_min =
3020 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3021
3022 static ssize_t
3023 sync_max_show(mddev_t *mddev, char *page)
3024 {
3025         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3026                        mddev->sync_speed_max ? "local": "system");
3027 }
3028
3029 static ssize_t
3030 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3031 {
3032         int max;
3033         char *e;
3034         if (strncmp(buf, "system", 6)==0) {
3035                 mddev->sync_speed_max = 0;
3036                 return len;
3037         }
3038         max = simple_strtoul(buf, &e, 10);
3039         if (buf == e || (*e && *e != '\n') || max <= 0)
3040                 return -EINVAL;
3041         mddev->sync_speed_max = max;
3042         return len;
3043 }
3044
3045 static struct md_sysfs_entry md_sync_max =
3046 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3047
3048 static ssize_t
3049 degraded_show(mddev_t *mddev, char *page)
3050 {
3051         return sprintf(page, "%d\n", mddev->degraded);
3052 }
3053 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3054
3055 static ssize_t
3056 sync_force_parallel_show(mddev_t *mddev, char *page)
3057 {
3058         return sprintf(page, "%d\n", mddev->parallel_resync);
3059 }
3060
3061 static ssize_t
3062 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3063 {
3064         long n;
3065
3066         if (strict_strtol(buf, 10, &n))
3067                 return -EINVAL;
3068
3069         if (n != 0 && n != 1)
3070                 return -EINVAL;
3071
3072         mddev->parallel_resync = n;
3073
3074         if (mddev->sync_thread)
3075                 wake_up(&resync_wait);
3076
3077         return len;
3078 }
3079
3080 /* force parallel resync, even with shared block devices */
3081 static struct md_sysfs_entry md_sync_force_parallel =
3082 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3083        sync_force_parallel_show, sync_force_parallel_store);
3084
3085 static ssize_t
3086 sync_speed_show(mddev_t *mddev, char *page)
3087 {
3088         unsigned long resync, dt, db;
3089         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
3090         dt = ((jiffies - mddev->resync_mark) / HZ);
3091         if (!dt) dt++;
3092         db = resync - (mddev->resync_mark_cnt);
3093         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
3094 }
3095
3096 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3097
3098 static ssize_t
3099 sync_completed_show(mddev_t *mddev, char *page)
3100 {
3101         unsigned long max_blocks, resync;
3102
3103         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3104                 max_blocks = mddev->resync_max_sectors;
3105         else
3106                 max_blocks = mddev->size << 1;
3107
3108         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3109         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3110 }
3111
3112 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3113
3114 static ssize_t
3115 min_sync_show(mddev_t *mddev, char *page)
3116 {
3117         return sprintf(page, "%llu\n",
3118                        (unsigned long long)mddev->resync_min);
3119 }
3120 static ssize_t
3121 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3122 {
3123         unsigned long long min;
3124         if (strict_strtoull(buf, 10, &min))
3125                 return -EINVAL;
3126         if (min > mddev->resync_max)
3127                 return -EINVAL;
3128         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3129                 return -EBUSY;
3130
3131         /* Must be a multiple of chunk_size */
3132         if (mddev->chunk_size) {
3133                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3134                         return -EINVAL;
3135         }
3136         mddev->resync_min = min;
3137
3138         return len;
3139 }
3140
3141 static struct md_sysfs_entry md_min_sync =
3142 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3143
3144 static ssize_t
3145 max_sync_show(mddev_t *mddev, char *page)
3146 {
3147         if (mddev->resync_max == MaxSector)
3148                 return sprintf(page, "max\n");
3149         else
3150                 return sprintf(page, "%llu\n",
3151                                (unsigned long long)mddev->resync_max);
3152 }
3153 static ssize_t
3154 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3155 {
3156         if (strncmp(buf, "max", 3) == 0)
3157                 mddev->resync_max = MaxSector;
3158         else {
3159                 unsigned long long max;
3160                 if (strict_strtoull(buf, 10, &max))
3161                         return -EINVAL;
3162                 if (max < mddev->resync_min)
3163                         return -EINVAL;
3164                 if (max < mddev->resync_max &&
3165                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3166                         return -EBUSY;
3167
3168                 /* Must be a multiple of chunk_size */
3169                 if (mddev->chunk_size) {
3170                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3171                                 return -EINVAL;
3172                 }
3173                 mddev->resync_max = max;
3174         }
3175         wake_up(&mddev->recovery_wait);
3176         return len;
3177 }
3178
3179 static struct md_sysfs_entry md_max_sync =
3180 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3181
3182 static ssize_t
3183 suspend_lo_show(mddev_t *mddev, char *page)
3184 {
3185         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3186 }
3187
3188 static ssize_t
3189 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3190 {
3191         char *e;
3192         unsigned long long new = simple_strtoull(buf, &e, 10);
3193
3194         if (mddev->pers->quiesce == NULL)
3195                 return -EINVAL;
3196         if (buf == e || (*e && *e != '\n'))
3197                 return -EINVAL;
3198         if (new >= mddev->suspend_hi ||
3199             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3200                 mddev->suspend_lo = new;
3201                 mddev->pers->quiesce(mddev, 2);
3202                 return len;
3203         } else
3204                 return -EINVAL;
3205 }
3206 static struct md_sysfs_entry md_suspend_lo =
3207 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3208
3209
3210 static ssize_t
3211 suspend_hi_show(mddev_t *mddev, char *page)
3212 {
3213         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3214 }
3215
3216 static ssize_t
3217 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3218 {
3219         char *e;
3220         unsigned long long new = simple_strtoull(buf, &e, 10);
3221
3222         if (mddev->pers->quiesce == NULL)
3223                 return -EINVAL;
3224         if (buf == e || (*e && *e != '\n'))
3225                 return -EINVAL;
3226         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3227             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3228                 mddev->suspend_hi = new;
3229                 mddev->pers->quiesce(mddev, 1);
3230                 mddev->pers->quiesce(mddev, 0);
3231                 return len;
3232         } else
3233                 return -EINVAL;
3234 }
3235 static struct md_sysfs_entry md_suspend_hi =
3236 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3237
3238 static ssize_t
3239 reshape_position_show(mddev_t *mddev, char *page)
3240 {
3241         if (mddev->reshape_position != MaxSector)
3242                 return sprintf(page, "%llu\n",
3243                                (unsigned long long)mddev->reshape_position);
3244         strcpy(page, "none\n");
3245         return 5;
3246 }
3247
3248 static ssize_t
3249 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3250 {
3251         char *e;
3252         unsigned long long new = simple_strtoull(buf, &e, 10);
3253         if (mddev->pers)
3254                 return -EBUSY;
3255         if (buf == e || (*e && *e != '\n'))
3256                 return -EINVAL;
3257         mddev->reshape_position = new;
3258         mddev->delta_disks = 0;
3259         mddev->new_level = mddev->level;
3260         mddev->new_layout = mddev->layout;
3261         mddev->new_chunk = mddev->chunk_size;
3262         return len;
3263 }
3264
3265 static struct md_sysfs_entry md_reshape_position =
3266 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3267        reshape_position_store);
3268
3269
3270 static struct attribute *md_default_attrs[] = {
3271         &md_level.attr,
3272         &md_layout.attr,
3273         &md_raid_disks.attr,
3274         &md_chunk_size.attr,
3275         &md_size.attr,
3276         &md_resync_start.attr,
3277         &md_metadata.attr,
3278         &md_new_device.attr,
3279         &md_safe_delay.attr,
3280         &md_array_state.attr,
3281         &md_reshape_position.attr,
3282         NULL,
3283 };
3284
3285 static struct attribute *md_redundancy_attrs[] = {
3286         &md_scan_mode.attr,
3287         &md_mismatches.attr,
3288         &md_sync_min.attr,
3289         &md_sync_max.attr,
3290         &md_sync_speed.attr,
3291         &md_sync_force_parallel.attr,
3292         &md_sync_completed.attr,
3293         &md_min_sync.attr,
3294         &md_max_sync.attr,
3295         &md_suspend_lo.attr,
3296         &md_suspend_hi.attr,
3297         &md_bitmap.attr,
3298         &md_degraded.attr,
3299         NULL,
3300 };
3301 static struct attribute_group md_redundancy_group = {
3302         .name = NULL,
3303         .attrs = md_redundancy_attrs,
3304 };
3305
3306
3307 static ssize_t
3308 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3309 {
3310         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3311         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3312         ssize_t rv;
3313
3314         if (!entry->show)
3315                 return -EIO;
3316         rv = mddev_lock(mddev);
3317         if (!rv) {
3318                 rv = entry->show(mddev, page);
3319                 mddev_unlock(mddev);
3320         }
3321         return rv;
3322 }
3323
3324 static ssize_t
3325 md_attr_store(struct kobject *kobj, struct attribute *attr,
3326               const char *page, size_t length)
3327 {
3328         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3329         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3330         ssize_t rv;
3331
3332         if (!entry->store)
3333                 return -EIO;
3334         if (!capable(CAP_SYS_ADMIN))
3335                 return -EACCES;
3336         rv = mddev_lock(mddev);
3337         if (!rv) {
3338                 rv = entry->store(mddev, page, length);
3339                 mddev_unlock(mddev);
3340         }
3341         return rv;
3342 }
3343
3344 static void md_free(struct kobject *ko)
3345 {
3346         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3347         kfree(mddev);
3348 }
3349
3350 static struct sysfs_ops md_sysfs_ops = {
3351         .show   = md_attr_show,
3352         .store  = md_attr_store,
3353 };
3354 static struct kobj_type md_ktype = {
3355         .release        = md_free,
3356         .sysfs_ops      = &md_sysfs_ops,
3357         .default_attrs  = md_default_attrs,
3358 };
3359
3360 int mdp_major = 0;
3361
3362 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3363 {
3364         static DEFINE_MUTEX(disks_mutex);
3365         mddev_t *mddev = mddev_find(dev);
3366         struct gendisk *disk;
3367         int partitioned = (MAJOR(dev) != MD_MAJOR);
3368         int shift = partitioned ? MdpMinorShift : 0;
3369         int unit = MINOR(dev) >> shift;
3370         int error;
3371
3372         if (!mddev)
3373                 return NULL;
3374
3375         mutex_lock(&disks_mutex);
3376         if (mddev->gendisk) {
3377                 mutex_unlock(&disks_mutex);
3378                 mddev_put(mddev);
3379                 return NULL;
3380         }
3381         disk = alloc_disk(1 << shift);
3382         if (!disk) {
3383                 mutex_unlock(&disks_mutex);
3384                 mddev_put(mddev);
3385                 return NULL;
3386         }
3387         disk->major = MAJOR(dev);
3388         disk->first_minor = unit << shift;
3389         if (partitioned)
3390                 sprintf(disk->disk_name, "md_d%d", unit);
3391         else
3392                 sprintf(disk->disk_name, "md%d", unit);
3393         disk->fops = &md_fops;
3394         disk->private_data = mddev;
3395         disk->queue = mddev->queue;
3396         add_disk(disk);
3397         mddev->gendisk = disk;
3398         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3399                                      "%s", "md");
3400         mutex_unlock(&disks_mutex);
3401         if (error)
3402                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3403                        disk->disk_name);
3404         else
3405                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3406         return NULL;
3407 }
3408
3409 static void md_safemode_timeout(unsigned long data)
3410 {
3411         mddev_t *mddev = (mddev_t *) data;
3412
3413         if (!atomic_read(&mddev->writes_pending)) {
3414                 mddev->safemode = 1;
3415                 if (mddev->external)
3416                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3417         }
3418         md_wakeup_thread(mddev->thread);
3419 }
3420
3421 static int start_dirty_degraded;
3422
3423 static int do_md_run(mddev_t * mddev)
3424 {
3425         int err;
3426         int chunk_size;
3427         struct list_head *tmp;
3428         mdk_rdev_t *rdev;
3429         struct gendisk *disk;
3430         struct mdk_personality *pers;
3431         char b[BDEVNAME_SIZE];
3432
3433         if (list_empty(&mddev->disks))
3434                 /* cannot run an array with no devices.. */
3435                 return -EINVAL;
3436
3437         if (mddev->pers)
3438                 return -EBUSY;
3439
3440         /*
3441          * Analyze all RAID superblock(s)
3442          */
3443         if (!mddev->raid_disks) {
3444                 if (!mddev->persistent)
3445                         return -EINVAL;
3446                 analyze_sbs(mddev);
3447         }
3448
3449         chunk_size = mddev->chunk_size;
3450
3451         if (chunk_size) {
3452                 if (chunk_size > MAX_CHUNK_SIZE) {
3453                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3454                                 chunk_size, MAX_CHUNK_SIZE);
3455                         return -EINVAL;
3456                 }
3457                 /*
3458                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3459                  */
3460                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3461                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3462                         return -EINVAL;
3463                 }
3464                 if (chunk_size < PAGE_SIZE) {
3465                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3466                                 chunk_size, PAGE_SIZE);
3467                         return -EINVAL;
3468                 }
3469
3470                 /* devices must have minimum size of one chunk */
3471                 rdev_for_each(rdev, tmp, mddev) {
3472                         if (test_bit(Faulty, &rdev->flags))
3473                                 continue;
3474                         if (rdev->size < chunk_size / 1024) {
3475                                 printk(KERN_WARNING
3476                                         "md: Dev %s smaller than chunk_size:"
3477                                         " %lluk < %dk\n",
3478                                         bdevname(rdev->bdev,b),
3479                                         (unsigned long long)rdev->size,
3480                                         chunk_size / 1024);
3481                                 return -EINVAL;
3482                         }
3483                 }
3484         }
3485
3486 #ifdef CONFIG_KMOD
3487         if (mddev->level != LEVEL_NONE)
3488                 request_module("md-level-%d", mddev->level);
3489         else if (mddev->clevel[0])
3490                 request_module("md-%s", mddev->clevel);
3491 #endif
3492
3493         /*
3494          * Drop all container device buffers, from now on
3495          * the only valid external interface is through the md
3496          * device.
3497          */
3498         rdev_for_each(rdev, tmp, mddev) {
3499                 if (test_bit(Faulty, &rdev->flags))
3500                         continue;
3501                 sync_blockdev(rdev->bdev);
3502                 invalidate_bdev(rdev->bdev);
3503
3504                 /* perform some consistency tests on the device.
3505                  * We don't want the data to overlap the metadata,
3506                  * Internal Bitmap issues has handled elsewhere.
3507                  */
3508                 if (rdev->data_offset < rdev->sb_offset) {
3509                         if (mddev->size &&
3510                             rdev->data_offset + mddev->size*2
3511                             > rdev->sb_offset*2) {
3512                                 printk("md: %s: data overlaps metadata\n",
3513                                        mdname(mddev));
3514                                 return -EINVAL;
3515                         }
3516                 } else {
3517                         if (rdev->sb_offset*2 + rdev->sb_size/512
3518                             > rdev->data_offset) {
3519                                 printk("md: %s: metadata overlaps data\n",
3520                                        mdname(mddev));
3521                                 return -EINVAL;
3522                         }
3523                 }
3524         }
3525
3526         md_probe(mddev->unit, NULL, NULL);
3527         disk = mddev->gendisk;
3528         if (!disk)
3529                 return -ENOMEM;
3530
3531         spin_lock(&pers_lock);
3532         pers = find_pers(mddev->level, mddev->clevel);
3533         if (!pers || !try_module_get(pers->owner)) {
3534                 spin_unlock(&pers_lock);
3535                 if (mddev->level != LEVEL_NONE)
3536                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3537                                mddev->level);
3538                 else
3539                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3540                                mddev->clevel);
3541                 return -EINVAL;
3542         }
3543         mddev->pers = pers;
3544         spin_unlock(&pers_lock);
3545         mddev->level = pers->level;
3546         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3547
3548         if (mddev->reshape_position != MaxSector &&
3549             pers->start_reshape == NULL) {
3550                 /* This personality cannot handle reshaping... */
3551                 mddev->pers = NULL;
3552                 module_put(pers->owner);
3553                 return -EINVAL;
3554         }
3555
3556         if (pers->sync_request) {
3557                 /* Warn if this is a potentially silly
3558                  * configuration.
3559                  */
3560                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3561                 mdk_rdev_t *rdev2;
3562                 struct list_head *tmp2;
3563                 int warned = 0;
3564                 rdev_for_each(rdev, tmp, mddev) {
3565                         rdev_for_each(rdev2, tmp2, mddev) {
3566                                 if (rdev < rdev2 &&
3567                                     rdev->bdev->bd_contains ==
3568                                     rdev2->bdev->bd_contains) {
3569                                         printk(KERN_WARNING
3570                                                "%s: WARNING: %s appears to be"
3571                                                " on the same physical disk as"
3572                                                " %s.\n",
3573                                                mdname(mddev),
3574                                                bdevname(rdev->bdev,b),
3575                                                bdevname(rdev2->bdev,b2));
3576                                         warned = 1;
3577                                 }
3578                         }
3579                 }
3580                 if (warned)
3581                         printk(KERN_WARNING
3582                                "True protection against single-disk"
3583                                " failure might be compromised.\n");
3584         }
3585
3586         mddev->recovery = 0;
3587         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3588         mddev->barriers_work = 1;
3589         mddev->ok_start_degraded = start_dirty_degraded;
3590
3591         if (start_readonly)
3592                 mddev->ro = 2; /* read-only, but switch on first write */
3593
3594         err = mddev->pers->run(mddev);
3595         if (!err && mddev->pers->sync_request) {
3596                 err = bitmap_create(mddev);
3597                 if (err) {
3598                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3599                                mdname(mddev), err);
3600                         mddev->pers->stop(mddev);
3601                 }
3602         }
3603         if (err) {
3604                 printk(KERN_ERR "md: pers->run() failed ...\n");
3605                 module_put(mddev->pers->owner);
3606                 mddev->pers = NULL;
3607                 bitmap_destroy(mddev);
3608                 return err;
3609         }
3610         if (mddev->pers->sync_request) {
3611                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3612                         printk(KERN_WARNING
3613                                "md: cannot register extra attributes for %s\n",
3614                                mdname(mddev));
3615         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3616                 mddev->ro = 0;
3617
3618         atomic_set(&mddev->writes_pending,0);
3619         mddev->safemode = 0;
3620         mddev->safemode_timer.function = md_safemode_timeout;
3621         mddev->safemode_timer.data = (unsigned long) mddev;
3622         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3623         mddev->in_sync = 1;
3624
3625         rdev_for_each(rdev, tmp, mddev)
3626                 if (rdev->raid_disk >= 0) {
3627                         char nm[20];
3628                         sprintf(nm, "rd%d", rdev->raid_disk);
3629                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3630                                 printk("md: cannot register %s for %s\n",
3631                                        nm, mdname(mddev));
3632                 }
3633
3634         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3635
3636         if (mddev->flags)
3637                 md_update_sb(mddev, 0);
3638
3639         set_capacity(disk, mddev->array_size<<1);
3640
3641         /* If we call blk_queue_make_request here, it will
3642          * re-initialise max_sectors etc which may have been
3643          * refined inside -> run.  So just set the bits we need to set.
3644          * Most initialisation happended when we called
3645          * blk_queue_make_request(..., md_fail_request)
3646          * earlier.
3647          */
3648         mddev->queue->queuedata = mddev;
3649         mddev->queue->make_request_fn = mddev->pers->make_request;
3650
3651         /* If there is a partially-recovered drive we need to
3652          * start recovery here.  If we leave it to md_check_recovery,
3653          * it will remove the drives and not do the right thing
3654          */
3655         if (mddev->degraded && !mddev->sync_thread) {
3656                 struct list_head *rtmp;
3657                 int spares = 0;
3658                 rdev_for_each(rdev, rtmp, mddev)
3659                         if (rdev->raid_disk >= 0 &&
3660                             !test_bit(In_sync, &rdev->flags) &&
3661                             !test_bit(Faulty, &rdev->flags))
3662                                 /* complete an interrupted recovery */
3663                                 spares++;
3664                 if (spares && mddev->pers->sync_request) {
3665                         mddev->recovery = 0;
3666                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3667                         mddev->sync_thread = md_register_thread(md_do_sync,
3668                                                                 mddev,
3669                                                                 "%s_resync");
3670                         if (!mddev->sync_thread) {
3671                                 printk(KERN_ERR "%s: could not start resync"
3672                                        " thread...\n",
3673                                        mdname(mddev));
3674                                 /* leave the spares where they are, it shouldn't hurt */
3675                                 mddev->recovery = 0;
3676                         }
3677                 }
3678         }
3679         md_wakeup_thread(mddev->thread);
3680         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3681
3682         mddev->changed = 1;
3683         md_new_event(mddev);
3684         sysfs_notify(&mddev->kobj, NULL, "array_state");
3685         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3686         return 0;
3687 }
3688
3689 static int restart_array(mddev_t *mddev)
3690 {
3691         struct gendisk *disk = mddev->gendisk;
3692         int err;
3693
3694         /*
3695          * Complain if it has no devices
3696          */
3697         err = -ENXIO;
3698         if (list_empty(&mddev->disks))
3699                 goto out;
3700
3701         if (mddev->pers) {
3702                 err = -EBUSY;
3703                 if (!mddev->ro)
3704                         goto out;
3705
3706                 mddev->safemode = 0;
3707                 mddev->ro = 0;
3708                 set_disk_ro(disk, 0);
3709
3710                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3711                         mdname(mddev));
3712                 /*
3713                  * Kick recovery or resync if necessary
3714                  */
3715                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3716                 md_wakeup_thread(mddev->thread);
3717                 md_wakeup_thread(mddev->sync_thread);
3718                 err = 0;
3719                 sysfs_notify(&mddev->kobj, NULL, "array_state");
3720
3721         } else
3722                 err = -EINVAL;
3723
3724 out:
3725         return err;
3726 }
3727
3728 /* similar to deny_write_access, but accounts for our holding a reference
3729  * to the file ourselves */
3730 static int deny_bitmap_write_access(struct file * file)
3731 {
3732         struct inode *inode = file->f_mapping->host;
3733
3734         spin_lock(&inode->i_lock);
3735         if (atomic_read(&inode->i_writecount) > 1) {
3736                 spin_unlock(&inode->i_lock);
3737                 return -ETXTBSY;
3738         }
3739         atomic_set(&inode->i_writecount, -1);
3740         spin_unlock(&inode->i_lock);
3741
3742         return 0;
3743 }
3744
3745 static void restore_bitmap_write_access(struct file *file)
3746 {
3747         struct inode *inode = file->f_mapping->host;
3748
3749         spin_lock(&inode->i_lock);
3750         atomic_set(&inode->i_writecount, 1);
3751         spin_unlock(&inode->i_lock);
3752 }
3753
3754 /* mode:
3755  *   0 - completely stop and dis-assemble array
3756  *   1 - switch to readonly
3757  *   2 - stop but do not disassemble array
3758  */
3759 static int do_md_stop(mddev_t * mddev, int mode)
3760 {
3761         int err = 0;
3762         struct gendisk *disk = mddev->gendisk;
3763
3764         if (mddev->pers) {
3765                 if (atomic_read(&mddev->active)>2) {
3766                         printk("md: %s still in use.\n",mdname(mddev));
3767                         return -EBUSY;
3768                 }
3769
3770                 if (mddev->sync_thread) {
3771                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3772                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3773                         md_unregister_thread(mddev->sync_thread);
3774                         mddev->sync_thread = NULL;
3775                 }
3776
3777                 del_timer_sync(&mddev->safemode_timer);
3778
3779                 invalidate_partition(disk, 0);
3780
3781                 switch(mode) {
3782                 case 1: /* readonly */
3783                         err  = -ENXIO;
3784                         if (mddev->ro==1)
3785                                 goto out;
3786                         mddev->ro = 1;
3787                         break;
3788                 case 0: /* disassemble */
3789                 case 2: /* stop */
3790                         bitmap_flush(mddev);
3791                         md_super_wait(mddev);
3792                         if (mddev->ro)
3793                                 set_disk_ro(disk, 0);
3794                         blk_queue_make_request(mddev->queue, md_fail_request);
3795                         mddev->pers->stop(mddev);
3796                         mddev->queue->merge_bvec_fn = NULL;
3797                         mddev->queue->unplug_fn = NULL;
3798                         mddev->queue->backing_dev_info.congested_fn = NULL;
3799                         if (mddev->pers->sync_request)
3800                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3801
3802                         module_put(mddev->pers->owner);
3803                         mddev->pers = NULL;
3804                         /* tell userspace to handle 'inactive' */
3805                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3806
3807                         set_capacity(disk, 0);
3808                         mddev->changed = 1;
3809
3810                         if (mddev->ro)
3811                                 mddev->ro = 0;
3812                 }
3813                 if (!mddev->in_sync || mddev->flags) {
3814                         /* mark array as shutdown cleanly */
3815                         mddev->in_sync = 1;
3816                         md_update_sb(mddev, 1);
3817                 }
3818                 if (mode == 1)
3819                         set_disk_ro(disk, 1);
3820                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3821         }
3822
3823         /*
3824          * Free resources if final stop
3825          */
3826         if (mode == 0) {
3827                 mdk_rdev_t *rdev;
3828                 struct list_head *tmp;
3829
3830                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3831
3832                 bitmap_destroy(mddev);
3833                 if (mddev->bitmap_file) {
3834                         restore_bitmap_write_access(mddev->bitmap_file);
3835                         fput(mddev->bitmap_file);
3836                         mddev->bitmap_file = NULL;
3837                 }
3838                 mddev->bitmap_offset = 0;
3839
3840                 rdev_for_each(rdev, tmp, mddev)
3841                         if (rdev->raid_disk >= 0) {
3842                                 char nm[20];
3843                                 sprintf(nm, "rd%d", rdev->raid_disk);
3844                                 sysfs_remove_link(&mddev->kobj, nm);
3845                         }
3846
3847                 /* make sure all md_delayed_delete calls have finished */
3848                 flush_scheduled_work();
3849
3850                 export_array(mddev);
3851
3852                 mddev->array_size = 0;
3853                 mddev->size = 0;
3854                 mddev->raid_disks = 0;
3855                 mddev->recovery_cp = 0;
3856                 mddev->resync_min = 0;
3857                 mddev->resync_max = MaxSector;
3858                 mddev->reshape_position = MaxSector;
3859                 mddev->external = 0;
3860                 mddev->persistent = 0;
3861                 mddev->level = LEVEL_NONE;
3862                 mddev->clevel[0] = 0;
3863                 mddev->flags = 0;
3864                 mddev->ro = 0;
3865                 mddev->metadata_type[0] = 0;
3866                 mddev->chunk_size = 0;
3867                 mddev->ctime = mddev->utime = 0;
3868                 mddev->layout = 0;
3869                 mddev->max_disks = 0;
3870                 mddev->events = 0;
3871                 mddev->delta_disks = 0;
3872                 mddev->new_level = LEVEL_NONE;
3873                 mddev->new_layout = 0;
3874                 mddev->new_chunk = 0;
3875                 mddev->curr_resync = 0;
3876                 mddev->resync_mismatches = 0;
3877                 mddev->suspend_lo = mddev->suspend_hi = 0;
3878                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3879                 mddev->recovery = 0;
3880                 mddev->in_sync = 0;
3881                 mddev->changed = 0;
3882                 mddev->degraded = 0;
3883                 mddev->barriers_work = 0;
3884                 mddev->safemode = 0;
3885
3886         } else if (mddev->pers)
3887                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3888                         mdname(mddev));
3889         err = 0;
3890         md_new_event(mddev);
3891         sysfs_notify(&mddev->kobj, NULL, "array_state");
3892 out:
3893         return err;
3894 }
3895
3896 #ifndef MODULE
3897 static void autorun_array(mddev_t *mddev)
3898 {
3899         mdk_rdev_t *rdev;
3900         struct list_head *tmp;
3901         int err;
3902
3903         if (list_empty(&mddev->disks))
3904                 return;
3905
3906         printk(KERN_INFO "md: running: ");
3907
3908         rdev_for_each(rdev, tmp, mddev) {
3909                 char b[BDEVNAME_SIZE];
3910                 printk("<%s>", bdevname(rdev->bdev,b));
3911         }
3912         printk("\n");
3913
3914         err = do_md_run (mddev);
3915         if (err) {
3916                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3917                 do_md_stop (mddev, 0);
3918         }
3919 }
3920
3921 /*
3922  * lets try to run arrays based on all disks that have arrived
3923  * until now. (those are in pending_raid_disks)
3924  *
3925  * the method: pick the first pending disk, collect all disks with
3926  * the same UUID, remove all from the pending list and put them into
3927  * the 'same_array' list. Then order this list based on superblock
3928  * update time (freshest comes first), kick out 'old' disks and
3929  * compare superblocks. If everything's fine then run it.
3930  *
3931  * If "unit" is allocated, then bump its reference count
3932  */
3933 static void autorun_devices(int part)
3934 {
3935         struct list_head *tmp;
3936         mdk_rdev_t *rdev0, *rdev;
3937         mddev_t *mddev;
3938         char b[BDEVNAME_SIZE];
3939
3940         printk(KERN_INFO "md: autorun ...\n");
3941         while (!list_empty(&pending_raid_disks)) {
3942                 int unit;
3943                 dev_t dev;
3944                 LIST_HEAD(candidates);
3945                 rdev0 = list_entry(pending_raid_disks.next,
3946                                          mdk_rdev_t, same_set);
3947
3948                 printk(KERN_INFO "md: considering %s ...\n",
3949                         bdevname(rdev0->bdev,b));
3950                 INIT_LIST_HEAD(&candidates);
3951                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3952                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3953                                 printk(KERN_INFO "md:  adding %s ...\n",
3954                                         bdevname(rdev->bdev,b));
3955                                 list_move(&rdev->same_set, &candidates);
3956                         }
3957                 /*
3958                  * now we have a set of devices, with all of them having
3959                  * mostly sane superblocks. It's time to allocate the
3960                  * mddev.
3961                  */
3962                 if (part) {
3963                         dev = MKDEV(mdp_major,
3964                                     rdev0->preferred_minor << MdpMinorShift);
3965                         unit = MINOR(dev) >> MdpMinorShift;
3966                 } else {
3967                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3968                         unit = MINOR(dev);
3969                 }
3970                 if (rdev0->preferred_minor != unit) {
3971                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3972                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3973                         break;
3974                 }
3975
3976                 md_probe(dev, NULL, NULL);
3977                 mddev = mddev_find(dev);
3978                 if (!mddev || !mddev->gendisk) {
3979                         if (mddev)
3980                                 mddev_put(mddev);
3981                         printk(KERN_ERR
3982                                 "md: cannot allocate memory for md drive.\n");
3983                         break;
3984                 }
3985                 if (mddev_lock(mddev))
3986                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3987                                mdname(mddev));
3988                 else if (mddev->raid_disks || mddev->major_version
3989                          || !list_empty(&mddev->disks)) {
3990                         printk(KERN_WARNING
3991                                 "md: %s already running, cannot run %s\n",
3992                                 mdname(mddev), bdevname(rdev0->bdev,b));
3993                         mddev_unlock(mddev);
3994                 } else {
3995                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3996                         mddev->persistent = 1;
3997                         rdev_for_each_list(rdev, tmp, candidates) {
3998                                 list_del_init(&rdev->same_set);
3999                                 if (bind_rdev_to_array(rdev, mddev))
4000                                         export_rdev(rdev);
4001                         }
4002                         autorun_array(mddev);
4003                         mddev_unlock(mddev);
4004                 }
4005                 /* on success, candidates will be empty, on error
4006                  * it won't...
4007                  */
4008                 rdev_for_each_list(rdev, tmp, candidates)
4009                         export_rdev(rdev);
4010                 mddev_put(mddev);
4011         }
4012         printk(KERN_INFO "md: ... autorun DONE.\n");
4013 }
4014 #endif /* !MODULE */
4015
4016 static int get_version(void __user * arg)
4017 {
4018         mdu_version_t ver;
4019
4020         ver.major = MD_MAJOR_VERSION;
4021         ver.minor = MD_MINOR_VERSION;
4022         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4023
4024         if (copy_to_user(arg, &ver, sizeof(ver)))
4025                 return -EFAULT;
4026
4027         return 0;
4028 }
4029
4030 static int get_array_info(mddev_t * mddev, void __user * arg)
4031 {
4032         mdu_array_info_t info;
4033         int nr,working,active,failed,spare;
4034         mdk_rdev_t *rdev;
4035         struct list_head *tmp;
4036
4037         nr=working=active=failed=spare=0;
4038         rdev_for_each(rdev, tmp, mddev) {
4039                 nr++;
4040                 if (test_bit(Faulty, &rdev->flags))
4041                         failed++;
4042                 else {
4043                         working++;
4044                         if (test_bit(In_sync, &rdev->flags))
4045                                 active++;
4046                         else
4047                                 spare++;
4048                 }
4049         }
4050
4051         info.major_version = mddev->major_version;
4052         info.minor_version = mddev->minor_version;
4053         info.patch_version = MD_PATCHLEVEL_VERSION;
4054         info.ctime         = mddev->ctime;
4055         info.level         = mddev->level;
4056         info.size          = mddev->size;
4057         if (info.size != mddev->size) /* overflow */
4058                 info.size = -1;
4059         info.nr_disks      = nr;
4060         info.raid_disks    = mddev->raid_disks;
4061         info.md_minor      = mddev->md_minor;
4062         info.not_persistent= !mddev->persistent;
4063
4064         info.utime         = mddev->utime;
4065         info.state         = 0;
4066         if (mddev->in_sync)
4067                 info.state = (1<<MD_SB_CLEAN);
4068         if (mddev->bitmap && mddev->bitmap_offset)
4069                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4070         info.active_disks  = active;
4071         info.working_disks = working;
4072         info.failed_disks  = failed;
4073         info.spare_disks   = spare;
4074
4075         info.layout        = mddev->layout;
4076         info.chunk_size    = mddev->chunk_size;
4077
4078         if (copy_to_user(arg, &info, sizeof(info)))
4079                 return -EFAULT;
4080
4081         return 0;
4082 }
4083
4084 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4085 {
4086         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4087         char *ptr, *buf = NULL;
4088         int err = -ENOMEM;
4089
4090         md_allow_write(mddev);
4091
4092         file = kmalloc(sizeof(*file), GFP_KERNEL);
4093         if (!file)
4094                 goto out;
4095
4096         /* bitmap disabled, zero the first byte and copy out */
4097         if (!mddev->bitmap || !mddev->bitmap->file) {
4098                 file->pathname[0] = '\0';
4099                 goto copy_out;
4100         }
4101
4102         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4103         if (!buf)
4104                 goto out;
4105
4106         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4107         if (IS_ERR(ptr))
4108                 goto out;
4109
4110         strcpy(file->pathname, ptr);
4111
4112 copy_out:
4113         err = 0;
4114         if (copy_to_user(arg, file, sizeof(*file)))
4115                 err = -EFAULT;
4116 out:
4117         kfree(buf);
4118         kfree(file);
4119         return err;
4120 }
4121
4122 static int get_disk_info(mddev_t * mddev, void __user * arg)
4123 {
4124         mdu_disk_info_t info;
4125         unsigned int nr;
4126         mdk_rdev_t *rdev;
4127
4128         if (copy_from_user(&info, arg, sizeof(info)))
4129                 return -EFAULT;
4130
4131         nr = info.number;
4132
4133         rdev = find_rdev_nr(mddev, nr);
4134         if (rdev) {
4135                 info.major = MAJOR(rdev->bdev->bd_dev);
4136                 info.minor = MINOR(rdev->bdev->bd_dev);
4137                 info.raid_disk = rdev->raid_disk;
4138                 info.state = 0;
4139                 if (test_bit(Faulty, &rdev->flags))
4140                         info.state |= (1<<MD_DISK_FAULTY);
4141                 else if (test_bit(In_sync, &rdev->flags)) {
4142                         info.state |= (1<<MD_DISK_ACTIVE);
4143                         info.state |= (1<<MD_DISK_SYNC);
4144                 }
4145                 if (test_bit(WriteMostly, &rdev->flags))
4146                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4147         } else {
4148                 info.major = info.minor = 0;
4149                 info.raid_disk = -1;
4150                 info.state = (1<<MD_DISK_REMOVED);
4151         }
4152
4153         if (copy_to_user(arg, &info, sizeof(info)))
4154                 return -EFAULT;
4155
4156         return 0;
4157 }
4158
4159 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4160 {
4161         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4162         mdk_rdev_t *rdev;
4163         dev_t dev = MKDEV(info->major,info->minor);
4164
4165         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4166                 return -EOVERFLOW;
4167
4168         if (!mddev->raid_disks) {
4169                 int err;
4170                 /* expecting a device which has a superblock */
4171                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4172                 if (IS_ERR(rdev)) {
4173                         printk(KERN_WARNING
4174                                 "md: md_import_device returned %ld\n",
4175                                 PTR_ERR(rdev));
4176                         return PTR_ERR(rdev);
4177                 }
4178                 if (!list_empty(&mddev->disks)) {
4179                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4180                                                         mdk_rdev_t, same_set);
4181                         int err = super_types[mddev->major_version]
4182                                 .load_super(rdev, rdev0, mddev->minor_version);
4183                         if (err < 0) {
4184                                 printk(KERN_WARNING
4185                                         "md: %s has different UUID to %s\n",
4186                                         bdevname(rdev->bdev,b),
4187                                         bdevname(rdev0->bdev,b2));
4188                                 export_rdev(rdev);
4189                                 return -EINVAL;
4190                         }
4191                 }
4192                 err = bind_rdev_to_array(rdev, mddev);
4193                 if (err)
4194                         export_rdev(rdev);
4195                 return err;
4196         }
4197
4198         /*
4199          * add_new_disk can be used once the array is assembled
4200          * to add "hot spares".  They must already have a superblock
4201          * written
4202          */
4203         if (mddev->pers) {
4204                 int err;
4205                 if (!mddev->pers->hot_add_disk) {
4206                         printk(KERN_WARNING
4207                                 "%s: personality does not support diskops!\n",
4208                                mdname(mddev));
4209                         return -EINVAL;
4210                 }
4211                 if (mddev->persistent)
4212                         rdev = md_import_device(dev, mddev->major_version,
4213                                                 mddev->minor_version);
4214                 else
4215                         rdev = md_import_device(dev, -1, -1);
4216                 if (IS_ERR(rdev)) {
4217                         printk(KERN_WARNING
4218                                 "md: md_import_device returned %ld\n",
4219                                 PTR_ERR(rdev));
4220                         return PTR_ERR(rdev);
4221                 }
4222                 /* set save_raid_disk if appropriate */
4223                 if (!mddev->persistent) {
4224                         if (info->state & (1<<MD_DISK_SYNC)  &&
4225                             info->raid_disk < mddev->raid_disks)
4226                                 rdev->raid_disk = info->raid_disk;
4227                         else
4228                                 rdev->raid_disk = -1;
4229                 } else
4230                         super_types[mddev->major_version].
4231                                 validate_super(mddev, rdev);
4232                 rdev->saved_raid_disk = rdev->raid_disk;
4233
4234                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4235                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4236                         set_bit(WriteMostly, &rdev->flags);
4237
4238                 rdev->raid_disk = -1;
4239                 err = bind_rdev_to_array(rdev, mddev);
4240                 if (!err && !mddev->pers->hot_remove_disk) {
4241                         /* If there is hot_add_disk but no hot_remove_disk
4242                          * then added disks for geometry changes,
4243                          * and should be added immediately.
4244                          */
4245                         super_types[mddev->major_version].
4246                                 validate_super(mddev, rdev);
4247                         err = mddev->pers->hot_add_disk(mddev, rdev);
4248                         if (err)
4249                                 unbind_rdev_from_array(rdev);
4250                 }
4251                 if (err)
4252                         export_rdev(rdev);
4253
4254                 md_update_sb(mddev, 1);
4255                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4256                 md_wakeup_thread(mddev->thread);
4257                 return err;
4258         }
4259
4260         /* otherwise, add_new_disk is only allowed
4261          * for major_version==0 superblocks
4262          */
4263         if (mddev->major_version != 0) {
4264                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4265                        mdname(mddev));
4266                 return -EINVAL;
4267         }
4268
4269         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4270                 int err;
4271                 rdev = md_import_device (dev, -1, 0);
4272                 if (IS_ERR(rdev)) {
4273                         printk(KERN_WARNING
4274                                 "md: error, md_import_device() returned %ld\n",
4275                                 PTR_ERR(rdev));
4276                         return PTR_ERR(rdev);
4277                 }
4278                 rdev->desc_nr = info->number;
4279                 if (info->raid_disk < mddev->raid_disks)
4280                         rdev->raid_disk = info->raid_disk;
4281                 else
4282                         rdev->raid_disk = -1;
4283
4284                 if (rdev->raid_disk < mddev->raid_disks)
4285                         if (info->state & (1<<MD_DISK_SYNC))
4286                                 set_bit(In_sync, &rdev->flags);
4287
4288                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4289                         set_bit(WriteMostly, &rdev->flags);
4290
4291                 if (!mddev->persistent) {
4292                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4293                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4294                 } else
4295                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4296                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4297
4298                 err = bind_rdev_to_array(rdev, mddev);
4299                 if (err) {
4300                         export_rdev(rdev);
4301                         return err;
4302                 }
4303         }
4304
4305         return 0;
4306 }
4307
4308 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4309 {
4310         char b[BDEVNAME_SIZE];
4311         mdk_rdev_t *rdev;
4312
4313         rdev = find_rdev(mddev, dev);
4314         if (!rdev)
4315                 return -ENXIO;
4316
4317         if (rdev->raid_disk >= 0)
4318                 goto busy;
4319
4320         kick_rdev_from_array(rdev);
4321         md_update_sb(mddev, 1);
4322         md_new_event(mddev);
4323
4324         return 0;
4325 busy:
4326         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4327                 bdevname(rdev->bdev,b), mdname(mddev));
4328         return -EBUSY;
4329 }
4330
4331 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4332 {
4333         char b[BDEVNAME_SIZE];
4334         int err;
4335         unsigned int size;
4336         mdk_rdev_t *rdev;
4337
4338         if (!mddev->pers)
4339                 return -ENODEV;
4340
4341         if (mddev->major_version != 0) {
4342                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4343                         " version-0 superblocks.\n",
4344                         mdname(mddev));
4345                 return -EINVAL;
4346         }
4347         if (!mddev->pers->hot_add_disk) {
4348                 printk(KERN_WARNING
4349                         "%s: personality does not support diskops!\n",
4350                         mdname(mddev));
4351                 return -EINVAL;
4352         }
4353
4354         rdev = md_import_device (dev, -1, 0);
4355         if (IS_ERR(rdev)) {
4356                 printk(KERN_WARNING
4357                         "md: error, md_import_device() returned %ld\n",
4358                         PTR_ERR(rdev));
4359                 return -EINVAL;
4360         }
4361
4362         if (mddev->persistent)
4363                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4364         else
4365                 rdev->sb_offset =
4366                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4367
4368         size = calc_dev_size(rdev, mddev->chunk_size);
4369         rdev->size = size;
4370
4371         if (test_bit(Faulty, &rdev->flags)) {
4372                 printk(KERN_WARNING
4373                         "md: can not hot-add faulty %s disk to %s!\n",
4374                         bdevname(rdev->bdev,b), mdname(mddev));
4375                 err = -EINVAL;
4376                 goto abort_export;
4377         }
4378         clear_bit(In_sync, &rdev->flags);
4379         rdev->desc_nr = -1;
4380         rdev->saved_raid_disk = -1;
4381         err = bind_rdev_to_array(rdev, mddev);
4382         if (err)
4383                 goto abort_export;
4384
4385         /*
4386          * The rest should better be atomic, we can have disk failures
4387          * noticed in interrupt contexts ...
4388          */
4389
4390         if (rdev->desc_nr == mddev->max_disks) {
4391                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4392                         mdname(mddev));
4393                 err = -EBUSY;
4394                 goto abort_unbind_export;
4395         }
4396
4397         rdev->raid_disk = -1;
4398
4399         md_update_sb(mddev, 1);
4400
4401         /*
4402          * Kick recovery, maybe this spare has to be added to the
4403          * array immediately.
4404          */
4405         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4406         md_wakeup_thread(mddev->thread);
4407         md_new_event(mddev);
4408         return 0;
4409
4410 abort_unbind_export:
4411         unbind_rdev_from_array(rdev);
4412
4413 abort_export:
4414         export_rdev(rdev);
4415         return err;
4416 }
4417
4418 static int set_bitmap_file(mddev_t *mddev, int fd)
4419 {
4420         int err;
4421
4422         if (mddev->pers) {
4423                 if (!mddev->pers->quiesce)
4424                         return -EBUSY;
4425                 if (mddev->recovery || mddev->sync_thread)
4426                         return -EBUSY;
4427                 /* we should be able to change the bitmap.. */
4428         }
4429
4430
4431         if (fd >= 0) {
4432                 if (mddev->bitmap)
4433                         return -EEXIST; /* cannot add when bitmap is present */
4434                 mddev->bitmap_file = fget(fd);
4435
4436                 if (mddev->bitmap_file == NULL) {
4437                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4438                                mdname(mddev));
4439                         return -EBADF;
4440                 }
4441
4442                 err = deny_bitmap_write_access(mddev->bitmap_file);
4443                 if (err) {
4444                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4445                                mdname(mddev));
4446                         fput(mddev->bitmap_file);
4447                         mddev->bitmap_file = NULL;
4448                         return err;
4449                 }
4450                 mddev->bitmap_offset = 0; /* file overrides offset */
4451         } else if (mddev->bitmap == NULL)
4452                 return -ENOENT; /* cannot remove what isn't there */
4453         err = 0;
4454         if (mddev->pers) {
4455                 mddev->pers->quiesce(mddev, 1);
4456                 if (fd >= 0)
4457                         err = bitmap_create(mddev);
4458                 if (fd < 0 || err) {
4459                         bitmap_destroy(mddev);
4460                         fd = -1; /* make sure to put the file */
4461                 }
4462                 mddev->pers->quiesce(mddev, 0);
4463         }
4464         if (fd < 0) {
4465                 if (mddev->bitmap_file) {
4466                         restore_bitmap_write_access(mddev->bitmap_file);
4467                         fput(mddev->bitmap_file);
4468                 }
4469                 mddev->bitmap_file = NULL;
4470         }
4471
4472         return err;
4473 }
4474
4475 /*
4476  * set_array_info is used two different ways
4477  * The original usage is when creating a new array.
4478  * In this usage, raid_disks is > 0 and it together with
4479  *  level, size, not_persistent,layout,chunksize determine the
4480  *  shape of the array.
4481  *  This will always create an array with a type-0.90.0 superblock.
4482  * The newer usage is when assembling an array.
4483  *  In this case raid_disks will be 0, and the major_version field is
4484  *  use to determine which style super-blocks are to be found on the devices.
4485  *  The minor and patch _version numbers are also kept incase the
4486  *  super_block handler wishes to interpret them.
4487  */
4488 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4489 {
4490
4491         if (info->raid_disks == 0) {
4492                 /* just setting version number for superblock loading */
4493                 if (info->major_version < 0 ||
4494                     info->major_version >= ARRAY_SIZE(super_types) ||
4495                     super_types[info->major_version].name == NULL) {
4496                         /* maybe try to auto-load a module? */
4497                         printk(KERN_INFO
4498                                 "md: superblock version %d not known\n",
4499                                 info->major_version);
4500                         return -EINVAL;
4501                 }
4502                 mddev->major_version = info->major_version;
4503                 mddev->minor_version = info->minor_version;
4504                 mddev->patch_version = info->patch_version;
4505                 mddev->persistent = !info->not_persistent;
4506                 return 0;
4507         }
4508         mddev->major_version = MD_MAJOR_VERSION;
4509         mddev->minor_version = MD_MINOR_VERSION;
4510         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4511         mddev->ctime         = get_seconds();
4512
4513         mddev->level         = info->level;
4514         mddev->clevel[0]     = 0;
4515         mddev->size          = info->size;
4516         mddev->raid_disks    = info->raid_disks;
4517         /* don't set md_minor, it is determined by which /dev/md* was
4518          * openned
4519          */
4520         if (info->state & (1<<MD_SB_CLEAN))
4521                 mddev->recovery_cp = MaxSector;
4522         else
4523                 mddev->recovery_cp = 0;
4524         mddev->persistent    = ! info->not_persistent;
4525         mddev->external      = 0;
4526
4527         mddev->layout        = info->layout;
4528         mddev->chunk_size    = info->chunk_size;
4529
4530         mddev->max_disks     = MD_SB_DISKS;
4531
4532         if (mddev->persistent)
4533                 mddev->flags         = 0;
4534         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4535
4536         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4537         mddev->bitmap_offset = 0;
4538
4539         mddev->reshape_position = MaxSector;
4540
4541         /*
4542          * Generate a 128 bit UUID
4543          */
4544         get_random_bytes(mddev->uuid, 16);
4545
4546         mddev->new_level = mddev->level;
4547         mddev->new_chunk = mddev->chunk_size;
4548         mddev->new_layout = mddev->layout;
4549         mddev->delta_disks = 0;
4550
4551         return 0;
4552 }
4553
4554 static int update_size(mddev_t *mddev, unsigned long size)
4555 {
4556         mdk_rdev_t * rdev;
4557         int rv;
4558         struct list_head *tmp;
4559         int fit = (size == 0);
4560
4561         if (mddev->pers->resize == NULL)
4562                 return -EINVAL;
4563         /* The "size" is the amount of each device that is used.
4564          * This can only make sense for arrays with redundancy.
4565          * linear and raid0 always use whatever space is available
4566          * We can only consider changing the size if no resync
4567          * or reconstruction is happening, and if the new size
4568          * is acceptable. It must fit before the sb_offset or,
4569          * if that is <data_offset, it must fit before the
4570          * size of each device.
4571          * If size is zero, we find the largest size that fits.
4572          */
4573         if (mddev->sync_thread)
4574                 return -EBUSY;
4575         rdev_for_each(rdev, tmp, mddev) {
4576                 sector_t avail;
4577                 avail = rdev->size * 2;
4578
4579                 if (fit && (size == 0 || size > avail/2))
4580                         size = avail/2;
4581                 if (avail < ((sector_t)size << 1))
4582                         return -ENOSPC;
4583         }
4584         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4585         if (!rv) {
4586                 struct block_device *bdev;
4587
4588                 bdev = bdget_disk(mddev->gendisk, 0);
4589                 if (bdev) {
4590                         mutex_lock(&bdev->bd_inode->i_mutex);
4591                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4592                         mutex_unlock(&bdev->bd_inode->i_mutex);
4593                         bdput(bdev);
4594                 }
4595         }
4596         return rv;
4597 }
4598
4599 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4600 {
4601         int rv;
4602         /* change the number of raid disks */
4603         if (mddev->pers->check_reshape == NULL)
4604                 return -EINVAL;
4605         if (raid_disks <= 0 ||
4606             raid_disks >= mddev->max_disks)
4607                 return -EINVAL;
4608         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4609                 return -EBUSY;
4610         mddev->delta_disks = raid_disks - mddev->raid_disks;
4611
4612         rv = mddev->pers->check_reshape(mddev);
4613         return rv;
4614 }
4615
4616
4617 /*
4618  * update_array_info is used to change the configuration of an
4619  * on-line array.
4620  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4621  * fields in the info are checked against the array.
4622  * Any differences that cannot be handled will cause an error.
4623  * Normally, only one change can be managed at a time.
4624  */
4625 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4626 {
4627         int rv = 0;
4628         int cnt = 0;
4629         int state = 0;
4630
4631         /* calculate expected state,ignoring low bits */
4632         if (mddev->bitmap && mddev->bitmap_offset)
4633                 state |= (1 << MD_SB_BITMAP_PRESENT);
4634
4635         if (mddev->major_version != info->major_version ||
4636             mddev->minor_version != info->minor_version ||
4637 /*          mddev->patch_version != info->patch_version || */
4638             mddev->ctime         != info->ctime         ||
4639             mddev->level         != info->level         ||
4640 /*          mddev->layout        != info->layout        || */
4641             !mddev->persistent   != info->not_persistent||
4642             mddev->chunk_size    != info->chunk_size    ||
4643             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4644             ((state^info->state) & 0xfffffe00)
4645                 )
4646                 return -EINVAL;
4647         /* Check there is only one change */
4648         if (info->size >= 0 && mddev->size != info->size) cnt++;
4649         if (mddev->raid_disks != info->raid_disks) cnt++;
4650         if (mddev->layout != info->layout) cnt++;
4651         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4652         if (cnt == 0) return 0;
4653         if (cnt > 1) return -EINVAL;
4654
4655         if (mddev->layout != info->layout) {
4656                 /* Change layout
4657                  * we don't need to do anything at the md level, the
4658                  * personality will take care of it all.
4659                  */
4660                 if (mddev->pers->reconfig == NULL)
4661                         return -EINVAL;
4662                 else
4663                         return mddev->pers->reconfig(mddev, info->layout, -1);
4664         }
4665         if (info->size >= 0 && mddev->size != info->size)
4666                 rv = update_size(mddev, info->size);
4667
4668         if (mddev->raid_disks    != info->raid_disks)
4669                 rv = update_raid_disks(mddev, info->raid_disks);
4670
4671         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4672                 if (mddev->pers->quiesce == NULL)
4673                         return -EINVAL;
4674                 if (mddev->recovery || mddev->sync_thread)
4675                         return -EBUSY;
4676                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4677                         /* add the bitmap */
4678                         if (mddev->bitmap)
4679                                 return -EEXIST;
4680                         if (mddev->default_bitmap_offset == 0)
4681                                 return -EINVAL;
4682                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4683                         mddev->pers->quiesce(mddev, 1);
4684                         rv = bitmap_create(mddev);
4685                         if (rv)
4686                                 bitmap_destroy(mddev);
4687                         mddev->pers->quiesce(mddev, 0);
4688                 } else {
4689                         /* remove the bitmap */
4690                         if (!mddev->bitmap)
4691                                 return -ENOENT;
4692                         if (mddev->bitmap->file)
4693                                 return -EINVAL;
4694                         mddev->pers->quiesce(mddev, 1);
4695                         bitmap_destroy(mddev);
4696                         mddev->pers->quiesce(mddev, 0);
4697                         mddev->bitmap_offset = 0;
4698                 }
4699         }
4700         md_update_sb(mddev, 1);
4701         return rv;
4702 }
4703
4704 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4705 {
4706         mdk_rdev_t *rdev;
4707
4708         if (mddev->pers == NULL)
4709                 return -ENODEV;
4710
4711         rdev = find_rdev(mddev, dev);
4712         if (!rdev)
4713                 return -ENODEV;
4714
4715         md_error(mddev, rdev);
4716         return 0;
4717 }
4718
4719 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4720 {
4721         mddev_t *mddev = bdev->bd_disk->private_data;
4722
4723         geo->heads = 2;
4724         geo->sectors = 4;
4725         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4726         return 0;
4727 }
4728
4729 static int md_ioctl(struct inode *inode, struct file *file,
4730                         unsigned int cmd, unsigned long arg)
4731 {
4732         int err = 0;
4733         void __user *argp = (void __user *)arg;
4734         mddev_t *mddev = NULL;
4735
4736         if (!capable(CAP_SYS_ADMIN))
4737                 return -EACCES;
4738
4739         /*
4740          * Commands dealing with the RAID driver but not any
4741          * particular array:
4742          */
4743         switch (cmd)
4744         {
4745                 case RAID_VERSION:
4746                         err = get_version(argp);
4747                         goto done;
4748
4749                 case PRINT_RAID_DEBUG:
4750                         err = 0;
4751                         md_print_devices();
4752                         goto done;
4753
4754 #ifndef MODULE
4755                 case RAID_AUTORUN:
4756                         err = 0;
4757                         autostart_arrays(arg);
4758                         goto done;
4759 #endif
4760                 default:;
4761         }
4762
4763         /*
4764          * Commands creating/starting a new array:
4765          */
4766
4767         mddev = inode->i_bdev->bd_disk->private_data;
4768
4769         if (!mddev) {
4770                 BUG();
4771                 goto abort;
4772         }
4773
4774         err = mddev_lock(mddev);
4775         if (err) {
4776                 printk(KERN_INFO
4777                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4778                         err, cmd);
4779                 goto abort;
4780         }
4781
4782         switch (cmd)
4783         {
4784                 case SET_ARRAY_INFO:
4785                         {
4786                                 mdu_array_info_t info;
4787                                 if (!arg)
4788                                         memset(&info, 0, sizeof(info));
4789                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4790                                         err = -EFAULT;
4791                                         goto abort_unlock;
4792                                 }
4793                                 if (mddev->pers) {
4794                                         err = update_array_info(mddev, &info);
4795                                         if (err) {
4796                                                 printk(KERN_WARNING "md: couldn't update"
4797                                                        " array info. %d\n", err);
4798                                                 goto abort_unlock;
4799                                         }
4800                                         goto done_unlock;
4801                                 }
4802                                 if (!list_empty(&mddev->disks)) {
4803                                         printk(KERN_WARNING
4804                                                "md: array %s already has disks!\n",
4805                                                mdname(mddev));
4806                                         err = -EBUSY;
4807                                         goto abort_unlock;
4808                                 }
4809                                 if (mddev->raid_disks) {
4810                                         printk(KERN_WARNING
4811                                                "md: array %s already initialised!\n",
4812                                                mdname(mddev));
4813                                         err = -EBUSY;
4814                                         goto abort_unlock;
4815                                 }
4816                                 err = set_array_info(mddev, &info);
4817                                 if (err) {
4818                                         printk(KERN_WARNING "md: couldn't set"
4819                                                " array info. %d\n", err);
4820                                         goto abort_unlock;
4821                                 }
4822                         }
4823                         goto done_unlock;
4824
4825                 default:;
4826         }
4827
4828         /*
4829          * Commands querying/configuring an existing array:
4830          */
4831         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4832          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4833         if ((!mddev->raid_disks && !mddev->external)
4834             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4835             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4836             && cmd != GET_BITMAP_FILE) {
4837                 err = -ENODEV;
4838                 goto abort_unlock;
4839         }
4840
4841         /*
4842          * Commands even a read-only array can execute:
4843          */
4844         switch (cmd)
4845         {
4846                 case GET_ARRAY_INFO:
4847                         err = get_array_info(mddev, argp);
4848                         goto done_unlock;
4849
4850                 case GET_BITMAP_FILE:
4851                         err = get_bitmap_file(mddev, argp);
4852                         goto done_unlock;
4853
4854                 case GET_DISK_INFO:
4855                         err = get_disk_info(mddev, argp);
4856                         goto done_unlock;
4857
4858                 case RESTART_ARRAY_RW:
4859                         err = restart_array(mddev);
4860                         goto done_unlock;
4861
4862                 case STOP_ARRAY:
4863                         err = do_md_stop (mddev, 0);
4864                         goto done_unlock;
4865
4866                 case STOP_ARRAY_RO:
4867                         err = do_md_stop (mddev, 1);
4868                         goto done_unlock;
4869
4870         /*
4871          * We have a problem here : there is no easy way to give a CHS
4872          * virtual geometry. We currently pretend that we have a 2 heads
4873          * 4 sectors (with a BIG number of cylinders...). This drives
4874          * dosfs just mad... ;-)
4875          */
4876         }
4877
4878         /*
4879          * The remaining ioctls are changing the state of the
4880          * superblock, so we do not allow them on read-only arrays.
4881          * However non-MD ioctls (e.g. get-size) will still come through
4882          * here and hit the 'default' below, so only disallow
4883          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4884          */
4885         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4886             mddev->ro && mddev->pers) {
4887                 if (mddev->ro == 2) {
4888                         mddev->ro = 0;
4889                         sysfs_notify(&mddev->kobj, NULL, "array_state");
4890                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4891                         md_wakeup_thread(mddev->thread);
4892
4893                 } else {
4894                         err = -EROFS;
4895                         goto abort_unlock;
4896                 }
4897         }
4898
4899         switch (cmd)
4900         {
4901                 case ADD_NEW_DISK:
4902                 {
4903                         mdu_disk_info_t info;
4904                         if (copy_from_user(&info, argp, sizeof(info)))
4905                                 err = -EFAULT;
4906                         else
4907                                 err = add_new_disk(mddev, &info);
4908                         goto done_unlock;
4909                 }
4910
4911                 case HOT_REMOVE_DISK:
4912                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4913                         goto done_unlock;
4914
4915                 case HOT_ADD_DISK:
4916                         err = hot_add_disk(mddev, new_decode_dev(arg));
4917                         goto done_unlock;
4918
4919                 case SET_DISK_FAULTY:
4920                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4921                         goto done_unlock;
4922
4923                 case RUN_ARRAY:
4924                         err = do_md_run (mddev);
4925                         goto done_unlock;
4926
4927                 case SET_BITMAP_FILE:
4928                         err = set_bitmap_file(mddev, (int)arg);
4929                         goto done_unlock;
4930
4931                 default:
4932                         err = -EINVAL;
4933                         goto abort_unlock;
4934         }
4935
4936 done_unlock:
4937 abort_unlock:
4938         mddev_unlock(mddev);
4939
4940         return err;
4941 done:
4942         if (err)
4943                 MD_BUG();
4944 abort:
4945         return err;
4946 }
4947
4948 static int md_open(struct inode *inode, struct file *file)
4949 {
4950         /*
4951          * Succeed if we can lock the mddev, which confirms that
4952          * it isn't being stopped right now.
4953          */
4954         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4955         int err;
4956
4957         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4958                 goto out;
4959
4960         err = 0;
4961         mddev_get(mddev);
4962         mddev_unlock(mddev);
4963
4964         check_disk_change(inode->i_bdev);
4965  out:
4966         return err;
4967 }
4968
4969 static int md_release(struct inode *inode, struct file * file)
4970 {
4971         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4972
4973         BUG_ON(!mddev);
4974         mddev_put(mddev);
4975
4976         return 0;
4977 }
4978
4979 static int md_media_changed(struct gendisk *disk)
4980 {
4981         mddev_t *mddev = disk->private_data;
4982
4983         return mddev->changed;
4984 }
4985
4986 static int md_revalidate(struct gendisk *disk)
4987 {
4988         mddev_t *mddev = disk->private_data;
4989
4990         mddev->changed = 0;
4991         return 0;
4992 }
4993 static struct block_device_operations md_fops =
4994 {
4995         .owner          = THIS_MODULE,
4996         .open           = md_open,
4997         .release        = md_release,
4998         .ioctl          = md_ioctl,
4999         .getgeo         = md_getgeo,
5000         .media_changed  = md_media_changed,
5001         .revalidate_disk= md_revalidate,
5002 };
5003
5004 static int md_thread(void * arg)
5005 {
5006         mdk_thread_t *thread = arg;
5007
5008         /*
5009          * md_thread is a 'system-thread', it's priority should be very
5010          * high. We avoid resource deadlocks individually in each
5011          * raid personality. (RAID5 does preallocation) We also use RR and
5012          * the very same RT priority as kswapd, thus we will never get
5013          * into a priority inversion deadlock.
5014          *
5015          * we definitely have to have equal or higher priority than
5016          * bdflush, otherwise bdflush will deadlock if there are too
5017          * many dirty RAID5 blocks.
5018          */
5019
5020         allow_signal(SIGKILL);
5021         while (!kthread_should_stop()) {
5022
5023                 /* We need to wait INTERRUPTIBLE so that
5024                  * we don't add to the load-average.
5025                  * That means we need to be sure no signals are
5026                  * pending
5027                  */
5028                 if (signal_pending(current))
5029                         flush_signals(current);
5030
5031                 wait_event_interruptible_timeout
5032                         (thread->wqueue,
5033                          test_bit(THREAD_WAKEUP, &thread->flags)
5034                          || kthread_should_stop(),
5035                          thread->timeout);
5036
5037                 clear_bit(THREAD_WAKEUP, &thread->flags);
5038
5039                 thread->run(thread->mddev);
5040         }
5041
5042         return 0;
5043 }
5044
5045 void md_wakeup_thread(mdk_thread_t *thread)
5046 {
5047         if (thread) {
5048                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5049                 set_bit(THREAD_WAKEUP, &thread->flags);
5050                 wake_up(&thread->wqueue);
5051         }
5052 }
5053
5054 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5055                                  const char *name)
5056 {
5057         mdk_thread_t *thread;
5058
5059         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5060         if (!thread)
5061                 return NULL;
5062
5063         init_waitqueue_head(&thread->wqueue);
5064
5065         thread->run = run;
5066         thread->mddev = mddev;
5067         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5068         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5069         if (IS_ERR(thread->tsk)) {
5070                 kfree(thread);
5071                 return NULL;
5072         }
5073         return thread;
5074 }
5075
5076 void md_unregister_thread(mdk_thread_t *thread)
5077 {
5078         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5079
5080         kthread_stop(thread->tsk);
5081         kfree(thread);
5082 }
5083
5084 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5085 {
5086         if (!mddev) {
5087                 MD_BUG();
5088                 return;
5089         }
5090
5091         if (!rdev || test_bit(Faulty, &rdev->flags))
5092                 return;
5093
5094         if (mddev->external)
5095                 set_bit(Blocked, &rdev->flags);
5096 /*
5097         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5098                 mdname(mddev),
5099                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5100                 __builtin_return_address(0),__builtin_return_address(1),
5101                 __builtin_return_address(2),__builtin_return_address(3));
5102 */
5103         if (!mddev->pers)
5104                 return;
5105         if (!mddev->pers->error_handler)
5106                 return;
5107         mddev->pers->error_handler(mddev,rdev);
5108         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5109         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5110         md_wakeup_thread(mddev->thread);
5111         md_new_event_inintr(mddev);
5112 }
5113
5114 /* seq_file implementation /proc/mdstat */
5115
5116 static void status_unused(struct seq_file *seq)
5117 {
5118         int i = 0;
5119         mdk_rdev_t *rdev;
5120         struct list_head *tmp;
5121
5122         seq_printf(seq, "unused devices: ");
5123
5124         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5125                 char b[BDEVNAME_SIZE];
5126                 i++;
5127                 seq_printf(seq, "%s ",
5128                               bdevname(rdev->bdev,b));
5129         }
5130         if (!i)
5131                 seq_printf(seq, "<none>");
5132
5133         seq_printf(seq, "\n");
5134 }
5135
5136
5137 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5138 {
5139         sector_t max_blocks, resync, res;
5140         unsigned long dt, db, rt;
5141         int scale;
5142         unsigned int per_milli;
5143
5144         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5145
5146         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5147                 max_blocks = mddev->resync_max_sectors >> 1;
5148         else
5149                 max_blocks = mddev->size;
5150
5151         /*
5152          * Should not happen.
5153          */
5154         if (!max_blocks) {
5155                 MD_BUG();
5156                 return;
5157         }
5158         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5159          * in a sector_t, and (max_blocks>>scale) will fit in a
5160          * u32, as those are the requirements for sector_div.
5161          * Thus 'scale' must be at least 10
5162          */
5163         scale = 10;
5164         if (sizeof(sector_t) > sizeof(unsigned long)) {
5165                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5166                         scale++;
5167         }
5168         res = (resync>>scale)*1000;
5169         sector_div(res, (u32)((max_blocks>>scale)+1));
5170
5171         per_milli = res;
5172         {
5173                 int i, x = per_milli/50, y = 20-x;
5174                 seq_printf(seq, "[");
5175                 for (i = 0; i < x; i++)
5176                         seq_printf(seq, "=");
5177                 seq_printf(seq, ">");
5178                 for (i = 0; i < y; i++)
5179                         seq_printf(seq, ".");
5180                 seq_printf(seq, "] ");
5181         }
5182         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5183                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5184                     "reshape" :
5185                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5186                      "check" :
5187                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5188                       "resync" : "recovery"))),
5189                    per_milli/10, per_milli % 10,
5190                    (unsigned long long) resync,
5191                    (unsigned long long) max_blocks);
5192
5193         /*
5194          * We do not want to overflow, so the order of operands and
5195          * the * 100 / 100 trick are important. We do a +1 to be
5196          * safe against division by zero. We only estimate anyway.
5197          *
5198          * dt: time from mark until now
5199          * db: blocks written from mark until now
5200          * rt: remaining time
5201          */
5202         dt = ((jiffies - mddev->resync_mark) / HZ);
5203         if (!dt) dt++;
5204         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5205                 - mddev->resync_mark_cnt;
5206         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5207
5208         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5209
5210         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5211 }
5212
5213 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5214 {
5215         struct list_head *tmp;
5216         loff_t l = *pos;
5217         mddev_t *mddev;
5218
5219         if (l >= 0x10000)
5220                 return NULL;
5221         if (!l--)
5222                 /* header */
5223                 return (void*)1;
5224
5225         spin_lock(&all_mddevs_lock);
5226         list_for_each(tmp,&all_mddevs)
5227                 if (!l--) {
5228                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5229                         mddev_get(mddev);
5230                         spin_unlock(&all_mddevs_lock);
5231                         return mddev;
5232                 }
5233         spin_unlock(&all_mddevs_lock);
5234         if (!l--)
5235                 return (void*)2;/* tail */
5236         return NULL;
5237 }
5238
5239 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5240 {
5241         struct list_head *tmp;
5242         mddev_t *next_mddev, *mddev = v;
5243
5244         ++*pos;
5245         if (v == (void*)2)
5246                 return NULL;
5247
5248         spin_lock(&all_mddevs_lock);
5249         if (v == (void*)1)
5250                 tmp = all_mddevs.next;
5251         else
5252                 tmp = mddev->all_mddevs.next;
5253         if (tmp != &all_mddevs)
5254                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5255         else {
5256                 next_mddev = (void*)2;
5257                 *pos = 0x10000;
5258         }
5259         spin_unlock(&all_mddevs_lock);
5260
5261         if (v != (void*)1)
5262                 mddev_put(mddev);
5263         return next_mddev;
5264
5265 }
5266
5267 static void md_seq_stop(struct seq_file *seq, void *v)
5268 {
5269         mddev_t *mddev = v;
5270
5271         if (mddev && v != (void*)1 && v != (void*)2)
5272                 mddev_put(mddev);
5273 }
5274
5275 struct mdstat_info {
5276         int event;
5277 };
5278
5279 static int md_seq_show(struct seq_file *seq, void *v)
5280 {
5281         mddev_t *mddev = v;
5282         sector_t size;
5283         struct list_head *tmp2;
5284         mdk_rdev_t *rdev;
5285         struct mdstat_info *mi = seq->private;
5286         struct bitmap *bitmap;
5287
5288         if (v == (void*)1) {
5289                 struct mdk_personality *pers;
5290                 seq_printf(seq, "Personalities : ");
5291                 spin_lock(&pers_lock);
5292                 list_for_each_entry(pers, &pers_list, list)
5293                         seq_printf(seq, "[%s] ", pers->name);
5294
5295                 spin_unlock(&pers_lock);
5296                 seq_printf(seq, "\n");
5297                 mi->event = atomic_read(&md_event_count);
5298                 return 0;
5299         }
5300         if (v == (void*)2) {
5301                 status_unused(seq);
5302                 return 0;
5303         }
5304
5305         if (mddev_lock(mddev) < 0)
5306                 return -EINTR;
5307
5308         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5309                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5310                                                 mddev->pers ? "" : "in");
5311                 if (mddev->pers) {
5312                         if (mddev->ro==1)
5313                                 seq_printf(seq, " (read-only)");
5314                         if (mddev->ro==2)
5315                                 seq_printf(seq, " (auto-read-only)");
5316                         seq_printf(seq, " %s", mddev->pers->name);
5317                 }
5318
5319                 size = 0;
5320                 rdev_for_each(rdev, tmp2, mddev) {
5321                         char b[BDEVNAME_SIZE];
5322                         seq_printf(seq, " %s[%d]",
5323                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5324                         if (test_bit(WriteMostly, &rdev->flags))
5325                                 seq_printf(seq, "(W)");
5326                         if (test_bit(Faulty, &rdev->flags)) {
5327                                 seq_printf(seq, "(F)");
5328                                 continue;
5329                         } else if (rdev->raid_disk < 0)
5330                                 seq_printf(seq, "(S)"); /* spare */
5331                         size += rdev->size;
5332                 }
5333
5334                 if (!list_empty(&mddev->disks)) {
5335                         if (mddev->pers)
5336                                 seq_printf(seq, "\n      %llu blocks",
5337                                         (unsigned long long)mddev->array_size);
5338                         else
5339                                 seq_printf(seq, "\n      %llu blocks",
5340                                         (unsigned long long)size);
5341                 }
5342                 if (mddev->persistent) {
5343                         if (mddev->major_version != 0 ||
5344                             mddev->minor_version != 90) {
5345                                 seq_printf(seq," super %d.%d",
5346                                            mddev->major_version,
5347                                            mddev->minor_version);
5348                         }
5349                 } else if (mddev->external)
5350                         seq_printf(seq, " super external:%s",
5351                                    mddev->metadata_type);
5352                 else
5353                         seq_printf(seq, " super non-persistent");
5354
5355                 if (mddev->pers) {
5356                         mddev->pers->status (seq, mddev);
5357                         seq_printf(seq, "\n      ");
5358                         if (mddev->pers->sync_request) {
5359                                 if (mddev->curr_resync > 2) {
5360                                         status_resync (seq, mddev);
5361                                         seq_printf(seq, "\n      ");
5362                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5363                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5364                                 else if (mddev->recovery_cp < MaxSector)
5365                                         seq_printf(seq, "\tresync=PENDING\n      ");
5366                         }
5367                 } else
5368                         seq_printf(seq, "\n       ");
5369
5370                 if ((bitmap = mddev->bitmap)) {
5371                         unsigned long chunk_kb;
5372                         unsigned long flags;
5373                         spin_lock_irqsave(&bitmap->lock, flags);
5374                         chunk_kb = bitmap->chunksize >> 10;
5375                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5376                                 "%lu%s chunk",
5377                                 bitmap->pages - bitmap->missing_pages,
5378                                 bitmap->pages,
5379                                 (bitmap->pages - bitmap->missing_pages)
5380                                         << (PAGE_SHIFT - 10),
5381                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5382                                 chunk_kb ? "KB" : "B");
5383                         if (bitmap->file) {
5384                                 seq_printf(seq, ", file: ");
5385                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5386                         }
5387
5388                         seq_printf(seq, "\n");
5389                         spin_unlock_irqrestore(&bitmap->lock, flags);
5390                 }
5391
5392                 seq_printf(seq, "\n");
5393         }
5394         mddev_unlock(mddev);
5395
5396         return 0;
5397 }
5398
5399 static struct seq_operations md_seq_ops = {
5400         .start  = md_seq_start,
5401         .next   = md_seq_next,
5402         .stop   = md_seq_stop,
5403         .show   = md_seq_show,
5404 };
5405
5406 static int md_seq_open(struct inode *inode, struct file *file)
5407 {
5408         int error;
5409         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5410         if (mi == NULL)
5411                 return -ENOMEM;
5412
5413         error = seq_open(file, &md_seq_ops);
5414         if (error)
5415                 kfree(mi);
5416         else {
5417                 struct seq_file *p = file->private_data;
5418                 p->private = mi;
5419                 mi->event = atomic_read(&md_event_count);
5420         }
5421         return error;
5422 }
5423
5424 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5425 {
5426         struct seq_file *m = filp->private_data;
5427         struct mdstat_info *mi = m->private;
5428         int mask;
5429
5430         poll_wait(filp, &md_event_waiters, wait);
5431
5432         /* always allow read */
5433         mask = POLLIN | POLLRDNORM;
5434
5435         if (mi->event != atomic_read(&md_event_count))
5436                 mask |= POLLERR | POLLPRI;
5437         return mask;
5438 }
5439
5440 static const struct file_operations md_seq_fops = {
5441         .owner          = THIS_MODULE,
5442         .open           = md_seq_open,
5443         .read           = seq_read,
5444         .llseek         = seq_lseek,
5445         .release        = seq_release_private,
5446         .poll           = mdstat_poll,
5447 };
5448
5449 int register_md_personality(struct mdk_personality *p)
5450 {
5451         spin_lock(&pers_lock);
5452         list_add_tail(&p->list, &pers_list);
5453         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5454         spin_unlock(&pers_lock);
5455         return 0;
5456 }
5457
5458 int unregister_md_personality(struct mdk_personality *p)
5459 {
5460         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5461         spin_lock(&pers_lock);
5462         list_del_init(&p->list);
5463         spin_unlock(&pers_lock);
5464         return 0;
5465 }
5466
5467 static int is_mddev_idle(mddev_t *mddev)
5468 {
5469         mdk_rdev_t * rdev;
5470         struct list_head *tmp;
5471         int idle;
5472         long curr_events;
5473
5474         idle = 1;
5475         rdev_for_each(rdev, tmp, mddev) {
5476                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5477                 curr_events = disk_stat_read(disk, sectors[0]) +
5478                                 disk_stat_read(disk, sectors[1]) -
5479                                 atomic_read(&disk->sync_io);
5480                 /* sync IO will cause sync_io to increase before the disk_stats
5481                  * as sync_io is counted when a request starts, and
5482                  * disk_stats is counted when it completes.
5483                  * So resync activity will cause curr_events to be smaller than
5484                  * when there was no such activity.
5485                  * non-sync IO will cause disk_stat to increase without
5486                  * increasing sync_io so curr_events will (eventually)
5487                  * be larger than it was before.  Once it becomes
5488                  * substantially larger, the test below will cause
5489                  * the array to appear non-idle, and resync will slow
5490                  * down.
5491                  * If there is a lot of outstanding resync activity when
5492                  * we set last_event to curr_events, then all that activity
5493                  * completing might cause the array to appear non-idle
5494                  * and resync will be slowed down even though there might
5495                  * not have been non-resync activity.  This will only
5496                  * happen once though.  'last_events' will soon reflect
5497                  * the state where there is little or no outstanding
5498                  * resync requests, and further resync activity will
5499                  * always make curr_events less than last_events.
5500                  *
5501                  */
5502                 if (curr_events - rdev->last_events > 4096) {
5503                         rdev->last_events = curr_events;
5504                         idle = 0;
5505                 }
5506         }
5507         return idle;
5508 }
5509
5510 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5511 {
5512         /* another "blocks" (512byte) blocks have been synced */
5513         atomic_sub(blocks, &mddev->recovery_active);
5514         wake_up(&mddev->recovery_wait);
5515         if (!ok) {
5516                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5517                 md_wakeup_thread(mddev->thread);
5518                 // stop recovery, signal do_sync ....
5519         }
5520 }
5521
5522
5523 /* md_write_start(mddev, bi)
5524  * If we need to update some array metadata (e.g. 'active' flag
5525  * in superblock) before writing, schedule a superblock update
5526  * and wait for it to complete.
5527  */
5528 void md_write_start(mddev_t *mddev, struct bio *bi)
5529 {
5530         int did_change = 0;
5531         if (bio_data_dir(bi) != WRITE)
5532                 return;
5533
5534         BUG_ON(mddev->ro == 1);
5535         if (mddev->ro == 2) {
5536                 /* need to switch to read/write */
5537                 mddev->ro = 0;
5538                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5539                 md_wakeup_thread(mddev->thread);
5540                 md_wakeup_thread(mddev->sync_thread);
5541                 did_change = 1;
5542         }
5543         atomic_inc(&mddev->writes_pending);
5544         if (mddev->safemode == 1)
5545                 mddev->safemode = 0;
5546         if (mddev->in_sync) {
5547                 spin_lock_irq(&mddev->write_lock);
5548                 if (mddev->in_sync) {
5549                         mddev->in_sync = 0;
5550                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5551                         md_wakeup_thread(mddev->thread);
5552                         did_change = 1;
5553                 }
5554                 spin_unlock_irq(&mddev->write_lock);
5555         }
5556         if (did_change)
5557                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5558         wait_event(mddev->sb_wait,
5559                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5560                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5561 }
5562
5563 void md_write_end(mddev_t *mddev)
5564 {
5565         if (atomic_dec_and_test(&mddev->writes_pending)) {
5566                 if (mddev->safemode == 2)
5567                         md_wakeup_thread(mddev->thread);
5568                 else if (mddev->safemode_delay)
5569                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5570         }
5571 }
5572
5573 /* md_allow_write(mddev)
5574  * Calling this ensures that the array is marked 'active' so that writes
5575  * may proceed without blocking.  It is important to call this before
5576  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5577  * Must be called with mddev_lock held.
5578  */
5579 void md_allow_write(mddev_t *mddev)
5580 {
5581         if (!mddev->pers)
5582                 return;
5583         if (mddev->ro)
5584                 return;
5585         if (!mddev->pers->sync_request)
5586                 return;
5587
5588         spin_lock_irq(&mddev->write_lock);
5589         if (mddev->in_sync) {
5590                 mddev->in_sync = 0;
5591                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5592                 if (mddev->safemode_delay &&
5593                     mddev->safemode == 0)
5594                         mddev->safemode = 1;
5595                 spin_unlock_irq(&mddev->write_lock);
5596                 md_update_sb(mddev, 0);
5597
5598                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5599                 /* wait for the dirty state to be recorded in the metadata */
5600                 wait_event(mddev->sb_wait,
5601                            !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5602                            !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5603         } else
5604                 spin_unlock_irq(&mddev->write_lock);
5605 }
5606 EXPORT_SYMBOL_GPL(md_allow_write);
5607
5608 #define SYNC_MARKS      10
5609 #define SYNC_MARK_STEP  (3*HZ)
5610 void md_do_sync(mddev_t *mddev)
5611 {
5612         mddev_t *mddev2;
5613         unsigned int currspeed = 0,
5614                  window;
5615         sector_t max_sectors,j, io_sectors;
5616         unsigned long mark[SYNC_MARKS];
5617         sector_t mark_cnt[SYNC_MARKS];
5618         int last_mark,m;
5619         struct list_head *tmp;
5620         sector_t last_check;
5621         int skipped = 0;
5622         struct list_head *rtmp;
5623         mdk_rdev_t *rdev;
5624         char *desc;
5625
5626         /* just incase thread restarts... */
5627         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5628                 return;
5629         if (mddev->ro) /* never try to sync a read-only array */
5630                 return;
5631
5632         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5633                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5634                         desc = "data-check";
5635                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5636                         desc = "requested-resync";
5637                 else
5638                         desc = "resync";
5639         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5640                 desc = "reshape";
5641         else
5642                 desc = "recovery";
5643
5644         /* we overload curr_resync somewhat here.
5645          * 0 == not engaged in resync at all
5646          * 2 == checking that there is no conflict with another sync
5647          * 1 == like 2, but have yielded to allow conflicting resync to
5648          *              commense
5649          * other == active in resync - this many blocks
5650          *
5651          * Before starting a resync we must have set curr_resync to
5652          * 2, and then checked that every "conflicting" array has curr_resync
5653          * less than ours.  When we find one that is the same or higher
5654          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5655          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5656          * This will mean we have to start checking from the beginning again.
5657          *
5658          */
5659
5660         do {
5661                 mddev->curr_resync = 2;
5662
5663         try_again:
5664                 if (kthread_should_stop()) {
5665                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5666                         goto skip;
5667                 }
5668                 for_each_mddev(mddev2, tmp) {
5669                         if (mddev2 == mddev)
5670                                 continue;
5671                         if (!mddev->parallel_resync
5672                         &&  mddev2->curr_resync
5673                         &&  match_mddev_units(mddev, mddev2)) {
5674                                 DEFINE_WAIT(wq);
5675                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5676                                         /* arbitrarily yield */
5677                                         mddev->curr_resync = 1;
5678                                         wake_up(&resync_wait);
5679                                 }
5680                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5681                                         /* no need to wait here, we can wait the next
5682                                          * time 'round when curr_resync == 2
5683                                          */
5684                                         continue;
5685                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5686                                 if (!kthread_should_stop() &&
5687                                     mddev2->curr_resync >= mddev->curr_resync) {
5688                                         printk(KERN_INFO "md: delaying %s of %s"
5689                                                " until %s has finished (they"
5690                                                " share one or more physical units)\n",
5691                                                desc, mdname(mddev), mdname(mddev2));
5692                                         mddev_put(mddev2);
5693                                         schedule();
5694                                         finish_wait(&resync_wait, &wq);
5695                                         goto try_again;
5696                                 }
5697                                 finish_wait(&resync_wait, &wq);
5698                         }
5699                 }
5700         } while (mddev->curr_resync < 2);
5701
5702         j = 0;
5703         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5704                 /* resync follows the size requested by the personality,
5705                  * which defaults to physical size, but can be virtual size
5706                  */
5707                 max_sectors = mddev->resync_max_sectors;
5708                 mddev->resync_mismatches = 0;
5709                 /* we don't use the checkpoint if there's a bitmap */
5710                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5711                         j = mddev->resync_min;
5712                 else if (!mddev->bitmap)
5713                         j = mddev->recovery_cp;
5714
5715         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5716                 max_sectors = mddev->size << 1;
5717         else {
5718                 /* recovery follows the physical size of devices */
5719                 max_sectors = mddev->size << 1;
5720                 j = MaxSector;
5721                 rdev_for_each(rdev, rtmp, mddev)
5722                         if (rdev->raid_disk >= 0 &&
5723                             !test_bit(Faulty, &rdev->flags) &&
5724                             !test_bit(In_sync, &rdev->flags) &&
5725                             rdev->recovery_offset < j)
5726                                 j = rdev->recovery_offset;
5727         }
5728
5729         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5730         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5731                 " %d KB/sec/disk.\n", speed_min(mddev));
5732         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5733                "(but not more than %d KB/sec) for %s.\n",
5734                speed_max(mddev), desc);
5735
5736         is_mddev_idle(mddev); /* this also initializes IO event counters */
5737
5738         io_sectors = 0;
5739         for (m = 0; m < SYNC_MARKS; m++) {
5740                 mark[m] = jiffies;
5741                 mark_cnt[m] = io_sectors;
5742         }
5743         last_mark = 0;
5744         mddev->resync_mark = mark[last_mark];
5745         mddev->resync_mark_cnt = mark_cnt[last_mark];
5746
5747         /*
5748          * Tune reconstruction:
5749          */
5750         window = 32*(PAGE_SIZE/512);
5751         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5752                 window/2,(unsigned long long) max_sectors/2);
5753
5754         atomic_set(&mddev->recovery_active, 0);
5755         last_check = 0;
5756
5757         if (j>2) {
5758                 printk(KERN_INFO
5759                        "md: resuming %s of %s from checkpoint.\n",
5760                        desc, mdname(mddev));
5761                 mddev->curr_resync = j;
5762         }
5763
5764         while (j < max_sectors) {
5765                 sector_t sectors;
5766
5767                 skipped = 0;
5768                 if (j >= mddev->resync_max) {
5769                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5770                         wait_event(mddev->recovery_wait,
5771                                    mddev->resync_max > j
5772                                    || kthread_should_stop());
5773                 }
5774                 if (kthread_should_stop())
5775                         goto interrupted;
5776                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5777                                                   currspeed < speed_min(mddev));
5778                 if (sectors == 0) {
5779                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5780                         goto out;
5781                 }
5782
5783                 if (!skipped) { /* actual IO requested */
5784                         io_sectors += sectors;
5785                         atomic_add(sectors, &mddev->recovery_active);
5786                 }
5787
5788                 j += sectors;
5789                 if (j>1) mddev->curr_resync = j;
5790                 mddev->curr_mark_cnt = io_sectors;
5791                 if (last_check == 0)
5792                         /* this is the earliers that rebuilt will be
5793                          * visible in /proc/mdstat
5794                          */
5795                         md_new_event(mddev);
5796
5797                 if (last_check + window > io_sectors || j == max_sectors)
5798                         continue;
5799
5800                 last_check = io_sectors;
5801
5802                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5803                         break;
5804
5805         repeat:
5806                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5807                         /* step marks */
5808                         int next = (last_mark+1) % SYNC_MARKS;
5809
5810                         mddev->resync_mark = mark[next];
5811                         mddev->resync_mark_cnt = mark_cnt[next];
5812                         mark[next] = jiffies;
5813                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5814                         last_mark = next;
5815                 }
5816
5817
5818                 if (kthread_should_stop())
5819                         goto interrupted;
5820
5821
5822                 /*
5823                  * this loop exits only if either when we are slower than
5824                  * the 'hard' speed limit, or the system was IO-idle for
5825                  * a jiffy.
5826                  * the system might be non-idle CPU-wise, but we only care
5827                  * about not overloading the IO subsystem. (things like an
5828                  * e2fsck being done on the RAID array should execute fast)
5829                  */
5830                 blk_unplug(mddev->queue);
5831                 cond_resched();
5832
5833                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5834                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5835
5836                 if (currspeed > speed_min(mddev)) {
5837                         if ((currspeed > speed_max(mddev)) ||
5838                                         !is_mddev_idle(mddev)) {
5839                                 msleep(500);
5840                                 goto repeat;
5841                         }
5842                 }
5843         }
5844         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5845         /*
5846          * this also signals 'finished resyncing' to md_stop
5847          */
5848  out:
5849         blk_unplug(mddev->queue);
5850
5851         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5852
5853         /* tell personality that we are finished */
5854         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5855
5856         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5857             mddev->curr_resync > 2) {
5858                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5859                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5860                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5861                                         printk(KERN_INFO
5862                                                "md: checkpointing %s of %s.\n",
5863                                                desc, mdname(mddev));
5864                                         mddev->recovery_cp = mddev->curr_resync;
5865                                 }
5866                         } else
5867                                 mddev->recovery_cp = MaxSector;
5868                 } else {
5869                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5870                                 mddev->curr_resync = MaxSector;
5871                         rdev_for_each(rdev, rtmp, mddev)
5872                                 if (rdev->raid_disk >= 0 &&
5873                                     !test_bit(Faulty, &rdev->flags) &&
5874                                     !test_bit(In_sync, &rdev->flags) &&
5875                                     rdev->recovery_offset < mddev->curr_resync)
5876                                         rdev->recovery_offset = mddev->curr_resync;
5877                 }
5878         }
5879         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5880
5881  skip:
5882         mddev->curr_resync = 0;
5883         mddev->resync_min = 0;
5884         mddev->resync_max = MaxSector;
5885         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5886         wake_up(&resync_wait);
5887         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5888         md_wakeup_thread(mddev->thread);
5889         return;
5890
5891  interrupted:
5892         /*
5893          * got a signal, exit.
5894          */
5895         printk(KERN_INFO
5896                "md: md_do_sync() got signal ... exiting\n");
5897         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5898         goto out;
5899
5900 }
5901 EXPORT_SYMBOL_GPL(md_do_sync);
5902
5903
5904 static int remove_and_add_spares(mddev_t *mddev)
5905 {
5906         mdk_rdev_t *rdev;
5907         struct list_head *rtmp;
5908         int spares = 0;
5909
5910         rdev_for_each(rdev, rtmp, mddev)
5911                 if (rdev->raid_disk >= 0 &&
5912                     !test_bit(Blocked, &rdev->flags) &&
5913                     (test_bit(Faulty, &rdev->flags) ||
5914                      ! test_bit(In_sync, &rdev->flags)) &&
5915                     atomic_read(&rdev->nr_pending)==0) {
5916                         if (mddev->pers->hot_remove_disk(
5917                                     mddev, rdev->raid_disk)==0) {
5918                                 char nm[20];
5919                                 sprintf(nm,"rd%d", rdev->raid_disk);
5920                                 sysfs_remove_link(&mddev->kobj, nm);
5921                                 rdev->raid_disk = -1;
5922                         }
5923                 }
5924
5925         if (mddev->degraded) {
5926                 rdev_for_each(rdev, rtmp, mddev) {
5927                         if (rdev->raid_disk >= 0 &&
5928                             !test_bit(In_sync, &rdev->flags))
5929                                 spares++;
5930                         if (rdev->raid_disk < 0
5931                             && !test_bit(Faulty, &rdev->flags)) {
5932                                 rdev->recovery_offset = 0;
5933                                 if (mddev->pers->
5934                                     hot_add_disk(mddev, rdev) == 0) {
5935                                         char nm[20];
5936                                         sprintf(nm, "rd%d", rdev->raid_disk);
5937                                         if (sysfs_create_link(&mddev->kobj,
5938                                                               &rdev->kobj, nm))
5939                                                 printk(KERN_WARNING
5940                                                        "md: cannot register "
5941                                                        "%s for %s\n",
5942                                                        nm, mdname(mddev));
5943                                         spares++;
5944                                         md_new_event(mddev);
5945                                 } else
5946                                         break;
5947                         }
5948                 }
5949         }
5950         return spares;
5951 }
5952 /*
5953  * This routine is regularly called by all per-raid-array threads to
5954  * deal with generic issues like resync and super-block update.
5955  * Raid personalities that don't have a thread (linear/raid0) do not
5956  * need this as they never do any recovery or update the superblock.
5957  *
5958  * It does not do any resync itself, but rather "forks" off other threads
5959  * to do that as needed.
5960  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5961  * "->recovery" and create a thread at ->sync_thread.
5962  * When the thread finishes it sets MD_RECOVERY_DONE
5963  * and wakeups up this thread which will reap the thread and finish up.
5964  * This thread also removes any faulty devices (with nr_pending == 0).
5965  *
5966  * The overall approach is:
5967  *  1/ if the superblock needs updating, update it.
5968  *  2/ If a recovery thread is running, don't do anything else.
5969  *  3/ If recovery has finished, clean up, possibly marking spares active.
5970  *  4/ If there are any faulty devices, remove them.
5971  *  5/ If array is degraded, try to add spares devices
5972  *  6/ If array has spares or is not in-sync, start a resync thread.
5973  */
5974 void md_check_recovery(mddev_t *mddev)
5975 {
5976         mdk_rdev_t *rdev;
5977         struct list_head *rtmp;
5978
5979
5980         if (mddev->bitmap)
5981                 bitmap_daemon_work(mddev->bitmap);
5982
5983         if (mddev->ro)
5984                 return;
5985
5986         if (signal_pending(current)) {
5987                 if (mddev->pers->sync_request && !mddev->external) {
5988                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5989                                mdname(mddev));
5990                         mddev->safemode = 2;
5991                 }
5992                 flush_signals(current);
5993         }
5994
5995         if ( ! (
5996                 (mddev->flags && !mddev->external) ||
5997                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5998                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5999                 (mddev->external == 0 && mddev->safemode == 1) ||
6000                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6001                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6002                 ))
6003                 return;
6004
6005         if (mddev_trylock(mddev)) {
6006                 int spares = 0;
6007
6008                 if (!mddev->external) {
6009                         int did_change = 0;
6010                         spin_lock_irq(&mddev->write_lock);
6011                         if (mddev->safemode &&
6012                             !atomic_read(&mddev->writes_pending) &&
6013                             !mddev->in_sync &&
6014                             mddev->recovery_cp == MaxSector) {
6015                                 mddev->in_sync = 1;
6016                                 did_change = 1;
6017                                 if (mddev->persistent)
6018                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6019                         }
6020                         if (mddev->safemode == 1)
6021                                 mddev->safemode = 0;
6022                         spin_unlock_irq(&mddev->write_lock);
6023                         if (did_change)
6024                                 sysfs_notify(&mddev->kobj, NULL, "array_state");
6025                 }
6026
6027                 if (mddev->flags)
6028                         md_update_sb(mddev, 0);
6029
6030
6031                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6032                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6033                         /* resync/recovery still happening */
6034                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6035                         goto unlock;
6036                 }
6037                 if (mddev->sync_thread) {
6038                         /* resync has finished, collect result */
6039                         md_unregister_thread(mddev->sync_thread);
6040                         mddev->sync_thread = NULL;
6041                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6042                                 /* success...*/
6043                                 /* activate any spares */
6044                                 mddev->pers->spare_active(mddev);
6045                         }
6046                         md_update_sb(mddev, 1);
6047
6048                         /* if array is no-longer degraded, then any saved_raid_disk
6049                          * information must be scrapped
6050                          */
6051                         if (!mddev->degraded)
6052                                 rdev_for_each(rdev, rtmp, mddev)
6053                                         rdev->saved_raid_disk = -1;
6054
6055                         mddev->recovery = 0;
6056                         /* flag recovery needed just to double check */
6057                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6058                         md_new_event(mddev);
6059                         goto unlock;
6060                 }
6061                 /* Clear some bits that don't mean anything, but
6062                  * might be left set
6063                  */
6064                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6065                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6066                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6067
6068                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6069                         goto unlock;
6070                 /* no recovery is running.
6071                  * remove any failed drives, then
6072                  * add spares if possible.
6073                  * Spare are also removed and re-added, to allow
6074                  * the personality to fail the re-add.
6075                  */
6076
6077                 if (mddev->reshape_position != MaxSector) {
6078                         if (mddev->pers->check_reshape(mddev) != 0)
6079                                 /* Cannot proceed */
6080                                 goto unlock;
6081                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6082                 } else if ((spares = remove_and_add_spares(mddev))) {
6083                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6084                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6085                 } else if (mddev->recovery_cp < MaxSector) {
6086                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6087                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6088                         /* nothing to be done ... */
6089                         goto unlock;
6090
6091                 if (mddev->pers->sync_request) {
6092                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6093                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6094                                 /* We are adding a device or devices to an array
6095                                  * which has the bitmap stored on all devices.
6096                                  * So make sure all bitmap pages get written
6097                                  */
6098                                 bitmap_write_all(mddev->bitmap);
6099                         }
6100                         mddev->sync_thread = md_register_thread(md_do_sync,
6101                                                                 mddev,
6102                                                                 "%s_resync");
6103                         if (!mddev->sync_thread) {
6104                                 printk(KERN_ERR "%s: could not start resync"
6105                                         " thread...\n",
6106                                         mdname(mddev));
6107                                 /* leave the spares where they are, it shouldn't hurt */
6108                                 mddev->recovery = 0;
6109                         } else
6110                                 md_wakeup_thread(mddev->sync_thread);
6111                         md_new_event(mddev);
6112                 }
6113         unlock:
6114                 mddev_unlock(mddev);
6115         }
6116 }
6117
6118 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6119 {
6120         sysfs_notify(&rdev->kobj, NULL, "state");
6121         wait_event_timeout(rdev->blocked_wait,
6122                            !test_bit(Blocked, &rdev->flags),
6123                            msecs_to_jiffies(5000));
6124         rdev_dec_pending(rdev, mddev);
6125 }
6126 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6127
6128 static int md_notify_reboot(struct notifier_block *this,
6129                             unsigned long code, void *x)
6130 {
6131         struct list_head *tmp;
6132         mddev_t *mddev;
6133
6134         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6135
6136                 printk(KERN_INFO "md: stopping all md devices.\n");
6137
6138                 for_each_mddev(mddev, tmp)
6139                         if (mddev_trylock(mddev)) {
6140                                 do_md_stop (mddev, 1);
6141                                 mddev_unlock(mddev);
6142                         }
6143                 /*
6144                  * certain more exotic SCSI devices are known to be
6145                  * volatile wrt too early system reboots. While the
6146                  * right place to handle this issue is the given
6147                  * driver, we do want to have a safe RAID driver ...
6148                  */
6149                 mdelay(1000*1);
6150         }
6151         return NOTIFY_DONE;
6152 }
6153
6154 static struct notifier_block md_notifier = {
6155         .notifier_call  = md_notify_reboot,
6156         .next           = NULL,
6157         .priority       = INT_MAX, /* before any real devices */
6158 };
6159
6160 static void md_geninit(void)
6161 {
6162         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6163
6164         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6165 }
6166
6167 static int __init md_init(void)
6168 {
6169         if (register_blkdev(MAJOR_NR, "md"))
6170                 return -1;
6171         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6172                 unregister_blkdev(MAJOR_NR, "md");
6173                 return -1;
6174         }
6175         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6176                             md_probe, NULL, NULL);
6177         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6178                             md_probe, NULL, NULL);
6179
6180         register_reboot_notifier(&md_notifier);
6181         raid_table_header = register_sysctl_table(raid_root_table);
6182
6183         md_geninit();
6184         return (0);
6185 }
6186
6187
6188 #ifndef MODULE
6189
6190 /*
6191  * Searches all registered partitions for autorun RAID arrays
6192  * at boot time.
6193  */
6194
6195 static LIST_HEAD(all_detected_devices);
6196 struct detected_devices_node {
6197         struct list_head list;
6198         dev_t dev;
6199 };
6200
6201 void md_autodetect_dev(dev_t dev)
6202 {
6203         struct detected_devices_node *node_detected_dev;
6204
6205         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6206         if (node_detected_dev) {
6207                 node_detected_dev->dev = dev;
6208                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6209         } else {
6210                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6211                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6212         }
6213 }
6214
6215
6216 static void autostart_arrays(int part)
6217 {
6218         mdk_rdev_t *rdev;
6219         struct detected_devices_node *node_detected_dev;
6220         dev_t dev;
6221         int i_scanned, i_passed;
6222
6223         i_scanned = 0;
6224         i_passed = 0;
6225
6226         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6227
6228         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6229                 i_scanned++;
6230                 node_detected_dev = list_entry(all_detected_devices.next,
6231                                         struct detected_devices_node, list);
6232                 list_del(&node_detected_dev->list);
6233                 dev = node_detected_dev->dev;
6234                 kfree(node_detected_dev);
6235                 rdev = md_import_device(dev,0, 90);
6236                 if (IS_ERR(rdev))
6237                         continue;
6238
6239                 if (test_bit(Faulty, &rdev->flags)) {
6240                         MD_BUG();
6241                         continue;
6242                 }
6243                 set_bit(AutoDetected, &rdev->flags);
6244                 list_add(&rdev->same_set, &pending_raid_disks);
6245                 i_passed++;
6246         }
6247
6248         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6249                                                 i_scanned, i_passed);
6250
6251         autorun_devices(part);
6252 }
6253
6254 #endif /* !MODULE */
6255
6256 static __exit void md_exit(void)
6257 {
6258         mddev_t *mddev;
6259         struct list_head *tmp;
6260
6261         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6262         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6263
6264         unregister_blkdev(MAJOR_NR,"md");
6265         unregister_blkdev(mdp_major, "mdp");
6266         unregister_reboot_notifier(&md_notifier);
6267         unregister_sysctl_table(raid_table_header);
6268         remove_proc_entry("mdstat", NULL);
6269         for_each_mddev(mddev, tmp) {
6270                 struct gendisk *disk = mddev->gendisk;
6271                 if (!disk)
6272                         continue;
6273                 export_array(mddev);
6274                 del_gendisk(disk);
6275                 put_disk(disk);
6276                 mddev->gendisk = NULL;
6277                 mddev_put(mddev);
6278         }
6279 }
6280
6281 subsys_initcall(md_init);
6282 module_exit(md_exit)
6283
6284 static int get_ro(char *buffer, struct kernel_param *kp)
6285 {
6286         return sprintf(buffer, "%d", start_readonly);
6287 }
6288 static int set_ro(const char *val, struct kernel_param *kp)
6289 {
6290         char *e;
6291         int num = simple_strtoul(val, &e, 10);
6292         if (*val && (*e == '\0' || *e == '\n')) {
6293                 start_readonly = num;
6294                 return 0;
6295         }
6296         return -EINVAL;
6297 }
6298
6299 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6300 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6301
6302
6303 EXPORT_SYMBOL(register_md_personality);
6304 EXPORT_SYMBOL(unregister_md_personality);
6305 EXPORT_SYMBOL(md_error);
6306 EXPORT_SYMBOL(md_done_sync);
6307 EXPORT_SYMBOL(md_write_start);
6308 EXPORT_SYMBOL(md_write_end);
6309 EXPORT_SYMBOL(md_register_thread);
6310 EXPORT_SYMBOL(md_unregister_thread);
6311 EXPORT_SYMBOL(md_wakeup_thread);
6312 EXPORT_SYMBOL(md_check_recovery);
6313 MODULE_LICENSE("GPL");
6314 MODULE_ALIAS("md");
6315 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);