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) {
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 (slot != -1)
1944                         return -EBUSY;
1945                 if (rdev->raid_disk == -1)
1946                         return -EEXIST;
1947                 /* personality does all needed checks */
1948                 if (rdev->mddev->pers->hot_add_disk == NULL)
1949                         return -EINVAL;
1950                 err = rdev->mddev->pers->
1951                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
1952                 if (err)
1953                         return err;
1954                 sprintf(nm, "rd%d", rdev->raid_disk);
1955                 sysfs_remove_link(&rdev->mddev->kobj, nm);
1956                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1957                 md_wakeup_thread(rdev->mddev->thread);
1958         } else {
1959                 if (slot >= rdev->mddev->raid_disks)
1960                         return -ENOSPC;
1961                 rdev->raid_disk = slot;
1962                 /* assume it is working */
1963                 clear_bit(Faulty, &rdev->flags);
1964                 clear_bit(WriteMostly, &rdev->flags);
1965                 set_bit(In_sync, &rdev->flags);
1966         }
1967         return len;
1968 }
1969
1970
1971 static struct rdev_sysfs_entry rdev_slot =
1972 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1973
1974 static ssize_t
1975 offset_show(mdk_rdev_t *rdev, char *page)
1976 {
1977         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1978 }
1979
1980 static ssize_t
1981 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1982 {
1983         char *e;
1984         unsigned long long offset = simple_strtoull(buf, &e, 10);
1985         if (e==buf || (*e && *e != '\n'))
1986                 return -EINVAL;
1987         if (rdev->mddev->pers)
1988                 return -EBUSY;
1989         if (rdev->size && rdev->mddev->external)
1990                 /* Must set offset before size, so overlap checks
1991                  * can be sane */
1992                 return -EBUSY;
1993         rdev->data_offset = offset;
1994         return len;
1995 }
1996
1997 static struct rdev_sysfs_entry rdev_offset =
1998 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1999
2000 static ssize_t
2001 rdev_size_show(mdk_rdev_t *rdev, char *page)
2002 {
2003         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2004 }
2005
2006 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2007 {
2008         /* check if two start/length pairs overlap */
2009         if (s1+l1 <= s2)
2010                 return 0;
2011         if (s2+l2 <= s1)
2012                 return 0;
2013         return 1;
2014 }
2015
2016 static ssize_t
2017 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2018 {
2019         char *e;
2020         unsigned long long size = simple_strtoull(buf, &e, 10);
2021         unsigned long long oldsize = rdev->size;
2022         mddev_t *my_mddev = rdev->mddev;
2023
2024         if (e==buf || (*e && *e != '\n'))
2025                 return -EINVAL;
2026         if (my_mddev->pers)
2027                 return -EBUSY;
2028         rdev->size = size;
2029         if (size > oldsize && rdev->mddev->external) {
2030                 /* need to check that all other rdevs with the same ->bdev
2031                  * do not overlap.  We need to unlock the mddev to avoid
2032                  * a deadlock.  We have already changed rdev->size, and if
2033                  * we have to change it back, we will have the lock again.
2034                  */
2035                 mddev_t *mddev;
2036                 int overlap = 0;
2037                 struct list_head *tmp, *tmp2;
2038
2039                 mddev_unlock(my_mddev);
2040                 for_each_mddev(mddev, tmp) {
2041                         mdk_rdev_t *rdev2;
2042
2043                         mddev_lock(mddev);
2044                         rdev_for_each(rdev2, tmp2, mddev)
2045                                 if (test_bit(AllReserved, &rdev2->flags) ||
2046                                     (rdev->bdev == rdev2->bdev &&
2047                                      rdev != rdev2 &&
2048                                      overlaps(rdev->data_offset, rdev->size,
2049                                             rdev2->data_offset, rdev2->size))) {
2050                                         overlap = 1;
2051                                         break;
2052                                 }
2053                         mddev_unlock(mddev);
2054                         if (overlap) {
2055                                 mddev_put(mddev);
2056                                 break;
2057                         }
2058                 }
2059                 mddev_lock(my_mddev);
2060                 if (overlap) {
2061                         /* Someone else could have slipped in a size
2062                          * change here, but doing so is just silly.
2063                          * We put oldsize back because we *know* it is
2064                          * safe, and trust userspace not to race with
2065                          * itself
2066                          */
2067                         rdev->size = oldsize;
2068                         return -EBUSY;
2069                 }
2070         }
2071         if (size < my_mddev->size || my_mddev->size == 0)
2072                 my_mddev->size = size;
2073         return len;
2074 }
2075
2076 static struct rdev_sysfs_entry rdev_size =
2077 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2078
2079 static struct attribute *rdev_default_attrs[] = {
2080         &rdev_state.attr,
2081         &rdev_errors.attr,
2082         &rdev_slot.attr,
2083         &rdev_offset.attr,
2084         &rdev_size.attr,
2085         NULL,
2086 };
2087 static ssize_t
2088 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2089 {
2090         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2091         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2092         mddev_t *mddev = rdev->mddev;
2093         ssize_t rv;
2094
2095         if (!entry->show)
2096                 return -EIO;
2097
2098         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2099         if (!rv) {
2100                 if (rdev->mddev == NULL)
2101                         rv = -EBUSY;
2102                 else
2103                         rv = entry->show(rdev, page);
2104                 mddev_unlock(mddev);
2105         }
2106         return rv;
2107 }
2108
2109 static ssize_t
2110 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2111               const char *page, size_t length)
2112 {
2113         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2114         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2115         ssize_t rv;
2116         mddev_t *mddev = rdev->mddev;
2117
2118         if (!entry->store)
2119                 return -EIO;
2120         if (!capable(CAP_SYS_ADMIN))
2121                 return -EACCES;
2122         rv = mddev ? mddev_lock(mddev): -EBUSY;
2123         if (!rv) {
2124                 if (rdev->mddev == NULL)
2125                         rv = -EBUSY;
2126                 else
2127                         rv = entry->store(rdev, page, length);
2128                 mddev_unlock(mddev);
2129         }
2130         return rv;
2131 }
2132
2133 static void rdev_free(struct kobject *ko)
2134 {
2135         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2136         kfree(rdev);
2137 }
2138 static struct sysfs_ops rdev_sysfs_ops = {
2139         .show           = rdev_attr_show,
2140         .store          = rdev_attr_store,
2141 };
2142 static struct kobj_type rdev_ktype = {
2143         .release        = rdev_free,
2144         .sysfs_ops      = &rdev_sysfs_ops,
2145         .default_attrs  = rdev_default_attrs,
2146 };
2147
2148 /*
2149  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2150  *
2151  * mark the device faulty if:
2152  *
2153  *   - the device is nonexistent (zero size)
2154  *   - the device has no valid superblock
2155  *
2156  * a faulty rdev _never_ has rdev->sb set.
2157  */
2158 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2159 {
2160         char b[BDEVNAME_SIZE];
2161         int err;
2162         mdk_rdev_t *rdev;
2163         sector_t size;
2164
2165         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2166         if (!rdev) {
2167                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2168                 return ERR_PTR(-ENOMEM);
2169         }
2170
2171         if ((err = alloc_disk_sb(rdev)))
2172                 goto abort_free;
2173
2174         err = lock_rdev(rdev, newdev, super_format == -2);
2175         if (err)
2176                 goto abort_free;
2177
2178         kobject_init(&rdev->kobj, &rdev_ktype);
2179
2180         rdev->desc_nr = -1;
2181         rdev->saved_raid_disk = -1;
2182         rdev->raid_disk = -1;
2183         rdev->flags = 0;
2184         rdev->data_offset = 0;
2185         rdev->sb_events = 0;
2186         atomic_set(&rdev->nr_pending, 0);
2187         atomic_set(&rdev->read_errors, 0);
2188         atomic_set(&rdev->corrected_errors, 0);
2189
2190         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2191         if (!size) {
2192                 printk(KERN_WARNING
2193                         "md: %s has zero or unknown size, marking faulty!\n",
2194                         bdevname(rdev->bdev,b));
2195                 err = -EINVAL;
2196                 goto abort_free;
2197         }
2198
2199         if (super_format >= 0) {
2200                 err = super_types[super_format].
2201                         load_super(rdev, NULL, super_minor);
2202                 if (err == -EINVAL) {
2203                         printk(KERN_WARNING
2204                                 "md: %s does not have a valid v%d.%d "
2205                                "superblock, not importing!\n",
2206                                 bdevname(rdev->bdev,b),
2207                                super_format, super_minor);
2208                         goto abort_free;
2209                 }
2210                 if (err < 0) {
2211                         printk(KERN_WARNING
2212                                 "md: could not read %s's sb, not importing!\n",
2213                                 bdevname(rdev->bdev,b));
2214                         goto abort_free;
2215                 }
2216         }
2217
2218         INIT_LIST_HEAD(&rdev->same_set);
2219         init_waitqueue_head(&rdev->blocked_wait);
2220
2221         return rdev;
2222
2223 abort_free:
2224         if (rdev->sb_page) {
2225                 if (rdev->bdev)
2226                         unlock_rdev(rdev);
2227                 free_disk_sb(rdev);
2228         }
2229         kfree(rdev);
2230         return ERR_PTR(err);
2231 }
2232
2233 /*
2234  * Check a full RAID array for plausibility
2235  */
2236
2237
2238 static void analyze_sbs(mddev_t * mddev)
2239 {
2240         int i;
2241         struct list_head *tmp;
2242         mdk_rdev_t *rdev, *freshest;
2243         char b[BDEVNAME_SIZE];
2244
2245         freshest = NULL;
2246         rdev_for_each(rdev, tmp, mddev)
2247                 switch (super_types[mddev->major_version].
2248                         load_super(rdev, freshest, mddev->minor_version)) {
2249                 case 1:
2250                         freshest = rdev;
2251                         break;
2252                 case 0:
2253                         break;
2254                 default:
2255                         printk( KERN_ERR \
2256                                 "md: fatal superblock inconsistency in %s"
2257                                 " -- removing from array\n",
2258                                 bdevname(rdev->bdev,b));
2259                         kick_rdev_from_array(rdev);
2260                 }
2261
2262
2263         super_types[mddev->major_version].
2264                 validate_super(mddev, freshest);
2265
2266         i = 0;
2267         rdev_for_each(rdev, tmp, mddev) {
2268                 if (rdev != freshest)
2269                         if (super_types[mddev->major_version].
2270                             validate_super(mddev, rdev)) {
2271                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2272                                         " from array!\n",
2273                                         bdevname(rdev->bdev,b));
2274                                 kick_rdev_from_array(rdev);
2275                                 continue;
2276                         }
2277                 if (mddev->level == LEVEL_MULTIPATH) {
2278                         rdev->desc_nr = i++;
2279                         rdev->raid_disk = rdev->desc_nr;
2280                         set_bit(In_sync, &rdev->flags);
2281                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2282                         rdev->raid_disk = -1;
2283                         clear_bit(In_sync, &rdev->flags);
2284                 }
2285         }
2286
2287
2288
2289         if (mddev->recovery_cp != MaxSector &&
2290             mddev->level >= 1)
2291                 printk(KERN_ERR "md: %s: raid array is not clean"
2292                        " -- starting background reconstruction\n",
2293                        mdname(mddev));
2294
2295 }
2296
2297 static ssize_t
2298 safe_delay_show(mddev_t *mddev, char *page)
2299 {
2300         int msec = (mddev->safemode_delay*1000)/HZ;
2301         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2302 }
2303 static ssize_t
2304 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2305 {
2306         int scale=1;
2307         int dot=0;
2308         int i;
2309         unsigned long msec;
2310         char buf[30];
2311         char *e;
2312         /* remove a period, and count digits after it */
2313         if (len >= sizeof(buf))
2314                 return -EINVAL;
2315         strlcpy(buf, cbuf, len);
2316         buf[len] = 0;
2317         for (i=0; i<len; i++) {
2318                 if (dot) {
2319                         if (isdigit(buf[i])) {
2320                                 buf[i-1] = buf[i];
2321                                 scale *= 10;
2322                         }
2323                         buf[i] = 0;
2324                 } else if (buf[i] == '.') {
2325                         dot=1;
2326                         buf[i] = 0;
2327                 }
2328         }
2329         msec = simple_strtoul(buf, &e, 10);
2330         if (e == buf || (*e && *e != '\n'))
2331                 return -EINVAL;
2332         msec = (msec * 1000) / scale;
2333         if (msec == 0)
2334                 mddev->safemode_delay = 0;
2335         else {
2336                 mddev->safemode_delay = (msec*HZ)/1000;
2337                 if (mddev->safemode_delay == 0)
2338                         mddev->safemode_delay = 1;
2339         }
2340         return len;
2341 }
2342 static struct md_sysfs_entry md_safe_delay =
2343 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2344
2345 static ssize_t
2346 level_show(mddev_t *mddev, char *page)
2347 {
2348         struct mdk_personality *p = mddev->pers;
2349         if (p)
2350                 return sprintf(page, "%s\n", p->name);
2351         else if (mddev->clevel[0])
2352                 return sprintf(page, "%s\n", mddev->clevel);
2353         else if (mddev->level != LEVEL_NONE)
2354                 return sprintf(page, "%d\n", mddev->level);
2355         else
2356                 return 0;
2357 }
2358
2359 static ssize_t
2360 level_store(mddev_t *mddev, const char *buf, size_t len)
2361 {
2362         ssize_t rv = len;
2363         if (mddev->pers)
2364                 return -EBUSY;
2365         if (len == 0)
2366                 return 0;
2367         if (len >= sizeof(mddev->clevel))
2368                 return -ENOSPC;
2369         strncpy(mddev->clevel, buf, len);
2370         if (mddev->clevel[len-1] == '\n')
2371                 len--;
2372         mddev->clevel[len] = 0;
2373         mddev->level = LEVEL_NONE;
2374         return rv;
2375 }
2376
2377 static struct md_sysfs_entry md_level =
2378 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2379
2380
2381 static ssize_t
2382 layout_show(mddev_t *mddev, char *page)
2383 {
2384         /* just a number, not meaningful for all levels */
2385         if (mddev->reshape_position != MaxSector &&
2386             mddev->layout != mddev->new_layout)
2387                 return sprintf(page, "%d (%d)\n",
2388                                mddev->new_layout, mddev->layout);
2389         return sprintf(page, "%d\n", mddev->layout);
2390 }
2391
2392 static ssize_t
2393 layout_store(mddev_t *mddev, const char *buf, size_t len)
2394 {
2395         char *e;
2396         unsigned long n = simple_strtoul(buf, &e, 10);
2397
2398         if (!*buf || (*e && *e != '\n'))
2399                 return -EINVAL;
2400
2401         if (mddev->pers)
2402                 return -EBUSY;
2403         if (mddev->reshape_position != MaxSector)
2404                 mddev->new_layout = n;
2405         else
2406                 mddev->layout = n;
2407         return len;
2408 }
2409 static struct md_sysfs_entry md_layout =
2410 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2411
2412
2413 static ssize_t
2414 raid_disks_show(mddev_t *mddev, char *page)
2415 {
2416         if (mddev->raid_disks == 0)
2417                 return 0;
2418         if (mddev->reshape_position != MaxSector &&
2419             mddev->delta_disks != 0)
2420                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2421                                mddev->raid_disks - mddev->delta_disks);
2422         return sprintf(page, "%d\n", mddev->raid_disks);
2423 }
2424
2425 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2426
2427 static ssize_t
2428 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2429 {
2430         char *e;
2431         int rv = 0;
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                 rv = update_raid_disks(mddev, n);
2439         else if (mddev->reshape_position != MaxSector) {
2440                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2441                 mddev->delta_disks = n - olddisks;
2442                 mddev->raid_disks = n;
2443         } else
2444                 mddev->raid_disks = n;
2445         return rv ? rv : len;
2446 }
2447 static struct md_sysfs_entry md_raid_disks =
2448 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2449
2450 static ssize_t
2451 chunk_size_show(mddev_t *mddev, char *page)
2452 {
2453         if (mddev->reshape_position != MaxSector &&
2454             mddev->chunk_size != mddev->new_chunk)
2455                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2456                                mddev->chunk_size);
2457         return sprintf(page, "%d\n", mddev->chunk_size);
2458 }
2459
2460 static ssize_t
2461 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2462 {
2463         /* can only set chunk_size if array is not yet active */
2464         char *e;
2465         unsigned long n = simple_strtoul(buf, &e, 10);
2466
2467         if (!*buf || (*e && *e != '\n'))
2468                 return -EINVAL;
2469
2470         if (mddev->pers)
2471                 return -EBUSY;
2472         else if (mddev->reshape_position != MaxSector)
2473                 mddev->new_chunk = n;
2474         else
2475                 mddev->chunk_size = n;
2476         return len;
2477 }
2478 static struct md_sysfs_entry md_chunk_size =
2479 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2480
2481 static ssize_t
2482 resync_start_show(mddev_t *mddev, char *page)
2483 {
2484         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2485 }
2486
2487 static ssize_t
2488 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2489 {
2490         char *e;
2491         unsigned long long n = simple_strtoull(buf, &e, 10);
2492
2493         if (mddev->pers)
2494                 return -EBUSY;
2495         if (!*buf || (*e && *e != '\n'))
2496                 return -EINVAL;
2497
2498         mddev->recovery_cp = n;
2499         return len;
2500 }
2501 static struct md_sysfs_entry md_resync_start =
2502 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2503
2504 /*
2505  * The array state can be:
2506  *
2507  * clear
2508  *     No devices, no size, no level
2509  *     Equivalent to STOP_ARRAY ioctl
2510  * inactive
2511  *     May have some settings, but array is not active
2512  *        all IO results in error
2513  *     When written, doesn't tear down array, but just stops it
2514  * suspended (not supported yet)
2515  *     All IO requests will block. The array can be reconfigured.
2516  *     Writing this, if accepted, will block until array is quiessent
2517  * readonly
2518  *     no resync can happen.  no superblocks get written.
2519  *     write requests fail
2520  * read-auto
2521  *     like readonly, but behaves like 'clean' on a write request.
2522  *
2523  * clean - no pending writes, but otherwise active.
2524  *     When written to inactive array, starts without resync
2525  *     If a write request arrives then
2526  *       if metadata is known, mark 'dirty' and switch to 'active'.
2527  *       if not known, block and switch to write-pending
2528  *     If written to an active array that has pending writes, then fails.
2529  * active
2530  *     fully active: IO and resync can be happening.
2531  *     When written to inactive array, starts with resync
2532  *
2533  * write-pending
2534  *     clean, but writes are blocked waiting for 'active' to be written.
2535  *
2536  * active-idle
2537  *     like active, but no writes have been seen for a while (100msec).
2538  *
2539  */
2540 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2541                    write_pending, active_idle, bad_word};
2542 static char *array_states[] = {
2543         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2544         "write-pending", "active-idle", NULL };
2545
2546 static int match_word(const char *word, char **list)
2547 {
2548         int n;
2549         for (n=0; list[n]; n++)
2550                 if (cmd_match(word, list[n]))
2551                         break;
2552         return n;
2553 }
2554
2555 static ssize_t
2556 array_state_show(mddev_t *mddev, char *page)
2557 {
2558         enum array_state st = inactive;
2559
2560         if (mddev->pers)
2561                 switch(mddev->ro) {
2562                 case 1:
2563                         st = readonly;
2564                         break;
2565                 case 2:
2566                         st = read_auto;
2567                         break;
2568                 case 0:
2569                         if (mddev->in_sync)
2570                                 st = clean;
2571                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2572                                 st = write_pending;
2573                         else if (mddev->safemode)
2574                                 st = active_idle;
2575                         else
2576                                 st = active;
2577                 }
2578         else {
2579                 if (list_empty(&mddev->disks) &&
2580                     mddev->raid_disks == 0 &&
2581                     mddev->size == 0)
2582                         st = clear;
2583                 else
2584                         st = inactive;
2585         }
2586         return sprintf(page, "%s\n", array_states[st]);
2587 }
2588
2589 static int do_md_stop(mddev_t * mddev, int ro);
2590 static int do_md_run(mddev_t * mddev);
2591 static int restart_array(mddev_t *mddev);
2592
2593 static ssize_t
2594 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2595 {
2596         int err = -EINVAL;
2597         enum array_state st = match_word(buf, array_states);
2598         switch(st) {
2599         case bad_word:
2600                 break;
2601         case clear:
2602                 /* stopping an active array */
2603                 if (atomic_read(&mddev->active) > 1)
2604                         return -EBUSY;
2605                 err = do_md_stop(mddev, 0);
2606                 break;
2607         case inactive:
2608                 /* stopping an active array */
2609                 if (mddev->pers) {
2610                         if (atomic_read(&mddev->active) > 1)
2611                                 return -EBUSY;
2612                         err = do_md_stop(mddev, 2);
2613                 } else
2614                         err = 0; /* already inactive */
2615                 break;
2616         case suspended:
2617                 break; /* not supported yet */
2618         case readonly:
2619                 if (mddev->pers)
2620                         err = do_md_stop(mddev, 1);
2621                 else {
2622                         mddev->ro = 1;
2623                         set_disk_ro(mddev->gendisk, 1);
2624                         err = do_md_run(mddev);
2625                 }
2626                 break;
2627         case read_auto:
2628                 if (mddev->pers) {
2629                         if (mddev->ro != 1)
2630                                 err = do_md_stop(mddev, 1);
2631                         else
2632                                 err = restart_array(mddev);
2633                         if (err == 0) {
2634                                 mddev->ro = 2;
2635                                 set_disk_ro(mddev->gendisk, 0);
2636                         }
2637                 } else {
2638                         mddev->ro = 2;
2639                         err = do_md_run(mddev);
2640                 }
2641                 break;
2642         case clean:
2643                 if (mddev->pers) {
2644                         restart_array(mddev);
2645                         spin_lock_irq(&mddev->write_lock);
2646                         if (atomic_read(&mddev->writes_pending) == 0) {
2647                                 if (mddev->in_sync == 0) {
2648                                         mddev->in_sync = 1;
2649                                         if (mddev->safemode == 1)
2650                                                 mddev->safemode = 0;
2651                                         if (mddev->persistent)
2652                                                 set_bit(MD_CHANGE_CLEAN,
2653                                                         &mddev->flags);
2654                                 }
2655                                 err = 0;
2656                         } else
2657                                 err = -EBUSY;
2658                         spin_unlock_irq(&mddev->write_lock);
2659                 } else {
2660                         mddev->ro = 0;
2661                         mddev->recovery_cp = MaxSector;
2662                         err = do_md_run(mddev);
2663                 }
2664                 break;
2665         case active:
2666                 if (mddev->pers) {
2667                         restart_array(mddev);
2668                         if (mddev->external)
2669                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2670                         wake_up(&mddev->sb_wait);
2671                         err = 0;
2672                 } else {
2673                         mddev->ro = 0;
2674                         set_disk_ro(mddev->gendisk, 0);
2675                         err = do_md_run(mddev);
2676                 }
2677                 break;
2678         case write_pending:
2679         case active_idle:
2680                 /* these cannot be set */
2681                 break;
2682         }
2683         if (err)
2684                 return err;
2685         else
2686                 return len;
2687 }
2688 static struct md_sysfs_entry md_array_state =
2689 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2690
2691 static ssize_t
2692 null_show(mddev_t *mddev, char *page)
2693 {
2694         return -EINVAL;
2695 }
2696
2697 static ssize_t
2698 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2699 {
2700         /* buf must be %d:%d\n? giving major and minor numbers */
2701         /* The new device is added to the array.
2702          * If the array has a persistent superblock, we read the
2703          * superblock to initialise info and check validity.
2704          * Otherwise, only checking done is that in bind_rdev_to_array,
2705          * which mainly checks size.
2706          */
2707         char *e;
2708         int major = simple_strtoul(buf, &e, 10);
2709         int minor;
2710         dev_t dev;
2711         mdk_rdev_t *rdev;
2712         int err;
2713
2714         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2715                 return -EINVAL;
2716         minor = simple_strtoul(e+1, &e, 10);
2717         if (*e && *e != '\n')
2718                 return -EINVAL;
2719         dev = MKDEV(major, minor);
2720         if (major != MAJOR(dev) ||
2721             minor != MINOR(dev))
2722                 return -EOVERFLOW;
2723
2724
2725         if (mddev->persistent) {
2726                 rdev = md_import_device(dev, mddev->major_version,
2727                                         mddev->minor_version);
2728                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2729                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2730                                                        mdk_rdev_t, same_set);
2731                         err = super_types[mddev->major_version]
2732                                 .load_super(rdev, rdev0, mddev->minor_version);
2733                         if (err < 0)
2734                                 goto out;
2735                 }
2736         } else if (mddev->external)
2737                 rdev = md_import_device(dev, -2, -1);
2738         else
2739                 rdev = md_import_device(dev, -1, -1);
2740
2741         if (IS_ERR(rdev))
2742                 return PTR_ERR(rdev);
2743         err = bind_rdev_to_array(rdev, mddev);
2744  out:
2745         if (err)
2746                 export_rdev(rdev);
2747         return err ? err : len;
2748 }
2749
2750 static struct md_sysfs_entry md_new_device =
2751 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2752
2753 static ssize_t
2754 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2755 {
2756         char *end;
2757         unsigned long chunk, end_chunk;
2758
2759         if (!mddev->bitmap)
2760                 goto out;
2761         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2762         while (*buf) {
2763                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2764                 if (buf == end) break;
2765                 if (*end == '-') { /* range */
2766                         buf = end + 1;
2767                         end_chunk = simple_strtoul(buf, &end, 0);
2768                         if (buf == end) break;
2769                 }
2770                 if (*end && !isspace(*end)) break;
2771                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2772                 buf = end;
2773                 while (isspace(*buf)) buf++;
2774         }
2775         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2776 out:
2777         return len;
2778 }
2779
2780 static struct md_sysfs_entry md_bitmap =
2781 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2782
2783 static ssize_t
2784 size_show(mddev_t *mddev, char *page)
2785 {
2786         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2787 }
2788
2789 static int update_size(mddev_t *mddev, unsigned long size);
2790
2791 static ssize_t
2792 size_store(mddev_t *mddev, const char *buf, size_t len)
2793 {
2794         /* If array is inactive, we can reduce the component size, but
2795          * not increase it (except from 0).
2796          * If array is active, we can try an on-line resize
2797          */
2798         char *e;
2799         int err = 0;
2800         unsigned long long size = simple_strtoull(buf, &e, 10);
2801         if (!*buf || *buf == '\n' ||
2802             (*e && *e != '\n'))
2803                 return -EINVAL;
2804
2805         if (mddev->pers) {
2806                 err = update_size(mddev, size);
2807                 md_update_sb(mddev, 1);
2808         } else {
2809                 if (mddev->size == 0 ||
2810                     mddev->size > size)
2811                         mddev->size = size;
2812                 else
2813                         err = -ENOSPC;
2814         }
2815         return err ? err : len;
2816 }
2817
2818 static struct md_sysfs_entry md_size =
2819 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2820
2821
2822 /* Metdata version.
2823  * This is one of
2824  *   'none' for arrays with no metadata (good luck...)
2825  *   'external' for arrays with externally managed metadata,
2826  * or N.M for internally known formats
2827  */
2828 static ssize_t
2829 metadata_show(mddev_t *mddev, char *page)
2830 {
2831         if (mddev->persistent)
2832                 return sprintf(page, "%d.%d\n",
2833                                mddev->major_version, mddev->minor_version);
2834         else if (mddev->external)
2835                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2836         else
2837                 return sprintf(page, "none\n");
2838 }
2839
2840 static ssize_t
2841 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2842 {
2843         int major, minor;
2844         char *e;
2845         if (!list_empty(&mddev->disks))
2846                 return -EBUSY;
2847
2848         if (cmd_match(buf, "none")) {
2849                 mddev->persistent = 0;
2850                 mddev->external = 0;
2851                 mddev->major_version = 0;
2852                 mddev->minor_version = 90;
2853                 return len;
2854         }
2855         if (strncmp(buf, "external:", 9) == 0) {
2856                 size_t namelen = len-9;
2857                 if (namelen >= sizeof(mddev->metadata_type))
2858                         namelen = sizeof(mddev->metadata_type)-1;
2859                 strncpy(mddev->metadata_type, buf+9, namelen);
2860                 mddev->metadata_type[namelen] = 0;
2861                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2862                         mddev->metadata_type[--namelen] = 0;
2863                 mddev->persistent = 0;
2864                 mddev->external = 1;
2865                 mddev->major_version = 0;
2866                 mddev->minor_version = 90;
2867                 return len;
2868         }
2869         major = simple_strtoul(buf, &e, 10);
2870         if (e==buf || *e != '.')
2871                 return -EINVAL;
2872         buf = e+1;
2873         minor = simple_strtoul(buf, &e, 10);
2874         if (e==buf || (*e && *e != '\n') )
2875                 return -EINVAL;
2876         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2877                 return -ENOENT;
2878         mddev->major_version = major;
2879         mddev->minor_version = minor;
2880         mddev->persistent = 1;
2881         mddev->external = 0;
2882         return len;
2883 }
2884
2885 static struct md_sysfs_entry md_metadata =
2886 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2887
2888 static ssize_t
2889 action_show(mddev_t *mddev, char *page)
2890 {
2891         char *type = "idle";
2892         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2893             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2894                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2895                         type = "reshape";
2896                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2897                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2898                                 type = "resync";
2899                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2900                                 type = "check";
2901                         else
2902                                 type = "repair";
2903                 } else
2904                         type = "recover";
2905         }
2906         return sprintf(page, "%s\n", type);
2907 }
2908
2909 static ssize_t
2910 action_store(mddev_t *mddev, const char *page, size_t len)
2911 {
2912         if (!mddev->pers || !mddev->pers->sync_request)
2913                 return -EINVAL;
2914
2915         if (cmd_match(page, "idle")) {
2916                 if (mddev->sync_thread) {
2917                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2918                         md_unregister_thread(mddev->sync_thread);
2919                         mddev->sync_thread = NULL;
2920                         mddev->recovery = 0;
2921                 }
2922         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2923                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2924                 return -EBUSY;
2925         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2926                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2927         else if (cmd_match(page, "reshape")) {
2928                 int err;
2929                 if (mddev->pers->start_reshape == NULL)
2930                         return -EINVAL;
2931                 err = mddev->pers->start_reshape(mddev);
2932                 if (err)
2933                         return err;
2934         } else {
2935                 if (cmd_match(page, "check"))
2936                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2937                 else if (!cmd_match(page, "repair"))
2938                         return -EINVAL;
2939                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2940                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2941         }
2942         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2943         md_wakeup_thread(mddev->thread);
2944         return len;
2945 }
2946
2947 static ssize_t
2948 mismatch_cnt_show(mddev_t *mddev, char *page)
2949 {
2950         return sprintf(page, "%llu\n",
2951                        (unsigned long long) mddev->resync_mismatches);
2952 }
2953
2954 static struct md_sysfs_entry md_scan_mode =
2955 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2956
2957
2958 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2959
2960 static ssize_t
2961 sync_min_show(mddev_t *mddev, char *page)
2962 {
2963         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2964                        mddev->sync_speed_min ? "local": "system");
2965 }
2966
2967 static ssize_t
2968 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2969 {
2970         int min;
2971         char *e;
2972         if (strncmp(buf, "system", 6)==0) {
2973                 mddev->sync_speed_min = 0;
2974                 return len;
2975         }
2976         min = simple_strtoul(buf, &e, 10);
2977         if (buf == e || (*e && *e != '\n') || min <= 0)
2978                 return -EINVAL;
2979         mddev->sync_speed_min = min;
2980         return len;
2981 }
2982
2983 static struct md_sysfs_entry md_sync_min =
2984 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2985
2986 static ssize_t
2987 sync_max_show(mddev_t *mddev, char *page)
2988 {
2989         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2990                        mddev->sync_speed_max ? "local": "system");
2991 }
2992
2993 static ssize_t
2994 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2995 {
2996         int max;
2997         char *e;
2998         if (strncmp(buf, "system", 6)==0) {
2999                 mddev->sync_speed_max = 0;
3000                 return len;
3001         }
3002         max = simple_strtoul(buf, &e, 10);
3003         if (buf == e || (*e && *e != '\n') || max <= 0)
3004                 return -EINVAL;
3005         mddev->sync_speed_max = max;
3006         return len;
3007 }
3008
3009 static struct md_sysfs_entry md_sync_max =
3010 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3011
3012 static ssize_t
3013 degraded_show(mddev_t *mddev, char *page)
3014 {
3015         return sprintf(page, "%d\n", mddev->degraded);
3016 }
3017 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3018
3019 static ssize_t
3020 sync_force_parallel_show(mddev_t *mddev, char *page)
3021 {
3022         return sprintf(page, "%d\n", mddev->parallel_resync);
3023 }
3024
3025 static ssize_t
3026 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3027 {
3028         long n;
3029
3030         if (strict_strtol(buf, 10, &n))
3031                 return -EINVAL;
3032
3033         if (n != 0 && n != 1)
3034                 return -EINVAL;
3035
3036         mddev->parallel_resync = n;
3037
3038         if (mddev->sync_thread)
3039                 wake_up(&resync_wait);
3040
3041         return len;
3042 }
3043
3044 /* force parallel resync, even with shared block devices */
3045 static struct md_sysfs_entry md_sync_force_parallel =
3046 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3047        sync_force_parallel_show, sync_force_parallel_store);
3048
3049 static ssize_t
3050 sync_speed_show(mddev_t *mddev, char *page)
3051 {
3052         unsigned long resync, dt, db;
3053         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
3054         dt = ((jiffies - mddev->resync_mark) / HZ);
3055         if (!dt) dt++;
3056         db = resync - (mddev->resync_mark_cnt);
3057         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
3058 }
3059
3060 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3061
3062 static ssize_t
3063 sync_completed_show(mddev_t *mddev, char *page)
3064 {
3065         unsigned long max_blocks, resync;
3066
3067         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3068                 max_blocks = mddev->resync_max_sectors;
3069         else
3070                 max_blocks = mddev->size << 1;
3071
3072         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3073         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3074 }
3075
3076 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3077
3078 static ssize_t
3079 min_sync_show(mddev_t *mddev, char *page)
3080 {
3081         return sprintf(page, "%llu\n",
3082                        (unsigned long long)mddev->resync_min);
3083 }
3084 static ssize_t
3085 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3086 {
3087         unsigned long long min;
3088         if (strict_strtoull(buf, 10, &min))
3089                 return -EINVAL;
3090         if (min > mddev->resync_max)
3091                 return -EINVAL;
3092         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3093                 return -EBUSY;
3094
3095         /* Must be a multiple of chunk_size */
3096         if (mddev->chunk_size) {
3097                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3098                         return -EINVAL;
3099         }
3100         mddev->resync_min = min;
3101
3102         return len;
3103 }
3104
3105 static struct md_sysfs_entry md_min_sync =
3106 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3107
3108 static ssize_t
3109 max_sync_show(mddev_t *mddev, char *page)
3110 {
3111         if (mddev->resync_max == MaxSector)
3112                 return sprintf(page, "max\n");
3113         else
3114                 return sprintf(page, "%llu\n",
3115                                (unsigned long long)mddev->resync_max);
3116 }
3117 static ssize_t
3118 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3119 {
3120         if (strncmp(buf, "max", 3) == 0)
3121                 mddev->resync_max = MaxSector;
3122         else {
3123                 unsigned long long max;
3124                 if (strict_strtoull(buf, 10, &max))
3125                         return -EINVAL;
3126                 if (max < mddev->resync_min)
3127                         return -EINVAL;
3128                 if (max < mddev->resync_max &&
3129                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3130                         return -EBUSY;
3131
3132                 /* Must be a multiple of chunk_size */
3133                 if (mddev->chunk_size) {
3134                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3135                                 return -EINVAL;
3136                 }
3137                 mddev->resync_max = max;
3138         }
3139         wake_up(&mddev->recovery_wait);
3140         return len;
3141 }
3142
3143 static struct md_sysfs_entry md_max_sync =
3144 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3145
3146 static ssize_t
3147 suspend_lo_show(mddev_t *mddev, char *page)
3148 {
3149         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3150 }
3151
3152 static ssize_t
3153 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3154 {
3155         char *e;
3156         unsigned long long new = simple_strtoull(buf, &e, 10);
3157
3158         if (mddev->pers->quiesce == NULL)
3159                 return -EINVAL;
3160         if (buf == e || (*e && *e != '\n'))
3161                 return -EINVAL;
3162         if (new >= mddev->suspend_hi ||
3163             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3164                 mddev->suspend_lo = new;
3165                 mddev->pers->quiesce(mddev, 2);
3166                 return len;
3167         } else
3168                 return -EINVAL;
3169 }
3170 static struct md_sysfs_entry md_suspend_lo =
3171 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3172
3173
3174 static ssize_t
3175 suspend_hi_show(mddev_t *mddev, char *page)
3176 {
3177         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3178 }
3179
3180 static ssize_t
3181 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3182 {
3183         char *e;
3184         unsigned long long new = simple_strtoull(buf, &e, 10);
3185
3186         if (mddev->pers->quiesce == NULL)
3187                 return -EINVAL;
3188         if (buf == e || (*e && *e != '\n'))
3189                 return -EINVAL;
3190         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3191             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3192                 mddev->suspend_hi = new;
3193                 mddev->pers->quiesce(mddev, 1);
3194                 mddev->pers->quiesce(mddev, 0);
3195                 return len;
3196         } else
3197                 return -EINVAL;
3198 }
3199 static struct md_sysfs_entry md_suspend_hi =
3200 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3201
3202 static ssize_t
3203 reshape_position_show(mddev_t *mddev, char *page)
3204 {
3205         if (mddev->reshape_position != MaxSector)
3206                 return sprintf(page, "%llu\n",
3207                                (unsigned long long)mddev->reshape_position);
3208         strcpy(page, "none\n");
3209         return 5;
3210 }
3211
3212 static ssize_t
3213 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3214 {
3215         char *e;
3216         unsigned long long new = simple_strtoull(buf, &e, 10);
3217         if (mddev->pers)
3218                 return -EBUSY;
3219         if (buf == e || (*e && *e != '\n'))
3220                 return -EINVAL;
3221         mddev->reshape_position = new;
3222         mddev->delta_disks = 0;
3223         mddev->new_level = mddev->level;
3224         mddev->new_layout = mddev->layout;
3225         mddev->new_chunk = mddev->chunk_size;
3226         return len;
3227 }
3228
3229 static struct md_sysfs_entry md_reshape_position =
3230 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3231        reshape_position_store);
3232
3233
3234 static struct attribute *md_default_attrs[] = {
3235         &md_level.attr,
3236         &md_layout.attr,
3237         &md_raid_disks.attr,
3238         &md_chunk_size.attr,
3239         &md_size.attr,
3240         &md_resync_start.attr,
3241         &md_metadata.attr,
3242         &md_new_device.attr,
3243         &md_safe_delay.attr,
3244         &md_array_state.attr,
3245         &md_reshape_position.attr,
3246         NULL,
3247 };
3248
3249 static struct attribute *md_redundancy_attrs[] = {
3250         &md_scan_mode.attr,
3251         &md_mismatches.attr,
3252         &md_sync_min.attr,
3253         &md_sync_max.attr,
3254         &md_sync_speed.attr,
3255         &md_sync_force_parallel.attr,
3256         &md_sync_completed.attr,
3257         &md_min_sync.attr,
3258         &md_max_sync.attr,
3259         &md_suspend_lo.attr,
3260         &md_suspend_hi.attr,
3261         &md_bitmap.attr,
3262         &md_degraded.attr,
3263         NULL,
3264 };
3265 static struct attribute_group md_redundancy_group = {
3266         .name = NULL,
3267         .attrs = md_redundancy_attrs,
3268 };
3269
3270
3271 static ssize_t
3272 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3273 {
3274         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3275         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3276         ssize_t rv;
3277
3278         if (!entry->show)
3279                 return -EIO;
3280         rv = mddev_lock(mddev);
3281         if (!rv) {
3282                 rv = entry->show(mddev, page);
3283                 mddev_unlock(mddev);
3284         }
3285         return rv;
3286 }
3287
3288 static ssize_t
3289 md_attr_store(struct kobject *kobj, struct attribute *attr,
3290               const char *page, size_t length)
3291 {
3292         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3293         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3294         ssize_t rv;
3295
3296         if (!entry->store)
3297                 return -EIO;
3298         if (!capable(CAP_SYS_ADMIN))
3299                 return -EACCES;
3300         rv = mddev_lock(mddev);
3301         if (!rv) {
3302                 rv = entry->store(mddev, page, length);
3303                 mddev_unlock(mddev);
3304         }
3305         return rv;
3306 }
3307
3308 static void md_free(struct kobject *ko)
3309 {
3310         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3311         kfree(mddev);
3312 }
3313
3314 static struct sysfs_ops md_sysfs_ops = {
3315         .show   = md_attr_show,
3316         .store  = md_attr_store,
3317 };
3318 static struct kobj_type md_ktype = {
3319         .release        = md_free,
3320         .sysfs_ops      = &md_sysfs_ops,
3321         .default_attrs  = md_default_attrs,
3322 };
3323
3324 int mdp_major = 0;
3325
3326 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3327 {
3328         static DEFINE_MUTEX(disks_mutex);
3329         mddev_t *mddev = mddev_find(dev);
3330         struct gendisk *disk;
3331         int partitioned = (MAJOR(dev) != MD_MAJOR);
3332         int shift = partitioned ? MdpMinorShift : 0;
3333         int unit = MINOR(dev) >> shift;
3334         int error;
3335
3336         if (!mddev)
3337                 return NULL;
3338
3339         mutex_lock(&disks_mutex);
3340         if (mddev->gendisk) {
3341                 mutex_unlock(&disks_mutex);
3342                 mddev_put(mddev);
3343                 return NULL;
3344         }
3345         disk = alloc_disk(1 << shift);
3346         if (!disk) {
3347                 mutex_unlock(&disks_mutex);
3348                 mddev_put(mddev);
3349                 return NULL;
3350         }
3351         disk->major = MAJOR(dev);
3352         disk->first_minor = unit << shift;
3353         if (partitioned)
3354                 sprintf(disk->disk_name, "md_d%d", unit);
3355         else
3356                 sprintf(disk->disk_name, "md%d", unit);
3357         disk->fops = &md_fops;
3358         disk->private_data = mddev;
3359         disk->queue = mddev->queue;
3360         add_disk(disk);
3361         mddev->gendisk = disk;
3362         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3363                                      "%s", "md");
3364         mutex_unlock(&disks_mutex);
3365         if (error)
3366                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3367                        disk->disk_name);
3368         else
3369                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3370         return NULL;
3371 }
3372
3373 static void md_safemode_timeout(unsigned long data)
3374 {
3375         mddev_t *mddev = (mddev_t *) data;
3376
3377         mddev->safemode = 1;
3378         md_wakeup_thread(mddev->thread);
3379 }
3380
3381 static int start_dirty_degraded;
3382
3383 static int do_md_run(mddev_t * mddev)
3384 {
3385         int err;
3386         int chunk_size;
3387         struct list_head *tmp;
3388         mdk_rdev_t *rdev;
3389         struct gendisk *disk;
3390         struct mdk_personality *pers;
3391         char b[BDEVNAME_SIZE];
3392
3393         if (list_empty(&mddev->disks))
3394                 /* cannot run an array with no devices.. */
3395                 return -EINVAL;
3396
3397         if (mddev->pers)
3398                 return -EBUSY;
3399
3400         /*
3401          * Analyze all RAID superblock(s)
3402          */
3403         if (!mddev->raid_disks) {
3404                 if (!mddev->persistent)
3405                         return -EINVAL;
3406                 analyze_sbs(mddev);
3407         }
3408
3409         chunk_size = mddev->chunk_size;
3410
3411         if (chunk_size) {
3412                 if (chunk_size > MAX_CHUNK_SIZE) {
3413                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3414                                 chunk_size, MAX_CHUNK_SIZE);
3415                         return -EINVAL;
3416                 }
3417                 /*
3418                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3419                  */
3420                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3421                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3422                         return -EINVAL;
3423                 }
3424                 if (chunk_size < PAGE_SIZE) {
3425                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3426                                 chunk_size, PAGE_SIZE);
3427                         return -EINVAL;
3428                 }
3429
3430                 /* devices must have minimum size of one chunk */
3431                 rdev_for_each(rdev, tmp, mddev) {
3432                         if (test_bit(Faulty, &rdev->flags))
3433                                 continue;
3434                         if (rdev->size < chunk_size / 1024) {
3435                                 printk(KERN_WARNING
3436                                         "md: Dev %s smaller than chunk_size:"
3437                                         " %lluk < %dk\n",
3438                                         bdevname(rdev->bdev,b),
3439                                         (unsigned long long)rdev->size,
3440                                         chunk_size / 1024);
3441                                 return -EINVAL;
3442                         }
3443                 }
3444         }
3445
3446 #ifdef CONFIG_KMOD
3447         if (mddev->level != LEVEL_NONE)
3448                 request_module("md-level-%d", mddev->level);
3449         else if (mddev->clevel[0])
3450                 request_module("md-%s", mddev->clevel);
3451 #endif
3452
3453         /*
3454          * Drop all container device buffers, from now on
3455          * the only valid external interface is through the md
3456          * device.
3457          */
3458         rdev_for_each(rdev, tmp, mddev) {
3459                 if (test_bit(Faulty, &rdev->flags))
3460                         continue;
3461                 sync_blockdev(rdev->bdev);
3462                 invalidate_bdev(rdev->bdev);
3463
3464                 /* perform some consistency tests on the device.
3465                  * We don't want the data to overlap the metadata,
3466                  * Internal Bitmap issues has handled elsewhere.
3467                  */
3468                 if (rdev->data_offset < rdev->sb_offset) {
3469                         if (mddev->size &&
3470                             rdev->data_offset + mddev->size*2
3471                             > rdev->sb_offset*2) {
3472                                 printk("md: %s: data overlaps metadata\n",
3473                                        mdname(mddev));
3474                                 return -EINVAL;
3475                         }
3476                 } else {
3477                         if (rdev->sb_offset*2 + rdev->sb_size/512
3478                             > rdev->data_offset) {
3479                                 printk("md: %s: metadata overlaps data\n",
3480                                        mdname(mddev));
3481                                 return -EINVAL;
3482                         }
3483                 }
3484         }
3485
3486         md_probe(mddev->unit, NULL, NULL);
3487         disk = mddev->gendisk;
3488         if (!disk)
3489                 return -ENOMEM;
3490
3491         spin_lock(&pers_lock);
3492         pers = find_pers(mddev->level, mddev->clevel);
3493         if (!pers || !try_module_get(pers->owner)) {
3494                 spin_unlock(&pers_lock);
3495                 if (mddev->level != LEVEL_NONE)
3496                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3497                                mddev->level);
3498                 else
3499                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3500                                mddev->clevel);
3501                 return -EINVAL;
3502         }
3503         mddev->pers = pers;
3504         spin_unlock(&pers_lock);
3505         mddev->level = pers->level;
3506         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3507
3508         if (mddev->reshape_position != MaxSector &&
3509             pers->start_reshape == NULL) {
3510                 /* This personality cannot handle reshaping... */
3511                 mddev->pers = NULL;
3512                 module_put(pers->owner);
3513                 return -EINVAL;
3514         }
3515
3516         if (pers->sync_request) {
3517                 /* Warn if this is a potentially silly
3518                  * configuration.
3519                  */
3520                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3521                 mdk_rdev_t *rdev2;
3522                 struct list_head *tmp2;
3523                 int warned = 0;
3524                 rdev_for_each(rdev, tmp, mddev) {
3525                         rdev_for_each(rdev2, tmp2, mddev) {
3526                                 if (rdev < rdev2 &&
3527                                     rdev->bdev->bd_contains ==
3528                                     rdev2->bdev->bd_contains) {
3529                                         printk(KERN_WARNING
3530                                                "%s: WARNING: %s appears to be"
3531                                                " on the same physical disk as"
3532                                                " %s.\n",
3533                                                mdname(mddev),
3534                                                bdevname(rdev->bdev,b),
3535                                                bdevname(rdev2->bdev,b2));
3536                                         warned = 1;
3537                                 }
3538                         }
3539                 }
3540                 if (warned)
3541                         printk(KERN_WARNING
3542                                "True protection against single-disk"
3543                                " failure might be compromised.\n");
3544         }
3545
3546         mddev->recovery = 0;
3547         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3548         mddev->barriers_work = 1;
3549         mddev->ok_start_degraded = start_dirty_degraded;
3550
3551         if (start_readonly)
3552                 mddev->ro = 2; /* read-only, but switch on first write */
3553
3554         err = mddev->pers->run(mddev);
3555         if (!err && mddev->pers->sync_request) {
3556                 err = bitmap_create(mddev);
3557                 if (err) {
3558                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3559                                mdname(mddev), err);
3560                         mddev->pers->stop(mddev);
3561                 }
3562         }
3563         if (err) {
3564                 printk(KERN_ERR "md: pers->run() failed ...\n");
3565                 module_put(mddev->pers->owner);
3566                 mddev->pers = NULL;
3567                 bitmap_destroy(mddev);
3568                 return err;
3569         }
3570         if (mddev->pers->sync_request) {
3571                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3572                         printk(KERN_WARNING
3573                                "md: cannot register extra attributes for %s\n",
3574                                mdname(mddev));
3575         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3576                 mddev->ro = 0;
3577
3578         atomic_set(&mddev->writes_pending,0);
3579         mddev->safemode = 0;
3580         mddev->safemode_timer.function = md_safemode_timeout;
3581         mddev->safemode_timer.data = (unsigned long) mddev;
3582         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3583         mddev->in_sync = 1;
3584
3585         rdev_for_each(rdev, tmp, mddev)
3586                 if (rdev->raid_disk >= 0) {
3587                         char nm[20];
3588                         sprintf(nm, "rd%d", rdev->raid_disk);
3589                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3590                                 printk("md: cannot register %s for %s\n",
3591                                        nm, mdname(mddev));
3592                 }
3593
3594         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3595
3596         if (mddev->flags)
3597                 md_update_sb(mddev, 0);
3598
3599         set_capacity(disk, mddev->array_size<<1);
3600
3601         /* If we call blk_queue_make_request here, it will
3602          * re-initialise max_sectors etc which may have been
3603          * refined inside -> run.  So just set the bits we need to set.
3604          * Most initialisation happended when we called
3605          * blk_queue_make_request(..., md_fail_request)
3606          * earlier.
3607          */
3608         mddev->queue->queuedata = mddev;
3609         mddev->queue->make_request_fn = mddev->pers->make_request;
3610
3611         /* If there is a partially-recovered drive we need to
3612          * start recovery here.  If we leave it to md_check_recovery,
3613          * it will remove the drives and not do the right thing
3614          */
3615         if (mddev->degraded && !mddev->sync_thread) {
3616                 struct list_head *rtmp;
3617                 int spares = 0;
3618                 rdev_for_each(rdev, rtmp, mddev)
3619                         if (rdev->raid_disk >= 0 &&
3620                             !test_bit(In_sync, &rdev->flags) &&
3621                             !test_bit(Faulty, &rdev->flags))
3622                                 /* complete an interrupted recovery */
3623                                 spares++;
3624                 if (spares && mddev->pers->sync_request) {
3625                         mddev->recovery = 0;
3626                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3627                         mddev->sync_thread = md_register_thread(md_do_sync,
3628                                                                 mddev,
3629                                                                 "%s_resync");
3630                         if (!mddev->sync_thread) {
3631                                 printk(KERN_ERR "%s: could not start resync"
3632                                        " thread...\n",
3633                                        mdname(mddev));
3634                                 /* leave the spares where they are, it shouldn't hurt */
3635                                 mddev->recovery = 0;
3636                         }
3637                 }
3638         }
3639         md_wakeup_thread(mddev->thread);
3640         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3641
3642         mddev->changed = 1;
3643         md_new_event(mddev);
3644         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3645         return 0;
3646 }
3647
3648 static int restart_array(mddev_t *mddev)
3649 {
3650         struct gendisk *disk = mddev->gendisk;
3651         int err;
3652
3653         /*
3654          * Complain if it has no devices
3655          */
3656         err = -ENXIO;
3657         if (list_empty(&mddev->disks))
3658                 goto out;
3659
3660         if (mddev->pers) {
3661                 err = -EBUSY;
3662                 if (!mddev->ro)
3663                         goto out;
3664
3665                 mddev->safemode = 0;
3666                 mddev->ro = 0;
3667                 set_disk_ro(disk, 0);
3668
3669                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3670                         mdname(mddev));
3671                 /*
3672                  * Kick recovery or resync if necessary
3673                  */
3674                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3675                 md_wakeup_thread(mddev->thread);
3676                 md_wakeup_thread(mddev->sync_thread);
3677                 err = 0;
3678         } else
3679                 err = -EINVAL;
3680
3681 out:
3682         return err;
3683 }
3684
3685 /* similar to deny_write_access, but accounts for our holding a reference
3686  * to the file ourselves */
3687 static int deny_bitmap_write_access(struct file * file)
3688 {
3689         struct inode *inode = file->f_mapping->host;
3690
3691         spin_lock(&inode->i_lock);
3692         if (atomic_read(&inode->i_writecount) > 1) {
3693                 spin_unlock(&inode->i_lock);
3694                 return -ETXTBSY;
3695         }
3696         atomic_set(&inode->i_writecount, -1);
3697         spin_unlock(&inode->i_lock);
3698
3699         return 0;
3700 }
3701
3702 static void restore_bitmap_write_access(struct file *file)
3703 {
3704         struct inode *inode = file->f_mapping->host;
3705
3706         spin_lock(&inode->i_lock);
3707         atomic_set(&inode->i_writecount, 1);
3708         spin_unlock(&inode->i_lock);
3709 }
3710
3711 /* mode:
3712  *   0 - completely stop and dis-assemble array
3713  *   1 - switch to readonly
3714  *   2 - stop but do not disassemble array
3715  */
3716 static int do_md_stop(mddev_t * mddev, int mode)
3717 {
3718         int err = 0;
3719         struct gendisk *disk = mddev->gendisk;
3720
3721         if (mddev->pers) {
3722                 if (atomic_read(&mddev->active)>2) {
3723                         printk("md: %s still in use.\n",mdname(mddev));
3724                         return -EBUSY;
3725                 }
3726
3727                 if (mddev->sync_thread) {
3728                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3729                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3730                         md_unregister_thread(mddev->sync_thread);
3731                         mddev->sync_thread = NULL;
3732                 }
3733
3734                 del_timer_sync(&mddev->safemode_timer);
3735
3736                 invalidate_partition(disk, 0);
3737
3738                 switch(mode) {
3739                 case 1: /* readonly */
3740                         err  = -ENXIO;
3741                         if (mddev->ro==1)
3742                                 goto out;
3743                         mddev->ro = 1;
3744                         break;
3745                 case 0: /* disassemble */
3746                 case 2: /* stop */
3747                         bitmap_flush(mddev);
3748                         md_super_wait(mddev);
3749                         if (mddev->ro)
3750                                 set_disk_ro(disk, 0);
3751                         blk_queue_make_request(mddev->queue, md_fail_request);
3752                         mddev->pers->stop(mddev);
3753                         mddev->queue->merge_bvec_fn = NULL;
3754                         mddev->queue->unplug_fn = NULL;
3755                         mddev->queue->backing_dev_info.congested_fn = NULL;
3756                         if (mddev->pers->sync_request)
3757                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3758
3759                         module_put(mddev->pers->owner);
3760                         mddev->pers = NULL;
3761                         /* tell userspace to handle 'inactive' */
3762                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3763
3764                         set_capacity(disk, 0);
3765                         mddev->changed = 1;
3766
3767                         if (mddev->ro)
3768                                 mddev->ro = 0;
3769                 }
3770                 if (!mddev->in_sync || mddev->flags) {
3771                         /* mark array as shutdown cleanly */
3772                         mddev->in_sync = 1;
3773                         md_update_sb(mddev, 1);
3774                 }
3775                 if (mode == 1)
3776                         set_disk_ro(disk, 1);
3777                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3778         }
3779
3780         /*
3781          * Free resources if final stop
3782          */
3783         if (mode == 0) {
3784                 mdk_rdev_t *rdev;
3785                 struct list_head *tmp;
3786
3787                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3788
3789                 bitmap_destroy(mddev);
3790                 if (mddev->bitmap_file) {
3791                         restore_bitmap_write_access(mddev->bitmap_file);
3792                         fput(mddev->bitmap_file);
3793                         mddev->bitmap_file = NULL;
3794                 }
3795                 mddev->bitmap_offset = 0;
3796
3797                 rdev_for_each(rdev, tmp, mddev)
3798                         if (rdev->raid_disk >= 0) {
3799                                 char nm[20];
3800                                 sprintf(nm, "rd%d", rdev->raid_disk);
3801                                 sysfs_remove_link(&mddev->kobj, nm);
3802                         }
3803
3804                 /* make sure all md_delayed_delete calls have finished */
3805                 flush_scheduled_work();
3806
3807                 export_array(mddev);
3808
3809                 mddev->array_size = 0;
3810                 mddev->size = 0;
3811                 mddev->raid_disks = 0;
3812                 mddev->recovery_cp = 0;
3813                 mddev->resync_min = 0;
3814                 mddev->resync_max = MaxSector;
3815                 mddev->reshape_position = MaxSector;
3816                 mddev->external = 0;
3817                 mddev->persistent = 0;
3818                 mddev->level = LEVEL_NONE;
3819                 mddev->clevel[0] = 0;
3820                 mddev->flags = 0;
3821                 mddev->ro = 0;
3822                 mddev->metadata_type[0] = 0;
3823                 mddev->chunk_size = 0;
3824                 mddev->ctime = mddev->utime = 0;
3825                 mddev->layout = 0;
3826                 mddev->max_disks = 0;
3827                 mddev->events = 0;
3828                 mddev->delta_disks = 0;
3829                 mddev->new_level = LEVEL_NONE;
3830                 mddev->new_layout = 0;
3831                 mddev->new_chunk = 0;
3832                 mddev->curr_resync = 0;
3833                 mddev->resync_mismatches = 0;
3834                 mddev->suspend_lo = mddev->suspend_hi = 0;
3835                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3836                 mddev->recovery = 0;
3837                 mddev->in_sync = 0;
3838                 mddev->changed = 0;
3839                 mddev->degraded = 0;
3840                 mddev->barriers_work = 0;
3841                 mddev->safemode = 0;
3842
3843         } else if (mddev->pers)
3844                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3845                         mdname(mddev));
3846         err = 0;
3847         md_new_event(mddev);
3848 out:
3849         return err;
3850 }
3851
3852 #ifndef MODULE
3853 static void autorun_array(mddev_t *mddev)
3854 {
3855         mdk_rdev_t *rdev;
3856         struct list_head *tmp;
3857         int err;
3858
3859         if (list_empty(&mddev->disks))
3860                 return;
3861
3862         printk(KERN_INFO "md: running: ");
3863
3864         rdev_for_each(rdev, tmp, mddev) {
3865                 char b[BDEVNAME_SIZE];
3866                 printk("<%s>", bdevname(rdev->bdev,b));
3867         }
3868         printk("\n");
3869
3870         err = do_md_run (mddev);
3871         if (err) {
3872                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3873                 do_md_stop (mddev, 0);
3874         }
3875 }
3876
3877 /*
3878  * lets try to run arrays based on all disks that have arrived
3879  * until now. (those are in pending_raid_disks)
3880  *
3881  * the method: pick the first pending disk, collect all disks with
3882  * the same UUID, remove all from the pending list and put them into
3883  * the 'same_array' list. Then order this list based on superblock
3884  * update time (freshest comes first), kick out 'old' disks and
3885  * compare superblocks. If everything's fine then run it.
3886  *
3887  * If "unit" is allocated, then bump its reference count
3888  */
3889 static void autorun_devices(int part)
3890 {
3891         struct list_head *tmp;
3892         mdk_rdev_t *rdev0, *rdev;
3893         mddev_t *mddev;
3894         char b[BDEVNAME_SIZE];
3895
3896         printk(KERN_INFO "md: autorun ...\n");
3897         while (!list_empty(&pending_raid_disks)) {
3898                 int unit;
3899                 dev_t dev;
3900                 LIST_HEAD(candidates);
3901                 rdev0 = list_entry(pending_raid_disks.next,
3902                                          mdk_rdev_t, same_set);
3903
3904                 printk(KERN_INFO "md: considering %s ...\n",
3905                         bdevname(rdev0->bdev,b));
3906                 INIT_LIST_HEAD(&candidates);
3907                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3908                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3909                                 printk(KERN_INFO "md:  adding %s ...\n",
3910                                         bdevname(rdev->bdev,b));
3911                                 list_move(&rdev->same_set, &candidates);
3912                         }
3913                 /*
3914                  * now we have a set of devices, with all of them having
3915                  * mostly sane superblocks. It's time to allocate the
3916                  * mddev.
3917                  */
3918                 if (part) {
3919                         dev = MKDEV(mdp_major,
3920                                     rdev0->preferred_minor << MdpMinorShift);
3921                         unit = MINOR(dev) >> MdpMinorShift;
3922                 } else {
3923                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3924                         unit = MINOR(dev);
3925                 }
3926                 if (rdev0->preferred_minor != unit) {
3927                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3928                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3929                         break;
3930                 }
3931
3932                 md_probe(dev, NULL, NULL);
3933                 mddev = mddev_find(dev);
3934                 if (!mddev || !mddev->gendisk) {
3935                         if (mddev)
3936                                 mddev_put(mddev);
3937                         printk(KERN_ERR
3938                                 "md: cannot allocate memory for md drive.\n");
3939                         break;
3940                 }
3941                 if (mddev_lock(mddev))
3942                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3943                                mdname(mddev));
3944                 else if (mddev->raid_disks || mddev->major_version
3945                          || !list_empty(&mddev->disks)) {
3946                         printk(KERN_WARNING
3947                                 "md: %s already running, cannot run %s\n",
3948                                 mdname(mddev), bdevname(rdev0->bdev,b));
3949                         mddev_unlock(mddev);
3950                 } else {
3951                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3952                         mddev->persistent = 1;
3953                         rdev_for_each_list(rdev, tmp, candidates) {
3954                                 list_del_init(&rdev->same_set);
3955                                 if (bind_rdev_to_array(rdev, mddev))
3956                                         export_rdev(rdev);
3957                         }
3958                         autorun_array(mddev);
3959                         mddev_unlock(mddev);
3960                 }
3961                 /* on success, candidates will be empty, on error
3962                  * it won't...
3963                  */
3964                 rdev_for_each_list(rdev, tmp, candidates)
3965                         export_rdev(rdev);
3966                 mddev_put(mddev);
3967         }
3968         printk(KERN_INFO "md: ... autorun DONE.\n");
3969 }
3970 #endif /* !MODULE */
3971
3972 static int get_version(void __user * arg)
3973 {
3974         mdu_version_t ver;
3975
3976         ver.major = MD_MAJOR_VERSION;
3977         ver.minor = MD_MINOR_VERSION;
3978         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3979
3980         if (copy_to_user(arg, &ver, sizeof(ver)))
3981                 return -EFAULT;
3982
3983         return 0;
3984 }
3985
3986 static int get_array_info(mddev_t * mddev, void __user * arg)
3987 {
3988         mdu_array_info_t info;
3989         int nr,working,active,failed,spare;
3990         mdk_rdev_t *rdev;
3991         struct list_head *tmp;
3992
3993         nr=working=active=failed=spare=0;
3994         rdev_for_each(rdev, tmp, mddev) {
3995                 nr++;
3996                 if (test_bit(Faulty, &rdev->flags))
3997                         failed++;
3998                 else {
3999                         working++;
4000                         if (test_bit(In_sync, &rdev->flags))
4001                                 active++;
4002                         else
4003                                 spare++;
4004                 }
4005         }
4006
4007         info.major_version = mddev->major_version;
4008         info.minor_version = mddev->minor_version;
4009         info.patch_version = MD_PATCHLEVEL_VERSION;
4010         info.ctime         = mddev->ctime;
4011         info.level         = mddev->level;
4012         info.size          = mddev->size;
4013         if (info.size != mddev->size) /* overflow */
4014                 info.size = -1;
4015         info.nr_disks      = nr;
4016         info.raid_disks    = mddev->raid_disks;
4017         info.md_minor      = mddev->md_minor;
4018         info.not_persistent= !mddev->persistent;
4019
4020         info.utime         = mddev->utime;
4021         info.state         = 0;
4022         if (mddev->in_sync)
4023                 info.state = (1<<MD_SB_CLEAN);
4024         if (mddev->bitmap && mddev->bitmap_offset)
4025                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4026         info.active_disks  = active;
4027         info.working_disks = working;
4028         info.failed_disks  = failed;
4029         info.spare_disks   = spare;
4030
4031         info.layout        = mddev->layout;
4032         info.chunk_size    = mddev->chunk_size;
4033
4034         if (copy_to_user(arg, &info, sizeof(info)))
4035                 return -EFAULT;
4036
4037         return 0;
4038 }
4039
4040 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4041 {
4042         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4043         char *ptr, *buf = NULL;
4044         int err = -ENOMEM;
4045
4046         md_allow_write(mddev);
4047
4048         file = kmalloc(sizeof(*file), GFP_KERNEL);
4049         if (!file)
4050                 goto out;
4051
4052         /* bitmap disabled, zero the first byte and copy out */
4053         if (!mddev->bitmap || !mddev->bitmap->file) {
4054                 file->pathname[0] = '\0';
4055                 goto copy_out;
4056         }
4057
4058         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4059         if (!buf)
4060                 goto out;
4061
4062         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4063         if (IS_ERR(ptr))
4064                 goto out;
4065
4066         strcpy(file->pathname, ptr);
4067
4068 copy_out:
4069         err = 0;
4070         if (copy_to_user(arg, file, sizeof(*file)))
4071                 err = -EFAULT;
4072 out:
4073         kfree(buf);
4074         kfree(file);
4075         return err;
4076 }
4077
4078 static int get_disk_info(mddev_t * mddev, void __user * arg)
4079 {
4080         mdu_disk_info_t info;
4081         unsigned int nr;
4082         mdk_rdev_t *rdev;
4083
4084         if (copy_from_user(&info, arg, sizeof(info)))
4085                 return -EFAULT;
4086
4087         nr = info.number;
4088
4089         rdev = find_rdev_nr(mddev, nr);
4090         if (rdev) {
4091                 info.major = MAJOR(rdev->bdev->bd_dev);
4092                 info.minor = MINOR(rdev->bdev->bd_dev);
4093                 info.raid_disk = rdev->raid_disk;
4094                 info.state = 0;
4095                 if (test_bit(Faulty, &rdev->flags))
4096                         info.state |= (1<<MD_DISK_FAULTY);
4097                 else if (test_bit(In_sync, &rdev->flags)) {
4098                         info.state |= (1<<MD_DISK_ACTIVE);
4099                         info.state |= (1<<MD_DISK_SYNC);
4100                 }
4101                 if (test_bit(WriteMostly, &rdev->flags))
4102                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4103         } else {
4104                 info.major = info.minor = 0;
4105                 info.raid_disk = -1;
4106                 info.state = (1<<MD_DISK_REMOVED);
4107         }
4108
4109         if (copy_to_user(arg, &info, sizeof(info)))
4110                 return -EFAULT;
4111
4112         return 0;
4113 }
4114
4115 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4116 {
4117         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4118         mdk_rdev_t *rdev;
4119         dev_t dev = MKDEV(info->major,info->minor);
4120
4121         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4122                 return -EOVERFLOW;
4123
4124         if (!mddev->raid_disks) {
4125                 int err;
4126                 /* expecting a device which has a superblock */
4127                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4128                 if (IS_ERR(rdev)) {
4129                         printk(KERN_WARNING
4130                                 "md: md_import_device returned %ld\n",
4131                                 PTR_ERR(rdev));
4132                         return PTR_ERR(rdev);
4133                 }
4134                 if (!list_empty(&mddev->disks)) {
4135                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4136                                                         mdk_rdev_t, same_set);
4137                         int err = super_types[mddev->major_version]
4138                                 .load_super(rdev, rdev0, mddev->minor_version);
4139                         if (err < 0) {
4140                                 printk(KERN_WARNING
4141                                         "md: %s has different UUID to %s\n",
4142                                         bdevname(rdev->bdev,b),
4143                                         bdevname(rdev0->bdev,b2));
4144                                 export_rdev(rdev);
4145                                 return -EINVAL;
4146                         }
4147                 }
4148                 err = bind_rdev_to_array(rdev, mddev);
4149                 if (err)
4150                         export_rdev(rdev);
4151                 return err;
4152         }
4153
4154         /*
4155          * add_new_disk can be used once the array is assembled
4156          * to add "hot spares".  They must already have a superblock
4157          * written
4158          */
4159         if (mddev->pers) {
4160                 int err;
4161                 if (!mddev->pers->hot_add_disk) {
4162                         printk(KERN_WARNING
4163                                 "%s: personality does not support diskops!\n",
4164                                mdname(mddev));
4165                         return -EINVAL;
4166                 }
4167                 if (mddev->persistent)
4168                         rdev = md_import_device(dev, mddev->major_version,
4169                                                 mddev->minor_version);
4170                 else
4171                         rdev = md_import_device(dev, -1, -1);
4172                 if (IS_ERR(rdev)) {
4173                         printk(KERN_WARNING
4174                                 "md: md_import_device returned %ld\n",
4175                                 PTR_ERR(rdev));
4176                         return PTR_ERR(rdev);
4177                 }
4178                 /* set save_raid_disk if appropriate */
4179                 if (!mddev->persistent) {
4180                         if (info->state & (1<<MD_DISK_SYNC)  &&
4181                             info->raid_disk < mddev->raid_disks)
4182                                 rdev->raid_disk = info->raid_disk;
4183                         else
4184                                 rdev->raid_disk = -1;
4185                 } else
4186                         super_types[mddev->major_version].
4187                                 validate_super(mddev, rdev);
4188                 rdev->saved_raid_disk = rdev->raid_disk;
4189
4190                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4191                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4192                         set_bit(WriteMostly, &rdev->flags);
4193
4194                 rdev->raid_disk = -1;
4195                 err = bind_rdev_to_array(rdev, mddev);
4196                 if (!err && !mddev->pers->hot_remove_disk) {
4197                         /* If there is hot_add_disk but no hot_remove_disk
4198                          * then added disks for geometry changes,
4199                          * and should be added immediately.
4200                          */
4201                         super_types[mddev->major_version].
4202                                 validate_super(mddev, rdev);
4203                         err = mddev->pers->hot_add_disk(mddev, rdev);
4204                         if (err)
4205                                 unbind_rdev_from_array(rdev);
4206                 }
4207                 if (err)
4208                         export_rdev(rdev);
4209
4210                 md_update_sb(mddev, 1);
4211                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4212                 md_wakeup_thread(mddev->thread);
4213                 return err;
4214         }
4215
4216         /* otherwise, add_new_disk is only allowed
4217          * for major_version==0 superblocks
4218          */
4219         if (mddev->major_version != 0) {
4220                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4221                        mdname(mddev));
4222                 return -EINVAL;
4223         }
4224
4225         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4226                 int err;
4227                 rdev = md_import_device (dev, -1, 0);
4228                 if (IS_ERR(rdev)) {
4229                         printk(KERN_WARNING
4230                                 "md: error, md_import_device() returned %ld\n",
4231                                 PTR_ERR(rdev));
4232                         return PTR_ERR(rdev);
4233                 }
4234                 rdev->desc_nr = info->number;
4235                 if (info->raid_disk < mddev->raid_disks)
4236                         rdev->raid_disk = info->raid_disk;
4237                 else
4238                         rdev->raid_disk = -1;
4239
4240                 if (rdev->raid_disk < mddev->raid_disks)
4241                         if (info->state & (1<<MD_DISK_SYNC))
4242                                 set_bit(In_sync, &rdev->flags);
4243
4244                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4245                         set_bit(WriteMostly, &rdev->flags);
4246
4247                 if (!mddev->persistent) {
4248                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4249                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4250                 } else
4251                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4252                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4253
4254                 err = bind_rdev_to_array(rdev, mddev);
4255                 if (err) {
4256                         export_rdev(rdev);
4257                         return err;
4258                 }
4259         }
4260
4261         return 0;
4262 }
4263
4264 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4265 {
4266         char b[BDEVNAME_SIZE];
4267         mdk_rdev_t *rdev;
4268
4269         if (!mddev->pers)
4270                 return -ENODEV;
4271
4272         rdev = find_rdev(mddev, dev);
4273         if (!rdev)
4274                 return -ENXIO;
4275
4276         if (rdev->raid_disk >= 0)
4277                 goto busy;
4278
4279         kick_rdev_from_array(rdev);
4280         md_update_sb(mddev, 1);
4281         md_new_event(mddev);
4282
4283         return 0;
4284 busy:
4285         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4286                 bdevname(rdev->bdev,b), mdname(mddev));
4287         return -EBUSY;
4288 }
4289
4290 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4291 {
4292         char b[BDEVNAME_SIZE];
4293         int err;
4294         unsigned int size;
4295         mdk_rdev_t *rdev;
4296
4297         if (!mddev->pers)
4298                 return -ENODEV;
4299
4300         if (mddev->major_version != 0) {
4301                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4302                         " version-0 superblocks.\n",
4303                         mdname(mddev));
4304                 return -EINVAL;
4305         }
4306         if (!mddev->pers->hot_add_disk) {
4307                 printk(KERN_WARNING
4308                         "%s: personality does not support diskops!\n",
4309                         mdname(mddev));
4310                 return -EINVAL;
4311         }
4312
4313         rdev = md_import_device (dev, -1, 0);
4314         if (IS_ERR(rdev)) {
4315                 printk(KERN_WARNING
4316                         "md: error, md_import_device() returned %ld\n",
4317                         PTR_ERR(rdev));
4318                 return -EINVAL;
4319         }
4320
4321         if (mddev->persistent)
4322                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4323         else
4324                 rdev->sb_offset =
4325                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4326
4327         size = calc_dev_size(rdev, mddev->chunk_size);
4328         rdev->size = size;
4329
4330         if (test_bit(Faulty, &rdev->flags)) {
4331                 printk(KERN_WARNING
4332                         "md: can not hot-add faulty %s disk to %s!\n",
4333                         bdevname(rdev->bdev,b), mdname(mddev));
4334                 err = -EINVAL;
4335                 goto abort_export;
4336         }
4337         clear_bit(In_sync, &rdev->flags);
4338         rdev->desc_nr = -1;
4339         rdev->saved_raid_disk = -1;
4340         err = bind_rdev_to_array(rdev, mddev);
4341         if (err)
4342                 goto abort_export;
4343
4344         /*
4345          * The rest should better be atomic, we can have disk failures
4346          * noticed in interrupt contexts ...
4347          */
4348
4349         if (rdev->desc_nr == mddev->max_disks) {
4350                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4351                         mdname(mddev));
4352                 err = -EBUSY;
4353                 goto abort_unbind_export;
4354         }
4355
4356         rdev->raid_disk = -1;
4357
4358         md_update_sb(mddev, 1);
4359
4360         /*
4361          * Kick recovery, maybe this spare has to be added to the
4362          * array immediately.
4363          */
4364         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4365         md_wakeup_thread(mddev->thread);
4366         md_new_event(mddev);
4367         return 0;
4368
4369 abort_unbind_export:
4370         unbind_rdev_from_array(rdev);
4371
4372 abort_export:
4373         export_rdev(rdev);
4374         return err;
4375 }
4376
4377 static int set_bitmap_file(mddev_t *mddev, int fd)
4378 {
4379         int err;
4380
4381         if (mddev->pers) {
4382                 if (!mddev->pers->quiesce)
4383                         return -EBUSY;
4384                 if (mddev->recovery || mddev->sync_thread)
4385                         return -EBUSY;
4386                 /* we should be able to change the bitmap.. */
4387         }
4388
4389
4390         if (fd >= 0) {
4391                 if (mddev->bitmap)
4392                         return -EEXIST; /* cannot add when bitmap is present */
4393                 mddev->bitmap_file = fget(fd);
4394
4395                 if (mddev->bitmap_file == NULL) {
4396                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4397                                mdname(mddev));
4398                         return -EBADF;
4399                 }
4400
4401                 err = deny_bitmap_write_access(mddev->bitmap_file);
4402                 if (err) {
4403                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4404                                mdname(mddev));
4405                         fput(mddev->bitmap_file);
4406                         mddev->bitmap_file = NULL;
4407                         return err;
4408                 }
4409                 mddev->bitmap_offset = 0; /* file overrides offset */
4410         } else if (mddev->bitmap == NULL)
4411                 return -ENOENT; /* cannot remove what isn't there */
4412         err = 0;
4413         if (mddev->pers) {
4414                 mddev->pers->quiesce(mddev, 1);
4415                 if (fd >= 0)
4416                         err = bitmap_create(mddev);
4417                 if (fd < 0 || err) {
4418                         bitmap_destroy(mddev);
4419                         fd = -1; /* make sure to put the file */
4420                 }
4421                 mddev->pers->quiesce(mddev, 0);
4422         }
4423         if (fd < 0) {
4424                 if (mddev->bitmap_file) {
4425                         restore_bitmap_write_access(mddev->bitmap_file);
4426                         fput(mddev->bitmap_file);
4427                 }
4428                 mddev->bitmap_file = NULL;
4429         }
4430
4431         return err;
4432 }
4433
4434 /*
4435  * set_array_info is used two different ways
4436  * The original usage is when creating a new array.
4437  * In this usage, raid_disks is > 0 and it together with
4438  *  level, size, not_persistent,layout,chunksize determine the
4439  *  shape of the array.
4440  *  This will always create an array with a type-0.90.0 superblock.
4441  * The newer usage is when assembling an array.
4442  *  In this case raid_disks will be 0, and the major_version field is
4443  *  use to determine which style super-blocks are to be found on the devices.
4444  *  The minor and patch _version numbers are also kept incase the
4445  *  super_block handler wishes to interpret them.
4446  */
4447 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4448 {
4449
4450         if (info->raid_disks == 0) {
4451                 /* just setting version number for superblock loading */
4452                 if (info->major_version < 0 ||
4453                     info->major_version >= ARRAY_SIZE(super_types) ||
4454                     super_types[info->major_version].name == NULL) {
4455                         /* maybe try to auto-load a module? */
4456                         printk(KERN_INFO
4457                                 "md: superblock version %d not known\n",
4458                                 info->major_version);
4459                         return -EINVAL;
4460                 }
4461                 mddev->major_version = info->major_version;
4462                 mddev->minor_version = info->minor_version;
4463                 mddev->patch_version = info->patch_version;
4464                 mddev->persistent = !info->not_persistent;
4465                 return 0;
4466         }
4467         mddev->major_version = MD_MAJOR_VERSION;
4468         mddev->minor_version = MD_MINOR_VERSION;
4469         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4470         mddev->ctime         = get_seconds();
4471
4472         mddev->level         = info->level;
4473         mddev->clevel[0]     = 0;
4474         mddev->size          = info->size;
4475         mddev->raid_disks    = info->raid_disks;
4476         /* don't set md_minor, it is determined by which /dev/md* was
4477          * openned
4478          */
4479         if (info->state & (1<<MD_SB_CLEAN))
4480                 mddev->recovery_cp = MaxSector;
4481         else
4482                 mddev->recovery_cp = 0;
4483         mddev->persistent    = ! info->not_persistent;
4484         mddev->external      = 0;
4485
4486         mddev->layout        = info->layout;
4487         mddev->chunk_size    = info->chunk_size;
4488
4489         mddev->max_disks     = MD_SB_DISKS;
4490
4491         if (mddev->persistent)
4492                 mddev->flags         = 0;
4493         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4494
4495         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4496         mddev->bitmap_offset = 0;
4497
4498         mddev->reshape_position = MaxSector;
4499
4500         /*
4501          * Generate a 128 bit UUID
4502          */
4503         get_random_bytes(mddev->uuid, 16);
4504
4505         mddev->new_level = mddev->level;
4506         mddev->new_chunk = mddev->chunk_size;
4507         mddev->new_layout = mddev->layout;
4508         mddev->delta_disks = 0;
4509
4510         return 0;
4511 }
4512
4513 static int update_size(mddev_t *mddev, unsigned long size)
4514 {
4515         mdk_rdev_t * rdev;
4516         int rv;
4517         struct list_head *tmp;
4518         int fit = (size == 0);
4519
4520         if (mddev->pers->resize == NULL)
4521                 return -EINVAL;
4522         /* The "size" is the amount of each device that is used.
4523          * This can only make sense for arrays with redundancy.
4524          * linear and raid0 always use whatever space is available
4525          * We can only consider changing the size if no resync
4526          * or reconstruction is happening, and if the new size
4527          * is acceptable. It must fit before the sb_offset or,
4528          * if that is <data_offset, it must fit before the
4529          * size of each device.
4530          * If size is zero, we find the largest size that fits.
4531          */
4532         if (mddev->sync_thread)
4533                 return -EBUSY;
4534         rdev_for_each(rdev, tmp, mddev) {
4535                 sector_t avail;
4536                 avail = rdev->size * 2;
4537
4538                 if (fit && (size == 0 || size > avail/2))
4539                         size = avail/2;
4540                 if (avail < ((sector_t)size << 1))
4541                         return -ENOSPC;
4542         }
4543         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4544         if (!rv) {
4545                 struct block_device *bdev;
4546
4547                 bdev = bdget_disk(mddev->gendisk, 0);
4548                 if (bdev) {
4549                         mutex_lock(&bdev->bd_inode->i_mutex);
4550                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4551                         mutex_unlock(&bdev->bd_inode->i_mutex);
4552                         bdput(bdev);
4553                 }
4554         }
4555         return rv;
4556 }
4557
4558 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4559 {
4560         int rv;
4561         /* change the number of raid disks */
4562         if (mddev->pers->check_reshape == NULL)
4563                 return -EINVAL;
4564         if (raid_disks <= 0 ||
4565             raid_disks >= mddev->max_disks)
4566                 return -EINVAL;
4567         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4568                 return -EBUSY;
4569         mddev->delta_disks = raid_disks - mddev->raid_disks;
4570
4571         rv = mddev->pers->check_reshape(mddev);
4572         return rv;
4573 }
4574
4575
4576 /*
4577  * update_array_info is used to change the configuration of an
4578  * on-line array.
4579  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4580  * fields in the info are checked against the array.
4581  * Any differences that cannot be handled will cause an error.
4582  * Normally, only one change can be managed at a time.
4583  */
4584 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4585 {
4586         int rv = 0;
4587         int cnt = 0;
4588         int state = 0;
4589
4590         /* calculate expected state,ignoring low bits */
4591         if (mddev->bitmap && mddev->bitmap_offset)
4592                 state |= (1 << MD_SB_BITMAP_PRESENT);
4593
4594         if (mddev->major_version != info->major_version ||
4595             mddev->minor_version != info->minor_version ||
4596 /*          mddev->patch_version != info->patch_version || */
4597             mddev->ctime         != info->ctime         ||
4598             mddev->level         != info->level         ||
4599 /*          mddev->layout        != info->layout        || */
4600             !mddev->persistent   != info->not_persistent||
4601             mddev->chunk_size    != info->chunk_size    ||
4602             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4603             ((state^info->state) & 0xfffffe00)
4604                 )
4605                 return -EINVAL;
4606         /* Check there is only one change */
4607         if (info->size >= 0 && mddev->size != info->size) cnt++;
4608         if (mddev->raid_disks != info->raid_disks) cnt++;
4609         if (mddev->layout != info->layout) cnt++;
4610         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4611         if (cnt == 0) return 0;
4612         if (cnt > 1) return -EINVAL;
4613
4614         if (mddev->layout != info->layout) {
4615                 /* Change layout
4616                  * we don't need to do anything at the md level, the
4617                  * personality will take care of it all.
4618                  */
4619                 if (mddev->pers->reconfig == NULL)
4620                         return -EINVAL;
4621                 else
4622                         return mddev->pers->reconfig(mddev, info->layout, -1);
4623         }
4624         if (info->size >= 0 && mddev->size != info->size)
4625                 rv = update_size(mddev, info->size);
4626
4627         if (mddev->raid_disks    != info->raid_disks)
4628                 rv = update_raid_disks(mddev, info->raid_disks);
4629
4630         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4631                 if (mddev->pers->quiesce == NULL)
4632                         return -EINVAL;
4633                 if (mddev->recovery || mddev->sync_thread)
4634                         return -EBUSY;
4635                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4636                         /* add the bitmap */
4637                         if (mddev->bitmap)
4638                                 return -EEXIST;
4639                         if (mddev->default_bitmap_offset == 0)
4640                                 return -EINVAL;
4641                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4642                         mddev->pers->quiesce(mddev, 1);
4643                         rv = bitmap_create(mddev);
4644                         if (rv)
4645                                 bitmap_destroy(mddev);
4646                         mddev->pers->quiesce(mddev, 0);
4647                 } else {
4648                         /* remove the bitmap */
4649                         if (!mddev->bitmap)
4650                                 return -ENOENT;
4651                         if (mddev->bitmap->file)
4652                                 return -EINVAL;
4653                         mddev->pers->quiesce(mddev, 1);
4654                         bitmap_destroy(mddev);
4655                         mddev->pers->quiesce(mddev, 0);
4656                         mddev->bitmap_offset = 0;
4657                 }
4658         }
4659         md_update_sb(mddev, 1);
4660         return rv;
4661 }
4662
4663 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4664 {
4665         mdk_rdev_t *rdev;
4666
4667         if (mddev->pers == NULL)
4668                 return -ENODEV;
4669
4670         rdev = find_rdev(mddev, dev);
4671         if (!rdev)
4672                 return -ENODEV;
4673
4674         md_error(mddev, rdev);
4675         return 0;
4676 }
4677
4678 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4679 {
4680         mddev_t *mddev = bdev->bd_disk->private_data;
4681
4682         geo->heads = 2;
4683         geo->sectors = 4;
4684         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4685         return 0;
4686 }
4687
4688 static int md_ioctl(struct inode *inode, struct file *file,
4689                         unsigned int cmd, unsigned long arg)
4690 {
4691         int err = 0;
4692         void __user *argp = (void __user *)arg;
4693         mddev_t *mddev = NULL;
4694
4695         if (!capable(CAP_SYS_ADMIN))
4696                 return -EACCES;
4697
4698         /*
4699          * Commands dealing with the RAID driver but not any
4700          * particular array:
4701          */
4702         switch (cmd)
4703         {
4704                 case RAID_VERSION:
4705                         err = get_version(argp);
4706                         goto done;
4707
4708                 case PRINT_RAID_DEBUG:
4709                         err = 0;
4710                         md_print_devices();
4711                         goto done;
4712
4713 #ifndef MODULE
4714                 case RAID_AUTORUN:
4715                         err = 0;
4716                         autostart_arrays(arg);
4717                         goto done;
4718 #endif
4719                 default:;
4720         }
4721
4722         /*
4723          * Commands creating/starting a new array:
4724          */
4725
4726         mddev = inode->i_bdev->bd_disk->private_data;
4727
4728         if (!mddev) {
4729                 BUG();
4730                 goto abort;
4731         }
4732
4733         err = mddev_lock(mddev);
4734         if (err) {
4735                 printk(KERN_INFO
4736                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4737                         err, cmd);
4738                 goto abort;
4739         }
4740
4741         switch (cmd)
4742         {
4743                 case SET_ARRAY_INFO:
4744                         {
4745                                 mdu_array_info_t info;
4746                                 if (!arg)
4747                                         memset(&info, 0, sizeof(info));
4748                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4749                                         err = -EFAULT;
4750                                         goto abort_unlock;
4751                                 }
4752                                 if (mddev->pers) {
4753                                         err = update_array_info(mddev, &info);
4754                                         if (err) {
4755                                                 printk(KERN_WARNING "md: couldn't update"
4756                                                        " array info. %d\n", err);
4757                                                 goto abort_unlock;
4758                                         }
4759                                         goto done_unlock;
4760                                 }
4761                                 if (!list_empty(&mddev->disks)) {
4762                                         printk(KERN_WARNING
4763                                                "md: array %s already has disks!\n",
4764                                                mdname(mddev));
4765                                         err = -EBUSY;
4766                                         goto abort_unlock;
4767                                 }
4768                                 if (mddev->raid_disks) {
4769                                         printk(KERN_WARNING
4770                                                "md: array %s already initialised!\n",
4771                                                mdname(mddev));
4772                                         err = -EBUSY;
4773                                         goto abort_unlock;
4774                                 }
4775                                 err = set_array_info(mddev, &info);
4776                                 if (err) {
4777                                         printk(KERN_WARNING "md: couldn't set"
4778                                                " array info. %d\n", err);
4779                                         goto abort_unlock;
4780                                 }
4781                         }
4782                         goto done_unlock;
4783
4784                 default:;
4785         }
4786
4787         /*
4788          * Commands querying/configuring an existing array:
4789          */
4790         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4791          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4792         if ((!mddev->raid_disks && !mddev->external)
4793             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4794             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4795             && cmd != GET_BITMAP_FILE) {
4796                 err = -ENODEV;
4797                 goto abort_unlock;
4798         }
4799
4800         /*
4801          * Commands even a read-only array can execute:
4802          */
4803         switch (cmd)
4804         {
4805                 case GET_ARRAY_INFO:
4806                         err = get_array_info(mddev, argp);
4807                         goto done_unlock;
4808
4809                 case GET_BITMAP_FILE:
4810                         err = get_bitmap_file(mddev, argp);
4811                         goto done_unlock;
4812
4813                 case GET_DISK_INFO:
4814                         err = get_disk_info(mddev, argp);
4815                         goto done_unlock;
4816
4817                 case RESTART_ARRAY_RW:
4818                         err = restart_array(mddev);
4819                         goto done_unlock;
4820
4821                 case STOP_ARRAY:
4822                         err = do_md_stop (mddev, 0);
4823                         goto done_unlock;
4824
4825                 case STOP_ARRAY_RO:
4826                         err = do_md_stop (mddev, 1);
4827                         goto done_unlock;
4828
4829         /*
4830          * We have a problem here : there is no easy way to give a CHS
4831          * virtual geometry. We currently pretend that we have a 2 heads
4832          * 4 sectors (with a BIG number of cylinders...). This drives
4833          * dosfs just mad... ;-)
4834          */
4835         }
4836
4837         /*
4838          * The remaining ioctls are changing the state of the
4839          * superblock, so we do not allow them on read-only arrays.
4840          * However non-MD ioctls (e.g. get-size) will still come through
4841          * here and hit the 'default' below, so only disallow
4842          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4843          */
4844         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4845             mddev->ro && mddev->pers) {
4846                 if (mddev->ro == 2) {
4847                         mddev->ro = 0;
4848                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4849                 md_wakeup_thread(mddev->thread);
4850
4851                 } else {
4852                         err = -EROFS;
4853                         goto abort_unlock;
4854                 }
4855         }
4856
4857         switch (cmd)
4858         {
4859                 case ADD_NEW_DISK:
4860                 {
4861                         mdu_disk_info_t info;
4862                         if (copy_from_user(&info, argp, sizeof(info)))
4863                                 err = -EFAULT;
4864                         else
4865                                 err = add_new_disk(mddev, &info);
4866                         goto done_unlock;
4867                 }
4868
4869                 case HOT_REMOVE_DISK:
4870                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4871                         goto done_unlock;
4872
4873                 case HOT_ADD_DISK:
4874                         err = hot_add_disk(mddev, new_decode_dev(arg));
4875                         goto done_unlock;
4876
4877                 case SET_DISK_FAULTY:
4878                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4879                         goto done_unlock;
4880
4881                 case RUN_ARRAY:
4882                         err = do_md_run (mddev);
4883                         goto done_unlock;
4884
4885                 case SET_BITMAP_FILE:
4886                         err = set_bitmap_file(mddev, (int)arg);
4887                         goto done_unlock;
4888
4889                 default:
4890                         err = -EINVAL;
4891                         goto abort_unlock;
4892         }
4893
4894 done_unlock:
4895 abort_unlock:
4896         mddev_unlock(mddev);
4897
4898         return err;
4899 done:
4900         if (err)
4901                 MD_BUG();
4902 abort:
4903         return err;
4904 }
4905
4906 static int md_open(struct inode *inode, struct file *file)
4907 {
4908         /*
4909          * Succeed if we can lock the mddev, which confirms that
4910          * it isn't being stopped right now.
4911          */
4912         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4913         int err;
4914
4915         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4916                 goto out;
4917
4918         err = 0;
4919         mddev_get(mddev);
4920         mddev_unlock(mddev);
4921
4922         check_disk_change(inode->i_bdev);
4923  out:
4924         return err;
4925 }
4926
4927 static int md_release(struct inode *inode, struct file * file)
4928 {
4929         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4930
4931         BUG_ON(!mddev);
4932         mddev_put(mddev);
4933
4934         return 0;
4935 }
4936
4937 static int md_media_changed(struct gendisk *disk)
4938 {
4939         mddev_t *mddev = disk->private_data;
4940
4941         return mddev->changed;
4942 }
4943
4944 static int md_revalidate(struct gendisk *disk)
4945 {
4946         mddev_t *mddev = disk->private_data;
4947
4948         mddev->changed = 0;
4949         return 0;
4950 }
4951 static struct block_device_operations md_fops =
4952 {
4953         .owner          = THIS_MODULE,
4954         .open           = md_open,
4955         .release        = md_release,
4956         .ioctl          = md_ioctl,
4957         .getgeo         = md_getgeo,
4958         .media_changed  = md_media_changed,
4959         .revalidate_disk= md_revalidate,
4960 };
4961
4962 static int md_thread(void * arg)
4963 {
4964         mdk_thread_t *thread = arg;
4965
4966         /*
4967          * md_thread is a 'system-thread', it's priority should be very
4968          * high. We avoid resource deadlocks individually in each
4969          * raid personality. (RAID5 does preallocation) We also use RR and
4970          * the very same RT priority as kswapd, thus we will never get
4971          * into a priority inversion deadlock.
4972          *
4973          * we definitely have to have equal or higher priority than
4974          * bdflush, otherwise bdflush will deadlock if there are too
4975          * many dirty RAID5 blocks.
4976          */
4977
4978         allow_signal(SIGKILL);
4979         while (!kthread_should_stop()) {
4980
4981                 /* We need to wait INTERRUPTIBLE so that
4982                  * we don't add to the load-average.
4983                  * That means we need to be sure no signals are
4984                  * pending
4985                  */
4986                 if (signal_pending(current))
4987                         flush_signals(current);
4988
4989                 wait_event_interruptible_timeout
4990                         (thread->wqueue,
4991                          test_bit(THREAD_WAKEUP, &thread->flags)
4992                          || kthread_should_stop(),
4993                          thread->timeout);
4994
4995                 clear_bit(THREAD_WAKEUP, &thread->flags);
4996
4997                 thread->run(thread->mddev);
4998         }
4999
5000         return 0;
5001 }
5002
5003 void md_wakeup_thread(mdk_thread_t *thread)
5004 {
5005         if (thread) {
5006                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5007                 set_bit(THREAD_WAKEUP, &thread->flags);
5008                 wake_up(&thread->wqueue);
5009         }
5010 }
5011
5012 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5013                                  const char *name)
5014 {
5015         mdk_thread_t *thread;
5016
5017         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5018         if (!thread)
5019                 return NULL;
5020
5021         init_waitqueue_head(&thread->wqueue);
5022
5023         thread->run = run;
5024         thread->mddev = mddev;
5025         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5026         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5027         if (IS_ERR(thread->tsk)) {
5028                 kfree(thread);
5029                 return NULL;
5030         }
5031         return thread;
5032 }
5033
5034 void md_unregister_thread(mdk_thread_t *thread)
5035 {
5036         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5037
5038         kthread_stop(thread->tsk);
5039         kfree(thread);
5040 }
5041
5042 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5043 {
5044         if (!mddev) {
5045                 MD_BUG();
5046                 return;
5047         }
5048
5049         if (!rdev || test_bit(Faulty, &rdev->flags))
5050                 return;
5051
5052         if (mddev->external)
5053                 set_bit(Blocked, &rdev->flags);
5054 /*
5055         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5056                 mdname(mddev),
5057                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5058                 __builtin_return_address(0),__builtin_return_address(1),
5059                 __builtin_return_address(2),__builtin_return_address(3));
5060 */
5061         if (!mddev->pers)
5062                 return;
5063         if (!mddev->pers->error_handler)
5064                 return;
5065         mddev->pers->error_handler(mddev,rdev);
5066         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5067         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5068         md_wakeup_thread(mddev->thread);
5069         md_new_event_inintr(mddev);
5070 }
5071
5072 /* seq_file implementation /proc/mdstat */
5073
5074 static void status_unused(struct seq_file *seq)
5075 {
5076         int i = 0;
5077         mdk_rdev_t *rdev;
5078         struct list_head *tmp;
5079
5080         seq_printf(seq, "unused devices: ");
5081
5082         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5083                 char b[BDEVNAME_SIZE];
5084                 i++;
5085                 seq_printf(seq, "%s ",
5086                               bdevname(rdev->bdev,b));
5087         }
5088         if (!i)
5089                 seq_printf(seq, "<none>");
5090
5091         seq_printf(seq, "\n");
5092 }
5093
5094
5095 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5096 {
5097         sector_t max_blocks, resync, res;
5098         unsigned long dt, db, rt;
5099         int scale;
5100         unsigned int per_milli;
5101
5102         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5103
5104         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5105                 max_blocks = mddev->resync_max_sectors >> 1;
5106         else
5107                 max_blocks = mddev->size;
5108
5109         /*
5110          * Should not happen.
5111          */
5112         if (!max_blocks) {
5113                 MD_BUG();
5114                 return;
5115         }
5116         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5117          * in a sector_t, and (max_blocks>>scale) will fit in a
5118          * u32, as those are the requirements for sector_div.
5119          * Thus 'scale' must be at least 10
5120          */
5121         scale = 10;
5122         if (sizeof(sector_t) > sizeof(unsigned long)) {
5123                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5124                         scale++;
5125         }
5126         res = (resync>>scale)*1000;
5127         sector_div(res, (u32)((max_blocks>>scale)+1));
5128
5129         per_milli = res;
5130         {
5131                 int i, x = per_milli/50, y = 20-x;
5132                 seq_printf(seq, "[");
5133                 for (i = 0; i < x; i++)
5134                         seq_printf(seq, "=");
5135                 seq_printf(seq, ">");
5136                 for (i = 0; i < y; i++)
5137                         seq_printf(seq, ".");
5138                 seq_printf(seq, "] ");
5139         }
5140         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5141                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5142                     "reshape" :
5143                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5144                      "check" :
5145                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5146                       "resync" : "recovery"))),
5147                    per_milli/10, per_milli % 10,
5148                    (unsigned long long) resync,
5149                    (unsigned long long) max_blocks);
5150
5151         /*
5152          * We do not want to overflow, so the order of operands and
5153          * the * 100 / 100 trick are important. We do a +1 to be
5154          * safe against division by zero. We only estimate anyway.
5155          *
5156          * dt: time from mark until now
5157          * db: blocks written from mark until now
5158          * rt: remaining time
5159          */
5160         dt = ((jiffies - mddev->resync_mark) / HZ);
5161         if (!dt) dt++;
5162         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5163                 - mddev->resync_mark_cnt;
5164         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5165
5166         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5167
5168         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5169 }
5170
5171 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5172 {
5173         struct list_head *tmp;
5174         loff_t l = *pos;
5175         mddev_t *mddev;
5176
5177         if (l >= 0x10000)
5178                 return NULL;
5179         if (!l--)
5180                 /* header */
5181                 return (void*)1;
5182
5183         spin_lock(&all_mddevs_lock);
5184         list_for_each(tmp,&all_mddevs)
5185                 if (!l--) {
5186                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5187                         mddev_get(mddev);
5188                         spin_unlock(&all_mddevs_lock);
5189                         return mddev;
5190                 }
5191         spin_unlock(&all_mddevs_lock);
5192         if (!l--)
5193                 return (void*)2;/* tail */
5194         return NULL;
5195 }
5196
5197 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5198 {
5199         struct list_head *tmp;
5200         mddev_t *next_mddev, *mddev = v;
5201
5202         ++*pos;
5203         if (v == (void*)2)
5204                 return NULL;
5205
5206         spin_lock(&all_mddevs_lock);
5207         if (v == (void*)1)
5208                 tmp = all_mddevs.next;
5209         else
5210                 tmp = mddev->all_mddevs.next;
5211         if (tmp != &all_mddevs)
5212                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5213         else {
5214                 next_mddev = (void*)2;
5215                 *pos = 0x10000;
5216         }
5217         spin_unlock(&all_mddevs_lock);
5218
5219         if (v != (void*)1)
5220                 mddev_put(mddev);
5221         return next_mddev;
5222
5223 }
5224
5225 static void md_seq_stop(struct seq_file *seq, void *v)
5226 {
5227         mddev_t *mddev = v;
5228
5229         if (mddev && v != (void*)1 && v != (void*)2)
5230                 mddev_put(mddev);
5231 }
5232
5233 struct mdstat_info {
5234         int event;
5235 };
5236
5237 static int md_seq_show(struct seq_file *seq, void *v)
5238 {
5239         mddev_t *mddev = v;
5240         sector_t size;
5241         struct list_head *tmp2;
5242         mdk_rdev_t *rdev;
5243         struct mdstat_info *mi = seq->private;
5244         struct bitmap *bitmap;
5245
5246         if (v == (void*)1) {
5247                 struct mdk_personality *pers;
5248                 seq_printf(seq, "Personalities : ");
5249                 spin_lock(&pers_lock);
5250                 list_for_each_entry(pers, &pers_list, list)
5251                         seq_printf(seq, "[%s] ", pers->name);
5252
5253                 spin_unlock(&pers_lock);
5254                 seq_printf(seq, "\n");
5255                 mi->event = atomic_read(&md_event_count);
5256                 return 0;
5257         }
5258         if (v == (void*)2) {
5259                 status_unused(seq);
5260                 return 0;
5261         }
5262
5263         if (mddev_lock(mddev) < 0)
5264                 return -EINTR;
5265
5266         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5267                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5268                                                 mddev->pers ? "" : "in");
5269                 if (mddev->pers) {
5270                         if (mddev->ro==1)
5271                                 seq_printf(seq, " (read-only)");
5272                         if (mddev->ro==2)
5273                                 seq_printf(seq, " (auto-read-only)");
5274                         seq_printf(seq, " %s", mddev->pers->name);
5275                 }
5276
5277                 size = 0;
5278                 rdev_for_each(rdev, tmp2, mddev) {
5279                         char b[BDEVNAME_SIZE];
5280                         seq_printf(seq, " %s[%d]",
5281                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5282                         if (test_bit(WriteMostly, &rdev->flags))
5283                                 seq_printf(seq, "(W)");
5284                         if (test_bit(Faulty, &rdev->flags)) {
5285                                 seq_printf(seq, "(F)");
5286                                 continue;
5287                         } else if (rdev->raid_disk < 0)
5288                                 seq_printf(seq, "(S)"); /* spare */
5289                         size += rdev->size;
5290                 }
5291
5292                 if (!list_empty(&mddev->disks)) {
5293                         if (mddev->pers)
5294                                 seq_printf(seq, "\n      %llu blocks",
5295                                         (unsigned long long)mddev->array_size);
5296                         else
5297                                 seq_printf(seq, "\n      %llu blocks",
5298                                         (unsigned long long)size);
5299                 }
5300                 if (mddev->persistent) {
5301                         if (mddev->major_version != 0 ||
5302                             mddev->minor_version != 90) {
5303                                 seq_printf(seq," super %d.%d",
5304                                            mddev->major_version,
5305                                            mddev->minor_version);
5306                         }
5307                 } else if (mddev->external)
5308                         seq_printf(seq, " super external:%s",
5309                                    mddev->metadata_type);
5310                 else
5311                         seq_printf(seq, " super non-persistent");
5312
5313                 if (mddev->pers) {
5314                         mddev->pers->status (seq, mddev);
5315                         seq_printf(seq, "\n      ");
5316                         if (mddev->pers->sync_request) {
5317                                 if (mddev->curr_resync > 2) {
5318                                         status_resync (seq, mddev);
5319                                         seq_printf(seq, "\n      ");
5320                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5321                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5322                                 else if (mddev->recovery_cp < MaxSector)
5323                                         seq_printf(seq, "\tresync=PENDING\n      ");
5324                         }
5325                 } else
5326                         seq_printf(seq, "\n       ");
5327
5328                 if ((bitmap = mddev->bitmap)) {
5329                         unsigned long chunk_kb;
5330                         unsigned long flags;
5331                         spin_lock_irqsave(&bitmap->lock, flags);
5332                         chunk_kb = bitmap->chunksize >> 10;
5333                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5334                                 "%lu%s chunk",
5335                                 bitmap->pages - bitmap->missing_pages,
5336                                 bitmap->pages,
5337                                 (bitmap->pages - bitmap->missing_pages)
5338                                         << (PAGE_SHIFT - 10),
5339                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5340                                 chunk_kb ? "KB" : "B");
5341                         if (bitmap->file) {
5342                                 seq_printf(seq, ", file: ");
5343                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5344                         }
5345
5346                         seq_printf(seq, "\n");
5347                         spin_unlock_irqrestore(&bitmap->lock, flags);
5348                 }
5349
5350                 seq_printf(seq, "\n");
5351         }
5352         mddev_unlock(mddev);
5353
5354         return 0;
5355 }
5356
5357 static struct seq_operations md_seq_ops = {
5358         .start  = md_seq_start,
5359         .next   = md_seq_next,
5360         .stop   = md_seq_stop,
5361         .show   = md_seq_show,
5362 };
5363
5364 static int md_seq_open(struct inode *inode, struct file *file)
5365 {
5366         int error;
5367         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5368         if (mi == NULL)
5369                 return -ENOMEM;
5370
5371         error = seq_open(file, &md_seq_ops);
5372         if (error)
5373                 kfree(mi);
5374         else {
5375                 struct seq_file *p = file->private_data;
5376                 p->private = mi;
5377                 mi->event = atomic_read(&md_event_count);
5378         }
5379         return error;
5380 }
5381
5382 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5383 {
5384         struct seq_file *m = filp->private_data;
5385         struct mdstat_info *mi = m->private;
5386         int mask;
5387
5388         poll_wait(filp, &md_event_waiters, wait);
5389
5390         /* always allow read */
5391         mask = POLLIN | POLLRDNORM;
5392
5393         if (mi->event != atomic_read(&md_event_count))
5394                 mask |= POLLERR | POLLPRI;
5395         return mask;
5396 }
5397
5398 static const struct file_operations md_seq_fops = {
5399         .owner          = THIS_MODULE,
5400         .open           = md_seq_open,
5401         .read           = seq_read,
5402         .llseek         = seq_lseek,
5403         .release        = seq_release_private,
5404         .poll           = mdstat_poll,
5405 };
5406
5407 int register_md_personality(struct mdk_personality *p)
5408 {
5409         spin_lock(&pers_lock);
5410         list_add_tail(&p->list, &pers_list);
5411         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5412         spin_unlock(&pers_lock);
5413         return 0;
5414 }
5415
5416 int unregister_md_personality(struct mdk_personality *p)
5417 {
5418         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5419         spin_lock(&pers_lock);
5420         list_del_init(&p->list);
5421         spin_unlock(&pers_lock);
5422         return 0;
5423 }
5424
5425 static int is_mddev_idle(mddev_t *mddev)
5426 {
5427         mdk_rdev_t * rdev;
5428         struct list_head *tmp;
5429         int idle;
5430         long curr_events;
5431
5432         idle = 1;
5433         rdev_for_each(rdev, tmp, mddev) {
5434                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5435                 curr_events = disk_stat_read(disk, sectors[0]) +
5436                                 disk_stat_read(disk, sectors[1]) -
5437                                 atomic_read(&disk->sync_io);
5438                 /* sync IO will cause sync_io to increase before the disk_stats
5439                  * as sync_io is counted when a request starts, and
5440                  * disk_stats is counted when it completes.
5441                  * So resync activity will cause curr_events to be smaller than
5442                  * when there was no such activity.
5443                  * non-sync IO will cause disk_stat to increase without
5444                  * increasing sync_io so curr_events will (eventually)
5445                  * be larger than it was before.  Once it becomes
5446                  * substantially larger, the test below will cause
5447                  * the array to appear non-idle, and resync will slow
5448                  * down.
5449                  * If there is a lot of outstanding resync activity when
5450                  * we set last_event to curr_events, then all that activity
5451                  * completing might cause the array to appear non-idle
5452                  * and resync will be slowed down even though there might
5453                  * not have been non-resync activity.  This will only
5454                  * happen once though.  'last_events' will soon reflect
5455                  * the state where there is little or no outstanding
5456                  * resync requests, and further resync activity will
5457                  * always make curr_events less than last_events.
5458                  *
5459                  */
5460                 if (curr_events - rdev->last_events > 4096) {
5461                         rdev->last_events = curr_events;
5462                         idle = 0;
5463                 }
5464         }
5465         return idle;
5466 }
5467
5468 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5469 {
5470         /* another "blocks" (512byte) blocks have been synced */
5471         atomic_sub(blocks, &mddev->recovery_active);
5472         wake_up(&mddev->recovery_wait);
5473         if (!ok) {
5474                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5475                 md_wakeup_thread(mddev->thread);
5476                 // stop recovery, signal do_sync ....
5477         }
5478 }
5479
5480
5481 /* md_write_start(mddev, bi)
5482  * If we need to update some array metadata (e.g. 'active' flag
5483  * in superblock) before writing, schedule a superblock update
5484  * and wait for it to complete.
5485  */
5486 void md_write_start(mddev_t *mddev, struct bio *bi)
5487 {
5488         if (bio_data_dir(bi) != WRITE)
5489                 return;
5490
5491         BUG_ON(mddev->ro == 1);
5492         if (mddev->ro == 2) {
5493                 /* need to switch to read/write */
5494                 mddev->ro = 0;
5495                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5496                 md_wakeup_thread(mddev->thread);
5497                 md_wakeup_thread(mddev->sync_thread);
5498         }
5499         atomic_inc(&mddev->writes_pending);
5500         if (mddev->safemode == 1)
5501                 mddev->safemode = 0;
5502         if (mddev->in_sync) {
5503                 spin_lock_irq(&mddev->write_lock);
5504                 if (mddev->in_sync) {
5505                         mddev->in_sync = 0;
5506                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5507                         md_wakeup_thread(mddev->thread);
5508                 }
5509                 spin_unlock_irq(&mddev->write_lock);
5510                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5511         }
5512         wait_event(mddev->sb_wait,
5513                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5514                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5515 }
5516
5517 void md_write_end(mddev_t *mddev)
5518 {
5519         if (atomic_dec_and_test(&mddev->writes_pending)) {
5520                 if (mddev->safemode == 2)
5521                         md_wakeup_thread(mddev->thread);
5522                 else if (mddev->safemode_delay)
5523                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5524         }
5525 }
5526
5527 /* md_allow_write(mddev)
5528  * Calling this ensures that the array is marked 'active' so that writes
5529  * may proceed without blocking.  It is important to call this before
5530  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5531  * Must be called with mddev_lock held.
5532  */
5533 void md_allow_write(mddev_t *mddev)
5534 {
5535         if (!mddev->pers)
5536                 return;
5537         if (mddev->ro)
5538                 return;
5539         if (!mddev->pers->sync_request)
5540                 return;
5541
5542         spin_lock_irq(&mddev->write_lock);
5543         if (mddev->in_sync) {
5544                 mddev->in_sync = 0;
5545                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5546                 if (mddev->safemode_delay &&
5547                     mddev->safemode == 0)
5548                         mddev->safemode = 1;
5549                 spin_unlock_irq(&mddev->write_lock);
5550                 md_update_sb(mddev, 0);
5551
5552                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5553                 /* wait for the dirty state to be recorded in the metadata */
5554                 wait_event(mddev->sb_wait,
5555                            !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5556                            !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5557         } else
5558                 spin_unlock_irq(&mddev->write_lock);
5559 }
5560 EXPORT_SYMBOL_GPL(md_allow_write);
5561
5562 #define SYNC_MARKS      10
5563 #define SYNC_MARK_STEP  (3*HZ)
5564 void md_do_sync(mddev_t *mddev)
5565 {
5566         mddev_t *mddev2;
5567         unsigned int currspeed = 0,
5568                  window;
5569         sector_t max_sectors,j, io_sectors;
5570         unsigned long mark[SYNC_MARKS];
5571         sector_t mark_cnt[SYNC_MARKS];
5572         int last_mark,m;
5573         struct list_head *tmp;
5574         sector_t last_check;
5575         int skipped = 0;
5576         struct list_head *rtmp;
5577         mdk_rdev_t *rdev;
5578         char *desc;
5579
5580         /* just incase thread restarts... */
5581         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5582                 return;
5583         if (mddev->ro) /* never try to sync a read-only array */
5584                 return;
5585
5586         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5587                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5588                         desc = "data-check";
5589                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5590                         desc = "requested-resync";
5591                 else
5592                         desc = "resync";
5593         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5594                 desc = "reshape";
5595         else
5596                 desc = "recovery";
5597
5598         /* we overload curr_resync somewhat here.
5599          * 0 == not engaged in resync at all
5600          * 2 == checking that there is no conflict with another sync
5601          * 1 == like 2, but have yielded to allow conflicting resync to
5602          *              commense
5603          * other == active in resync - this many blocks
5604          *
5605          * Before starting a resync we must have set curr_resync to
5606          * 2, and then checked that every "conflicting" array has curr_resync
5607          * less than ours.  When we find one that is the same or higher
5608          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5609          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5610          * This will mean we have to start checking from the beginning again.
5611          *
5612          */
5613
5614         do {
5615                 mddev->curr_resync = 2;
5616
5617         try_again:
5618                 if (kthread_should_stop()) {
5619                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5620                         goto skip;
5621                 }
5622                 for_each_mddev(mddev2, tmp) {
5623                         if (mddev2 == mddev)
5624                                 continue;
5625                         if (!mddev->parallel_resync
5626                         &&  mddev2->curr_resync
5627                         &&  match_mddev_units(mddev, mddev2)) {
5628                                 DEFINE_WAIT(wq);
5629                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5630                                         /* arbitrarily yield */
5631                                         mddev->curr_resync = 1;
5632                                         wake_up(&resync_wait);
5633                                 }
5634                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5635                                         /* no need to wait here, we can wait the next
5636                                          * time 'round when curr_resync == 2
5637                                          */
5638                                         continue;
5639                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5640                                 if (!kthread_should_stop() &&
5641                                     mddev2->curr_resync >= mddev->curr_resync) {
5642                                         printk(KERN_INFO "md: delaying %s of %s"
5643                                                " until %s has finished (they"
5644                                                " share one or more physical units)\n",
5645                                                desc, mdname(mddev), mdname(mddev2));
5646                                         mddev_put(mddev2);
5647                                         schedule();
5648                                         finish_wait(&resync_wait, &wq);
5649                                         goto try_again;
5650                                 }
5651                                 finish_wait(&resync_wait, &wq);
5652                         }
5653                 }
5654         } while (mddev->curr_resync < 2);
5655
5656         j = 0;
5657         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5658                 /* resync follows the size requested by the personality,
5659                  * which defaults to physical size, but can be virtual size
5660                  */
5661                 max_sectors = mddev->resync_max_sectors;
5662                 mddev->resync_mismatches = 0;
5663                 /* we don't use the checkpoint if there's a bitmap */
5664                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5665                         j = mddev->resync_min;
5666                 else if (!mddev->bitmap)
5667                         j = mddev->recovery_cp;
5668
5669         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5670                 max_sectors = mddev->size << 1;
5671         else {
5672                 /* recovery follows the physical size of devices */
5673                 max_sectors = mddev->size << 1;
5674                 j = MaxSector;
5675                 rdev_for_each(rdev, rtmp, mddev)
5676                         if (rdev->raid_disk >= 0 &&
5677                             !test_bit(Faulty, &rdev->flags) &&
5678                             !test_bit(In_sync, &rdev->flags) &&
5679                             rdev->recovery_offset < j)
5680                                 j = rdev->recovery_offset;
5681         }
5682
5683         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5684         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5685                 " %d KB/sec/disk.\n", speed_min(mddev));
5686         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5687                "(but not more than %d KB/sec) for %s.\n",
5688                speed_max(mddev), desc);
5689
5690         is_mddev_idle(mddev); /* this also initializes IO event counters */
5691
5692         io_sectors = 0;
5693         for (m = 0; m < SYNC_MARKS; m++) {
5694                 mark[m] = jiffies;
5695                 mark_cnt[m] = io_sectors;
5696         }
5697         last_mark = 0;
5698         mddev->resync_mark = mark[last_mark];
5699         mddev->resync_mark_cnt = mark_cnt[last_mark];
5700
5701         /*
5702          * Tune reconstruction:
5703          */
5704         window = 32*(PAGE_SIZE/512);
5705         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5706                 window/2,(unsigned long long) max_sectors/2);
5707
5708         atomic_set(&mddev->recovery_active, 0);
5709         last_check = 0;
5710
5711         if (j>2) {
5712                 printk(KERN_INFO
5713                        "md: resuming %s of %s from checkpoint.\n",
5714                        desc, mdname(mddev));
5715                 mddev->curr_resync = j;
5716         }
5717
5718         while (j < max_sectors) {
5719                 sector_t sectors;
5720
5721                 skipped = 0;
5722                 if (j >= mddev->resync_max) {
5723                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5724                         wait_event(mddev->recovery_wait,
5725                                    mddev->resync_max > j
5726                                    || kthread_should_stop());
5727                 }
5728                 if (kthread_should_stop())
5729                         goto interrupted;
5730                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5731                                                   currspeed < speed_min(mddev));
5732                 if (sectors == 0) {
5733                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5734                         goto out;
5735                 }
5736
5737                 if (!skipped) { /* actual IO requested */
5738                         io_sectors += sectors;
5739                         atomic_add(sectors, &mddev->recovery_active);
5740                 }
5741
5742                 j += sectors;
5743                 if (j>1) mddev->curr_resync = j;
5744                 mddev->curr_mark_cnt = io_sectors;
5745                 if (last_check == 0)
5746                         /* this is the earliers that rebuilt will be
5747                          * visible in /proc/mdstat
5748                          */
5749                         md_new_event(mddev);
5750
5751                 if (last_check + window > io_sectors || j == max_sectors)
5752                         continue;
5753
5754                 last_check = io_sectors;
5755
5756                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5757                         break;
5758
5759         repeat:
5760                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5761                         /* step marks */
5762                         int next = (last_mark+1) % SYNC_MARKS;
5763
5764                         mddev->resync_mark = mark[next];
5765                         mddev->resync_mark_cnt = mark_cnt[next];
5766                         mark[next] = jiffies;
5767                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5768                         last_mark = next;
5769                 }
5770
5771
5772                 if (kthread_should_stop())
5773                         goto interrupted;
5774
5775
5776                 /*
5777                  * this loop exits only if either when we are slower than
5778                  * the 'hard' speed limit, or the system was IO-idle for
5779                  * a jiffy.
5780                  * the system might be non-idle CPU-wise, but we only care
5781                  * about not overloading the IO subsystem. (things like an
5782                  * e2fsck being done on the RAID array should execute fast)
5783                  */
5784                 blk_unplug(mddev->queue);
5785                 cond_resched();
5786
5787                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5788                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5789
5790                 if (currspeed > speed_min(mddev)) {
5791                         if ((currspeed > speed_max(mddev)) ||
5792                                         !is_mddev_idle(mddev)) {
5793                                 msleep(500);
5794                                 goto repeat;
5795                         }
5796                 }
5797         }
5798         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5799         /*
5800          * this also signals 'finished resyncing' to md_stop
5801          */
5802  out:
5803         blk_unplug(mddev->queue);
5804
5805         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5806
5807         /* tell personality that we are finished */
5808         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5809
5810         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5811             mddev->curr_resync > 2) {
5812                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5813                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5814                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5815                                         printk(KERN_INFO
5816                                                "md: checkpointing %s of %s.\n",
5817                                                desc, mdname(mddev));
5818                                         mddev->recovery_cp = mddev->curr_resync;
5819                                 }
5820                         } else
5821                                 mddev->recovery_cp = MaxSector;
5822                 } else {
5823                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5824                                 mddev->curr_resync = MaxSector;
5825                         rdev_for_each(rdev, rtmp, mddev)
5826                                 if (rdev->raid_disk >= 0 &&
5827                                     !test_bit(Faulty, &rdev->flags) &&
5828                                     !test_bit(In_sync, &rdev->flags) &&
5829                                     rdev->recovery_offset < mddev->curr_resync)
5830                                         rdev->recovery_offset = mddev->curr_resync;
5831                 }
5832         }
5833         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5834
5835  skip:
5836         mddev->curr_resync = 0;
5837         mddev->resync_min = 0;
5838         mddev->resync_max = MaxSector;
5839         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5840         wake_up(&resync_wait);
5841         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5842         md_wakeup_thread(mddev->thread);
5843         return;
5844
5845  interrupted:
5846         /*
5847          * got a signal, exit.
5848          */
5849         printk(KERN_INFO
5850                "md: md_do_sync() got signal ... exiting\n");
5851         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5852         goto out;
5853
5854 }
5855 EXPORT_SYMBOL_GPL(md_do_sync);
5856
5857
5858 static int remove_and_add_spares(mddev_t *mddev)
5859 {
5860         mdk_rdev_t *rdev;
5861         struct list_head *rtmp;
5862         int spares = 0;
5863
5864         rdev_for_each(rdev, rtmp, mddev)
5865                 if (rdev->raid_disk >= 0 &&
5866                     !test_bit(Blocked, &rdev->flags) &&
5867                     (test_bit(Faulty, &rdev->flags) ||
5868                      ! test_bit(In_sync, &rdev->flags)) &&
5869                     atomic_read(&rdev->nr_pending)==0) {
5870                         if (mddev->pers->hot_remove_disk(
5871                                     mddev, rdev->raid_disk)==0) {
5872                                 char nm[20];
5873                                 sprintf(nm,"rd%d", rdev->raid_disk);
5874                                 sysfs_remove_link(&mddev->kobj, nm);
5875                                 rdev->raid_disk = -1;
5876                         }
5877                 }
5878
5879         if (mddev->degraded) {
5880                 rdev_for_each(rdev, rtmp, mddev) {
5881                         if (rdev->raid_disk >= 0 &&
5882                             !test_bit(In_sync, &rdev->flags))
5883                                 spares++;
5884                         if (rdev->raid_disk < 0
5885                             && !test_bit(Faulty, &rdev->flags)) {
5886                                 rdev->recovery_offset = 0;
5887                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5888                                         char nm[20];
5889                                         sprintf(nm, "rd%d", rdev->raid_disk);
5890                                         if (sysfs_create_link(&mddev->kobj,
5891                                                               &rdev->kobj, nm))
5892                                                 printk(KERN_WARNING
5893                                                        "md: cannot register "
5894                                                        "%s for %s\n",
5895                                                        nm, mdname(mddev));
5896                                         spares++;
5897                                         md_new_event(mddev);
5898                                 } else
5899                                         break;
5900                         }
5901                 }
5902         }
5903         return spares;
5904 }
5905 /*
5906  * This routine is regularly called by all per-raid-array threads to
5907  * deal with generic issues like resync and super-block update.
5908  * Raid personalities that don't have a thread (linear/raid0) do not
5909  * need this as they never do any recovery or update the superblock.
5910  *
5911  * It does not do any resync itself, but rather "forks" off other threads
5912  * to do that as needed.
5913  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5914  * "->recovery" and create a thread at ->sync_thread.
5915  * When the thread finishes it sets MD_RECOVERY_DONE
5916  * and wakeups up this thread which will reap the thread and finish up.
5917  * This thread also removes any faulty devices (with nr_pending == 0).
5918  *
5919  * The overall approach is:
5920  *  1/ if the superblock needs updating, update it.
5921  *  2/ If a recovery thread is running, don't do anything else.
5922  *  3/ If recovery has finished, clean up, possibly marking spares active.
5923  *  4/ If there are any faulty devices, remove them.
5924  *  5/ If array is degraded, try to add spares devices
5925  *  6/ If array has spares or is not in-sync, start a resync thread.
5926  */
5927 void md_check_recovery(mddev_t *mddev)
5928 {
5929         mdk_rdev_t *rdev;
5930         struct list_head *rtmp;
5931
5932
5933         if (mddev->bitmap)
5934                 bitmap_daemon_work(mddev->bitmap);
5935
5936         if (mddev->ro)
5937                 return;
5938
5939         if (signal_pending(current)) {
5940                 if (mddev->pers->sync_request && !mddev->external) {
5941                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5942                                mdname(mddev));
5943                         mddev->safemode = 2;
5944                 }
5945                 flush_signals(current);
5946         }
5947
5948         if ( ! (
5949                 (mddev->flags && !mddev->external) ||
5950                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5951                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5952                 (mddev->external == 0 && mddev->safemode == 1) ||
5953                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5954                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5955                 ))
5956                 return;
5957
5958         if (mddev_trylock(mddev)) {
5959                 int spares = 0;
5960
5961                 if (!mddev->external) {
5962                         spin_lock_irq(&mddev->write_lock);
5963                         if (mddev->safemode &&
5964                             !atomic_read(&mddev->writes_pending) &&
5965                             !mddev->in_sync &&
5966                             mddev->recovery_cp == MaxSector) {
5967                                 mddev->in_sync = 1;
5968                                 if (mddev->persistent)
5969                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5970                         }
5971                         if (mddev->safemode == 1)
5972                                 mddev->safemode = 0;
5973                         spin_unlock_irq(&mddev->write_lock);
5974                 }
5975
5976                 if (mddev->flags)
5977                         md_update_sb(mddev, 0);
5978
5979
5980                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5981                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5982                         /* resync/recovery still happening */
5983                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5984                         goto unlock;
5985                 }
5986                 if (mddev->sync_thread) {
5987                         /* resync has finished, collect result */
5988                         md_unregister_thread(mddev->sync_thread);
5989                         mddev->sync_thread = NULL;
5990                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5991                                 /* success...*/
5992                                 /* activate any spares */
5993                                 mddev->pers->spare_active(mddev);
5994                         }
5995                         md_update_sb(mddev, 1);
5996
5997                         /* if array is no-longer degraded, then any saved_raid_disk
5998                          * information must be scrapped
5999                          */
6000                         if (!mddev->degraded)
6001                                 rdev_for_each(rdev, rtmp, mddev)
6002                                         rdev->saved_raid_disk = -1;
6003
6004                         mddev->recovery = 0;
6005                         /* flag recovery needed just to double check */
6006                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6007                         md_new_event(mddev);
6008                         goto unlock;
6009                 }
6010                 /* Clear some bits that don't mean anything, but
6011                  * might be left set
6012                  */
6013                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6014                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6015                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6016
6017                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6018                         goto unlock;
6019                 /* no recovery is running.
6020                  * remove any failed drives, then
6021                  * add spares if possible.
6022                  * Spare are also removed and re-added, to allow
6023                  * the personality to fail the re-add.
6024                  */
6025
6026                 if (mddev->reshape_position != MaxSector) {
6027                         if (mddev->pers->check_reshape(mddev) != 0)
6028                                 /* Cannot proceed */
6029                                 goto unlock;
6030                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6031                 } else if ((spares = remove_and_add_spares(mddev))) {
6032                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6033                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6034                 } else if (mddev->recovery_cp < MaxSector) {
6035                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6036                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6037                         /* nothing to be done ... */
6038                         goto unlock;
6039
6040                 if (mddev->pers->sync_request) {
6041                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6042                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6043                                 /* We are adding a device or devices to an array
6044                                  * which has the bitmap stored on all devices.
6045                                  * So make sure all bitmap pages get written
6046                                  */
6047                                 bitmap_write_all(mddev->bitmap);
6048                         }
6049                         mddev->sync_thread = md_register_thread(md_do_sync,
6050                                                                 mddev,
6051                                                                 "%s_resync");
6052                         if (!mddev->sync_thread) {
6053                                 printk(KERN_ERR "%s: could not start resync"
6054                                         " thread...\n",
6055                                         mdname(mddev));
6056                                 /* leave the spares where they are, it shouldn't hurt */
6057                                 mddev->recovery = 0;
6058                         } else
6059                                 md_wakeup_thread(mddev->sync_thread);
6060                         md_new_event(mddev);
6061                 }
6062         unlock:
6063                 mddev_unlock(mddev);
6064         }
6065 }
6066
6067 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6068 {
6069         sysfs_notify(&rdev->kobj, NULL, "state");
6070         wait_event_timeout(rdev->blocked_wait,
6071                            !test_bit(Blocked, &rdev->flags),
6072                            msecs_to_jiffies(5000));
6073         rdev_dec_pending(rdev, mddev);
6074 }
6075 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6076
6077 static int md_notify_reboot(struct notifier_block *this,
6078                             unsigned long code, void *x)
6079 {
6080         struct list_head *tmp;
6081         mddev_t *mddev;
6082
6083         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6084
6085                 printk(KERN_INFO "md: stopping all md devices.\n");
6086
6087                 for_each_mddev(mddev, tmp)
6088                         if (mddev_trylock(mddev)) {
6089                                 do_md_stop (mddev, 1);
6090                                 mddev_unlock(mddev);
6091                         }
6092                 /*
6093                  * certain more exotic SCSI devices are known to be
6094                  * volatile wrt too early system reboots. While the
6095                  * right place to handle this issue is the given
6096                  * driver, we do want to have a safe RAID driver ...
6097                  */
6098                 mdelay(1000*1);
6099         }
6100         return NOTIFY_DONE;
6101 }
6102
6103 static struct notifier_block md_notifier = {
6104         .notifier_call  = md_notify_reboot,
6105         .next           = NULL,
6106         .priority       = INT_MAX, /* before any real devices */
6107 };
6108
6109 static void md_geninit(void)
6110 {
6111         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6112
6113         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6114 }
6115
6116 static int __init md_init(void)
6117 {
6118         if (register_blkdev(MAJOR_NR, "md"))
6119                 return -1;
6120         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6121                 unregister_blkdev(MAJOR_NR, "md");
6122                 return -1;
6123         }
6124         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6125                             md_probe, NULL, NULL);
6126         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6127                             md_probe, NULL, NULL);
6128
6129         register_reboot_notifier(&md_notifier);
6130         raid_table_header = register_sysctl_table(raid_root_table);
6131
6132         md_geninit();
6133         return (0);
6134 }
6135
6136
6137 #ifndef MODULE
6138
6139 /*
6140  * Searches all registered partitions for autorun RAID arrays
6141  * at boot time.
6142  */
6143
6144 static LIST_HEAD(all_detected_devices);
6145 struct detected_devices_node {
6146         struct list_head list;
6147         dev_t dev;
6148 };
6149
6150 void md_autodetect_dev(dev_t dev)
6151 {
6152         struct detected_devices_node *node_detected_dev;
6153
6154         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6155         if (node_detected_dev) {
6156                 node_detected_dev->dev = dev;
6157                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6158         } else {
6159                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6160                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6161         }
6162 }
6163
6164
6165 static void autostart_arrays(int part)
6166 {
6167         mdk_rdev_t *rdev;
6168         struct detected_devices_node *node_detected_dev;
6169         dev_t dev;
6170         int i_scanned, i_passed;
6171
6172         i_scanned = 0;
6173         i_passed = 0;
6174
6175         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6176
6177         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6178                 i_scanned++;
6179                 node_detected_dev = list_entry(all_detected_devices.next,
6180                                         struct detected_devices_node, list);
6181                 list_del(&node_detected_dev->list);
6182                 dev = node_detected_dev->dev;
6183                 kfree(node_detected_dev);
6184                 rdev = md_import_device(dev,0, 90);
6185                 if (IS_ERR(rdev))
6186                         continue;
6187
6188                 if (test_bit(Faulty, &rdev->flags)) {
6189                         MD_BUG();
6190                         continue;
6191                 }
6192                 set_bit(AutoDetected, &rdev->flags);
6193                 list_add(&rdev->same_set, &pending_raid_disks);
6194                 i_passed++;
6195         }
6196
6197         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6198                                                 i_scanned, i_passed);
6199
6200         autorun_devices(part);
6201 }
6202
6203 #endif /* !MODULE */
6204
6205 static __exit void md_exit(void)
6206 {
6207         mddev_t *mddev;
6208         struct list_head *tmp;
6209
6210         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6211         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6212
6213         unregister_blkdev(MAJOR_NR,"md");
6214         unregister_blkdev(mdp_major, "mdp");
6215         unregister_reboot_notifier(&md_notifier);
6216         unregister_sysctl_table(raid_table_header);
6217         remove_proc_entry("mdstat", NULL);
6218         for_each_mddev(mddev, tmp) {
6219                 struct gendisk *disk = mddev->gendisk;
6220                 if (!disk)
6221                         continue;
6222                 export_array(mddev);
6223                 del_gendisk(disk);
6224                 put_disk(disk);
6225                 mddev->gendisk = NULL;
6226                 mddev_put(mddev);
6227         }
6228 }
6229
6230 subsys_initcall(md_init);
6231 module_exit(md_exit)
6232
6233 static int get_ro(char *buffer, struct kernel_param *kp)
6234 {
6235         return sprintf(buffer, "%d", start_readonly);
6236 }
6237 static int set_ro(const char *val, struct kernel_param *kp)
6238 {
6239         char *e;
6240         int num = simple_strtoul(val, &e, 10);
6241         if (*val && (*e == '\0' || *e == '\n')) {
6242                 start_readonly = num;
6243                 return 0;
6244         }
6245         return -EINVAL;
6246 }
6247
6248 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6249 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6250
6251
6252 EXPORT_SYMBOL(register_md_personality);
6253 EXPORT_SYMBOL(unregister_md_personality);
6254 EXPORT_SYMBOL(md_error);
6255 EXPORT_SYMBOL(md_done_sync);
6256 EXPORT_SYMBOL(md_write_start);
6257 EXPORT_SYMBOL(md_write_end);
6258 EXPORT_SYMBOL(md_register_thread);
6259 EXPORT_SYMBOL(md_unregister_thread);
6260 EXPORT_SYMBOL(md_wakeup_thread);
6261 EXPORT_SYMBOL(md_check_recovery);
6262 MODULE_LICENSE("GPL");
6263 MODULE_ALIAS("md");
6264 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);