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/config.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/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45 #include <linux/poll.h>
  46
  47 #include <linux/init.h>
  48
  49 #include <linux/file.h>
  50
  51 #ifdef CONFIG_KMOD
  52 #include <linux/kmod.h>
  53 #endif
  54
  55 #include <asm/unaligned.h>
  56
  57 #define MAJOR_NR MD_MAJOR
  58 #define MD_DRIVER
  59
  60 /* 63 partitions with the alternate major number (mdp) */
  61 #define MdpMinorShift 6
  62
  63 #define DEBUG 0
  64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  65
  66
  67 #ifndef MODULE
  68 static void autostart_arrays (int part);
  69 #endif
  70
  71 static LIST_HEAD(pers_list);
  72 static DEFINE_SPINLOCK(pers_lock);
  73
  74 /*
  75  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  76  * is 1000 KB/sec, so the extra system load does not show up that much.
  77  * Increase it if you want to have more _guaranteed_ speed. Note that
  78  * the RAID driver will use the maximum available bandwidth if the IO
  79  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  80  * speed limit - in case reconstruction slows down your system despite
  81  * idle IO detection.
  82  *
  83  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  84  * or /sys/block/mdX/md/sync_speed_{min,max}
  85  */
  86
  87 static int sysctl_speed_limit_min = 1000;
  88 static int sysctl_speed_limit_max = 200000;
  89 static inline int speed_min(mddev_t *mddev)
  90 {
  91         return mddev->sync_speed_min ?
  92                 mddev->sync_speed_min : sysctl_speed_limit_min;
  93 }
  94
  95 static inline int speed_max(mddev_t *mddev)
  96 {
  97         return mddev->sync_speed_max ?
  98                 mddev->sync_speed_max : sysctl_speed_limit_max;
  99 }
 100
 101 static struct ctl_table_header *raid_table_header;
 102
 103 static ctl_table raid_table[] = {
 104         {
 105                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 106                 .procname       = "speed_limit_min",
 107                 .data           = &sysctl_speed_limit_min,
 108                 .maxlen         = sizeof(int),
 109                 .mode           = 0644,
 110                 .proc_handler   = &proc_dointvec,
 111         },
 112         {
 113                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 114                 .procname       = "speed_limit_max",
 115                 .data           = &sysctl_speed_limit_max,
 116                 .maxlen         = sizeof(int),
 117                 .mode           = 0644,
 118                 .proc_handler   = &proc_dointvec,
 119         },
 120         { .ctl_name = 0 }
 121 };
 122
 123 static ctl_table raid_dir_table[] = {
 124         {
 125                 .ctl_name       = DEV_RAID,
 126                 .procname       = "raid",
 127                 .maxlen         = 0,
 128                 .mode           = 0555,
 129                 .child          = raid_table,
 130         },
 131         { .ctl_name = 0 }
 132 };
 133
 134 static ctl_table raid_root_table[] = {
 135         {
 136                 .ctl_name       = CTL_DEV,
 137                 .procname       = "dev",
 138                 .maxlen         = 0,
 139                 .mode           = 0555,
 140                 .child          = raid_dir_table,
 141         },
 142         { .ctl_name = 0 }
 143 };
 144
 145 static struct block_device_operations md_fops;
 146
 147 static int start_readonly;
 148
 149 /*
 150  * We have a system wide 'event count' that is incremented
 151  * on any 'interesting' event, and readers of /proc/mdstat
 152  * can use 'poll' or 'select' to find out when the event
 153  * count increases.
 154  *
 155  * Events are:
 156  *  start array, stop array, error, add device, remove device,
 157  *  start build, activate spare
 158  */
 159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 160 static atomic_t md_event_count;
 161 static void md_new_event(mddev_t *mddev)
 162 {
 163         atomic_inc(&md_event_count);
 164         wake_up(&md_event_waiters);
 165 }
 166
 167 /*
 168  * Enables to iterate over all existing md arrays
 169  * all_mddevs_lock protects this list.
 170  */
 171 static LIST_HEAD(all_mddevs);
 172 static DEFINE_SPINLOCK(all_mddevs_lock);
 173
 174
 175 /*
 176  * iterates through all used mddevs in the system.
 177  * We take care to grab the all_mddevs_lock whenever navigating
 178  * the list, and to always hold a refcount when unlocked.
 179  * Any code which breaks out of this loop while own
 180  * a reference to the current mddev and must mddev_put it.
 181  */
 182 #define ITERATE_MDDEV(mddev,tmp)                                        \
 183                                                                         \
 184         for (({ spin_lock(&all_mddevs_lock);                            \
 185                 tmp = all_mddevs.next;                                  \
 186                 mddev = NULL;});                                        \
 187              ({ if (tmp != &all_mddevs)                                 \
 188                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 189                 spin_unlock(&all_mddevs_lock);                          \
 190                 if (mddev) mddev_put(mddev);                            \
 191                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 192                 tmp != &all_mddevs;});                                  \
 193              ({ spin_lock(&all_mddevs_lock);                            \
 194                 tmp = tmp->next;})                                      \
 195                 )
 196
 197
 198 static int md_fail_request (request_queue_t *q, struct bio *bio)
 199 {
 200         bio_io_error(bio, bio->bi_size);
 201         return 0;
 202 }
 203
 204 static inline mddev_t *mddev_get(mddev_t *mddev)
 205 {
 206         atomic_inc(&mddev->active);
 207         return mddev;
 208 }
 209
 210 static void mddev_put(mddev_t *mddev)
 211 {
 212         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 213                 return;
 214         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 215                 list_del(&mddev->all_mddevs);
 216                 blk_put_queue(mddev->queue);
 217                 kobject_unregister(&mddev->kobj);
 218         }
 219         spin_unlock(&all_mddevs_lock);
 220 }
 221
 222 static mddev_t * mddev_find(dev_t unit)
 223 {
 224         mddev_t *mddev, *new = NULL;
 225
 226  retry:
 227         spin_lock(&all_mddevs_lock);
 228         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 229                 if (mddev->unit == unit) {
 230                         mddev_get(mddev);
 231                         spin_unlock(&all_mddevs_lock);
 232                         kfree(new);
 233                         return mddev;
 234                 }
 235
 236         if (new) {
 237                 list_add(&new->all_mddevs, &all_mddevs);
 238                 spin_unlock(&all_mddevs_lock);
 239                 return new;
 240         }
 241         spin_unlock(&all_mddevs_lock);
 242
 243         new = kzalloc(sizeof(*new), GFP_KERNEL);
 244         if (!new)
 245                 return NULL;
 246
 247         new->unit = unit;
 248         if (MAJOR(unit) == MD_MAJOR)
 249                 new->md_minor = MINOR(unit);
 250         else
 251                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 252
 253         init_MUTEX(&new->reconfig_sem);
 254         INIT_LIST_HEAD(&new->disks);
 255         INIT_LIST_HEAD(&new->all_mddevs);
 256         init_timer(&new->safemode_timer);
 257         atomic_set(&new->active, 1);
 258         spin_lock_init(&new->write_lock);
 259         init_waitqueue_head(&new->sb_wait);
 260
 261         new->queue = blk_alloc_queue(GFP_KERNEL);
 262         if (!new->queue) {
 263                 kfree(new);
 264                 return NULL;
 265         }
 266
 267         blk_queue_make_request(new->queue, md_fail_request);
 268
 269         goto retry;
 270 }
 271
 272 static inline int mddev_lock(mddev_t * mddev)
 273 {
 274         return down_interruptible(&mddev->reconfig_sem);
 275 }
 276
 277 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 278 {
 279         down(&mddev->reconfig_sem);
 280 }
 281
 282 static inline int mddev_trylock(mddev_t * mddev)
 283 {
 284         return down_trylock(&mddev->reconfig_sem);
 285 }
 286
 287 static inline void mddev_unlock(mddev_t * mddev)
 288 {
 289         up(&mddev->reconfig_sem);
 290
 291         md_wakeup_thread(mddev->thread);
 292 }
 293
 294 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 295 {
 296         mdk_rdev_t * rdev;
 297         struct list_head *tmp;
 298
 299         ITERATE_RDEV(mddev,rdev,tmp) {
 300                 if (rdev->desc_nr == nr)
 301                         return rdev;
 302         }
 303         return NULL;
 304 }
 305
 306 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 307 {
 308         struct list_head *tmp;
 309         mdk_rdev_t *rdev;
 310
 311         ITERATE_RDEV(mddev,rdev,tmp) {
 312                 if (rdev->bdev->bd_dev == dev)
 313                         return rdev;
 314         }
 315         return NULL;
 316 }
 317
 318 static struct mdk_personality *find_pers(int level, char *clevel)
 319 {
 320         struct mdk_personality *pers;
 321         list_for_each_entry(pers, &pers_list, list) {
 322                 if (level != LEVEL_NONE && pers->level == level)
 323                         return pers;
 324                 if (strcmp(pers->name, clevel)==0)
 325                         return pers;
 326         }
 327         return NULL;
 328 }
 329
 330 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 331 {
 332         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 333         return MD_NEW_SIZE_BLOCKS(size);
 334 }
 335
 336 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 337 {
 338         sector_t size;
 339
 340         size = rdev->sb_offset;
 341
 342         if (chunk_size)
 343                 size &= ~((sector_t)chunk_size/1024 - 1);
 344         return size;
 345 }
 346
 347 static int alloc_disk_sb(mdk_rdev_t * rdev)
 348 {
 349         if (rdev->sb_page)
 350                 MD_BUG();
 351
 352         rdev->sb_page = alloc_page(GFP_KERNEL);
 353         if (!rdev->sb_page) {
 354                 printk(KERN_ALERT "md: out of memory.\n");
 355                 return -EINVAL;
 356         }
 357
 358         return 0;
 359 }
 360
 361 static void free_disk_sb(mdk_rdev_t * rdev)
 362 {
 363         if (rdev->sb_page) {
 364                 put_page(rdev->sb_page);
 365                 rdev->sb_loaded = 0;
 366                 rdev->sb_page = NULL;
 367                 rdev->sb_offset = 0;
 368                 rdev->size = 0;
 369         }
 370 }
 371
 372
 373 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 374 {
 375         mdk_rdev_t *rdev = bio->bi_private;
 376         mddev_t *mddev = rdev->mddev;
 377         if (bio->bi_size)
 378                 return 1;
 379
 380         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 381                 md_error(mddev, rdev);
 382
 383         if (atomic_dec_and_test(&mddev->pending_writes))
 384                 wake_up(&mddev->sb_wait);
 385         bio_put(bio);
 386         return 0;
 387 }
 388
 389 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 390 {
 391         struct bio *bio2 = bio->bi_private;
 392         mdk_rdev_t *rdev = bio2->bi_private;
 393         mddev_t *mddev = rdev->mddev;
 394         if (bio->bi_size)
 395                 return 1;
 396
 397         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 398             error == -EOPNOTSUPP) {
 399                 unsigned long flags;
 400                 /* barriers don't appear to be supported :-( */
 401                 set_bit(BarriersNotsupp, &rdev->flags);
 402                 mddev->barriers_work = 0;
 403                 spin_lock_irqsave(&mddev->write_lock, flags);
 404                 bio2->bi_next = mddev->biolist;
 405                 mddev->biolist = bio2;
 406                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 407                 wake_up(&mddev->sb_wait);
 408                 bio_put(bio);
 409                 return 0;
 410         }
 411         bio_put(bio2);
 412         bio->bi_private = rdev;
 413         return super_written(bio, bytes_done, error);
 414 }
 415
 416 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 417                    sector_t sector, int size, struct page *page)
 418 {
 419         /* write first size bytes of page to sector of rdev
 420          * Increment mddev->pending_writes before returning
 421          * and decrement it on completion, waking up sb_wait
 422          * if zero is reached.
 423          * If an error occurred, call md_error
 424          *
 425          * As we might need to resubmit the request if BIO_RW_BARRIER
 426          * causes ENOTSUPP, we allocate a spare bio...
 427          */
 428         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 429         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 430
 431         bio->bi_bdev = rdev->bdev;
 432         bio->bi_sector = sector;
 433         bio_add_page(bio, page, size, 0);
 434         bio->bi_private = rdev;
 435         bio->bi_end_io = super_written;
 436         bio->bi_rw = rw;
 437
 438         atomic_inc(&mddev->pending_writes);
 439         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 440                 struct bio *rbio;
 441                 rw |= (1<<BIO_RW_BARRIER);
 442                 rbio = bio_clone(bio, GFP_NOIO);
 443                 rbio->bi_private = bio;
 444                 rbio->bi_end_io = super_written_barrier;
 445                 submit_bio(rw, rbio);
 446         } else
 447                 submit_bio(rw, bio);
 448 }
 449
 450 void md_super_wait(mddev_t *mddev)
 451 {
 452         /* wait for all superblock writes that were scheduled to complete.
 453          * if any had to be retried (due to BARRIER problems), retry them
 454          */
 455         DEFINE_WAIT(wq);
 456         for(;;) {
 457                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 458                 if (atomic_read(&mddev->pending_writes)==0)
 459                         break;
 460                 while (mddev->biolist) {
 461                         struct bio *bio;
 462                         spin_lock_irq(&mddev->write_lock);
 463                         bio = mddev->biolist;
 464                         mddev->biolist = bio->bi_next ;
 465                         bio->bi_next = NULL;
 466                         spin_unlock_irq(&mddev->write_lock);
 467                         submit_bio(bio->bi_rw, bio);
 468                 }
 469                 schedule();
 470         }
 471         finish_wait(&mddev->sb_wait, &wq);
 472 }
 473
 474 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 475 {
 476         if (bio->bi_size)
 477                 return 1;
 478
 479         complete((struct completion*)bio->bi_private);
 480         return 0;
 481 }
 482
 483 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 484                    struct page *page, int rw)
 485 {
 486         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 487         struct completion event;
 488         int ret;
 489
 490         rw |= (1 << BIO_RW_SYNC);
 491
 492         bio->bi_bdev = bdev;
 493         bio->bi_sector = sector;
 494         bio_add_page(bio, page, size, 0);
 495         init_completion(&event);
 496         bio->bi_private = &event;
 497         bio->bi_end_io = bi_complete;
 498         submit_bio(rw, bio);
 499         wait_for_completion(&event);
 500
 501         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 502         bio_put(bio);
 503         return ret;
 504 }
 505 EXPORT_SYMBOL_GPL(sync_page_io);
 506
 507 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 508 {
 509         char b[BDEVNAME_SIZE];
 510         if (!rdev->sb_page) {
 511                 MD_BUG();
 512                 return -EINVAL;
 513         }
 514         if (rdev->sb_loaded)
 515                 return 0;
 516
 517
 518         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 519                 goto fail;
 520         rdev->sb_loaded = 1;
 521         return 0;
 522
 523 fail:
 524         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 525                 bdevname(rdev->bdev,b));
 526         return -EINVAL;
 527 }
 528
 529 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 530 {
 531         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 532                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 533                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 534                 (sb1->set_uuid3 == sb2->set_uuid3))
 535
 536                 return 1;
 537
 538         return 0;
 539 }
 540
 541
 542 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 543 {
 544         int ret;
 545         mdp_super_t *tmp1, *tmp2;
 546
 547         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 548         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 549
 550         if (!tmp1 || !tmp2) {
 551                 ret = 0;
 552                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 553                 goto abort;
 554         }
 555
 556         *tmp1 = *sb1;
 557         *tmp2 = *sb2;
 558
 559         /*
 560          * nr_disks is not constant
 561          */
 562         tmp1->nr_disks = 0;
 563         tmp2->nr_disks = 0;
 564
 565         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 566                 ret = 0;
 567         else
 568                 ret = 1;
 569
 570 abort:
 571         kfree(tmp1);
 572         kfree(tmp2);
 573         return ret;
 574 }
 575
 576 static unsigned int calc_sb_csum(mdp_super_t * sb)
 577 {
 578         unsigned int disk_csum, csum;
 579
 580         disk_csum = sb->sb_csum;
 581         sb->sb_csum = 0;
 582         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 583         sb->sb_csum = disk_csum;
 584         return csum;
 585 }
 586
 587
 588 /*
 589  * Handle superblock details.
 590  * We want to be able to handle multiple superblock formats
 591  * so we have a common interface to them all, and an array of
 592  * different handlers.
 593  * We rely on user-space to write the initial superblock, and support
 594  * reading and updating of superblocks.
 595  * Interface methods are:
 596  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 597  *      loads and validates a superblock on dev.
 598  *      if refdev != NULL, compare superblocks on both devices
 599  *    Return:
 600  *      0 - dev has a superblock that is compatible with refdev
 601  *      1 - dev has a superblock that is compatible and newer than refdev
 602  *          so dev should be used as the refdev in future
 603  *     -EINVAL superblock incompatible or invalid
 604  *     -othererror e.g. -EIO
 605  *
 606  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 607  *      Verify that dev is acceptable into mddev.
 608  *       The first time, mddev->raid_disks will be 0, and data from
 609  *       dev should be merged in.  Subsequent calls check that dev
 610  *       is new enough.  Return 0 or -EINVAL
 611  *
 612  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 613  *     Update the superblock for rdev with data in mddev
 614  *     This does not write to disc.
 615  *
 616  */
 617
 618 struct super_type  {
 619         char            *name;
 620         struct module   *owner;
 621         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 622         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 623         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 624 };
 625
 626 /*
 627  * load_super for 0.90.0
 628  */
 629 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 630 {
 631         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 632         mdp_super_t *sb;
 633         int ret;
 634         sector_t sb_offset;
 635
 636         /*
 637          * Calculate the position of the superblock,
 638          * it's at the end of the disk.
 639          *
 640          * It also happens to be a multiple of 4Kb.
 641          */
 642         sb_offset = calc_dev_sboffset(rdev->bdev);
 643         rdev->sb_offset = sb_offset;
 644
 645         ret = read_disk_sb(rdev, MD_SB_BYTES);
 646         if (ret) return ret;
 647
 648         ret = -EINVAL;
 649
 650         bdevname(rdev->bdev, b);
 651         sb = (mdp_super_t*)page_address(rdev->sb_page);
 652
 653         if (sb->md_magic != MD_SB_MAGIC) {
 654                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 655                        b);
 656                 goto abort;
 657         }
 658
 659         if (sb->major_version != 0 ||
 660             sb->minor_version != 90) {
 661                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 662                         sb->major_version, sb->minor_version,
 663                         b);
 664                 goto abort;
 665         }
 666
 667         if (sb->raid_disks <= 0)
 668                 goto abort;
 669
 670         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 671                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 672                         b);
 673                 goto abort;
 674         }
 675
 676         rdev->preferred_minor = sb->md_minor;
 677         rdev->data_offset = 0;
 678         rdev->sb_size = MD_SB_BYTES;
 679
 680         if (sb->level == LEVEL_MULTIPATH)
 681                 rdev->desc_nr = -1;
 682         else
 683                 rdev->desc_nr = sb->this_disk.number;
 684
 685         if (refdev == 0)
 686                 ret = 1;
 687         else {
 688                 __u64 ev1, ev2;
 689                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 690                 if (!uuid_equal(refsb, sb)) {
 691                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 692                                 b, bdevname(refdev->bdev,b2));
 693                         goto abort;
 694                 }
 695                 if (!sb_equal(refsb, sb)) {
 696                         printk(KERN_WARNING "md: %s has same UUID"
 697                                " but different superblock to %s\n",
 698                                b, bdevname(refdev->bdev, b2));
 699                         goto abort;
 700                 }
 701                 ev1 = md_event(sb);
 702                 ev2 = md_event(refsb);
 703                 if (ev1 > ev2)
 704                         ret = 1;
 705                 else
 706                         ret = 0;
 707         }
 708         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 709
 710         if (rdev->size < sb->size && sb->level > 1)
 711                 /* "this cannot possibly happen" ... */
 712                 ret = -EINVAL;
 713
 714  abort:
 715         return ret;
 716 }
 717
 718 /*
 719  * validate_super for 0.90.0
 720  */
 721 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 722 {
 723         mdp_disk_t *desc;
 724         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 725
 726         rdev->raid_disk = -1;
 727         rdev->flags = 0;
 728         if (mddev->raid_disks == 0) {
 729                 mddev->major_version = 0;
 730                 mddev->minor_version = sb->minor_version;
 731                 mddev->patch_version = sb->patch_version;
 732                 mddev->persistent = ! sb->not_persistent;
 733                 mddev->chunk_size = sb->chunk_size;
 734                 mddev->ctime = sb->ctime;
 735                 mddev->utime = sb->utime;
 736                 mddev->level = sb->level;
 737                 mddev->clevel[0] = 0;
 738                 mddev->layout = sb->layout;
 739                 mddev->raid_disks = sb->raid_disks;
 740                 mddev->size = sb->size;
 741                 mddev->events = md_event(sb);
 742                 mddev->bitmap_offset = 0;
 743                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 744
 745                 if (sb->state & (1<<MD_SB_CLEAN))
 746                         mddev->recovery_cp = MaxSector;
 747                 else {
 748                         if (sb->events_hi == sb->cp_events_hi &&
 749                                 sb->events_lo == sb->cp_events_lo) {
 750                                 mddev->recovery_cp = sb->recovery_cp;
 751                         } else
 752                                 mddev->recovery_cp = 0;
 753                 }
 754
 755                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 756                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 757                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 758                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 759
 760                 mddev->max_disks = MD_SB_DISKS;
 761
 762                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 763                     mddev->bitmap_file == NULL) {
 764                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
 765                             && mddev->level != 10) {
 766                                 /* FIXME use a better test */
 767                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
 768                                 return -EINVAL;
 769                         }
 770                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 771                 }
 772
 773         } else if (mddev->pers == NULL) {
 774                 /* Insist on good event counter while assembling */
 775                 __u64 ev1 = md_event(sb);
 776                 ++ev1;
 777                 if (ev1 < mddev->events)
 778                         return -EINVAL;
 779         } else if (mddev->bitmap) {
 780                 /* if adding to array with a bitmap, then we can accept an
 781                  * older device ... but not too old.
 782                  */
 783                 __u64 ev1 = md_event(sb);
 784                 if (ev1 < mddev->bitmap->events_cleared)
 785                         return 0;
 786         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 787                 return 0;
 788
 789         if (mddev->level != LEVEL_MULTIPATH) {
 790                 desc = sb->disks + rdev->desc_nr;
 791
 792                 if (desc->state & (1<<MD_DISK_FAULTY))
 793                         set_bit(Faulty, &rdev->flags);
 794                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 795                          desc->raid_disk < mddev->raid_disks) {
 796                         set_bit(In_sync, &rdev->flags);
 797                         rdev->raid_disk = desc->raid_disk;
 798                 }
 799                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 800                         set_bit(WriteMostly, &rdev->flags);
 801         } else /* MULTIPATH are always insync */
 802                 set_bit(In_sync, &rdev->flags);
 803         return 0;
 804 }
 805
 806 /*
 807  * sync_super for 0.90.0
 808  */
 809 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 810 {
 811         mdp_super_t *sb;
 812         struct list_head *tmp;
 813         mdk_rdev_t *rdev2;
 814         int next_spare = mddev->raid_disks;
 815
 816
 817         /* make rdev->sb match mddev data..
 818          *
 819          * 1/ zero out disks
 820          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 821          * 3/ any empty disks < next_spare become removed
 822          *
 823          * disks[0] gets initialised to REMOVED because
 824          * we cannot be sure from other fields if it has
 825          * been initialised or not.
 826          */
 827         int i;
 828         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 829
 830         rdev->sb_size = MD_SB_BYTES;
 831
 832         sb = (mdp_super_t*)page_address(rdev->sb_page);
 833
 834         memset(sb, 0, sizeof(*sb));
 835
 836         sb->md_magic = MD_SB_MAGIC;
 837         sb->major_version = mddev->major_version;
 838         sb->minor_version = mddev->minor_version;
 839         sb->patch_version = mddev->patch_version;
 840         sb->gvalid_words  = 0; /* ignored */
 841         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 842         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 843         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 844         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 845
 846         sb->ctime = mddev->ctime;
 847         sb->level = mddev->level;
 848         sb->size  = mddev->size;
 849         sb->raid_disks = mddev->raid_disks;
 850         sb->md_minor = mddev->md_minor;
 851         sb->not_persistent = !mddev->persistent;
 852         sb->utime = mddev->utime;
 853         sb->state = 0;
 854         sb->events_hi = (mddev->events>>32);
 855         sb->events_lo = (u32)mddev->events;
 856
 857         if (mddev->in_sync)
 858         {
 859                 sb->recovery_cp = mddev->recovery_cp;
 860                 sb->cp_events_hi = (mddev->events>>32);
 861                 sb->cp_events_lo = (u32)mddev->events;
 862                 if (mddev->recovery_cp == MaxSector)
 863                         sb->state = (1<< MD_SB_CLEAN);
 864         } else
 865                 sb->recovery_cp = 0;
 866
 867         sb->layout = mddev->layout;
 868         sb->chunk_size = mddev->chunk_size;
 869
 870         if (mddev->bitmap && mddev->bitmap_file == NULL)
 871                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 872
 873         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 874         ITERATE_RDEV(mddev,rdev2,tmp) {
 875                 mdp_disk_t *d;
 876                 int desc_nr;
 877                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 878                     && !test_bit(Faulty, &rdev2->flags))
 879                         desc_nr = rdev2->raid_disk;
 880                 else
 881                         desc_nr = next_spare++;
 882                 rdev2->desc_nr = desc_nr;
 883                 d = &sb->disks[rdev2->desc_nr];
 884                 nr_disks++;
 885                 d->number = rdev2->desc_nr;
 886                 d->major = MAJOR(rdev2->bdev->bd_dev);
 887                 d->minor = MINOR(rdev2->bdev->bd_dev);
 888                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 889                     && !test_bit(Faulty, &rdev2->flags))
 890                         d->raid_disk = rdev2->raid_disk;
 891                 else
 892                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 893                 if (test_bit(Faulty, &rdev2->flags)) {
 894                         d->state = (1<<MD_DISK_FAULTY);
 895                         failed++;
 896                 } else if (test_bit(In_sync, &rdev2->flags)) {
 897                         d->state = (1<<MD_DISK_ACTIVE);
 898                         d->state |= (1<<MD_DISK_SYNC);
 899                         active++;
 900                         working++;
 901                 } else {
 902                         d->state = 0;
 903                         spare++;
 904                         working++;
 905                 }
 906                 if (test_bit(WriteMostly, &rdev2->flags))
 907                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 908         }
 909         /* now set the "removed" and "faulty" bits on any missing devices */
 910         for (i=0 ; i < mddev->raid_disks ; i++) {
 911                 mdp_disk_t *d = &sb->disks[i];
 912                 if (d->state == 0 && d->number == 0) {
 913                         d->number = i;
 914                         d->raid_disk = i;
 915                         d->state = (1<<MD_DISK_REMOVED);
 916                         d->state |= (1<<MD_DISK_FAULTY);
 917                         failed++;
 918                 }
 919         }
 920         sb->nr_disks = nr_disks;
 921         sb->active_disks = active;
 922         sb->working_disks = working;
 923         sb->failed_disks = failed;
 924         sb->spare_disks = spare;
 925
 926         sb->this_disk = sb->disks[rdev->desc_nr];
 927         sb->sb_csum = calc_sb_csum(sb);
 928 }
 929
 930 /*
 931  * version 1 superblock
 932  */
 933
 934 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 935 {
 936         unsigned int disk_csum, csum;
 937         unsigned long long newcsum;
 938         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 939         unsigned int *isuper = (unsigned int*)sb;
 940         int i;
 941
 942         disk_csum = sb->sb_csum;
 943         sb->sb_csum = 0;
 944         newcsum = 0;
 945         for (i=0; size>=4; size -= 4 )
 946                 newcsum += le32_to_cpu(*isuper++);
 947
 948         if (size == 2)
 949                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 950
 951         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 952         sb->sb_csum = disk_csum;
 953         return cpu_to_le32(csum);
 954 }
 955
 956 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 957 {
 958         struct mdp_superblock_1 *sb;
 959         int ret;
 960         sector_t sb_offset;
 961         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 962         int bmask;
 963
 964         /*
 965          * Calculate the position of the superblock.
 966          * It is always aligned to a 4K boundary and
 967          * depeding on minor_version, it can be:
 968          * 0: At least 8K, but less than 12K, from end of device
 969          * 1: At start of device
 970          * 2: 4K from start of device.
 971          */
 972         switch(minor_version) {
 973         case 0:
 974                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 975                 sb_offset -= 8*2;
 976                 sb_offset &= ~(sector_t)(4*2-1);
 977                 /* convert from sectors to K */
 978                 sb_offset /= 2;
 979                 break;
 980         case 1:
 981                 sb_offset = 0;
 982                 break;
 983         case 2:
 984                 sb_offset = 4;
 985                 break;
 986         default:
 987                 return -EINVAL;
 988         }
 989         rdev->sb_offset = sb_offset;
 990
 991         /* superblock is rarely larger than 1K, but it can be larger,
 992          * and it is safe to read 4k, so we do that
 993          */
 994         ret = read_disk_sb(rdev, 4096);
 995         if (ret) return ret;
 996
 997
 998         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 999
1000         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1001             sb->major_version != cpu_to_le32(1) ||
1002             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1003             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1004             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1005                 return -EINVAL;
1006
1007         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1008                 printk("md: invalid superblock checksum on %s\n",
1009                         bdevname(rdev->bdev,b));
1010                 return -EINVAL;
1011         }
1012         if (le64_to_cpu(sb->data_size) < 10) {
1013                 printk("md: data_size too small on %s\n",
1014                        bdevname(rdev->bdev,b));
1015                 return -EINVAL;
1016         }
1017         rdev->preferred_minor = 0xffff;
1018         rdev->data_offset = le64_to_cpu(sb->data_offset);
1019         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1020
1021         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1022         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1023         if (rdev->sb_size & bmask)
1024                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1025
1026         if (refdev == 0)
1027                 return 1;
1028         else {
1029                 __u64 ev1, ev2;
1030                 struct mdp_superblock_1 *refsb =
1031                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1032
1033                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1034                     sb->level != refsb->level ||
1035                     sb->layout != refsb->layout ||
1036                     sb->chunksize != refsb->chunksize) {
1037                         printk(KERN_WARNING "md: %s has strangely different"
1038                                 " superblock to %s\n",
1039                                 bdevname(rdev->bdev,b),
1040                                 bdevname(refdev->bdev,b2));
1041                         return -EINVAL;
1042                 }
1043                 ev1 = le64_to_cpu(sb->events);
1044                 ev2 = le64_to_cpu(refsb->events);
1045
1046                 if (ev1 > ev2)
1047                         return 1;
1048         }
1049         if (minor_version)
1050                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1051         else
1052                 rdev->size = rdev->sb_offset;
1053         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1054                 return -EINVAL;
1055         rdev->size = le64_to_cpu(sb->data_size)/2;
1056         if (le32_to_cpu(sb->chunksize))
1057                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1058
1059         if (le32_to_cpu(sb->size) > rdev->size*2)
1060                 return -EINVAL;
1061         return 0;
1062 }
1063
1064 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1065 {
1066         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1067
1068         rdev->raid_disk = -1;
1069         rdev->flags = 0;
1070         if (mddev->raid_disks == 0) {
1071                 mddev->major_version = 1;
1072                 mddev->patch_version = 0;
1073                 mddev->persistent = 1;
1074                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1075                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1076                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1077                 mddev->level = le32_to_cpu(sb->level);
1078                 mddev->clevel[0] = 0;
1079                 mddev->layout = le32_to_cpu(sb->layout);
1080                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1081                 mddev->size = le64_to_cpu(sb->size)/2;
1082                 mddev->events = le64_to_cpu(sb->events);
1083                 mddev->bitmap_offset = 0;
1084                 mddev->default_bitmap_offset = 1024;
1085
1086                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1087                 memcpy(mddev->uuid, sb->set_uuid, 16);
1088
1089                 mddev->max_disks =  (4096-256)/2;
1090
1091                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1092                     mddev->bitmap_file == NULL ) {
1093                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1094                             && mddev->level != 10) {
1095                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1096                                 return -EINVAL;
1097                         }
1098                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1099                 }
1100         } else if (mddev->pers == NULL) {
1101                 /* Insist of good event counter while assembling */
1102                 __u64 ev1 = le64_to_cpu(sb->events);
1103                 ++ev1;
1104                 if (ev1 < mddev->events)
1105                         return -EINVAL;
1106         } else if (mddev->bitmap) {
1107                 /* If adding to array with a bitmap, then we can accept an
1108                  * older device, but not too old.
1109                  */
1110                 __u64 ev1 = le64_to_cpu(sb->events);
1111                 if (ev1 < mddev->bitmap->events_cleared)
1112                         return 0;
1113         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1114                 return 0;
1115
1116         if (mddev->level != LEVEL_MULTIPATH) {
1117                 int role;
1118                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1119                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1120                 switch(role) {
1121                 case 0xffff: /* spare */
1122                         break;
1123                 case 0xfffe: /* faulty */
1124                         set_bit(Faulty, &rdev->flags);
1125                         break;
1126                 default:
1127                         set_bit(In_sync, &rdev->flags);
1128                         rdev->raid_disk = role;
1129                         break;
1130                 }
1131                 if (sb->devflags & WriteMostly1)
1132                         set_bit(WriteMostly, &rdev->flags);
1133         } else /* MULTIPATH are always insync */
1134                 set_bit(In_sync, &rdev->flags);
1135
1136         return 0;
1137 }
1138
1139 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1140 {
1141         struct mdp_superblock_1 *sb;
1142         struct list_head *tmp;
1143         mdk_rdev_t *rdev2;
1144         int max_dev, i;
1145         /* make rdev->sb match mddev and rdev data. */
1146
1147         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1148
1149         sb->feature_map = 0;
1150         sb->pad0 = 0;
1151         memset(sb->pad1, 0, sizeof(sb->pad1));
1152         memset(sb->pad2, 0, sizeof(sb->pad2));
1153         memset(sb->pad3, 0, sizeof(sb->pad3));
1154
1155         sb->utime = cpu_to_le64((__u64)mddev->utime);
1156         sb->events = cpu_to_le64(mddev->events);
1157         if (mddev->in_sync)
1158                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1159         else
1160                 sb->resync_offset = cpu_to_le64(0);
1161
1162         sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1163
1164         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1165                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1166                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1167         }
1168
1169         max_dev = 0;
1170         ITERATE_RDEV(mddev,rdev2,tmp)
1171                 if (rdev2->desc_nr+1 > max_dev)
1172                         max_dev = rdev2->desc_nr+1;
1173
1174         sb->max_dev = cpu_to_le32(max_dev);
1175         for (i=0; i<max_dev;i++)
1176                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1177
1178         ITERATE_RDEV(mddev,rdev2,tmp) {
1179                 i = rdev2->desc_nr;
1180                 if (test_bit(Faulty, &rdev2->flags))
1181                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1182                 else if (test_bit(In_sync, &rdev2->flags))
1183                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1184                 else
1185                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1186         }
1187
1188         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1189         sb->sb_csum = calc_sb_1_csum(sb);
1190 }
1191
1192
1193 static struct super_type super_types[] = {
1194         [0] = {
1195                 .name   = "0.90.0",
1196                 .owner  = THIS_MODULE,
1197                 .load_super     = super_90_load,
1198                 .validate_super = super_90_validate,
1199                 .sync_super     = super_90_sync,
1200         },
1201         [1] = {
1202                 .name   = "md-1",
1203                 .owner  = THIS_MODULE,
1204                 .load_super     = super_1_load,
1205                 .validate_super = super_1_validate,
1206                 .sync_super     = super_1_sync,
1207         },
1208 };
1209
1210 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1211 {
1212         struct list_head *tmp;
1213         mdk_rdev_t *rdev;
1214
1215         ITERATE_RDEV(mddev,rdev,tmp)
1216                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1217                         return rdev;
1218
1219         return NULL;
1220 }
1221
1222 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1223 {
1224         struct list_head *tmp;
1225         mdk_rdev_t *rdev;
1226
1227         ITERATE_RDEV(mddev1,rdev,tmp)
1228                 if (match_dev_unit(mddev2, rdev))
1229                         return 1;
1230
1231         return 0;
1232 }
1233
1234 static LIST_HEAD(pending_raid_disks);
1235
1236 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1237 {
1238         mdk_rdev_t *same_pdev;
1239         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1240         struct kobject *ko;
1241
1242         if (rdev->mddev) {
1243                 MD_BUG();
1244                 return -EINVAL;
1245         }
1246         /* make sure rdev->size exceeds mddev->size */
1247         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1248                 if (mddev->pers)
1249                         /* Cannot change size, so fail */
1250                         return -ENOSPC;
1251                 else
1252                         mddev->size = rdev->size;
1253         }
1254         same_pdev = match_dev_unit(mddev, rdev);
1255         if (same_pdev)
1256                 printk(KERN_WARNING
1257                         "%s: WARNING: %s appears to be on the same physical"
1258                         " disk as %s. True\n     protection against single-disk"
1259                         " failure might be compromised.\n",
1260                         mdname(mddev), bdevname(rdev->bdev,b),
1261                         bdevname(same_pdev->bdev,b2));
1262
1263         /* Verify rdev->desc_nr is unique.
1264          * If it is -1, assign a free number, else
1265          * check number is not in use
1266          */
1267         if (rdev->desc_nr < 0) {
1268                 int choice = 0;
1269                 if (mddev->pers) choice = mddev->raid_disks;
1270                 while (find_rdev_nr(mddev, choice))
1271                         choice++;
1272                 rdev->desc_nr = choice;
1273         } else {
1274                 if (find_rdev_nr(mddev, rdev->desc_nr))
1275                         return -EBUSY;
1276         }
1277         bdevname(rdev->bdev,b);
1278         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1279                 return -ENOMEM;
1280
1281         list_add(&rdev->same_set, &mddev->disks);
1282         rdev->mddev = mddev;
1283         printk(KERN_INFO "md: bind<%s>\n", b);
1284
1285         rdev->kobj.parent = &mddev->kobj;
1286         kobject_add(&rdev->kobj);
1287
1288         if (rdev->bdev->bd_part)
1289                 ko = &rdev->bdev->bd_part->kobj;
1290         else
1291                 ko = &rdev->bdev->bd_disk->kobj;
1292         sysfs_create_link(&rdev->kobj, ko, "block");
1293         return 0;
1294 }
1295
1296 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1297 {
1298         char b[BDEVNAME_SIZE];
1299         if (!rdev->mddev) {
1300                 MD_BUG();
1301                 return;
1302         }
1303         list_del_init(&rdev->same_set);
1304         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1305         rdev->mddev = NULL;
1306         sysfs_remove_link(&rdev->kobj, "block");
1307         kobject_del(&rdev->kobj);
1308 }
1309
1310 /*
1311  * prevent the device from being mounted, repartitioned or
1312  * otherwise reused by a RAID array (or any other kernel
1313  * subsystem), by bd_claiming the device.
1314  */
1315 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1316 {
1317         int err = 0;
1318         struct block_device *bdev;
1319         char b[BDEVNAME_SIZE];
1320
1321         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1322         if (IS_ERR(bdev)) {
1323                 printk(KERN_ERR "md: could not open %s.\n",
1324                         __bdevname(dev, b));
1325                 return PTR_ERR(bdev);
1326         }
1327         err = bd_claim(bdev, rdev);
1328         if (err) {
1329                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1330                         bdevname(bdev, b));
1331                 blkdev_put(bdev);
1332                 return err;
1333         }
1334         rdev->bdev = bdev;
1335         return err;
1336 }
1337
1338 static void unlock_rdev(mdk_rdev_t *rdev)
1339 {
1340         struct block_device *bdev = rdev->bdev;
1341         rdev->bdev = NULL;
1342         if (!bdev)
1343                 MD_BUG();
1344         bd_release(bdev);
1345         blkdev_put(bdev);
1346 }
1347
1348 void md_autodetect_dev(dev_t dev);
1349
1350 static void export_rdev(mdk_rdev_t * rdev)
1351 {
1352         char b[BDEVNAME_SIZE];
1353         printk(KERN_INFO "md: export_rdev(%s)\n",
1354                 bdevname(rdev->bdev,b));
1355         if (rdev->mddev)
1356                 MD_BUG();
1357         free_disk_sb(rdev);
1358         list_del_init(&rdev->same_set);
1359 #ifndef MODULE
1360         md_autodetect_dev(rdev->bdev->bd_dev);
1361 #endif
1362         unlock_rdev(rdev);
1363         kobject_put(&rdev->kobj);
1364 }
1365
1366 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1367 {
1368         unbind_rdev_from_array(rdev);
1369         export_rdev(rdev);
1370 }
1371
1372 static void export_array(mddev_t *mddev)
1373 {
1374         struct list_head *tmp;
1375         mdk_rdev_t *rdev;
1376
1377         ITERATE_RDEV(mddev,rdev,tmp) {
1378                 if (!rdev->mddev) {
1379                         MD_BUG();
1380                         continue;
1381                 }
1382                 kick_rdev_from_array(rdev);
1383         }
1384         if (!list_empty(&mddev->disks))
1385                 MD_BUG();
1386         mddev->raid_disks = 0;
1387         mddev->major_version = 0;
1388 }
1389
1390 static void print_desc(mdp_disk_t *desc)
1391 {
1392         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1393                 desc->major,desc->minor,desc->raid_disk,desc->state);
1394 }
1395
1396 static void print_sb(mdp_super_t *sb)
1397 {
1398         int i;
1399
1400         printk(KERN_INFO
1401                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1402                 sb->major_version, sb->minor_version, sb->patch_version,
1403                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1404                 sb->ctime);
1405         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1406                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1407                 sb->md_minor, sb->layout, sb->chunk_size);
1408         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1409                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1410                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1411                 sb->failed_disks, sb->spare_disks,
1412                 sb->sb_csum, (unsigned long)sb->events_lo);
1413
1414         printk(KERN_INFO);
1415         for (i = 0; i < MD_SB_DISKS; i++) {
1416                 mdp_disk_t *desc;
1417
1418                 desc = sb->disks + i;
1419                 if (desc->number || desc->major || desc->minor ||
1420                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1421                         printk("     D %2d: ", i);
1422                         print_desc(desc);
1423                 }
1424         }
1425         printk(KERN_INFO "md:     THIS: ");
1426         print_desc(&sb->this_disk);
1427
1428 }
1429
1430 static void print_rdev(mdk_rdev_t *rdev)
1431 {
1432         char b[BDEVNAME_SIZE];
1433         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1434                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1435                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1436                 rdev->desc_nr);
1437         if (rdev->sb_loaded) {
1438                 printk(KERN_INFO "md: rdev superblock:\n");
1439                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1440         } else
1441                 printk(KERN_INFO "md: no rdev superblock!\n");
1442 }
1443
1444 void md_print_devices(void)
1445 {
1446         struct list_head *tmp, *tmp2;
1447         mdk_rdev_t *rdev;
1448         mddev_t *mddev;
1449         char b[BDEVNAME_SIZE];
1450
1451         printk("\n");
1452         printk("md:     **********************************\n");
1453         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1454         printk("md:     **********************************\n");
1455         ITERATE_MDDEV(mddev,tmp) {
1456
1457                 if (mddev->bitmap)
1458                         bitmap_print_sb(mddev->bitmap);
1459                 else
1460                         printk("%s: ", mdname(mddev));
1461                 ITERATE_RDEV(mddev,rdev,tmp2)
1462                         printk("<%s>", bdevname(rdev->bdev,b));
1463                 printk("\n");
1464
1465                 ITERATE_RDEV(mddev,rdev,tmp2)
1466                         print_rdev(rdev);
1467         }
1468         printk("md:     **********************************\n");
1469         printk("\n");
1470 }
1471
1472
1473 static void sync_sbs(mddev_t * mddev)
1474 {
1475         mdk_rdev_t *rdev;
1476         struct list_head *tmp;
1477
1478         ITERATE_RDEV(mddev,rdev,tmp) {
1479                 super_types[mddev->major_version].
1480                         sync_super(mddev, rdev);
1481                 rdev->sb_loaded = 1;
1482         }
1483 }
1484
1485 static void md_update_sb(mddev_t * mddev)
1486 {
1487         int err;
1488         struct list_head *tmp;
1489         mdk_rdev_t *rdev;
1490         int sync_req;
1491
1492 repeat:
1493         spin_lock_irq(&mddev->write_lock);
1494         sync_req = mddev->in_sync;
1495         mddev->utime = get_seconds();
1496         mddev->events ++;
1497
1498         if (!mddev->events) {
1499                 /*
1500                  * oops, this 64-bit counter should never wrap.
1501                  * Either we are in around ~1 trillion A.C., assuming
1502                  * 1 reboot per second, or we have a bug:
1503                  */
1504                 MD_BUG();
1505                 mddev->events --;
1506         }
1507         mddev->sb_dirty = 2;
1508         sync_sbs(mddev);
1509
1510         /*
1511          * do not write anything to disk if using
1512          * nonpersistent superblocks
1513          */
1514         if (!mddev->persistent) {
1515                 mddev->sb_dirty = 0;
1516                 spin_unlock_irq(&mddev->write_lock);
1517                 wake_up(&mddev->sb_wait);
1518                 return;
1519         }
1520         spin_unlock_irq(&mddev->write_lock);
1521
1522         dprintk(KERN_INFO
1523                 "md: updating %s RAID superblock on device (in sync %d)\n",
1524                 mdname(mddev),mddev->in_sync);
1525
1526         err = bitmap_update_sb(mddev->bitmap);
1527         ITERATE_RDEV(mddev,rdev,tmp) {
1528                 char b[BDEVNAME_SIZE];
1529                 dprintk(KERN_INFO "md: ");
1530                 if (test_bit(Faulty, &rdev->flags))
1531                         dprintk("(skipping faulty ");
1532
1533                 dprintk("%s ", bdevname(rdev->bdev,b));
1534                 if (!test_bit(Faulty, &rdev->flags)) {
1535                         md_super_write(mddev,rdev,
1536                                        rdev->sb_offset<<1, rdev->sb_size,
1537                                        rdev->sb_page);
1538                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1539                                 bdevname(rdev->bdev,b),
1540                                 (unsigned long long)rdev->sb_offset);
1541
1542                 } else
1543                         dprintk(")\n");
1544                 if (mddev->level == LEVEL_MULTIPATH)
1545                         /* only need to write one superblock... */
1546                         break;
1547         }
1548         md_super_wait(mddev);
1549         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1550
1551         spin_lock_irq(&mddev->write_lock);
1552         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1553                 /* have to write it out again */
1554                 spin_unlock_irq(&mddev->write_lock);
1555                 goto repeat;
1556         }
1557         mddev->sb_dirty = 0;
1558         spin_unlock_irq(&mddev->write_lock);
1559         wake_up(&mddev->sb_wait);
1560
1561 }
1562
1563 /* words written to sysfs files may, or my not, be \n terminated.
1564  * We want to accept with case. For this we use cmd_match.
1565  */
1566 static int cmd_match(const char *cmd, const char *str)
1567 {
1568         /* See if cmd, written into a sysfs file, matches
1569          * str.  They must either be the same, or cmd can
1570          * have a trailing newline
1571          */
1572         while (*cmd && *str && *cmd == *str) {
1573                 cmd++;
1574                 str++;
1575         }
1576         if (*cmd == '\n')
1577                 cmd++;
1578         if (*str || *cmd)
1579                 return 0;
1580         return 1;
1581 }
1582
1583 struct rdev_sysfs_entry {
1584         struct attribute attr;
1585         ssize_t (*show)(mdk_rdev_t *, char *);
1586         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1587 };
1588
1589 static ssize_t
1590 state_show(mdk_rdev_t *rdev, char *page)
1591 {
1592         char *sep = "";
1593         int len=0;
1594
1595         if (test_bit(Faulty, &rdev->flags)) {
1596                 len+= sprintf(page+len, "%sfaulty",sep);
1597                 sep = ",";
1598         }
1599         if (test_bit(In_sync, &rdev->flags)) {
1600                 len += sprintf(page+len, "%sin_sync",sep);
1601                 sep = ",";
1602         }
1603         if (!test_bit(Faulty, &rdev->flags) &&
1604             !test_bit(In_sync, &rdev->flags)) {
1605                 len += sprintf(page+len, "%sspare", sep);
1606                 sep = ",";
1607         }
1608         return len+sprintf(page+len, "\n");
1609 }
1610
1611 static struct rdev_sysfs_entry
1612 rdev_state = __ATTR_RO(state);
1613
1614 static ssize_t
1615 super_show(mdk_rdev_t *rdev, char *page)
1616 {
1617         if (rdev->sb_loaded && rdev->sb_size) {
1618                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1619                 return rdev->sb_size;
1620         } else
1621                 return 0;
1622 }
1623 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1624
1625 static ssize_t
1626 errors_show(mdk_rdev_t *rdev, char *page)
1627 {
1628         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1629 }
1630
1631 static ssize_t
1632 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1633 {
1634         char *e;
1635         unsigned long n = simple_strtoul(buf, &e, 10);
1636         if (*buf && (*e == 0 || *e == '\n')) {
1637                 atomic_set(&rdev->corrected_errors, n);
1638                 return len;
1639         }
1640         return -EINVAL;
1641 }
1642 static struct rdev_sysfs_entry rdev_errors =
1643 __ATTR(errors, 0644, errors_show, errors_store);
1644
1645 static ssize_t
1646 slot_show(mdk_rdev_t *rdev, char *page)
1647 {
1648         if (rdev->raid_disk < 0)
1649                 return sprintf(page, "none\n");
1650         else
1651                 return sprintf(page, "%d\n", rdev->raid_disk);
1652 }
1653
1654 static ssize_t
1655 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1656 {
1657         char *e;
1658         int slot = simple_strtoul(buf, &e, 10);
1659         if (strncmp(buf, "none", 4)==0)
1660                 slot = -1;
1661         else if (e==buf || (*e && *e!= '\n'))
1662                 return -EINVAL;
1663         if (rdev->mddev->pers)
1664                 /* Cannot set slot in active array (yet) */
1665                 return -EBUSY;
1666         if (slot >= rdev->mddev->raid_disks)
1667                 return -ENOSPC;
1668         rdev->raid_disk = slot;
1669         /* assume it is working */
1670         rdev->flags = 0;
1671         set_bit(In_sync, &rdev->flags);
1672         return len;
1673 }
1674
1675
1676 static struct rdev_sysfs_entry rdev_slot =
1677 __ATTR(slot, 0644, slot_show, slot_store);
1678
1679 static ssize_t
1680 offset_show(mdk_rdev_t *rdev, char *page)
1681 {
1682         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1683 }
1684
1685 static ssize_t
1686 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1687 {
1688         char *e;
1689         unsigned long long offset = simple_strtoull(buf, &e, 10);
1690         if (e==buf || (*e && *e != '\n'))
1691                 return -EINVAL;
1692         if (rdev->mddev->pers)
1693                 return -EBUSY;
1694         rdev->data_offset = offset;
1695         return len;
1696 }
1697
1698 static struct rdev_sysfs_entry rdev_offset =
1699 __ATTR(offset, 0644, offset_show, offset_store);
1700
1701 static ssize_t
1702 rdev_size_show(mdk_rdev_t *rdev, char *page)
1703 {
1704         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1705 }
1706
1707 static ssize_t
1708 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1709 {
1710         char *e;
1711         unsigned long long size = simple_strtoull(buf, &e, 10);
1712         if (e==buf || (*e && *e != '\n'))
1713                 return -EINVAL;
1714         if (rdev->mddev->pers)
1715                 return -EBUSY;
1716         rdev->size = size;
1717         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1718                 rdev->mddev->size = size;
1719         return len;
1720 }
1721
1722 static struct rdev_sysfs_entry rdev_size =
1723 __ATTR(size, 0644, rdev_size_show, rdev_size_store);
1724
1725 static struct attribute *rdev_default_attrs[] = {
1726         &rdev_state.attr,
1727         &rdev_super.attr,
1728         &rdev_errors.attr,
1729         &rdev_slot.attr,
1730         &rdev_offset.attr,
1731         &rdev_size.attr,
1732         NULL,
1733 };
1734 static ssize_t
1735 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1736 {
1737         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1738         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1739
1740         if (!entry->show)
1741                 return -EIO;
1742         return entry->show(rdev, page);
1743 }
1744
1745 static ssize_t
1746 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1747               const char *page, size_t length)
1748 {
1749         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1750         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1751
1752         if (!entry->store)
1753                 return -EIO;
1754         return entry->store(rdev, page, length);
1755 }
1756
1757 static void rdev_free(struct kobject *ko)
1758 {
1759         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1760         kfree(rdev);
1761 }
1762 static struct sysfs_ops rdev_sysfs_ops = {
1763         .show           = rdev_attr_show,
1764         .store          = rdev_attr_store,
1765 };
1766 static struct kobj_type rdev_ktype = {
1767         .release        = rdev_free,
1768         .sysfs_ops      = &rdev_sysfs_ops,
1769         .default_attrs  = rdev_default_attrs,
1770 };
1771
1772 /*
1773  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1774  *
1775  * mark the device faulty if:
1776  *
1777  *   - the device is nonexistent (zero size)
1778  *   - the device has no valid superblock
1779  *
1780  * a faulty rdev _never_ has rdev->sb set.
1781  */
1782 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1783 {
1784         char b[BDEVNAME_SIZE];
1785         int err;
1786         mdk_rdev_t *rdev;
1787         sector_t size;
1788
1789         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1790         if (!rdev) {
1791                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1792                 return ERR_PTR(-ENOMEM);
1793         }
1794
1795         if ((err = alloc_disk_sb(rdev)))
1796                 goto abort_free;
1797
1798         err = lock_rdev(rdev, newdev);
1799         if (err)
1800                 goto abort_free;
1801
1802         rdev->kobj.parent = NULL;
1803         rdev->kobj.ktype = &rdev_ktype;
1804         kobject_init(&rdev->kobj);
1805
1806         rdev->desc_nr = -1;
1807         rdev->flags = 0;
1808         rdev->data_offset = 0;
1809         atomic_set(&rdev->nr_pending, 0);
1810         atomic_set(&rdev->read_errors, 0);
1811         atomic_set(&rdev->corrected_errors, 0);
1812
1813         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1814         if (!size) {
1815                 printk(KERN_WARNING
1816                         "md: %s has zero or unknown size, marking faulty!\n",
1817                         bdevname(rdev->bdev,b));
1818                 err = -EINVAL;
1819                 goto abort_free;
1820         }
1821
1822         if (super_format >= 0) {
1823                 err = super_types[super_format].
1824                         load_super(rdev, NULL, super_minor);
1825                 if (err == -EINVAL) {
1826                         printk(KERN_WARNING
1827                                 "md: %s has invalid sb, not importing!\n",
1828                                 bdevname(rdev->bdev,b));
1829                         goto abort_free;
1830                 }
1831                 if (err < 0) {
1832                         printk(KERN_WARNING
1833                                 "md: could not read %s's sb, not importing!\n",
1834                                 bdevname(rdev->bdev,b));
1835                         goto abort_free;
1836                 }
1837         }
1838         INIT_LIST_HEAD(&rdev->same_set);
1839
1840         return rdev;
1841
1842 abort_free:
1843         if (rdev->sb_page) {
1844                 if (rdev->bdev)
1845                         unlock_rdev(rdev);
1846                 free_disk_sb(rdev);
1847         }
1848         kfree(rdev);
1849         return ERR_PTR(err);
1850 }
1851
1852 /*
1853  * Check a full RAID array for plausibility
1854  */
1855
1856
1857 static void analyze_sbs(mddev_t * mddev)
1858 {
1859         int i;
1860         struct list_head *tmp;
1861         mdk_rdev_t *rdev, *freshest;
1862         char b[BDEVNAME_SIZE];
1863
1864         freshest = NULL;
1865         ITERATE_RDEV(mddev,rdev,tmp)
1866                 switch (super_types[mddev->major_version].
1867                         load_super(rdev, freshest, mddev->minor_version)) {
1868                 case 1:
1869                         freshest = rdev;
1870                         break;
1871                 case 0:
1872                         break;
1873                 default:
1874                         printk( KERN_ERR \
1875                                 "md: fatal superblock inconsistency in %s"
1876                                 " -- removing from array\n",
1877                                 bdevname(rdev->bdev,b));
1878                         kick_rdev_from_array(rdev);
1879                 }
1880
1881
1882         super_types[mddev->major_version].
1883                 validate_super(mddev, freshest);
1884
1885         i = 0;
1886         ITERATE_RDEV(mddev,rdev,tmp) {
1887                 if (rdev != freshest)
1888                         if (super_types[mddev->major_version].
1889                             validate_super(mddev, rdev)) {
1890                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1891                                         " from array!\n",
1892                                         bdevname(rdev->bdev,b));
1893                                 kick_rdev_from_array(rdev);
1894                                 continue;
1895                         }
1896                 if (mddev->level == LEVEL_MULTIPATH) {
1897                         rdev->desc_nr = i++;
1898                         rdev->raid_disk = rdev->desc_nr;
1899                         set_bit(In_sync, &rdev->flags);
1900                 }
1901         }
1902
1903
1904
1905         if (mddev->recovery_cp != MaxSector &&
1906             mddev->level >= 1)
1907                 printk(KERN_ERR "md: %s: raid array is not clean"
1908                        " -- starting background reconstruction\n",
1909                        mdname(mddev));
1910
1911 }
1912
1913 static ssize_t
1914 level_show(mddev_t *mddev, char *page)
1915 {
1916         struct mdk_personality *p = mddev->pers;
1917         if (p)
1918                 return sprintf(page, "%s\n", p->name);
1919         else if (mddev->clevel[0])
1920                 return sprintf(page, "%s\n", mddev->clevel);
1921         else if (mddev->level != LEVEL_NONE)
1922                 return sprintf(page, "%d\n", mddev->level);
1923         else
1924                 return 0;
1925 }
1926
1927 static ssize_t
1928 level_store(mddev_t *mddev, const char *buf, size_t len)
1929 {
1930         int rv = len;
1931         if (mddev->pers)
1932                 return -EBUSY;
1933         if (len == 0)
1934                 return 0;
1935         if (len >= sizeof(mddev->clevel))
1936                 return -ENOSPC;
1937         strncpy(mddev->clevel, buf, len);
1938         if (mddev->clevel[len-1] == '\n')
1939                 len--;
1940         mddev->clevel[len] = 0;
1941         mddev->level = LEVEL_NONE;
1942         return rv;
1943 }
1944
1945 static struct md_sysfs_entry md_level =
1946 __ATTR(level, 0644, level_show, level_store);
1947
1948 static ssize_t
1949 raid_disks_show(mddev_t *mddev, char *page)
1950 {
1951         if (mddev->raid_disks == 0)
1952                 return 0;
1953         return sprintf(page, "%d\n", mddev->raid_disks);
1954 }
1955
1956 static int update_raid_disks(mddev_t *mddev, int raid_disks);
1957
1958 static ssize_t
1959 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
1960 {
1961         /* can only set raid_disks if array is not yet active */
1962         char *e;
1963         int rv = 0;
1964         unsigned long n = simple_strtoul(buf, &e, 10);
1965
1966         if (!*buf || (*e && *e != '\n'))
1967                 return -EINVAL;
1968
1969         if (mddev->pers)
1970                 rv = update_raid_disks(mddev, n);
1971         else
1972                 mddev->raid_disks = n;
1973         return rv ? rv : len;
1974 }
1975 static struct md_sysfs_entry md_raid_disks =
1976 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
1977
1978 static ssize_t
1979 chunk_size_show(mddev_t *mddev, char *page)
1980 {
1981         return sprintf(page, "%d\n", mddev->chunk_size);
1982 }
1983
1984 static ssize_t
1985 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1986 {
1987         /* can only set chunk_size if array is not yet active */
1988         char *e;
1989         unsigned long n = simple_strtoul(buf, &e, 10);
1990
1991         if (mddev->pers)
1992                 return -EBUSY;
1993         if (!*buf || (*e && *e != '\n'))
1994                 return -EINVAL;
1995
1996         mddev->chunk_size = n;
1997         return len;
1998 }
1999 static struct md_sysfs_entry md_chunk_size =
2000 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2001
2002 static ssize_t
2003 null_show(mddev_t *mddev, char *page)
2004 {
2005         return -EINVAL;
2006 }
2007
2008 static ssize_t
2009 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2010 {
2011         /* buf must be %d:%d\n? giving major and minor numbers */
2012         /* The new device is added to the array.
2013          * If the array has a persistent superblock, we read the
2014          * superblock to initialise info and check validity.
2015          * Otherwise, only checking done is that in bind_rdev_to_array,
2016          * which mainly checks size.
2017          */
2018         char *e;
2019         int major = simple_strtoul(buf, &e, 10);
2020         int minor;
2021         dev_t dev;
2022         mdk_rdev_t *rdev;
2023         int err;
2024
2025         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2026                 return -EINVAL;
2027         minor = simple_strtoul(e+1, &e, 10);
2028         if (*e && *e != '\n')
2029                 return -EINVAL;
2030         dev = MKDEV(major, minor);
2031         if (major != MAJOR(dev) ||
2032             minor != MINOR(dev))
2033                 return -EOVERFLOW;
2034
2035
2036         if (mddev->persistent) {
2037                 rdev = md_import_device(dev, mddev->major_version,
2038                                         mddev->minor_version);
2039                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2040                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2041                                                        mdk_rdev_t, same_set);
2042                         err = super_types[mddev->major_version]
2043                                 .load_super(rdev, rdev0, mddev->minor_version);
2044                         if (err < 0)
2045                                 goto out;
2046                 }
2047         } else
2048                 rdev = md_import_device(dev, -1, -1);
2049
2050         if (IS_ERR(rdev))
2051                 return PTR_ERR(rdev);
2052         err = bind_rdev_to_array(rdev, mddev);
2053  out:
2054         if (err)
2055                 export_rdev(rdev);
2056         return err ? err : len;
2057 }
2058
2059 static struct md_sysfs_entry md_new_device =
2060 __ATTR(new_dev, 0200, null_show, new_dev_store);
2061
2062 static ssize_t
2063 size_show(mddev_t *mddev, char *page)
2064 {
2065         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2066 }
2067
2068 static int update_size(mddev_t *mddev, unsigned long size);
2069
2070 static ssize_t
2071 size_store(mddev_t *mddev, const char *buf, size_t len)
2072 {
2073         /* If array is inactive, we can reduce the component size, but
2074          * not increase it (except from 0).
2075          * If array is active, we can try an on-line resize
2076          */
2077         char *e;
2078         int err = 0;
2079         unsigned long long size = simple_strtoull(buf, &e, 10);
2080         if (!*buf || *buf == '\n' ||
2081             (*e && *e != '\n'))
2082                 return -EINVAL;
2083
2084         if (mddev->pers) {
2085                 err = update_size(mddev, size);
2086                 md_update_sb(mddev);
2087         } else {
2088                 if (mddev->size == 0 ||
2089                     mddev->size > size)
2090                         mddev->size = size;
2091                 else
2092                         err = -ENOSPC;
2093         }
2094         return err ? err : len;
2095 }
2096
2097 static struct md_sysfs_entry md_size =
2098 __ATTR(component_size, 0644, size_show, size_store);
2099
2100
2101 /* Metdata version.
2102  * This is either 'none' for arrays with externally managed metadata,
2103  * or N.M for internally known formats
2104  */
2105 static ssize_t
2106 metadata_show(mddev_t *mddev, char *page)
2107 {
2108         if (mddev->persistent)
2109                 return sprintf(page, "%d.%d\n",
2110                                mddev->major_version, mddev->minor_version);
2111         else
2112                 return sprintf(page, "none\n");
2113 }
2114
2115 static ssize_t
2116 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2117 {
2118         int major, minor;
2119         char *e;
2120         if (!list_empty(&mddev->disks))
2121                 return -EBUSY;
2122
2123         if (cmd_match(buf, "none")) {
2124                 mddev->persistent = 0;
2125                 mddev->major_version = 0;
2126                 mddev->minor_version = 90;
2127                 return len;
2128         }
2129         major = simple_strtoul(buf, &e, 10);
2130         if (e==buf || *e != '.')
2131                 return -EINVAL;
2132         buf = e+1;
2133         minor = simple_strtoul(buf, &e, 10);
2134         if (e==buf || *e != '\n')
2135                 return -EINVAL;
2136         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2137             super_types[major].name == NULL)
2138                 return -ENOENT;
2139         mddev->major_version = major;
2140         mddev->minor_version = minor;
2141         mddev->persistent = 1;
2142         return len;
2143 }
2144
2145 static struct md_sysfs_entry md_metadata =
2146 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
2147
2148 static ssize_t
2149 action_show(mddev_t *mddev, char *page)
2150 {
2151         char *type = "idle";
2152         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2153             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2154                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2155                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2156                                 type = "resync";
2157                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2158                                 type = "check";
2159                         else
2160                                 type = "repair";
2161                 } else
2162                         type = "recover";
2163         }
2164         return sprintf(page, "%s\n", type);
2165 }
2166
2167 static ssize_t
2168 action_store(mddev_t *mddev, const char *page, size_t len)
2169 {
2170         if (!mddev->pers || !mddev->pers->sync_request)
2171                 return -EINVAL;
2172
2173         if (cmd_match(page, "idle")) {
2174                 if (mddev->sync_thread) {
2175                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2176                         md_unregister_thread(mddev->sync_thread);
2177                         mddev->sync_thread = NULL;
2178                         mddev->recovery = 0;
2179                 }
2180         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2181                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2182                 return -EBUSY;
2183         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2184                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2185         else {
2186                 if (cmd_match(page, "check"))
2187                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2188                 else if (cmd_match(page, "repair"))
2189                         return -EINVAL;
2190                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2191                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2192         }
2193         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2194         md_wakeup_thread(mddev->thread);
2195         return len;
2196 }
2197
2198 static ssize_t
2199 mismatch_cnt_show(mddev_t *mddev, char *page)
2200 {
2201         return sprintf(page, "%llu\n",
2202                        (unsigned long long) mddev->resync_mismatches);
2203 }
2204
2205 static struct md_sysfs_entry
2206 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2207
2208
2209 static struct md_sysfs_entry
2210 md_mismatches = __ATTR_RO(mismatch_cnt);
2211
2212 static ssize_t
2213 sync_min_show(mddev_t *mddev, char *page)
2214 {
2215         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2216                        mddev->sync_speed_min ? "local": "system");
2217 }
2218
2219 static ssize_t
2220 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2221 {
2222         int min;
2223         char *e;
2224         if (strncmp(buf, "system", 6)==0) {
2225                 mddev->sync_speed_min = 0;
2226                 return len;
2227         }
2228         min = simple_strtoul(buf, &e, 10);
2229         if (buf == e || (*e && *e != '\n') || min <= 0)
2230                 return -EINVAL;
2231         mddev->sync_speed_min = min;
2232         return len;
2233 }
2234
2235 static struct md_sysfs_entry md_sync_min =
2236 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2237
2238 static ssize_t
2239 sync_max_show(mddev_t *mddev, char *page)
2240 {
2241         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2242                        mddev->sync_speed_max ? "local": "system");
2243 }
2244
2245 static ssize_t
2246 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2247 {
2248         int max;
2249         char *e;
2250         if (strncmp(buf, "system", 6)==0) {
2251                 mddev->sync_speed_max = 0;
2252                 return len;
2253         }
2254         max = simple_strtoul(buf, &e, 10);
2255         if (buf == e || (*e && *e != '\n') || max <= 0)
2256                 return -EINVAL;
2257         mddev->sync_speed_max = max;
2258         return len;
2259 }
2260
2261 static struct md_sysfs_entry md_sync_max =
2262 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2263
2264
2265 static ssize_t
2266 sync_speed_show(mddev_t *mddev, char *page)
2267 {
2268         unsigned long resync, dt, db;
2269         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2270         dt = ((jiffies - mddev->resync_mark) / HZ);
2271         if (!dt) dt++;
2272         db = resync - (mddev->resync_mark_cnt);
2273         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2274 }
2275
2276 static struct md_sysfs_entry
2277 md_sync_speed = __ATTR_RO(sync_speed);
2278
2279 static ssize_t
2280 sync_completed_show(mddev_t *mddev, char *page)
2281 {
2282         unsigned long max_blocks, resync;
2283
2284         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2285                 max_blocks = mddev->resync_max_sectors;
2286         else
2287                 max_blocks = mddev->size << 1;
2288
2289         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2290         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2291 }
2292
2293 static struct md_sysfs_entry
2294 md_sync_completed = __ATTR_RO(sync_completed);
2295
2296 static struct attribute *md_default_attrs[] = {
2297         &md_level.attr,
2298         &md_raid_disks.attr,
2299         &md_chunk_size.attr,
2300         &md_size.attr,
2301         &md_metadata.attr,
2302         &md_new_device.attr,
2303         NULL,
2304 };
2305
2306 static struct attribute *md_redundancy_attrs[] = {
2307         &md_scan_mode.attr,
2308         &md_mismatches.attr,
2309         &md_sync_min.attr,
2310         &md_sync_max.attr,
2311         &md_sync_speed.attr,
2312         &md_sync_completed.attr,
2313         NULL,
2314 };
2315 static struct attribute_group md_redundancy_group = {
2316         .name = NULL,
2317         .attrs = md_redundancy_attrs,
2318 };
2319
2320
2321 static ssize_t
2322 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2323 {
2324         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2325         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2326         ssize_t rv;
2327
2328         if (!entry->show)
2329                 return -EIO;
2330         mddev_lock(mddev);
2331         rv = entry->show(mddev, page);
2332         mddev_unlock(mddev);
2333         return rv;
2334 }
2335
2336 static ssize_t
2337 md_attr_store(struct kobject *kobj, struct attribute *attr,
2338               const char *page, size_t length)
2339 {
2340         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2341         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2342         ssize_t rv;
2343
2344         if (!entry->store)
2345                 return -EIO;
2346         mddev_lock(mddev);
2347         rv = entry->store(mddev, page, length);
2348         mddev_unlock(mddev);
2349         return rv;
2350 }
2351
2352 static void md_free(struct kobject *ko)
2353 {
2354         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2355         kfree(mddev);
2356 }
2357
2358 static struct sysfs_ops md_sysfs_ops = {
2359         .show   = md_attr_show,
2360         .store  = md_attr_store,
2361 };
2362 static struct kobj_type md_ktype = {
2363         .release        = md_free,
2364         .sysfs_ops      = &md_sysfs_ops,
2365         .default_attrs  = md_default_attrs,
2366 };
2367
2368 int mdp_major = 0;
2369
2370 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2371 {
2372         static DECLARE_MUTEX(disks_sem);
2373         mddev_t *mddev = mddev_find(dev);
2374         struct gendisk *disk;
2375         int partitioned = (MAJOR(dev) != MD_MAJOR);
2376         int shift = partitioned ? MdpMinorShift : 0;
2377         int unit = MINOR(dev) >> shift;
2378
2379         if (!mddev)
2380                 return NULL;
2381
2382         down(&disks_sem);
2383         if (mddev->gendisk) {
2384                 up(&disks_sem);
2385                 mddev_put(mddev);
2386                 return NULL;
2387         }
2388         disk = alloc_disk(1 << shift);
2389         if (!disk) {
2390                 up(&disks_sem);
2391                 mddev_put(mddev);
2392                 return NULL;
2393         }
2394         disk->major = MAJOR(dev);
2395         disk->first_minor = unit << shift;
2396         if (partitioned) {
2397                 sprintf(disk->disk_name, "md_d%d", unit);
2398                 sprintf(disk->devfs_name, "md/d%d", unit);
2399         } else {
2400                 sprintf(disk->disk_name, "md%d", unit);
2401                 sprintf(disk->devfs_name, "md/%d", unit);
2402         }
2403         disk->fops = &md_fops;
2404         disk->private_data = mddev;
2405         disk->queue = mddev->queue;
2406         add_disk(disk);
2407         mddev->gendisk = disk;
2408         up(&disks_sem);
2409         mddev->kobj.parent = &disk->kobj;
2410         mddev->kobj.k_name = NULL;
2411         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2412         mddev->kobj.ktype = &md_ktype;
2413         kobject_register(&mddev->kobj);
2414         return NULL;
2415 }
2416
2417 void md_wakeup_thread(mdk_thread_t *thread);
2418
2419 static void md_safemode_timeout(unsigned long data)
2420 {
2421         mddev_t *mddev = (mddev_t *) data;
2422
2423         mddev->safemode = 1;
2424         md_wakeup_thread(mddev->thread);
2425 }
2426
2427 static int start_dirty_degraded;
2428
2429 static int do_md_run(mddev_t * mddev)
2430 {
2431         int err;
2432         int chunk_size;
2433         struct list_head *tmp;
2434         mdk_rdev_t *rdev;
2435         struct gendisk *disk;
2436         struct mdk_personality *pers;
2437         char b[BDEVNAME_SIZE];
2438
2439         if (list_empty(&mddev->disks))
2440                 /* cannot run an array with no devices.. */
2441                 return -EINVAL;
2442
2443         if (mddev->pers)
2444                 return -EBUSY;
2445
2446         /*
2447          * Analyze all RAID superblock(s)
2448          */
2449         if (!mddev->raid_disks)
2450                 analyze_sbs(mddev);
2451
2452         chunk_size = mddev->chunk_size;
2453
2454         if (chunk_size) {
2455                 if (chunk_size > MAX_CHUNK_SIZE) {
2456                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2457                                 chunk_size, MAX_CHUNK_SIZE);
2458                         return -EINVAL;
2459                 }
2460                 /*
2461                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2462                  */
2463                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2464                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2465                         return -EINVAL;
2466                 }
2467                 if (chunk_size < PAGE_SIZE) {
2468                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2469                                 chunk_size, PAGE_SIZE);
2470                         return -EINVAL;
2471                 }
2472
2473                 /* devices must have minimum size of one chunk */
2474                 ITERATE_RDEV(mddev,rdev,tmp) {
2475                         if (test_bit(Faulty, &rdev->flags))
2476                                 continue;
2477                         if (rdev->size < chunk_size / 1024) {
2478                                 printk(KERN_WARNING
2479                                         "md: Dev %s smaller than chunk_size:"
2480                                         " %lluk < %dk\n",
2481                                         bdevname(rdev->bdev,b),
2482                                         (unsigned long long)rdev->size,
2483                                         chunk_size / 1024);
2484                                 return -EINVAL;
2485                         }
2486                 }
2487         }
2488
2489 #ifdef CONFIG_KMOD
2490         if (mddev->level != LEVEL_NONE)
2491                 request_module("md-level-%d", mddev->level);
2492         else if (mddev->clevel[0])
2493                 request_module("md-%s", mddev->clevel);
2494 #endif
2495
2496         /*
2497          * Drop all container device buffers, from now on
2498          * the only valid external interface is through the md
2499          * device.
2500          * Also find largest hardsector size
2501          */
2502         ITERATE_RDEV(mddev,rdev,tmp) {
2503                 if (test_bit(Faulty, &rdev->flags))
2504                         continue;
2505                 sync_blockdev(rdev->bdev);
2506                 invalidate_bdev(rdev->bdev, 0);
2507         }
2508
2509         md_probe(mddev->unit, NULL, NULL);
2510         disk = mddev->gendisk;
2511         if (!disk)
2512                 return -ENOMEM;
2513
2514         spin_lock(&pers_lock);
2515         pers = find_pers(mddev->level, mddev->clevel);
2516         if (!pers || !try_module_get(pers->owner)) {
2517                 spin_unlock(&pers_lock);
2518                 if (mddev->level != LEVEL_NONE)
2519                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2520                                mddev->level);
2521                 else
2522                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2523                                mddev->clevel);
2524                 return -EINVAL;
2525         }
2526         mddev->pers = pers;
2527         spin_unlock(&pers_lock);
2528         mddev->level = pers->level;
2529         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2530
2531         mddev->recovery = 0;
2532         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2533         mddev->barriers_work = 1;
2534         mddev->ok_start_degraded = start_dirty_degraded;
2535
2536         if (start_readonly)
2537                 mddev->ro = 2; /* read-only, but switch on first write */
2538
2539         err = mddev->pers->run(mddev);
2540         if (!err && mddev->pers->sync_request) {
2541                 err = bitmap_create(mddev);
2542                 if (err) {
2543                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2544                                mdname(mddev), err);
2545                         mddev->pers->stop(mddev);
2546                 }
2547         }
2548         if (err) {
2549                 printk(KERN_ERR "md: pers->run() failed ...\n");
2550                 module_put(mddev->pers->owner);
2551                 mddev->pers = NULL;
2552                 bitmap_destroy(mddev);
2553                 return err;
2554         }
2555         if (mddev->pers->sync_request)
2556                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2557         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2558                 mddev->ro = 0;
2559
2560         atomic_set(&mddev->writes_pending,0);
2561         mddev->safemode = 0;
2562         mddev->safemode_timer.function = md_safemode_timeout;
2563         mddev->safemode_timer.data = (unsigned long) mddev;
2564         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2565         mddev->in_sync = 1;
2566
2567         ITERATE_RDEV(mddev,rdev,tmp)
2568                 if (rdev->raid_disk >= 0) {
2569                         char nm[20];
2570                         sprintf(nm, "rd%d", rdev->raid_disk);
2571                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2572                 }
2573
2574         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2575         md_wakeup_thread(mddev->thread);
2576
2577         if (mddev->sb_dirty)
2578                 md_update_sb(mddev);
2579
2580         set_capacity(disk, mddev->array_size<<1);
2581
2582         /* If we call blk_queue_make_request here, it will
2583          * re-initialise max_sectors etc which may have been
2584          * refined inside -> run.  So just set the bits we need to set.
2585          * Most initialisation happended when we called
2586          * blk_queue_make_request(..., md_fail_request)
2587          * earlier.
2588          */
2589         mddev->queue->queuedata = mddev;
2590         mddev->queue->make_request_fn = mddev->pers->make_request;
2591
2592         mddev->changed = 1;
2593         md_new_event(mddev);
2594         return 0;
2595 }
2596
2597 static int restart_array(mddev_t *mddev)
2598 {
2599         struct gendisk *disk = mddev->gendisk;
2600         int err;
2601
2602         /*
2603          * Complain if it has no devices
2604          */
2605         err = -ENXIO;
2606         if (list_empty(&mddev->disks))
2607                 goto out;
2608
2609         if (mddev->pers) {
2610                 err = -EBUSY;
2611                 if (!mddev->ro)
2612                         goto out;
2613
2614                 mddev->safemode = 0;
2615                 mddev->ro = 0;
2616                 set_disk_ro(disk, 0);
2617
2618                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2619                         mdname(mddev));
2620                 /*
2621                  * Kick recovery or resync if necessary
2622                  */
2623                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2624                 md_wakeup_thread(mddev->thread);
2625                 err = 0;
2626         } else {
2627                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2628                         mdname(mddev));
2629                 err = -EINVAL;
2630         }
2631
2632 out:
2633         return err;
2634 }
2635
2636 static int do_md_stop(mddev_t * mddev, int ro)
2637 {
2638         int err = 0;
2639         struct gendisk *disk = mddev->gendisk;
2640
2641         if (mddev->pers) {
2642                 if (atomic_read(&mddev->active)>2) {
2643                         printk("md: %s still in use.\n",mdname(mddev));
2644                         return -EBUSY;
2645                 }
2646
2647                 if (mddev->sync_thread) {
2648                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2649                         md_unregister_thread(mddev->sync_thread);
2650                         mddev->sync_thread = NULL;
2651                 }
2652
2653                 del_timer_sync(&mddev->safemode_timer);
2654
2655                 invalidate_partition(disk, 0);
2656
2657                 if (ro) {
2658                         err  = -ENXIO;
2659                         if (mddev->ro==1)
2660                                 goto out;
2661                         mddev->ro = 1;
2662                 } else {
2663                         bitmap_flush(mddev);
2664                         md_super_wait(mddev);
2665                         if (mddev->ro)
2666                                 set_disk_ro(disk, 0);
2667                         blk_queue_make_request(mddev->queue, md_fail_request);
2668                         mddev->pers->stop(mddev);
2669                         if (mddev->pers->sync_request)
2670                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2671
2672                         module_put(mddev->pers->owner);
2673                         mddev->pers = NULL;
2674                         if (mddev->ro)
2675                                 mddev->ro = 0;
2676                 }
2677                 if (!mddev->in_sync) {
2678                         /* mark array as shutdown cleanly */
2679                         mddev->in_sync = 1;
2680                         md_update_sb(mddev);
2681                 }
2682                 if (ro)
2683                         set_disk_ro(disk, 1);
2684         }
2685
2686         bitmap_destroy(mddev);
2687         if (mddev->bitmap_file) {
2688                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2689                 fput(mddev->bitmap_file);
2690                 mddev->bitmap_file = NULL;
2691         }
2692         mddev->bitmap_offset = 0;
2693
2694         /*
2695          * Free resources if final stop
2696          */
2697         if (!ro) {
2698                 mdk_rdev_t *rdev;
2699                 struct list_head *tmp;
2700                 struct gendisk *disk;
2701                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2702
2703                 ITERATE_RDEV(mddev,rdev,tmp)
2704                         if (rdev->raid_disk >= 0) {
2705                                 char nm[20];
2706                                 sprintf(nm, "rd%d", rdev->raid_disk);
2707                                 sysfs_remove_link(&mddev->kobj, nm);
2708                         }
2709
2710                 export_array(mddev);
2711
2712                 mddev->array_size = 0;
2713                 disk = mddev->gendisk;
2714                 if (disk)
2715                         set_capacity(disk, 0);
2716                 mddev->changed = 1;
2717         } else
2718                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2719                         mdname(mddev));
2720         err = 0;
2721         md_new_event(mddev);
2722 out:
2723         return err;
2724 }
2725
2726 static void autorun_array(mddev_t *mddev)
2727 {
2728         mdk_rdev_t *rdev;
2729         struct list_head *tmp;
2730         int err;
2731
2732         if (list_empty(&mddev->disks))
2733                 return;
2734
2735         printk(KERN_INFO "md: running: ");
2736
2737         ITERATE_RDEV(mddev,rdev,tmp) {
2738                 char b[BDEVNAME_SIZE];
2739                 printk("<%s>", bdevname(rdev->bdev,b));
2740         }
2741         printk("\n");
2742
2743         err = do_md_run (mddev);
2744         if (err) {
2745                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2746                 do_md_stop (mddev, 0);
2747         }
2748 }
2749
2750 /*
2751  * lets try to run arrays based on all disks that have arrived
2752  * until now. (those are in pending_raid_disks)
2753  *
2754  * the method: pick the first pending disk, collect all disks with
2755  * the same UUID, remove all from the pending list and put them into
2756  * the 'same_array' list. Then order this list based on superblock
2757  * update time (freshest comes first), kick out 'old' disks and
2758  * compare superblocks. If everything's fine then run it.
2759  *
2760  * If "unit" is allocated, then bump its reference count
2761  */
2762 static void autorun_devices(int part)
2763 {
2764         struct list_head candidates;
2765         struct list_head *tmp;
2766         mdk_rdev_t *rdev0, *rdev;
2767         mddev_t *mddev;
2768         char b[BDEVNAME_SIZE];
2769
2770         printk(KERN_INFO "md: autorun ...\n");
2771         while (!list_empty(&pending_raid_disks)) {
2772                 dev_t dev;
2773                 rdev0 = list_entry(pending_raid_disks.next,
2774                                          mdk_rdev_t, same_set);
2775
2776                 printk(KERN_INFO "md: considering %s ...\n",
2777                         bdevname(rdev0->bdev,b));
2778                 INIT_LIST_HEAD(&candidates);
2779                 ITERATE_RDEV_PENDING(rdev,tmp)
2780                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2781                                 printk(KERN_INFO "md:  adding %s ...\n",
2782                                         bdevname(rdev->bdev,b));
2783                                 list_move(&rdev->same_set, &candidates);
2784                         }
2785                 /*
2786                  * now we have a set of devices, with all of them having
2787                  * mostly sane superblocks. It's time to allocate the
2788                  * mddev.
2789                  */
2790                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2791                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2792                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2793                         break;
2794                 }
2795                 if (part)
2796                         dev = MKDEV(mdp_major,
2797                                     rdev0->preferred_minor << MdpMinorShift);
2798                 else
2799                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2800
2801                 md_probe(dev, NULL, NULL);
2802                 mddev = mddev_find(dev);
2803                 if (!mddev) {
2804                         printk(KERN_ERR
2805                                 "md: cannot allocate memory for md drive.\n");
2806                         break;
2807                 }
2808                 if (mddev_lock(mddev))
2809                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2810                                mdname(mddev));
2811                 else if (mddev->raid_disks || mddev->major_version
2812                          || !list_empty(&mddev->disks)) {
2813                         printk(KERN_WARNING
2814                                 "md: %s already running, cannot run %s\n",
2815                                 mdname(mddev), bdevname(rdev0->bdev,b));
2816                         mddev_unlock(mddev);
2817                 } else {
2818                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2819                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2820                                 list_del_init(&rdev->same_set);
2821                                 if (bind_rdev_to_array(rdev, mddev))
2822                                         export_rdev(rdev);
2823                         }
2824                         autorun_array(mddev);
2825                         mddev_unlock(mddev);
2826                 }
2827                 /* on success, candidates will be empty, on error
2828                  * it won't...
2829                  */
2830                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2831                         export_rdev(rdev);
2832                 mddev_put(mddev);
2833         }
2834         printk(KERN_INFO "md: ... autorun DONE.\n");
2835 }
2836
2837 /*
2838  * import RAID devices based on one partition
2839  * if possible, the array gets run as well.
2840  */
2841
2842 static int autostart_array(dev_t startdev)
2843 {
2844         char b[BDEVNAME_SIZE];
2845         int err = -EINVAL, i;
2846         mdp_super_t *sb = NULL;
2847         mdk_rdev_t *start_rdev = NULL, *rdev;
2848
2849         start_rdev = md_import_device(startdev, 0, 0);
2850         if (IS_ERR(start_rdev))
2851                 return err;
2852
2853
2854         /* NOTE: this can only work for 0.90.0 superblocks */
2855         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2856         if (sb->major_version != 0 ||
2857             sb->minor_version != 90 ) {
2858                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2859                 export_rdev(start_rdev);
2860                 return err;
2861         }
2862
2863         if (test_bit(Faulty, &start_rdev->flags)) {
2864                 printk(KERN_WARNING
2865                         "md: can not autostart based on faulty %s!\n",
2866                         bdevname(start_rdev->bdev,b));
2867                 export_rdev(start_rdev);
2868                 return err;
2869         }
2870         list_add(&start_rdev->same_set, &pending_raid_disks);
2871
2872         for (i = 0; i < MD_SB_DISKS; i++) {
2873                 mdp_disk_t *desc = sb->disks + i;
2874                 dev_t dev = MKDEV(desc->major, desc->minor);
2875
2876                 if (!dev)
2877                         continue;
2878                 if (dev == startdev)
2879                         continue;
2880                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2881                         continue;
2882                 rdev = md_import_device(dev, 0, 0);
2883                 if (IS_ERR(rdev))
2884                         continue;
2885
2886                 list_add(&rdev->same_set, &pending_raid_disks);
2887         }
2888
2889         /*
2890          * possibly return codes
2891          */
2892         autorun_devices(0);
2893         return 0;
2894
2895 }
2896
2897
2898 static int get_version(void __user * arg)
2899 {
2900         mdu_version_t ver;
2901
2902         ver.major = MD_MAJOR_VERSION;
2903         ver.minor = MD_MINOR_VERSION;
2904         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2905
2906         if (copy_to_user(arg, &ver, sizeof(ver)))
2907                 return -EFAULT;
2908
2909         return 0;
2910 }
2911
2912 static int get_array_info(mddev_t * mddev, void __user * arg)
2913 {
2914         mdu_array_info_t info;
2915         int nr,working,active,failed,spare;
2916         mdk_rdev_t *rdev;
2917         struct list_head *tmp;
2918
2919         nr=working=active=failed=spare=0;
2920         ITERATE_RDEV(mddev,rdev,tmp) {
2921                 nr++;
2922                 if (test_bit(Faulty, &rdev->flags))
2923                         failed++;
2924                 else {
2925                         working++;
2926                         if (test_bit(In_sync, &rdev->flags))
2927                                 active++;
2928                         else
2929                                 spare++;
2930                 }
2931         }
2932
2933         info.major_version = mddev->major_version;
2934         info.minor_version = mddev->minor_version;
2935         info.patch_version = MD_PATCHLEVEL_VERSION;
2936         info.ctime         = mddev->ctime;
2937         info.level         = mddev->level;
2938         info.size          = mddev->size;
2939         info.nr_disks      = nr;
2940         info.raid_disks    = mddev->raid_disks;
2941         info.md_minor      = mddev->md_minor;
2942         info.not_persistent= !mddev->persistent;
2943
2944         info.utime         = mddev->utime;
2945         info.state         = 0;
2946         if (mddev->in_sync)
2947                 info.state = (1<<MD_SB_CLEAN);
2948         if (mddev->bitmap && mddev->bitmap_offset)
2949                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2950         info.active_disks  = active;
2951         info.working_disks = working;
2952         info.failed_disks  = failed;
2953         info.spare_disks   = spare;
2954
2955         info.layout        = mddev->layout;
2956         info.chunk_size    = mddev->chunk_size;
2957
2958         if (copy_to_user(arg, &info, sizeof(info)))
2959                 return -EFAULT;
2960
2961         return 0;
2962 }
2963
2964 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2965 {
2966         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2967         char *ptr, *buf = NULL;
2968         int err = -ENOMEM;
2969
2970         file = kmalloc(sizeof(*file), GFP_KERNEL);
2971         if (!file)
2972                 goto out;
2973
2974         /* bitmap disabled, zero the first byte and copy out */
2975         if (!mddev->bitmap || !mddev->bitmap->file) {
2976                 file->pathname[0] = '\0';
2977                 goto copy_out;
2978         }
2979
2980         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2981         if (!buf)
2982                 goto out;
2983
2984         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2985         if (!ptr)
2986                 goto out;
2987
2988         strcpy(file->pathname, ptr);
2989
2990 copy_out:
2991         err = 0;
2992         if (copy_to_user(arg, file, sizeof(*file)))
2993                 err = -EFAULT;
2994 out:
2995         kfree(buf);
2996         kfree(file);
2997         return err;
2998 }
2999
3000 static int get_disk_info(mddev_t * mddev, void __user * arg)
3001 {
3002         mdu_disk_info_t info;
3003         unsigned int nr;
3004         mdk_rdev_t *rdev;
3005
3006         if (copy_from_user(&info, arg, sizeof(info)))
3007                 return -EFAULT;
3008
3009         nr = info.number;
3010
3011         rdev = find_rdev_nr(mddev, nr);
3012         if (rdev) {
3013                 info.major = MAJOR(rdev->bdev->bd_dev);
3014                 info.minor = MINOR(rdev->bdev->bd_dev);
3015                 info.raid_disk = rdev->raid_disk;
3016                 info.state = 0;
3017                 if (test_bit(Faulty, &rdev->flags))
3018                         info.state |= (1<<MD_DISK_FAULTY);
3019                 else if (test_bit(In_sync, &rdev->flags)) {
3020                         info.state |= (1<<MD_DISK_ACTIVE);
3021                         info.state |= (1<<MD_DISK_SYNC);
3022                 }
3023                 if (test_bit(WriteMostly, &rdev->flags))
3024                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3025         } else {
3026                 info.major = info.minor = 0;
3027                 info.raid_disk = -1;
3028                 info.state = (1<<MD_DISK_REMOVED);
3029         }
3030
3031         if (copy_to_user(arg, &info, sizeof(info)))
3032                 return -EFAULT;
3033
3034         return 0;
3035 }
3036
3037 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3038 {
3039         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3040         mdk_rdev_t *rdev;
3041         dev_t dev = MKDEV(info->major,info->minor);
3042
3043         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3044                 return -EOVERFLOW;
3045
3046         if (!mddev->raid_disks) {
3047                 int err;
3048                 /* expecting a device which has a superblock */
3049                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3050                 if (IS_ERR(rdev)) {
3051                         printk(KERN_WARNING
3052                                 "md: md_import_device returned %ld\n",
3053                                 PTR_ERR(rdev));
3054                         return PTR_ERR(rdev);
3055                 }
3056                 if (!list_empty(&mddev->disks)) {
3057                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3058                                                         mdk_rdev_t, same_set);
3059                         int err = super_types[mddev->major_version]
3060                                 .load_super(rdev, rdev0, mddev->minor_version);
3061                         if (err < 0) {
3062                                 printk(KERN_WARNING
3063                                         "md: %s has different UUID to %s\n",
3064                                         bdevname(rdev->bdev,b),
3065                                         bdevname(rdev0->bdev,b2));
3066                                 export_rdev(rdev);
3067                                 return -EINVAL;
3068                         }
3069                 }
3070                 err = bind_rdev_to_array(rdev, mddev);
3071                 if (err)
3072                         export_rdev(rdev);
3073                 return err;
3074         }
3075
3076         /*
3077          * add_new_disk can be used once the array is assembled
3078          * to add "hot spares".  They must already have a superblock
3079          * written
3080          */
3081         if (mddev->pers) {
3082                 int err;
3083                 if (!mddev->pers->hot_add_disk) {
3084                         printk(KERN_WARNING
3085                                 "%s: personality does not support diskops!\n",
3086                                mdname(mddev));
3087                         return -EINVAL;
3088                 }
3089                 if (mddev->persistent)
3090                         rdev = md_import_device(dev, mddev->major_version,
3091                                                 mddev->minor_version);
3092                 else
3093                         rdev = md_import_device(dev, -1, -1);
3094                 if (IS_ERR(rdev)) {
3095                         printk(KERN_WARNING
3096                                 "md: md_import_device returned %ld\n",
3097                                 PTR_ERR(rdev));
3098                         return PTR_ERR(rdev);
3099                 }
3100                 /* set save_raid_disk if appropriate */
3101                 if (!mddev->persistent) {
3102                         if (info->state & (1<<MD_DISK_SYNC)  &&
3103                             info->raid_disk < mddev->raid_disks)
3104                                 rdev->raid_disk = info->raid_disk;
3105                         else
3106                                 rdev->raid_disk = -1;
3107                 } else
3108                         super_types[mddev->major_version].
3109                                 validate_super(mddev, rdev);
3110                 rdev->saved_raid_disk = rdev->raid_disk;
3111
3112                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3113                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3114                         set_bit(WriteMostly, &rdev->flags);
3115
3116                 rdev->raid_disk = -1;
3117                 err = bind_rdev_to_array(rdev, mddev);
3118                 if (err)
3119                         export_rdev(rdev);
3120
3121                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3122                 md_wakeup_thread(mddev->thread);
3123                 return err;
3124         }
3125
3126         /* otherwise, add_new_disk is only allowed
3127          * for major_version==0 superblocks
3128          */
3129         if (mddev->major_version != 0) {
3130                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3131                        mdname(mddev));
3132                 return -EINVAL;
3133         }
3134
3135         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3136                 int err;
3137                 rdev = md_import_device (dev, -1, 0);
3138                 if (IS_ERR(rdev)) {
3139                         printk(KERN_WARNING
3140                                 "md: error, md_import_device() returned %ld\n",
3141                                 PTR_ERR(rdev));
3142                         return PTR_ERR(rdev);
3143                 }
3144                 rdev->desc_nr = info->number;
3145                 if (info->raid_disk < mddev->raid_disks)
3146                         rdev->raid_disk = info->raid_disk;
3147                 else
3148                         rdev->raid_disk = -1;
3149
3150                 rdev->flags = 0;
3151
3152                 if (rdev->raid_disk < mddev->raid_disks)
3153                         if (info->state & (1<<MD_DISK_SYNC))
3154                                 set_bit(In_sync, &rdev->flags);
3155
3156                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3157                         set_bit(WriteMostly, &rdev->flags);
3158
3159                 if (!mddev->persistent) {
3160                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3161                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3162                 } else
3163                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3164                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3165
3166                 err = bind_rdev_to_array(rdev, mddev);
3167                 if (err) {
3168                         export_rdev(rdev);
3169                         return err;
3170                 }
3171         }
3172
3173         return 0;
3174 }
3175
3176 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3177 {
3178         char b[BDEVNAME_SIZE];
3179         mdk_rdev_t *rdev;
3180
3181         if (!mddev->pers)
3182                 return -ENODEV;
3183
3184         rdev = find_rdev(mddev, dev);
3185         if (!rdev)
3186                 return -ENXIO;
3187
3188         if (rdev->raid_disk >= 0)
3189                 goto busy;
3190
3191         kick_rdev_from_array(rdev);
3192         md_update_sb(mddev);
3193         md_new_event(mddev);
3194
3195         return 0;
3196 busy:
3197         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3198                 bdevname(rdev->bdev,b), mdname(mddev));
3199         return -EBUSY;
3200 }
3201
3202 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3203 {
3204         char b[BDEVNAME_SIZE];
3205         int err;
3206         unsigned int size;
3207         mdk_rdev_t *rdev;
3208
3209         if (!mddev->pers)
3210                 return -ENODEV;
3211
3212         if (mddev->major_version != 0) {
3213                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3214                         " version-0 superblocks.\n",
3215                         mdname(mddev));
3216                 return -EINVAL;
3217         }
3218         if (!mddev->pers->hot_add_disk) {
3219                 printk(KERN_WARNING
3220                         "%s: personality does not support diskops!\n",
3221                         mdname(mddev));
3222                 return -EINVAL;
3223         }
3224
3225         rdev = md_import_device (dev, -1, 0);
3226         if (IS_ERR(rdev)) {
3227                 printk(KERN_WARNING
3228                         "md: error, md_import_device() returned %ld\n",
3229                         PTR_ERR(rdev));
3230                 return -EINVAL;
3231         }
3232
3233         if (mddev->persistent)
3234                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3235         else
3236                 rdev->sb_offset =
3237                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3238
3239         size = calc_dev_size(rdev, mddev->chunk_size);
3240         rdev->size = size;
3241
3242         if (test_bit(Faulty, &rdev->flags)) {
3243                 printk(KERN_WARNING
3244                         "md: can not hot-add faulty %s disk to %s!\n",
3245                         bdevname(rdev->bdev,b), mdname(mddev));
3246                 err = -EINVAL;
3247                 goto abort_export;
3248         }
3249         clear_bit(In_sync, &rdev->flags);
3250         rdev->desc_nr = -1;
3251         err = bind_rdev_to_array(rdev, mddev);
3252         if (err)
3253                 goto abort_export;
3254
3255         /*
3256          * The rest should better be atomic, we can have disk failures
3257          * noticed in interrupt contexts ...
3258          */
3259
3260         if (rdev->desc_nr == mddev->max_disks) {
3261                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3262                         mdname(mddev));
3263                 err = -EBUSY;
3264                 goto abort_unbind_export;
3265         }
3266
3267         rdev->raid_disk = -1;
3268
3269         md_update_sb(mddev);
3270
3271         /*
3272          * Kick recovery, maybe this spare has to be added to the
3273          * array immediately.
3274          */
3275         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3276         md_wakeup_thread(mddev->thread);
3277         md_new_event(mddev);
3278         return 0;
3279
3280 abort_unbind_export:
3281         unbind_rdev_from_array(rdev);
3282
3283 abort_export:
3284         export_rdev(rdev);
3285         return err;
3286 }
3287
3288 /* similar to deny_write_access, but accounts for our holding a reference
3289  * to the file ourselves */
3290 static int deny_bitmap_write_access(struct file * file)
3291 {
3292         struct inode *inode = file->f_mapping->host;
3293
3294         spin_lock(&inode->i_lock);
3295         if (atomic_read(&inode->i_writecount) > 1) {
3296                 spin_unlock(&inode->i_lock);
3297                 return -ETXTBSY;
3298         }
3299         atomic_set(&inode->i_writecount, -1);
3300         spin_unlock(&inode->i_lock);
3301
3302         return 0;
3303 }
3304
3305 static int set_bitmap_file(mddev_t *mddev, int fd)
3306 {
3307         int err;
3308
3309         if (mddev->pers) {
3310                 if (!mddev->pers->quiesce)
3311                         return -EBUSY;
3312                 if (mddev->recovery || mddev->sync_thread)
3313                         return -EBUSY;
3314                 /* we should be able to change the bitmap.. */
3315         }
3316
3317
3318         if (fd >= 0) {
3319                 if (mddev->bitmap)
3320                         return -EEXIST; /* cannot add when bitmap is present */
3321                 mddev->bitmap_file = fget(fd);
3322
3323                 if (mddev->bitmap_file == NULL) {
3324                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3325                                mdname(mddev));
3326                         return -EBADF;
3327                 }
3328
3329                 err = deny_bitmap_write_access(mddev->bitmap_file);
3330                 if (err) {
3331                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3332                                mdname(mddev));
3333                         fput(mddev->bitmap_file);
3334                         mddev->bitmap_file = NULL;
3335                         return err;
3336                 }
3337                 mddev->bitmap_offset = 0; /* file overrides offset */
3338         } else if (mddev->bitmap == NULL)
3339                 return -ENOENT; /* cannot remove what isn't there */
3340         err = 0;
3341         if (mddev->pers) {
3342                 mddev->pers->quiesce(mddev, 1);
3343                 if (fd >= 0)
3344                         err = bitmap_create(mddev);
3345                 if (fd < 0 || err)
3346                         bitmap_destroy(mddev);
3347                 mddev->pers->quiesce(mddev, 0);
3348         } else if (fd < 0) {
3349                 if (mddev->bitmap_file)
3350                         fput(mddev->bitmap_file);
3351                 mddev->bitmap_file = NULL;
3352         }
3353
3354         return err;
3355 }
3356
3357 /*
3358  * set_array_info is used two different ways
3359  * The original usage is when creating a new array.
3360  * In this usage, raid_disks is > 0 and it together with
3361  *  level, size, not_persistent,layout,chunksize determine the
3362  *  shape of the array.
3363  *  This will always create an array with a type-0.90.0 superblock.
3364  * The newer usage is when assembling an array.
3365  *  In this case raid_disks will be 0, and the major_version field is
3366  *  use to determine which style super-blocks are to be found on the devices.
3367  *  The minor and patch _version numbers are also kept incase the
3368  *  super_block handler wishes to interpret them.
3369  */
3370 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3371 {
3372
3373         if (info->raid_disks == 0) {
3374                 /* just setting version number for superblock loading */
3375                 if (info->major_version < 0 ||
3376                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3377                     super_types[info->major_version].name == NULL) {
3378                         /* maybe try to auto-load a module? */
3379                         printk(KERN_INFO
3380                                 "md: superblock version %d not known\n",
3381                                 info->major_version);
3382                         return -EINVAL;
3383                 }
3384                 mddev->major_version = info->major_version;
3385                 mddev->minor_version = info->minor_version;
3386                 mddev->patch_version = info->patch_version;
3387                 return 0;
3388         }
3389         mddev->major_version = MD_MAJOR_VERSION;
3390         mddev->minor_version = MD_MINOR_VERSION;
3391         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3392         mddev->ctime         = get_seconds();
3393
3394         mddev->level         = info->level;
3395         mddev->size          = info->size;
3396         mddev->raid_disks    = info->raid_disks;
3397         /* don't set md_minor, it is determined by which /dev/md* was
3398          * openned
3399          */
3400         if (info->state & (1<<MD_SB_CLEAN))
3401                 mddev->recovery_cp = MaxSector;
3402         else
3403                 mddev->recovery_cp = 0;
3404         mddev->persistent    = ! info->not_persistent;
3405
3406         mddev->layout        = info->layout;
3407         mddev->chunk_size    = info->chunk_size;
3408
3409         mddev->max_disks     = MD_SB_DISKS;
3410
3411         mddev->sb_dirty      = 1;
3412
3413         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3414         mddev->bitmap_offset = 0;
3415
3416         /*
3417          * Generate a 128 bit UUID
3418          */
3419         get_random_bytes(mddev->uuid, 16);
3420
3421         return 0;
3422 }
3423
3424 static int update_size(mddev_t *mddev, unsigned long size)
3425 {
3426         mdk_rdev_t * rdev;
3427         int rv;
3428         struct list_head *tmp;
3429
3430         if (mddev->pers->resize == NULL)
3431                 return -EINVAL;
3432         /* The "size" is the amount of each device that is used.
3433          * This can only make sense for arrays with redundancy.
3434          * linear and raid0 always use whatever space is available
3435          * We can only consider changing the size if no resync
3436          * or reconstruction is happening, and if the new size
3437          * is acceptable. It must fit before the sb_offset or,
3438          * if that is <data_offset, it must fit before the
3439          * size of each device.
3440          * If size is zero, we find the largest size that fits.
3441          */
3442         if (mddev->sync_thread)
3443                 return -EBUSY;
3444         ITERATE_RDEV(mddev,rdev,tmp) {
3445                 sector_t avail;
3446                 int fit = (size == 0);
3447                 if (rdev->sb_offset > rdev->data_offset)
3448                         avail = (rdev->sb_offset*2) - rdev->data_offset;
3449                 else
3450                         avail = get_capacity(rdev->bdev->bd_disk)
3451                                 - rdev->data_offset;
3452                 if (fit && (size == 0 || size > avail/2))
3453                         size = avail/2;
3454                 if (avail < ((sector_t)size << 1))
3455                         return -ENOSPC;
3456         }
3457         rv = mddev->pers->resize(mddev, (sector_t)size *2);
3458         if (!rv) {
3459                 struct block_device *bdev;
3460
3461                 bdev = bdget_disk(mddev->gendisk, 0);
3462                 if (bdev) {
3463                         mutex_lock(&bdev->bd_inode->i_mutex);
3464                         i_size_write(bdev->bd_inode, mddev->array_size << 10);
3465                         mutex_unlock(&bdev->bd_inode->i_mutex);
3466                         bdput(bdev);
3467                 }
3468         }
3469         return rv;
3470 }
3471
3472 static int update_raid_disks(mddev_t *mddev, int raid_disks)
3473 {
3474         int rv;
3475         /* change the number of raid disks */
3476         if (mddev->pers->reshape == NULL)
3477                 return -EINVAL;
3478         if (raid_disks <= 0 ||
3479             raid_disks >= mddev->max_disks)
3480                 return -EINVAL;
3481         if (mddev->sync_thread)
3482                 return -EBUSY;
3483         rv = mddev->pers->reshape(mddev, raid_disks);
3484         if (!rv) {
3485                 struct block_device *bdev;
3486
3487                 bdev = bdget_disk(mddev->gendisk, 0);
3488                 if (bdev) {
3489                         mutex_lock(&bdev->bd_inode->i_mutex);
3490                         i_size_write(bdev->bd_inode, mddev->array_size << 10);
3491                         mutex_unlock(&bdev->bd_inode->i_mutex);
3492                         bdput(bdev);
3493                 }
3494         }
3495         return rv;
3496 }
3497
3498
3499 /*
3500  * update_array_info is used to change the configuration of an
3501  * on-line array.
3502  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3503  * fields in the info are checked against the array.
3504  * Any differences that cannot be handled will cause an error.
3505  * Normally, only one change can be managed at a time.
3506  */
3507 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3508 {
3509         int rv = 0;
3510         int cnt = 0;
3511         int state = 0;
3512
3513         /* calculate expected state,ignoring low bits */
3514         if (mddev->bitmap && mddev->bitmap_offset)
3515                 state |= (1 << MD_SB_BITMAP_PRESENT);
3516
3517         if (mddev->major_version != info->major_version ||
3518             mddev->minor_version != info->minor_version ||
3519 /*          mddev->patch_version != info->patch_version || */
3520             mddev->ctime         != info->ctime         ||
3521             mddev->level         != info->level         ||
3522 /*          mddev->layout        != info->layout        || */
3523             !mddev->persistent   != info->not_persistent||
3524             mddev->chunk_size    != info->chunk_size    ||
3525             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3526             ((state^info->state) & 0xfffffe00)
3527                 )
3528                 return -EINVAL;
3529         /* Check there is only one change */
3530         if (mddev->size != info->size) cnt++;
3531         if (mddev->raid_disks != info->raid_disks) cnt++;
3532         if (mddev->layout != info->layout) cnt++;
3533         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3534         if (cnt == 0) return 0;
3535         if (cnt > 1) return -EINVAL;
3536
3537         if (mddev->layout != info->layout) {
3538                 /* Change layout
3539                  * we don't need to do anything at the md level, the
3540                  * personality will take care of it all.
3541                  */
3542                 if (mddev->pers->reconfig == NULL)
3543                         return -EINVAL;
3544                 else
3545                         return mddev->pers->reconfig(mddev, info->layout, -1);
3546         }
3547         if (mddev->size != info->size)
3548                 rv = update_size(mddev, info->size);
3549
3550         if (mddev->raid_disks    != info->raid_disks)
3551                 rv = update_raid_disks(mddev, info->raid_disks);
3552
3553         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3554                 if (mddev->pers->quiesce == NULL)
3555                         return -EINVAL;
3556                 if (mddev->recovery || mddev->sync_thread)
3557                         return -EBUSY;
3558                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3559                         /* add the bitmap */
3560                         if (mddev->bitmap)
3561                                 return -EEXIST;
3562                         if (mddev->default_bitmap_offset == 0)
3563                                 return -EINVAL;
3564                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3565                         mddev->pers->quiesce(mddev, 1);
3566                         rv = bitmap_create(mddev);
3567                         if (rv)
3568                                 bitmap_destroy(mddev);
3569                         mddev->pers->quiesce(mddev, 0);
3570                 } else {
3571                         /* remove the bitmap */
3572                         if (!mddev->bitmap)
3573                                 return -ENOENT;
3574                         if (mddev->bitmap->file)
3575                                 return -EINVAL;
3576                         mddev->pers->quiesce(mddev, 1);
3577                         bitmap_destroy(mddev);
3578                         mddev->pers->quiesce(mddev, 0);
3579                         mddev->bitmap_offset = 0;
3580                 }
3581         }
3582         md_update_sb(mddev);
3583         return rv;
3584 }
3585
3586 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3587 {
3588         mdk_rdev_t *rdev;
3589
3590         if (mddev->pers == NULL)
3591                 return -ENODEV;
3592
3593         rdev = find_rdev(mddev, dev);
3594         if (!rdev)
3595                 return -ENODEV;
3596
3597         md_error(mddev, rdev);
3598         return 0;
3599 }
3600
3601 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3602 {
3603         mddev_t *mddev = bdev->bd_disk->private_data;
3604
3605         geo->heads = 2;
3606         geo->sectors = 4;
3607         geo->cylinders = get_capacity(mddev->gendisk) / 8;
3608         return 0;
3609 }
3610
3611 static int md_ioctl(struct inode *inode, struct file *file,
3612                         unsigned int cmd, unsigned long arg)
3613 {
3614         int err = 0;
3615         void __user *argp = (void __user *)arg;
3616         mddev_t *mddev = NULL;
3617
3618         if (!capable(CAP_SYS_ADMIN))
3619                 return -EACCES;
3620
3621         /*
3622          * Commands dealing with the RAID driver but not any
3623          * particular array:
3624          */
3625         switch (cmd)
3626         {
3627                 case RAID_VERSION:
3628                         err = get_version(argp);
3629                         goto done;
3630
3631                 case PRINT_RAID_DEBUG:
3632                         err = 0;
3633                         md_print_devices();
3634                         goto done;
3635
3636 #ifndef MODULE
3637                 case RAID_AUTORUN:
3638                         err = 0;
3639                         autostart_arrays(arg);
3640                         goto done;
3641 #endif
3642                 default:;
3643         }
3644
3645         /*
3646          * Commands creating/starting a new array:
3647          */
3648
3649         mddev = inode->i_bdev->bd_disk->private_data;
3650
3651         if (!mddev) {
3652                 BUG();
3653                 goto abort;
3654         }
3655
3656
3657         if (cmd == START_ARRAY) {
3658                 /* START_ARRAY doesn't need to lock the array as autostart_array
3659                  * does the locking, and it could even be a different array
3660                  */
3661                 static int cnt = 3;
3662                 if (cnt > 0 ) {
3663                         printk(KERN_WARNING
3664                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3665                                "This will not be supported beyond July 2006\n",
3666                                current->comm, current->pid);
3667                         cnt--;
3668                 }
3669                 err = autostart_array(new_decode_dev(arg));
3670                 if (err) {
3671                         printk(KERN_WARNING "md: autostart failed!\n");
3672                         goto abort;
3673                 }
3674                 goto done;
3675         }
3676
3677         err = mddev_lock(mddev);
3678         if (err) {
3679                 printk(KERN_INFO
3680                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3681                         err, cmd);
3682                 goto abort;
3683         }
3684
3685         switch (cmd)
3686         {
3687                 case SET_ARRAY_INFO:
3688                         {
3689                                 mdu_array_info_t info;
3690                                 if (!arg)
3691                                         memset(&info, 0, sizeof(info));
3692                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3693                                         err = -EFAULT;
3694                                         goto abort_unlock;
3695                                 }
3696                                 if (mddev->pers) {
3697                                         err = update_array_info(mddev, &info);
3698                                         if (err) {
3699                                                 printk(KERN_WARNING "md: couldn't update"
3700                                                        " array info. %d\n", err);
3701                                                 goto abort_unlock;
3702                                         }
3703                                         goto done_unlock;
3704                                 }
3705                                 if (!list_empty(&mddev->disks)) {
3706                                         printk(KERN_WARNING
3707                                                "md: array %s already has disks!\n",
3708                                                mdname(mddev));
3709                                         err = -EBUSY;
3710                                         goto abort_unlock;
3711                                 }
3712                                 if (mddev->raid_disks) {
3713                                         printk(KERN_WARNING
3714                                                "md: array %s already initialised!\n",
3715                                                mdname(mddev));
3716                                         err = -EBUSY;
3717                                         goto abort_unlock;
3718                                 }
3719                                 err = set_array_info(mddev, &info);
3720                                 if (err) {
3721                                         printk(KERN_WARNING "md: couldn't set"
3722                                                " array info. %d\n", err);
3723                                         goto abort_unlock;
3724                                 }
3725                         }
3726                         goto done_unlock;
3727
3728                 default:;
3729         }
3730
3731         /*
3732          * Commands querying/configuring an existing array:
3733          */
3734         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3735          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3736         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3737                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3738                 err = -ENODEV;
3739                 goto abort_unlock;
3740         }
3741
3742         /*
3743          * Commands even a read-only array can execute:
3744          */
3745         switch (cmd)
3746         {
3747                 case GET_ARRAY_INFO:
3748                         err = get_array_info(mddev, argp);
3749                         goto done_unlock;
3750
3751                 case GET_BITMAP_FILE:
3752                         err = get_bitmap_file(mddev, argp);
3753                         goto done_unlock;
3754
3755                 case GET_DISK_INFO:
3756                         err = get_disk_info(mddev, argp);
3757                         goto done_unlock;
3758
3759                 case RESTART_ARRAY_RW:
3760                         err = restart_array(mddev);
3761                         goto done_unlock;
3762
3763                 case STOP_ARRAY:
3764                         err = do_md_stop (mddev, 0);
3765                         goto done_unlock;
3766
3767                 case STOP_ARRAY_RO:
3768                         err = do_md_stop (mddev, 1);
3769                         goto done_unlock;
3770
3771         /*
3772          * We have a problem here : there is no easy way to give a CHS
3773          * virtual geometry. We currently pretend that we have a 2 heads
3774          * 4 sectors (with a BIG number of cylinders...). This drives
3775          * dosfs just mad... ;-)
3776          */
3777         }
3778
3779         /*
3780          * The remaining ioctls are changing the state of the
3781          * superblock, so we do not allow them on read-only arrays.
3782          * However non-MD ioctls (e.g. get-size) will still come through
3783          * here and hit the 'default' below, so only disallow
3784          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3785          */
3786         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3787             mddev->ro && mddev->pers) {
3788                 if (mddev->ro == 2) {
3789                         mddev->ro = 0;
3790                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3791                 md_wakeup_thread(mddev->thread);
3792
3793                 } else {
3794                         err = -EROFS;
3795                         goto abort_unlock;
3796                 }
3797         }
3798
3799         switch (cmd)
3800         {
3801                 case ADD_NEW_DISK:
3802                 {
3803                         mdu_disk_info_t info;
3804                         if (copy_from_user(&info, argp, sizeof(info)))
3805                                 err = -EFAULT;
3806                         else
3807                                 err = add_new_disk(mddev, &info);
3808                         goto done_unlock;
3809                 }
3810
3811                 case HOT_REMOVE_DISK:
3812                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3813                         goto done_unlock;
3814
3815                 case HOT_ADD_DISK:
3816                         err = hot_add_disk(mddev, new_decode_dev(arg));
3817                         goto done_unlock;
3818
3819                 case SET_DISK_FAULTY:
3820                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3821                         goto done_unlock;
3822
3823                 case RUN_ARRAY:
3824                         err = do_md_run (mddev);
3825                         goto done_unlock;
3826
3827                 case SET_BITMAP_FILE:
3828                         err = set_bitmap_file(mddev, (int)arg);
3829                         goto done_unlock;
3830
3831                 default:
3832                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3833                                 printk(KERN_WARNING "md: %s(pid %d) used"
3834                                         " obsolete MD ioctl, upgrade your"
3835                                         " software to use new ictls.\n",
3836                                         current->comm, current->pid);
3837                         err = -EINVAL;
3838                         goto abort_unlock;
3839         }
3840
3841 done_unlock:
3842 abort_unlock:
3843         mddev_unlock(mddev);
3844
3845         return err;
3846 done:
3847         if (err)
3848                 MD_BUG();
3849 abort:
3850         return err;
3851 }
3852
3853 static int md_open(struct inode *inode, struct file *file)
3854 {
3855         /*
3856          * Succeed if we can lock the mddev, which confirms that
3857          * it isn't being stopped right now.
3858          */
3859         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3860         int err;
3861
3862         if ((err = mddev_lock(mddev)))
3863                 goto out;
3864
3865         err = 0;
3866         mddev_get(mddev);
3867         mddev_unlock(mddev);
3868
3869         check_disk_change(inode->i_bdev);
3870  out:
3871         return err;
3872 }
3873
3874 static int md_release(struct inode *inode, struct file * file)
3875 {
3876         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3877
3878         if (!mddev)
3879                 BUG();
3880         mddev_put(mddev);
3881
3882         return 0;
3883 }
3884
3885 static int md_media_changed(struct gendisk *disk)
3886 {
3887         mddev_t *mddev = disk->private_data;
3888
3889         return mddev->changed;
3890 }
3891
3892 static int md_revalidate(struct gendisk *disk)
3893 {
3894         mddev_t *mddev = disk->private_data;
3895
3896         mddev->changed = 0;
3897         return 0;
3898 }
3899 static struct block_device_operations md_fops =
3900 {
3901         .owner          = THIS_MODULE,
3902         .open           = md_open,
3903         .release        = md_release,
3904         .ioctl          = md_ioctl,
3905         .getgeo         = md_getgeo,
3906         .media_changed  = md_media_changed,
3907         .revalidate_disk= md_revalidate,
3908 };
3909
3910 static int md_thread(void * arg)
3911 {
3912         mdk_thread_t *thread = arg;
3913
3914         /*
3915          * md_thread is a 'system-thread', it's priority should be very
3916          * high. We avoid resource deadlocks individually in each
3917          * raid personality. (RAID5 does preallocation) We also use RR and
3918          * the very same RT priority as kswapd, thus we will never get
3919          * into a priority inversion deadlock.
3920          *
3921          * we definitely have to have equal or higher priority than
3922          * bdflush, otherwise bdflush will deadlock if there are too
3923          * many dirty RAID5 blocks.
3924          */
3925
3926         allow_signal(SIGKILL);
3927         while (!kthread_should_stop()) {
3928
3929                 /* We need to wait INTERRUPTIBLE so that
3930                  * we don't add to the load-average.
3931                  * That means we need to be sure no signals are
3932                  * pending
3933                  */
3934                 if (signal_pending(current))
3935                         flush_signals(current);
3936
3937                 wait_event_interruptible_timeout
3938                         (thread->wqueue,
3939                          test_bit(THREAD_WAKEUP, &thread->flags)
3940                          || kthread_should_stop(),
3941                          thread->timeout);
3942                 try_to_freeze();
3943
3944                 clear_bit(THREAD_WAKEUP, &thread->flags);
3945
3946                 thread->run(thread->mddev);
3947         }
3948
3949         return 0;
3950 }
3951
3952 void md_wakeup_thread(mdk_thread_t *thread)
3953 {
3954         if (thread) {
3955                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3956                 set_bit(THREAD_WAKEUP, &thread->flags);
3957                 wake_up(&thread->wqueue);
3958         }
3959 }
3960
3961 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3962                                  const char *name)
3963 {
3964         mdk_thread_t *thread;
3965
3966         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3967         if (!thread)
3968                 return NULL;
3969
3970         init_waitqueue_head(&thread->wqueue);
3971
3972         thread->run = run;
3973         thread->mddev = mddev;
3974         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3975         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3976         if (IS_ERR(thread->tsk)) {
3977                 kfree(thread);
3978                 return NULL;
3979         }
3980         return thread;
3981 }
3982
3983 void md_unregister_thread(mdk_thread_t *thread)
3984 {
3985         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3986
3987         kthread_stop(thread->tsk);
3988         kfree(thread);
3989 }
3990
3991 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3992 {
3993         if (!mddev) {
3994                 MD_BUG();
3995                 return;
3996         }
3997
3998         if (!rdev || test_bit(Faulty, &rdev->flags))
3999                 return;
4000 /*
4001         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4002                 mdname(mddev),
4003                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4004                 __builtin_return_address(0),__builtin_return_address(1),
4005                 __builtin_return_address(2),__builtin_return_address(3));
4006 */
4007         if (!mddev->pers->error_handler)
4008                 return;
4009         mddev->pers->error_handler(mddev,rdev);
4010         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4011         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4012         md_wakeup_thread(mddev->thread);
4013         md_new_event(mddev);
4014 }
4015
4016 /* seq_file implementation /proc/mdstat */
4017
4018 static void status_unused(struct seq_file *seq)
4019 {
4020         int i = 0;
4021         mdk_rdev_t *rdev;
4022         struct list_head *tmp;
4023
4024         seq_printf(seq, "unused devices: ");
4025
4026         ITERATE_RDEV_PENDING(rdev,tmp) {
4027                 char b[BDEVNAME_SIZE];
4028                 i++;
4029                 seq_printf(seq, "%s ",
4030                               bdevname(rdev->bdev,b));
4031         }
4032         if (!i)
4033                 seq_printf(seq, "<none>");
4034
4035         seq_printf(seq, "\n");
4036 }
4037
4038
4039 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4040 {
4041         unsigned long max_blocks, resync, res, dt, db, rt;
4042
4043         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4044
4045         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4046                 max_blocks = mddev->resync_max_sectors >> 1;
4047         else
4048                 max_blocks = mddev->size;
4049
4050         /*
4051          * Should not happen.
4052          */
4053         if (!max_blocks) {
4054                 MD_BUG();
4055                 return;
4056         }
4057         res = (resync/1024)*1000/(max_blocks/1024 + 1);
4058         {
4059                 int i, x = res/50, y = 20-x;
4060                 seq_printf(seq, "[");
4061                 for (i = 0; i < x; i++)
4062                         seq_printf(seq, "=");
4063                 seq_printf(seq, ">");
4064                 for (i = 0; i < y; i++)
4065                         seq_printf(seq, ".");
4066                 seq_printf(seq, "] ");
4067         }
4068         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
4069                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4070                        "resync" : "recovery"),
4071                       res/10, res % 10, resync, max_blocks);
4072
4073         /*
4074          * We do not want to overflow, so the order of operands and
4075          * the * 100 / 100 trick are important. We do a +1 to be
4076          * safe against division by zero. We only estimate anyway.
4077          *
4078          * dt: time from mark until now
4079          * db: blocks written from mark until now
4080          * rt: remaining time
4081          */
4082         dt = ((jiffies - mddev->resync_mark) / HZ);
4083         if (!dt) dt++;
4084         db = resync - (mddev->resync_mark_cnt/2);
4085         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
4086
4087         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4088
4089         seq_printf(seq, " speed=%ldK/sec", db/dt);
4090 }
4091
4092 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4093 {
4094         struct list_head *tmp;
4095         loff_t l = *pos;
4096         mddev_t *mddev;
4097
4098         if (l >= 0x10000)
4099                 return NULL;
4100         if (!l--)
4101                 /* header */
4102                 return (void*)1;
4103
4104         spin_lock(&all_mddevs_lock);
4105         list_for_each(tmp,&all_mddevs)
4106                 if (!l--) {
4107                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4108                         mddev_get(mddev);
4109                         spin_unlock(&all_mddevs_lock);
4110                         return mddev;
4111                 }
4112         spin_unlock(&all_mddevs_lock);
4113         if (!l--)
4114                 return (void*)2;/* tail */
4115         return NULL;
4116 }
4117
4118 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4119 {
4120         struct list_head *tmp;
4121         mddev_t *next_mddev, *mddev = v;
4122
4123         ++*pos;
4124         if (v == (void*)2)
4125                 return NULL;
4126
4127         spin_lock(&all_mddevs_lock);
4128         if (v == (void*)1)
4129                 tmp = all_mddevs.next;
4130         else
4131                 tmp = mddev->all_mddevs.next;
4132         if (tmp != &all_mddevs)
4133                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4134         else {
4135                 next_mddev = (void*)2;
4136                 *pos = 0x10000;
4137         }
4138         spin_unlock(&all_mddevs_lock);
4139
4140         if (v != (void*)1)
4141                 mddev_put(mddev);
4142         return next_mddev;
4143
4144 }
4145
4146 static void md_seq_stop(struct seq_file *seq, void *v)
4147 {
4148         mddev_t *mddev = v;
4149
4150         if (mddev && v != (void*)1 && v != (void*)2)
4151                 mddev_put(mddev);
4152 }
4153
4154 struct mdstat_info {
4155         int event;
4156 };
4157
4158 static int md_seq_show(struct seq_file *seq, void *v)
4159 {
4160         mddev_t *mddev = v;
4161         sector_t size;
4162         struct list_head *tmp2;
4163         mdk_rdev_t *rdev;
4164         struct mdstat_info *mi = seq->private;
4165         struct bitmap *bitmap;
4166
4167         if (v == (void*)1) {
4168                 struct mdk_personality *pers;
4169                 seq_printf(seq, "Personalities : ");
4170                 spin_lock(&pers_lock);
4171                 list_for_each_entry(pers, &pers_list, list)
4172                         seq_printf(seq, "[%s] ", pers->name);
4173
4174                 spin_unlock(&pers_lock);
4175                 seq_printf(seq, "\n");
4176                 mi->event = atomic_read(&md_event_count);
4177                 return 0;
4178         }
4179         if (v == (void*)2) {
4180                 status_unused(seq);
4181                 return 0;
4182         }
4183
4184         if (mddev_lock(mddev)!=0)
4185                 return -EINTR;
4186         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4187                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4188                                                 mddev->pers ? "" : "in");
4189                 if (mddev->pers) {
4190                         if (mddev->ro==1)
4191                                 seq_printf(seq, " (read-only)");
4192                         if (mddev->ro==2)
4193                                 seq_printf(seq, "(auto-read-only)");
4194                         seq_printf(seq, " %s", mddev->pers->name);
4195                 }
4196
4197                 size = 0;
4198                 ITERATE_RDEV(mddev,rdev,tmp2) {
4199                         char b[BDEVNAME_SIZE];
4200                         seq_printf(seq, " %s[%d]",
4201                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4202                         if (test_bit(WriteMostly, &rdev->flags))
4203                                 seq_printf(seq, "(W)");
4204                         if (test_bit(Faulty, &rdev->flags)) {
4205                                 seq_printf(seq, "(F)");
4206                                 continue;
4207                         } else if (rdev->raid_disk < 0)
4208                                 seq_printf(seq, "(S)"); /* spare */
4209                         size += rdev->size;
4210                 }
4211
4212                 if (!list_empty(&mddev->disks)) {
4213                         if (mddev->pers)
4214                                 seq_printf(seq, "\n      %llu blocks",
4215                                         (unsigned long long)mddev->array_size);
4216                         else
4217                                 seq_printf(seq, "\n      %llu blocks",
4218                                         (unsigned long long)size);
4219                 }
4220                 if (mddev->persistent) {
4221                         if (mddev->major_version != 0 ||
4222                             mddev->minor_version != 90) {
4223                                 seq_printf(seq," super %d.%d",
4224                                            mddev->major_version,
4225                                            mddev->minor_version);
4226                         }
4227                 } else
4228                         seq_printf(seq, " super non-persistent");
4229
4230                 if (mddev->pers) {
4231                         mddev->pers->status (seq, mddev);
4232                         seq_printf(seq, "\n      ");
4233                         if (mddev->pers->sync_request) {
4234                                 if (mddev->curr_resync > 2) {
4235                                         status_resync (seq, mddev);
4236                                         seq_printf(seq, "\n      ");
4237                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4238                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4239                                 else if (mddev->recovery_cp < MaxSector)
4240                                         seq_printf(seq, "\tresync=PENDING\n      ");
4241                         }
4242                 } else
4243                         seq_printf(seq, "\n       ");
4244
4245                 if ((bitmap = mddev->bitmap)) {
4246                         unsigned long chunk_kb;
4247                         unsigned long flags;
4248                         spin_lock_irqsave(&bitmap->lock, flags);
4249                         chunk_kb = bitmap->chunksize >> 10;
4250                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4251                                 "%lu%s chunk",
4252                                 bitmap->pages - bitmap->missing_pages,
4253                                 bitmap->pages,
4254                                 (bitmap->pages - bitmap->missing_pages)
4255                                         << (PAGE_SHIFT - 10),
4256                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4257                                 chunk_kb ? "KB" : "B");
4258                         if (bitmap->file) {
4259                                 seq_printf(seq, ", file: ");
4260                                 seq_path(seq, bitmap->file->f_vfsmnt,
4261                                          bitmap->file->f_dentry," \t\n");
4262                         }
4263
4264                         seq_printf(seq, "\n");
4265                         spin_unlock_irqrestore(&bitmap->lock, flags);
4266                 }
4267
4268                 seq_printf(seq, "\n");
4269         }
4270         mddev_unlock(mddev);
4271
4272         return 0;
4273 }
4274
4275 static struct seq_operations md_seq_ops = {
4276         .start  = md_seq_start,
4277         .next   = md_seq_next,
4278         .stop   = md_seq_stop,
4279         .show   = md_seq_show,
4280 };
4281
4282 static int md_seq_open(struct inode *inode, struct file *file)
4283 {
4284         int error;
4285         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4286         if (mi == NULL)
4287                 return -ENOMEM;
4288
4289         error = seq_open(file, &md_seq_ops);
4290         if (error)
4291                 kfree(mi);
4292         else {
4293                 struct seq_file *p = file->private_data;
4294                 p->private = mi;
4295                 mi->event = atomic_read(&md_event_count);
4296         }
4297         return error;
4298 }
4299
4300 static int md_seq_release(struct inode *inode, struct file *file)
4301 {
4302         struct seq_file *m = file->private_data;
4303         struct mdstat_info *mi = m->private;
4304         m->private = NULL;
4305         kfree(mi);
4306         return seq_release(inode, file);
4307 }
4308
4309 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4310 {
4311         struct seq_file *m = filp->private_data;
4312         struct mdstat_info *mi = m->private;
4313         int mask;
4314
4315         poll_wait(filp, &md_event_waiters, wait);
4316
4317         /* always allow read */
4318         mask = POLLIN | POLLRDNORM;
4319
4320         if (mi->event != atomic_read(&md_event_count))
4321                 mask |= POLLERR | POLLPRI;
4322         return mask;
4323 }
4324
4325 static struct file_operations md_seq_fops = {
4326         .open           = md_seq_open,
4327         .read           = seq_read,
4328         .llseek         = seq_lseek,
4329         .release        = md_seq_release,
4330         .poll           = mdstat_poll,
4331 };
4332
4333 int register_md_personality(struct mdk_personality *p)
4334 {
4335         spin_lock(&pers_lock);
4336         list_add_tail(&p->list, &pers_list);
4337         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4338         spin_unlock(&pers_lock);
4339         return 0;
4340 }
4341
4342 int unregister_md_personality(struct mdk_personality *p)
4343 {
4344         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4345         spin_lock(&pers_lock);
4346         list_del_init(&p->list);
4347         spin_unlock(&pers_lock);
4348         return 0;
4349 }
4350
4351 static int is_mddev_idle(mddev_t *mddev)
4352 {
4353         mdk_rdev_t * rdev;
4354         struct list_head *tmp;
4355         int idle;
4356         unsigned long curr_events;
4357
4358         idle = 1;
4359         ITERATE_RDEV(mddev,rdev,tmp) {
4360                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4361                 curr_events = disk_stat_read(disk, sectors[0]) +
4362                                 disk_stat_read(disk, sectors[1]) -
4363                                 atomic_read(&disk->sync_io);
4364                 /* The difference between curr_events and last_events
4365                  * will be affected by any new non-sync IO (making
4366                  * curr_events bigger) and any difference in the amount of
4367                  * in-flight syncio (making current_events bigger or smaller)
4368                  * The amount in-flight is currently limited to
4369                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4370                  * which is at most 4096 sectors.
4371                  * These numbers are fairly fragile and should be made
4372                  * more robust, probably by enforcing the
4373                  * 'window size' that md_do_sync sort-of uses.
4374                  *
4375                  * Note: the following is an unsigned comparison.
4376                  */
4377                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4378                         rdev->last_events = curr_events;
4379                         idle = 0;
4380                 }
4381         }
4382         return idle;
4383 }
4384
4385 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4386 {
4387         /* another "blocks" (512byte) blocks have been synced */
4388         atomic_sub(blocks, &mddev->recovery_active);
4389         wake_up(&mddev->recovery_wait);
4390         if (!ok) {
4391                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4392                 md_wakeup_thread(mddev->thread);
4393                 // stop recovery, signal do_sync ....
4394         }
4395 }
4396
4397
4398 /* md_write_start(mddev, bi)
4399  * If we need to update some array metadata (e.g. 'active' flag
4400  * in superblock) before writing, schedule a superblock update
4401  * and wait for it to complete.
4402  */
4403 void md_write_start(mddev_t *mddev, struct bio *bi)
4404 {
4405         if (bio_data_dir(bi) != WRITE)
4406                 return;
4407
4408         BUG_ON(mddev->ro == 1);
4409         if (mddev->ro == 2) {
4410                 /* need to switch to read/write */
4411                 mddev->ro = 0;
4412                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4413                 md_wakeup_thread(mddev->thread);
4414         }
4415         atomic_inc(&mddev->writes_pending);
4416         if (mddev->in_sync) {
4417                 spin_lock_irq(&mddev->write_lock);
4418                 if (mddev->in_sync) {
4419                         mddev->in_sync = 0;
4420                         mddev->sb_dirty = 1;
4421                         md_wakeup_thread(mddev->thread);
4422                 }
4423                 spin_unlock_irq(&mddev->write_lock);
4424         }
4425         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4426 }
4427
4428 void md_write_end(mddev_t *mddev)
4429 {
4430         if (atomic_dec_and_test(&mddev->writes_pending)) {
4431                 if (mddev->safemode == 2)
4432                         md_wakeup_thread(mddev->thread);
4433                 else
4434                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4435         }
4436 }
4437
4438 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4439
4440 #define SYNC_MARKS      10
4441 #define SYNC_MARK_STEP  (3*HZ)
4442 static void md_do_sync(mddev_t *mddev)
4443 {
4444         mddev_t *mddev2;
4445         unsigned int currspeed = 0,
4446                  window;
4447         sector_t max_sectors,j, io_sectors;
4448         unsigned long mark[SYNC_MARKS];
4449         sector_t mark_cnt[SYNC_MARKS];
4450         int last_mark,m;
4451         struct list_head *tmp;
4452         sector_t last_check;
4453         int skipped = 0;
4454
4455         /* just incase thread restarts... */
4456         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4457                 return;
4458
4459         /* we overload curr_resync somewhat here.
4460          * 0 == not engaged in resync at all
4461          * 2 == checking that there is no conflict with another sync
4462          * 1 == like 2, but have yielded to allow conflicting resync to
4463          *              commense
4464          * other == active in resync - this many blocks
4465          *
4466          * Before starting a resync we must have set curr_resync to
4467          * 2, and then checked that every "conflicting" array has curr_resync
4468          * less than ours.  When we find one that is the same or higher
4469          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4470          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4471          * This will mean we have to start checking from the beginning again.
4472          *
4473          */
4474
4475         do {
4476                 mddev->curr_resync = 2;
4477
4478         try_again:
4479                 if (kthread_should_stop()) {
4480                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4481                         goto skip;
4482                 }
4483                 ITERATE_MDDEV(mddev2,tmp) {
4484                         if (mddev2 == mddev)
4485                                 continue;
4486                         if (mddev2->curr_resync &&
4487                             match_mddev_units(mddev,mddev2)) {
4488                                 DEFINE_WAIT(wq);
4489                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4490                                         /* arbitrarily yield */
4491                                         mddev->curr_resync = 1;
4492                                         wake_up(&resync_wait);
4493                                 }
4494                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4495                                         /* no need to wait here, we can wait the next
4496                                          * time 'round when curr_resync == 2
4497                                          */
4498                                         continue;
4499                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4500                                 if (!kthread_should_stop() &&
4501                                     mddev2->curr_resync >= mddev->curr_resync) {
4502                                         printk(KERN_INFO "md: delaying resync of %s"
4503                                                " until %s has finished resync (they"
4504                                                " share one or more physical units)\n",
4505                                                mdname(mddev), mdname(mddev2));
4506                                         mddev_put(mddev2);
4507                                         schedule();
4508                                         finish_wait(&resync_wait, &wq);
4509                                         goto try_again;
4510                                 }
4511                                 finish_wait(&resync_wait, &wq);
4512                         }
4513                 }
4514         } while (mddev->curr_resync < 2);
4515
4516         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4517                 /* resync follows the size requested by the personality,
4518                  * which defaults to physical size, but can be virtual size
4519                  */
4520                 max_sectors = mddev->resync_max_sectors;
4521                 mddev->resync_mismatches = 0;
4522         } else
4523                 /* recovery follows the physical size of devices */
4524                 max_sectors = mddev->size << 1;
4525
4526         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4527         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4528                 " %d KB/sec/disc.\n", speed_min(mddev));
4529         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4530                "(but not more than %d KB/sec) for reconstruction.\n",
4531                speed_max(mddev));
4532
4533         is_mddev_idle(mddev); /* this also initializes IO event counters */
4534         /* we don't use the checkpoint if there's a bitmap */
4535         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4536             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4537                 j = mddev->recovery_cp;
4538         else
4539                 j = 0;
4540         io_sectors = 0;
4541         for (m = 0; m < SYNC_MARKS; m++) {
4542                 mark[m] = jiffies;
4543                 mark_cnt[m] = io_sectors;
4544         }
4545         last_mark = 0;
4546         mddev->resync_mark = mark[last_mark];
4547         mddev->resync_mark_cnt = mark_cnt[last_mark];
4548
4549         /*
4550          * Tune reconstruction:
4551          */
4552         window = 32*(PAGE_SIZE/512);
4553         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4554                 window/2,(unsigned long long) max_sectors/2);
4555
4556         atomic_set(&mddev->recovery_active, 0);
4557         init_waitqueue_head(&mddev->recovery_wait);
4558         last_check = 0;
4559
4560         if (j>2) {
4561                 printk(KERN_INFO
4562                         "md: resuming recovery of %s from checkpoint.\n",
4563                         mdname(mddev));
4564                 mddev->curr_resync = j;
4565         }
4566
4567         while (j < max_sectors) {
4568                 sector_t sectors;
4569
4570                 skipped = 0;
4571                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4572                                             currspeed < speed_min(mddev));
4573                 if (sectors == 0) {
4574                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4575                         goto out;
4576                 }
4577
4578                 if (!skipped) { /* actual IO requested */
4579                         io_sectors += sectors;
4580                         atomic_add(sectors, &mddev->recovery_active);
4581                 }
4582
4583                 j += sectors;
4584                 if (j>1) mddev->curr_resync = j;
4585                 if (last_check == 0)
4586                         /* this is the earliers that rebuilt will be
4587                          * visible in /proc/mdstat
4588                          */
4589                         md_new_event(mddev);
4590
4591                 if (last_check + window > io_sectors || j == max_sectors)
4592                         continue;
4593
4594                 last_check = io_sectors;
4595
4596                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4597                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4598                         break;
4599
4600         repeat:
4601                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4602                         /* step marks */
4603                         int next = (last_mark+1) % SYNC_MARKS;
4604
4605                         mddev->resync_mark = mark[next];
4606                         mddev->resync_mark_cnt = mark_cnt[next];
4607                         mark[next] = jiffies;
4608                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4609                         last_mark = next;
4610                 }
4611
4612
4613                 if (kthread_should_stop()) {
4614                         /*
4615                          * got a signal, exit.
4616                          */
4617                         printk(KERN_INFO
4618                                 "md: md_do_sync() got signal ... exiting\n");
4619                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4620                         goto out;
4621                 }
4622
4623                 /*
4624                  * this loop exits only if either when we are slower than
4625                  * the 'hard' speed limit, or the system was IO-idle for
4626                  * a jiffy.
4627                  * the system might be non-idle CPU-wise, but we only care
4628                  * about not overloading the IO subsystem. (things like an
4629                  * e2fsck being done on the RAID array should execute fast)
4630                  */
4631                 mddev->queue->unplug_fn(mddev->queue);
4632                 cond_resched();
4633
4634                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4635                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4636
4637                 if (currspeed > speed_min(mddev)) {
4638                         if ((currspeed > speed_max(mddev)) ||
4639                                         !is_mddev_idle(mddev)) {
4640                                 msleep(500);
4641                                 goto repeat;
4642                         }
4643                 }
4644         }
4645         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4646         /*
4647          * this also signals 'finished resyncing' to md_stop
4648          */
4649  out:
4650         mddev->queue->unplug_fn(mddev->queue);
4651
4652         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4653
4654         /* tell personality that we are finished */
4655         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4656
4657         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4658             mddev->curr_resync > 2 &&
4659             mddev->curr_resync >= mddev->recovery_cp) {
4660                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4661                         printk(KERN_INFO
4662                                 "md: checkpointing recovery of %s.\n",
4663                                 mdname(mddev));
4664                         mddev->recovery_cp = mddev->curr_resync;
4665                 } else
4666                         mddev->recovery_cp = MaxSector;
4667         }
4668
4669  skip:
4670         mddev->curr_resync = 0;
4671         wake_up(&resync_wait);
4672         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4673         md_wakeup_thread(mddev->thread);
4674 }
4675
4676
4677 /*
4678  * This routine is regularly called by all per-raid-array threads to
4679  * deal with generic issues like resync and super-block update.
4680  * Raid personalities that don't have a thread (linear/raid0) do not
4681  * need this as they never do any recovery or update the superblock.
4682  *
4683  * It does not do any resync itself, but rather "forks" off other threads
4684  * to do that as needed.
4685  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4686  * "->recovery" and create a thread at ->sync_thread.
4687  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4688  * and wakeups up this thread which will reap the thread and finish up.
4689  * This thread also removes any faulty devices (with nr_pending == 0).
4690  *
4691  * The overall approach is:
4692  *  1/ if the superblock needs updating, update it.
4693  *  2/ If a recovery thread is running, don't do anything else.
4694  *  3/ If recovery has finished, clean up, possibly marking spares active.
4695  *  4/ If there are any faulty devices, remove them.
4696  *  5/ If array is degraded, try to add spares devices
4697  *  6/ If array has spares or is not in-sync, start a resync thread.
4698  */
4699 void md_check_recovery(mddev_t *mddev)
4700 {
4701         mdk_rdev_t *rdev;
4702         struct list_head *rtmp;
4703
4704
4705         if (mddev->bitmap)
4706                 bitmap_daemon_work(mddev->bitmap);
4707
4708         if (mddev->ro)
4709                 return;
4710
4711         if (signal_pending(current)) {
4712                 if (mddev->pers->sync_request) {
4713                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4714                                mdname(mddev));
4715                         mddev->safemode = 2;
4716                 }
4717                 flush_signals(current);
4718         }
4719
4720         if ( ! (
4721                 mddev->sb_dirty ||
4722                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4723                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4724                 (mddev->safemode == 1) ||
4725                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4726                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4727                 ))
4728                 return;
4729
4730         if (mddev_trylock(mddev)==0) {
4731                 int spares =0;
4732
4733                 spin_lock_irq(&mddev->write_lock);
4734                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4735                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4736                         mddev->in_sync = 1;
4737                         mddev->sb_dirty = 1;
4738                 }
4739                 if (mddev->safemode == 1)
4740                         mddev->safemode = 0;
4741                 spin_unlock_irq(&mddev->write_lock);
4742
4743                 if (mddev->sb_dirty)
4744                         md_update_sb(mddev);
4745
4746
4747                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4748                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4749                         /* resync/recovery still happening */
4750                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4751                         goto unlock;
4752                 }
4753                 if (mddev->sync_thread) {
4754                         /* resync has finished, collect result */
4755                         md_unregister_thread(mddev->sync_thread);
4756                         mddev->sync_thread = NULL;
4757                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4758                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4759                                 /* success...*/
4760                                 /* activate any spares */
4761                                 mddev->pers->spare_active(mddev);
4762                         }
4763                         md_update_sb(mddev);
4764
4765                         /* if array is no-longer degraded, then any saved_raid_disk
4766                          * information must be scrapped
4767                          */
4768                         if (!mddev->degraded)
4769                                 ITERATE_RDEV(mddev,rdev,rtmp)
4770                                         rdev->saved_raid_disk = -1;
4771
4772                         mddev->recovery = 0;
4773                         /* flag recovery needed just to double check */
4774                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4775                         md_new_event(mddev);
4776                         goto unlock;
4777                 }
4778                 /* Clear some bits that don't mean anything, but
4779                  * might be left set
4780                  */
4781                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4782                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4783                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4784                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4785
4786                 /* no recovery is running.
4787                  * remove any failed drives, then
4788                  * add spares if possible.
4789                  * Spare are also removed and re-added, to allow
4790                  * the personality to fail the re-add.
4791                  */
4792                 ITERATE_RDEV(mddev,rdev,rtmp)
4793                         if (rdev->raid_disk >= 0 &&
4794                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4795                             atomic_read(&rdev->nr_pending)==0) {
4796                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4797                                         char nm[20];
4798                                         sprintf(nm,"rd%d", rdev->raid_disk);
4799                                         sysfs_remove_link(&mddev->kobj, nm);
4800                                         rdev->raid_disk = -1;
4801                                 }
4802                         }
4803
4804                 if (mddev->degraded) {
4805                         ITERATE_RDEV(mddev,rdev,rtmp)
4806                                 if (rdev->raid_disk < 0
4807                                     && !test_bit(Faulty, &rdev->flags)) {
4808                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4809                                                 char nm[20];
4810                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4811                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4812                                                 spares++;
4813                                                 md_new_event(mddev);
4814                                         } else
4815                                                 break;
4816                                 }
4817                 }
4818
4819                 if (spares) {
4820                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4821                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4822                 } else if (mddev->recovery_cp < MaxSector) {
4823                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4824                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4825                         /* nothing to be done ... */
4826                         goto unlock;
4827
4828                 if (mddev->pers->sync_request) {
4829                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4830                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4831                                 /* We are adding a device or devices to an array
4832                                  * which has the bitmap stored on all devices.
4833                                  * So make sure all bitmap pages get written
4834                                  */
4835                                 bitmap_write_all(mddev->bitmap);
4836                         }
4837                         mddev->sync_thread = md_register_thread(md_do_sync,
4838                                                                 mddev,
4839                                                                 "%s_resync");
4840                         if (!mddev->sync_thread) {
4841                                 printk(KERN_ERR "%s: could not start resync"
4842                                         " thread...\n",
4843                                         mdname(mddev));
4844                                 /* leave the spares where they are, it shouldn't hurt */
4845                                 mddev->recovery = 0;
4846                         } else
4847                                 md_wakeup_thread(mddev->sync_thread);
4848                         md_new_event(mddev);
4849                 }
4850         unlock:
4851                 mddev_unlock(mddev);
4852         }
4853 }
4854
4855 static int md_notify_reboot(struct notifier_block *this,
4856                             unsigned long code, void *x)
4857 {
4858         struct list_head *tmp;
4859         mddev_t *mddev;
4860
4861         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4862
4863                 printk(KERN_INFO "md: stopping all md devices.\n");
4864
4865                 ITERATE_MDDEV(mddev,tmp)
4866                         if (mddev_trylock(mddev)==0)
4867                                 do_md_stop (mddev, 1);
4868                 /*
4869                  * certain more exotic SCSI devices are known to be
4870                  * volatile wrt too early system reboots. While the
4871                  * right place to handle this issue is the given
4872                  * driver, we do want to have a safe RAID driver ...
4873                  */
4874                 mdelay(1000*1);
4875         }
4876         return NOTIFY_DONE;
4877 }
4878
4879 static struct notifier_block md_notifier = {
4880         .notifier_call  = md_notify_reboot,
4881         .next           = NULL,
4882         .priority       = INT_MAX, /* before any real devices */
4883 };
4884
4885 static void md_geninit(void)
4886 {
4887         struct proc_dir_entry *p;
4888
4889         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4890
4891         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4892         if (p)
4893                 p->proc_fops = &md_seq_fops;
4894 }
4895
4896 static int __init md_init(void)
4897 {
4898         int minor;
4899
4900         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4901                         " MD_SB_DISKS=%d\n",
4902                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4903                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4904         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4905                         BITMAP_MINOR);
4906
4907         if (register_blkdev(MAJOR_NR, "md"))
4908                 return -1;
4909         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4910                 unregister_blkdev(MAJOR_NR, "md");
4911                 return -1;
4912         }
4913         devfs_mk_dir("md");
4914         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4915                                 md_probe, NULL, NULL);
4916         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4917                             md_probe, NULL, NULL);
4918
4919         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4920                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4921                                 S_IFBLK|S_IRUSR|S_IWUSR,
4922                                 "md/%d", minor);
4923
4924         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4925                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4926                               S_IFBLK|S_IRUSR|S_IWUSR,
4927                               "md/mdp%d", minor);
4928
4929
4930         register_reboot_notifier(&md_notifier);
4931         raid_table_header = register_sysctl_table(raid_root_table, 1);
4932
4933         md_geninit();
4934         return (0);
4935 }
4936
4937
4938 #ifndef MODULE
4939
4940 /*
4941  * Searches all registered partitions for autorun RAID arrays
4942  * at boot time.
4943  */
4944 static dev_t detected_devices[128];
4945 static int dev_cnt;
4946
4947 void md_autodetect_dev(dev_t dev)
4948 {
4949         if (dev_cnt >= 0 && dev_cnt < 127)
4950                 detected_devices[dev_cnt++] = dev;
4951 }
4952
4953
4954 static void autostart_arrays(int part)
4955 {
4956         mdk_rdev_t *rdev;
4957         int i;
4958
4959         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4960
4961         for (i = 0; i < dev_cnt; i++) {
4962                 dev_t dev = detected_devices[i];
4963
4964                 rdev = md_import_device(dev,0, 0);
4965                 if (IS_ERR(rdev))
4966                         continue;
4967
4968                 if (test_bit(Faulty, &rdev->flags)) {
4969                         MD_BUG();
4970                         continue;
4971                 }
4972                 list_add(&rdev->same_set, &pending_raid_disks);
4973         }
4974         dev_cnt = 0;
4975
4976         autorun_devices(part);
4977 }
4978
4979 #endif
4980
4981 static __exit void md_exit(void)
4982 {
4983         mddev_t *mddev;
4984         struct list_head *tmp;
4985         int i;
4986         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4987         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4988         for (i=0; i < MAX_MD_DEVS; i++)
4989                 devfs_remove("md/%d", i);
4990         for (i=0; i < MAX_MD_DEVS; i++)
4991                 devfs_remove("md/d%d", i);
4992
4993         devfs_remove("md");
4994
4995         unregister_blkdev(MAJOR_NR,"md");
4996         unregister_blkdev(mdp_major, "mdp");
4997         unregister_reboot_notifier(&md_notifier);
4998         unregister_sysctl_table(raid_table_header);
4999         remove_proc_entry("mdstat", NULL);
5000         ITERATE_MDDEV(mddev,tmp) {
5001                 struct gendisk *disk = mddev->gendisk;
5002                 if (!disk)
5003                         continue;
5004                 export_array(mddev);
5005                 del_gendisk(disk);
5006                 put_disk(disk);
5007                 mddev->gendisk = NULL;
5008                 mddev_put(mddev);
5009         }
5010 }
5011
5012 module_init(md_init)
5013 module_exit(md_exit)
5014
5015 static int get_ro(char *buffer, struct kernel_param *kp)
5016 {
5017         return sprintf(buffer, "%d", start_readonly);
5018 }
5019 static int set_ro(const char *val, struct kernel_param *kp)
5020 {
5021         char *e;
5022         int num = simple_strtoul(val, &e, 10);
5023         if (*val && (*e == '\0' || *e == '\n')) {
5024                 start_readonly = num;
5025                 return 0;
5026         }
5027         return -EINVAL;
5028 }
5029
5030 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
5031 module_param(start_dirty_degraded, int, 0644);
5032
5033
5034 EXPORT_SYMBOL(register_md_personality);
5035 EXPORT_SYMBOL(unregister_md_personality);
5036 EXPORT_SYMBOL(md_error);
5037 EXPORT_SYMBOL(md_done_sync);
5038 EXPORT_SYMBOL(md_write_start);
5039 EXPORT_SYMBOL(md_write_end);
5040 EXPORT_SYMBOL(md_register_thread);
5041 EXPORT_SYMBOL(md_unregister_thread);
5042 EXPORT_SYMBOL(md_wakeup_thread);
5043 EXPORT_SYMBOL(md_print_devices);
5044 EXPORT_SYMBOL(md_check_recovery);
5045 MODULE_LICENSE("GPL");
5046 MODULE_ALIAS("md");
5047 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);