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 != 1) {
1981                         rdev->raid_disk = -1;
1982                         if (err == 0)
1983                                 return -EEXIST;
1984                         return err;
1985                 }
1986                 sprintf(nm, "rd%d", rdev->raid_disk);
1987                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
1988                         printk(KERN_WARNING
1989                                "md: cannot register "
1990                                "%s for %s\n",
1991                                nm, mdname(rdev->mddev));
1992
1993                 /* don't wakeup anyone, leave that to userspace. */
1994         } else {
1995                 if (slot >= rdev->mddev->raid_disks)
1996                         return -ENOSPC;
1997                 rdev->raid_disk = slot;
1998                 /* assume it is working */
1999                 clear_bit(Faulty, &rdev->flags);
2000                 clear_bit(WriteMostly, &rdev->flags);
2001                 set_bit(In_sync, &rdev->flags);
2002         }
2003         return len;
2004 }
2005
2006
2007 static struct rdev_sysfs_entry rdev_slot =
2008 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2009
2010 static ssize_t
2011 offset_show(mdk_rdev_t *rdev, char *page)
2012 {
2013         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2014 }
2015
2016 static ssize_t
2017 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2018 {
2019         char *e;
2020         unsigned long long offset = simple_strtoull(buf, &e, 10);
2021         if (e==buf || (*e && *e != '\n'))
2022                 return -EINVAL;
2023         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2024                 return -EBUSY;
2025         if (rdev->size && rdev->mddev->external)
2026                 /* Must set offset before size, so overlap checks
2027                  * can be sane */
2028                 return -EBUSY;
2029         rdev->data_offset = offset;
2030         return len;
2031 }
2032
2033 static struct rdev_sysfs_entry rdev_offset =
2034 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2035
2036 static ssize_t
2037 rdev_size_show(mdk_rdev_t *rdev, char *page)
2038 {
2039         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2040 }
2041
2042 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2043 {
2044         /* check if two start/length pairs overlap */
2045         if (s1+l1 <= s2)
2046                 return 0;
2047         if (s2+l2 <= s1)
2048                 return 0;
2049         return 1;
2050 }
2051
2052 static ssize_t
2053 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2054 {
2055         char *e;
2056         unsigned long long size = simple_strtoull(buf, &e, 10);
2057         unsigned long long oldsize = rdev->size;
2058         mddev_t *my_mddev = rdev->mddev;
2059
2060         if (e==buf || (*e && *e != '\n'))
2061                 return -EINVAL;
2062         if (my_mddev->pers && rdev->raid_disk >= 0)
2063                 return -EBUSY;
2064         rdev->size = size;
2065         if (size > oldsize && rdev->mddev->external) {
2066                 /* need to check that all other rdevs with the same ->bdev
2067                  * do not overlap.  We need to unlock the mddev to avoid
2068                  * a deadlock.  We have already changed rdev->size, and if
2069                  * we have to change it back, we will have the lock again.
2070                  */
2071                 mddev_t *mddev;
2072                 int overlap = 0;
2073                 struct list_head *tmp, *tmp2;
2074
2075                 mddev_unlock(my_mddev);
2076                 for_each_mddev(mddev, tmp) {
2077                         mdk_rdev_t *rdev2;
2078
2079                         mddev_lock(mddev);
2080                         rdev_for_each(rdev2, tmp2, mddev)
2081                                 if (test_bit(AllReserved, &rdev2->flags) ||
2082                                     (rdev->bdev == rdev2->bdev &&
2083                                      rdev != rdev2 &&
2084                                      overlaps(rdev->data_offset, rdev->size,
2085                                             rdev2->data_offset, rdev2->size))) {
2086                                         overlap = 1;
2087                                         break;
2088                                 }
2089                         mddev_unlock(mddev);
2090                         if (overlap) {
2091                                 mddev_put(mddev);
2092                                 break;
2093                         }
2094                 }
2095                 mddev_lock(my_mddev);
2096                 if (overlap) {
2097                         /* Someone else could have slipped in a size
2098                          * change here, but doing so is just silly.
2099                          * We put oldsize back because we *know* it is
2100                          * safe, and trust userspace not to race with
2101                          * itself
2102                          */
2103                         rdev->size = oldsize;
2104                         return -EBUSY;
2105                 }
2106         }
2107         if (size < my_mddev->size || my_mddev->size == 0)
2108                 my_mddev->size = size;
2109         return len;
2110 }
2111
2112 static struct rdev_sysfs_entry rdev_size =
2113 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2114
2115 static struct attribute *rdev_default_attrs[] = {
2116         &rdev_state.attr,
2117         &rdev_errors.attr,
2118         &rdev_slot.attr,
2119         &rdev_offset.attr,
2120         &rdev_size.attr,
2121         NULL,
2122 };
2123 static ssize_t
2124 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2125 {
2126         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2127         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2128         mddev_t *mddev = rdev->mddev;
2129         ssize_t rv;
2130
2131         if (!entry->show)
2132                 return -EIO;
2133
2134         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2135         if (!rv) {
2136                 if (rdev->mddev == NULL)
2137                         rv = -EBUSY;
2138                 else
2139                         rv = entry->show(rdev, page);
2140                 mddev_unlock(mddev);
2141         }
2142         return rv;
2143 }
2144
2145 static ssize_t
2146 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2147               const char *page, size_t length)
2148 {
2149         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2150         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2151         ssize_t rv;
2152         mddev_t *mddev = rdev->mddev;
2153
2154         if (!entry->store)
2155                 return -EIO;
2156         if (!capable(CAP_SYS_ADMIN))
2157                 return -EACCES;
2158         rv = mddev ? mddev_lock(mddev): -EBUSY;
2159         if (!rv) {
2160                 if (rdev->mddev == NULL)
2161                         rv = -EBUSY;
2162                 else
2163                         rv = entry->store(rdev, page, length);
2164                 mddev_unlock(mddev);
2165         }
2166         return rv;
2167 }
2168
2169 static void rdev_free(struct kobject *ko)
2170 {
2171         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2172         kfree(rdev);
2173 }
2174 static struct sysfs_ops rdev_sysfs_ops = {
2175         .show           = rdev_attr_show,
2176         .store          = rdev_attr_store,
2177 };
2178 static struct kobj_type rdev_ktype = {
2179         .release        = rdev_free,
2180         .sysfs_ops      = &rdev_sysfs_ops,
2181         .default_attrs  = rdev_default_attrs,
2182 };
2183
2184 /*
2185  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2186  *
2187  * mark the device faulty if:
2188  *
2189  *   - the device is nonexistent (zero size)
2190  *   - the device has no valid superblock
2191  *
2192  * a faulty rdev _never_ has rdev->sb set.
2193  */
2194 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2195 {
2196         char b[BDEVNAME_SIZE];
2197         int err;
2198         mdk_rdev_t *rdev;
2199         sector_t size;
2200
2201         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2202         if (!rdev) {
2203                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2204                 return ERR_PTR(-ENOMEM);
2205         }
2206
2207         if ((err = alloc_disk_sb(rdev)))
2208                 goto abort_free;
2209
2210         err = lock_rdev(rdev, newdev, super_format == -2);
2211         if (err)
2212                 goto abort_free;
2213
2214         kobject_init(&rdev->kobj, &rdev_ktype);
2215
2216         rdev->desc_nr = -1;
2217         rdev->saved_raid_disk = -1;
2218         rdev->raid_disk = -1;
2219         rdev->flags = 0;
2220         rdev->data_offset = 0;
2221         rdev->sb_events = 0;
2222         atomic_set(&rdev->nr_pending, 0);
2223         atomic_set(&rdev->read_errors, 0);
2224         atomic_set(&rdev->corrected_errors, 0);
2225
2226         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2227         if (!size) {
2228                 printk(KERN_WARNING
2229                         "md: %s has zero or unknown size, marking faulty!\n",
2230                         bdevname(rdev->bdev,b));
2231                 err = -EINVAL;
2232                 goto abort_free;
2233         }
2234
2235         if (super_format >= 0) {
2236                 err = super_types[super_format].
2237                         load_super(rdev, NULL, super_minor);
2238                 if (err == -EINVAL) {
2239                         printk(KERN_WARNING
2240                                 "md: %s does not have a valid v%d.%d "
2241                                "superblock, not importing!\n",
2242                                 bdevname(rdev->bdev,b),
2243                                super_format, super_minor);
2244                         goto abort_free;
2245                 }
2246                 if (err < 0) {
2247                         printk(KERN_WARNING
2248                                 "md: could not read %s's sb, not importing!\n",
2249                                 bdevname(rdev->bdev,b));
2250                         goto abort_free;
2251                 }
2252         }
2253
2254         INIT_LIST_HEAD(&rdev->same_set);
2255         init_waitqueue_head(&rdev->blocked_wait);
2256
2257         return rdev;
2258
2259 abort_free:
2260         if (rdev->sb_page) {
2261                 if (rdev->bdev)
2262                         unlock_rdev(rdev);
2263                 free_disk_sb(rdev);
2264         }
2265         kfree(rdev);
2266         return ERR_PTR(err);
2267 }
2268
2269 /*
2270  * Check a full RAID array for plausibility
2271  */
2272
2273
2274 static void analyze_sbs(mddev_t * mddev)
2275 {
2276         int i;
2277         struct list_head *tmp;
2278         mdk_rdev_t *rdev, *freshest;
2279         char b[BDEVNAME_SIZE];
2280
2281         freshest = NULL;
2282         rdev_for_each(rdev, tmp, mddev)
2283                 switch (super_types[mddev->major_version].
2284                         load_super(rdev, freshest, mddev->minor_version)) {
2285                 case 1:
2286                         freshest = rdev;
2287                         break;
2288                 case 0:
2289                         break;
2290                 default:
2291                         printk( KERN_ERR \
2292                                 "md: fatal superblock inconsistency in %s"
2293                                 " -- removing from array\n",
2294                                 bdevname(rdev->bdev,b));
2295                         kick_rdev_from_array(rdev);
2296                 }
2297
2298
2299         super_types[mddev->major_version].
2300                 validate_super(mddev, freshest);
2301
2302         i = 0;
2303         rdev_for_each(rdev, tmp, mddev) {
2304                 if (rdev != freshest)
2305                         if (super_types[mddev->major_version].
2306                             validate_super(mddev, rdev)) {
2307                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2308                                         " from array!\n",
2309                                         bdevname(rdev->bdev,b));
2310                                 kick_rdev_from_array(rdev);
2311                                 continue;
2312                         }
2313                 if (mddev->level == LEVEL_MULTIPATH) {
2314                         rdev->desc_nr = i++;
2315                         rdev->raid_disk = rdev->desc_nr;
2316                         set_bit(In_sync, &rdev->flags);
2317                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2318                         rdev->raid_disk = -1;
2319                         clear_bit(In_sync, &rdev->flags);
2320                 }
2321         }
2322
2323
2324
2325         if (mddev->recovery_cp != MaxSector &&
2326             mddev->level >= 1)
2327                 printk(KERN_ERR "md: %s: raid array is not clean"
2328                        " -- starting background reconstruction\n",
2329                        mdname(mddev));
2330
2331 }
2332
2333 static ssize_t
2334 safe_delay_show(mddev_t *mddev, char *page)
2335 {
2336         int msec = (mddev->safemode_delay*1000)/HZ;
2337         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2338 }
2339 static ssize_t
2340 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2341 {
2342         int scale=1;
2343         int dot=0;
2344         int i;
2345         unsigned long msec;
2346         char buf[30];
2347         char *e;
2348         /* remove a period, and count digits after it */
2349         if (len >= sizeof(buf))
2350                 return -EINVAL;
2351         strlcpy(buf, cbuf, len);
2352         buf[len] = 0;
2353         for (i=0; i<len; i++) {
2354                 if (dot) {
2355                         if (isdigit(buf[i])) {
2356                                 buf[i-1] = buf[i];
2357                                 scale *= 10;
2358                         }
2359                         buf[i] = 0;
2360                 } else if (buf[i] == '.') {
2361                         dot=1;
2362                         buf[i] = 0;
2363                 }
2364         }
2365         msec = simple_strtoul(buf, &e, 10);
2366         if (e == buf || (*e && *e != '\n'))
2367                 return -EINVAL;
2368         msec = (msec * 1000) / scale;
2369         if (msec == 0)
2370                 mddev->safemode_delay = 0;
2371         else {
2372                 mddev->safemode_delay = (msec*HZ)/1000;
2373                 if (mddev->safemode_delay == 0)
2374                         mddev->safemode_delay = 1;
2375         }
2376         return len;
2377 }
2378 static struct md_sysfs_entry md_safe_delay =
2379 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2380
2381 static ssize_t
2382 level_show(mddev_t *mddev, char *page)
2383 {
2384         struct mdk_personality *p = mddev->pers;
2385         if (p)
2386                 return sprintf(page, "%s\n", p->name);
2387         else if (mddev->clevel[0])
2388                 return sprintf(page, "%s\n", mddev->clevel);
2389         else if (mddev->level != LEVEL_NONE)
2390                 return sprintf(page, "%d\n", mddev->level);
2391         else
2392                 return 0;
2393 }
2394
2395 static ssize_t
2396 level_store(mddev_t *mddev, const char *buf, size_t len)
2397 {
2398         ssize_t rv = len;
2399         if (mddev->pers)
2400                 return -EBUSY;
2401         if (len == 0)
2402                 return 0;
2403         if (len >= sizeof(mddev->clevel))
2404                 return -ENOSPC;
2405         strncpy(mddev->clevel, buf, len);
2406         if (mddev->clevel[len-1] == '\n')
2407                 len--;
2408         mddev->clevel[len] = 0;
2409         mddev->level = LEVEL_NONE;
2410         return rv;
2411 }
2412
2413 static struct md_sysfs_entry md_level =
2414 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2415
2416
2417 static ssize_t
2418 layout_show(mddev_t *mddev, char *page)
2419 {
2420         /* just a number, not meaningful for all levels */
2421         if (mddev->reshape_position != MaxSector &&
2422             mddev->layout != mddev->new_layout)
2423                 return sprintf(page, "%d (%d)\n",
2424                                mddev->new_layout, mddev->layout);
2425         return sprintf(page, "%d\n", mddev->layout);
2426 }
2427
2428 static ssize_t
2429 layout_store(mddev_t *mddev, const char *buf, size_t len)
2430 {
2431         char *e;
2432         unsigned long n = simple_strtoul(buf, &e, 10);
2433
2434         if (!*buf || (*e && *e != '\n'))
2435                 return -EINVAL;
2436
2437         if (mddev->pers)
2438                 return -EBUSY;
2439         if (mddev->reshape_position != MaxSector)
2440                 mddev->new_layout = n;
2441         else
2442                 mddev->layout = n;
2443         return len;
2444 }
2445 static struct md_sysfs_entry md_layout =
2446 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2447
2448
2449 static ssize_t
2450 raid_disks_show(mddev_t *mddev, char *page)
2451 {
2452         if (mddev->raid_disks == 0)
2453                 return 0;
2454         if (mddev->reshape_position != MaxSector &&
2455             mddev->delta_disks != 0)
2456                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2457                                mddev->raid_disks - mddev->delta_disks);
2458         return sprintf(page, "%d\n", mddev->raid_disks);
2459 }
2460
2461 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2462
2463 static ssize_t
2464 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2465 {
2466         char *e;
2467         int rv = 0;
2468         unsigned long n = simple_strtoul(buf, &e, 10);
2469
2470         if (!*buf || (*e && *e != '\n'))
2471                 return -EINVAL;
2472
2473         if (mddev->pers)
2474                 rv = update_raid_disks(mddev, n);
2475         else if (mddev->reshape_position != MaxSector) {
2476                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2477                 mddev->delta_disks = n - olddisks;
2478                 mddev->raid_disks = n;
2479         } else
2480                 mddev->raid_disks = n;
2481         return rv ? rv : len;
2482 }
2483 static struct md_sysfs_entry md_raid_disks =
2484 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2485
2486 static ssize_t
2487 chunk_size_show(mddev_t *mddev, char *page)
2488 {
2489         if (mddev->reshape_position != MaxSector &&
2490             mddev->chunk_size != mddev->new_chunk)
2491                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2492                                mddev->chunk_size);
2493         return sprintf(page, "%d\n", mddev->chunk_size);
2494 }
2495
2496 static ssize_t
2497 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2498 {
2499         /* can only set chunk_size if array is not yet active */
2500         char *e;
2501         unsigned long n = simple_strtoul(buf, &e, 10);
2502
2503         if (!*buf || (*e && *e != '\n'))
2504                 return -EINVAL;
2505
2506         if (mddev->pers)
2507                 return -EBUSY;
2508         else if (mddev->reshape_position != MaxSector)
2509                 mddev->new_chunk = n;
2510         else
2511                 mddev->chunk_size = n;
2512         return len;
2513 }
2514 static struct md_sysfs_entry md_chunk_size =
2515 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2516
2517 static ssize_t
2518 resync_start_show(mddev_t *mddev, char *page)
2519 {
2520         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2521 }
2522
2523 static ssize_t
2524 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2525 {
2526         char *e;
2527         unsigned long long n = simple_strtoull(buf, &e, 10);
2528
2529         if (mddev->pers)
2530                 return -EBUSY;
2531         if (!*buf || (*e && *e != '\n'))
2532                 return -EINVAL;
2533
2534         mddev->recovery_cp = n;
2535         return len;
2536 }
2537 static struct md_sysfs_entry md_resync_start =
2538 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2539
2540 /*
2541  * The array state can be:
2542  *
2543  * clear
2544  *     No devices, no size, no level
2545  *     Equivalent to STOP_ARRAY ioctl
2546  * inactive
2547  *     May have some settings, but array is not active
2548  *        all IO results in error
2549  *     When written, doesn't tear down array, but just stops it
2550  * suspended (not supported yet)
2551  *     All IO requests will block. The array can be reconfigured.
2552  *     Writing this, if accepted, will block until array is quiessent
2553  * readonly
2554  *     no resync can happen.  no superblocks get written.
2555  *     write requests fail
2556  * read-auto
2557  *     like readonly, but behaves like 'clean' on a write request.
2558  *
2559  * clean - no pending writes, but otherwise active.
2560  *     When written to inactive array, starts without resync
2561  *     If a write request arrives then
2562  *       if metadata is known, mark 'dirty' and switch to 'active'.
2563  *       if not known, block and switch to write-pending
2564  *     If written to an active array that has pending writes, then fails.
2565  * active
2566  *     fully active: IO and resync can be happening.
2567  *     When written to inactive array, starts with resync
2568  *
2569  * write-pending
2570  *     clean, but writes are blocked waiting for 'active' to be written.
2571  *
2572  * active-idle
2573  *     like active, but no writes have been seen for a while (100msec).
2574  *
2575  */
2576 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2577                    write_pending, active_idle, bad_word};
2578 static char *array_states[] = {
2579         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2580         "write-pending", "active-idle", NULL };
2581
2582 static int match_word(const char *word, char **list)
2583 {
2584         int n;
2585         for (n=0; list[n]; n++)
2586                 if (cmd_match(word, list[n]))
2587                         break;
2588         return n;
2589 }
2590
2591 static ssize_t
2592 array_state_show(mddev_t *mddev, char *page)
2593 {
2594         enum array_state st = inactive;
2595
2596         if (mddev->pers)
2597                 switch(mddev->ro) {
2598                 case 1:
2599                         st = readonly;
2600                         break;
2601                 case 2:
2602                         st = read_auto;
2603                         break;
2604                 case 0:
2605                         if (mddev->in_sync)
2606                                 st = clean;
2607                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2608                                 st = write_pending;
2609                         else if (mddev->safemode)
2610                                 st = active_idle;
2611                         else
2612                                 st = active;
2613                 }
2614         else {
2615                 if (list_empty(&mddev->disks) &&
2616                     mddev->raid_disks == 0 &&
2617                     mddev->size == 0)
2618                         st = clear;
2619                 else
2620                         st = inactive;
2621         }
2622         return sprintf(page, "%s\n", array_states[st]);
2623 }
2624
2625 static int do_md_stop(mddev_t * mddev, int ro);
2626 static int do_md_run(mddev_t * mddev);
2627 static int restart_array(mddev_t *mddev);
2628
2629 static ssize_t
2630 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2631 {
2632         int err = -EINVAL;
2633         enum array_state st = match_word(buf, array_states);
2634         switch(st) {
2635         case bad_word:
2636                 break;
2637         case clear:
2638                 /* stopping an active array */
2639                 if (atomic_read(&mddev->active) > 1)
2640                         return -EBUSY;
2641                 err = do_md_stop(mddev, 0);
2642                 break;
2643         case inactive:
2644                 /* stopping an active array */
2645                 if (mddev->pers) {
2646                         if (atomic_read(&mddev->active) > 1)
2647                                 return -EBUSY;
2648                         err = do_md_stop(mddev, 2);
2649                 } else
2650                         err = 0; /* already inactive */
2651                 break;
2652         case suspended:
2653                 break; /* not supported yet */
2654         case readonly:
2655                 if (mddev->pers)
2656                         err = do_md_stop(mddev, 1);
2657                 else {
2658                         mddev->ro = 1;
2659                         set_disk_ro(mddev->gendisk, 1);
2660                         err = do_md_run(mddev);
2661                 }
2662                 break;
2663         case read_auto:
2664                 if (mddev->pers) {
2665                         if (mddev->ro != 1)
2666                                 err = do_md_stop(mddev, 1);
2667                         else
2668                                 err = restart_array(mddev);
2669                         if (err == 0) {
2670                                 mddev->ro = 2;
2671                                 set_disk_ro(mddev->gendisk, 0);
2672                         }
2673                 } else {
2674                         mddev->ro = 2;
2675                         err = do_md_run(mddev);
2676                 }
2677                 break;
2678         case clean:
2679                 if (mddev->pers) {
2680                         restart_array(mddev);
2681                         spin_lock_irq(&mddev->write_lock);
2682                         if (atomic_read(&mddev->writes_pending) == 0) {
2683                                 if (mddev->in_sync == 0) {
2684                                         mddev->in_sync = 1;
2685                                         if (mddev->safemode == 1)
2686                                                 mddev->safemode = 0;
2687                                         if (mddev->persistent)
2688                                                 set_bit(MD_CHANGE_CLEAN,
2689                                                         &mddev->flags);
2690                                 }
2691                                 err = 0;
2692                         } else
2693                                 err = -EBUSY;
2694                         spin_unlock_irq(&mddev->write_lock);
2695                 } else {
2696                         mddev->ro = 0;
2697                         mddev->recovery_cp = MaxSector;
2698                         err = do_md_run(mddev);
2699                 }
2700                 break;
2701         case active:
2702                 if (mddev->pers) {
2703                         restart_array(mddev);
2704                         if (mddev->external)
2705                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2706                         wake_up(&mddev->sb_wait);
2707                         err = 0;
2708                 } else {
2709                         mddev->ro = 0;
2710                         set_disk_ro(mddev->gendisk, 0);
2711                         err = do_md_run(mddev);
2712                 }
2713                 break;
2714         case write_pending:
2715         case active_idle:
2716                 /* these cannot be set */
2717                 break;
2718         }
2719         if (err)
2720                 return err;
2721         else
2722                 return len;
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         mddev->safemode = 1;
3414         md_wakeup_thread(mddev->thread);
3415 }
3416
3417 static int start_dirty_degraded;
3418
3419 static int do_md_run(mddev_t * mddev)
3420 {
3421         int err;
3422         int chunk_size;
3423         struct list_head *tmp;
3424         mdk_rdev_t *rdev;
3425         struct gendisk *disk;
3426         struct mdk_personality *pers;
3427         char b[BDEVNAME_SIZE];
3428
3429         if (list_empty(&mddev->disks))
3430                 /* cannot run an array with no devices.. */
3431                 return -EINVAL;
3432
3433         if (mddev->pers)
3434                 return -EBUSY;
3435
3436         /*
3437          * Analyze all RAID superblock(s)
3438          */
3439         if (!mddev->raid_disks) {
3440                 if (!mddev->persistent)
3441                         return -EINVAL;
3442                 analyze_sbs(mddev);
3443         }
3444
3445         chunk_size = mddev->chunk_size;
3446
3447         if (chunk_size) {
3448                 if (chunk_size > MAX_CHUNK_SIZE) {
3449                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3450                                 chunk_size, MAX_CHUNK_SIZE);
3451                         return -EINVAL;
3452                 }
3453                 /*
3454                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3455                  */
3456                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3457                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3458                         return -EINVAL;
3459                 }
3460                 if (chunk_size < PAGE_SIZE) {
3461                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3462                                 chunk_size, PAGE_SIZE);
3463                         return -EINVAL;
3464                 }
3465
3466                 /* devices must have minimum size of one chunk */
3467                 rdev_for_each(rdev, tmp, mddev) {
3468                         if (test_bit(Faulty, &rdev->flags))
3469                                 continue;
3470                         if (rdev->size < chunk_size / 1024) {
3471                                 printk(KERN_WARNING
3472                                         "md: Dev %s smaller than chunk_size:"
3473                                         " %lluk < %dk\n",
3474                                         bdevname(rdev->bdev,b),
3475                                         (unsigned long long)rdev->size,
3476                                         chunk_size / 1024);
3477                                 return -EINVAL;
3478                         }
3479                 }
3480         }
3481
3482 #ifdef CONFIG_KMOD
3483         if (mddev->level != LEVEL_NONE)
3484                 request_module("md-level-%d", mddev->level);
3485         else if (mddev->clevel[0])
3486                 request_module("md-%s", mddev->clevel);
3487 #endif
3488
3489         /*
3490          * Drop all container device buffers, from now on
3491          * the only valid external interface is through the md
3492          * device.
3493          */
3494         rdev_for_each(rdev, tmp, mddev) {
3495                 if (test_bit(Faulty, &rdev->flags))
3496                         continue;
3497                 sync_blockdev(rdev->bdev);
3498                 invalidate_bdev(rdev->bdev);
3499
3500                 /* perform some consistency tests on the device.
3501                  * We don't want the data to overlap the metadata,
3502                  * Internal Bitmap issues has handled elsewhere.
3503                  */
3504                 if (rdev->data_offset < rdev->sb_offset) {
3505                         if (mddev->size &&
3506                             rdev->data_offset + mddev->size*2
3507                             > rdev->sb_offset*2) {
3508                                 printk("md: %s: data overlaps metadata\n",
3509                                        mdname(mddev));
3510                                 return -EINVAL;
3511                         }
3512                 } else {
3513                         if (rdev->sb_offset*2 + rdev->sb_size/512
3514                             > rdev->data_offset) {
3515                                 printk("md: %s: metadata overlaps data\n",
3516                                        mdname(mddev));
3517                                 return -EINVAL;
3518                         }
3519                 }
3520         }
3521
3522         md_probe(mddev->unit, NULL, NULL);
3523         disk = mddev->gendisk;
3524         if (!disk)
3525                 return -ENOMEM;
3526
3527         spin_lock(&pers_lock);
3528         pers = find_pers(mddev->level, mddev->clevel);
3529         if (!pers || !try_module_get(pers->owner)) {
3530                 spin_unlock(&pers_lock);
3531                 if (mddev->level != LEVEL_NONE)
3532                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3533                                mddev->level);
3534                 else
3535                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3536                                mddev->clevel);
3537                 return -EINVAL;
3538         }
3539         mddev->pers = pers;
3540         spin_unlock(&pers_lock);
3541         mddev->level = pers->level;
3542         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3543
3544         if (mddev->reshape_position != MaxSector &&
3545             pers->start_reshape == NULL) {
3546                 /* This personality cannot handle reshaping... */
3547                 mddev->pers = NULL;
3548                 module_put(pers->owner);
3549                 return -EINVAL;
3550         }
3551
3552         if (pers->sync_request) {
3553                 /* Warn if this is a potentially silly
3554                  * configuration.
3555                  */
3556                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3557                 mdk_rdev_t *rdev2;
3558                 struct list_head *tmp2;
3559                 int warned = 0;
3560                 rdev_for_each(rdev, tmp, mddev) {
3561                         rdev_for_each(rdev2, tmp2, mddev) {
3562                                 if (rdev < rdev2 &&
3563                                     rdev->bdev->bd_contains ==
3564                                     rdev2->bdev->bd_contains) {
3565                                         printk(KERN_WARNING
3566                                                "%s: WARNING: %s appears to be"
3567                                                " on the same physical disk as"
3568                                                " %s.\n",
3569                                                mdname(mddev),
3570                                                bdevname(rdev->bdev,b),
3571                                                bdevname(rdev2->bdev,b2));
3572                                         warned = 1;
3573                                 }
3574                         }
3575                 }
3576                 if (warned)
3577                         printk(KERN_WARNING
3578                                "True protection against single-disk"
3579                                " failure might be compromised.\n");
3580         }
3581
3582         mddev->recovery = 0;
3583         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3584         mddev->barriers_work = 1;
3585         mddev->ok_start_degraded = start_dirty_degraded;
3586
3587         if (start_readonly)
3588                 mddev->ro = 2; /* read-only, but switch on first write */
3589
3590         err = mddev->pers->run(mddev);
3591         if (!err && mddev->pers->sync_request) {
3592                 err = bitmap_create(mddev);
3593                 if (err) {
3594                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3595                                mdname(mddev), err);
3596                         mddev->pers->stop(mddev);
3597                 }
3598         }
3599         if (err) {
3600                 printk(KERN_ERR "md: pers->run() failed ...\n");
3601                 module_put(mddev->pers->owner);
3602                 mddev->pers = NULL;
3603                 bitmap_destroy(mddev);
3604                 return err;
3605         }
3606         if (mddev->pers->sync_request) {
3607                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3608                         printk(KERN_WARNING
3609                                "md: cannot register extra attributes for %s\n",
3610                                mdname(mddev));
3611         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3612                 mddev->ro = 0;
3613
3614         atomic_set(&mddev->writes_pending,0);
3615         mddev->safemode = 0;
3616         mddev->safemode_timer.function = md_safemode_timeout;
3617         mddev->safemode_timer.data = (unsigned long) mddev;
3618         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3619         mddev->in_sync = 1;
3620
3621         rdev_for_each(rdev, tmp, mddev)
3622                 if (rdev->raid_disk >= 0) {
3623                         char nm[20];
3624                         sprintf(nm, "rd%d", rdev->raid_disk);
3625                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3626                                 printk("md: cannot register %s for %s\n",
3627                                        nm, mdname(mddev));
3628                 }
3629
3630         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3631
3632         if (mddev->flags)
3633                 md_update_sb(mddev, 0);
3634
3635         set_capacity(disk, mddev->array_size<<1);
3636
3637         /* If we call blk_queue_make_request here, it will
3638          * re-initialise max_sectors etc which may have been
3639          * refined inside -> run.  So just set the bits we need to set.
3640          * Most initialisation happended when we called
3641          * blk_queue_make_request(..., md_fail_request)
3642          * earlier.
3643          */
3644         mddev->queue->queuedata = mddev;
3645         mddev->queue->make_request_fn = mddev->pers->make_request;
3646
3647         /* If there is a partially-recovered drive we need to
3648          * start recovery here.  If we leave it to md_check_recovery,
3649          * it will remove the drives and not do the right thing
3650          */
3651         if (mddev->degraded && !mddev->sync_thread) {
3652                 struct list_head *rtmp;
3653                 int spares = 0;
3654                 rdev_for_each(rdev, rtmp, mddev)
3655                         if (rdev->raid_disk >= 0 &&
3656                             !test_bit(In_sync, &rdev->flags) &&
3657                             !test_bit(Faulty, &rdev->flags))
3658                                 /* complete an interrupted recovery */
3659                                 spares++;
3660                 if (spares && mddev->pers->sync_request) {
3661                         mddev->recovery = 0;
3662                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3663                         mddev->sync_thread = md_register_thread(md_do_sync,
3664                                                                 mddev,
3665                                                                 "%s_resync");
3666                         if (!mddev->sync_thread) {
3667                                 printk(KERN_ERR "%s: could not start resync"
3668                                        " thread...\n",
3669                                        mdname(mddev));
3670                                 /* leave the spares where they are, it shouldn't hurt */
3671                                 mddev->recovery = 0;
3672                         }
3673                 }
3674         }
3675         md_wakeup_thread(mddev->thread);
3676         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3677
3678         mddev->changed = 1;
3679         md_new_event(mddev);
3680         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3681         return 0;
3682 }
3683
3684 static int restart_array(mddev_t *mddev)
3685 {
3686         struct gendisk *disk = mddev->gendisk;
3687         int err;
3688
3689         /*
3690          * Complain if it has no devices
3691          */
3692         err = -ENXIO;
3693         if (list_empty(&mddev->disks))
3694                 goto out;
3695
3696         if (mddev->pers) {
3697                 err = -EBUSY;
3698                 if (!mddev->ro)
3699                         goto out;
3700
3701                 mddev->safemode = 0;
3702                 mddev->ro = 0;
3703                 set_disk_ro(disk, 0);
3704
3705                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3706                         mdname(mddev));
3707                 /*
3708                  * Kick recovery or resync if necessary
3709                  */
3710                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3711                 md_wakeup_thread(mddev->thread);
3712                 md_wakeup_thread(mddev->sync_thread);
3713                 err = 0;
3714         } else
3715                 err = -EINVAL;
3716
3717 out:
3718         return err;
3719 }
3720
3721 /* similar to deny_write_access, but accounts for our holding a reference
3722  * to the file ourselves */
3723 static int deny_bitmap_write_access(struct file * file)
3724 {
3725         struct inode *inode = file->f_mapping->host;
3726
3727         spin_lock(&inode->i_lock);
3728         if (atomic_read(&inode->i_writecount) > 1) {
3729                 spin_unlock(&inode->i_lock);
3730                 return -ETXTBSY;
3731         }
3732         atomic_set(&inode->i_writecount, -1);
3733         spin_unlock(&inode->i_lock);
3734
3735         return 0;
3736 }
3737
3738 static void restore_bitmap_write_access(struct file *file)
3739 {
3740         struct inode *inode = file->f_mapping->host;
3741
3742         spin_lock(&inode->i_lock);
3743         atomic_set(&inode->i_writecount, 1);
3744         spin_unlock(&inode->i_lock);
3745 }
3746
3747 /* mode:
3748  *   0 - completely stop and dis-assemble array
3749  *   1 - switch to readonly
3750  *   2 - stop but do not disassemble array
3751  */
3752 static int do_md_stop(mddev_t * mddev, int mode)
3753 {
3754         int err = 0;
3755         struct gendisk *disk = mddev->gendisk;
3756
3757         if (mddev->pers) {
3758                 if (atomic_read(&mddev->active)>2) {
3759                         printk("md: %s still in use.\n",mdname(mddev));
3760                         return -EBUSY;
3761                 }
3762
3763                 if (mddev->sync_thread) {
3764                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3765                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3766                         md_unregister_thread(mddev->sync_thread);
3767                         mddev->sync_thread = NULL;
3768                 }
3769
3770                 del_timer_sync(&mddev->safemode_timer);
3771
3772                 invalidate_partition(disk, 0);
3773
3774                 switch(mode) {
3775                 case 1: /* readonly */
3776                         err  = -ENXIO;
3777                         if (mddev->ro==1)
3778                                 goto out;
3779                         mddev->ro = 1;
3780                         break;
3781                 case 0: /* disassemble */
3782                 case 2: /* stop */
3783                         bitmap_flush(mddev);
3784                         md_super_wait(mddev);
3785                         if (mddev->ro)
3786                                 set_disk_ro(disk, 0);
3787                         blk_queue_make_request(mddev->queue, md_fail_request);
3788                         mddev->pers->stop(mddev);
3789                         mddev->queue->merge_bvec_fn = NULL;
3790                         mddev->queue->unplug_fn = NULL;
3791                         mddev->queue->backing_dev_info.congested_fn = NULL;
3792                         if (mddev->pers->sync_request)
3793                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3794
3795                         module_put(mddev->pers->owner);
3796                         mddev->pers = NULL;
3797                         /* tell userspace to handle 'inactive' */
3798                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3799
3800                         set_capacity(disk, 0);
3801                         mddev->changed = 1;
3802
3803                         if (mddev->ro)
3804                                 mddev->ro = 0;
3805                 }
3806                 if (!mddev->in_sync || mddev->flags) {
3807                         /* mark array as shutdown cleanly */
3808                         mddev->in_sync = 1;
3809                         md_update_sb(mddev, 1);
3810                 }
3811                 if (mode == 1)
3812                         set_disk_ro(disk, 1);
3813                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3814         }
3815
3816         /*
3817          * Free resources if final stop
3818          */
3819         if (mode == 0) {
3820                 mdk_rdev_t *rdev;
3821                 struct list_head *tmp;
3822
3823                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3824
3825                 bitmap_destroy(mddev);
3826                 if (mddev->bitmap_file) {
3827                         restore_bitmap_write_access(mddev->bitmap_file);
3828                         fput(mddev->bitmap_file);
3829                         mddev->bitmap_file = NULL;
3830                 }
3831                 mddev->bitmap_offset = 0;
3832
3833                 rdev_for_each(rdev, tmp, mddev)
3834                         if (rdev->raid_disk >= 0) {
3835                                 char nm[20];
3836                                 sprintf(nm, "rd%d", rdev->raid_disk);
3837                                 sysfs_remove_link(&mddev->kobj, nm);
3838                         }
3839
3840                 /* make sure all md_delayed_delete calls have finished */
3841                 flush_scheduled_work();
3842
3843                 export_array(mddev);
3844
3845                 mddev->array_size = 0;
3846                 mddev->size = 0;
3847                 mddev->raid_disks = 0;
3848                 mddev->recovery_cp = 0;
3849                 mddev->resync_min = 0;
3850                 mddev->resync_max = MaxSector;
3851                 mddev->reshape_position = MaxSector;
3852                 mddev->external = 0;
3853                 mddev->persistent = 0;
3854                 mddev->level = LEVEL_NONE;
3855                 mddev->clevel[0] = 0;
3856                 mddev->flags = 0;
3857                 mddev->ro = 0;
3858                 mddev->metadata_type[0] = 0;
3859                 mddev->chunk_size = 0;
3860                 mddev->ctime = mddev->utime = 0;
3861                 mddev->layout = 0;
3862                 mddev->max_disks = 0;
3863                 mddev->events = 0;
3864                 mddev->delta_disks = 0;
3865                 mddev->new_level = LEVEL_NONE;
3866                 mddev->new_layout = 0;
3867                 mddev->new_chunk = 0;
3868                 mddev->curr_resync = 0;
3869                 mddev->resync_mismatches = 0;
3870                 mddev->suspend_lo = mddev->suspend_hi = 0;
3871                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3872                 mddev->recovery = 0;
3873                 mddev->in_sync = 0;
3874                 mddev->changed = 0;
3875                 mddev->degraded = 0;
3876                 mddev->barriers_work = 0;
3877                 mddev->safemode = 0;
3878
3879         } else if (mddev->pers)
3880                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3881                         mdname(mddev));
3882         err = 0;
3883         md_new_event(mddev);
3884 out:
3885         return err;
3886 }
3887
3888 #ifndef MODULE
3889 static void autorun_array(mddev_t *mddev)
3890 {
3891         mdk_rdev_t *rdev;
3892         struct list_head *tmp;
3893         int err;
3894
3895         if (list_empty(&mddev->disks))
3896                 return;
3897
3898         printk(KERN_INFO "md: running: ");
3899
3900         rdev_for_each(rdev, tmp, mddev) {
3901                 char b[BDEVNAME_SIZE];
3902                 printk("<%s>", bdevname(rdev->bdev,b));
3903         }
3904         printk("\n");
3905
3906         err = do_md_run (mddev);
3907         if (err) {
3908                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3909                 do_md_stop (mddev, 0);
3910         }
3911 }
3912
3913 /*
3914  * lets try to run arrays based on all disks that have arrived
3915  * until now. (those are in pending_raid_disks)
3916  *
3917  * the method: pick the first pending disk, collect all disks with
3918  * the same UUID, remove all from the pending list and put them into
3919  * the 'same_array' list. Then order this list based on superblock
3920  * update time (freshest comes first), kick out 'old' disks and
3921  * compare superblocks. If everything's fine then run it.
3922  *
3923  * If "unit" is allocated, then bump its reference count
3924  */
3925 static void autorun_devices(int part)
3926 {
3927         struct list_head *tmp;
3928         mdk_rdev_t *rdev0, *rdev;
3929         mddev_t *mddev;
3930         char b[BDEVNAME_SIZE];
3931
3932         printk(KERN_INFO "md: autorun ...\n");
3933         while (!list_empty(&pending_raid_disks)) {
3934                 int unit;
3935                 dev_t dev;
3936                 LIST_HEAD(candidates);
3937                 rdev0 = list_entry(pending_raid_disks.next,
3938                                          mdk_rdev_t, same_set);
3939
3940                 printk(KERN_INFO "md: considering %s ...\n",
3941                         bdevname(rdev0->bdev,b));
3942                 INIT_LIST_HEAD(&candidates);
3943                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3944                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3945                                 printk(KERN_INFO "md:  adding %s ...\n",
3946                                         bdevname(rdev->bdev,b));
3947                                 list_move(&rdev->same_set, &candidates);
3948                         }
3949                 /*
3950                  * now we have a set of devices, with all of them having
3951                  * mostly sane superblocks. It's time to allocate the
3952                  * mddev.
3953                  */
3954                 if (part) {
3955                         dev = MKDEV(mdp_major,
3956                                     rdev0->preferred_minor << MdpMinorShift);
3957                         unit = MINOR(dev) >> MdpMinorShift;
3958                 } else {
3959                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3960                         unit = MINOR(dev);
3961                 }
3962                 if (rdev0->preferred_minor != unit) {
3963                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3964                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3965                         break;
3966                 }
3967
3968                 md_probe(dev, NULL, NULL);
3969                 mddev = mddev_find(dev);
3970                 if (!mddev || !mddev->gendisk) {
3971                         if (mddev)
3972                                 mddev_put(mddev);
3973                         printk(KERN_ERR
3974                                 "md: cannot allocate memory for md drive.\n");
3975                         break;
3976                 }
3977                 if (mddev_lock(mddev))
3978                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3979                                mdname(mddev));
3980                 else if (mddev->raid_disks || mddev->major_version
3981                          || !list_empty(&mddev->disks)) {
3982                         printk(KERN_WARNING
3983                                 "md: %s already running, cannot run %s\n",
3984                                 mdname(mddev), bdevname(rdev0->bdev,b));
3985                         mddev_unlock(mddev);
3986                 } else {
3987                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3988                         mddev->persistent = 1;
3989                         rdev_for_each_list(rdev, tmp, candidates) {
3990                                 list_del_init(&rdev->same_set);
3991                                 if (bind_rdev_to_array(rdev, mddev))
3992                                         export_rdev(rdev);
3993                         }
3994                         autorun_array(mddev);
3995                         mddev_unlock(mddev);
3996                 }
3997                 /* on success, candidates will be empty, on error
3998                  * it won't...
3999                  */
4000                 rdev_for_each_list(rdev, tmp, candidates)
4001                         export_rdev(rdev);
4002                 mddev_put(mddev);
4003         }
4004         printk(KERN_INFO "md: ... autorun DONE.\n");
4005 }
4006 #endif /* !MODULE */
4007
4008 static int get_version(void __user * arg)
4009 {
4010         mdu_version_t ver;
4011
4012         ver.major = MD_MAJOR_VERSION;
4013         ver.minor = MD_MINOR_VERSION;
4014         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4015
4016         if (copy_to_user(arg, &ver, sizeof(ver)))
4017                 return -EFAULT;
4018
4019         return 0;
4020 }
4021
4022 static int get_array_info(mddev_t * mddev, void __user * arg)
4023 {
4024         mdu_array_info_t info;
4025         int nr,working,active,failed,spare;
4026         mdk_rdev_t *rdev;
4027         struct list_head *tmp;
4028
4029         nr=working=active=failed=spare=0;
4030         rdev_for_each(rdev, tmp, mddev) {
4031                 nr++;
4032                 if (test_bit(Faulty, &rdev->flags))
4033                         failed++;
4034                 else {
4035                         working++;
4036                         if (test_bit(In_sync, &rdev->flags))
4037                                 active++;
4038                         else
4039                                 spare++;
4040                 }
4041         }
4042
4043         info.major_version = mddev->major_version;
4044         info.minor_version = mddev->minor_version;
4045         info.patch_version = MD_PATCHLEVEL_VERSION;
4046         info.ctime         = mddev->ctime;
4047         info.level         = mddev->level;
4048         info.size          = mddev->size;
4049         if (info.size != mddev->size) /* overflow */
4050                 info.size = -1;
4051         info.nr_disks      = nr;
4052         info.raid_disks    = mddev->raid_disks;
4053         info.md_minor      = mddev->md_minor;
4054         info.not_persistent= !mddev->persistent;
4055
4056         info.utime         = mddev->utime;
4057         info.state         = 0;
4058         if (mddev->in_sync)
4059                 info.state = (1<<MD_SB_CLEAN);
4060         if (mddev->bitmap && mddev->bitmap_offset)
4061                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4062         info.active_disks  = active;
4063         info.working_disks = working;
4064         info.failed_disks  = failed;
4065         info.spare_disks   = spare;
4066
4067         info.layout        = mddev->layout;
4068         info.chunk_size    = mddev->chunk_size;
4069
4070         if (copy_to_user(arg, &info, sizeof(info)))
4071                 return -EFAULT;
4072
4073         return 0;
4074 }
4075
4076 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4077 {
4078         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4079         char *ptr, *buf = NULL;
4080         int err = -ENOMEM;
4081
4082         md_allow_write(mddev);
4083
4084         file = kmalloc(sizeof(*file), GFP_KERNEL);
4085         if (!file)
4086                 goto out;
4087
4088         /* bitmap disabled, zero the first byte and copy out */
4089         if (!mddev->bitmap || !mddev->bitmap->file) {
4090                 file->pathname[0] = '\0';
4091                 goto copy_out;
4092         }
4093
4094         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4095         if (!buf)
4096                 goto out;
4097
4098         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4099         if (IS_ERR(ptr))
4100                 goto out;
4101
4102         strcpy(file->pathname, ptr);
4103
4104 copy_out:
4105         err = 0;
4106         if (copy_to_user(arg, file, sizeof(*file)))
4107                 err = -EFAULT;
4108 out:
4109         kfree(buf);
4110         kfree(file);
4111         return err;
4112 }
4113
4114 static int get_disk_info(mddev_t * mddev, void __user * arg)
4115 {
4116         mdu_disk_info_t info;
4117         unsigned int nr;
4118         mdk_rdev_t *rdev;
4119
4120         if (copy_from_user(&info, arg, sizeof(info)))
4121                 return -EFAULT;
4122
4123         nr = info.number;
4124
4125         rdev = find_rdev_nr(mddev, nr);
4126         if (rdev) {
4127                 info.major = MAJOR(rdev->bdev->bd_dev);
4128                 info.minor = MINOR(rdev->bdev->bd_dev);
4129                 info.raid_disk = rdev->raid_disk;
4130                 info.state = 0;
4131                 if (test_bit(Faulty, &rdev->flags))
4132                         info.state |= (1<<MD_DISK_FAULTY);
4133                 else if (test_bit(In_sync, &rdev->flags)) {
4134                         info.state |= (1<<MD_DISK_ACTIVE);
4135                         info.state |= (1<<MD_DISK_SYNC);
4136                 }
4137                 if (test_bit(WriteMostly, &rdev->flags))
4138                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4139         } else {
4140                 info.major = info.minor = 0;
4141                 info.raid_disk = -1;
4142                 info.state = (1<<MD_DISK_REMOVED);
4143         }
4144
4145         if (copy_to_user(arg, &info, sizeof(info)))
4146                 return -EFAULT;
4147
4148         return 0;
4149 }
4150
4151 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4152 {
4153         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4154         mdk_rdev_t *rdev;
4155         dev_t dev = MKDEV(info->major,info->minor);
4156
4157         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4158                 return -EOVERFLOW;
4159
4160         if (!mddev->raid_disks) {
4161                 int err;
4162                 /* expecting a device which has a superblock */
4163                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4164                 if (IS_ERR(rdev)) {
4165                         printk(KERN_WARNING
4166                                 "md: md_import_device returned %ld\n",
4167                                 PTR_ERR(rdev));
4168                         return PTR_ERR(rdev);
4169                 }
4170                 if (!list_empty(&mddev->disks)) {
4171                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4172                                                         mdk_rdev_t, same_set);
4173                         int err = super_types[mddev->major_version]
4174                                 .load_super(rdev, rdev0, mddev->minor_version);
4175                         if (err < 0) {
4176                                 printk(KERN_WARNING
4177                                         "md: %s has different UUID to %s\n",
4178                                         bdevname(rdev->bdev,b),
4179                                         bdevname(rdev0->bdev,b2));
4180                                 export_rdev(rdev);
4181                                 return -EINVAL;
4182                         }
4183                 }
4184                 err = bind_rdev_to_array(rdev, mddev);
4185                 if (err)
4186                         export_rdev(rdev);
4187                 return err;
4188         }
4189
4190         /*
4191          * add_new_disk can be used once the array is assembled
4192          * to add "hot spares".  They must already have a superblock
4193          * written
4194          */
4195         if (mddev->pers) {
4196                 int err;
4197                 if (!mddev->pers->hot_add_disk) {
4198                         printk(KERN_WARNING
4199                                 "%s: personality does not support diskops!\n",
4200                                mdname(mddev));
4201                         return -EINVAL;
4202                 }
4203                 if (mddev->persistent)
4204                         rdev = md_import_device(dev, mddev->major_version,
4205                                                 mddev->minor_version);
4206                 else
4207                         rdev = md_import_device(dev, -1, -1);
4208                 if (IS_ERR(rdev)) {
4209                         printk(KERN_WARNING
4210                                 "md: md_import_device returned %ld\n",
4211                                 PTR_ERR(rdev));
4212                         return PTR_ERR(rdev);
4213                 }
4214                 /* set save_raid_disk if appropriate */
4215                 if (!mddev->persistent) {
4216                         if (info->state & (1<<MD_DISK_SYNC)  &&
4217                             info->raid_disk < mddev->raid_disks)
4218                                 rdev->raid_disk = info->raid_disk;
4219                         else
4220                                 rdev->raid_disk = -1;
4221                 } else
4222                         super_types[mddev->major_version].
4223                                 validate_super(mddev, rdev);
4224                 rdev->saved_raid_disk = rdev->raid_disk;
4225
4226                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4227                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4228                         set_bit(WriteMostly, &rdev->flags);
4229
4230                 rdev->raid_disk = -1;
4231                 err = bind_rdev_to_array(rdev, mddev);
4232                 if (!err && !mddev->pers->hot_remove_disk) {
4233                         /* If there is hot_add_disk but no hot_remove_disk
4234                          * then added disks for geometry changes,
4235                          * and should be added immediately.
4236                          */
4237                         super_types[mddev->major_version].
4238                                 validate_super(mddev, rdev);
4239                         err = mddev->pers->hot_add_disk(mddev, rdev);
4240                         if (err)
4241                                 unbind_rdev_from_array(rdev);
4242                 }
4243                 if (err)
4244                         export_rdev(rdev);
4245
4246                 md_update_sb(mddev, 1);
4247                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4248                 md_wakeup_thread(mddev->thread);
4249                 return err;
4250         }
4251
4252         /* otherwise, add_new_disk is only allowed
4253          * for major_version==0 superblocks
4254          */
4255         if (mddev->major_version != 0) {
4256                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4257                        mdname(mddev));
4258                 return -EINVAL;
4259         }
4260
4261         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4262                 int err;
4263                 rdev = md_import_device (dev, -1, 0);
4264                 if (IS_ERR(rdev)) {
4265                         printk(KERN_WARNING
4266                                 "md: error, md_import_device() returned %ld\n",
4267                                 PTR_ERR(rdev));
4268                         return PTR_ERR(rdev);
4269                 }
4270                 rdev->desc_nr = info->number;
4271                 if (info->raid_disk < mddev->raid_disks)
4272                         rdev->raid_disk = info->raid_disk;
4273                 else
4274                         rdev->raid_disk = -1;
4275
4276                 if (rdev->raid_disk < mddev->raid_disks)
4277                         if (info->state & (1<<MD_DISK_SYNC))
4278                                 set_bit(In_sync, &rdev->flags);
4279
4280                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4281                         set_bit(WriteMostly, &rdev->flags);
4282
4283                 if (!mddev->persistent) {
4284                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4285                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4286                 } else
4287                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4288                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4289
4290                 err = bind_rdev_to_array(rdev, mddev);
4291                 if (err) {
4292                         export_rdev(rdev);
4293                         return err;
4294                 }
4295         }
4296
4297         return 0;
4298 }
4299
4300 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4301 {
4302         char b[BDEVNAME_SIZE];
4303         mdk_rdev_t *rdev;
4304
4305         if (!mddev->pers)
4306                 return -ENODEV;
4307
4308         rdev = find_rdev(mddev, dev);
4309         if (!rdev)
4310                 return -ENXIO;
4311
4312         if (rdev->raid_disk >= 0)
4313                 goto busy;
4314
4315         kick_rdev_from_array(rdev);
4316         md_update_sb(mddev, 1);
4317         md_new_event(mddev);
4318
4319         return 0;
4320 busy:
4321         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4322                 bdevname(rdev->bdev,b), mdname(mddev));
4323         return -EBUSY;
4324 }
4325
4326 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4327 {
4328         char b[BDEVNAME_SIZE];
4329         int err;
4330         unsigned int size;
4331         mdk_rdev_t *rdev;
4332
4333         if (!mddev->pers)
4334                 return -ENODEV;
4335
4336         if (mddev->major_version != 0) {
4337                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4338                         " version-0 superblocks.\n",
4339                         mdname(mddev));
4340                 return -EINVAL;
4341         }
4342         if (!mddev->pers->hot_add_disk) {
4343                 printk(KERN_WARNING
4344                         "%s: personality does not support diskops!\n",
4345                         mdname(mddev));
4346                 return -EINVAL;
4347         }
4348
4349         rdev = md_import_device (dev, -1, 0);
4350         if (IS_ERR(rdev)) {
4351                 printk(KERN_WARNING
4352                         "md: error, md_import_device() returned %ld\n",
4353                         PTR_ERR(rdev));
4354                 return -EINVAL;
4355         }
4356
4357         if (mddev->persistent)
4358                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4359         else
4360                 rdev->sb_offset =
4361                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4362
4363         size = calc_dev_size(rdev, mddev->chunk_size);
4364         rdev->size = size;
4365
4366         if (test_bit(Faulty, &rdev->flags)) {
4367                 printk(KERN_WARNING
4368                         "md: can not hot-add faulty %s disk to %s!\n",
4369                         bdevname(rdev->bdev,b), mdname(mddev));
4370                 err = -EINVAL;
4371                 goto abort_export;
4372         }
4373         clear_bit(In_sync, &rdev->flags);
4374         rdev->desc_nr = -1;
4375         rdev->saved_raid_disk = -1;
4376         err = bind_rdev_to_array(rdev, mddev);
4377         if (err)
4378                 goto abort_export;
4379
4380         /*
4381          * The rest should better be atomic, we can have disk failures
4382          * noticed in interrupt contexts ...
4383          */
4384
4385         if (rdev->desc_nr == mddev->max_disks) {
4386                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4387                         mdname(mddev));
4388                 err = -EBUSY;
4389                 goto abort_unbind_export;
4390         }
4391
4392         rdev->raid_disk = -1;
4393
4394         md_update_sb(mddev, 1);
4395
4396         /*
4397          * Kick recovery, maybe this spare has to be added to the
4398          * array immediately.
4399          */
4400         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4401         md_wakeup_thread(mddev->thread);
4402         md_new_event(mddev);
4403         return 0;
4404
4405 abort_unbind_export:
4406         unbind_rdev_from_array(rdev);
4407
4408 abort_export:
4409         export_rdev(rdev);
4410         return err;
4411 }
4412
4413 static int set_bitmap_file(mddev_t *mddev, int fd)
4414 {
4415         int err;
4416
4417         if (mddev->pers) {
4418                 if (!mddev->pers->quiesce)
4419                         return -EBUSY;
4420                 if (mddev->recovery || mddev->sync_thread)
4421                         return -EBUSY;
4422                 /* we should be able to change the bitmap.. */
4423         }
4424
4425
4426         if (fd >= 0) {
4427                 if (mddev->bitmap)
4428                         return -EEXIST; /* cannot add when bitmap is present */
4429                 mddev->bitmap_file = fget(fd);
4430
4431                 if (mddev->bitmap_file == NULL) {
4432                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4433                                mdname(mddev));
4434                         return -EBADF;
4435                 }
4436
4437                 err = deny_bitmap_write_access(mddev->bitmap_file);
4438                 if (err) {
4439                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4440                                mdname(mddev));
4441                         fput(mddev->bitmap_file);
4442                         mddev->bitmap_file = NULL;
4443                         return err;
4444                 }
4445                 mddev->bitmap_offset = 0; /* file overrides offset */
4446         } else if (mddev->bitmap == NULL)
4447                 return -ENOENT; /* cannot remove what isn't there */
4448         err = 0;
4449         if (mddev->pers) {
4450                 mddev->pers->quiesce(mddev, 1);
4451                 if (fd >= 0)
4452                         err = bitmap_create(mddev);
4453                 if (fd < 0 || err) {
4454                         bitmap_destroy(mddev);
4455                         fd = -1; /* make sure to put the file */
4456                 }
4457                 mddev->pers->quiesce(mddev, 0);
4458         }
4459         if (fd < 0) {
4460                 if (mddev->bitmap_file) {
4461                         restore_bitmap_write_access(mddev->bitmap_file);
4462                         fput(mddev->bitmap_file);
4463                 }
4464                 mddev->bitmap_file = NULL;
4465         }
4466
4467         return err;
4468 }
4469
4470 /*
4471  * set_array_info is used two different ways
4472  * The original usage is when creating a new array.
4473  * In this usage, raid_disks is > 0 and it together with
4474  *  level, size, not_persistent,layout,chunksize determine the
4475  *  shape of the array.
4476  *  This will always create an array with a type-0.90.0 superblock.
4477  * The newer usage is when assembling an array.
4478  *  In this case raid_disks will be 0, and the major_version field is
4479  *  use to determine which style super-blocks are to be found on the devices.
4480  *  The minor and patch _version numbers are also kept incase the
4481  *  super_block handler wishes to interpret them.
4482  */
4483 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4484 {
4485
4486         if (info->raid_disks == 0) {
4487                 /* just setting version number for superblock loading */
4488                 if (info->major_version < 0 ||
4489                     info->major_version >= ARRAY_SIZE(super_types) ||
4490                     super_types[info->major_version].name == NULL) {
4491                         /* maybe try to auto-load a module? */
4492                         printk(KERN_INFO
4493                                 "md: superblock version %d not known\n",
4494                                 info->major_version);
4495                         return -EINVAL;
4496                 }
4497                 mddev->major_version = info->major_version;
4498                 mddev->minor_version = info->minor_version;
4499                 mddev->patch_version = info->patch_version;
4500                 mddev->persistent = !info->not_persistent;
4501                 return 0;
4502         }
4503         mddev->major_version = MD_MAJOR_VERSION;
4504         mddev->minor_version = MD_MINOR_VERSION;
4505         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4506         mddev->ctime         = get_seconds();
4507
4508         mddev->level         = info->level;
4509         mddev->clevel[0]     = 0;
4510         mddev->size          = info->size;
4511         mddev->raid_disks    = info->raid_disks;
4512         /* don't set md_minor, it is determined by which /dev/md* was
4513          * openned
4514          */
4515         if (info->state & (1<<MD_SB_CLEAN))
4516                 mddev->recovery_cp = MaxSector;
4517         else
4518                 mddev->recovery_cp = 0;
4519         mddev->persistent    = ! info->not_persistent;
4520         mddev->external      = 0;
4521
4522         mddev->layout        = info->layout;
4523         mddev->chunk_size    = info->chunk_size;
4524
4525         mddev->max_disks     = MD_SB_DISKS;
4526
4527         if (mddev->persistent)
4528                 mddev->flags         = 0;
4529         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4530
4531         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4532         mddev->bitmap_offset = 0;
4533
4534         mddev->reshape_position = MaxSector;
4535
4536         /*
4537          * Generate a 128 bit UUID
4538          */
4539         get_random_bytes(mddev->uuid, 16);
4540
4541         mddev->new_level = mddev->level;
4542         mddev->new_chunk = mddev->chunk_size;
4543         mddev->new_layout = mddev->layout;
4544         mddev->delta_disks = 0;
4545
4546         return 0;
4547 }
4548
4549 static int update_size(mddev_t *mddev, unsigned long size)
4550 {
4551         mdk_rdev_t * rdev;
4552         int rv;
4553         struct list_head *tmp;
4554         int fit = (size == 0);
4555
4556         if (mddev->pers->resize == NULL)
4557                 return -EINVAL;
4558         /* The "size" is the amount of each device that is used.
4559          * This can only make sense for arrays with redundancy.
4560          * linear and raid0 always use whatever space is available
4561          * We can only consider changing the size if no resync
4562          * or reconstruction is happening, and if the new size
4563          * is acceptable. It must fit before the sb_offset or,
4564          * if that is <data_offset, it must fit before the
4565          * size of each device.
4566          * If size is zero, we find the largest size that fits.
4567          */
4568         if (mddev->sync_thread)
4569                 return -EBUSY;
4570         rdev_for_each(rdev, tmp, mddev) {
4571                 sector_t avail;
4572                 avail = rdev->size * 2;
4573
4574                 if (fit && (size == 0 || size > avail/2))
4575                         size = avail/2;
4576                 if (avail < ((sector_t)size << 1))
4577                         return -ENOSPC;
4578         }
4579         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4580         if (!rv) {
4581                 struct block_device *bdev;
4582
4583                 bdev = bdget_disk(mddev->gendisk, 0);
4584                 if (bdev) {
4585                         mutex_lock(&bdev->bd_inode->i_mutex);
4586                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4587                         mutex_unlock(&bdev->bd_inode->i_mutex);
4588                         bdput(bdev);
4589                 }
4590         }
4591         return rv;
4592 }
4593
4594 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4595 {
4596         int rv;
4597         /* change the number of raid disks */
4598         if (mddev->pers->check_reshape == NULL)
4599                 return -EINVAL;
4600         if (raid_disks <= 0 ||
4601             raid_disks >= mddev->max_disks)
4602                 return -EINVAL;
4603         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4604                 return -EBUSY;
4605         mddev->delta_disks = raid_disks - mddev->raid_disks;
4606
4607         rv = mddev->pers->check_reshape(mddev);
4608         return rv;
4609 }
4610
4611
4612 /*
4613  * update_array_info is used to change the configuration of an
4614  * on-line array.
4615  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4616  * fields in the info are checked against the array.
4617  * Any differences that cannot be handled will cause an error.
4618  * Normally, only one change can be managed at a time.
4619  */
4620 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4621 {
4622         int rv = 0;
4623         int cnt = 0;
4624         int state = 0;
4625
4626         /* calculate expected state,ignoring low bits */
4627         if (mddev->bitmap && mddev->bitmap_offset)
4628                 state |= (1 << MD_SB_BITMAP_PRESENT);
4629
4630         if (mddev->major_version != info->major_version ||
4631             mddev->minor_version != info->minor_version ||
4632 /*          mddev->patch_version != info->patch_version || */
4633             mddev->ctime         != info->ctime         ||
4634             mddev->level         != info->level         ||
4635 /*          mddev->layout        != info->layout        || */
4636             !mddev->persistent   != info->not_persistent||
4637             mddev->chunk_size    != info->chunk_size    ||
4638             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4639             ((state^info->state) & 0xfffffe00)
4640                 )
4641                 return -EINVAL;
4642         /* Check there is only one change */
4643         if (info->size >= 0 && mddev->size != info->size) cnt++;
4644         if (mddev->raid_disks != info->raid_disks) cnt++;
4645         if (mddev->layout != info->layout) cnt++;
4646         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4647         if (cnt == 0) return 0;
4648         if (cnt > 1) return -EINVAL;
4649
4650         if (mddev->layout != info->layout) {
4651                 /* Change layout
4652                  * we don't need to do anything at the md level, the
4653                  * personality will take care of it all.
4654                  */
4655                 if (mddev->pers->reconfig == NULL)
4656                         return -EINVAL;
4657                 else
4658                         return mddev->pers->reconfig(mddev, info->layout, -1);
4659         }
4660         if (info->size >= 0 && mddev->size != info->size)
4661                 rv = update_size(mddev, info->size);
4662
4663         if (mddev->raid_disks    != info->raid_disks)
4664                 rv = update_raid_disks(mddev, info->raid_disks);
4665
4666         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4667                 if (mddev->pers->quiesce == NULL)
4668                         return -EINVAL;
4669                 if (mddev->recovery || mddev->sync_thread)
4670                         return -EBUSY;
4671                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4672                         /* add the bitmap */
4673                         if (mddev->bitmap)
4674                                 return -EEXIST;
4675                         if (mddev->default_bitmap_offset == 0)
4676                                 return -EINVAL;
4677                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4678                         mddev->pers->quiesce(mddev, 1);
4679                         rv = bitmap_create(mddev);
4680                         if (rv)
4681                                 bitmap_destroy(mddev);
4682                         mddev->pers->quiesce(mddev, 0);
4683                 } else {
4684                         /* remove the bitmap */
4685                         if (!mddev->bitmap)
4686                                 return -ENOENT;
4687                         if (mddev->bitmap->file)
4688                                 return -EINVAL;
4689                         mddev->pers->quiesce(mddev, 1);
4690                         bitmap_destroy(mddev);
4691                         mddev->pers->quiesce(mddev, 0);
4692                         mddev->bitmap_offset = 0;
4693                 }
4694         }
4695         md_update_sb(mddev, 1);
4696         return rv;
4697 }
4698
4699 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4700 {
4701         mdk_rdev_t *rdev;
4702
4703         if (mddev->pers == NULL)
4704                 return -ENODEV;
4705
4706         rdev = find_rdev(mddev, dev);
4707         if (!rdev)
4708                 return -ENODEV;
4709
4710         md_error(mddev, rdev);
4711         return 0;
4712 }
4713
4714 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4715 {
4716         mddev_t *mddev = bdev->bd_disk->private_data;
4717
4718         geo->heads = 2;
4719         geo->sectors = 4;
4720         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4721         return 0;
4722 }
4723
4724 static int md_ioctl(struct inode *inode, struct file *file,
4725                         unsigned int cmd, unsigned long arg)
4726 {
4727         int err = 0;
4728         void __user *argp = (void __user *)arg;
4729         mddev_t *mddev = NULL;
4730
4731         if (!capable(CAP_SYS_ADMIN))
4732                 return -EACCES;
4733
4734         /*
4735          * Commands dealing with the RAID driver but not any
4736          * particular array:
4737          */
4738         switch (cmd)
4739         {
4740                 case RAID_VERSION:
4741                         err = get_version(argp);
4742                         goto done;
4743
4744                 case PRINT_RAID_DEBUG:
4745                         err = 0;
4746                         md_print_devices();
4747                         goto done;
4748
4749 #ifndef MODULE
4750                 case RAID_AUTORUN:
4751                         err = 0;
4752                         autostart_arrays(arg);
4753                         goto done;
4754 #endif
4755                 default:;
4756         }
4757
4758         /*
4759          * Commands creating/starting a new array:
4760          */
4761
4762         mddev = inode->i_bdev->bd_disk->private_data;
4763
4764         if (!mddev) {
4765                 BUG();
4766                 goto abort;
4767         }
4768
4769         err = mddev_lock(mddev);
4770         if (err) {
4771                 printk(KERN_INFO
4772                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4773                         err, cmd);
4774                 goto abort;
4775         }
4776
4777         switch (cmd)
4778         {
4779                 case SET_ARRAY_INFO:
4780                         {
4781                                 mdu_array_info_t info;
4782                                 if (!arg)
4783                                         memset(&info, 0, sizeof(info));
4784                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4785                                         err = -EFAULT;
4786                                         goto abort_unlock;
4787                                 }
4788                                 if (mddev->pers) {
4789                                         err = update_array_info(mddev, &info);
4790                                         if (err) {
4791                                                 printk(KERN_WARNING "md: couldn't update"
4792                                                        " array info. %d\n", err);
4793                                                 goto abort_unlock;
4794                                         }
4795                                         goto done_unlock;
4796                                 }
4797                                 if (!list_empty(&mddev->disks)) {
4798                                         printk(KERN_WARNING
4799                                                "md: array %s already has disks!\n",
4800                                                mdname(mddev));
4801                                         err = -EBUSY;
4802                                         goto abort_unlock;
4803                                 }
4804                                 if (mddev->raid_disks) {
4805                                         printk(KERN_WARNING
4806                                                "md: array %s already initialised!\n",
4807                                                mdname(mddev));
4808                                         err = -EBUSY;
4809                                         goto abort_unlock;
4810                                 }
4811                                 err = set_array_info(mddev, &info);
4812                                 if (err) {
4813                                         printk(KERN_WARNING "md: couldn't set"
4814                                                " array info. %d\n", err);
4815                                         goto abort_unlock;
4816                                 }
4817                         }
4818                         goto done_unlock;
4819
4820                 default:;
4821         }
4822
4823         /*
4824          * Commands querying/configuring an existing array:
4825          */
4826         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4827          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4828         if ((!mddev->raid_disks && !mddev->external)
4829             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4830             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4831             && cmd != GET_BITMAP_FILE) {
4832                 err = -ENODEV;
4833                 goto abort_unlock;
4834         }
4835
4836         /*
4837          * Commands even a read-only array can execute:
4838          */
4839         switch (cmd)
4840         {
4841                 case GET_ARRAY_INFO:
4842                         err = get_array_info(mddev, argp);
4843                         goto done_unlock;
4844
4845                 case GET_BITMAP_FILE:
4846                         err = get_bitmap_file(mddev, argp);
4847                         goto done_unlock;
4848
4849                 case GET_DISK_INFO:
4850                         err = get_disk_info(mddev, argp);
4851                         goto done_unlock;
4852
4853                 case RESTART_ARRAY_RW:
4854                         err = restart_array(mddev);
4855                         goto done_unlock;
4856
4857                 case STOP_ARRAY:
4858                         err = do_md_stop (mddev, 0);
4859                         goto done_unlock;
4860
4861                 case STOP_ARRAY_RO:
4862                         err = do_md_stop (mddev, 1);
4863                         goto done_unlock;
4864
4865         /*
4866          * We have a problem here : there is no easy way to give a CHS
4867          * virtual geometry. We currently pretend that we have a 2 heads
4868          * 4 sectors (with a BIG number of cylinders...). This drives
4869          * dosfs just mad... ;-)
4870          */
4871         }
4872
4873         /*
4874          * The remaining ioctls are changing the state of the
4875          * superblock, so we do not allow them on read-only arrays.
4876          * However non-MD ioctls (e.g. get-size) will still come through
4877          * here and hit the 'default' below, so only disallow
4878          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4879          */
4880         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4881             mddev->ro && mddev->pers) {
4882                 if (mddev->ro == 2) {
4883                         mddev->ro = 0;
4884                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4885                 md_wakeup_thread(mddev->thread);
4886
4887                 } else {
4888                         err = -EROFS;
4889                         goto abort_unlock;
4890                 }
4891         }
4892
4893         switch (cmd)
4894         {
4895                 case ADD_NEW_DISK:
4896                 {
4897                         mdu_disk_info_t info;
4898                         if (copy_from_user(&info, argp, sizeof(info)))
4899                                 err = -EFAULT;
4900                         else
4901                                 err = add_new_disk(mddev, &info);
4902                         goto done_unlock;
4903                 }
4904
4905                 case HOT_REMOVE_DISK:
4906                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4907                         goto done_unlock;
4908
4909                 case HOT_ADD_DISK:
4910                         err = hot_add_disk(mddev, new_decode_dev(arg));
4911                         goto done_unlock;
4912
4913                 case SET_DISK_FAULTY:
4914                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4915                         goto done_unlock;
4916
4917                 case RUN_ARRAY:
4918                         err = do_md_run (mddev);
4919                         goto done_unlock;
4920
4921                 case SET_BITMAP_FILE:
4922                         err = set_bitmap_file(mddev, (int)arg);
4923                         goto done_unlock;
4924
4925                 default:
4926                         err = -EINVAL;
4927                         goto abort_unlock;
4928         }
4929
4930 done_unlock:
4931 abort_unlock:
4932         mddev_unlock(mddev);
4933
4934         return err;
4935 done:
4936         if (err)
4937                 MD_BUG();
4938 abort:
4939         return err;
4940 }
4941
4942 static int md_open(struct inode *inode, struct file *file)
4943 {
4944         /*
4945          * Succeed if we can lock the mddev, which confirms that
4946          * it isn't being stopped right now.
4947          */
4948         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4949         int err;
4950
4951         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4952                 goto out;
4953
4954         err = 0;
4955         mddev_get(mddev);
4956         mddev_unlock(mddev);
4957
4958         check_disk_change(inode->i_bdev);
4959  out:
4960         return err;
4961 }
4962
4963 static int md_release(struct inode *inode, struct file * file)
4964 {
4965         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4966
4967         BUG_ON(!mddev);
4968         mddev_put(mddev);
4969
4970         return 0;
4971 }
4972
4973 static int md_media_changed(struct gendisk *disk)
4974 {
4975         mddev_t *mddev = disk->private_data;
4976
4977         return mddev->changed;
4978 }
4979
4980 static int md_revalidate(struct gendisk *disk)
4981 {
4982         mddev_t *mddev = disk->private_data;
4983
4984         mddev->changed = 0;
4985         return 0;
4986 }
4987 static struct block_device_operations md_fops =
4988 {
4989         .owner          = THIS_MODULE,
4990         .open           = md_open,
4991         .release        = md_release,
4992         .ioctl          = md_ioctl,
4993         .getgeo         = md_getgeo,
4994         .media_changed  = md_media_changed,
4995         .revalidate_disk= md_revalidate,
4996 };
4997
4998 static int md_thread(void * arg)
4999 {
5000         mdk_thread_t *thread = arg;
5001
5002         /*
5003          * md_thread is a 'system-thread', it's priority should be very
5004          * high. We avoid resource deadlocks individually in each
5005          * raid personality. (RAID5 does preallocation) We also use RR and
5006          * the very same RT priority as kswapd, thus we will never get
5007          * into a priority inversion deadlock.
5008          *
5009          * we definitely have to have equal or higher priority than
5010          * bdflush, otherwise bdflush will deadlock if there are too
5011          * many dirty RAID5 blocks.
5012          */
5013
5014         allow_signal(SIGKILL);
5015         while (!kthread_should_stop()) {
5016
5017                 /* We need to wait INTERRUPTIBLE so that
5018                  * we don't add to the load-average.
5019                  * That means we need to be sure no signals are
5020                  * pending
5021                  */
5022                 if (signal_pending(current))
5023                         flush_signals(current);
5024
5025                 wait_event_interruptible_timeout
5026                         (thread->wqueue,
5027                          test_bit(THREAD_WAKEUP, &thread->flags)
5028                          || kthread_should_stop(),
5029                          thread->timeout);
5030
5031                 clear_bit(THREAD_WAKEUP, &thread->flags);
5032
5033                 thread->run(thread->mddev);
5034         }
5035
5036         return 0;
5037 }
5038
5039 void md_wakeup_thread(mdk_thread_t *thread)
5040 {
5041         if (thread) {
5042                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5043                 set_bit(THREAD_WAKEUP, &thread->flags);
5044                 wake_up(&thread->wqueue);
5045         }
5046 }
5047
5048 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5049                                  const char *name)
5050 {
5051         mdk_thread_t *thread;
5052
5053         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5054         if (!thread)
5055                 return NULL;
5056
5057         init_waitqueue_head(&thread->wqueue);
5058
5059         thread->run = run;
5060         thread->mddev = mddev;
5061         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5062         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5063         if (IS_ERR(thread->tsk)) {
5064                 kfree(thread);
5065                 return NULL;
5066         }
5067         return thread;
5068 }
5069
5070 void md_unregister_thread(mdk_thread_t *thread)
5071 {
5072         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5073
5074         kthread_stop(thread->tsk);
5075         kfree(thread);
5076 }
5077
5078 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5079 {
5080         if (!mddev) {
5081                 MD_BUG();
5082                 return;
5083         }
5084
5085         if (!rdev || test_bit(Faulty, &rdev->flags))
5086                 return;
5087
5088         if (mddev->external)
5089                 set_bit(Blocked, &rdev->flags);
5090 /*
5091         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5092                 mdname(mddev),
5093                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5094                 __builtin_return_address(0),__builtin_return_address(1),
5095                 __builtin_return_address(2),__builtin_return_address(3));
5096 */
5097         if (!mddev->pers)
5098                 return;
5099         if (!mddev->pers->error_handler)
5100                 return;
5101         mddev->pers->error_handler(mddev,rdev);
5102         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5103         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5104         md_wakeup_thread(mddev->thread);
5105         md_new_event_inintr(mddev);
5106 }
5107
5108 /* seq_file implementation /proc/mdstat */
5109
5110 static void status_unused(struct seq_file *seq)
5111 {
5112         int i = 0;
5113         mdk_rdev_t *rdev;
5114         struct list_head *tmp;
5115
5116         seq_printf(seq, "unused devices: ");
5117
5118         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5119                 char b[BDEVNAME_SIZE];
5120                 i++;
5121                 seq_printf(seq, "%s ",
5122                               bdevname(rdev->bdev,b));
5123         }
5124         if (!i)
5125                 seq_printf(seq, "<none>");
5126
5127         seq_printf(seq, "\n");
5128 }
5129
5130
5131 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5132 {
5133         sector_t max_blocks, resync, res;
5134         unsigned long dt, db, rt;
5135         int scale;
5136         unsigned int per_milli;
5137
5138         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5139
5140         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5141                 max_blocks = mddev->resync_max_sectors >> 1;
5142         else
5143                 max_blocks = mddev->size;
5144
5145         /*
5146          * Should not happen.
5147          */
5148         if (!max_blocks) {
5149                 MD_BUG();
5150                 return;
5151         }
5152         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5153          * in a sector_t, and (max_blocks>>scale) will fit in a
5154          * u32, as those are the requirements for sector_div.
5155          * Thus 'scale' must be at least 10
5156          */
5157         scale = 10;
5158         if (sizeof(sector_t) > sizeof(unsigned long)) {
5159                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5160                         scale++;
5161         }
5162         res = (resync>>scale)*1000;
5163         sector_div(res, (u32)((max_blocks>>scale)+1));
5164
5165         per_milli = res;
5166         {
5167                 int i, x = per_milli/50, y = 20-x;
5168                 seq_printf(seq, "[");
5169                 for (i = 0; i < x; i++)
5170                         seq_printf(seq, "=");
5171                 seq_printf(seq, ">");
5172                 for (i = 0; i < y; i++)
5173                         seq_printf(seq, ".");
5174                 seq_printf(seq, "] ");
5175         }
5176         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5177                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5178                     "reshape" :
5179                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5180                      "check" :
5181                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5182                       "resync" : "recovery"))),
5183                    per_milli/10, per_milli % 10,
5184                    (unsigned long long) resync,
5185                    (unsigned long long) max_blocks);
5186
5187         /*
5188          * We do not want to overflow, so the order of operands and
5189          * the * 100 / 100 trick are important. We do a +1 to be
5190          * safe against division by zero. We only estimate anyway.
5191          *
5192          * dt: time from mark until now
5193          * db: blocks written from mark until now
5194          * rt: remaining time
5195          */
5196         dt = ((jiffies - mddev->resync_mark) / HZ);
5197         if (!dt) dt++;
5198         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5199                 - mddev->resync_mark_cnt;
5200         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5201
5202         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5203
5204         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5205 }
5206
5207 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5208 {
5209         struct list_head *tmp;
5210         loff_t l = *pos;
5211         mddev_t *mddev;
5212
5213         if (l >= 0x10000)
5214                 return NULL;
5215         if (!l--)
5216                 /* header */
5217                 return (void*)1;
5218
5219         spin_lock(&all_mddevs_lock);
5220         list_for_each(tmp,&all_mddevs)
5221                 if (!l--) {
5222                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5223                         mddev_get(mddev);
5224                         spin_unlock(&all_mddevs_lock);
5225                         return mddev;
5226                 }
5227         spin_unlock(&all_mddevs_lock);
5228         if (!l--)
5229                 return (void*)2;/* tail */
5230         return NULL;
5231 }
5232
5233 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5234 {
5235         struct list_head *tmp;
5236         mddev_t *next_mddev, *mddev = v;
5237
5238         ++*pos;
5239         if (v == (void*)2)
5240                 return NULL;
5241
5242         spin_lock(&all_mddevs_lock);
5243         if (v == (void*)1)
5244                 tmp = all_mddevs.next;
5245         else
5246                 tmp = mddev->all_mddevs.next;
5247         if (tmp != &all_mddevs)
5248                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5249         else {
5250                 next_mddev = (void*)2;
5251                 *pos = 0x10000;
5252         }
5253         spin_unlock(&all_mddevs_lock);
5254
5255         if (v != (void*)1)
5256                 mddev_put(mddev);
5257         return next_mddev;
5258
5259 }
5260
5261 static void md_seq_stop(struct seq_file *seq, void *v)
5262 {
5263         mddev_t *mddev = v;
5264
5265         if (mddev && v != (void*)1 && v != (void*)2)
5266                 mddev_put(mddev);
5267 }
5268
5269 struct mdstat_info {
5270         int event;
5271 };
5272
5273 static int md_seq_show(struct seq_file *seq, void *v)
5274 {
5275         mddev_t *mddev = v;
5276         sector_t size;
5277         struct list_head *tmp2;
5278         mdk_rdev_t *rdev;
5279         struct mdstat_info *mi = seq->private;
5280         struct bitmap *bitmap;
5281
5282         if (v == (void*)1) {
5283                 struct mdk_personality *pers;
5284                 seq_printf(seq, "Personalities : ");
5285                 spin_lock(&pers_lock);
5286                 list_for_each_entry(pers, &pers_list, list)
5287                         seq_printf(seq, "[%s] ", pers->name);
5288
5289                 spin_unlock(&pers_lock);
5290                 seq_printf(seq, "\n");
5291                 mi->event = atomic_read(&md_event_count);
5292                 return 0;
5293         }
5294         if (v == (void*)2) {
5295                 status_unused(seq);
5296                 return 0;
5297         }
5298
5299         if (mddev_lock(mddev) < 0)
5300                 return -EINTR;
5301
5302         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5303                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5304                                                 mddev->pers ? "" : "in");
5305                 if (mddev->pers) {
5306                         if (mddev->ro==1)
5307                                 seq_printf(seq, " (read-only)");
5308                         if (mddev->ro==2)
5309                                 seq_printf(seq, " (auto-read-only)");
5310                         seq_printf(seq, " %s", mddev->pers->name);
5311                 }
5312
5313                 size = 0;
5314                 rdev_for_each(rdev, tmp2, mddev) {
5315                         char b[BDEVNAME_SIZE];
5316                         seq_printf(seq, " %s[%d]",
5317                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5318                         if (test_bit(WriteMostly, &rdev->flags))
5319                                 seq_printf(seq, "(W)");
5320                         if (test_bit(Faulty, &rdev->flags)) {
5321                                 seq_printf(seq, "(F)");
5322                                 continue;
5323                         } else if (rdev->raid_disk < 0)
5324                                 seq_printf(seq, "(S)"); /* spare */
5325                         size += rdev->size;
5326                 }
5327
5328                 if (!list_empty(&mddev->disks)) {
5329                         if (mddev->pers)
5330                                 seq_printf(seq, "\n      %llu blocks",
5331                                         (unsigned long long)mddev->array_size);
5332                         else
5333                                 seq_printf(seq, "\n      %llu blocks",
5334                                         (unsigned long long)size);
5335                 }
5336                 if (mddev->persistent) {
5337                         if (mddev->major_version != 0 ||
5338                             mddev->minor_version != 90) {
5339                                 seq_printf(seq," super %d.%d",
5340                                            mddev->major_version,
5341                                            mddev->minor_version);
5342                         }
5343                 } else if (mddev->external)
5344                         seq_printf(seq, " super external:%s",
5345                                    mddev->metadata_type);
5346                 else
5347                         seq_printf(seq, " super non-persistent");
5348
5349                 if (mddev->pers) {
5350                         mddev->pers->status (seq, mddev);
5351                         seq_printf(seq, "\n      ");
5352                         if (mddev->pers->sync_request) {
5353                                 if (mddev->curr_resync > 2) {
5354                                         status_resync (seq, mddev);
5355                                         seq_printf(seq, "\n      ");
5356                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5357                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5358                                 else if (mddev->recovery_cp < MaxSector)
5359                                         seq_printf(seq, "\tresync=PENDING\n      ");
5360                         }
5361                 } else
5362                         seq_printf(seq, "\n       ");
5363
5364                 if ((bitmap = mddev->bitmap)) {
5365                         unsigned long chunk_kb;
5366                         unsigned long flags;
5367                         spin_lock_irqsave(&bitmap->lock, flags);
5368                         chunk_kb = bitmap->chunksize >> 10;
5369                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5370                                 "%lu%s chunk",
5371                                 bitmap->pages - bitmap->missing_pages,
5372                                 bitmap->pages,
5373                                 (bitmap->pages - bitmap->missing_pages)
5374                                         << (PAGE_SHIFT - 10),
5375                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5376                                 chunk_kb ? "KB" : "B");
5377                         if (bitmap->file) {
5378                                 seq_printf(seq, ", file: ");
5379                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5380                         }
5381
5382                         seq_printf(seq, "\n");
5383                         spin_unlock_irqrestore(&bitmap->lock, flags);
5384                 }
5385
5386                 seq_printf(seq, "\n");
5387         }
5388         mddev_unlock(mddev);
5389
5390         return 0;
5391 }
5392
5393 static struct seq_operations md_seq_ops = {
5394         .start  = md_seq_start,
5395         .next   = md_seq_next,
5396         .stop   = md_seq_stop,
5397         .show   = md_seq_show,
5398 };
5399
5400 static int md_seq_open(struct inode *inode, struct file *file)
5401 {
5402         int error;
5403         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5404         if (mi == NULL)
5405                 return -ENOMEM;
5406
5407         error = seq_open(file, &md_seq_ops);
5408         if (error)
5409                 kfree(mi);
5410         else {
5411                 struct seq_file *p = file->private_data;
5412                 p->private = mi;
5413                 mi->event = atomic_read(&md_event_count);
5414         }
5415         return error;
5416 }
5417
5418 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5419 {
5420         struct seq_file *m = filp->private_data;
5421         struct mdstat_info *mi = m->private;
5422         int mask;
5423
5424         poll_wait(filp, &md_event_waiters, wait);
5425
5426         /* always allow read */
5427         mask = POLLIN | POLLRDNORM;
5428
5429         if (mi->event != atomic_read(&md_event_count))
5430                 mask |= POLLERR | POLLPRI;
5431         return mask;
5432 }
5433
5434 static const struct file_operations md_seq_fops = {
5435         .owner          = THIS_MODULE,
5436         .open           = md_seq_open,
5437         .read           = seq_read,
5438         .llseek         = seq_lseek,
5439         .release        = seq_release_private,
5440         .poll           = mdstat_poll,
5441 };
5442
5443 int register_md_personality(struct mdk_personality *p)
5444 {
5445         spin_lock(&pers_lock);
5446         list_add_tail(&p->list, &pers_list);
5447         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5448         spin_unlock(&pers_lock);
5449         return 0;
5450 }
5451
5452 int unregister_md_personality(struct mdk_personality *p)
5453 {
5454         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5455         spin_lock(&pers_lock);
5456         list_del_init(&p->list);
5457         spin_unlock(&pers_lock);
5458         return 0;
5459 }
5460
5461 static int is_mddev_idle(mddev_t *mddev)
5462 {
5463         mdk_rdev_t * rdev;
5464         struct list_head *tmp;
5465         int idle;
5466         long curr_events;
5467
5468         idle = 1;
5469         rdev_for_each(rdev, tmp, mddev) {
5470                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5471                 curr_events = disk_stat_read(disk, sectors[0]) +
5472                                 disk_stat_read(disk, sectors[1]) -
5473                                 atomic_read(&disk->sync_io);
5474                 /* sync IO will cause sync_io to increase before the disk_stats
5475                  * as sync_io is counted when a request starts, and
5476                  * disk_stats is counted when it completes.
5477                  * So resync activity will cause curr_events to be smaller than
5478                  * when there was no such activity.
5479                  * non-sync IO will cause disk_stat to increase without
5480                  * increasing sync_io so curr_events will (eventually)
5481                  * be larger than it was before.  Once it becomes
5482                  * substantially larger, the test below will cause
5483                  * the array to appear non-idle, and resync will slow
5484                  * down.
5485                  * If there is a lot of outstanding resync activity when
5486                  * we set last_event to curr_events, then all that activity
5487                  * completing might cause the array to appear non-idle
5488                  * and resync will be slowed down even though there might
5489                  * not have been non-resync activity.  This will only
5490                  * happen once though.  'last_events' will soon reflect
5491                  * the state where there is little or no outstanding
5492                  * resync requests, and further resync activity will
5493                  * always make curr_events less than last_events.
5494                  *
5495                  */
5496                 if (curr_events - rdev->last_events > 4096) {
5497                         rdev->last_events = curr_events;
5498                         idle = 0;
5499                 }
5500         }
5501         return idle;
5502 }
5503
5504 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5505 {
5506         /* another "blocks" (512byte) blocks have been synced */
5507         atomic_sub(blocks, &mddev->recovery_active);
5508         wake_up(&mddev->recovery_wait);
5509         if (!ok) {
5510                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5511                 md_wakeup_thread(mddev->thread);
5512                 // stop recovery, signal do_sync ....
5513         }
5514 }
5515
5516
5517 /* md_write_start(mddev, bi)
5518  * If we need to update some array metadata (e.g. 'active' flag
5519  * in superblock) before writing, schedule a superblock update
5520  * and wait for it to complete.
5521  */
5522 void md_write_start(mddev_t *mddev, struct bio *bi)
5523 {
5524         if (bio_data_dir(bi) != WRITE)
5525                 return;
5526
5527         BUG_ON(mddev->ro == 1);
5528         if (mddev->ro == 2) {
5529                 /* need to switch to read/write */
5530                 mddev->ro = 0;
5531                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5532                 md_wakeup_thread(mddev->thread);
5533                 md_wakeup_thread(mddev->sync_thread);
5534         }
5535         atomic_inc(&mddev->writes_pending);
5536         if (mddev->safemode == 1)
5537                 mddev->safemode = 0;
5538         if (mddev->in_sync) {
5539                 spin_lock_irq(&mddev->write_lock);
5540                 if (mddev->in_sync) {
5541                         mddev->in_sync = 0;
5542                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5543                         md_wakeup_thread(mddev->thread);
5544                 }
5545                 spin_unlock_irq(&mddev->write_lock);
5546                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5547         }
5548         wait_event(mddev->sb_wait,
5549                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5550                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5551 }
5552
5553 void md_write_end(mddev_t *mddev)
5554 {
5555         if (atomic_dec_and_test(&mddev->writes_pending)) {
5556                 if (mddev->safemode == 2)
5557                         md_wakeup_thread(mddev->thread);
5558                 else if (mddev->safemode_delay)
5559                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5560         }
5561 }
5562
5563 /* md_allow_write(mddev)
5564  * Calling this ensures that the array is marked 'active' so that writes
5565  * may proceed without blocking.  It is important to call this before
5566  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5567  * Must be called with mddev_lock held.
5568  */
5569 void md_allow_write(mddev_t *mddev)
5570 {
5571         if (!mddev->pers)
5572                 return;
5573         if (mddev->ro)
5574                 return;
5575         if (!mddev->pers->sync_request)
5576                 return;
5577
5578         spin_lock_irq(&mddev->write_lock);
5579         if (mddev->in_sync) {
5580                 mddev->in_sync = 0;
5581                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5582                 if (mddev->safemode_delay &&
5583                     mddev->safemode == 0)
5584                         mddev->safemode = 1;
5585                 spin_unlock_irq(&mddev->write_lock);
5586                 md_update_sb(mddev, 0);
5587
5588                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5589                 /* wait for the dirty state to be recorded in the metadata */
5590                 wait_event(mddev->sb_wait,
5591                            !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5592                            !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5593         } else
5594                 spin_unlock_irq(&mddev->write_lock);
5595 }
5596 EXPORT_SYMBOL_GPL(md_allow_write);
5597
5598 #define SYNC_MARKS      10
5599 #define SYNC_MARK_STEP  (3*HZ)
5600 void md_do_sync(mddev_t *mddev)
5601 {
5602         mddev_t *mddev2;
5603         unsigned int currspeed = 0,
5604                  window;
5605         sector_t max_sectors,j, io_sectors;
5606         unsigned long mark[SYNC_MARKS];
5607         sector_t mark_cnt[SYNC_MARKS];
5608         int last_mark,m;
5609         struct list_head *tmp;
5610         sector_t last_check;
5611         int skipped = 0;
5612         struct list_head *rtmp;
5613         mdk_rdev_t *rdev;
5614         char *desc;
5615
5616         /* just incase thread restarts... */
5617         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5618                 return;
5619         if (mddev->ro) /* never try to sync a read-only array */
5620                 return;
5621
5622         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5623                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5624                         desc = "data-check";
5625                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5626                         desc = "requested-resync";
5627                 else
5628                         desc = "resync";
5629         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5630                 desc = "reshape";
5631         else
5632                 desc = "recovery";
5633
5634         /* we overload curr_resync somewhat here.
5635          * 0 == not engaged in resync at all
5636          * 2 == checking that there is no conflict with another sync
5637          * 1 == like 2, but have yielded to allow conflicting resync to
5638          *              commense
5639          * other == active in resync - this many blocks
5640          *
5641          * Before starting a resync we must have set curr_resync to
5642          * 2, and then checked that every "conflicting" array has curr_resync
5643          * less than ours.  When we find one that is the same or higher
5644          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5645          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5646          * This will mean we have to start checking from the beginning again.
5647          *
5648          */
5649
5650         do {
5651                 mddev->curr_resync = 2;
5652
5653         try_again:
5654                 if (kthread_should_stop()) {
5655                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5656                         goto skip;
5657                 }
5658                 for_each_mddev(mddev2, tmp) {
5659                         if (mddev2 == mddev)
5660                                 continue;
5661                         if (!mddev->parallel_resync
5662                         &&  mddev2->curr_resync
5663                         &&  match_mddev_units(mddev, mddev2)) {
5664                                 DEFINE_WAIT(wq);
5665                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5666                                         /* arbitrarily yield */
5667                                         mddev->curr_resync = 1;
5668                                         wake_up(&resync_wait);
5669                                 }
5670                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5671                                         /* no need to wait here, we can wait the next
5672                                          * time 'round when curr_resync == 2
5673                                          */
5674                                         continue;
5675                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5676                                 if (!kthread_should_stop() &&
5677                                     mddev2->curr_resync >= mddev->curr_resync) {
5678                                         printk(KERN_INFO "md: delaying %s of %s"
5679                                                " until %s has finished (they"
5680                                                " share one or more physical units)\n",
5681                                                desc, mdname(mddev), mdname(mddev2));
5682                                         mddev_put(mddev2);
5683                                         schedule();
5684                                         finish_wait(&resync_wait, &wq);
5685                                         goto try_again;
5686                                 }
5687                                 finish_wait(&resync_wait, &wq);
5688                         }
5689                 }
5690         } while (mddev->curr_resync < 2);
5691
5692         j = 0;
5693         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5694                 /* resync follows the size requested by the personality,
5695                  * which defaults to physical size, but can be virtual size
5696                  */
5697                 max_sectors = mddev->resync_max_sectors;
5698                 mddev->resync_mismatches = 0;
5699                 /* we don't use the checkpoint if there's a bitmap */
5700                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5701                         j = mddev->resync_min;
5702                 else if (!mddev->bitmap)
5703                         j = mddev->recovery_cp;
5704
5705         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5706                 max_sectors = mddev->size << 1;
5707         else {
5708                 /* recovery follows the physical size of devices */
5709                 max_sectors = mddev->size << 1;
5710                 j = MaxSector;
5711                 rdev_for_each(rdev, rtmp, mddev)
5712                         if (rdev->raid_disk >= 0 &&
5713                             !test_bit(Faulty, &rdev->flags) &&
5714                             !test_bit(In_sync, &rdev->flags) &&
5715                             rdev->recovery_offset < j)
5716                                 j = rdev->recovery_offset;
5717         }
5718
5719         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5720         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5721                 " %d KB/sec/disk.\n", speed_min(mddev));
5722         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5723                "(but not more than %d KB/sec) for %s.\n",
5724                speed_max(mddev), desc);
5725
5726         is_mddev_idle(mddev); /* this also initializes IO event counters */
5727
5728         io_sectors = 0;
5729         for (m = 0; m < SYNC_MARKS; m++) {
5730                 mark[m] = jiffies;
5731                 mark_cnt[m] = io_sectors;
5732         }
5733         last_mark = 0;
5734         mddev->resync_mark = mark[last_mark];
5735         mddev->resync_mark_cnt = mark_cnt[last_mark];
5736
5737         /*
5738          * Tune reconstruction:
5739          */
5740         window = 32*(PAGE_SIZE/512);
5741         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5742                 window/2,(unsigned long long) max_sectors/2);
5743
5744         atomic_set(&mddev->recovery_active, 0);
5745         last_check = 0;
5746
5747         if (j>2) {
5748                 printk(KERN_INFO
5749                        "md: resuming %s of %s from checkpoint.\n",
5750                        desc, mdname(mddev));
5751                 mddev->curr_resync = j;
5752         }
5753
5754         while (j < max_sectors) {
5755                 sector_t sectors;
5756
5757                 skipped = 0;
5758                 if (j >= mddev->resync_max) {
5759                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5760                         wait_event(mddev->recovery_wait,
5761                                    mddev->resync_max > j
5762                                    || kthread_should_stop());
5763                 }
5764                 if (kthread_should_stop())
5765                         goto interrupted;
5766                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5767                                                   currspeed < speed_min(mddev));
5768                 if (sectors == 0) {
5769                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5770                         goto out;
5771                 }
5772
5773                 if (!skipped) { /* actual IO requested */
5774                         io_sectors += sectors;
5775                         atomic_add(sectors, &mddev->recovery_active);
5776                 }
5777
5778                 j += sectors;
5779                 if (j>1) mddev->curr_resync = j;
5780                 mddev->curr_mark_cnt = io_sectors;
5781                 if (last_check == 0)
5782                         /* this is the earliers that rebuilt will be
5783                          * visible in /proc/mdstat
5784                          */
5785                         md_new_event(mddev);
5786
5787                 if (last_check + window > io_sectors || j == max_sectors)
5788                         continue;
5789
5790                 last_check = io_sectors;
5791
5792                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5793                         break;
5794
5795         repeat:
5796                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5797                         /* step marks */
5798                         int next = (last_mark+1) % SYNC_MARKS;
5799
5800                         mddev->resync_mark = mark[next];
5801                         mddev->resync_mark_cnt = mark_cnt[next];
5802                         mark[next] = jiffies;
5803                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5804                         last_mark = next;
5805                 }
5806
5807
5808                 if (kthread_should_stop())
5809                         goto interrupted;
5810
5811
5812                 /*
5813                  * this loop exits only if either when we are slower than
5814                  * the 'hard' speed limit, or the system was IO-idle for
5815                  * a jiffy.
5816                  * the system might be non-idle CPU-wise, but we only care
5817                  * about not overloading the IO subsystem. (things like an
5818                  * e2fsck being done on the RAID array should execute fast)
5819                  */
5820                 blk_unplug(mddev->queue);
5821                 cond_resched();
5822
5823                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5824                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5825
5826                 if (currspeed > speed_min(mddev)) {
5827                         if ((currspeed > speed_max(mddev)) ||
5828                                         !is_mddev_idle(mddev)) {
5829                                 msleep(500);
5830                                 goto repeat;
5831                         }
5832                 }
5833         }
5834         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5835         /*
5836          * this also signals 'finished resyncing' to md_stop
5837          */
5838  out:
5839         blk_unplug(mddev->queue);
5840
5841         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5842
5843         /* tell personality that we are finished */
5844         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5845
5846         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5847             mddev->curr_resync > 2) {
5848                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5849                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5850                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5851                                         printk(KERN_INFO
5852                                                "md: checkpointing %s of %s.\n",
5853                                                desc, mdname(mddev));
5854                                         mddev->recovery_cp = mddev->curr_resync;
5855                                 }
5856                         } else
5857                                 mddev->recovery_cp = MaxSector;
5858                 } else {
5859                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5860                                 mddev->curr_resync = MaxSector;
5861                         rdev_for_each(rdev, rtmp, mddev)
5862                                 if (rdev->raid_disk >= 0 &&
5863                                     !test_bit(Faulty, &rdev->flags) &&
5864                                     !test_bit(In_sync, &rdev->flags) &&
5865                                     rdev->recovery_offset < mddev->curr_resync)
5866                                         rdev->recovery_offset = mddev->curr_resync;
5867                 }
5868         }
5869         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5870
5871  skip:
5872         mddev->curr_resync = 0;
5873         mddev->resync_min = 0;
5874         mddev->resync_max = MaxSector;
5875         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5876         wake_up(&resync_wait);
5877         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5878         md_wakeup_thread(mddev->thread);
5879         return;
5880
5881  interrupted:
5882         /*
5883          * got a signal, exit.
5884          */
5885         printk(KERN_INFO
5886                "md: md_do_sync() got signal ... exiting\n");
5887         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5888         goto out;
5889
5890 }
5891 EXPORT_SYMBOL_GPL(md_do_sync);
5892
5893
5894 static int remove_and_add_spares(mddev_t *mddev)
5895 {
5896         mdk_rdev_t *rdev;
5897         struct list_head *rtmp;
5898         int spares = 0;
5899
5900         rdev_for_each(rdev, rtmp, mddev)
5901                 if (rdev->raid_disk >= 0 &&
5902                     !test_bit(Blocked, &rdev->flags) &&
5903                     (test_bit(Faulty, &rdev->flags) ||
5904                      ! test_bit(In_sync, &rdev->flags)) &&
5905                     atomic_read(&rdev->nr_pending)==0) {
5906                         if (mddev->pers->hot_remove_disk(
5907                                     mddev, rdev->raid_disk)==0) {
5908                                 char nm[20];
5909                                 sprintf(nm,"rd%d", rdev->raid_disk);
5910                                 sysfs_remove_link(&mddev->kobj, nm);
5911                                 rdev->raid_disk = -1;
5912                         }
5913                 }
5914
5915         if (mddev->degraded) {
5916                 rdev_for_each(rdev, rtmp, mddev) {
5917                         if (rdev->raid_disk >= 0 &&
5918                             !test_bit(In_sync, &rdev->flags))
5919                                 spares++;
5920                         if (rdev->raid_disk < 0
5921                             && !test_bit(Faulty, &rdev->flags)) {
5922                                 rdev->recovery_offset = 0;
5923                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5924                                         char nm[20];
5925                                         sprintf(nm, "rd%d", rdev->raid_disk);
5926                                         if (sysfs_create_link(&mddev->kobj,
5927                                                               &rdev->kobj, nm))
5928                                                 printk(KERN_WARNING
5929                                                        "md: cannot register "
5930                                                        "%s for %s\n",
5931                                                        nm, mdname(mddev));
5932                                         spares++;
5933                                         md_new_event(mddev);
5934                                 } else
5935                                         break;
5936                         }
5937                 }
5938         }
5939         return spares;
5940 }
5941 /*
5942  * This routine is regularly called by all per-raid-array threads to
5943  * deal with generic issues like resync and super-block update.
5944  * Raid personalities that don't have a thread (linear/raid0) do not
5945  * need this as they never do any recovery or update the superblock.
5946  *
5947  * It does not do any resync itself, but rather "forks" off other threads
5948  * to do that as needed.
5949  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5950  * "->recovery" and create a thread at ->sync_thread.
5951  * When the thread finishes it sets MD_RECOVERY_DONE
5952  * and wakeups up this thread which will reap the thread and finish up.
5953  * This thread also removes any faulty devices (with nr_pending == 0).
5954  *
5955  * The overall approach is:
5956  *  1/ if the superblock needs updating, update it.
5957  *  2/ If a recovery thread is running, don't do anything else.
5958  *  3/ If recovery has finished, clean up, possibly marking spares active.
5959  *  4/ If there are any faulty devices, remove them.
5960  *  5/ If array is degraded, try to add spares devices
5961  *  6/ If array has spares or is not in-sync, start a resync thread.
5962  */
5963 void md_check_recovery(mddev_t *mddev)
5964 {
5965         mdk_rdev_t *rdev;
5966         struct list_head *rtmp;
5967
5968
5969         if (mddev->bitmap)
5970                 bitmap_daemon_work(mddev->bitmap);
5971
5972         if (mddev->ro)
5973                 return;
5974
5975         if (signal_pending(current)) {
5976                 if (mddev->pers->sync_request && !mddev->external) {
5977                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5978                                mdname(mddev));
5979                         mddev->safemode = 2;
5980                 }
5981                 flush_signals(current);
5982         }
5983
5984         if ( ! (
5985                 (mddev->flags && !mddev->external) ||
5986                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5987                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5988                 (mddev->external == 0 && mddev->safemode == 1) ||
5989                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5990                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5991                 ))
5992                 return;
5993
5994         if (mddev_trylock(mddev)) {
5995                 int spares = 0;
5996
5997                 if (!mddev->external) {
5998                         spin_lock_irq(&mddev->write_lock);
5999                         if (mddev->safemode &&
6000                             !atomic_read(&mddev->writes_pending) &&
6001                             !mddev->in_sync &&
6002                             mddev->recovery_cp == MaxSector) {
6003                                 mddev->in_sync = 1;
6004                                 if (mddev->persistent)
6005                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6006                         }
6007                         if (mddev->safemode == 1)
6008                                 mddev->safemode = 0;
6009                         spin_unlock_irq(&mddev->write_lock);
6010                 }
6011
6012                 if (mddev->flags)
6013                         md_update_sb(mddev, 0);
6014
6015
6016                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6017                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6018                         /* resync/recovery still happening */
6019                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6020                         goto unlock;
6021                 }
6022                 if (mddev->sync_thread) {
6023                         /* resync has finished, collect result */
6024                         md_unregister_thread(mddev->sync_thread);
6025                         mddev->sync_thread = NULL;
6026                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6027                                 /* success...*/
6028                                 /* activate any spares */
6029                                 mddev->pers->spare_active(mddev);
6030                         }
6031                         md_update_sb(mddev, 1);
6032
6033                         /* if array is no-longer degraded, then any saved_raid_disk
6034                          * information must be scrapped
6035                          */
6036                         if (!mddev->degraded)
6037                                 rdev_for_each(rdev, rtmp, mddev)
6038                                         rdev->saved_raid_disk = -1;
6039
6040                         mddev->recovery = 0;
6041                         /* flag recovery needed just to double check */
6042                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6043                         md_new_event(mddev);
6044                         goto unlock;
6045                 }
6046                 /* Clear some bits that don't mean anything, but
6047                  * might be left set
6048                  */
6049                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6050                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6051                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6052
6053                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6054                         goto unlock;
6055                 /* no recovery is running.
6056                  * remove any failed drives, then
6057                  * add spares if possible.
6058                  * Spare are also removed and re-added, to allow
6059                  * the personality to fail the re-add.
6060                  */
6061
6062                 if (mddev->reshape_position != MaxSector) {
6063                         if (mddev->pers->check_reshape(mddev) != 0)
6064                                 /* Cannot proceed */
6065                                 goto unlock;
6066                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6067                 } else if ((spares = remove_and_add_spares(mddev))) {
6068                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6069                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6070                 } else if (mddev->recovery_cp < MaxSector) {
6071                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6072                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6073                         /* nothing to be done ... */
6074                         goto unlock;
6075
6076                 if (mddev->pers->sync_request) {
6077                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6078                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6079                                 /* We are adding a device or devices to an array
6080                                  * which has the bitmap stored on all devices.
6081                                  * So make sure all bitmap pages get written
6082                                  */
6083                                 bitmap_write_all(mddev->bitmap);
6084                         }
6085                         mddev->sync_thread = md_register_thread(md_do_sync,
6086                                                                 mddev,
6087                                                                 "%s_resync");
6088                         if (!mddev->sync_thread) {
6089                                 printk(KERN_ERR "%s: could not start resync"
6090                                         " thread...\n",
6091                                         mdname(mddev));
6092                                 /* leave the spares where they are, it shouldn't hurt */
6093                                 mddev->recovery = 0;
6094                         } else
6095                                 md_wakeup_thread(mddev->sync_thread);
6096                         md_new_event(mddev);
6097                 }
6098         unlock:
6099                 mddev_unlock(mddev);
6100         }
6101 }
6102
6103 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6104 {
6105         sysfs_notify(&rdev->kobj, NULL, "state");
6106         wait_event_timeout(rdev->blocked_wait,
6107                            !test_bit(Blocked, &rdev->flags),
6108                            msecs_to_jiffies(5000));
6109         rdev_dec_pending(rdev, mddev);
6110 }
6111 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6112
6113 static int md_notify_reboot(struct notifier_block *this,
6114                             unsigned long code, void *x)
6115 {
6116         struct list_head *tmp;
6117         mddev_t *mddev;
6118
6119         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6120
6121                 printk(KERN_INFO "md: stopping all md devices.\n");
6122
6123                 for_each_mddev(mddev, tmp)
6124                         if (mddev_trylock(mddev)) {
6125                                 do_md_stop (mddev, 1);
6126                                 mddev_unlock(mddev);
6127                         }
6128                 /*
6129                  * certain more exotic SCSI devices are known to be
6130                  * volatile wrt too early system reboots. While the
6131                  * right place to handle this issue is the given
6132                  * driver, we do want to have a safe RAID driver ...
6133                  */
6134                 mdelay(1000*1);
6135         }
6136         return NOTIFY_DONE;
6137 }
6138
6139 static struct notifier_block md_notifier = {
6140         .notifier_call  = md_notify_reboot,
6141         .next           = NULL,
6142         .priority       = INT_MAX, /* before any real devices */
6143 };
6144
6145 static void md_geninit(void)
6146 {
6147         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6148
6149         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6150 }
6151
6152 static int __init md_init(void)
6153 {
6154         if (register_blkdev(MAJOR_NR, "md"))
6155                 return -1;
6156         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6157                 unregister_blkdev(MAJOR_NR, "md");
6158                 return -1;
6159         }
6160         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6161                             md_probe, NULL, NULL);
6162         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6163                             md_probe, NULL, NULL);
6164
6165         register_reboot_notifier(&md_notifier);
6166         raid_table_header = register_sysctl_table(raid_root_table);
6167
6168         md_geninit();
6169         return (0);
6170 }
6171
6172
6173 #ifndef MODULE
6174
6175 /*
6176  * Searches all registered partitions for autorun RAID arrays
6177  * at boot time.
6178  */
6179
6180 static LIST_HEAD(all_detected_devices);
6181 struct detected_devices_node {
6182         struct list_head list;
6183         dev_t dev;
6184 };
6185
6186 void md_autodetect_dev(dev_t dev)
6187 {
6188         struct detected_devices_node *node_detected_dev;
6189
6190         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6191         if (node_detected_dev) {
6192                 node_detected_dev->dev = dev;
6193                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6194         } else {
6195                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6196                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6197         }
6198 }
6199
6200
6201 static void autostart_arrays(int part)
6202 {
6203         mdk_rdev_t *rdev;
6204         struct detected_devices_node *node_detected_dev;
6205         dev_t dev;
6206         int i_scanned, i_passed;
6207
6208         i_scanned = 0;
6209         i_passed = 0;
6210
6211         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6212
6213         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6214                 i_scanned++;
6215                 node_detected_dev = list_entry(all_detected_devices.next,
6216                                         struct detected_devices_node, list);
6217                 list_del(&node_detected_dev->list);
6218                 dev = node_detected_dev->dev;
6219                 kfree(node_detected_dev);
6220                 rdev = md_import_device(dev,0, 90);
6221                 if (IS_ERR(rdev))
6222                         continue;
6223
6224                 if (test_bit(Faulty, &rdev->flags)) {
6225                         MD_BUG();
6226                         continue;
6227                 }
6228                 set_bit(AutoDetected, &rdev->flags);
6229                 list_add(&rdev->same_set, &pending_raid_disks);
6230                 i_passed++;
6231         }
6232
6233         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6234                                                 i_scanned, i_passed);
6235
6236         autorun_devices(part);
6237 }
6238
6239 #endif /* !MODULE */
6240
6241 static __exit void md_exit(void)
6242 {
6243         mddev_t *mddev;
6244         struct list_head *tmp;
6245
6246         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6247         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6248
6249         unregister_blkdev(MAJOR_NR,"md");
6250         unregister_blkdev(mdp_major, "mdp");
6251         unregister_reboot_notifier(&md_notifier);
6252         unregister_sysctl_table(raid_table_header);
6253         remove_proc_entry("mdstat", NULL);
6254         for_each_mddev(mddev, tmp) {
6255                 struct gendisk *disk = mddev->gendisk;
6256                 if (!disk)
6257                         continue;
6258                 export_array(mddev);
6259                 del_gendisk(disk);
6260                 put_disk(disk);
6261                 mddev->gendisk = NULL;
6262                 mddev_put(mddev);
6263         }
6264 }
6265
6266 subsys_initcall(md_init);
6267 module_exit(md_exit)
6268
6269 static int get_ro(char *buffer, struct kernel_param *kp)
6270 {
6271         return sprintf(buffer, "%d", start_readonly);
6272 }
6273 static int set_ro(const char *val, struct kernel_param *kp)
6274 {
6275         char *e;
6276         int num = simple_strtoul(val, &e, 10);
6277         if (*val && (*e == '\0' || *e == '\n')) {
6278                 start_readonly = num;
6279                 return 0;
6280         }
6281         return -EINVAL;
6282 }
6283
6284 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6285 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6286
6287
6288 EXPORT_SYMBOL(register_md_personality);
6289 EXPORT_SYMBOL(unregister_md_personality);
6290 EXPORT_SYMBOL(md_error);
6291 EXPORT_SYMBOL(md_done_sync);
6292 EXPORT_SYMBOL(md_write_start);
6293 EXPORT_SYMBOL(md_write_end);
6294 EXPORT_SYMBOL(md_register_thread);
6295 EXPORT_SYMBOL(md_unregister_thread);
6296 EXPORT_SYMBOL(md_wakeup_thread);
6297 EXPORT_SYMBOL(md_check_recovery);
6298 MODULE_LICENSE("GPL");
6299 MODULE_ALIAS("md");
6300 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);