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