drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/kernel.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/buffer_head.h> /* for invalidate_bdev */
  43 #include <linux/poll.h>
  44 #include <linux/mutex.h>
  45 #include <linux/ctype.h>
  46 #include <linux/freezer.h>
  47
  48 #include <linux/init.h>
  49
  50 #include <linux/file.h>
  51
  52 #ifdef CONFIG_KMOD
  53 #include <linux/kmod.h>
  54 #endif
  55
  56 #include <asm/unaligned.h>
  57
  58 #define MAJOR_NR MD_MAJOR
  59 #define MD_DRIVER
  60
  61 /* 63 partitions with the alternate major number (mdp) */
  62 #define MdpMinorShift 6
  63
  64 #define DEBUG 0
  65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  66
  67
  68 #ifndef MODULE
  69 static void autostart_arrays (int part);
  70 #endif
  71
  72 static LIST_HEAD(pers_list);
  73 static DEFINE_SPINLOCK(pers_lock);
  74
  75 static void md_print_devices(void);
  76
  77 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  78
  79 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  80
  81 /*
  82  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  83  * is 1000 KB/sec, so the extra system load does not show up that much.
  84  * Increase it if you want to have more _guaranteed_ speed. Note that
  85  * the RAID driver will use the maximum available bandwidth if the IO
  86  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  87  * speed limit - in case reconstruction slows down your system despite
  88  * idle IO detection.
  89  *
  90  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  91  * or /sys/block/mdX/md/sync_speed_{min,max}
  92  */
  93
  94 static int sysctl_speed_limit_min = 1000;
  95 static int sysctl_speed_limit_max = 200000;
  96 static inline int speed_min(mddev_t *mddev)
  97 {
  98         return mddev->sync_speed_min ?
  99                 mddev->sync_speed_min : sysctl_speed_limit_min;
 100 }
 101
 102 static inline int speed_max(mddev_t *mddev)
 103 {
 104         return mddev->sync_speed_max ?
 105                 mddev->sync_speed_max : sysctl_speed_limit_max;
 106 }
 107
 108 static struct ctl_table_header *raid_table_header;
 109
 110 static ctl_table raid_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 113                 .procname       = "speed_limit_min",
 114                 .data           = &sysctl_speed_limit_min,
 115                 .maxlen         = sizeof(int),
 116                 .mode           = S_IRUGO|S_IWUSR,
 117                 .proc_handler   = &proc_dointvec,
 118         },
 119         {
 120                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 121                 .procname       = "speed_limit_max",
 122                 .data           = &sysctl_speed_limit_max,
 123                 .maxlen         = sizeof(int),
 124                 .mode           = S_IRUGO|S_IWUSR,
 125                 .proc_handler   = &proc_dointvec,
 126         },
 127         { .ctl_name = 0 }
 128 };
 129
 130 static ctl_table raid_dir_table[] = {
 131         {
 132                 .ctl_name       = DEV_RAID,
 133                 .procname       = "raid",
 134                 .maxlen         = 0,
 135                 .mode           = S_IRUGO|S_IXUGO,
 136                 .child          = raid_table,
 137         },
 138         { .ctl_name = 0 }
 139 };
 140
 141 static ctl_table raid_root_table[] = {
 142         {
 143                 .ctl_name       = CTL_DEV,
 144                 .procname       = "dev",
 145                 .maxlen         = 0,
 146                 .mode           = 0555,
 147                 .child          = raid_dir_table,
 148         },
 149         { .ctl_name = 0 }
 150 };
 151
 152 static struct block_device_operations md_fops;
 153
 154 static int start_readonly;
 155
 156 /*
 157  * We have a system wide 'event count' that is incremented
 158  * on any 'interesting' event, and readers of /proc/mdstat
 159  * can use 'poll' or 'select' to find out when the event
 160  * count increases.
 161  *
 162  * Events are:
 163  *  start array, stop array, error, add device, remove device,
 164  *  start build, activate spare
 165  */
 166 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 167 static atomic_t md_event_count;
 168 void md_new_event(mddev_t *mddev)
 169 {
 170         atomic_inc(&md_event_count);
 171         wake_up(&md_event_waiters);
 172         sysfs_notify(&mddev->kobj, NULL, "sync_action");
 173 }
 174 EXPORT_SYMBOL_GPL(md_new_event);
 175
 176 /* Alternate version that can be called from interrupts
 177  * when calling sysfs_notify isn't needed.
 178  */
 179 static void md_new_event_inintr(mddev_t *mddev)
 180 {
 181         atomic_inc(&md_event_count);
 182         wake_up(&md_event_waiters);
 183 }
 184
 185 /*
 186  * Enables to iterate over all existing md arrays
 187  * all_mddevs_lock protects this list.
 188  */
 189 static LIST_HEAD(all_mddevs);
 190 static DEFINE_SPINLOCK(all_mddevs_lock);
 191
 192
 193 /*
 194  * iterates through all used mddevs in the system.
 195  * We take care to grab the all_mddevs_lock whenever navigating
 196  * the list, and to always hold a refcount when unlocked.
 197  * Any code which breaks out of this loop while own
 198  * a reference to the current mddev and must mddev_put it.
 199  */
 200 #define for_each_mddev(mddev,tmp)                                       \
 201                                                                         \
 202         for (({ spin_lock(&all_mddevs_lock);                            \
 203                 tmp = all_mddevs.next;                                  \
 204                 mddev = NULL;});                                        \
 205              ({ if (tmp != &all_mddevs)                                 \
 206                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 207                 spin_unlock(&all_mddevs_lock);                          \
 208                 if (mddev) mddev_put(mddev);                            \
 209                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 210                 tmp != &all_mddevs;});                                  \
 211              ({ spin_lock(&all_mddevs_lock);                            \
 212                 tmp = tmp->next;})                                      \
 213                 )
 214
 215
 216 static int md_fail_request (struct request_queue *q, struct bio *bio)
 217 {
 218         bio_io_error(bio);
 219         return 0;
 220 }
 221
 222 static inline mddev_t *mddev_get(mddev_t *mddev)
 223 {
 224         atomic_inc(&mddev->active);
 225         return mddev;
 226 }
 227
 228 static void mddev_put(mddev_t *mddev)
 229 {
 230         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 231                 return;
 232         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 233                 list_del(&mddev->all_mddevs);
 234                 spin_unlock(&all_mddevs_lock);
 235                 blk_cleanup_queue(mddev->queue);
 236                 kobject_put(&mddev->kobj);
 237         } else
 238                 spin_unlock(&all_mddevs_lock);
 239 }
 240
 241 static mddev_t * mddev_find(dev_t unit)
 242 {
 243         mddev_t *mddev, *new = NULL;
 244
 245  retry:
 246         spin_lock(&all_mddevs_lock);
 247         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 248                 if (mddev->unit == unit) {
 249                         mddev_get(mddev);
 250                         spin_unlock(&all_mddevs_lock);
 251                         kfree(new);
 252                         return mddev;
 253                 }
 254
 255         if (new) {
 256                 list_add(&new->all_mddevs, &all_mddevs);
 257                 spin_unlock(&all_mddevs_lock);
 258                 return new;
 259         }
 260         spin_unlock(&all_mddevs_lock);
 261
 262         new = kzalloc(sizeof(*new), GFP_KERNEL);
 263         if (!new)
 264                 return NULL;
 265
 266         new->unit = unit;
 267         if (MAJOR(unit) == MD_MAJOR)
 268                 new->md_minor = MINOR(unit);
 269         else
 270                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 271
 272         mutex_init(&new->reconfig_mutex);
 273         INIT_LIST_HEAD(&new->disks);
 274         INIT_LIST_HEAD(&new->all_mddevs);
 275         init_timer(&new->safemode_timer);
 276         atomic_set(&new->active, 1);
 277         spin_lock_init(&new->write_lock);
 278         init_waitqueue_head(&new->sb_wait);
 279         init_waitqueue_head(&new->recovery_wait);
 280         new->reshape_position = MaxSector;
 281         new->resync_min = 0;
 282         new->resync_max = MaxSector;
 283         new->level = LEVEL_NONE;
 284
 285         new->queue = blk_alloc_queue(GFP_KERNEL);
 286         if (!new->queue) {
 287                 kfree(new);
 288                 return NULL;
 289         }
 290         /* Can be unlocked because the queue is new: no concurrency */
 291         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
 292
 293         blk_queue_make_request(new->queue, md_fail_request);
 294
 295         goto retry;
 296 }
 297
 298 static inline int mddev_lock(mddev_t * mddev)
 299 {
 300         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 301 }
 302
 303 static inline int mddev_trylock(mddev_t * mddev)
 304 {
 305         return mutex_trylock(&mddev->reconfig_mutex);
 306 }
 307
 308 static inline void mddev_unlock(mddev_t * mddev)
 309 {
 310         mutex_unlock(&mddev->reconfig_mutex);
 311
 312         md_wakeup_thread(mddev->thread);
 313 }
 314
 315 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 316 {
 317         mdk_rdev_t * rdev;
 318         struct list_head *tmp;
 319
 320         rdev_for_each(rdev, tmp, mddev) {
 321                 if (rdev->desc_nr == nr)
 322                         return rdev;
 323         }
 324         return NULL;
 325 }
 326
 327 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 328 {
 329         struct list_head *tmp;
 330         mdk_rdev_t *rdev;
 331
 332         rdev_for_each(rdev, tmp, mddev) {
 333                 if (rdev->bdev->bd_dev == dev)
 334                         return rdev;
 335         }
 336         return NULL;
 337 }
 338
 339 static struct mdk_personality *find_pers(int level, char *clevel)
 340 {
 341         struct mdk_personality *pers;
 342         list_for_each_entry(pers, &pers_list, list) {
 343                 if (level != LEVEL_NONE && pers->level == level)
 344                         return pers;
 345                 if (strcmp(pers->name, clevel)==0)
 346                         return pers;
 347         }
 348         return NULL;
 349 }
 350
 351 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 352 {
 353         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 354         return MD_NEW_SIZE_BLOCKS(size);
 355 }
 356
 357 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 358 {
 359         sector_t size;
 360
 361         size = rdev->sb_offset;
 362
 363         if (chunk_size)
 364                 size &= ~((sector_t)chunk_size/1024 - 1);
 365         return size;
 366 }
 367
 368 static int alloc_disk_sb(mdk_rdev_t * rdev)
 369 {
 370         if (rdev->sb_page)
 371                 MD_BUG();
 372
 373         rdev->sb_page = alloc_page(GFP_KERNEL);
 374         if (!rdev->sb_page) {
 375                 printk(KERN_ALERT "md: out of memory.\n");
 376                 return -EINVAL;
 377         }
 378
 379         return 0;
 380 }
 381
 382 static void free_disk_sb(mdk_rdev_t * rdev)
 383 {
 384         if (rdev->sb_page) {
 385                 put_page(rdev->sb_page);
 386                 rdev->sb_loaded = 0;
 387                 rdev->sb_page = NULL;
 388                 rdev->sb_offset = 0;
 389                 rdev->size = 0;
 390         }
 391 }
 392
 393
 394 static void super_written(struct bio *bio, int error)
 395 {
 396         mdk_rdev_t *rdev = bio->bi_private;
 397         mddev_t *mddev = rdev->mddev;
 398
 399         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 400                 printk("md: super_written gets error=%d, uptodate=%d\n",
 401                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 402                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 403                 md_error(mddev, rdev);
 404         }
 405
 406         if (atomic_dec_and_test(&mddev->pending_writes))
 407                 wake_up(&mddev->sb_wait);
 408         bio_put(bio);
 409 }
 410
 411 static void super_written_barrier(struct bio *bio, int error)
 412 {
 413         struct bio *bio2 = bio->bi_private;
 414         mdk_rdev_t *rdev = bio2->bi_private;
 415         mddev_t *mddev = rdev->mddev;
 416
 417         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 418             error == -EOPNOTSUPP) {
 419                 unsigned long flags;
 420                 /* barriers don't appear to be supported :-( */
 421                 set_bit(BarriersNotsupp, &rdev->flags);
 422                 mddev->barriers_work = 0;
 423                 spin_lock_irqsave(&mddev->write_lock, flags);
 424                 bio2->bi_next = mddev->biolist;
 425                 mddev->biolist = bio2;
 426                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 427                 wake_up(&mddev->sb_wait);
 428                 bio_put(bio);
 429         } else {
 430                 bio_put(bio2);
 431                 bio->bi_private = rdev;
 432                 super_written(bio, error);
 433         }
 434 }
 435
 436 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 437                    sector_t sector, int size, struct page *page)
 438 {
 439         /* write first size bytes of page to sector of rdev
 440          * Increment mddev->pending_writes before returning
 441          * and decrement it on completion, waking up sb_wait
 442          * if zero is reached.
 443          * If an error occurred, call md_error
 444          *
 445          * As we might need to resubmit the request if BIO_RW_BARRIER
 446          * causes ENOTSUPP, we allocate a spare bio...
 447          */
 448         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 449         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 450
 451         bio->bi_bdev = rdev->bdev;
 452         bio->bi_sector = sector;
 453         bio_add_page(bio, page, size, 0);
 454         bio->bi_private = rdev;
 455         bio->bi_end_io = super_written;
 456         bio->bi_rw = rw;
 457
 458         atomic_inc(&mddev->pending_writes);
 459         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 460                 struct bio *rbio;
 461                 rw |= (1<<BIO_RW_BARRIER);
 462                 rbio = bio_clone(bio, GFP_NOIO);
 463                 rbio->bi_private = bio;
 464                 rbio->bi_end_io = super_written_barrier;
 465                 submit_bio(rw, rbio);
 466         } else
 467                 submit_bio(rw, bio);
 468 }
 469
 470 void md_super_wait(mddev_t *mddev)
 471 {
 472         /* wait for all superblock writes that were scheduled to complete.
 473          * if any had to be retried (due to BARRIER problems), retry them
 474          */
 475         DEFINE_WAIT(wq);
 476         for(;;) {
 477                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 478                 if (atomic_read(&mddev->pending_writes)==0)
 479                         break;
 480                 while (mddev->biolist) {
 481                         struct bio *bio;
 482                         spin_lock_irq(&mddev->write_lock);
 483                         bio = mddev->biolist;
 484                         mddev->biolist = bio->bi_next ;
 485                         bio->bi_next = NULL;
 486                         spin_unlock_irq(&mddev->write_lock);
 487                         submit_bio(bio->bi_rw, bio);
 488                 }
 489                 schedule();
 490         }
 491         finish_wait(&mddev->sb_wait, &wq);
 492 }
 493
 494 static void bi_complete(struct bio *bio, int error)
 495 {
 496         complete((struct completion*)bio->bi_private);
 497 }
 498
 499 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 500                    struct page *page, int rw)
 501 {
 502         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 503         struct completion event;
 504         int ret;
 505
 506         rw |= (1 << BIO_RW_SYNC);
 507
 508         bio->bi_bdev = bdev;
 509         bio->bi_sector = sector;
 510         bio_add_page(bio, page, size, 0);
 511         init_completion(&event);
 512         bio->bi_private = &event;
 513         bio->bi_end_io = bi_complete;
 514         submit_bio(rw, bio);
 515         wait_for_completion(&event);
 516
 517         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 518         bio_put(bio);
 519         return ret;
 520 }
 521 EXPORT_SYMBOL_GPL(sync_page_io);
 522
 523 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 524 {
 525         char b[BDEVNAME_SIZE];
 526         if (!rdev->sb_page) {
 527                 MD_BUG();
 528                 return -EINVAL;
 529         }
 530         if (rdev->sb_loaded)
 531                 return 0;
 532
 533
 534         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 535                 goto fail;
 536         rdev->sb_loaded = 1;
 537         return 0;
 538
 539 fail:
 540         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 541                 bdevname(rdev->bdev,b));
 542         return -EINVAL;
 543 }
 544
 545 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 546 {
 547         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 548                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 549                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 550                 (sb1->set_uuid3 == sb2->set_uuid3))
 551
 552                 return 1;
 553
 554         return 0;
 555 }
 556
 557
 558 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 559 {
 560         int ret;
 561         mdp_super_t *tmp1, *tmp2;
 562
 563         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 564         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 565
 566         if (!tmp1 || !tmp2) {
 567                 ret = 0;
 568                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 569                 goto abort;
 570         }
 571
 572         *tmp1 = *sb1;
 573         *tmp2 = *sb2;
 574
 575         /*
 576          * nr_disks is not constant
 577          */
 578         tmp1->nr_disks = 0;
 579         tmp2->nr_disks = 0;
 580
 581         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 582                 ret = 0;
 583         else
 584                 ret = 1;
 585
 586 abort:
 587         kfree(tmp1);
 588         kfree(tmp2);
 589         return ret;
 590 }
 591
 592
 593 static u32 md_csum_fold(u32 csum)
 594 {
 595         csum = (csum & 0xffff) + (csum >> 16);
 596         return (csum & 0xffff) + (csum >> 16);
 597 }
 598
 599 static unsigned int calc_sb_csum(mdp_super_t * sb)
 600 {
 601         u64 newcsum = 0;
 602         u32 *sb32 = (u32*)sb;
 603         int i;
 604         unsigned int disk_csum, csum;
 605
 606         disk_csum = sb->sb_csum;
 607         sb->sb_csum = 0;
 608
 609         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 610                 newcsum += sb32[i];
 611         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 612
 613
 614 #ifdef CONFIG_ALPHA
 615         /* This used to use csum_partial, which was wrong for several
 616          * reasons including that different results are returned on
 617          * different architectures.  It isn't critical that we get exactly
 618          * the same return value as before (we always csum_fold before
 619          * testing, and that removes any differences).  However as we
 620          * know that csum_partial always returned a 16bit value on
 621          * alphas, do a fold to maximise conformity to previous behaviour.
 622          */
 623         sb->sb_csum = md_csum_fold(disk_csum);
 624 #else
 625         sb->sb_csum = disk_csum;
 626 #endif
 627         return csum;
 628 }
 629
 630
 631 /*
 632  * Handle superblock details.
 633  * We want to be able to handle multiple superblock formats
 634  * so we have a common interface to them all, and an array of
 635  * different handlers.
 636  * We rely on user-space to write the initial superblock, and support
 637  * reading and updating of superblocks.
 638  * Interface methods are:
 639  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 640  *      loads and validates a superblock on dev.
 641  *      if refdev != NULL, compare superblocks on both devices
 642  *    Return:
 643  *      0 - dev has a superblock that is compatible with refdev
 644  *      1 - dev has a superblock that is compatible and newer than refdev
 645  *          so dev should be used as the refdev in future
 646  *     -EINVAL superblock incompatible or invalid
 647  *     -othererror e.g. -EIO
 648  *
 649  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 650  *      Verify that dev is acceptable into mddev.
 651  *       The first time, mddev->raid_disks will be 0, and data from
 652  *       dev should be merged in.  Subsequent calls check that dev
 653  *       is new enough.  Return 0 or -EINVAL
 654  *
 655  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 656  *     Update the superblock for rdev with data in mddev
 657  *     This does not write to disc.
 658  *
 659  */
 660
 661 struct super_type  {
 662         char            *name;
 663         struct module   *owner;
 664         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 665         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 666         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 667 };
 668
 669 /*
 670  * load_super for 0.90.0
 671  */
 672 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 673 {
 674         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 675         mdp_super_t *sb;
 676         int ret;
 677         sector_t sb_offset;
 678
 679         /*
 680          * Calculate the position of the superblock,
 681          * it's at the end of the disk.
 682          *
 683          * It also happens to be a multiple of 4Kb.
 684          */
 685         sb_offset = calc_dev_sboffset(rdev->bdev);
 686         rdev->sb_offset = sb_offset;
 687
 688         ret = read_disk_sb(rdev, MD_SB_BYTES);
 689         if (ret) return ret;
 690
 691         ret = -EINVAL;
 692
 693         bdevname(rdev->bdev, b);
 694         sb = (mdp_super_t*)page_address(rdev->sb_page);
 695
 696         if (sb->md_magic != MD_SB_MAGIC) {
 697                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 698                        b);
 699                 goto abort;
 700         }
 701
 702         if (sb->major_version != 0 ||
 703             sb->minor_version < 90 ||
 704             sb->minor_version > 91) {
 705                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 706                         sb->major_version, sb->minor_version,
 707                         b);
 708                 goto abort;
 709         }
 710
 711         if (sb->raid_disks <= 0)
 712                 goto abort;
 713
 714         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 715                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 716                         b);
 717                 goto abort;
 718         }
 719
 720         rdev->preferred_minor = sb->md_minor;
 721         rdev->data_offset = 0;
 722         rdev->sb_size = MD_SB_BYTES;
 723
 724         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 725                 if (sb->level != 1 && sb->level != 4
 726                     && sb->level != 5 && sb->level != 6
 727                     && sb->level != 10) {
 728                         /* FIXME use a better test */
 729                         printk(KERN_WARNING
 730                                "md: bitmaps not supported for this level.\n");
 731                         goto abort;
 732                 }
 733         }
 734
 735         if (sb->level == LEVEL_MULTIPATH)
 736                 rdev->desc_nr = -1;
 737         else
 738                 rdev->desc_nr = sb->this_disk.number;
 739
 740         if (!refdev) {
 741                 ret = 1;
 742         } else {
 743                 __u64 ev1, ev2;
 744                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 745                 if (!uuid_equal(refsb, sb)) {
 746                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 747                                 b, bdevname(refdev->bdev,b2));
 748                         goto abort;
 749                 }
 750                 if (!sb_equal(refsb, sb)) {
 751                         printk(KERN_WARNING "md: %s has same UUID"
 752                                " but different superblock to %s\n",
 753                                b, bdevname(refdev->bdev, b2));
 754                         goto abort;
 755                 }
 756                 ev1 = md_event(sb);
 757                 ev2 = md_event(refsb);
 758                 if (ev1 > ev2)
 759                         ret = 1;
 760                 else
 761                         ret = 0;
 762         }
 763         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 764
 765         if (rdev->size < sb->size && sb->level > 1)
 766                 /* "this cannot possibly happen" ... */
 767                 ret = -EINVAL;
 768
 769  abort:
 770         return ret;
 771 }
 772
 773 /*
 774  * validate_super for 0.90.0
 775  */
 776 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 777 {
 778         mdp_disk_t *desc;
 779         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 780         __u64 ev1 = md_event(sb);
 781
 782         rdev->raid_disk = -1;
 783         clear_bit(Faulty, &rdev->flags);
 784         clear_bit(In_sync, &rdev->flags);
 785         clear_bit(WriteMostly, &rdev->flags);
 786         clear_bit(BarriersNotsupp, &rdev->flags);
 787
 788         if (mddev->raid_disks == 0) {
 789                 mddev->major_version = 0;
 790                 mddev->minor_version = sb->minor_version;
 791                 mddev->patch_version = sb->patch_version;
 792                 mddev->external = 0;
 793                 mddev->chunk_size = sb->chunk_size;
 794                 mddev->ctime = sb->ctime;
 795                 mddev->utime = sb->utime;
 796                 mddev->level = sb->level;
 797                 mddev->clevel[0] = 0;
 798                 mddev->layout = sb->layout;
 799                 mddev->raid_disks = sb->raid_disks;
 800                 mddev->size = sb->size;
 801                 mddev->events = ev1;
 802                 mddev->bitmap_offset = 0;
 803                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 804
 805                 if (mddev->minor_version >= 91) {
 806                         mddev->reshape_position = sb->reshape_position;
 807                         mddev->delta_disks = sb->delta_disks;
 808                         mddev->new_level = sb->new_level;
 809                         mddev->new_layout = sb->new_layout;
 810                         mddev->new_chunk = sb->new_chunk;
 811                 } else {
 812                         mddev->reshape_position = MaxSector;
 813                         mddev->delta_disks = 0;
 814                         mddev->new_level = mddev->level;
 815                         mddev->new_layout = mddev->layout;
 816                         mddev->new_chunk = mddev->chunk_size;
 817                 }
 818
 819                 if (sb->state & (1<<MD_SB_CLEAN))
 820                         mddev->recovery_cp = MaxSector;
 821                 else {
 822                         if (sb->events_hi == sb->cp_events_hi &&
 823                                 sb->events_lo == sb->cp_events_lo) {
 824                                 mddev->recovery_cp = sb->recovery_cp;
 825                         } else
 826                                 mddev->recovery_cp = 0;
 827                 }
 828
 829                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 830                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 831                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 832                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 833
 834                 mddev->max_disks = MD_SB_DISKS;
 835
 836                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 837                     mddev->bitmap_file == NULL)
 838                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 839
 840         } else if (mddev->pers == NULL) {
 841                 /* Insist on good event counter while assembling */
 842                 ++ev1;
 843                 if (ev1 < mddev->events)
 844                         return -EINVAL;
 845         } else if (mddev->bitmap) {
 846                 /* if adding to array with a bitmap, then we can accept an
 847                  * older device ... but not too old.
 848                  */
 849                 if (ev1 < mddev->bitmap->events_cleared)
 850                         return 0;
 851         } else {
 852                 if (ev1 < mddev->events)
 853                         /* just a hot-add of a new device, leave raid_disk at -1 */
 854                         return 0;
 855         }
 856
 857         if (mddev->level != LEVEL_MULTIPATH) {
 858                 desc = sb->disks + rdev->desc_nr;
 859
 860                 if (desc->state & (1<<MD_DISK_FAULTY))
 861                         set_bit(Faulty, &rdev->flags);
 862                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
 863                             desc->raid_disk < mddev->raid_disks */) {
 864                         set_bit(In_sync, &rdev->flags);
 865                         rdev->raid_disk = desc->raid_disk;
 866                 }
 867                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 868                         set_bit(WriteMostly, &rdev->flags);
 869         } else /* MULTIPATH are always insync */
 870                 set_bit(In_sync, &rdev->flags);
 871         return 0;
 872 }
 873
 874 /*
 875  * sync_super for 0.90.0
 876  */
 877 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 878 {
 879         mdp_super_t *sb;
 880         struct list_head *tmp;
 881         mdk_rdev_t *rdev2;
 882         int next_spare = mddev->raid_disks;
 883
 884
 885         /* make rdev->sb match mddev data..
 886          *
 887          * 1/ zero out disks
 888          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 889          * 3/ any empty disks < next_spare become removed
 890          *
 891          * disks[0] gets initialised to REMOVED because
 892          * we cannot be sure from other fields if it has
 893          * been initialised or not.
 894          */
 895         int i;
 896         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 897
 898         rdev->sb_size = MD_SB_BYTES;
 899
 900         sb = (mdp_super_t*)page_address(rdev->sb_page);
 901
 902         memset(sb, 0, sizeof(*sb));
 903
 904         sb->md_magic = MD_SB_MAGIC;
 905         sb->major_version = mddev->major_version;
 906         sb->patch_version = mddev->patch_version;
 907         sb->gvalid_words  = 0; /* ignored */
 908         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 909         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 910         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 911         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 912
 913         sb->ctime = mddev->ctime;
 914         sb->level = mddev->level;
 915         sb->size  = mddev->size;
 916         sb->raid_disks = mddev->raid_disks;
 917         sb->md_minor = mddev->md_minor;
 918         sb->not_persistent = 0;
 919         sb->utime = mddev->utime;
 920         sb->state = 0;
 921         sb->events_hi = (mddev->events>>32);
 922         sb->events_lo = (u32)mddev->events;
 923
 924         if (mddev->reshape_position == MaxSector)
 925                 sb->minor_version = 90;
 926         else {
 927                 sb->minor_version = 91;
 928                 sb->reshape_position = mddev->reshape_position;
 929                 sb->new_level = mddev->new_level;
 930                 sb->delta_disks = mddev->delta_disks;
 931                 sb->new_layout = mddev->new_layout;
 932                 sb->new_chunk = mddev->new_chunk;
 933         }
 934         mddev->minor_version = sb->minor_version;
 935         if (mddev->in_sync)
 936         {
 937                 sb->recovery_cp = mddev->recovery_cp;
 938                 sb->cp_events_hi = (mddev->events>>32);
 939                 sb->cp_events_lo = (u32)mddev->events;
 940                 if (mddev->recovery_cp == MaxSector)
 941                         sb->state = (1<< MD_SB_CLEAN);
 942         } else
 943                 sb->recovery_cp = 0;
 944
 945         sb->layout = mddev->layout;
 946         sb->chunk_size = mddev->chunk_size;
 947
 948         if (mddev->bitmap && mddev->bitmap_file == NULL)
 949                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 950
 951         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 952         rdev_for_each(rdev2, tmp, mddev) {
 953                 mdp_disk_t *d;
 954                 int desc_nr;
 955                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 956                     && !test_bit(Faulty, &rdev2->flags))
 957                         desc_nr = rdev2->raid_disk;
 958                 else
 959                         desc_nr = next_spare++;
 960                 rdev2->desc_nr = desc_nr;
 961                 d = &sb->disks[rdev2->desc_nr];
 962                 nr_disks++;
 963                 d->number = rdev2->desc_nr;
 964                 d->major = MAJOR(rdev2->bdev->bd_dev);
 965                 d->minor = MINOR(rdev2->bdev->bd_dev);
 966                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 967                     && !test_bit(Faulty, &rdev2->flags))
 968                         d->raid_disk = rdev2->raid_disk;
 969                 else
 970                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 971                 if (test_bit(Faulty, &rdev2->flags))
 972                         d->state = (1<<MD_DISK_FAULTY);
 973                 else if (test_bit(In_sync, &rdev2->flags)) {
 974                         d->state = (1<<MD_DISK_ACTIVE);
 975                         d->state |= (1<<MD_DISK_SYNC);
 976                         active++;
 977                         working++;
 978                 } else {
 979                         d->state = 0;
 980                         spare++;
 981                         working++;
 982                 }
 983                 if (test_bit(WriteMostly, &rdev2->flags))
 984                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 985         }
 986         /* now set the "removed" and "faulty" bits on any missing devices */
 987         for (i=0 ; i < mddev->raid_disks ; i++) {
 988                 mdp_disk_t *d = &sb->disks[i];
 989                 if (d->state == 0 && d->number == 0) {
 990                         d->number = i;
 991                         d->raid_disk = i;
 992                         d->state = (1<<MD_DISK_REMOVED);
 993                         d->state |= (1<<MD_DISK_FAULTY);
 994                         failed++;
 995                 }
 996         }
 997         sb->nr_disks = nr_disks;
 998         sb->active_disks = active;
 999         sb->working_disks = working;
1000         sb->failed_disks = failed;
1001         sb->spare_disks = spare;
1002
1003         sb->this_disk = sb->disks[rdev->desc_nr];
1004         sb->sb_csum = calc_sb_csum(sb);
1005 }
1006
1007 /*
1008  * version 1 superblock
1009  */
1010
1011 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1012 {
1013         __le32 disk_csum;
1014         u32 csum;
1015         unsigned long long newcsum;
1016         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1017         __le32 *isuper = (__le32*)sb;
1018         int i;
1019
1020         disk_csum = sb->sb_csum;
1021         sb->sb_csum = 0;
1022         newcsum = 0;
1023         for (i=0; size>=4; size -= 4 )
1024                 newcsum += le32_to_cpu(*isuper++);
1025
1026         if (size == 2)
1027                 newcsum += le16_to_cpu(*(__le16*) isuper);
1028
1029         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1030         sb->sb_csum = disk_csum;
1031         return cpu_to_le32(csum);
1032 }
1033
1034 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1035 {
1036         struct mdp_superblock_1 *sb;
1037         int ret;
1038         sector_t sb_offset;
1039         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1040         int bmask;
1041
1042         /*
1043          * Calculate the position of the superblock.
1044          * It is always aligned to a 4K boundary and
1045          * depeding on minor_version, it can be:
1046          * 0: At least 8K, but less than 12K, from end of device
1047          * 1: At start of device
1048          * 2: 4K from start of device.
1049          */
1050         switch(minor_version) {
1051         case 0:
1052                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1053                 sb_offset -= 8*2;
1054                 sb_offset &= ~(sector_t)(4*2-1);
1055                 /* convert from sectors to K */
1056                 sb_offset /= 2;
1057                 break;
1058         case 1:
1059                 sb_offset = 0;
1060                 break;
1061         case 2:
1062                 sb_offset = 4;
1063                 break;
1064         default:
1065                 return -EINVAL;
1066         }
1067         rdev->sb_offset = sb_offset;
1068
1069         /* superblock is rarely larger than 1K, but it can be larger,
1070          * and it is safe to read 4k, so we do that
1071          */
1072         ret = read_disk_sb(rdev, 4096);
1073         if (ret) return ret;
1074
1075
1076         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1077
1078         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1079             sb->major_version != cpu_to_le32(1) ||
1080             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1081             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1082             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1083                 return -EINVAL;
1084
1085         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1086                 printk("md: invalid superblock checksum on %s\n",
1087                         bdevname(rdev->bdev,b));
1088                 return -EINVAL;
1089         }
1090         if (le64_to_cpu(sb->data_size) < 10) {
1091                 printk("md: data_size too small on %s\n",
1092                        bdevname(rdev->bdev,b));
1093                 return -EINVAL;
1094         }
1095         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1096                 if (sb->level != cpu_to_le32(1) &&
1097                     sb->level != cpu_to_le32(4) &&
1098                     sb->level != cpu_to_le32(5) &&
1099                     sb->level != cpu_to_le32(6) &&
1100                     sb->level != cpu_to_le32(10)) {
1101                         printk(KERN_WARNING
1102                                "md: bitmaps not supported for this level.\n");
1103                         return -EINVAL;
1104                 }
1105         }
1106
1107         rdev->preferred_minor = 0xffff;
1108         rdev->data_offset = le64_to_cpu(sb->data_offset);
1109         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1110
1111         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1112         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1113         if (rdev->sb_size & bmask)
1114                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1115
1116         if (minor_version
1117             && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1118                 return -EINVAL;
1119
1120         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1121                 rdev->desc_nr = -1;
1122         else
1123                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1124
1125         if (!refdev) {
1126                 ret = 1;
1127         } else {
1128                 __u64 ev1, ev2;
1129                 struct mdp_superblock_1 *refsb =
1130                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1131
1132                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1133                     sb->level != refsb->level ||
1134                     sb->layout != refsb->layout ||
1135                     sb->chunksize != refsb->chunksize) {
1136                         printk(KERN_WARNING "md: %s has strangely different"
1137                                 " superblock to %s\n",
1138                                 bdevname(rdev->bdev,b),
1139                                 bdevname(refdev->bdev,b2));
1140                         return -EINVAL;
1141                 }
1142                 ev1 = le64_to_cpu(sb->events);
1143                 ev2 = le64_to_cpu(refsb->events);
1144
1145                 if (ev1 > ev2)
1146                         ret = 1;
1147                 else
1148                         ret = 0;
1149         }
1150         if (minor_version)
1151                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1152         else
1153                 rdev->size = rdev->sb_offset;
1154         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1155                 return -EINVAL;
1156         rdev->size = le64_to_cpu(sb->data_size)/2;
1157         if (le32_to_cpu(sb->chunksize))
1158                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1159
1160         if (le64_to_cpu(sb->size) > rdev->size*2)
1161                 return -EINVAL;
1162         return ret;
1163 }
1164
1165 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1166 {
1167         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1168         __u64 ev1 = le64_to_cpu(sb->events);
1169
1170         rdev->raid_disk = -1;
1171         clear_bit(Faulty, &rdev->flags);
1172         clear_bit(In_sync, &rdev->flags);
1173         clear_bit(WriteMostly, &rdev->flags);
1174         clear_bit(BarriersNotsupp, &rdev->flags);
1175
1176         if (mddev->raid_disks == 0) {
1177                 mddev->major_version = 1;
1178                 mddev->patch_version = 0;
1179                 mddev->external = 0;
1180                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1181                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1182                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1183                 mddev->level = le32_to_cpu(sb->level);
1184                 mddev->clevel[0] = 0;
1185                 mddev->layout = le32_to_cpu(sb->layout);
1186                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1187                 mddev->size = le64_to_cpu(sb->size)/2;
1188                 mddev->events = ev1;
1189                 mddev->bitmap_offset = 0;
1190                 mddev->default_bitmap_offset = 1024 >> 9;
1191
1192                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1193                 memcpy(mddev->uuid, sb->set_uuid, 16);
1194
1195                 mddev->max_disks =  (4096-256)/2;
1196
1197                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1198                     mddev->bitmap_file == NULL )
1199                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1200
1201                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1202                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1203                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1204                         mddev->new_level = le32_to_cpu(sb->new_level);
1205                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1206                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1207                 } else {
1208                         mddev->reshape_position = MaxSector;
1209                         mddev->delta_disks = 0;
1210                         mddev->new_level = mddev->level;
1211                         mddev->new_layout = mddev->layout;
1212                         mddev->new_chunk = mddev->chunk_size;
1213                 }
1214
1215         } else if (mddev->pers == NULL) {
1216                 /* Insist of good event counter while assembling */
1217                 ++ev1;
1218                 if (ev1 < mddev->events)
1219                         return -EINVAL;
1220         } else if (mddev->bitmap) {
1221                 /* If adding to array with a bitmap, then we can accept an
1222                  * older device, but not too old.
1223                  */
1224                 if (ev1 < mddev->bitmap->events_cleared)
1225                         return 0;
1226         } else {
1227                 if (ev1 < mddev->events)
1228                         /* just a hot-add of a new device, leave raid_disk at -1 */
1229                         return 0;
1230         }
1231         if (mddev->level != LEVEL_MULTIPATH) {
1232                 int role;
1233                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1234                 switch(role) {
1235                 case 0xffff: /* spare */
1236                         break;
1237                 case 0xfffe: /* faulty */
1238                         set_bit(Faulty, &rdev->flags);
1239                         break;
1240                 default:
1241                         if ((le32_to_cpu(sb->feature_map) &
1242                              MD_FEATURE_RECOVERY_OFFSET))
1243                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1244                         else
1245                                 set_bit(In_sync, &rdev->flags);
1246                         rdev->raid_disk = role;
1247                         break;
1248                 }
1249                 if (sb->devflags & WriteMostly1)
1250                         set_bit(WriteMostly, &rdev->flags);
1251         } else /* MULTIPATH are always insync */
1252                 set_bit(In_sync, &rdev->flags);
1253
1254         return 0;
1255 }
1256
1257 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1258 {
1259         struct mdp_superblock_1 *sb;
1260         struct list_head *tmp;
1261         mdk_rdev_t *rdev2;
1262         int max_dev, i;
1263         /* make rdev->sb match mddev and rdev data. */
1264
1265         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1266
1267         sb->feature_map = 0;
1268         sb->pad0 = 0;
1269         sb->recovery_offset = cpu_to_le64(0);
1270         memset(sb->pad1, 0, sizeof(sb->pad1));
1271         memset(sb->pad2, 0, sizeof(sb->pad2));
1272         memset(sb->pad3, 0, sizeof(sb->pad3));
1273
1274         sb->utime = cpu_to_le64((__u64)mddev->utime);
1275         sb->events = cpu_to_le64(mddev->events);
1276         if (mddev->in_sync)
1277                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1278         else
1279                 sb->resync_offset = cpu_to_le64(0);
1280
1281         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1282
1283         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1284         sb->size = cpu_to_le64(mddev->size<<1);
1285
1286         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1287                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1288                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1289         }
1290
1291         if (rdev->raid_disk >= 0 &&
1292             !test_bit(In_sync, &rdev->flags) &&
1293             rdev->recovery_offset > 0) {
1294                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1295                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1296         }
1297
1298         if (mddev->reshape_position != MaxSector) {
1299                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1300                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1301                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1302                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1303                 sb->new_level = cpu_to_le32(mddev->new_level);
1304                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1305         }
1306
1307         max_dev = 0;
1308         rdev_for_each(rdev2, tmp, mddev)
1309                 if (rdev2->desc_nr+1 > max_dev)
1310                         max_dev = rdev2->desc_nr+1;
1311
1312         if (max_dev > le32_to_cpu(sb->max_dev))
1313                 sb->max_dev = cpu_to_le32(max_dev);
1314         for (i=0; i<max_dev;i++)
1315                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1316
1317         rdev_for_each(rdev2, tmp, mddev) {
1318                 i = rdev2->desc_nr;
1319                 if (test_bit(Faulty, &rdev2->flags))
1320                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1321                 else if (test_bit(In_sync, &rdev2->flags))
1322                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1323                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1324                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1325                 else
1326                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1327         }
1328
1329         sb->sb_csum = calc_sb_1_csum(sb);
1330 }
1331
1332
1333 static struct super_type super_types[] = {
1334         [0] = {
1335                 .name   = "0.90.0",
1336                 .owner  = THIS_MODULE,
1337                 .load_super     = super_90_load,
1338                 .validate_super = super_90_validate,
1339                 .sync_super     = super_90_sync,
1340         },
1341         [1] = {
1342                 .name   = "md-1",
1343                 .owner  = THIS_MODULE,
1344                 .load_super     = super_1_load,
1345                 .validate_super = super_1_validate,
1346                 .sync_super     = super_1_sync,
1347         },
1348 };
1349
1350 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1351 {
1352         struct list_head *tmp, *tmp2;
1353         mdk_rdev_t *rdev, *rdev2;
1354
1355         rdev_for_each(rdev, tmp, mddev1)
1356                 rdev_for_each(rdev2, tmp2, mddev2)
1357                         if (rdev->bdev->bd_contains ==
1358                             rdev2->bdev->bd_contains)
1359                                 return 1;
1360
1361         return 0;
1362 }
1363
1364 static LIST_HEAD(pending_raid_disks);
1365
1366 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1367 {
1368         char b[BDEVNAME_SIZE];
1369         struct kobject *ko;
1370         char *s;
1371         int err;
1372
1373         if (rdev->mddev) {
1374                 MD_BUG();
1375                 return -EINVAL;
1376         }
1377
1378         /* prevent duplicates */
1379         if (find_rdev(mddev, rdev->bdev->bd_dev))
1380                 return -EEXIST;
1381
1382         /* make sure rdev->size exceeds mddev->size */
1383         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1384                 if (mddev->pers) {
1385                         /* Cannot change size, so fail
1386                          * If mddev->level <= 0, then we don't care
1387                          * about aligning sizes (e.g. linear)
1388                          */
1389                         if (mddev->level > 0)
1390                                 return -ENOSPC;
1391                 } else
1392                         mddev->size = rdev->size;
1393         }
1394
1395         /* Verify rdev->desc_nr is unique.
1396          * If it is -1, assign a free number, else
1397          * check number is not in use
1398          */
1399         if (rdev->desc_nr < 0) {
1400                 int choice = 0;
1401                 if (mddev->pers) choice = mddev->raid_disks;
1402                 while (find_rdev_nr(mddev, choice))
1403                         choice++;
1404                 rdev->desc_nr = choice;
1405         } else {
1406                 if (find_rdev_nr(mddev, rdev->desc_nr))
1407                         return -EBUSY;
1408         }
1409         bdevname(rdev->bdev,b);
1410         while ( (s=strchr(b, '/')) != NULL)
1411                 *s = '!';
1412
1413         rdev->mddev = mddev;
1414         printk(KERN_INFO "md: bind<%s>\n", b);
1415
1416         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1417                 goto fail;
1418
1419         if (rdev->bdev->bd_part)
1420                 ko = &rdev->bdev->bd_part->dev.kobj;
1421         else
1422                 ko = &rdev->bdev->bd_disk->dev.kobj;
1423         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1424                 kobject_del(&rdev->kobj);
1425                 goto fail;
1426         }
1427         list_add(&rdev->same_set, &mddev->disks);
1428         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1429         return 0;
1430
1431  fail:
1432         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1433                b, mdname(mddev));
1434         return err;
1435 }
1436
1437 static void md_delayed_delete(struct work_struct *ws)
1438 {
1439         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1440         kobject_del(&rdev->kobj);
1441         kobject_put(&rdev->kobj);
1442 }
1443
1444 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1445 {
1446         char b[BDEVNAME_SIZE];
1447         if (!rdev->mddev) {
1448                 MD_BUG();
1449                 return;
1450         }
1451         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1452         list_del_init(&rdev->same_set);
1453         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1454         rdev->mddev = NULL;
1455         sysfs_remove_link(&rdev->kobj, "block");
1456
1457         /* We need to delay this, otherwise we can deadlock when
1458          * writing to 'remove' to "dev/state"
1459          */
1460         INIT_WORK(&rdev->del_work, md_delayed_delete);
1461         kobject_get(&rdev->kobj);
1462         schedule_work(&rdev->del_work);
1463 }
1464
1465 /*
1466  * prevent the device from being mounted, repartitioned or
1467  * otherwise reused by a RAID array (or any other kernel
1468  * subsystem), by bd_claiming the device.
1469  */
1470 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1471 {
1472         int err = 0;
1473         struct block_device *bdev;
1474         char b[BDEVNAME_SIZE];
1475
1476         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1477         if (IS_ERR(bdev)) {
1478                 printk(KERN_ERR "md: could not open %s.\n",
1479                         __bdevname(dev, b));
1480                 return PTR_ERR(bdev);
1481         }
1482         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1483         if (err) {
1484                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1485                         bdevname(bdev, b));
1486                 blkdev_put(bdev);
1487                 return err;
1488         }
1489         if (!shared)
1490                 set_bit(AllReserved, &rdev->flags);
1491         rdev->bdev = bdev;
1492         return err;
1493 }
1494
1495 static void unlock_rdev(mdk_rdev_t *rdev)
1496 {
1497         struct block_device *bdev = rdev->bdev;
1498         rdev->bdev = NULL;
1499         if (!bdev)
1500                 MD_BUG();
1501         bd_release(bdev);
1502         blkdev_put(bdev);
1503 }
1504
1505 void md_autodetect_dev(dev_t dev);
1506
1507 static void export_rdev(mdk_rdev_t * rdev)
1508 {
1509         char b[BDEVNAME_SIZE];
1510         printk(KERN_INFO "md: export_rdev(%s)\n",
1511                 bdevname(rdev->bdev,b));
1512         if (rdev->mddev)
1513                 MD_BUG();
1514         free_disk_sb(rdev);
1515         list_del_init(&rdev->same_set);
1516 #ifndef MODULE
1517         if (test_bit(AutoDetected, &rdev->flags))
1518                 md_autodetect_dev(rdev->bdev->bd_dev);
1519 #endif
1520         unlock_rdev(rdev);
1521         kobject_put(&rdev->kobj);
1522 }
1523
1524 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1525 {
1526         unbind_rdev_from_array(rdev);
1527         export_rdev(rdev);
1528 }
1529
1530 static void export_array(mddev_t *mddev)
1531 {
1532         struct list_head *tmp;
1533         mdk_rdev_t *rdev;
1534
1535         rdev_for_each(rdev, tmp, mddev) {
1536                 if (!rdev->mddev) {
1537                         MD_BUG();
1538                         continue;
1539                 }
1540                 kick_rdev_from_array(rdev);
1541         }
1542         if (!list_empty(&mddev->disks))
1543                 MD_BUG();
1544         mddev->raid_disks = 0;
1545         mddev->major_version = 0;
1546 }
1547
1548 static void print_desc(mdp_disk_t *desc)
1549 {
1550         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1551                 desc->major,desc->minor,desc->raid_disk,desc->state);
1552 }
1553
1554 static void print_sb(mdp_super_t *sb)
1555 {
1556         int i;
1557
1558         printk(KERN_INFO
1559                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1560                 sb->major_version, sb->minor_version, sb->patch_version,
1561                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1562                 sb->ctime);
1563         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1564                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1565                 sb->md_minor, sb->layout, sb->chunk_size);
1566         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1567                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1568                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1569                 sb->failed_disks, sb->spare_disks,
1570                 sb->sb_csum, (unsigned long)sb->events_lo);
1571
1572         printk(KERN_INFO);
1573         for (i = 0; i < MD_SB_DISKS; i++) {
1574                 mdp_disk_t *desc;
1575
1576                 desc = sb->disks + i;
1577                 if (desc->number || desc->major || desc->minor ||
1578                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1579                         printk("     D %2d: ", i);
1580                         print_desc(desc);
1581                 }
1582         }
1583         printk(KERN_INFO "md:     THIS: ");
1584         print_desc(&sb->this_disk);
1585
1586 }
1587
1588 static void print_rdev(mdk_rdev_t *rdev)
1589 {
1590         char b[BDEVNAME_SIZE];
1591         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1592                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1593                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1594                 rdev->desc_nr);
1595         if (rdev->sb_loaded) {
1596                 printk(KERN_INFO "md: rdev superblock:\n");
1597                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1598         } else
1599                 printk(KERN_INFO "md: no rdev superblock!\n");
1600 }
1601
1602 static void md_print_devices(void)
1603 {
1604         struct list_head *tmp, *tmp2;
1605         mdk_rdev_t *rdev;
1606         mddev_t *mddev;
1607         char b[BDEVNAME_SIZE];
1608
1609         printk("\n");
1610         printk("md:     **********************************\n");
1611         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1612         printk("md:     **********************************\n");
1613         for_each_mddev(mddev, tmp) {
1614
1615                 if (mddev->bitmap)
1616                         bitmap_print_sb(mddev->bitmap);
1617                 else
1618                         printk("%s: ", mdname(mddev));
1619                 rdev_for_each(rdev, tmp2, mddev)
1620                         printk("<%s>", bdevname(rdev->bdev,b));
1621                 printk("\n");
1622
1623                 rdev_for_each(rdev, tmp2, mddev)
1624                         print_rdev(rdev);
1625         }
1626         printk("md:     **********************************\n");
1627         printk("\n");
1628 }
1629
1630
1631 static void sync_sbs(mddev_t * mddev, int nospares)
1632 {
1633         /* Update each superblock (in-memory image), but
1634          * if we are allowed to, skip spares which already
1635          * have the right event counter, or have one earlier
1636          * (which would mean they aren't being marked as dirty
1637          * with the rest of the array)
1638          */
1639         mdk_rdev_t *rdev;
1640         struct list_head *tmp;
1641
1642         rdev_for_each(rdev, tmp, mddev) {
1643                 if (rdev->sb_events == mddev->events ||
1644                     (nospares &&
1645                      rdev->raid_disk < 0 &&
1646                      (rdev->sb_events&1)==0 &&
1647                      rdev->sb_events+1 == mddev->events)) {
1648                         /* Don't update this superblock */
1649                         rdev->sb_loaded = 2;
1650                 } else {
1651                         super_types[mddev->major_version].
1652                                 sync_super(mddev, rdev);
1653                         rdev->sb_loaded = 1;
1654                 }
1655         }
1656 }
1657
1658 static void md_update_sb(mddev_t * mddev, int force_change)
1659 {
1660         struct list_head *tmp;
1661         mdk_rdev_t *rdev;
1662         int sync_req;
1663         int nospares = 0;
1664
1665         if (mddev->external)
1666                 return;
1667 repeat:
1668         spin_lock_irq(&mddev->write_lock);
1669
1670         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1671         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1672                 force_change = 1;
1673         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1674                 /* just a clean<-> dirty transition, possibly leave spares alone,
1675                  * though if events isn't the right even/odd, we will have to do
1676                  * spares after all
1677                  */
1678                 nospares = 1;
1679         if (force_change)
1680                 nospares = 0;
1681         if (mddev->degraded)
1682                 /* If the array is degraded, then skipping spares is both
1683                  * dangerous and fairly pointless.
1684                  * Dangerous because a device that was removed from the array
1685                  * might have a event_count that still looks up-to-date,
1686                  * so it can be re-added without a resync.
1687                  * Pointless because if there are any spares to skip,
1688                  * then a recovery will happen and soon that array won't
1689                  * be degraded any more and the spare can go back to sleep then.
1690                  */
1691                 nospares = 0;
1692
1693         sync_req = mddev->in_sync;
1694         mddev->utime = get_seconds();
1695
1696         /* If this is just a dirty<->clean transition, and the array is clean
1697          * and 'events' is odd, we can roll back to the previous clean state */
1698         if (nospares
1699             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1700             && (mddev->events & 1)
1701             && mddev->events != 1)
1702                 mddev->events--;
1703         else {
1704                 /* otherwise we have to go forward and ... */
1705                 mddev->events ++;
1706                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1707                         /* .. if the array isn't clean, insist on an odd 'events' */
1708                         if ((mddev->events&1)==0) {
1709                                 mddev->events++;
1710                                 nospares = 0;
1711                         }
1712                 } else {
1713                         /* otherwise insist on an even 'events' (for clean states) */
1714                         if ((mddev->events&1)) {
1715                                 mddev->events++;
1716                                 nospares = 0;
1717                         }
1718                 }
1719         }
1720
1721         if (!mddev->events) {
1722                 /*
1723                  * oops, this 64-bit counter should never wrap.
1724                  * Either we are in around ~1 trillion A.C., assuming
1725                  * 1 reboot per second, or we have a bug:
1726                  */
1727                 MD_BUG();
1728                 mddev->events --;
1729         }
1730
1731         /*
1732          * do not write anything to disk if using
1733          * nonpersistent superblocks
1734          */
1735         if (!mddev->persistent) {
1736                 if (!mddev->external)
1737                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1738
1739                 spin_unlock_irq(&mddev->write_lock);
1740                 wake_up(&mddev->sb_wait);
1741                 return;
1742         }
1743         sync_sbs(mddev, nospares);
1744         spin_unlock_irq(&mddev->write_lock);
1745
1746         dprintk(KERN_INFO
1747                 "md: updating %s RAID superblock on device (in sync %d)\n",
1748                 mdname(mddev),mddev->in_sync);
1749
1750         bitmap_update_sb(mddev->bitmap);
1751         rdev_for_each(rdev, tmp, mddev) {
1752                 char b[BDEVNAME_SIZE];
1753                 dprintk(KERN_INFO "md: ");
1754                 if (rdev->sb_loaded != 1)
1755                         continue; /* no noise on spare devices */
1756                 if (test_bit(Faulty, &rdev->flags))
1757                         dprintk("(skipping faulty ");
1758
1759                 dprintk("%s ", bdevname(rdev->bdev,b));
1760                 if (!test_bit(Faulty, &rdev->flags)) {
1761                         md_super_write(mddev,rdev,
1762                                        rdev->sb_offset<<1, rdev->sb_size,
1763                                        rdev->sb_page);
1764                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1765                                 bdevname(rdev->bdev,b),
1766                                 (unsigned long long)rdev->sb_offset);
1767                         rdev->sb_events = mddev->events;
1768
1769                 } else
1770                         dprintk(")\n");
1771                 if (mddev->level == LEVEL_MULTIPATH)
1772                         /* only need to write one superblock... */
1773                         break;
1774         }
1775         md_super_wait(mddev);
1776         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1777
1778         spin_lock_irq(&mddev->write_lock);
1779         if (mddev->in_sync != sync_req ||
1780             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1781                 /* have to write it out again */
1782                 spin_unlock_irq(&mddev->write_lock);
1783                 goto repeat;
1784         }
1785         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1786         spin_unlock_irq(&mddev->write_lock);
1787         wake_up(&mddev->sb_wait);
1788
1789 }
1790
1791 /* words written to sysfs files may, or my not, be \n terminated.
1792  * We want to accept with case. For this we use cmd_match.
1793  */
1794 static int cmd_match(const char *cmd, const char *str)
1795 {
1796         /* See if cmd, written into a sysfs file, matches
1797          * str.  They must either be the same, or cmd can
1798          * have a trailing newline
1799          */
1800         while (*cmd && *str && *cmd == *str) {
1801                 cmd++;
1802                 str++;
1803         }
1804         if (*cmd == '\n')
1805                 cmd++;
1806         if (*str || *cmd)
1807                 return 0;
1808         return 1;
1809 }
1810
1811 struct rdev_sysfs_entry {
1812         struct attribute attr;
1813         ssize_t (*show)(mdk_rdev_t *, char *);
1814         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1815 };
1816
1817 static ssize_t
1818 state_show(mdk_rdev_t *rdev, char *page)
1819 {
1820         char *sep = "";
1821         size_t len = 0;
1822
1823         if (test_bit(Faulty, &rdev->flags)) {
1824                 len+= sprintf(page+len, "%sfaulty",sep);
1825                 sep = ",";
1826         }
1827         if (test_bit(In_sync, &rdev->flags)) {
1828                 len += sprintf(page+len, "%sin_sync",sep);
1829                 sep = ",";
1830         }
1831         if (test_bit(WriteMostly, &rdev->flags)) {
1832                 len += sprintf(page+len, "%swrite_mostly",sep);
1833                 sep = ",";
1834         }
1835         if (test_bit(Blocked, &rdev->flags)) {
1836                 len += sprintf(page+len, "%sblocked", sep);
1837                 sep = ",";
1838         }
1839         if (!test_bit(Faulty, &rdev->flags) &&
1840             !test_bit(In_sync, &rdev->flags)) {
1841                 len += sprintf(page+len, "%sspare", sep);
1842                 sep = ",";
1843         }
1844         return len+sprintf(page+len, "\n");
1845 }
1846
1847 static ssize_t
1848 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1849 {
1850         /* can write
1851          *  faulty  - simulates and error
1852          *  remove  - disconnects the device
1853          *  writemostly - sets write_mostly
1854          *  -writemostly - clears write_mostly
1855          *  blocked - sets the Blocked flag
1856          *  -blocked - clears the Blocked flag
1857          */
1858         int err = -EINVAL;
1859         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1860                 md_error(rdev->mddev, rdev);
1861                 err = 0;
1862         } else if (cmd_match(buf, "remove")) {
1863                 if (rdev->raid_disk >= 0)
1864                         err = -EBUSY;
1865                 else {
1866                         mddev_t *mddev = rdev->mddev;
1867                         kick_rdev_from_array(rdev);
1868                         if (mddev->pers)
1869                                 md_update_sb(mddev, 1);
1870                         md_new_event(mddev);
1871                         err = 0;
1872                 }
1873         } else if (cmd_match(buf, "writemostly")) {
1874                 set_bit(WriteMostly, &rdev->flags);
1875                 err = 0;
1876         } else if (cmd_match(buf, "-writemostly")) {
1877                 clear_bit(WriteMostly, &rdev->flags);
1878                 err = 0;
1879         } else if (cmd_match(buf, "blocked")) {
1880                 set_bit(Blocked, &rdev->flags);
1881                 err = 0;
1882         } else if (cmd_match(buf, "-blocked")) {
1883                 clear_bit(Blocked, &rdev->flags);
1884                 wake_up(&rdev->blocked_wait);
1885                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1886                 md_wakeup_thread(rdev->mddev->thread);
1887
1888                 err = 0;
1889         }
1890         return err ? err : len;
1891 }
1892 static struct rdev_sysfs_entry rdev_state =
1893 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1894
1895 static ssize_t
1896 errors_show(mdk_rdev_t *rdev, char *page)
1897 {
1898         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1899 }
1900
1901 static ssize_t
1902 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1903 {
1904         char *e;
1905         unsigned long n = simple_strtoul(buf, &e, 10);
1906         if (*buf && (*e == 0 || *e == '\n')) {
1907                 atomic_set(&rdev->corrected_errors, n);
1908                 return len;
1909         }
1910         return -EINVAL;
1911 }
1912 static struct rdev_sysfs_entry rdev_errors =
1913 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1914
1915 static ssize_t
1916 slot_show(mdk_rdev_t *rdev, char *page)
1917 {
1918         if (rdev->raid_disk < 0)
1919                 return sprintf(page, "none\n");
1920         else
1921                 return sprintf(page, "%d\n", rdev->raid_disk);
1922 }
1923
1924 static ssize_t
1925 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1926 {
1927         char *e;
1928         int err;
1929         char nm[20];
1930         int slot = simple_strtoul(buf, &e, 10);
1931         if (strncmp(buf, "none", 4)==0)
1932                 slot = -1;
1933         else if (e==buf || (*e && *e!= '\n'))
1934                 return -EINVAL;
1935         if (rdev->mddev->pers && slot == -1) {
1936                 /* Setting 'slot' on an active array requires also
1937                  * updating the 'rd%d' link, and communicating
1938                  * with the personality with ->hot_*_disk.
1939                  * For now we only support removing
1940                  * failed/spare devices.  This normally happens automatically,
1941                  * but not when the metadata is externally managed.
1942                  */
1943                 if (rdev->raid_disk == -1)
1944                         return -EEXIST;
1945                 /* personality does all needed checks */
1946                 if (rdev->mddev->pers->hot_add_disk == NULL)
1947                         return -EINVAL;
1948                 err = rdev->mddev->pers->
1949                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
1950                 if (err)
1951                         return err;
1952                 sprintf(nm, "rd%d", rdev->raid_disk);
1953                 sysfs_remove_link(&rdev->mddev->kobj, nm);
1954                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1955                 md_wakeup_thread(rdev->mddev->thread);
1956         } else if (rdev->mddev->pers) {
1957                 mdk_rdev_t *rdev2;
1958                 struct list_head *tmp;
1959                 /* Activating a spare .. or possibly reactivating
1960                  * if we every get bitmaps working here.
1961                  */
1962
1963                 if (rdev->raid_disk != -1)
1964                         return -EBUSY;
1965
1966                 if (rdev->mddev->pers->hot_add_disk == NULL)
1967                         return -EINVAL;
1968
1969                 rdev_for_each(rdev2, tmp, rdev->mddev)
1970                         if (rdev2->raid_disk == slot)
1971                                 return -EEXIST;
1972
1973                 rdev->raid_disk = slot;
1974                 if (test_bit(In_sync, &rdev->flags))
1975                         rdev->saved_raid_disk = slot;
1976                 else
1977                         rdev->saved_raid_disk = -1;
1978                 err = rdev->mddev->pers->
1979                         hot_add_disk(rdev->mddev, rdev);
1980                 if (err) {
1981                         rdev->raid_disk = -1;
1982                         return err;
1983                 }
1984                 sprintf(nm, "rd%d", rdev->raid_disk);
1985                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
1986                         printk(KERN_WARNING
1987                                "md: cannot register "
1988                                "%s for %s\n",
1989                                nm, mdname(rdev->mddev));
1990
1991                 /* don't wakeup anyone, leave that to userspace. */
1992         } else {
1993                 if (slot >= rdev->mddev->raid_disks)
1994                         return -ENOSPC;
1995                 rdev->raid_disk = slot;
1996                 /* assume it is working */
1997                 clear_bit(Faulty, &rdev->flags);
1998                 clear_bit(WriteMostly, &rdev->flags);
1999                 set_bit(In_sync, &rdev->flags);
2000         }
2001         return len;
2002 }
2003
2004
2005 static struct rdev_sysfs_entry rdev_slot =
2006 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2007
2008 static ssize_t
2009 offset_show(mdk_rdev_t *rdev, char *page)
2010 {
2011         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2012 }
2013
2014 static ssize_t
2015 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2016 {
2017         char *e;
2018         unsigned long long offset = simple_strtoull(buf, &e, 10);
2019         if (e==buf || (*e && *e != '\n'))
2020                 return -EINVAL;
2021         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2022                 return -EBUSY;
2023         if (rdev->size && rdev->mddev->external)
2024                 /* Must set offset before size, so overlap checks
2025                  * can be sane */
2026                 return -EBUSY;
2027         rdev->data_offset = offset;
2028         return len;
2029 }
2030
2031 static struct rdev_sysfs_entry rdev_offset =
2032 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2033
2034 static ssize_t
2035 rdev_size_show(mdk_rdev_t *rdev, char *page)
2036 {
2037         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2038 }
2039
2040 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2041 {
2042         /* check if two start/length pairs overlap */
2043         if (s1+l1 <= s2)
2044                 return 0;
2045         if (s2+l2 <= s1)
2046                 return 0;
2047         return 1;
2048 }
2049
2050 static ssize_t
2051 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2052 {
2053         char *e;
2054         unsigned long long size = simple_strtoull(buf, &e, 10);
2055         unsigned long long oldsize = rdev->size;
2056         mddev_t *my_mddev = rdev->mddev;
2057
2058         if (e==buf || (*e && *e != '\n'))
2059                 return -EINVAL;
2060         if (my_mddev->pers && rdev->raid_disk >= 0)
2061                 return -EBUSY;
2062         rdev->size = size;
2063         if (size > oldsize && rdev->mddev->external) {
2064                 /* need to check that all other rdevs with the same ->bdev
2065                  * do not overlap.  We need to unlock the mddev to avoid
2066                  * a deadlock.  We have already changed rdev->size, and if
2067                  * we have to change it back, we will have the lock again.
2068                  */
2069                 mddev_t *mddev;
2070                 int overlap = 0;
2071                 struct list_head *tmp, *tmp2;
2072
2073                 mddev_unlock(my_mddev);
2074                 for_each_mddev(mddev, tmp) {
2075                         mdk_rdev_t *rdev2;
2076
2077                         mddev_lock(mddev);
2078                         rdev_for_each(rdev2, tmp2, mddev)
2079                                 if (test_bit(AllReserved, &rdev2->flags) ||
2080                                     (rdev->bdev == rdev2->bdev &&
2081                                      rdev != rdev2 &&
2082                                      overlaps(rdev->data_offset, rdev->size,
2083                                             rdev2->data_offset, rdev2->size))) {
2084                                         overlap = 1;
2085                                         break;
2086                                 }
2087                         mddev_unlock(mddev);
2088                         if (overlap) {
2089                                 mddev_put(mddev);
2090                                 break;
2091                         }
2092                 }
2093                 mddev_lock(my_mddev);
2094                 if (overlap) {
2095                         /* Someone else could have slipped in a size
2096                          * change here, but doing so is just silly.
2097                          * We put oldsize back because we *know* it is
2098                          * safe, and trust userspace not to race with
2099                          * itself
2100                          */
2101                         rdev->size = oldsize;
2102                         return -EBUSY;
2103                 }
2104         }
2105         if (size < my_mddev->size || my_mddev->size == 0)
2106                 my_mddev->size = size;
2107         return len;
2108 }
2109
2110 static struct rdev_sysfs_entry rdev_size =
2111 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2112
2113 static struct attribute *rdev_default_attrs[] = {
2114         &rdev_state.attr,
2115         &rdev_errors.attr,
2116         &rdev_slot.attr,
2117         &rdev_offset.attr,
2118         &rdev_size.attr,
2119         NULL,
2120 };
2121 static ssize_t
2122 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2123 {
2124         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2125         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2126         mddev_t *mddev = rdev->mddev;
2127         ssize_t rv;
2128
2129         if (!entry->show)
2130                 return -EIO;
2131
2132         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2133         if (!rv) {
2134                 if (rdev->mddev == NULL)
2135                         rv = -EBUSY;
2136                 else
2137                         rv = entry->show(rdev, page);
2138                 mddev_unlock(mddev);
2139         }
2140         return rv;
2141 }
2142
2143 static ssize_t
2144 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2145               const char *page, size_t length)
2146 {
2147         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2148         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2149         ssize_t rv;
2150         mddev_t *mddev = rdev->mddev;
2151
2152         if (!entry->store)
2153                 return -EIO;
2154         if (!capable(CAP_SYS_ADMIN))
2155                 return -EACCES;
2156         rv = mddev ? mddev_lock(mddev): -EBUSY;
2157         if (!rv) {
2158                 if (rdev->mddev == NULL)
2159                         rv = -EBUSY;
2160                 else
2161                         rv = entry->store(rdev, page, length);
2162                 mddev_unlock(mddev);
2163         }
2164         return rv;
2165 }
2166
2167 static void rdev_free(struct kobject *ko)
2168 {
2169         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2170         kfree(rdev);
2171 }
2172 static struct sysfs_ops rdev_sysfs_ops = {
2173         .show           = rdev_attr_show,
2174         .store          = rdev_attr_store,
2175 };
2176 static struct kobj_type rdev_ktype = {
2177         .release        = rdev_free,
2178         .sysfs_ops      = &rdev_sysfs_ops,
2179         .default_attrs  = rdev_default_attrs,
2180 };
2181
2182 /*
2183  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2184  *
2185  * mark the device faulty if:
2186  *
2187  *   - the device is nonexistent (zero size)
2188  *   - the device has no valid superblock
2189  *
2190  * a faulty rdev _never_ has rdev->sb set.
2191  */
2192 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2193 {
2194         char b[BDEVNAME_SIZE];
2195         int err;
2196         mdk_rdev_t *rdev;
2197         sector_t size;
2198
2199         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2200         if (!rdev) {
2201                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2202                 return ERR_PTR(-ENOMEM);
2203         }
2204
2205         if ((err = alloc_disk_sb(rdev)))
2206                 goto abort_free;
2207
2208         err = lock_rdev(rdev, newdev, super_format == -2);
2209         if (err)
2210                 goto abort_free;
2211
2212         kobject_init(&rdev->kobj, &rdev_ktype);
2213
2214         rdev->desc_nr = -1;
2215         rdev->saved_raid_disk = -1;
2216         rdev->raid_disk = -1;
2217         rdev->flags = 0;
2218         rdev->data_offset = 0;
2219         rdev->sb_events = 0;
2220         atomic_set(&rdev->nr_pending, 0);
2221         atomic_set(&rdev->read_errors, 0);
2222         atomic_set(&rdev->corrected_errors, 0);
2223
2224         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2225         if (!size) {
2226                 printk(KERN_WARNING
2227                         "md: %s has zero or unknown size, marking faulty!\n",
2228                         bdevname(rdev->bdev,b));
2229                 err = -EINVAL;
2230                 goto abort_free;
2231         }
2232
2233         if (super_format >= 0) {
2234                 err = super_types[super_format].
2235                         load_super(rdev, NULL, super_minor);
2236                 if (err == -EINVAL) {
2237                         printk(KERN_WARNING
2238                                 "md: %s does not have a valid v%d.%d "
2239                                "superblock, not importing!\n",
2240                                 bdevname(rdev->bdev,b),
2241                                super_format, super_minor);
2242                         goto abort_free;
2243                 }
2244                 if (err < 0) {
2245                         printk(KERN_WARNING
2246                                 "md: could not read %s's sb, not importing!\n",
2247                                 bdevname(rdev->bdev,b));
2248                         goto abort_free;
2249                 }
2250         }
2251
2252         INIT_LIST_HEAD(&rdev->same_set);
2253         init_waitqueue_head(&rdev->blocked_wait);
2254
2255         return rdev;
2256
2257 abort_free:
2258         if (rdev->sb_page) {
2259                 if (rdev->bdev)
2260                         unlock_rdev(rdev);
2261                 free_disk_sb(rdev);
2262         }
2263         kfree(rdev);
2264         return ERR_PTR(err);
2265 }
2266
2267 /*
2268  * Check a full RAID array for plausibility
2269  */
2270
2271
2272 static void analyze_sbs(mddev_t * mddev)
2273 {
2274         int i;
2275         struct list_head *tmp;
2276         mdk_rdev_t *rdev, *freshest;
2277         char b[BDEVNAME_SIZE];
2278
2279         freshest = NULL;
2280         rdev_for_each(rdev, tmp, mddev)
2281                 switch (super_types[mddev->major_version].
2282                         load_super(rdev, freshest, mddev->minor_version)) {
2283                 case 1:
2284                         freshest = rdev;
2285                         break;
2286                 case 0:
2287                         break;
2288                 default:
2289                         printk( KERN_ERR \
2290                                 "md: fatal superblock inconsistency in %s"
2291                                 " -- removing from array\n",
2292                                 bdevname(rdev->bdev,b));
2293                         kick_rdev_from_array(rdev);
2294                 }
2295
2296
2297         super_types[mddev->major_version].
2298                 validate_super(mddev, freshest);
2299
2300         i = 0;
2301         rdev_for_each(rdev, tmp, mddev) {
2302                 if (rdev != freshest)
2303                         if (super_types[mddev->major_version].
2304                             validate_super(mddev, rdev)) {
2305                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2306                                         " from array!\n",
2307                                         bdevname(rdev->bdev,b));
2308                                 kick_rdev_from_array(rdev);
2309                                 continue;
2310                         }
2311                 if (mddev->level == LEVEL_MULTIPATH) {
2312                         rdev->desc_nr = i++;
2313                         rdev->raid_disk = rdev->desc_nr;
2314                         set_bit(In_sync, &rdev->flags);
2315                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2316                         rdev->raid_disk = -1;
2317                         clear_bit(In_sync, &rdev->flags);
2318                 }
2319         }
2320
2321
2322
2323         if (mddev->recovery_cp != MaxSector &&
2324             mddev->level >= 1)
2325                 printk(KERN_ERR "md: %s: raid array is not clean"
2326                        " -- starting background reconstruction\n",
2327                        mdname(mddev));
2328
2329 }
2330
2331 static ssize_t
2332 safe_delay_show(mddev_t *mddev, char *page)
2333 {
2334         int msec = (mddev->safemode_delay*1000)/HZ;
2335         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2336 }
2337 static ssize_t
2338 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2339 {
2340         int scale=1;
2341         int dot=0;
2342         int i;
2343         unsigned long msec;
2344         char buf[30];
2345         char *e;
2346         /* remove a period, and count digits after it */
2347         if (len >= sizeof(buf))
2348                 return -EINVAL;
2349         strlcpy(buf, cbuf, len);
2350         buf[len] = 0;
2351         for (i=0; i<len; i++) {
2352                 if (dot) {
2353                         if (isdigit(buf[i])) {
2354                                 buf[i-1] = buf[i];
2355                                 scale *= 10;
2356                         }
2357                         buf[i] = 0;
2358                 } else if (buf[i] == '.') {
2359                         dot=1;
2360                         buf[i] = 0;
2361                 }
2362         }
2363         msec = simple_strtoul(buf, &e, 10);
2364         if (e == buf || (*e && *e != '\n'))
2365                 return -EINVAL;
2366         msec = (msec * 1000) / scale;
2367         if (msec == 0)
2368                 mddev->safemode_delay = 0;
2369         else {
2370                 mddev->safemode_delay = (msec*HZ)/1000;
2371                 if (mddev->safemode_delay == 0)
2372                         mddev->safemode_delay = 1;
2373         }
2374         return len;
2375 }
2376 static struct md_sysfs_entry md_safe_delay =
2377 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2378
2379 static ssize_t
2380 level_show(mddev_t *mddev, char *page)
2381 {
2382         struct mdk_personality *p = mddev->pers;
2383         if (p)
2384                 return sprintf(page, "%s\n", p->name);
2385         else if (mddev->clevel[0])
2386                 return sprintf(page, "%s\n", mddev->clevel);
2387         else if (mddev->level != LEVEL_NONE)
2388                 return sprintf(page, "%d\n", mddev->level);
2389         else
2390                 return 0;
2391 }
2392
2393 static ssize_t
2394 level_store(mddev_t *mddev, const char *buf, size_t len)
2395 {
2396         ssize_t rv = len;
2397         if (mddev->pers)
2398                 return -EBUSY;
2399         if (len == 0)
2400                 return 0;
2401         if (len >= sizeof(mddev->clevel))
2402                 return -ENOSPC;
2403         strncpy(mddev->clevel, buf, len);
2404         if (mddev->clevel[len-1] == '\n')
2405                 len--;
2406         mddev->clevel[len] = 0;
2407         mddev->level = LEVEL_NONE;
2408         return rv;
2409 }
2410
2411 static struct md_sysfs_entry md_level =
2412 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2413
2414
2415 static ssize_t
2416 layout_show(mddev_t *mddev, char *page)
2417 {
2418         /* just a number, not meaningful for all levels */
2419         if (mddev->reshape_position != MaxSector &&
2420             mddev->layout != mddev->new_layout)
2421                 return sprintf(page, "%d (%d)\n",
2422                                mddev->new_layout, mddev->layout);
2423         return sprintf(page, "%d\n", mddev->layout);
2424 }
2425
2426 static ssize_t
2427 layout_store(mddev_t *mddev, const char *buf, size_t len)
2428 {
2429         char *e;
2430         unsigned long n = simple_strtoul(buf, &e, 10);
2431
2432         if (!*buf || (*e && *e != '\n'))
2433                 return -EINVAL;
2434
2435         if (mddev->pers)
2436                 return -EBUSY;
2437         if (mddev->reshape_position != MaxSector)
2438                 mddev->new_layout = n;
2439         else
2440                 mddev->layout = n;
2441         return len;
2442 }
2443 static struct md_sysfs_entry md_layout =
2444 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2445
2446
2447 static ssize_t
2448 raid_disks_show(mddev_t *mddev, char *page)
2449 {
2450         if (mddev->raid_disks == 0)
2451                 return 0;
2452         if (mddev->reshape_position != MaxSector &&
2453             mddev->delta_disks != 0)
2454                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2455                                mddev->raid_disks - mddev->delta_disks);
2456         return sprintf(page, "%d\n", mddev->raid_disks);
2457 }
2458
2459 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2460
2461 static ssize_t
2462 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2463 {
2464         char *e;
2465         int rv = 0;
2466         unsigned long n = simple_strtoul(buf, &e, 10);
2467
2468         if (!*buf || (*e && *e != '\n'))
2469                 return -EINVAL;
2470
2471         if (mddev->pers)
2472                 rv = update_raid_disks(mddev, n);
2473         else if (mddev->reshape_position != MaxSector) {
2474                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2475                 mddev->delta_disks = n - olddisks;
2476                 mddev->raid_disks = n;
2477         } else
2478                 mddev->raid_disks = n;
2479         return rv ? rv : len;
2480 }
2481 static struct md_sysfs_entry md_raid_disks =
2482 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2483
2484 static ssize_t
2485 chunk_size_show(mddev_t *mddev, char *page)
2486 {
2487         if (mddev->reshape_position != MaxSector &&
2488             mddev->chunk_size != mddev->new_chunk)
2489                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2490                                mddev->chunk_size);
2491         return sprintf(page, "%d\n", mddev->chunk_size);
2492 }
2493
2494 static ssize_t
2495 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2496 {
2497         /* can only set chunk_size if array is not yet active */
2498         char *e;
2499         unsigned long n = simple_strtoul(buf, &e, 10);
2500
2501         if (!*buf || (*e && *e != '\n'))
2502                 return -EINVAL;
2503
2504         if (mddev->pers)
2505                 return -EBUSY;
2506         else if (mddev->reshape_position != MaxSector)
2507                 mddev->new_chunk = n;
2508         else
2509                 mddev->chunk_size = n;
2510         return len;
2511 }
2512 static struct md_sysfs_entry md_chunk_size =
2513 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2514
2515 static ssize_t
2516 resync_start_show(mddev_t *mddev, char *page)
2517 {
2518         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2519 }
2520
2521 static ssize_t
2522 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2523 {
2524         char *e;
2525         unsigned long long n = simple_strtoull(buf, &e, 10);
2526
2527         if (mddev->pers)
2528                 return -EBUSY;
2529         if (!*buf || (*e && *e != '\n'))
2530                 return -EINVAL;
2531
2532         mddev->recovery_cp = n;
2533         return len;
2534 }
2535 static struct md_sysfs_entry md_resync_start =
2536 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2537
2538 /*
2539  * The array state can be:
2540  *
2541  * clear
2542  *     No devices, no size, no level
2543  *     Equivalent to STOP_ARRAY ioctl
2544  * inactive
2545  *     May have some settings, but array is not active
2546  *        all IO results in error
2547  *     When written, doesn't tear down array, but just stops it
2548  * suspended (not supported yet)
2549  *     All IO requests will block. The array can be reconfigured.
2550  *     Writing this, if accepted, will block until array is quiessent
2551  * readonly
2552  *     no resync can happen.  no superblocks get written.
2553  *     write requests fail
2554  * read-auto
2555  *     like readonly, but behaves like 'clean' on a write request.
2556  *
2557  * clean - no pending writes, but otherwise active.
2558  *     When written to inactive array, starts without resync
2559  *     If a write request arrives then
2560  *       if metadata is known, mark 'dirty' and switch to 'active'.
2561  *       if not known, block and switch to write-pending
2562  *     If written to an active array that has pending writes, then fails.
2563  * active
2564  *     fully active: IO and resync can be happening.
2565  *     When written to inactive array, starts with resync
2566  *
2567  * write-pending
2568  *     clean, but writes are blocked waiting for 'active' to be written.
2569  *
2570  * active-idle
2571  *     like active, but no writes have been seen for a while (100msec).
2572  *
2573  */
2574 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2575                    write_pending, active_idle, bad_word};
2576 static char *array_states[] = {
2577         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2578         "write-pending", "active-idle", NULL };
2579
2580 static int match_word(const char *word, char **list)
2581 {
2582         int n;
2583         for (n=0; list[n]; n++)
2584                 if (cmd_match(word, list[n]))
2585                         break;
2586         return n;
2587 }
2588
2589 static ssize_t
2590 array_state_show(mddev_t *mddev, char *page)
2591 {
2592         enum array_state st = inactive;
2593
2594         if (mddev->pers)
2595                 switch(mddev->ro) {
2596                 case 1:
2597                         st = readonly;
2598                         break;
2599                 case 2:
2600                         st = read_auto;
2601                         break;
2602                 case 0:
2603                         if (mddev->in_sync)
2604                                 st = clean;
2605                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2606                                 st = write_pending;
2607                         else if (mddev->safemode)
2608                                 st = active_idle;
2609                         else
2610                                 st = active;
2611                 }
2612         else {
2613                 if (list_empty(&mddev->disks) &&
2614                     mddev->raid_disks == 0 &&
2615                     mddev->size == 0)
2616                         st = clear;
2617                 else
2618                         st = inactive;
2619         }
2620         return sprintf(page, "%s\n", array_states[st]);
2621 }
2622
2623 static int do_md_stop(mddev_t * mddev, int ro);
2624 static int do_md_run(mddev_t * mddev);
2625 static int restart_array(mddev_t *mddev);
2626
2627 static ssize_t
2628 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2629 {
2630         int err = -EINVAL;
2631         enum array_state st = match_word(buf, array_states);
2632         switch(st) {
2633         case bad_word:
2634                 break;
2635         case clear:
2636                 /* stopping an active array */
2637                 if (atomic_read(&mddev->active) > 1)
2638                         return -EBUSY;
2639                 err = do_md_stop(mddev, 0);
2640                 break;
2641         case inactive:
2642                 /* stopping an active array */
2643                 if (mddev->pers) {
2644                         if (atomic_read(&mddev->active) > 1)
2645                                 return -EBUSY;
2646                         err = do_md_stop(mddev, 2);
2647                 } else
2648                         err = 0; /* already inactive */
2649                 break;
2650         case suspended:
2651                 break; /* not supported yet */
2652         case readonly:
2653                 if (mddev->pers)
2654                         err = do_md_stop(mddev, 1);
2655                 else {
2656                         mddev->ro = 1;
2657                         set_disk_ro(mddev->gendisk, 1);
2658                         err = do_md_run(mddev);
2659                 }
2660                 break;
2661         case read_auto:
2662                 if (mddev->pers) {
2663                         if (mddev->ro != 1)
2664                                 err = do_md_stop(mddev, 1);
2665                         else
2666                                 err = restart_array(mddev);
2667                         if (err == 0) {
2668                                 mddev->ro = 2;
2669                                 set_disk_ro(mddev->gendisk, 0);
2670                         }
2671                 } else {
2672                         mddev->ro = 2;
2673                         err = do_md_run(mddev);
2674                 }
2675                 break;
2676         case clean:
2677                 if (mddev->pers) {
2678                         restart_array(mddev);
2679                         spin_lock_irq(&mddev->write_lock);
2680                         if (atomic_read(&mddev->writes_pending) == 0) {
2681                                 if (mddev->in_sync == 0) {
2682                                         mddev->in_sync = 1;
2683                                         if (mddev->safemode == 1)
2684                                                 mddev->safemode = 0;
2685                                         if (mddev->persistent)
2686                                                 set_bit(MD_CHANGE_CLEAN,
2687                                                         &mddev->flags);
2688                                 }
2689                                 err = 0;
2690                         } else
2691                                 err = -EBUSY;
2692                         spin_unlock_irq(&mddev->write_lock);
2693                 } else {
2694                         mddev->ro = 0;
2695                         mddev->recovery_cp = MaxSector;
2696                         err = do_md_run(mddev);
2697                 }
2698                 break;
2699         case active:
2700                 if (mddev->pers) {
2701                         restart_array(mddev);
2702                         if (mddev->external)
2703                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2704                         wake_up(&mddev->sb_wait);
2705                         err = 0;
2706                 } else {
2707                         mddev->ro = 0;
2708                         set_disk_ro(mddev->gendisk, 0);
2709                         err = do_md_run(mddev);
2710                 }
2711                 break;
2712         case write_pending:
2713         case active_idle:
2714                 /* these cannot be set */
2715                 break;
2716         }
2717         if (err)
2718                 return err;
2719         else
2720                 return len;
2721 }
2722 static struct md_sysfs_entry md_array_state =
2723 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2724
2725 static ssize_t
2726 null_show(mddev_t *mddev, char *page)
2727 {
2728         return -EINVAL;
2729 }
2730
2731 static ssize_t
2732 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2733 {
2734         /* buf must be %d:%d\n? giving major and minor numbers */
2735         /* The new device is added to the array.
2736          * If the array has a persistent superblock, we read the
2737          * superblock to initialise info and check validity.
2738          * Otherwise, only checking done is that in bind_rdev_to_array,
2739          * which mainly checks size.
2740          */
2741         char *e;
2742         int major = simple_strtoul(buf, &e, 10);
2743         int minor;
2744         dev_t dev;
2745         mdk_rdev_t *rdev;
2746         int err;
2747
2748         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2749                 return -EINVAL;
2750         minor = simple_strtoul(e+1, &e, 10);
2751         if (*e && *e != '\n')
2752                 return -EINVAL;
2753         dev = MKDEV(major, minor);
2754         if (major != MAJOR(dev) ||
2755             minor != MINOR(dev))
2756                 return -EOVERFLOW;
2757
2758
2759         if (mddev->persistent) {
2760                 rdev = md_import_device(dev, mddev->major_version,
2761                                         mddev->minor_version);
2762                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2763                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2764                                                        mdk_rdev_t, same_set);
2765                         err = super_types[mddev->major_version]
2766                                 .load_super(rdev, rdev0, mddev->minor_version);
2767                         if (err < 0)
2768                                 goto out;
2769                 }
2770         } else if (mddev->external)
2771                 rdev = md_import_device(dev, -2, -1);
2772         else
2773                 rdev = md_import_device(dev, -1, -1);
2774
2775         if (IS_ERR(rdev))
2776                 return PTR_ERR(rdev);
2777         err = bind_rdev_to_array(rdev, mddev);
2778  out:
2779         if (err)
2780                 export_rdev(rdev);
2781         return err ? err : len;
2782 }
2783
2784 static struct md_sysfs_entry md_new_device =
2785 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2786
2787 static ssize_t
2788 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2789 {
2790         char *end;
2791         unsigned long chunk, end_chunk;
2792
2793         if (!mddev->bitmap)
2794                 goto out;
2795         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2796         while (*buf) {
2797                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2798                 if (buf == end) break;
2799                 if (*end == '-') { /* range */
2800                         buf = end + 1;
2801                         end_chunk = simple_strtoul(buf, &end, 0);
2802                         if (buf == end) break;
2803                 }
2804                 if (*end && !isspace(*end)) break;
2805                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2806                 buf = end;
2807                 while (isspace(*buf)) buf++;
2808         }
2809         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2810 out:
2811         return len;
2812 }
2813
2814 static struct md_sysfs_entry md_bitmap =
2815 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2816
2817 static ssize_t
2818 size_show(mddev_t *mddev, char *page)
2819 {
2820         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2821 }
2822
2823 static int update_size(mddev_t *mddev, unsigned long size);
2824
2825 static ssize_t
2826 size_store(mddev_t *mddev, const char *buf, size_t len)
2827 {
2828         /* If array is inactive, we can reduce the component size, but
2829          * not increase it (except from 0).
2830          * If array is active, we can try an on-line resize
2831          */
2832         char *e;
2833         int err = 0;
2834         unsigned long long size = simple_strtoull(buf, &e, 10);
2835         if (!*buf || *buf == '\n' ||
2836             (*e && *e != '\n'))
2837                 return -EINVAL;
2838
2839         if (mddev->pers) {
2840                 err = update_size(mddev, size);
2841                 md_update_sb(mddev, 1);
2842         } else {
2843                 if (mddev->size == 0 ||
2844                     mddev->size > size)
2845                         mddev->size = size;
2846                 else
2847                         err = -ENOSPC;
2848         }
2849         return err ? err : len;
2850 }
2851
2852 static struct md_sysfs_entry md_size =
2853 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2854
2855
2856 /* Metdata version.
2857  * This is one of
2858  *   'none' for arrays with no metadata (good luck...)
2859  *   'external' for arrays with externally managed metadata,
2860  * or N.M for internally known formats
2861  */
2862 static ssize_t
2863 metadata_show(mddev_t *mddev, char *page)
2864 {
2865         if (mddev->persistent)
2866                 return sprintf(page, "%d.%d\n",
2867                                mddev->major_version, mddev->minor_version);
2868         else if (mddev->external)
2869                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2870         else
2871                 return sprintf(page, "none\n");
2872 }
2873
2874 static ssize_t
2875 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2876 {
2877         int major, minor;
2878         char *e;
2879         if (!list_empty(&mddev->disks))
2880                 return -EBUSY;
2881
2882         if (cmd_match(buf, "none")) {
2883                 mddev->persistent = 0;
2884                 mddev->external = 0;
2885                 mddev->major_version = 0;
2886                 mddev->minor_version = 90;
2887                 return len;
2888         }
2889         if (strncmp(buf, "external:", 9) == 0) {
2890                 size_t namelen = len-9;
2891                 if (namelen >= sizeof(mddev->metadata_type))
2892                         namelen = sizeof(mddev->metadata_type)-1;
2893                 strncpy(mddev->metadata_type, buf+9, namelen);
2894                 mddev->metadata_type[namelen] = 0;
2895                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2896                         mddev->metadata_type[--namelen] = 0;
2897                 mddev->persistent = 0;
2898                 mddev->external = 1;
2899                 mddev->major_version = 0;
2900                 mddev->minor_version = 90;
2901                 return len;
2902         }
2903         major = simple_strtoul(buf, &e, 10);
2904         if (e==buf || *e != '.')
2905                 return -EINVAL;
2906         buf = e+1;
2907         minor = simple_strtoul(buf, &e, 10);
2908         if (e==buf || (*e && *e != '\n') )
2909                 return -EINVAL;
2910         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2911                 return -ENOENT;
2912         mddev->major_version = major;
2913         mddev->minor_version = minor;
2914         mddev->persistent = 1;
2915         mddev->external = 0;
2916         return len;
2917 }
2918
2919 static struct md_sysfs_entry md_metadata =
2920 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2921
2922 static ssize_t
2923 action_show(mddev_t *mddev, char *page)
2924 {
2925         char *type = "idle";
2926         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2927             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2928                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2929                         type = "reshape";
2930                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2931                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2932                                 type = "resync";
2933                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2934                                 type = "check";
2935                         else
2936                                 type = "repair";
2937                 } else
2938                         type = "recover";
2939         }
2940         return sprintf(page, "%s\n", type);
2941 }
2942
2943 static ssize_t
2944 action_store(mddev_t *mddev, const char *page, size_t len)
2945 {
2946         if (!mddev->pers || !mddev->pers->sync_request)
2947                 return -EINVAL;
2948
2949         if (cmd_match(page, "idle")) {
2950                 if (mddev->sync_thread) {
2951                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2952                         md_unregister_thread(mddev->sync_thread);
2953                         mddev->sync_thread = NULL;
2954                         mddev->recovery = 0;
2955                 }
2956         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2957                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2958                 return -EBUSY;
2959         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2960                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2961         else if (cmd_match(page, "reshape")) {
2962                 int err;
2963                 if (mddev->pers->start_reshape == NULL)
2964                         return -EINVAL;
2965                 err = mddev->pers->start_reshape(mddev);
2966                 if (err)
2967                         return err;
2968         } else {
2969                 if (cmd_match(page, "check"))
2970                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2971                 else if (!cmd_match(page, "repair"))
2972                         return -EINVAL;
2973                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2974                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2975         }
2976         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2977         md_wakeup_thread(mddev->thread);
2978         return len;
2979 }
2980
2981 static ssize_t
2982 mismatch_cnt_show(mddev_t *mddev, char *page)
2983 {
2984         return sprintf(page, "%llu\n",
2985                        (unsigned long long) mddev->resync_mismatches);
2986 }
2987
2988 static struct md_sysfs_entry md_scan_mode =
2989 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2990
2991
2992 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2993
2994 static ssize_t
2995 sync_min_show(mddev_t *mddev, char *page)
2996 {
2997         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2998                        mddev->sync_speed_min ? "local": "system");
2999 }
3000
3001 static ssize_t
3002 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3003 {
3004         int min;
3005         char *e;
3006         if (strncmp(buf, "system", 6)==0) {
3007                 mddev->sync_speed_min = 0;
3008                 return len;
3009         }
3010         min = simple_strtoul(buf, &e, 10);
3011         if (buf == e || (*e && *e != '\n') || min <= 0)
3012                 return -EINVAL;
3013         mddev->sync_speed_min = min;
3014         return len;
3015 }
3016
3017 static struct md_sysfs_entry md_sync_min =
3018 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3019
3020 static ssize_t
3021 sync_max_show(mddev_t *mddev, char *page)
3022 {
3023         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3024                        mddev->sync_speed_max ? "local": "system");
3025 }
3026
3027 static ssize_t
3028 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3029 {
3030         int max;
3031         char *e;
3032         if (strncmp(buf, "system", 6)==0) {
3033                 mddev->sync_speed_max = 0;
3034                 return len;
3035         }
3036         max = simple_strtoul(buf, &e, 10);
3037         if (buf == e || (*e && *e != '\n') || max <= 0)
3038                 return -EINVAL;
3039         mddev->sync_speed_max = max;
3040         return len;
3041 }
3042
3043 static struct md_sysfs_entry md_sync_max =
3044 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3045
3046 static ssize_t
3047 degraded_show(mddev_t *mddev, char *page)
3048 {
3049         return sprintf(page, "%d\n", mddev->degraded);
3050 }
3051 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3052
3053 static ssize_t
3054 sync_force_parallel_show(mddev_t *mddev, char *page)
3055 {
3056         return sprintf(page, "%d\n", mddev->parallel_resync);
3057 }
3058
3059 static ssize_t
3060 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3061 {
3062         long n;
3063
3064         if (strict_strtol(buf, 10, &n))
3065                 return -EINVAL;
3066
3067         if (n != 0 && n != 1)
3068                 return -EINVAL;
3069
3070         mddev->parallel_resync = n;
3071
3072         if (mddev->sync_thread)
3073                 wake_up(&resync_wait);
3074
3075         return len;
3076 }
3077
3078 /* force parallel resync, even with shared block devices */
3079 static struct md_sysfs_entry md_sync_force_parallel =
3080 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3081        sync_force_parallel_show, sync_force_parallel_store);
3082
3083 static ssize_t
3084 sync_speed_show(mddev_t *mddev, char *page)
3085 {
3086         unsigned long resync, dt, db;
3087         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
3088         dt = ((jiffies - mddev->resync_mark) / HZ);
3089         if (!dt) dt++;
3090         db = resync - (mddev->resync_mark_cnt);
3091         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
3092 }
3093
3094 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3095
3096 static ssize_t
3097 sync_completed_show(mddev_t *mddev, char *page)
3098 {
3099         unsigned long max_blocks, resync;
3100
3101         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3102                 max_blocks = mddev->resync_max_sectors;
3103         else
3104                 max_blocks = mddev->size << 1;
3105
3106         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3107         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3108 }
3109
3110 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3111
3112 static ssize_t
3113 min_sync_show(mddev_t *mddev, char *page)
3114 {
3115         return sprintf(page, "%llu\n",
3116                        (unsigned long long)mddev->resync_min);
3117 }
3118 static ssize_t
3119 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3120 {
3121         unsigned long long min;
3122         if (strict_strtoull(buf, 10, &min))
3123                 return -EINVAL;
3124         if (min > mddev->resync_max)
3125                 return -EINVAL;
3126         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3127                 return -EBUSY;
3128
3129         /* Must be a multiple of chunk_size */
3130         if (mddev->chunk_size) {
3131                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3132                         return -EINVAL;
3133         }
3134         mddev->resync_min = min;
3135
3136         return len;
3137 }
3138
3139 static struct md_sysfs_entry md_min_sync =
3140 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3141
3142 static ssize_t
3143 max_sync_show(mddev_t *mddev, char *page)
3144 {
3145         if (mddev->resync_max == MaxSector)
3146                 return sprintf(page, "max\n");
3147         else
3148                 return sprintf(page, "%llu\n",
3149                                (unsigned long long)mddev->resync_max);
3150 }
3151 static ssize_t
3152 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3153 {
3154         if (strncmp(buf, "max", 3) == 0)
3155                 mddev->resync_max = MaxSector;
3156         else {
3157                 unsigned long long max;
3158                 if (strict_strtoull(buf, 10, &max))
3159                         return -EINVAL;
3160                 if (max < mddev->resync_min)
3161                         return -EINVAL;
3162                 if (max < mddev->resync_max &&
3163                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3164                         return -EBUSY;
3165
3166                 /* Must be a multiple of chunk_size */
3167                 if (mddev->chunk_size) {
3168                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3169                                 return -EINVAL;
3170                 }
3171                 mddev->resync_max = max;
3172         }
3173         wake_up(&mddev->recovery_wait);
3174         return len;
3175 }
3176
3177 static struct md_sysfs_entry md_max_sync =
3178 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3179
3180 static ssize_t
3181 suspend_lo_show(mddev_t *mddev, char *page)
3182 {
3183         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3184 }
3185
3186 static ssize_t
3187 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3188 {
3189         char *e;
3190         unsigned long long new = simple_strtoull(buf, &e, 10);
3191
3192         if (mddev->pers->quiesce == NULL)
3193                 return -EINVAL;
3194         if (buf == e || (*e && *e != '\n'))
3195                 return -EINVAL;
3196         if (new >= mddev->suspend_hi ||
3197             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3198                 mddev->suspend_lo = new;
3199                 mddev->pers->quiesce(mddev, 2);
3200                 return len;
3201         } else
3202                 return -EINVAL;
3203 }
3204 static struct md_sysfs_entry md_suspend_lo =
3205 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3206
3207
3208 static ssize_t
3209 suspend_hi_show(mddev_t *mddev, char *page)
3210 {
3211         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3212 }
3213
3214 static ssize_t
3215 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3216 {
3217         char *e;
3218         unsigned long long new = simple_strtoull(buf, &e, 10);
3219
3220         if (mddev->pers->quiesce == NULL)
3221                 return -EINVAL;
3222         if (buf == e || (*e && *e != '\n'))
3223                 return -EINVAL;
3224         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3225             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3226                 mddev->suspend_hi = new;
3227                 mddev->pers->quiesce(mddev, 1);
3228                 mddev->pers->quiesce(mddev, 0);
3229                 return len;
3230         } else
3231                 return -EINVAL;
3232 }
3233 static struct md_sysfs_entry md_suspend_hi =
3234 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3235
3236 static ssize_t
3237 reshape_position_show(mddev_t *mddev, char *page)
3238 {
3239         if (mddev->reshape_position != MaxSector)
3240                 return sprintf(page, "%llu\n",
3241                                (unsigned long long)mddev->reshape_position);
3242         strcpy(page, "none\n");
3243         return 5;
3244 }
3245
3246 static ssize_t
3247 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3248 {
3249         char *e;
3250         unsigned long long new = simple_strtoull(buf, &e, 10);
3251         if (mddev->pers)
3252                 return -EBUSY;
3253         if (buf == e || (*e && *e != '\n'))
3254                 return -EINVAL;
3255         mddev->reshape_position = new;
3256         mddev->delta_disks = 0;
3257         mddev->new_level = mddev->level;
3258         mddev->new_layout = mddev->layout;
3259         mddev->new_chunk = mddev->chunk_size;
3260         return len;
3261 }
3262
3263 static struct md_sysfs_entry md_reshape_position =
3264 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3265        reshape_position_store);
3266
3267
3268 static struct attribute *md_default_attrs[] = {
3269         &md_level.attr,
3270         &md_layout.attr,
3271         &md_raid_disks.attr,
3272         &md_chunk_size.attr,
3273         &md_size.attr,
3274         &md_resync_start.attr,
3275         &md_metadata.attr,
3276         &md_new_device.attr,
3277         &md_safe_delay.attr,
3278         &md_array_state.attr,
3279         &md_reshape_position.attr,
3280         NULL,
3281 };
3282
3283 static struct attribute *md_redundancy_attrs[] = {
3284         &md_scan_mode.attr,
3285         &md_mismatches.attr,
3286         &md_sync_min.attr,
3287         &md_sync_max.attr,
3288         &md_sync_speed.attr,
3289         &md_sync_force_parallel.attr,
3290         &md_sync_completed.attr,
3291         &md_min_sync.attr,
3292         &md_max_sync.attr,
3293         &md_suspend_lo.attr,
3294         &md_suspend_hi.attr,
3295         &md_bitmap.attr,
3296         &md_degraded.attr,
3297         NULL,
3298 };
3299 static struct attribute_group md_redundancy_group = {
3300         .name = NULL,
3301         .attrs = md_redundancy_attrs,
3302 };
3303
3304
3305 static ssize_t
3306 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3307 {
3308         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3309         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3310         ssize_t rv;
3311
3312         if (!entry->show)
3313                 return -EIO;
3314         rv = mddev_lock(mddev);
3315         if (!rv) {
3316                 rv = entry->show(mddev, page);
3317                 mddev_unlock(mddev);
3318         }
3319         return rv;
3320 }
3321
3322 static ssize_t
3323 md_attr_store(struct kobject *kobj, struct attribute *attr,
3324               const char *page, size_t length)
3325 {
3326         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3327         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3328         ssize_t rv;
3329
3330         if (!entry->store)
3331                 return -EIO;
3332         if (!capable(CAP_SYS_ADMIN))
3333                 return -EACCES;
3334         rv = mddev_lock(mddev);
3335         if (!rv) {
3336                 rv = entry->store(mddev, page, length);
3337                 mddev_unlock(mddev);
3338         }
3339         return rv;
3340 }
3341
3342 static void md_free(struct kobject *ko)
3343 {
3344         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3345         kfree(mddev);
3346 }
3347
3348 static struct sysfs_ops md_sysfs_ops = {
3349         .show   = md_attr_show,
3350         .store  = md_attr_store,
3351 };
3352 static struct kobj_type md_ktype = {
3353         .release        = md_free,
3354         .sysfs_ops      = &md_sysfs_ops,
3355         .default_attrs  = md_default_attrs,
3356 };
3357
3358 int mdp_major = 0;
3359
3360 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3361 {
3362         static DEFINE_MUTEX(disks_mutex);
3363         mddev_t *mddev = mddev_find(dev);
3364         struct gendisk *disk;
3365         int partitioned = (MAJOR(dev) != MD_MAJOR);
3366         int shift = partitioned ? MdpMinorShift : 0;
3367         int unit = MINOR(dev) >> shift;
3368         int error;
3369
3370         if (!mddev)
3371                 return NULL;
3372
3373         mutex_lock(&disks_mutex);
3374         if (mddev->gendisk) {
3375                 mutex_unlock(&disks_mutex);
3376                 mddev_put(mddev);
3377                 return NULL;
3378         }
3379         disk = alloc_disk(1 << shift);
3380         if (!disk) {
3381                 mutex_unlock(&disks_mutex);
3382                 mddev_put(mddev);
3383                 return NULL;
3384         }
3385         disk->major = MAJOR(dev);
3386         disk->first_minor = unit << shift;
3387         if (partitioned)
3388                 sprintf(disk->disk_name, "md_d%d", unit);
3389         else
3390                 sprintf(disk->disk_name, "md%d", unit);
3391         disk->fops = &md_fops;
3392         disk->private_data = mddev;
3393         disk->queue = mddev->queue;
3394         add_disk(disk);
3395         mddev->gendisk = disk;
3396         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3397                                      "%s", "md");
3398         mutex_unlock(&disks_mutex);
3399         if (error)
3400                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3401                        disk->disk_name);
3402         else
3403                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3404         return NULL;
3405 }
3406
3407 static void md_safemode_timeout(unsigned long data)
3408 {
3409         mddev_t *mddev = (mddev_t *) data;
3410
3411         mddev->safemode = 1;
3412         md_wakeup_thread(mddev->thread);
3413 }
3414
3415 static int start_dirty_degraded;
3416
3417 static int do_md_run(mddev_t * mddev)
3418 {
3419         int err;
3420         int chunk_size;
3421         struct list_head *tmp;
3422         mdk_rdev_t *rdev;
3423         struct gendisk *disk;
3424         struct mdk_personality *pers;
3425         char b[BDEVNAME_SIZE];
3426
3427         if (list_empty(&mddev->disks))
3428                 /* cannot run an array with no devices.. */
3429                 return -EINVAL;
3430
3431         if (mddev->pers)
3432                 return -EBUSY;
3433
3434         /*
3435          * Analyze all RAID superblock(s)
3436          */
3437         if (!mddev->raid_disks) {
3438                 if (!mddev->persistent)
3439                         return -EINVAL;
3440                 analyze_sbs(mddev);
3441         }
3442
3443         chunk_size = mddev->chunk_size;
3444
3445         if (chunk_size) {
3446                 if (chunk_size > MAX_CHUNK_SIZE) {
3447                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3448                                 chunk_size, MAX_CHUNK_SIZE);
3449                         return -EINVAL;
3450                 }
3451                 /*
3452                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3453                  */
3454                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3455                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3456                         return -EINVAL;
3457                 }
3458                 if (chunk_size < PAGE_SIZE) {
3459                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3460                                 chunk_size, PAGE_SIZE);
3461                         return -EINVAL;
3462                 }
3463
3464                 /* devices must have minimum size of one chunk */
3465                 rdev_for_each(rdev, tmp, mddev) {
3466                         if (test_bit(Faulty, &rdev->flags))
3467                                 continue;
3468                         if (rdev->size < chunk_size / 1024) {
3469                                 printk(KERN_WARNING
3470                                         "md: Dev %s smaller than chunk_size:"
3471                                         " %lluk < %dk\n",
3472                                         bdevname(rdev->bdev,b),
3473                                         (unsigned long long)rdev->size,
3474                                         chunk_size / 1024);
3475                                 return -EINVAL;
3476                         }
3477                 }
3478         }
3479
3480 #ifdef CONFIG_KMOD
3481         if (mddev->level != LEVEL_NONE)
3482                 request_module("md-level-%d", mddev->level);
3483         else if (mddev->clevel[0])
3484                 request_module("md-%s", mddev->clevel);
3485 #endif
3486
3487         /*
3488          * Drop all container device buffers, from now on
3489          * the only valid external interface is through the md
3490          * device.
3491          */
3492         rdev_for_each(rdev, tmp, mddev) {
3493                 if (test_bit(Faulty, &rdev->flags))
3494                         continue;
3495                 sync_blockdev(rdev->bdev);
3496                 invalidate_bdev(rdev->bdev);
3497
3498                 /* perform some consistency tests on the device.
3499                  * We don't want the data to overlap the metadata,
3500                  * Internal Bitmap issues has handled elsewhere.
3501                  */
3502                 if (rdev->data_offset < rdev->sb_offset) {
3503                         if (mddev->size &&
3504                             rdev->data_offset + mddev->size*2
3505                             > rdev->sb_offset*2) {
3506                                 printk("md: %s: data overlaps metadata\n",
3507                                        mdname(mddev));
3508                                 return -EINVAL;
3509                         }
3510                 } else {
3511                         if (rdev->sb_offset*2 + rdev->sb_size/512
3512                             > rdev->data_offset) {
3513                                 printk("md: %s: metadata overlaps data\n",
3514                                        mdname(mddev));
3515                                 return -EINVAL;
3516                         }
3517                 }
3518         }
3519
3520         md_probe(mddev->unit, NULL, NULL);
3521         disk = mddev->gendisk;
3522         if (!disk)
3523                 return -ENOMEM;
3524
3525         spin_lock(&pers_lock);
3526         pers = find_pers(mddev->level, mddev->clevel);
3527         if (!pers || !try_module_get(pers->owner)) {
3528                 spin_unlock(&pers_lock);
3529                 if (mddev->level != LEVEL_NONE)
3530                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3531                                mddev->level);
3532                 else
3533                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3534                                mddev->clevel);
3535                 return -EINVAL;
3536         }
3537         mddev->pers = pers;
3538         spin_unlock(&pers_lock);
3539         mddev->level = pers->level;
3540         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3541
3542         if (mddev->reshape_position != MaxSector &&
3543             pers->start_reshape == NULL) {
3544                 /* This personality cannot handle reshaping... */
3545                 mddev->pers = NULL;
3546                 module_put(pers->owner);
3547                 return -EINVAL;
3548         }
3549
3550         if (pers->sync_request) {
3551                 /* Warn if this is a potentially silly
3552                  * configuration.
3553                  */
3554                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3555                 mdk_rdev_t *rdev2;
3556                 struct list_head *tmp2;
3557                 int warned = 0;
3558                 rdev_for_each(rdev, tmp, mddev) {
3559                         rdev_for_each(rdev2, tmp2, mddev) {
3560                                 if (rdev < rdev2 &&
3561                                     rdev->bdev->bd_contains ==
3562                                     rdev2->bdev->bd_contains) {
3563                                         printk(KERN_WARNING
3564                                                "%s: WARNING: %s appears to be"
3565                                                " on the same physical disk as"
3566                                                " %s.\n",
3567                                                mdname(mddev),
3568                                                bdevname(rdev->bdev,b),
3569                                                bdevname(rdev2->bdev,b2));
3570                                         warned = 1;
3571                                 }
3572                         }
3573                 }
3574                 if (warned)
3575                         printk(KERN_WARNING
3576                                "True protection against single-disk"
3577                                " failure might be compromised.\n");
3578         }
3579
3580         mddev->recovery = 0;
3581         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3582         mddev->barriers_work = 1;
3583         mddev->ok_start_degraded = start_dirty_degraded;
3584
3585         if (start_readonly)
3586                 mddev->ro = 2; /* read-only, but switch on first write */
3587
3588         err = mddev->pers->run(mddev);
3589         if (!err && mddev->pers->sync_request) {
3590                 err = bitmap_create(mddev);
3591                 if (err) {
3592                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3593                                mdname(mddev), err);
3594                         mddev->pers->stop(mddev);
3595                 }
3596         }
3597         if (err) {
3598                 printk(KERN_ERR "md: pers->run() failed ...\n");
3599                 module_put(mddev->pers->owner);
3600                 mddev->pers = NULL;
3601                 bitmap_destroy(mddev);
3602                 return err;
3603         }
3604         if (mddev->pers->sync_request) {
3605                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3606                         printk(KERN_WARNING
3607                                "md: cannot register extra attributes for %s\n",
3608                                mdname(mddev));
3609         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3610                 mddev->ro = 0;
3611
3612         atomic_set(&mddev->writes_pending,0);
3613         mddev->safemode = 0;
3614         mddev->safemode_timer.function = md_safemode_timeout;
3615         mddev->safemode_timer.data = (unsigned long) mddev;
3616         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3617         mddev->in_sync = 1;
3618
3619         rdev_for_each(rdev, tmp, mddev)
3620                 if (rdev->raid_disk >= 0) {
3621                         char nm[20];
3622                         sprintf(nm, "rd%d", rdev->raid_disk);
3623                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3624                                 printk("md: cannot register %s for %s\n",
3625                                        nm, mdname(mddev));
3626                 }
3627
3628         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3629
3630         if (mddev->flags)
3631                 md_update_sb(mddev, 0);
3632
3633         set_capacity(disk, mddev->array_size<<1);
3634
3635         /* If we call blk_queue_make_request here, it will
3636          * re-initialise max_sectors etc which may have been
3637          * refined inside -> run.  So just set the bits we need to set.
3638          * Most initialisation happended when we called
3639          * blk_queue_make_request(..., md_fail_request)
3640          * earlier.
3641          */
3642         mddev->queue->queuedata = mddev;
3643         mddev->queue->make_request_fn = mddev->pers->make_request;
3644
3645         /* If there is a partially-recovered drive we need to
3646          * start recovery here.  If we leave it to md_check_recovery,
3647          * it will remove the drives and not do the right thing
3648          */
3649         if (mddev->degraded && !mddev->sync_thread) {
3650                 struct list_head *rtmp;
3651                 int spares = 0;
3652                 rdev_for_each(rdev, rtmp, mddev)
3653                         if (rdev->raid_disk >= 0 &&
3654                             !test_bit(In_sync, &rdev->flags) &&
3655                             !test_bit(Faulty, &rdev->flags))
3656                                 /* complete an interrupted recovery */
3657                                 spares++;
3658                 if (spares && mddev->pers->sync_request) {
3659                         mddev->recovery = 0;
3660                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3661                         mddev->sync_thread = md_register_thread(md_do_sync,
3662                                                                 mddev,
3663                                                                 "%s_resync");
3664                         if (!mddev->sync_thread) {
3665                                 printk(KERN_ERR "%s: could not start resync"
3666                                        " thread...\n",
3667                                        mdname(mddev));
3668                                 /* leave the spares where they are, it shouldn't hurt */
3669                                 mddev->recovery = 0;
3670                         }
3671                 }
3672         }
3673         md_wakeup_thread(mddev->thread);
3674         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3675
3676         mddev->changed = 1;
3677         md_new_event(mddev);
3678         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3679         return 0;
3680 }
3681
3682 static int restart_array(mddev_t *mddev)
3683 {
3684         struct gendisk *disk = mddev->gendisk;
3685         int err;
3686
3687         /*
3688          * Complain if it has no devices
3689          */
3690         err = -ENXIO;
3691         if (list_empty(&mddev->disks))
3692                 goto out;
3693
3694         if (mddev->pers) {
3695                 err = -EBUSY;
3696                 if (!mddev->ro)
3697                         goto out;
3698
3699                 mddev->safemode = 0;
3700                 mddev->ro = 0;
3701                 set_disk_ro(disk, 0);
3702
3703                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3704                         mdname(mddev));
3705                 /*
3706                  * Kick recovery or resync if necessary
3707                  */
3708                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3709                 md_wakeup_thread(mddev->thread);
3710                 md_wakeup_thread(mddev->sync_thread);
3711                 err = 0;
3712         } else
3713                 err = -EINVAL;
3714
3715 out:
3716         return err;
3717 }
3718
3719 /* similar to deny_write_access, but accounts for our holding a reference
3720  * to the file ourselves */
3721 static int deny_bitmap_write_access(struct file * file)
3722 {
3723         struct inode *inode = file->f_mapping->host;
3724
3725         spin_lock(&inode->i_lock);
3726         if (atomic_read(&inode->i_writecount) > 1) {
3727                 spin_unlock(&inode->i_lock);
3728                 return -ETXTBSY;
3729         }
3730         atomic_set(&inode->i_writecount, -1);
3731         spin_unlock(&inode->i_lock);
3732
3733         return 0;
3734 }
3735
3736 static void restore_bitmap_write_access(struct file *file)
3737 {
3738         struct inode *inode = file->f_mapping->host;
3739
3740         spin_lock(&inode->i_lock);
3741         atomic_set(&inode->i_writecount, 1);
3742         spin_unlock(&inode->i_lock);
3743 }
3744
3745 /* mode:
3746  *   0 - completely stop and dis-assemble array
3747  *   1 - switch to readonly
3748  *   2 - stop but do not disassemble array
3749  */
3750 static int do_md_stop(mddev_t * mddev, int mode)
3751 {
3752         int err = 0;
3753         struct gendisk *disk = mddev->gendisk;
3754
3755         if (mddev->pers) {
3756                 if (atomic_read(&mddev->active)>2) {
3757                         printk("md: %s still in use.\n",mdname(mddev));
3758                         return -EBUSY;
3759                 }
3760
3761                 if (mddev->sync_thread) {
3762                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3763                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3764                         md_unregister_thread(mddev->sync_thread);
3765                         mddev->sync_thread = NULL;
3766                 }
3767
3768                 del_timer_sync(&mddev->safemode_timer);
3769
3770                 invalidate_partition(disk, 0);
3771
3772                 switch(mode) {
3773                 case 1: /* readonly */
3774                         err  = -ENXIO;
3775                         if (mddev->ro==1)
3776                                 goto out;
3777                         mddev->ro = 1;
3778                         break;
3779                 case 0: /* disassemble */
3780                 case 2: /* stop */
3781                         bitmap_flush(mddev);
3782                         md_super_wait(mddev);
3783                         if (mddev->ro)
3784                                 set_disk_ro(disk, 0);
3785                         blk_queue_make_request(mddev->queue, md_fail_request);
3786                         mddev->pers->stop(mddev);
3787                         mddev->queue->merge_bvec_fn = NULL;
3788                         mddev->queue->unplug_fn = NULL;
3789                         mddev->queue->backing_dev_info.congested_fn = NULL;
3790                         if (mddev->pers->sync_request)
3791                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3792
3793                         module_put(mddev->pers->owner);
3794                         mddev->pers = NULL;
3795                         /* tell userspace to handle 'inactive' */
3796                         sysfs_notify(&mddev->kobj, NULL, "array_state");
3797
3798                         set_capacity(disk, 0);
3799                         mddev->changed = 1;
3800
3801                         if (mddev->ro)
3802                                 mddev->ro = 0;
3803                 }
3804                 if (!mddev->in_sync || mddev->flags) {
3805                         /* mark array as shutdown cleanly */
3806                         mddev->in_sync = 1;
3807                         md_update_sb(mddev, 1);
3808                 }
3809                 if (mode == 1)
3810                         set_disk_ro(disk, 1);
3811                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3812         }
3813
3814         /*
3815          * Free resources if final stop
3816          */
3817         if (mode == 0) {
3818                 mdk_rdev_t *rdev;
3819                 struct list_head *tmp;
3820
3821                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3822
3823                 bitmap_destroy(mddev);
3824                 if (mddev->bitmap_file) {
3825                         restore_bitmap_write_access(mddev->bitmap_file);
3826                         fput(mddev->bitmap_file);
3827                         mddev->bitmap_file = NULL;
3828                 }
3829                 mddev->bitmap_offset = 0;
3830
3831                 rdev_for_each(rdev, tmp, mddev)
3832                         if (rdev->raid_disk >= 0) {
3833                                 char nm[20];
3834                                 sprintf(nm, "rd%d", rdev->raid_disk);
3835                                 sysfs_remove_link(&mddev->kobj, nm);
3836                         }
3837
3838                 /* make sure all md_delayed_delete calls have finished */
3839                 flush_scheduled_work();
3840
3841                 export_array(mddev);
3842
3843                 mddev->array_size = 0;
3844                 mddev->size = 0;
3845                 mddev->raid_disks = 0;
3846                 mddev->recovery_cp = 0;
3847                 mddev->resync_min = 0;
3848                 mddev->resync_max = MaxSector;
3849                 mddev->reshape_position = MaxSector;
3850                 mddev->external = 0;
3851                 mddev->persistent = 0;
3852                 mddev->level = LEVEL_NONE;
3853                 mddev->clevel[0] = 0;
3854                 mddev->flags = 0;
3855                 mddev->ro = 0;
3856                 mddev->metadata_type[0] = 0;
3857                 mddev->chunk_size = 0;
3858                 mddev->ctime = mddev->utime = 0;
3859                 mddev->layout = 0;
3860                 mddev->max_disks = 0;
3861                 mddev->events = 0;
3862                 mddev->delta_disks = 0;
3863                 mddev->new_level = LEVEL_NONE;
3864                 mddev->new_layout = 0;
3865                 mddev->new_chunk = 0;
3866                 mddev->curr_resync = 0;
3867                 mddev->resync_mismatches = 0;
3868                 mddev->suspend_lo = mddev->suspend_hi = 0;
3869                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3870                 mddev->recovery = 0;
3871                 mddev->in_sync = 0;
3872                 mddev->changed = 0;
3873                 mddev->degraded = 0;
3874                 mddev->barriers_work = 0;
3875                 mddev->safemode = 0;
3876
3877         } else if (mddev->pers)
3878                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3879                         mdname(mddev));
3880         err = 0;
3881         md_new_event(mddev);
3882 out:
3883         return err;
3884 }
3885
3886 #ifndef MODULE
3887 static void autorun_array(mddev_t *mddev)
3888 {
3889         mdk_rdev_t *rdev;
3890         struct list_head *tmp;
3891         int err;
3892
3893         if (list_empty(&mddev->disks))
3894                 return;
3895
3896         printk(KERN_INFO "md: running: ");
3897
3898         rdev_for_each(rdev, tmp, mddev) {
3899                 char b[BDEVNAME_SIZE];
3900                 printk("<%s>", bdevname(rdev->bdev,b));
3901         }
3902         printk("\n");
3903
3904         err = do_md_run (mddev);
3905         if (err) {
3906                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3907                 do_md_stop (mddev, 0);
3908         }
3909 }
3910
3911 /*
3912  * lets try to run arrays based on all disks that have arrived
3913  * until now. (those are in pending_raid_disks)
3914  *
3915  * the method: pick the first pending disk, collect all disks with
3916  * the same UUID, remove all from the pending list and put them into
3917  * the 'same_array' list. Then order this list based on superblock
3918  * update time (freshest comes first), kick out 'old' disks and
3919  * compare superblocks. If everything's fine then run it.
3920  *
3921  * If "unit" is allocated, then bump its reference count
3922  */
3923 static void autorun_devices(int part)
3924 {
3925         struct list_head *tmp;
3926         mdk_rdev_t *rdev0, *rdev;
3927         mddev_t *mddev;
3928         char b[BDEVNAME_SIZE];
3929
3930         printk(KERN_INFO "md: autorun ...\n");
3931         while (!list_empty(&pending_raid_disks)) {
3932                 int unit;
3933                 dev_t dev;
3934                 LIST_HEAD(candidates);
3935                 rdev0 = list_entry(pending_raid_disks.next,
3936                                          mdk_rdev_t, same_set);
3937
3938                 printk(KERN_INFO "md: considering %s ...\n",
3939                         bdevname(rdev0->bdev,b));
3940                 INIT_LIST_HEAD(&candidates);
3941                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3942                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3943                                 printk(KERN_INFO "md:  adding %s ...\n",
3944                                         bdevname(rdev->bdev,b));
3945                                 list_move(&rdev->same_set, &candidates);
3946                         }
3947                 /*
3948                  * now we have a set of devices, with all of them having
3949                  * mostly sane superblocks. It's time to allocate the
3950                  * mddev.
3951                  */
3952                 if (part) {
3953                         dev = MKDEV(mdp_major,
3954                                     rdev0->preferred_minor << MdpMinorShift);
3955                         unit = MINOR(dev) >> MdpMinorShift;
3956                 } else {
3957                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3958                         unit = MINOR(dev);
3959                 }
3960                 if (rdev0->preferred_minor != unit) {
3961                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3962                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3963                         break;
3964                 }
3965
3966                 md_probe(dev, NULL, NULL);
3967                 mddev = mddev_find(dev);
3968                 if (!mddev || !mddev->gendisk) {
3969                         if (mddev)
3970                                 mddev_put(mddev);
3971                         printk(KERN_ERR
3972                                 "md: cannot allocate memory for md drive.\n");
3973                         break;
3974                 }
3975                 if (mddev_lock(mddev))
3976                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3977                                mdname(mddev));
3978                 else if (mddev->raid_disks || mddev->major_version
3979                          || !list_empty(&mddev->disks)) {
3980                         printk(KERN_WARNING
3981                                 "md: %s already running, cannot run %s\n",
3982                                 mdname(mddev), bdevname(rdev0->bdev,b));
3983                         mddev_unlock(mddev);
3984                 } else {
3985                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3986                         mddev->persistent = 1;
3987                         rdev_for_each_list(rdev, tmp, candidates) {
3988                                 list_del_init(&rdev->same_set);
3989                                 if (bind_rdev_to_array(rdev, mddev))
3990                                         export_rdev(rdev);
3991                         }
3992                         autorun_array(mddev);
3993                         mddev_unlock(mddev);
3994                 }
3995                 /* on success, candidates will be empty, on error
3996                  * it won't...
3997                  */
3998                 rdev_for_each_list(rdev, tmp, candidates)
3999                         export_rdev(rdev);
4000                 mddev_put(mddev);
4001         }
4002         printk(KERN_INFO "md: ... autorun DONE.\n");
4003 }
4004 #endif /* !MODULE */
4005
4006 static int get_version(void __user * arg)
4007 {
4008         mdu_version_t ver;
4009
4010         ver.major = MD_MAJOR_VERSION;
4011         ver.minor = MD_MINOR_VERSION;
4012         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4013
4014         if (copy_to_user(arg, &ver, sizeof(ver)))
4015                 return -EFAULT;
4016
4017         return 0;
4018 }
4019
4020 static int get_array_info(mddev_t * mddev, void __user * arg)
4021 {
4022         mdu_array_info_t info;
4023         int nr,working,active,failed,spare;
4024         mdk_rdev_t *rdev;
4025         struct list_head *tmp;
4026
4027         nr=working=active=failed=spare=0;
4028         rdev_for_each(rdev, tmp, mddev) {
4029                 nr++;
4030                 if (test_bit(Faulty, &rdev->flags))
4031                         failed++;
4032                 else {
4033                         working++;
4034                         if (test_bit(In_sync, &rdev->flags))
4035                                 active++;
4036                         else
4037                                 spare++;
4038                 }
4039         }
4040
4041         info.major_version = mddev->major_version;
4042         info.minor_version = mddev->minor_version;
4043         info.patch_version = MD_PATCHLEVEL_VERSION;
4044         info.ctime         = mddev->ctime;
4045         info.level         = mddev->level;
4046         info.size          = mddev->size;
4047         if (info.size != mddev->size) /* overflow */
4048                 info.size = -1;
4049         info.nr_disks      = nr;
4050         info.raid_disks    = mddev->raid_disks;
4051         info.md_minor      = mddev->md_minor;
4052         info.not_persistent= !mddev->persistent;
4053
4054         info.utime         = mddev->utime;
4055         info.state         = 0;
4056         if (mddev->in_sync)
4057                 info.state = (1<<MD_SB_CLEAN);
4058         if (mddev->bitmap && mddev->bitmap_offset)
4059                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4060         info.active_disks  = active;
4061         info.working_disks = working;
4062         info.failed_disks  = failed;
4063         info.spare_disks   = spare;
4064
4065         info.layout        = mddev->layout;
4066         info.chunk_size    = mddev->chunk_size;
4067
4068         if (copy_to_user(arg, &info, sizeof(info)))
4069                 return -EFAULT;
4070
4071         return 0;
4072 }
4073
4074 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4075 {
4076         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4077         char *ptr, *buf = NULL;
4078         int err = -ENOMEM;
4079
4080         md_allow_write(mddev);
4081
4082         file = kmalloc(sizeof(*file), GFP_KERNEL);
4083         if (!file)
4084                 goto out;
4085
4086         /* bitmap disabled, zero the first byte and copy out */
4087         if (!mddev->bitmap || !mddev->bitmap->file) {
4088                 file->pathname[0] = '\0';
4089                 goto copy_out;
4090         }
4091
4092         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4093         if (!buf)
4094                 goto out;
4095
4096         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4097         if (IS_ERR(ptr))
4098                 goto out;
4099
4100         strcpy(file->pathname, ptr);
4101
4102 copy_out:
4103         err = 0;
4104         if (copy_to_user(arg, file, sizeof(*file)))
4105                 err = -EFAULT;
4106 out:
4107         kfree(buf);
4108         kfree(file);
4109         return err;
4110 }
4111
4112 static int get_disk_info(mddev_t * mddev, void __user * arg)
4113 {
4114         mdu_disk_info_t info;
4115         unsigned int nr;
4116         mdk_rdev_t *rdev;
4117
4118         if (copy_from_user(&info, arg, sizeof(info)))
4119                 return -EFAULT;
4120
4121         nr = info.number;
4122
4123         rdev = find_rdev_nr(mddev, nr);
4124         if (rdev) {
4125                 info.major = MAJOR(rdev->bdev->bd_dev);
4126                 info.minor = MINOR(rdev->bdev->bd_dev);
4127                 info.raid_disk = rdev->raid_disk;
4128                 info.state = 0;
4129                 if (test_bit(Faulty, &rdev->flags))
4130                         info.state |= (1<<MD_DISK_FAULTY);
4131                 else if (test_bit(In_sync, &rdev->flags)) {
4132                         info.state |= (1<<MD_DISK_ACTIVE);
4133                         info.state |= (1<<MD_DISK_SYNC);
4134                 }
4135                 if (test_bit(WriteMostly, &rdev->flags))
4136                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4137         } else {
4138                 info.major = info.minor = 0;
4139                 info.raid_disk = -1;
4140                 info.state = (1<<MD_DISK_REMOVED);
4141         }
4142
4143         if (copy_to_user(arg, &info, sizeof(info)))
4144                 return -EFAULT;
4145
4146         return 0;
4147 }
4148
4149 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4150 {
4151         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4152         mdk_rdev_t *rdev;
4153         dev_t dev = MKDEV(info->major,info->minor);
4154
4155         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4156                 return -EOVERFLOW;
4157
4158         if (!mddev->raid_disks) {
4159                 int err;
4160                 /* expecting a device which has a superblock */
4161                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4162                 if (IS_ERR(rdev)) {
4163                         printk(KERN_WARNING
4164                                 "md: md_import_device returned %ld\n",
4165                                 PTR_ERR(rdev));
4166                         return PTR_ERR(rdev);
4167                 }
4168                 if (!list_empty(&mddev->disks)) {
4169                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4170                                                         mdk_rdev_t, same_set);
4171                         int err = super_types[mddev->major_version]
4172                                 .load_super(rdev, rdev0, mddev->minor_version);
4173                         if (err < 0) {
4174                                 printk(KERN_WARNING
4175                                         "md: %s has different UUID to %s\n",
4176                                         bdevname(rdev->bdev,b),
4177                                         bdevname(rdev0->bdev,b2));
4178                                 export_rdev(rdev);
4179                                 return -EINVAL;
4180                         }
4181                 }
4182                 err = bind_rdev_to_array(rdev, mddev);
4183                 if (err)
4184                         export_rdev(rdev);
4185                 return err;
4186         }
4187
4188         /*
4189          * add_new_disk can be used once the array is assembled
4190          * to add "hot spares".  They must already have a superblock
4191          * written
4192          */
4193         if (mddev->pers) {
4194                 int err;
4195                 if (!mddev->pers->hot_add_disk) {
4196                         printk(KERN_WARNING
4197                                 "%s: personality does not support diskops!\n",
4198                                mdname(mddev));
4199                         return -EINVAL;
4200                 }
4201                 if (mddev->persistent)
4202                         rdev = md_import_device(dev, mddev->major_version,
4203                                                 mddev->minor_version);
4204                 else
4205                         rdev = md_import_device(dev, -1, -1);
4206                 if (IS_ERR(rdev)) {
4207                         printk(KERN_WARNING
4208                                 "md: md_import_device returned %ld\n",
4209                                 PTR_ERR(rdev));
4210                         return PTR_ERR(rdev);
4211                 }
4212                 /* set save_raid_disk if appropriate */
4213                 if (!mddev->persistent) {
4214                         if (info->state & (1<<MD_DISK_SYNC)  &&
4215                             info->raid_disk < mddev->raid_disks)
4216                                 rdev->raid_disk = info->raid_disk;
4217                         else
4218                                 rdev->raid_disk = -1;
4219                 } else
4220                         super_types[mddev->major_version].
4221                                 validate_super(mddev, rdev);
4222                 rdev->saved_raid_disk = rdev->raid_disk;
4223
4224                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4225                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4226                         set_bit(WriteMostly, &rdev->flags);
4227
4228                 rdev->raid_disk = -1;
4229                 err = bind_rdev_to_array(rdev, mddev);
4230                 if (!err && !mddev->pers->hot_remove_disk) {
4231                         /* If there is hot_add_disk but no hot_remove_disk
4232                          * then added disks for geometry changes,
4233                          * and should be added immediately.
4234                          */
4235                         super_types[mddev->major_version].
4236                                 validate_super(mddev, rdev);
4237                         err = mddev->pers->hot_add_disk(mddev, rdev);
4238                         if (err)
4239                                 unbind_rdev_from_array(rdev);
4240                 }
4241                 if (err)
4242                         export_rdev(rdev);
4243
4244                 md_update_sb(mddev, 1);
4245                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4246                 md_wakeup_thread(mddev->thread);
4247                 return err;
4248         }
4249
4250         /* otherwise, add_new_disk is only allowed
4251          * for major_version==0 superblocks
4252          */
4253         if (mddev->major_version != 0) {
4254                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4255                        mdname(mddev));
4256                 return -EINVAL;
4257         }
4258
4259         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4260                 int err;
4261                 rdev = md_import_device (dev, -1, 0);
4262                 if (IS_ERR(rdev)) {
4263                         printk(KERN_WARNING
4264                                 "md: error, md_import_device() returned %ld\n",
4265                                 PTR_ERR(rdev));
4266                         return PTR_ERR(rdev);
4267                 }
4268                 rdev->desc_nr = info->number;
4269                 if (info->raid_disk < mddev->raid_disks)
4270                         rdev->raid_disk = info->raid_disk;
4271                 else
4272                         rdev->raid_disk = -1;
4273
4274                 if (rdev->raid_disk < mddev->raid_disks)
4275                         if (info->state & (1<<MD_DISK_SYNC))
4276                                 set_bit(In_sync, &rdev->flags);
4277
4278                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4279                         set_bit(WriteMostly, &rdev->flags);
4280
4281                 if (!mddev->persistent) {
4282                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4283                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4284                 } else
4285                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4286                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4287
4288                 err = bind_rdev_to_array(rdev, mddev);
4289                 if (err) {
4290                         export_rdev(rdev);
4291                         return err;
4292                 }
4293         }
4294
4295         return 0;
4296 }
4297
4298 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4299 {
4300         char b[BDEVNAME_SIZE];
4301         mdk_rdev_t *rdev;
4302
4303         rdev = find_rdev(mddev, dev);
4304         if (!rdev)
4305                 return -ENXIO;
4306
4307         if (rdev->raid_disk >= 0)
4308                 goto busy;
4309
4310         kick_rdev_from_array(rdev);
4311         md_update_sb(mddev, 1);
4312         md_new_event(mddev);
4313
4314         return 0;
4315 busy:
4316         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4317                 bdevname(rdev->bdev,b), mdname(mddev));
4318         return -EBUSY;
4319 }
4320
4321 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4322 {
4323         char b[BDEVNAME_SIZE];
4324         int err;
4325         unsigned int size;
4326         mdk_rdev_t *rdev;
4327
4328         if (!mddev->pers)
4329                 return -ENODEV;
4330
4331         if (mddev->major_version != 0) {
4332                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4333                         " version-0 superblocks.\n",
4334                         mdname(mddev));
4335                 return -EINVAL;
4336         }
4337         if (!mddev->pers->hot_add_disk) {
4338                 printk(KERN_WARNING
4339                         "%s: personality does not support diskops!\n",
4340                         mdname(mddev));
4341                 return -EINVAL;
4342         }
4343
4344         rdev = md_import_device (dev, -1, 0);
4345         if (IS_ERR(rdev)) {
4346                 printk(KERN_WARNING
4347                         "md: error, md_import_device() returned %ld\n",
4348                         PTR_ERR(rdev));
4349                 return -EINVAL;
4350         }
4351
4352         if (mddev->persistent)
4353                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4354         else
4355                 rdev->sb_offset =
4356                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4357
4358         size = calc_dev_size(rdev, mddev->chunk_size);
4359         rdev->size = size;
4360
4361         if (test_bit(Faulty, &rdev->flags)) {
4362                 printk(KERN_WARNING
4363                         "md: can not hot-add faulty %s disk to %s!\n",
4364                         bdevname(rdev->bdev,b), mdname(mddev));
4365                 err = -EINVAL;
4366                 goto abort_export;
4367         }
4368         clear_bit(In_sync, &rdev->flags);
4369         rdev->desc_nr = -1;
4370         rdev->saved_raid_disk = -1;
4371         err = bind_rdev_to_array(rdev, mddev);
4372         if (err)
4373                 goto abort_export;
4374
4375         /*
4376          * The rest should better be atomic, we can have disk failures
4377          * noticed in interrupt contexts ...
4378          */
4379
4380         if (rdev->desc_nr == mddev->max_disks) {
4381                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4382                         mdname(mddev));
4383                 err = -EBUSY;
4384                 goto abort_unbind_export;
4385         }
4386
4387         rdev->raid_disk = -1;
4388
4389         md_update_sb(mddev, 1);
4390
4391         /*
4392          * Kick recovery, maybe this spare has to be added to the
4393          * array immediately.
4394          */
4395         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4396         md_wakeup_thread(mddev->thread);
4397         md_new_event(mddev);
4398         return 0;
4399
4400 abort_unbind_export:
4401         unbind_rdev_from_array(rdev);
4402
4403 abort_export:
4404         export_rdev(rdev);
4405         return err;
4406 }
4407
4408 static int set_bitmap_file(mddev_t *mddev, int fd)
4409 {
4410         int err;
4411
4412         if (mddev->pers) {
4413                 if (!mddev->pers->quiesce)
4414                         return -EBUSY;
4415                 if (mddev->recovery || mddev->sync_thread)
4416                         return -EBUSY;
4417                 /* we should be able to change the bitmap.. */
4418         }
4419
4420
4421         if (fd >= 0) {
4422                 if (mddev->bitmap)
4423                         return -EEXIST; /* cannot add when bitmap is present */
4424                 mddev->bitmap_file = fget(fd);
4425
4426                 if (mddev->bitmap_file == NULL) {
4427                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4428                                mdname(mddev));
4429                         return -EBADF;
4430                 }
4431
4432                 err = deny_bitmap_write_access(mddev->bitmap_file);
4433                 if (err) {
4434                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4435                                mdname(mddev));
4436                         fput(mddev->bitmap_file);
4437                         mddev->bitmap_file = NULL;
4438                         return err;
4439                 }
4440                 mddev->bitmap_offset = 0; /* file overrides offset */
4441         } else if (mddev->bitmap == NULL)
4442                 return -ENOENT; /* cannot remove what isn't there */
4443         err = 0;
4444         if (mddev->pers) {
4445                 mddev->pers->quiesce(mddev, 1);
4446                 if (fd >= 0)
4447                         err = bitmap_create(mddev);
4448                 if (fd < 0 || err) {
4449                         bitmap_destroy(mddev);
4450                         fd = -1; /* make sure to put the file */
4451                 }
4452                 mddev->pers->quiesce(mddev, 0);
4453         }
4454         if (fd < 0) {
4455                 if (mddev->bitmap_file) {
4456                         restore_bitmap_write_access(mddev->bitmap_file);
4457                         fput(mddev->bitmap_file);
4458                 }
4459                 mddev->bitmap_file = NULL;
4460         }
4461
4462         return err;
4463 }
4464
4465 /*
4466  * set_array_info is used two different ways
4467  * The original usage is when creating a new array.
4468  * In this usage, raid_disks is > 0 and it together with
4469  *  level, size, not_persistent,layout,chunksize determine the
4470  *  shape of the array.
4471  *  This will always create an array with a type-0.90.0 superblock.
4472  * The newer usage is when assembling an array.
4473  *  In this case raid_disks will be 0, and the major_version field is
4474  *  use to determine which style super-blocks are to be found on the devices.
4475  *  The minor and patch _version numbers are also kept incase the
4476  *  super_block handler wishes to interpret them.
4477  */
4478 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4479 {
4480
4481         if (info->raid_disks == 0) {
4482                 /* just setting version number for superblock loading */
4483                 if (info->major_version < 0 ||
4484                     info->major_version >= ARRAY_SIZE(super_types) ||
4485                     super_types[info->major_version].name == NULL) {
4486                         /* maybe try to auto-load a module? */
4487                         printk(KERN_INFO
4488                                 "md: superblock version %d not known\n",
4489                                 info->major_version);
4490                         return -EINVAL;
4491                 }
4492                 mddev->major_version = info->major_version;
4493                 mddev->minor_version = info->minor_version;
4494                 mddev->patch_version = info->patch_version;
4495                 mddev->persistent = !info->not_persistent;
4496                 return 0;
4497         }
4498         mddev->major_version = MD_MAJOR_VERSION;
4499         mddev->minor_version = MD_MINOR_VERSION;
4500         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4501         mddev->ctime         = get_seconds();
4502
4503         mddev->level         = info->level;
4504         mddev->clevel[0]     = 0;
4505         mddev->size          = info->size;
4506         mddev->raid_disks    = info->raid_disks;
4507         /* don't set md_minor, it is determined by which /dev/md* was
4508          * openned
4509          */
4510         if (info->state & (1<<MD_SB_CLEAN))
4511                 mddev->recovery_cp = MaxSector;
4512         else
4513                 mddev->recovery_cp = 0;
4514         mddev->persistent    = ! info->not_persistent;
4515         mddev->external      = 0;
4516
4517         mddev->layout        = info->layout;
4518         mddev->chunk_size    = info->chunk_size;
4519
4520         mddev->max_disks     = MD_SB_DISKS;
4521
4522         if (mddev->persistent)
4523                 mddev->flags         = 0;
4524         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4525
4526         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4527         mddev->bitmap_offset = 0;
4528
4529         mddev->reshape_position = MaxSector;
4530
4531         /*
4532          * Generate a 128 bit UUID
4533          */
4534         get_random_bytes(mddev->uuid, 16);
4535
4536         mddev->new_level = mddev->level;
4537         mddev->new_chunk = mddev->chunk_size;
4538         mddev->new_layout = mddev->layout;
4539         mddev->delta_disks = 0;
4540
4541         return 0;
4542 }
4543
4544 static int update_size(mddev_t *mddev, unsigned long size)
4545 {
4546         mdk_rdev_t * rdev;
4547         int rv;
4548         struct list_head *tmp;
4549         int fit = (size == 0);
4550
4551         if (mddev->pers->resize == NULL)
4552                 return -EINVAL;
4553         /* The "size" is the amount of each device that is used.
4554          * This can only make sense for arrays with redundancy.
4555          * linear and raid0 always use whatever space is available
4556          * We can only consider changing the size if no resync
4557          * or reconstruction is happening, and if the new size
4558          * is acceptable. It must fit before the sb_offset or,
4559          * if that is <data_offset, it must fit before the
4560          * size of each device.
4561          * If size is zero, we find the largest size that fits.
4562          */
4563         if (mddev->sync_thread)
4564                 return -EBUSY;
4565         rdev_for_each(rdev, tmp, mddev) {
4566                 sector_t avail;
4567                 avail = rdev->size * 2;
4568
4569                 if (fit && (size == 0 || size > avail/2))
4570                         size = avail/2;
4571                 if (avail < ((sector_t)size << 1))
4572                         return -ENOSPC;
4573         }
4574         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4575         if (!rv) {
4576                 struct block_device *bdev;
4577
4578                 bdev = bdget_disk(mddev->gendisk, 0);
4579                 if (bdev) {
4580                         mutex_lock(&bdev->bd_inode->i_mutex);
4581                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4582                         mutex_unlock(&bdev->bd_inode->i_mutex);
4583                         bdput(bdev);
4584                 }
4585         }
4586         return rv;
4587 }
4588
4589 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4590 {
4591         int rv;
4592         /* change the number of raid disks */
4593         if (mddev->pers->check_reshape == NULL)
4594                 return -EINVAL;
4595         if (raid_disks <= 0 ||
4596             raid_disks >= mddev->max_disks)
4597                 return -EINVAL;
4598         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4599                 return -EBUSY;
4600         mddev->delta_disks = raid_disks - mddev->raid_disks;
4601
4602         rv = mddev->pers->check_reshape(mddev);
4603         return rv;
4604 }
4605
4606
4607 /*
4608  * update_array_info is used to change the configuration of an
4609  * on-line array.
4610  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4611  * fields in the info are checked against the array.
4612  * Any differences that cannot be handled will cause an error.
4613  * Normally, only one change can be managed at a time.
4614  */
4615 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4616 {
4617         int rv = 0;
4618         int cnt = 0;
4619         int state = 0;
4620
4621         /* calculate expected state,ignoring low bits */
4622         if (mddev->bitmap && mddev->bitmap_offset)
4623                 state |= (1 << MD_SB_BITMAP_PRESENT);
4624
4625         if (mddev->major_version != info->major_version ||
4626             mddev->minor_version != info->minor_version ||
4627 /*          mddev->patch_version != info->patch_version || */
4628             mddev->ctime         != info->ctime         ||
4629             mddev->level         != info->level         ||
4630 /*          mddev->layout        != info->layout        || */
4631             !mddev->persistent   != info->not_persistent||
4632             mddev->chunk_size    != info->chunk_size    ||
4633             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4634             ((state^info->state) & 0xfffffe00)
4635                 )
4636                 return -EINVAL;
4637         /* Check there is only one change */
4638         if (info->size >= 0 && mddev->size != info->size) cnt++;
4639         if (mddev->raid_disks != info->raid_disks) cnt++;
4640         if (mddev->layout != info->layout) cnt++;
4641         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4642         if (cnt == 0) return 0;
4643         if (cnt > 1) return -EINVAL;
4644
4645         if (mddev->layout != info->layout) {
4646                 /* Change layout
4647                  * we don't need to do anything at the md level, the
4648                  * personality will take care of it all.
4649                  */
4650                 if (mddev->pers->reconfig == NULL)
4651                         return -EINVAL;
4652                 else
4653                         return mddev->pers->reconfig(mddev, info->layout, -1);
4654         }
4655         if (info->size >= 0 && mddev->size != info->size)
4656                 rv = update_size(mddev, info->size);
4657
4658         if (mddev->raid_disks    != info->raid_disks)
4659                 rv = update_raid_disks(mddev, info->raid_disks);
4660
4661         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4662                 if (mddev->pers->quiesce == NULL)
4663                         return -EINVAL;
4664                 if (mddev->recovery || mddev->sync_thread)
4665                         return -EBUSY;
4666                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4667                         /* add the bitmap */
4668                         if (mddev->bitmap)
4669                                 return -EEXIST;
4670                         if (mddev->default_bitmap_offset == 0)
4671                                 return -EINVAL;
4672                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4673                         mddev->pers->quiesce(mddev, 1);
4674                         rv = bitmap_create(mddev);
4675                         if (rv)
4676                                 bitmap_destroy(mddev);
4677                         mddev->pers->quiesce(mddev, 0);
4678                 } else {
4679                         /* remove the bitmap */
4680                         if (!mddev->bitmap)
4681                                 return -ENOENT;
4682                         if (mddev->bitmap->file)
4683                                 return -EINVAL;
4684                         mddev->pers->quiesce(mddev, 1);
4685                         bitmap_destroy(mddev);
4686                         mddev->pers->quiesce(mddev, 0);
4687                         mddev->bitmap_offset = 0;
4688                 }
4689         }
4690         md_update_sb(mddev, 1);
4691         return rv;
4692 }
4693
4694 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4695 {
4696         mdk_rdev_t *rdev;
4697
4698         if (mddev->pers == NULL)
4699                 return -ENODEV;
4700
4701         rdev = find_rdev(mddev, dev);
4702         if (!rdev)
4703                 return -ENODEV;
4704
4705         md_error(mddev, rdev);
4706         return 0;
4707 }
4708
4709 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4710 {
4711         mddev_t *mddev = bdev->bd_disk->private_data;
4712
4713         geo->heads = 2;
4714         geo->sectors = 4;
4715         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4716         return 0;
4717 }
4718
4719 static int md_ioctl(struct inode *inode, struct file *file,
4720                         unsigned int cmd, unsigned long arg)
4721 {
4722         int err = 0;
4723         void __user *argp = (void __user *)arg;
4724         mddev_t *mddev = NULL;
4725
4726         if (!capable(CAP_SYS_ADMIN))
4727                 return -EACCES;
4728
4729         /*
4730          * Commands dealing with the RAID driver but not any
4731          * particular array:
4732          */
4733         switch (cmd)
4734         {
4735                 case RAID_VERSION:
4736                         err = get_version(argp);
4737                         goto done;
4738
4739                 case PRINT_RAID_DEBUG:
4740                         err = 0;
4741                         md_print_devices();
4742                         goto done;
4743
4744 #ifndef MODULE
4745                 case RAID_AUTORUN:
4746                         err = 0;
4747                         autostart_arrays(arg);
4748                         goto done;
4749 #endif
4750                 default:;
4751         }
4752
4753         /*
4754          * Commands creating/starting a new array:
4755          */
4756
4757         mddev = inode->i_bdev->bd_disk->private_data;
4758
4759         if (!mddev) {
4760                 BUG();
4761                 goto abort;
4762         }
4763
4764         err = mddev_lock(mddev);
4765         if (err) {
4766                 printk(KERN_INFO
4767                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4768                         err, cmd);
4769                 goto abort;
4770         }
4771
4772         switch (cmd)
4773         {
4774                 case SET_ARRAY_INFO:
4775                         {
4776                                 mdu_array_info_t info;
4777                                 if (!arg)
4778                                         memset(&info, 0, sizeof(info));
4779                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4780                                         err = -EFAULT;
4781                                         goto abort_unlock;
4782                                 }
4783                                 if (mddev->pers) {
4784                                         err = update_array_info(mddev, &info);
4785                                         if (err) {
4786                                                 printk(KERN_WARNING "md: couldn't update"
4787                                                        " array info. %d\n", err);
4788                                                 goto abort_unlock;
4789                                         }
4790                                         goto done_unlock;
4791                                 }
4792                                 if (!list_empty(&mddev->disks)) {
4793                                         printk(KERN_WARNING
4794                                                "md: array %s already has disks!\n",
4795                                                mdname(mddev));
4796                                         err = -EBUSY;
4797                                         goto abort_unlock;
4798                                 }
4799                                 if (mddev->raid_disks) {
4800                                         printk(KERN_WARNING
4801                                                "md: array %s already initialised!\n",
4802                                                mdname(mddev));
4803                                         err = -EBUSY;
4804                                         goto abort_unlock;
4805                                 }
4806                                 err = set_array_info(mddev, &info);
4807                                 if (err) {
4808                                         printk(KERN_WARNING "md: couldn't set"
4809                                                " array info. %d\n", err);
4810                                         goto abort_unlock;
4811                                 }
4812                         }
4813                         goto done_unlock;
4814
4815                 default:;
4816         }
4817
4818         /*
4819          * Commands querying/configuring an existing array:
4820          */
4821         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4822          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4823         if ((!mddev->raid_disks && !mddev->external)
4824             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4825             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4826             && cmd != GET_BITMAP_FILE) {
4827                 err = -ENODEV;
4828                 goto abort_unlock;
4829         }
4830
4831         /*
4832          * Commands even a read-only array can execute:
4833          */
4834         switch (cmd)
4835         {
4836                 case GET_ARRAY_INFO:
4837                         err = get_array_info(mddev, argp);
4838                         goto done_unlock;
4839
4840                 case GET_BITMAP_FILE:
4841                         err = get_bitmap_file(mddev, argp);
4842                         goto done_unlock;
4843
4844                 case GET_DISK_INFO:
4845                         err = get_disk_info(mddev, argp);
4846                         goto done_unlock;
4847
4848                 case RESTART_ARRAY_RW:
4849                         err = restart_array(mddev);
4850                         goto done_unlock;
4851
4852                 case STOP_ARRAY:
4853                         err = do_md_stop (mddev, 0);
4854                         goto done_unlock;
4855
4856                 case STOP_ARRAY_RO:
4857                         err = do_md_stop (mddev, 1);
4858                         goto done_unlock;
4859
4860         /*
4861          * We have a problem here : there is no easy way to give a CHS
4862          * virtual geometry. We currently pretend that we have a 2 heads
4863          * 4 sectors (with a BIG number of cylinders...). This drives
4864          * dosfs just mad... ;-)
4865          */
4866         }
4867
4868         /*
4869          * The remaining ioctls are changing the state of the
4870          * superblock, so we do not allow them on read-only arrays.
4871          * However non-MD ioctls (e.g. get-size) will still come through
4872          * here and hit the 'default' below, so only disallow
4873          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4874          */
4875         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4876             mddev->ro && mddev->pers) {
4877                 if (mddev->ro == 2) {
4878                         mddev->ro = 0;
4879                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4880                 md_wakeup_thread(mddev->thread);
4881
4882                 } else {
4883                         err = -EROFS;
4884                         goto abort_unlock;
4885                 }
4886         }
4887
4888         switch (cmd)
4889         {
4890                 case ADD_NEW_DISK:
4891                 {
4892                         mdu_disk_info_t info;
4893                         if (copy_from_user(&info, argp, sizeof(info)))
4894                                 err = -EFAULT;
4895                         else
4896                                 err = add_new_disk(mddev, &info);
4897                         goto done_unlock;
4898                 }
4899
4900                 case HOT_REMOVE_DISK:
4901                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4902                         goto done_unlock;
4903
4904                 case HOT_ADD_DISK:
4905                         err = hot_add_disk(mddev, new_decode_dev(arg));
4906                         goto done_unlock;
4907
4908                 case SET_DISK_FAULTY:
4909                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4910                         goto done_unlock;
4911
4912                 case RUN_ARRAY:
4913                         err = do_md_run (mddev);
4914                         goto done_unlock;
4915
4916                 case SET_BITMAP_FILE:
4917                         err = set_bitmap_file(mddev, (int)arg);
4918                         goto done_unlock;
4919
4920                 default:
4921                         err = -EINVAL;
4922                         goto abort_unlock;
4923         }
4924
4925 done_unlock:
4926 abort_unlock:
4927         mddev_unlock(mddev);
4928
4929         return err;
4930 done:
4931         if (err)
4932                 MD_BUG();
4933 abort:
4934         return err;
4935 }
4936
4937 static int md_open(struct inode *inode, struct file *file)
4938 {
4939         /*
4940          * Succeed if we can lock the mddev, which confirms that
4941          * it isn't being stopped right now.
4942          */
4943         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4944         int err;
4945
4946         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4947                 goto out;
4948
4949         err = 0;
4950         mddev_get(mddev);
4951         mddev_unlock(mddev);
4952
4953         check_disk_change(inode->i_bdev);
4954  out:
4955         return err;
4956 }
4957
4958 static int md_release(struct inode *inode, struct file * file)
4959 {
4960         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4961
4962         BUG_ON(!mddev);
4963         mddev_put(mddev);
4964
4965         return 0;
4966 }
4967
4968 static int md_media_changed(struct gendisk *disk)
4969 {
4970         mddev_t *mddev = disk->private_data;
4971
4972         return mddev->changed;
4973 }
4974
4975 static int md_revalidate(struct gendisk *disk)
4976 {
4977         mddev_t *mddev = disk->private_data;
4978
4979         mddev->changed = 0;
4980         return 0;
4981 }
4982 static struct block_device_operations md_fops =
4983 {
4984         .owner          = THIS_MODULE,
4985         .open           = md_open,
4986         .release        = md_release,
4987         .ioctl          = md_ioctl,
4988         .getgeo         = md_getgeo,
4989         .media_changed  = md_media_changed,
4990         .revalidate_disk= md_revalidate,
4991 };
4992
4993 static int md_thread(void * arg)
4994 {
4995         mdk_thread_t *thread = arg;
4996
4997         /*
4998          * md_thread is a 'system-thread', it's priority should be very
4999          * high. We avoid resource deadlocks individually in each
5000          * raid personality. (RAID5 does preallocation) We also use RR and
5001          * the very same RT priority as kswapd, thus we will never get
5002          * into a priority inversion deadlock.
5003          *
5004          * we definitely have to have equal or higher priority than
5005          * bdflush, otherwise bdflush will deadlock if there are too
5006          * many dirty RAID5 blocks.
5007          */
5008
5009         allow_signal(SIGKILL);
5010         while (!kthread_should_stop()) {
5011
5012                 /* We need to wait INTERRUPTIBLE so that
5013                  * we don't add to the load-average.
5014                  * That means we need to be sure no signals are
5015                  * pending
5016                  */
5017                 if (signal_pending(current))
5018                         flush_signals(current);
5019
5020                 wait_event_interruptible_timeout
5021                         (thread->wqueue,
5022                          test_bit(THREAD_WAKEUP, &thread->flags)
5023                          || kthread_should_stop(),
5024                          thread->timeout);
5025
5026                 clear_bit(THREAD_WAKEUP, &thread->flags);
5027
5028                 thread->run(thread->mddev);
5029         }
5030
5031         return 0;
5032 }
5033
5034 void md_wakeup_thread(mdk_thread_t *thread)
5035 {
5036         if (thread) {
5037                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5038                 set_bit(THREAD_WAKEUP, &thread->flags);
5039                 wake_up(&thread->wqueue);
5040         }
5041 }
5042
5043 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5044                                  const char *name)
5045 {
5046         mdk_thread_t *thread;
5047
5048         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5049         if (!thread)
5050                 return NULL;
5051
5052         init_waitqueue_head(&thread->wqueue);
5053
5054         thread->run = run;
5055         thread->mddev = mddev;
5056         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5057         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5058         if (IS_ERR(thread->tsk)) {
5059                 kfree(thread);
5060                 return NULL;
5061         }
5062         return thread;
5063 }
5064
5065 void md_unregister_thread(mdk_thread_t *thread)
5066 {
5067         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5068
5069         kthread_stop(thread->tsk);
5070         kfree(thread);
5071 }
5072
5073 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5074 {
5075         if (!mddev) {
5076                 MD_BUG();
5077                 return;
5078         }
5079
5080         if (!rdev || test_bit(Faulty, &rdev->flags))
5081                 return;
5082
5083         if (mddev->external)
5084                 set_bit(Blocked, &rdev->flags);
5085 /*
5086         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5087                 mdname(mddev),
5088                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5089                 __builtin_return_address(0),__builtin_return_address(1),
5090                 __builtin_return_address(2),__builtin_return_address(3));
5091 */
5092         if (!mddev->pers)
5093                 return;
5094         if (!mddev->pers->error_handler)
5095                 return;
5096         mddev->pers->error_handler(mddev,rdev);
5097         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5098         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5099         md_wakeup_thread(mddev->thread);
5100         md_new_event_inintr(mddev);
5101 }
5102
5103 /* seq_file implementation /proc/mdstat */
5104
5105 static void status_unused(struct seq_file *seq)
5106 {
5107         int i = 0;
5108         mdk_rdev_t *rdev;
5109         struct list_head *tmp;
5110
5111         seq_printf(seq, "unused devices: ");
5112
5113         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5114                 char b[BDEVNAME_SIZE];
5115                 i++;
5116                 seq_printf(seq, "%s ",
5117                               bdevname(rdev->bdev,b));
5118         }
5119         if (!i)
5120                 seq_printf(seq, "<none>");
5121
5122         seq_printf(seq, "\n");
5123 }
5124
5125
5126 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5127 {
5128         sector_t max_blocks, resync, res;
5129         unsigned long dt, db, rt;
5130         int scale;
5131         unsigned int per_milli;
5132
5133         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5134
5135         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5136                 max_blocks = mddev->resync_max_sectors >> 1;
5137         else
5138                 max_blocks = mddev->size;
5139
5140         /*
5141          * Should not happen.
5142          */
5143         if (!max_blocks) {
5144                 MD_BUG();
5145                 return;
5146         }
5147         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5148          * in a sector_t, and (max_blocks>>scale) will fit in a
5149          * u32, as those are the requirements for sector_div.
5150          * Thus 'scale' must be at least 10
5151          */
5152         scale = 10;
5153         if (sizeof(sector_t) > sizeof(unsigned long)) {
5154                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5155                         scale++;
5156         }
5157         res = (resync>>scale)*1000;
5158         sector_div(res, (u32)((max_blocks>>scale)+1));
5159
5160         per_milli = res;
5161         {
5162                 int i, x = per_milli/50, y = 20-x;
5163                 seq_printf(seq, "[");
5164                 for (i = 0; i < x; i++)
5165                         seq_printf(seq, "=");
5166                 seq_printf(seq, ">");
5167                 for (i = 0; i < y; i++)
5168                         seq_printf(seq, ".");
5169                 seq_printf(seq, "] ");
5170         }
5171         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5172                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5173                     "reshape" :
5174                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5175                      "check" :
5176                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5177                       "resync" : "recovery"))),
5178                    per_milli/10, per_milli % 10,
5179                    (unsigned long long) resync,
5180                    (unsigned long long) max_blocks);
5181
5182         /*
5183          * We do not want to overflow, so the order of operands and
5184          * the * 100 / 100 trick are important. We do a +1 to be
5185          * safe against division by zero. We only estimate anyway.
5186          *
5187          * dt: time from mark until now
5188          * db: blocks written from mark until now
5189          * rt: remaining time
5190          */
5191         dt = ((jiffies - mddev->resync_mark) / HZ);
5192         if (!dt) dt++;
5193         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5194                 - mddev->resync_mark_cnt;
5195         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5196
5197         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5198
5199         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5200 }
5201
5202 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5203 {
5204         struct list_head *tmp;
5205         loff_t l = *pos;
5206         mddev_t *mddev;
5207
5208         if (l >= 0x10000)
5209                 return NULL;
5210         if (!l--)
5211                 /* header */
5212                 return (void*)1;
5213
5214         spin_lock(&all_mddevs_lock);
5215         list_for_each(tmp,&all_mddevs)
5216                 if (!l--) {
5217                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5218                         mddev_get(mddev);
5219                         spin_unlock(&all_mddevs_lock);
5220                         return mddev;
5221                 }
5222         spin_unlock(&all_mddevs_lock);
5223         if (!l--)
5224                 return (void*)2;/* tail */
5225         return NULL;
5226 }
5227
5228 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5229 {
5230         struct list_head *tmp;
5231         mddev_t *next_mddev, *mddev = v;
5232
5233         ++*pos;
5234         if (v == (void*)2)
5235                 return NULL;
5236
5237         spin_lock(&all_mddevs_lock);
5238         if (v == (void*)1)
5239                 tmp = all_mddevs.next;
5240         else
5241                 tmp = mddev->all_mddevs.next;
5242         if (tmp != &all_mddevs)
5243                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5244         else {
5245                 next_mddev = (void*)2;
5246                 *pos = 0x10000;
5247         }
5248         spin_unlock(&all_mddevs_lock);
5249
5250         if (v != (void*)1)
5251                 mddev_put(mddev);
5252         return next_mddev;
5253
5254 }
5255
5256 static void md_seq_stop(struct seq_file *seq, void *v)
5257 {
5258         mddev_t *mddev = v;
5259
5260         if (mddev && v != (void*)1 && v != (void*)2)
5261                 mddev_put(mddev);
5262 }
5263
5264 struct mdstat_info {
5265         int event;
5266 };
5267
5268 static int md_seq_show(struct seq_file *seq, void *v)
5269 {
5270         mddev_t *mddev = v;
5271         sector_t size;
5272         struct list_head *tmp2;
5273         mdk_rdev_t *rdev;
5274         struct mdstat_info *mi = seq->private;
5275         struct bitmap *bitmap;
5276
5277         if (v == (void*)1) {
5278                 struct mdk_personality *pers;
5279                 seq_printf(seq, "Personalities : ");
5280                 spin_lock(&pers_lock);
5281                 list_for_each_entry(pers, &pers_list, list)
5282                         seq_printf(seq, "[%s] ", pers->name);
5283
5284                 spin_unlock(&pers_lock);
5285                 seq_printf(seq, "\n");
5286                 mi->event = atomic_read(&md_event_count);
5287                 return 0;
5288         }
5289         if (v == (void*)2) {
5290                 status_unused(seq);
5291                 return 0;
5292         }
5293
5294         if (mddev_lock(mddev) < 0)
5295                 return -EINTR;
5296
5297         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5298                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5299                                                 mddev->pers ? "" : "in");
5300                 if (mddev->pers) {
5301                         if (mddev->ro==1)
5302                                 seq_printf(seq, " (read-only)");
5303                         if (mddev->ro==2)
5304                                 seq_printf(seq, " (auto-read-only)");
5305                         seq_printf(seq, " %s", mddev->pers->name);
5306                 }
5307
5308                 size = 0;
5309                 rdev_for_each(rdev, tmp2, mddev) {
5310                         char b[BDEVNAME_SIZE];
5311                         seq_printf(seq, " %s[%d]",
5312                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5313                         if (test_bit(WriteMostly, &rdev->flags))
5314                                 seq_printf(seq, "(W)");
5315                         if (test_bit(Faulty, &rdev->flags)) {
5316                                 seq_printf(seq, "(F)");
5317                                 continue;
5318                         } else if (rdev->raid_disk < 0)
5319                                 seq_printf(seq, "(S)"); /* spare */
5320                         size += rdev->size;
5321                 }
5322
5323                 if (!list_empty(&mddev->disks)) {
5324                         if (mddev->pers)
5325                                 seq_printf(seq, "\n      %llu blocks",
5326                                         (unsigned long long)mddev->array_size);
5327                         else
5328                                 seq_printf(seq, "\n      %llu blocks",
5329                                         (unsigned long long)size);
5330                 }
5331                 if (mddev->persistent) {
5332                         if (mddev->major_version != 0 ||
5333                             mddev->minor_version != 90) {
5334                                 seq_printf(seq," super %d.%d",
5335                                            mddev->major_version,
5336                                            mddev->minor_version);
5337                         }
5338                 } else if (mddev->external)
5339                         seq_printf(seq, " super external:%s",
5340                                    mddev->metadata_type);
5341                 else
5342                         seq_printf(seq, " super non-persistent");
5343
5344                 if (mddev->pers) {
5345                         mddev->pers->status (seq, mddev);
5346                         seq_printf(seq, "\n      ");
5347                         if (mddev->pers->sync_request) {
5348                                 if (mddev->curr_resync > 2) {
5349                                         status_resync (seq, mddev);
5350                                         seq_printf(seq, "\n      ");
5351                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5352                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5353                                 else if (mddev->recovery_cp < MaxSector)
5354                                         seq_printf(seq, "\tresync=PENDING\n      ");
5355                         }
5356                 } else
5357                         seq_printf(seq, "\n       ");
5358
5359                 if ((bitmap = mddev->bitmap)) {
5360                         unsigned long chunk_kb;
5361                         unsigned long flags;
5362                         spin_lock_irqsave(&bitmap->lock, flags);
5363                         chunk_kb = bitmap->chunksize >> 10;
5364                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5365                                 "%lu%s chunk",
5366                                 bitmap->pages - bitmap->missing_pages,
5367                                 bitmap->pages,
5368                                 (bitmap->pages - bitmap->missing_pages)
5369                                         << (PAGE_SHIFT - 10),
5370                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5371                                 chunk_kb ? "KB" : "B");
5372                         if (bitmap->file) {
5373                                 seq_printf(seq, ", file: ");
5374                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5375                         }
5376
5377                         seq_printf(seq, "\n");
5378                         spin_unlock_irqrestore(&bitmap->lock, flags);
5379                 }
5380
5381                 seq_printf(seq, "\n");
5382         }
5383         mddev_unlock(mddev);
5384
5385         return 0;
5386 }
5387
5388 static struct seq_operations md_seq_ops = {
5389         .start  = md_seq_start,
5390         .next   = md_seq_next,
5391         .stop   = md_seq_stop,
5392         .show   = md_seq_show,
5393 };
5394
5395 static int md_seq_open(struct inode *inode, struct file *file)
5396 {
5397         int error;
5398         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5399         if (mi == NULL)
5400                 return -ENOMEM;
5401
5402         error = seq_open(file, &md_seq_ops);
5403         if (error)
5404                 kfree(mi);
5405         else {
5406                 struct seq_file *p = file->private_data;
5407                 p->private = mi;
5408                 mi->event = atomic_read(&md_event_count);
5409         }
5410         return error;
5411 }
5412
5413 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5414 {
5415         struct seq_file *m = filp->private_data;
5416         struct mdstat_info *mi = m->private;
5417         int mask;
5418
5419         poll_wait(filp, &md_event_waiters, wait);
5420
5421         /* always allow read */
5422         mask = POLLIN | POLLRDNORM;
5423
5424         if (mi->event != atomic_read(&md_event_count))
5425                 mask |= POLLERR | POLLPRI;
5426         return mask;
5427 }
5428
5429 static const struct file_operations md_seq_fops = {
5430         .owner          = THIS_MODULE,
5431         .open           = md_seq_open,
5432         .read           = seq_read,
5433         .llseek         = seq_lseek,
5434         .release        = seq_release_private,
5435         .poll           = mdstat_poll,
5436 };
5437
5438 int register_md_personality(struct mdk_personality *p)
5439 {
5440         spin_lock(&pers_lock);
5441         list_add_tail(&p->list, &pers_list);
5442         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5443         spin_unlock(&pers_lock);
5444         return 0;
5445 }
5446
5447 int unregister_md_personality(struct mdk_personality *p)
5448 {
5449         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5450         spin_lock(&pers_lock);
5451         list_del_init(&p->list);
5452         spin_unlock(&pers_lock);
5453         return 0;
5454 }
5455
5456 static int is_mddev_idle(mddev_t *mddev)
5457 {
5458         mdk_rdev_t * rdev;
5459         struct list_head *tmp;
5460         int idle;
5461         long curr_events;
5462
5463         idle = 1;
5464         rdev_for_each(rdev, tmp, mddev) {
5465                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5466                 curr_events = disk_stat_read(disk, sectors[0]) +
5467                                 disk_stat_read(disk, sectors[1]) -
5468                                 atomic_read(&disk->sync_io);
5469                 /* sync IO will cause sync_io to increase before the disk_stats
5470                  * as sync_io is counted when a request starts, and
5471                  * disk_stats is counted when it completes.
5472                  * So resync activity will cause curr_events to be smaller than
5473                  * when there was no such activity.
5474                  * non-sync IO will cause disk_stat to increase without
5475                  * increasing sync_io so curr_events will (eventually)
5476                  * be larger than it was before.  Once it becomes
5477                  * substantially larger, the test below will cause
5478                  * the array to appear non-idle, and resync will slow
5479                  * down.
5480                  * If there is a lot of outstanding resync activity when
5481                  * we set last_event to curr_events, then all that activity
5482                  * completing might cause the array to appear non-idle
5483                  * and resync will be slowed down even though there might
5484                  * not have been non-resync activity.  This will only
5485                  * happen once though.  'last_events' will soon reflect
5486                  * the state where there is little or no outstanding
5487                  * resync requests, and further resync activity will
5488                  * always make curr_events less than last_events.
5489                  *
5490                  */
5491                 if (curr_events - rdev->last_events > 4096) {
5492                         rdev->last_events = curr_events;
5493                         idle = 0;
5494                 }
5495         }
5496         return idle;
5497 }
5498
5499 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5500 {
5501         /* another "blocks" (512byte) blocks have been synced */
5502         atomic_sub(blocks, &mddev->recovery_active);
5503         wake_up(&mddev->recovery_wait);
5504         if (!ok) {
5505                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5506                 md_wakeup_thread(mddev->thread);
5507                 // stop recovery, signal do_sync ....
5508         }
5509 }
5510
5511
5512 /* md_write_start(mddev, bi)
5513  * If we need to update some array metadata (e.g. 'active' flag
5514  * in superblock) before writing, schedule a superblock update
5515  * and wait for it to complete.
5516  */
5517 void md_write_start(mddev_t *mddev, struct bio *bi)
5518 {
5519         if (bio_data_dir(bi) != WRITE)
5520                 return;
5521
5522         BUG_ON(mddev->ro == 1);
5523         if (mddev->ro == 2) {
5524                 /* need to switch to read/write */
5525                 mddev->ro = 0;
5526                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5527                 md_wakeup_thread(mddev->thread);
5528                 md_wakeup_thread(mddev->sync_thread);
5529         }
5530         atomic_inc(&mddev->writes_pending);
5531         if (mddev->safemode == 1)
5532                 mddev->safemode = 0;
5533         if (mddev->in_sync) {
5534                 spin_lock_irq(&mddev->write_lock);
5535                 if (mddev->in_sync) {
5536                         mddev->in_sync = 0;
5537                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5538                         md_wakeup_thread(mddev->thread);
5539                 }
5540                 spin_unlock_irq(&mddev->write_lock);
5541                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5542         }
5543         wait_event(mddev->sb_wait,
5544                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5545                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5546 }
5547
5548 void md_write_end(mddev_t *mddev)
5549 {
5550         if (atomic_dec_and_test(&mddev->writes_pending)) {
5551                 if (mddev->safemode == 2)
5552                         md_wakeup_thread(mddev->thread);
5553                 else if (mddev->safemode_delay)
5554                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5555         }
5556 }
5557
5558 /* md_allow_write(mddev)
5559  * Calling this ensures that the array is marked 'active' so that writes
5560  * may proceed without blocking.  It is important to call this before
5561  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5562  * Must be called with mddev_lock held.
5563  */
5564 void md_allow_write(mddev_t *mddev)
5565 {
5566         if (!mddev->pers)
5567                 return;
5568         if (mddev->ro)
5569                 return;
5570         if (!mddev->pers->sync_request)
5571                 return;
5572
5573         spin_lock_irq(&mddev->write_lock);
5574         if (mddev->in_sync) {
5575                 mddev->in_sync = 0;
5576                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5577                 if (mddev->safemode_delay &&
5578                     mddev->safemode == 0)
5579                         mddev->safemode = 1;
5580                 spin_unlock_irq(&mddev->write_lock);
5581                 md_update_sb(mddev, 0);
5582
5583                 sysfs_notify(&mddev->kobj, NULL, "array_state");
5584                 /* wait for the dirty state to be recorded in the metadata */
5585                 wait_event(mddev->sb_wait,
5586                            !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5587                            !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5588         } else
5589                 spin_unlock_irq(&mddev->write_lock);
5590 }
5591 EXPORT_SYMBOL_GPL(md_allow_write);
5592
5593 #define SYNC_MARKS      10
5594 #define SYNC_MARK_STEP  (3*HZ)
5595 void md_do_sync(mddev_t *mddev)
5596 {
5597         mddev_t *mddev2;
5598         unsigned int currspeed = 0,
5599                  window;
5600         sector_t max_sectors,j, io_sectors;
5601         unsigned long mark[SYNC_MARKS];
5602         sector_t mark_cnt[SYNC_MARKS];
5603         int last_mark,m;
5604         struct list_head *tmp;
5605         sector_t last_check;
5606         int skipped = 0;
5607         struct list_head *rtmp;
5608         mdk_rdev_t *rdev;
5609         char *desc;
5610
5611         /* just incase thread restarts... */
5612         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5613                 return;
5614         if (mddev->ro) /* never try to sync a read-only array */
5615                 return;
5616
5617         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5618                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5619                         desc = "data-check";
5620                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5621                         desc = "requested-resync";
5622                 else
5623                         desc = "resync";
5624         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5625                 desc = "reshape";
5626         else
5627                 desc = "recovery";
5628
5629         /* we overload curr_resync somewhat here.
5630          * 0 == not engaged in resync at all
5631          * 2 == checking that there is no conflict with another sync
5632          * 1 == like 2, but have yielded to allow conflicting resync to
5633          *              commense
5634          * other == active in resync - this many blocks
5635          *
5636          * Before starting a resync we must have set curr_resync to
5637          * 2, and then checked that every "conflicting" array has curr_resync
5638          * less than ours.  When we find one that is the same or higher
5639          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5640          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5641          * This will mean we have to start checking from the beginning again.
5642          *
5643          */
5644
5645         do {
5646                 mddev->curr_resync = 2;
5647
5648         try_again:
5649                 if (kthread_should_stop()) {
5650                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5651                         goto skip;
5652                 }
5653                 for_each_mddev(mddev2, tmp) {
5654                         if (mddev2 == mddev)
5655                                 continue;
5656                         if (!mddev->parallel_resync
5657                         &&  mddev2->curr_resync
5658                         &&  match_mddev_units(mddev, mddev2)) {
5659                                 DEFINE_WAIT(wq);
5660                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5661                                         /* arbitrarily yield */
5662                                         mddev->curr_resync = 1;
5663                                         wake_up(&resync_wait);
5664                                 }
5665                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5666                                         /* no need to wait here, we can wait the next
5667                                          * time 'round when curr_resync == 2
5668                                          */
5669                                         continue;
5670                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5671                                 if (!kthread_should_stop() &&
5672                                     mddev2->curr_resync >= mddev->curr_resync) {
5673                                         printk(KERN_INFO "md: delaying %s of %s"
5674                                                " until %s has finished (they"
5675                                                " share one or more physical units)\n",
5676                                                desc, mdname(mddev), mdname(mddev2));
5677                                         mddev_put(mddev2);
5678                                         schedule();
5679                                         finish_wait(&resync_wait, &wq);
5680                                         goto try_again;
5681                                 }
5682                                 finish_wait(&resync_wait, &wq);
5683                         }
5684                 }
5685         } while (mddev->curr_resync < 2);
5686
5687         j = 0;
5688         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5689                 /* resync follows the size requested by the personality,
5690                  * which defaults to physical size, but can be virtual size
5691                  */
5692                 max_sectors = mddev->resync_max_sectors;
5693                 mddev->resync_mismatches = 0;
5694                 /* we don't use the checkpoint if there's a bitmap */
5695                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5696                         j = mddev->resync_min;
5697                 else if (!mddev->bitmap)
5698                         j = mddev->recovery_cp;
5699
5700         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5701                 max_sectors = mddev->size << 1;
5702         else {
5703                 /* recovery follows the physical size of devices */
5704                 max_sectors = mddev->size << 1;
5705                 j = MaxSector;
5706                 rdev_for_each(rdev, rtmp, mddev)
5707                         if (rdev->raid_disk >= 0 &&
5708                             !test_bit(Faulty, &rdev->flags) &&
5709                             !test_bit(In_sync, &rdev->flags) &&
5710                             rdev->recovery_offset < j)
5711                                 j = rdev->recovery_offset;
5712         }
5713
5714         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5715         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5716                 " %d KB/sec/disk.\n", speed_min(mddev));
5717         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5718                "(but not more than %d KB/sec) for %s.\n",
5719                speed_max(mddev), desc);
5720
5721         is_mddev_idle(mddev); /* this also initializes IO event counters */
5722
5723         io_sectors = 0;
5724         for (m = 0; m < SYNC_MARKS; m++) {
5725                 mark[m] = jiffies;
5726                 mark_cnt[m] = io_sectors;
5727         }
5728         last_mark = 0;
5729         mddev->resync_mark = mark[last_mark];
5730         mddev->resync_mark_cnt = mark_cnt[last_mark];
5731
5732         /*
5733          * Tune reconstruction:
5734          */
5735         window = 32*(PAGE_SIZE/512);
5736         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5737                 window/2,(unsigned long long) max_sectors/2);
5738
5739         atomic_set(&mddev->recovery_active, 0);
5740         last_check = 0;
5741
5742         if (j>2) {
5743                 printk(KERN_INFO
5744                        "md: resuming %s of %s from checkpoint.\n",
5745                        desc, mdname(mddev));
5746                 mddev->curr_resync = j;
5747         }
5748
5749         while (j < max_sectors) {
5750                 sector_t sectors;
5751
5752                 skipped = 0;
5753                 if (j >= mddev->resync_max) {
5754                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5755                         wait_event(mddev->recovery_wait,
5756                                    mddev->resync_max > j
5757                                    || kthread_should_stop());
5758                 }
5759                 if (kthread_should_stop())
5760                         goto interrupted;
5761                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5762                                                   currspeed < speed_min(mddev));
5763                 if (sectors == 0) {
5764                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5765                         goto out;
5766                 }
5767
5768                 if (!skipped) { /* actual IO requested */
5769                         io_sectors += sectors;
5770                         atomic_add(sectors, &mddev->recovery_active);
5771                 }
5772
5773                 j += sectors;
5774                 if (j>1) mddev->curr_resync = j;
5775                 mddev->curr_mark_cnt = io_sectors;
5776                 if (last_check == 0)
5777                         /* this is the earliers that rebuilt will be
5778                          * visible in /proc/mdstat
5779                          */
5780                         md_new_event(mddev);
5781
5782                 if (last_check + window > io_sectors || j == max_sectors)
5783                         continue;
5784
5785                 last_check = io_sectors;
5786
5787                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5788                         break;
5789
5790         repeat:
5791                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5792                         /* step marks */
5793                         int next = (last_mark+1) % SYNC_MARKS;
5794
5795                         mddev->resync_mark = mark[next];
5796                         mddev->resync_mark_cnt = mark_cnt[next];
5797                         mark[next] = jiffies;
5798                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5799                         last_mark = next;
5800                 }
5801
5802
5803                 if (kthread_should_stop())
5804                         goto interrupted;
5805
5806
5807                 /*
5808                  * this loop exits only if either when we are slower than
5809                  * the 'hard' speed limit, or the system was IO-idle for
5810                  * a jiffy.
5811                  * the system might be non-idle CPU-wise, but we only care
5812                  * about not overloading the IO subsystem. (things like an
5813                  * e2fsck being done on the RAID array should execute fast)
5814                  */
5815                 blk_unplug(mddev->queue);
5816                 cond_resched();
5817
5818                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5819                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5820
5821                 if (currspeed > speed_min(mddev)) {
5822                         if ((currspeed > speed_max(mddev)) ||
5823                                         !is_mddev_idle(mddev)) {
5824                                 msleep(500);
5825                                 goto repeat;
5826                         }
5827                 }
5828         }
5829         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5830         /*
5831          * this also signals 'finished resyncing' to md_stop
5832          */
5833  out:
5834         blk_unplug(mddev->queue);
5835
5836         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5837
5838         /* tell personality that we are finished */
5839         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5840
5841         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5842             mddev->curr_resync > 2) {
5843                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5844                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5845                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5846                                         printk(KERN_INFO
5847                                                "md: checkpointing %s of %s.\n",
5848                                                desc, mdname(mddev));
5849                                         mddev->recovery_cp = mddev->curr_resync;
5850                                 }
5851                         } else
5852                                 mddev->recovery_cp = MaxSector;
5853                 } else {
5854                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5855                                 mddev->curr_resync = MaxSector;
5856                         rdev_for_each(rdev, rtmp, mddev)
5857                                 if (rdev->raid_disk >= 0 &&
5858                                     !test_bit(Faulty, &rdev->flags) &&
5859                                     !test_bit(In_sync, &rdev->flags) &&
5860                                     rdev->recovery_offset < mddev->curr_resync)
5861                                         rdev->recovery_offset = mddev->curr_resync;
5862                 }
5863         }
5864         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5865
5866  skip:
5867         mddev->curr_resync = 0;
5868         mddev->resync_min = 0;
5869         mddev->resync_max = MaxSector;
5870         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5871         wake_up(&resync_wait);
5872         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5873         md_wakeup_thread(mddev->thread);
5874         return;
5875
5876  interrupted:
5877         /*
5878          * got a signal, exit.
5879          */
5880         printk(KERN_INFO
5881                "md: md_do_sync() got signal ... exiting\n");
5882         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5883         goto out;
5884
5885 }
5886 EXPORT_SYMBOL_GPL(md_do_sync);
5887
5888
5889 static int remove_and_add_spares(mddev_t *mddev)
5890 {
5891         mdk_rdev_t *rdev;
5892         struct list_head *rtmp;
5893         int spares = 0;
5894
5895         rdev_for_each(rdev, rtmp, mddev)
5896                 if (rdev->raid_disk >= 0 &&
5897                     !test_bit(Blocked, &rdev->flags) &&
5898                     (test_bit(Faulty, &rdev->flags) ||
5899                      ! test_bit(In_sync, &rdev->flags)) &&
5900                     atomic_read(&rdev->nr_pending)==0) {
5901                         if (mddev->pers->hot_remove_disk(
5902                                     mddev, rdev->raid_disk)==0) {
5903                                 char nm[20];
5904                                 sprintf(nm,"rd%d", rdev->raid_disk);
5905                                 sysfs_remove_link(&mddev->kobj, nm);
5906                                 rdev->raid_disk = -1;
5907                         }
5908                 }
5909
5910         if (mddev->degraded) {
5911                 rdev_for_each(rdev, rtmp, mddev) {
5912                         if (rdev->raid_disk >= 0 &&
5913                             !test_bit(In_sync, &rdev->flags))
5914                                 spares++;
5915                         if (rdev->raid_disk < 0
5916                             && !test_bit(Faulty, &rdev->flags)) {
5917                                 rdev->recovery_offset = 0;
5918                                 if (mddev->pers->
5919                                     hot_add_disk(mddev, rdev) == 0) {
5920                                         char nm[20];
5921                                         sprintf(nm, "rd%d", rdev->raid_disk);
5922                                         if (sysfs_create_link(&mddev->kobj,
5923                                                               &rdev->kobj, nm))
5924                                                 printk(KERN_WARNING
5925                                                        "md: cannot register "
5926                                                        "%s for %s\n",
5927                                                        nm, mdname(mddev));
5928                                         spares++;
5929                                         md_new_event(mddev);
5930                                 } else
5931                                         break;
5932                         }
5933                 }
5934         }
5935         return spares;
5936 }
5937 /*
5938  * This routine is regularly called by all per-raid-array threads to
5939  * deal with generic issues like resync and super-block update.
5940  * Raid personalities that don't have a thread (linear/raid0) do not
5941  * need this as they never do any recovery or update the superblock.
5942  *
5943  * It does not do any resync itself, but rather "forks" off other threads
5944  * to do that as needed.
5945  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5946  * "->recovery" and create a thread at ->sync_thread.
5947  * When the thread finishes it sets MD_RECOVERY_DONE
5948  * and wakeups up this thread which will reap the thread and finish up.
5949  * This thread also removes any faulty devices (with nr_pending == 0).
5950  *
5951  * The overall approach is:
5952  *  1/ if the superblock needs updating, update it.
5953  *  2/ If a recovery thread is running, don't do anything else.
5954  *  3/ If recovery has finished, clean up, possibly marking spares active.
5955  *  4/ If there are any faulty devices, remove them.
5956  *  5/ If array is degraded, try to add spares devices
5957  *  6/ If array has spares or is not in-sync, start a resync thread.
5958  */
5959 void md_check_recovery(mddev_t *mddev)
5960 {
5961         mdk_rdev_t *rdev;
5962         struct list_head *rtmp;
5963
5964
5965         if (mddev->bitmap)
5966                 bitmap_daemon_work(mddev->bitmap);
5967
5968         if (mddev->ro)
5969                 return;
5970
5971         if (signal_pending(current)) {
5972                 if (mddev->pers->sync_request && !mddev->external) {
5973                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5974                                mdname(mddev));
5975                         mddev->safemode = 2;
5976                 }
5977                 flush_signals(current);
5978         }
5979
5980         if ( ! (
5981                 (mddev->flags && !mddev->external) ||
5982                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5983                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5984                 (mddev->external == 0 && mddev->safemode == 1) ||
5985                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5986                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5987                 ))
5988                 return;
5989
5990         if (mddev_trylock(mddev)) {
5991                 int spares = 0;
5992
5993                 if (!mddev->external) {
5994                         spin_lock_irq(&mddev->write_lock);
5995                         if (mddev->safemode &&
5996                             !atomic_read(&mddev->writes_pending) &&
5997                             !mddev->in_sync &&
5998                             mddev->recovery_cp == MaxSector) {
5999                                 mddev->in_sync = 1;
6000                                 if (mddev->persistent)
6001                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6002                         }
6003                         if (mddev->safemode == 1)
6004                                 mddev->safemode = 0;
6005                         spin_unlock_irq(&mddev->write_lock);
6006                 }
6007
6008                 if (mddev->flags)
6009                         md_update_sb(mddev, 0);
6010
6011
6012                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6013                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6014                         /* resync/recovery still happening */
6015                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6016                         goto unlock;
6017                 }
6018                 if (mddev->sync_thread) {
6019                         /* resync has finished, collect result */
6020                         md_unregister_thread(mddev->sync_thread);
6021                         mddev->sync_thread = NULL;
6022                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6023                                 /* success...*/
6024                                 /* activate any spares */
6025                                 mddev->pers->spare_active(mddev);
6026                         }
6027                         md_update_sb(mddev, 1);
6028
6029                         /* if array is no-longer degraded, then any saved_raid_disk
6030                          * information must be scrapped
6031                          */
6032                         if (!mddev->degraded)
6033                                 rdev_for_each(rdev, rtmp, mddev)
6034                                         rdev->saved_raid_disk = -1;
6035
6036                         mddev->recovery = 0;
6037                         /* flag recovery needed just to double check */
6038                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6039                         md_new_event(mddev);
6040                         goto unlock;
6041                 }
6042                 /* Clear some bits that don't mean anything, but
6043                  * might be left set
6044                  */
6045                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6046                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6047                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6048
6049                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6050                         goto unlock;
6051                 /* no recovery is running.
6052                  * remove any failed drives, then
6053                  * add spares if possible.
6054                  * Spare are also removed and re-added, to allow
6055                  * the personality to fail the re-add.
6056                  */
6057
6058                 if (mddev->reshape_position != MaxSector) {
6059                         if (mddev->pers->check_reshape(mddev) != 0)
6060                                 /* Cannot proceed */
6061                                 goto unlock;
6062                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6063                 } else if ((spares = remove_and_add_spares(mddev))) {
6064                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6065                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6066                 } else if (mddev->recovery_cp < MaxSector) {
6067                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6068                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6069                         /* nothing to be done ... */
6070                         goto unlock;
6071
6072                 if (mddev->pers->sync_request) {
6073                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6074                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6075                                 /* We are adding a device or devices to an array
6076                                  * which has the bitmap stored on all devices.
6077                                  * So make sure all bitmap pages get written
6078                                  */
6079                                 bitmap_write_all(mddev->bitmap);
6080                         }
6081                         mddev->sync_thread = md_register_thread(md_do_sync,
6082                                                                 mddev,
6083                                                                 "%s_resync");
6084                         if (!mddev->sync_thread) {
6085                                 printk(KERN_ERR "%s: could not start resync"
6086                                         " thread...\n",
6087                                         mdname(mddev));
6088                                 /* leave the spares where they are, it shouldn't hurt */
6089                                 mddev->recovery = 0;
6090                         } else
6091                                 md_wakeup_thread(mddev->sync_thread);
6092                         md_new_event(mddev);
6093                 }
6094         unlock:
6095                 mddev_unlock(mddev);
6096         }
6097 }
6098
6099 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6100 {
6101         sysfs_notify(&rdev->kobj, NULL, "state");
6102         wait_event_timeout(rdev->blocked_wait,
6103                            !test_bit(Blocked, &rdev->flags),
6104                            msecs_to_jiffies(5000));
6105         rdev_dec_pending(rdev, mddev);
6106 }
6107 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6108
6109 static int md_notify_reboot(struct notifier_block *this,
6110                             unsigned long code, void *x)
6111 {
6112         struct list_head *tmp;
6113         mddev_t *mddev;
6114
6115         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6116
6117                 printk(KERN_INFO "md: stopping all md devices.\n");
6118
6119                 for_each_mddev(mddev, tmp)
6120                         if (mddev_trylock(mddev)) {
6121                                 do_md_stop (mddev, 1);
6122                                 mddev_unlock(mddev);
6123                         }
6124                 /*
6125                  * certain more exotic SCSI devices are known to be
6126                  * volatile wrt too early system reboots. While the
6127                  * right place to handle this issue is the given
6128                  * driver, we do want to have a safe RAID driver ...
6129                  */
6130                 mdelay(1000*1);
6131         }
6132         return NOTIFY_DONE;
6133 }
6134
6135 static struct notifier_block md_notifier = {
6136         .notifier_call  = md_notify_reboot,
6137         .next           = NULL,
6138         .priority       = INT_MAX, /* before any real devices */
6139 };
6140
6141 static void md_geninit(void)
6142 {
6143         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6144
6145         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6146 }
6147
6148 static int __init md_init(void)
6149 {
6150         if (register_blkdev(MAJOR_NR, "md"))
6151                 return -1;
6152         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6153                 unregister_blkdev(MAJOR_NR, "md");
6154                 return -1;
6155         }
6156         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6157                             md_probe, NULL, NULL);
6158         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6159                             md_probe, NULL, NULL);
6160
6161         register_reboot_notifier(&md_notifier);
6162         raid_table_header = register_sysctl_table(raid_root_table);
6163
6164         md_geninit();
6165         return (0);
6166 }
6167
6168
6169 #ifndef MODULE
6170
6171 /*
6172  * Searches all registered partitions for autorun RAID arrays
6173  * at boot time.
6174  */
6175
6176 static LIST_HEAD(all_detected_devices);
6177 struct detected_devices_node {
6178         struct list_head list;
6179         dev_t dev;
6180 };
6181
6182 void md_autodetect_dev(dev_t dev)
6183 {
6184         struct detected_devices_node *node_detected_dev;
6185
6186         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6187         if (node_detected_dev) {
6188                 node_detected_dev->dev = dev;
6189                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6190         } else {
6191                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6192                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6193         }
6194 }
6195
6196
6197 static void autostart_arrays(int part)
6198 {
6199         mdk_rdev_t *rdev;
6200         struct detected_devices_node *node_detected_dev;
6201         dev_t dev;
6202         int i_scanned, i_passed;
6203
6204         i_scanned = 0;
6205         i_passed = 0;
6206
6207         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6208
6209         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6210                 i_scanned++;
6211                 node_detected_dev = list_entry(all_detected_devices.next,
6212                                         struct detected_devices_node, list);
6213                 list_del(&node_detected_dev->list);
6214                 dev = node_detected_dev->dev;
6215                 kfree(node_detected_dev);
6216                 rdev = md_import_device(dev,0, 90);
6217                 if (IS_ERR(rdev))
6218                         continue;
6219
6220                 if (test_bit(Faulty, &rdev->flags)) {
6221                         MD_BUG();
6222                         continue;
6223                 }
6224                 set_bit(AutoDetected, &rdev->flags);
6225                 list_add(&rdev->same_set, &pending_raid_disks);
6226                 i_passed++;
6227         }
6228
6229         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6230                                                 i_scanned, i_passed);
6231
6232         autorun_devices(part);
6233 }
6234
6235 #endif /* !MODULE */
6236
6237 static __exit void md_exit(void)
6238 {
6239         mddev_t *mddev;
6240         struct list_head *tmp;
6241
6242         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6243         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6244
6245         unregister_blkdev(MAJOR_NR,"md");
6246         unregister_blkdev(mdp_major, "mdp");
6247         unregister_reboot_notifier(&md_notifier);
6248         unregister_sysctl_table(raid_table_header);
6249         remove_proc_entry("mdstat", NULL);
6250         for_each_mddev(mddev, tmp) {
6251                 struct gendisk *disk = mddev->gendisk;
6252                 if (!disk)
6253                         continue;
6254                 export_array(mddev);
6255                 del_gendisk(disk);
6256                 put_disk(disk);
6257                 mddev->gendisk = NULL;
6258                 mddev_put(mddev);
6259         }
6260 }
6261
6262 subsys_initcall(md_init);
6263 module_exit(md_exit)
6264
6265 static int get_ro(char *buffer, struct kernel_param *kp)
6266 {
6267         return sprintf(buffer, "%d", start_readonly);
6268 }
6269 static int set_ro(const char *val, struct kernel_param *kp)
6270 {
6271         char *e;
6272         int num = simple_strtoul(val, &e, 10);
6273         if (*val && (*e == '\0' || *e == '\n')) {
6274                 start_readonly = num;
6275                 return 0;
6276         }
6277         return -EINVAL;
6278 }
6279
6280 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6281 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6282
6283
6284 EXPORT_SYMBOL(register_md_personality);
6285 EXPORT_SYMBOL(unregister_md_personality);
6286 EXPORT_SYMBOL(md_error);
6287 EXPORT_SYMBOL(md_done_sync);
6288 EXPORT_SYMBOL(md_write_start);
6289 EXPORT_SYMBOL(md_write_end);
6290 EXPORT_SYMBOL(md_register_thread);
6291 EXPORT_SYMBOL(md_unregister_thread);
6292 EXPORT_SYMBOL(md_wakeup_thread);
6293 EXPORT_SYMBOL(md_check_recovery);
6294 MODULE_LICENSE("GPL");
6295 MODULE_ALIAS("md");
6296 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);