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 != 4
 765                             && mddev->level != 5 && mddev->level != 6
 766                             && mddev->level != 10) {
 767                                 /* FIXME use a better test */
 768                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
 769                                 return -EINVAL;
 770                         }
 771                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 772                 }
 773
 774         } else if (mddev->pers == NULL) {
 775                 /* Insist on good event counter while assembling */
 776                 __u64 ev1 = md_event(sb);
 777                 ++ev1;
 778                 if (ev1 < mddev->events)
 779                         return -EINVAL;
 780         } else if (mddev->bitmap) {
 781                 /* if adding to array with a bitmap, then we can accept an
 782                  * older device ... but not too old.
 783                  */
 784                 __u64 ev1 = md_event(sb);
 785                 if (ev1 < mddev->bitmap->events_cleared)
 786                         return 0;
 787         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 788                 return 0;
 789
 790         if (mddev->level != LEVEL_MULTIPATH) {
 791                 desc = sb->disks + rdev->desc_nr;
 792
 793                 if (desc->state & (1<<MD_DISK_FAULTY))
 794                         set_bit(Faulty, &rdev->flags);
 795                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 796                          desc->raid_disk < mddev->raid_disks) {
 797                         set_bit(In_sync, &rdev->flags);
 798                         rdev->raid_disk = desc->raid_disk;
 799                 }
 800                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 801                         set_bit(WriteMostly, &rdev->flags);
 802         } else /* MULTIPATH are always insync */
 803                 set_bit(In_sync, &rdev->flags);
 804         return 0;
 805 }
 806
 807 /*
 808  * sync_super for 0.90.0
 809  */
 810 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 811 {
 812         mdp_super_t *sb;
 813         struct list_head *tmp;
 814         mdk_rdev_t *rdev2;
 815         int next_spare = mddev->raid_disks;
 816
 817
 818         /* make rdev->sb match mddev data..
 819          *
 820          * 1/ zero out disks
 821          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 822          * 3/ any empty disks < next_spare become removed
 823          *
 824          * disks[0] gets initialised to REMOVED because
 825          * we cannot be sure from other fields if it has
 826          * been initialised or not.
 827          */
 828         int i;
 829         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 830
 831         rdev->sb_size = MD_SB_BYTES;
 832
 833         sb = (mdp_super_t*)page_address(rdev->sb_page);
 834
 835         memset(sb, 0, sizeof(*sb));
 836
 837         sb->md_magic = MD_SB_MAGIC;
 838         sb->major_version = mddev->major_version;
 839         sb->minor_version = mddev->minor_version;
 840         sb->patch_version = mddev->patch_version;
 841         sb->gvalid_words  = 0; /* ignored */
 842         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 843         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 844         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 845         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 846
 847         sb->ctime = mddev->ctime;
 848         sb->level = mddev->level;
 849         sb->size  = mddev->size;
 850         sb->raid_disks = mddev->raid_disks;
 851         sb->md_minor = mddev->md_minor;
 852         sb->not_persistent = !mddev->persistent;
 853         sb->utime = mddev->utime;
 854         sb->state = 0;
 855         sb->events_hi = (mddev->events>>32);
 856         sb->events_lo = (u32)mddev->events;
 857
 858         if (mddev->in_sync)
 859         {
 860                 sb->recovery_cp = mddev->recovery_cp;
 861                 sb->cp_events_hi = (mddev->events>>32);
 862                 sb->cp_events_lo = (u32)mddev->events;
 863                 if (mddev->recovery_cp == MaxSector)
 864                         sb->state = (1<< MD_SB_CLEAN);
 865         } else
 866                 sb->recovery_cp = 0;
 867
 868         sb->layout = mddev->layout;
 869         sb->chunk_size = mddev->chunk_size;
 870
 871         if (mddev->bitmap && mddev->bitmap_file == NULL)
 872                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 873
 874         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 875         ITERATE_RDEV(mddev,rdev2,tmp) {
 876                 mdp_disk_t *d;
 877                 int desc_nr;
 878                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 879                     && !test_bit(Faulty, &rdev2->flags))
 880                         desc_nr = rdev2->raid_disk;
 881                 else
 882                         desc_nr = next_spare++;
 883                 rdev2->desc_nr = desc_nr;
 884                 d = &sb->disks[rdev2->desc_nr];
 885                 nr_disks++;
 886                 d->number = rdev2->desc_nr;
 887                 d->major = MAJOR(rdev2->bdev->bd_dev);
 888                 d->minor = MINOR(rdev2->bdev->bd_dev);
 889                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 890                     && !test_bit(Faulty, &rdev2->flags))
 891                         d->raid_disk = rdev2->raid_disk;
 892                 else
 893                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 894                 if (test_bit(Faulty, &rdev2->flags)) {
 895                         d->state = (1<<MD_DISK_FAULTY);
 896                         failed++;
 897                 } else if (test_bit(In_sync, &rdev2->flags)) {
 898                         d->state = (1<<MD_DISK_ACTIVE);
 899                         d->state |= (1<<MD_DISK_SYNC);
 900                         active++;
 901                         working++;
 902                 } else {
 903                         d->state = 0;
 904                         spare++;
 905                         working++;
 906                 }
 907                 if (test_bit(WriteMostly, &rdev2->flags))
 908                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 909         }
 910         /* now set the "removed" and "faulty" bits on any missing devices */
 911         for (i=0 ; i < mddev->raid_disks ; i++) {
 912                 mdp_disk_t *d = &sb->disks[i];
 913                 if (d->state == 0 && d->number == 0) {
 914                         d->number = i;
 915                         d->raid_disk = i;
 916                         d->state = (1<<MD_DISK_REMOVED);
 917                         d->state |= (1<<MD_DISK_FAULTY);
 918                         failed++;
 919                 }
 920         }
 921         sb->nr_disks = nr_disks;
 922         sb->active_disks = active;
 923         sb->working_disks = working;
 924         sb->failed_disks = failed;
 925         sb->spare_disks = spare;
 926
 927         sb->this_disk = sb->disks[rdev->desc_nr];
 928         sb->sb_csum = calc_sb_csum(sb);
 929 }
 930
 931 /*
 932  * version 1 superblock
 933  */
 934
 935 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 936 {
 937         unsigned int disk_csum, csum;
 938         unsigned long long newcsum;
 939         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 940         unsigned int *isuper = (unsigned int*)sb;
 941         int i;
 942
 943         disk_csum = sb->sb_csum;
 944         sb->sb_csum = 0;
 945         newcsum = 0;
 946         for (i=0; size>=4; size -= 4 )
 947                 newcsum += le32_to_cpu(*isuper++);
 948
 949         if (size == 2)
 950                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 951
 952         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 953         sb->sb_csum = disk_csum;
 954         return cpu_to_le32(csum);
 955 }
 956
 957 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 958 {
 959         struct mdp_superblock_1 *sb;
 960         int ret;
 961         sector_t sb_offset;
 962         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 963         int bmask;
 964
 965         /*
 966          * Calculate the position of the superblock.
 967          * It is always aligned to a 4K boundary and
 968          * depeding on minor_version, it can be:
 969          * 0: At least 8K, but less than 12K, from end of device
 970          * 1: At start of device
 971          * 2: 4K from start of device.
 972          */
 973         switch(minor_version) {
 974         case 0:
 975                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 976                 sb_offset -= 8*2;
 977                 sb_offset &= ~(sector_t)(4*2-1);
 978                 /* convert from sectors to K */
 979                 sb_offset /= 2;
 980                 break;
 981         case 1:
 982                 sb_offset = 0;
 983                 break;
 984         case 2:
 985                 sb_offset = 4;
 986                 break;
 987         default:
 988                 return -EINVAL;
 989         }
 990         rdev->sb_offset = sb_offset;
 991
 992         /* superblock is rarely larger than 1K, but it can be larger,
 993          * and it is safe to read 4k, so we do that
 994          */
 995         ret = read_disk_sb(rdev, 4096);
 996         if (ret) return ret;
 997
 998
 999         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1000
1001         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1002             sb->major_version != cpu_to_le32(1) ||
1003             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1004             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1005             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1006                 return -EINVAL;
1007
1008         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1009                 printk("md: invalid superblock checksum on %s\n",
1010                         bdevname(rdev->bdev,b));
1011                 return -EINVAL;
1012         }
1013         if (le64_to_cpu(sb->data_size) < 10) {
1014                 printk("md: data_size too small on %s\n",
1015                        bdevname(rdev->bdev,b));
1016                 return -EINVAL;
1017         }
1018         rdev->preferred_minor = 0xffff;
1019         rdev->data_offset = le64_to_cpu(sb->data_offset);
1020         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1021
1022         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1023         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1024         if (rdev->sb_size & bmask)
1025                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1026
1027         if (refdev == 0)
1028                 ret = 1;
1029         else {
1030                 __u64 ev1, ev2;
1031                 struct mdp_superblock_1 *refsb =
1032                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1033
1034                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1035                     sb->level != refsb->level ||
1036                     sb->layout != refsb->layout ||
1037                     sb->chunksize != refsb->chunksize) {
1038                         printk(KERN_WARNING "md: %s has strangely different"
1039                                 " superblock to %s\n",
1040                                 bdevname(rdev->bdev,b),
1041                                 bdevname(refdev->bdev,b2));
1042                         return -EINVAL;
1043                 }
1044                 ev1 = le64_to_cpu(sb->events);
1045                 ev2 = le64_to_cpu(refsb->events);
1046
1047                 if (ev1 > ev2)
1048                         ret = 1;
1049                 else
1050                         ret = 0;
1051         }
1052         if (minor_version)
1053                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1054         else
1055                 rdev->size = rdev->sb_offset;
1056         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1057                 return -EINVAL;
1058         rdev->size = le64_to_cpu(sb->data_size)/2;
1059         if (le32_to_cpu(sb->chunksize))
1060                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1061
1062         if (le32_to_cpu(sb->size) > rdev->size*2)
1063                 return -EINVAL;
1064         return ret;
1065 }
1066
1067 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1068 {
1069         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1070
1071         rdev->raid_disk = -1;
1072         rdev->flags = 0;
1073         if (mddev->raid_disks == 0) {
1074                 mddev->major_version = 1;
1075                 mddev->patch_version = 0;
1076                 mddev->persistent = 1;
1077                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1078                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1079                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1080                 mddev->level = le32_to_cpu(sb->level);
1081                 mddev->clevel[0] = 0;
1082                 mddev->layout = le32_to_cpu(sb->layout);
1083                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1084                 mddev->size = le64_to_cpu(sb->size)/2;
1085                 mddev->events = le64_to_cpu(sb->events);
1086                 mddev->bitmap_offset = 0;
1087                 mddev->default_bitmap_offset = 1024 >> 9;
1088
1089                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1090                 memcpy(mddev->uuid, sb->set_uuid, 16);
1091
1092                 mddev->max_disks =  (4096-256)/2;
1093
1094                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1095                     mddev->bitmap_file == NULL ) {
1096                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1097                             && mddev->level != 4
1098                             && mddev->level != 10) {
1099                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1100                                 return -EINVAL;
1101                         }
1102                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1103                 }
1104         } else if (mddev->pers == NULL) {
1105                 /* Insist of good event counter while assembling */
1106                 __u64 ev1 = le64_to_cpu(sb->events);
1107                 ++ev1;
1108                 if (ev1 < mddev->events)
1109                         return -EINVAL;
1110         } else if (mddev->bitmap) {
1111                 /* If adding to array with a bitmap, then we can accept an
1112                  * older device, but not too old.
1113                  */
1114                 __u64 ev1 = le64_to_cpu(sb->events);
1115                 if (ev1 < mddev->bitmap->events_cleared)
1116                         return 0;
1117         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1118                 return 0;
1119
1120         if (mddev->level != LEVEL_MULTIPATH) {
1121                 int role;
1122                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1123                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1124                 switch(role) {
1125                 case 0xffff: /* spare */
1126                         break;
1127                 case 0xfffe: /* faulty */
1128                         set_bit(Faulty, &rdev->flags);
1129                         break;
1130                 default:
1131                         set_bit(In_sync, &rdev->flags);
1132                         rdev->raid_disk = role;
1133                         break;
1134                 }
1135                 if (sb->devflags & WriteMostly1)
1136                         set_bit(WriteMostly, &rdev->flags);
1137         } else /* MULTIPATH are always insync */
1138                 set_bit(In_sync, &rdev->flags);
1139
1140         return 0;
1141 }
1142
1143 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1144 {
1145         struct mdp_superblock_1 *sb;
1146         struct list_head *tmp;
1147         mdk_rdev_t *rdev2;
1148         int max_dev, i;
1149         /* make rdev->sb match mddev and rdev data. */
1150
1151         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1152
1153         sb->feature_map = 0;
1154         sb->pad0 = 0;
1155         memset(sb->pad1, 0, sizeof(sb->pad1));
1156         memset(sb->pad2, 0, sizeof(sb->pad2));
1157         memset(sb->pad3, 0, sizeof(sb->pad3));
1158
1159         sb->utime = cpu_to_le64((__u64)mddev->utime);
1160         sb->events = cpu_to_le64(mddev->events);
1161         if (mddev->in_sync)
1162                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1163         else
1164                 sb->resync_offset = cpu_to_le64(0);
1165
1166         sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1167
1168         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1169         sb->size = cpu_to_le64(mddev->size<<1);
1170
1171         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1172                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1173                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1174         }
1175
1176         max_dev = 0;
1177         ITERATE_RDEV(mddev,rdev2,tmp)
1178                 if (rdev2->desc_nr+1 > max_dev)
1179                         max_dev = rdev2->desc_nr+1;
1180
1181         sb->max_dev = cpu_to_le32(max_dev);
1182         for (i=0; i<max_dev;i++)
1183                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1184
1185         ITERATE_RDEV(mddev,rdev2,tmp) {
1186                 i = rdev2->desc_nr;
1187                 if (test_bit(Faulty, &rdev2->flags))
1188                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1189                 else if (test_bit(In_sync, &rdev2->flags))
1190                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1191                 else
1192                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1193         }
1194
1195         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1196         sb->sb_csum = calc_sb_1_csum(sb);
1197 }
1198
1199
1200 static struct super_type super_types[] = {
1201         [0] = {
1202                 .name   = "0.90.0",
1203                 .owner  = THIS_MODULE,
1204                 .load_super     = super_90_load,
1205                 .validate_super = super_90_validate,
1206                 .sync_super     = super_90_sync,
1207         },
1208         [1] = {
1209                 .name   = "md-1",
1210                 .owner  = THIS_MODULE,
1211                 .load_super     = super_1_load,
1212                 .validate_super = super_1_validate,
1213                 .sync_super     = super_1_sync,
1214         },
1215 };
1216
1217 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1218 {
1219         struct list_head *tmp;
1220         mdk_rdev_t *rdev;
1221
1222         ITERATE_RDEV(mddev,rdev,tmp)
1223                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1224                         return rdev;
1225
1226         return NULL;
1227 }
1228
1229 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1230 {
1231         struct list_head *tmp;
1232         mdk_rdev_t *rdev;
1233
1234         ITERATE_RDEV(mddev1,rdev,tmp)
1235                 if (match_dev_unit(mddev2, rdev))
1236                         return 1;
1237
1238         return 0;
1239 }
1240
1241 static LIST_HEAD(pending_raid_disks);
1242
1243 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1244 {
1245         mdk_rdev_t *same_pdev;
1246         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1247         struct kobject *ko;
1248         char *s;
1249
1250         if (rdev->mddev) {
1251                 MD_BUG();
1252                 return -EINVAL;
1253         }
1254         /* make sure rdev->size exceeds mddev->size */
1255         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1256                 if (mddev->pers)
1257                         /* Cannot change size, so fail */
1258                         return -ENOSPC;
1259                 else
1260                         mddev->size = rdev->size;
1261         }
1262         same_pdev = match_dev_unit(mddev, rdev);
1263         if (same_pdev)
1264                 printk(KERN_WARNING
1265                         "%s: WARNING: %s appears to be on the same physical"
1266                         " disk as %s. True\n     protection against single-disk"
1267                         " failure might be compromised.\n",
1268                         mdname(mddev), bdevname(rdev->bdev,b),
1269                         bdevname(same_pdev->bdev,b2));
1270
1271         /* Verify rdev->desc_nr is unique.
1272          * If it is -1, assign a free number, else
1273          * check number is not in use
1274          */
1275         if (rdev->desc_nr < 0) {
1276                 int choice = 0;
1277                 if (mddev->pers) choice = mddev->raid_disks;
1278                 while (find_rdev_nr(mddev, choice))
1279                         choice++;
1280                 rdev->desc_nr = choice;
1281         } else {
1282                 if (find_rdev_nr(mddev, rdev->desc_nr))
1283                         return -EBUSY;
1284         }
1285         bdevname(rdev->bdev,b);
1286         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1287                 return -ENOMEM;
1288         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1289                 *s = '!';
1290
1291         list_add(&rdev->same_set, &mddev->disks);
1292         rdev->mddev = mddev;
1293         printk(KERN_INFO "md: bind<%s>\n", b);
1294
1295         rdev->kobj.parent = &mddev->kobj;
1296         kobject_add(&rdev->kobj);
1297
1298         if (rdev->bdev->bd_part)
1299                 ko = &rdev->bdev->bd_part->kobj;
1300         else
1301                 ko = &rdev->bdev->bd_disk->kobj;
1302         sysfs_create_link(&rdev->kobj, ko, "block");
1303         return 0;
1304 }
1305
1306 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1307 {
1308         char b[BDEVNAME_SIZE];
1309         if (!rdev->mddev) {
1310                 MD_BUG();
1311                 return;
1312         }
1313         list_del_init(&rdev->same_set);
1314         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1315         rdev->mddev = NULL;
1316         sysfs_remove_link(&rdev->kobj, "block");
1317         kobject_del(&rdev->kobj);
1318 }
1319
1320 /*
1321  * prevent the device from being mounted, repartitioned or
1322  * otherwise reused by a RAID array (or any other kernel
1323  * subsystem), by bd_claiming the device.
1324  */
1325 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1326 {
1327         int err = 0;
1328         struct block_device *bdev;
1329         char b[BDEVNAME_SIZE];
1330
1331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1332         if (IS_ERR(bdev)) {
1333                 printk(KERN_ERR "md: could not open %s.\n",
1334                         __bdevname(dev, b));
1335                 return PTR_ERR(bdev);
1336         }
1337         err = bd_claim(bdev, rdev);
1338         if (err) {
1339                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1340                         bdevname(bdev, b));
1341                 blkdev_put(bdev);
1342                 return err;
1343         }
1344         rdev->bdev = bdev;
1345         return err;
1346 }
1347
1348 static void unlock_rdev(mdk_rdev_t *rdev)
1349 {
1350         struct block_device *bdev = rdev->bdev;
1351         rdev->bdev = NULL;
1352         if (!bdev)
1353                 MD_BUG();
1354         bd_release(bdev);
1355         blkdev_put(bdev);
1356 }
1357
1358 void md_autodetect_dev(dev_t dev);
1359
1360 static void export_rdev(mdk_rdev_t * rdev)
1361 {
1362         char b[BDEVNAME_SIZE];
1363         printk(KERN_INFO "md: export_rdev(%s)\n",
1364                 bdevname(rdev->bdev,b));
1365         if (rdev->mddev)
1366                 MD_BUG();
1367         free_disk_sb(rdev);
1368         list_del_init(&rdev->same_set);
1369 #ifndef MODULE
1370         md_autodetect_dev(rdev->bdev->bd_dev);
1371 #endif
1372         unlock_rdev(rdev);
1373         kobject_put(&rdev->kobj);
1374 }
1375
1376 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1377 {
1378         unbind_rdev_from_array(rdev);
1379         export_rdev(rdev);
1380 }
1381
1382 static void export_array(mddev_t *mddev)
1383 {
1384         struct list_head *tmp;
1385         mdk_rdev_t *rdev;
1386
1387         ITERATE_RDEV(mddev,rdev,tmp) {
1388                 if (!rdev->mddev) {
1389                         MD_BUG();
1390                         continue;
1391                 }
1392                 kick_rdev_from_array(rdev);
1393         }
1394         if (!list_empty(&mddev->disks))
1395                 MD_BUG();
1396         mddev->raid_disks = 0;
1397         mddev->major_version = 0;
1398 }
1399
1400 static void print_desc(mdp_disk_t *desc)
1401 {
1402         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1403                 desc->major,desc->minor,desc->raid_disk,desc->state);
1404 }
1405
1406 static void print_sb(mdp_super_t *sb)
1407 {
1408         int i;
1409
1410         printk(KERN_INFO
1411                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1412                 sb->major_version, sb->minor_version, sb->patch_version,
1413                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1414                 sb->ctime);
1415         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1416                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1417                 sb->md_minor, sb->layout, sb->chunk_size);
1418         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1419                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1420                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1421                 sb->failed_disks, sb->spare_disks,
1422                 sb->sb_csum, (unsigned long)sb->events_lo);
1423
1424         printk(KERN_INFO);
1425         for (i = 0; i < MD_SB_DISKS; i++) {
1426                 mdp_disk_t *desc;
1427
1428                 desc = sb->disks + i;
1429                 if (desc->number || desc->major || desc->minor ||
1430                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1431                         printk("     D %2d: ", i);
1432                         print_desc(desc);
1433                 }
1434         }
1435         printk(KERN_INFO "md:     THIS: ");
1436         print_desc(&sb->this_disk);
1437
1438 }
1439
1440 static void print_rdev(mdk_rdev_t *rdev)
1441 {
1442         char b[BDEVNAME_SIZE];
1443         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1444                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1445                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1446                 rdev->desc_nr);
1447         if (rdev->sb_loaded) {
1448                 printk(KERN_INFO "md: rdev superblock:\n");
1449                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1450         } else
1451                 printk(KERN_INFO "md: no rdev superblock!\n");
1452 }
1453
1454 void md_print_devices(void)
1455 {
1456         struct list_head *tmp, *tmp2;
1457         mdk_rdev_t *rdev;
1458         mddev_t *mddev;
1459         char b[BDEVNAME_SIZE];
1460
1461         printk("\n");
1462         printk("md:     **********************************\n");
1463         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1464         printk("md:     **********************************\n");
1465         ITERATE_MDDEV(mddev,tmp) {
1466
1467                 if (mddev->bitmap)
1468                         bitmap_print_sb(mddev->bitmap);
1469                 else
1470                         printk("%s: ", mdname(mddev));
1471                 ITERATE_RDEV(mddev,rdev,tmp2)
1472                         printk("<%s>", bdevname(rdev->bdev,b));
1473                 printk("\n");
1474
1475                 ITERATE_RDEV(mddev,rdev,tmp2)
1476                         print_rdev(rdev);
1477         }
1478         printk("md:     **********************************\n");
1479         printk("\n");
1480 }
1481
1482
1483 static void sync_sbs(mddev_t * mddev)
1484 {
1485         mdk_rdev_t *rdev;
1486         struct list_head *tmp;
1487
1488         ITERATE_RDEV(mddev,rdev,tmp) {
1489                 super_types[mddev->major_version].
1490                         sync_super(mddev, rdev);
1491                 rdev->sb_loaded = 1;
1492         }
1493 }
1494
1495 static void md_update_sb(mddev_t * mddev)
1496 {
1497         int err;
1498         struct list_head *tmp;
1499         mdk_rdev_t *rdev;
1500         int sync_req;
1501
1502 repeat:
1503         spin_lock_irq(&mddev->write_lock);
1504         sync_req = mddev->in_sync;
1505         mddev->utime = get_seconds();
1506         mddev->events ++;
1507
1508         if (!mddev->events) {
1509                 /*
1510                  * oops, this 64-bit counter should never wrap.
1511                  * Either we are in around ~1 trillion A.C., assuming
1512                  * 1 reboot per second, or we have a bug:
1513                  */
1514                 MD_BUG();
1515                 mddev->events --;
1516         }
1517         mddev->sb_dirty = 2;
1518         sync_sbs(mddev);
1519
1520         /*
1521          * do not write anything to disk if using
1522          * nonpersistent superblocks
1523          */
1524         if (!mddev->persistent) {
1525                 mddev->sb_dirty = 0;
1526                 spin_unlock_irq(&mddev->write_lock);
1527                 wake_up(&mddev->sb_wait);
1528                 return;
1529         }
1530         spin_unlock_irq(&mddev->write_lock);
1531
1532         dprintk(KERN_INFO
1533                 "md: updating %s RAID superblock on device (in sync %d)\n",
1534                 mdname(mddev),mddev->in_sync);
1535
1536         err = bitmap_update_sb(mddev->bitmap);
1537         ITERATE_RDEV(mddev,rdev,tmp) {
1538                 char b[BDEVNAME_SIZE];
1539                 dprintk(KERN_INFO "md: ");
1540                 if (test_bit(Faulty, &rdev->flags))
1541                         dprintk("(skipping faulty ");
1542
1543                 dprintk("%s ", bdevname(rdev->bdev,b));
1544                 if (!test_bit(Faulty, &rdev->flags)) {
1545                         md_super_write(mddev,rdev,
1546                                        rdev->sb_offset<<1, rdev->sb_size,
1547                                        rdev->sb_page);
1548                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1549                                 bdevname(rdev->bdev,b),
1550                                 (unsigned long long)rdev->sb_offset);
1551
1552                 } else
1553                         dprintk(")\n");
1554                 if (mddev->level == LEVEL_MULTIPATH)
1555                         /* only need to write one superblock... */
1556                         break;
1557         }
1558         md_super_wait(mddev);
1559         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1560
1561         spin_lock_irq(&mddev->write_lock);
1562         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1563                 /* have to write it out again */
1564                 spin_unlock_irq(&mddev->write_lock);
1565                 goto repeat;
1566         }
1567         mddev->sb_dirty = 0;
1568         spin_unlock_irq(&mddev->write_lock);
1569         wake_up(&mddev->sb_wait);
1570
1571 }
1572
1573 /* words written to sysfs files may, or my not, be \n terminated.
1574  * We want to accept with case. For this we use cmd_match.
1575  */
1576 static int cmd_match(const char *cmd, const char *str)
1577 {
1578         /* See if cmd, written into a sysfs file, matches
1579          * str.  They must either be the same, or cmd can
1580          * have a trailing newline
1581          */
1582         while (*cmd && *str && *cmd == *str) {
1583                 cmd++;
1584                 str++;
1585         }
1586         if (*cmd == '\n')
1587                 cmd++;
1588         if (*str || *cmd)
1589                 return 0;
1590         return 1;
1591 }
1592
1593 struct rdev_sysfs_entry {
1594         struct attribute attr;
1595         ssize_t (*show)(mdk_rdev_t *, char *);
1596         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1597 };
1598
1599 static ssize_t
1600 state_show(mdk_rdev_t *rdev, char *page)
1601 {
1602         char *sep = "";
1603         int len=0;
1604
1605         if (test_bit(Faulty, &rdev->flags)) {
1606                 len+= sprintf(page+len, "%sfaulty",sep);
1607                 sep = ",";
1608         }
1609         if (test_bit(In_sync, &rdev->flags)) {
1610                 len += sprintf(page+len, "%sin_sync",sep);
1611                 sep = ",";
1612         }
1613         if (!test_bit(Faulty, &rdev->flags) &&
1614             !test_bit(In_sync, &rdev->flags)) {
1615                 len += sprintf(page+len, "%sspare", sep);
1616                 sep = ",";
1617         }
1618         return len+sprintf(page+len, "\n");
1619 }
1620
1621 static struct rdev_sysfs_entry
1622 rdev_state = __ATTR_RO(state);
1623
1624 static ssize_t
1625 super_show(mdk_rdev_t *rdev, char *page)
1626 {
1627         if (rdev->sb_loaded && rdev->sb_size) {
1628                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1629                 return rdev->sb_size;
1630         } else
1631                 return 0;
1632 }
1633 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1634
1635 static ssize_t
1636 errors_show(mdk_rdev_t *rdev, char *page)
1637 {
1638         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1639 }
1640
1641 static ssize_t
1642 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1643 {
1644         char *e;
1645         unsigned long n = simple_strtoul(buf, &e, 10);
1646         if (*buf && (*e == 0 || *e == '\n')) {
1647                 atomic_set(&rdev->corrected_errors, n);
1648                 return len;
1649         }
1650         return -EINVAL;
1651 }
1652 static struct rdev_sysfs_entry rdev_errors =
1653 __ATTR(errors, 0644, errors_show, errors_store);
1654
1655 static ssize_t
1656 slot_show(mdk_rdev_t *rdev, char *page)
1657 {
1658         if (rdev->raid_disk < 0)
1659                 return sprintf(page, "none\n");
1660         else
1661                 return sprintf(page, "%d\n", rdev->raid_disk);
1662 }
1663
1664 static ssize_t
1665 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1666 {
1667         char *e;
1668         int slot = simple_strtoul(buf, &e, 10);
1669         if (strncmp(buf, "none", 4)==0)
1670                 slot = -1;
1671         else if (e==buf || (*e && *e!= '\n'))
1672                 return -EINVAL;
1673         if (rdev->mddev->pers)
1674                 /* Cannot set slot in active array (yet) */
1675                 return -EBUSY;
1676         if (slot >= rdev->mddev->raid_disks)
1677                 return -ENOSPC;
1678         rdev->raid_disk = slot;
1679         /* assume it is working */
1680         rdev->flags = 0;
1681         set_bit(In_sync, &rdev->flags);
1682         return len;
1683 }
1684
1685
1686 static struct rdev_sysfs_entry rdev_slot =
1687 __ATTR(slot, 0644, slot_show, slot_store);
1688
1689 static ssize_t
1690 offset_show(mdk_rdev_t *rdev, char *page)
1691 {
1692         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1693 }
1694
1695 static ssize_t
1696 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1697 {
1698         char *e;
1699         unsigned long long offset = simple_strtoull(buf, &e, 10);
1700         if (e==buf || (*e && *e != '\n'))
1701                 return -EINVAL;
1702         if (rdev->mddev->pers)
1703                 return -EBUSY;
1704         rdev->data_offset = offset;
1705         return len;
1706 }
1707
1708 static struct rdev_sysfs_entry rdev_offset =
1709 __ATTR(offset, 0644, offset_show, offset_store);
1710
1711 static ssize_t
1712 rdev_size_show(mdk_rdev_t *rdev, char *page)
1713 {
1714         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1715 }
1716
1717 static ssize_t
1718 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1719 {
1720         char *e;
1721         unsigned long long size = simple_strtoull(buf, &e, 10);
1722         if (e==buf || (*e && *e != '\n'))
1723                 return -EINVAL;
1724         if (rdev->mddev->pers)
1725                 return -EBUSY;
1726         rdev->size = size;
1727         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1728                 rdev->mddev->size = size;
1729         return len;
1730 }
1731
1732 static struct rdev_sysfs_entry rdev_size =
1733 __ATTR(size, 0644, rdev_size_show, rdev_size_store);
1734
1735 static struct attribute *rdev_default_attrs[] = {
1736         &rdev_state.attr,
1737         &rdev_super.attr,
1738         &rdev_errors.attr,
1739         &rdev_slot.attr,
1740         &rdev_offset.attr,
1741         &rdev_size.attr,
1742         NULL,
1743 };
1744 static ssize_t
1745 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1746 {
1747         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1748         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1749
1750         if (!entry->show)
1751                 return -EIO;
1752         return entry->show(rdev, page);
1753 }
1754
1755 static ssize_t
1756 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1757               const char *page, size_t length)
1758 {
1759         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1760         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1761
1762         if (!entry->store)
1763                 return -EIO;
1764         return entry->store(rdev, page, length);
1765 }
1766
1767 static void rdev_free(struct kobject *ko)
1768 {
1769         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1770         kfree(rdev);
1771 }
1772 static struct sysfs_ops rdev_sysfs_ops = {
1773         .show           = rdev_attr_show,
1774         .store          = rdev_attr_store,
1775 };
1776 static struct kobj_type rdev_ktype = {
1777         .release        = rdev_free,
1778         .sysfs_ops      = &rdev_sysfs_ops,
1779         .default_attrs  = rdev_default_attrs,
1780 };
1781
1782 /*
1783  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1784  *
1785  * mark the device faulty if:
1786  *
1787  *   - the device is nonexistent (zero size)
1788  *   - the device has no valid superblock
1789  *
1790  * a faulty rdev _never_ has rdev->sb set.
1791  */
1792 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1793 {
1794         char b[BDEVNAME_SIZE];
1795         int err;
1796         mdk_rdev_t *rdev;
1797         sector_t size;
1798
1799         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1800         if (!rdev) {
1801                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1802                 return ERR_PTR(-ENOMEM);
1803         }
1804
1805         if ((err = alloc_disk_sb(rdev)))
1806                 goto abort_free;
1807
1808         err = lock_rdev(rdev, newdev);
1809         if (err)
1810                 goto abort_free;
1811
1812         rdev->kobj.parent = NULL;
1813         rdev->kobj.ktype = &rdev_ktype;
1814         kobject_init(&rdev->kobj);
1815
1816         rdev->desc_nr = -1;
1817         rdev->saved_raid_disk = -1;
1818         rdev->flags = 0;
1819         rdev->data_offset = 0;
1820         atomic_set(&rdev->nr_pending, 0);
1821         atomic_set(&rdev->read_errors, 0);
1822         atomic_set(&rdev->corrected_errors, 0);
1823
1824         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1825         if (!size) {
1826                 printk(KERN_WARNING
1827                         "md: %s has zero or unknown size, marking faulty!\n",
1828                         bdevname(rdev->bdev,b));
1829                 err = -EINVAL;
1830                 goto abort_free;
1831         }
1832
1833         if (super_format >= 0) {
1834                 err = super_types[super_format].
1835                         load_super(rdev, NULL, super_minor);
1836                 if (err == -EINVAL) {
1837                         printk(KERN_WARNING
1838                                 "md: %s has invalid sb, not importing!\n",
1839                                 bdevname(rdev->bdev,b));
1840                         goto abort_free;
1841                 }
1842                 if (err < 0) {
1843                         printk(KERN_WARNING
1844                                 "md: could not read %s's sb, not importing!\n",
1845                                 bdevname(rdev->bdev,b));
1846                         goto abort_free;
1847                 }
1848         }
1849         INIT_LIST_HEAD(&rdev->same_set);
1850
1851         return rdev;
1852
1853 abort_free:
1854         if (rdev->sb_page) {
1855                 if (rdev->bdev)
1856                         unlock_rdev(rdev);
1857                 free_disk_sb(rdev);
1858         }
1859         kfree(rdev);
1860         return ERR_PTR(err);
1861 }
1862
1863 /*
1864  * Check a full RAID array for plausibility
1865  */
1866
1867
1868 static void analyze_sbs(mddev_t * mddev)
1869 {
1870         int i;
1871         struct list_head *tmp;
1872         mdk_rdev_t *rdev, *freshest;
1873         char b[BDEVNAME_SIZE];
1874
1875         freshest = NULL;
1876         ITERATE_RDEV(mddev,rdev,tmp)
1877                 switch (super_types[mddev->major_version].
1878                         load_super(rdev, freshest, mddev->minor_version)) {
1879                 case 1:
1880                         freshest = rdev;
1881                         break;
1882                 case 0:
1883                         break;
1884                 default:
1885                         printk( KERN_ERR \
1886                                 "md: fatal superblock inconsistency in %s"
1887                                 " -- removing from array\n",
1888                                 bdevname(rdev->bdev,b));
1889                         kick_rdev_from_array(rdev);
1890                 }
1891
1892
1893         super_types[mddev->major_version].
1894                 validate_super(mddev, freshest);
1895
1896         i = 0;
1897         ITERATE_RDEV(mddev,rdev,tmp) {
1898                 if (rdev != freshest)
1899                         if (super_types[mddev->major_version].
1900                             validate_super(mddev, rdev)) {
1901                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1902                                         " from array!\n",
1903                                         bdevname(rdev->bdev,b));
1904                                 kick_rdev_from_array(rdev);
1905                                 continue;
1906                         }
1907                 if (mddev->level == LEVEL_MULTIPATH) {
1908                         rdev->desc_nr = i++;
1909                         rdev->raid_disk = rdev->desc_nr;
1910                         set_bit(In_sync, &rdev->flags);
1911                 }
1912         }
1913
1914
1915
1916         if (mddev->recovery_cp != MaxSector &&
1917             mddev->level >= 1)
1918                 printk(KERN_ERR "md: %s: raid array is not clean"
1919                        " -- starting background reconstruction\n",
1920                        mdname(mddev));
1921
1922 }
1923
1924 static ssize_t
1925 level_show(mddev_t *mddev, char *page)
1926 {
1927         struct mdk_personality *p = mddev->pers;
1928         if (p)
1929                 return sprintf(page, "%s\n", p->name);
1930         else if (mddev->clevel[0])
1931                 return sprintf(page, "%s\n", mddev->clevel);
1932         else if (mddev->level != LEVEL_NONE)
1933                 return sprintf(page, "%d\n", mddev->level);
1934         else
1935                 return 0;
1936 }
1937
1938 static ssize_t
1939 level_store(mddev_t *mddev, const char *buf, size_t len)
1940 {
1941         int rv = len;
1942         if (mddev->pers)
1943                 return -EBUSY;
1944         if (len == 0)
1945                 return 0;
1946         if (len >= sizeof(mddev->clevel))
1947                 return -ENOSPC;
1948         strncpy(mddev->clevel, buf, len);
1949         if (mddev->clevel[len-1] == '\n')
1950                 len--;
1951         mddev->clevel[len] = 0;
1952         mddev->level = LEVEL_NONE;
1953         return rv;
1954 }
1955
1956 static struct md_sysfs_entry md_level =
1957 __ATTR(level, 0644, level_show, level_store);
1958
1959 static ssize_t
1960 raid_disks_show(mddev_t *mddev, char *page)
1961 {
1962         if (mddev->raid_disks == 0)
1963                 return 0;
1964         return sprintf(page, "%d\n", mddev->raid_disks);
1965 }
1966
1967 static int update_raid_disks(mddev_t *mddev, int raid_disks);
1968
1969 static ssize_t
1970 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
1971 {
1972         /* can only set raid_disks if array is not yet active */
1973         char *e;
1974         int rv = 0;
1975         unsigned long n = simple_strtoul(buf, &e, 10);
1976
1977         if (!*buf || (*e && *e != '\n'))
1978                 return -EINVAL;
1979
1980         if (mddev->pers)
1981                 rv = update_raid_disks(mddev, n);
1982         else
1983                 mddev->raid_disks = n;
1984         return rv ? rv : len;
1985 }
1986 static struct md_sysfs_entry md_raid_disks =
1987 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
1988
1989 static ssize_t
1990 chunk_size_show(mddev_t *mddev, char *page)
1991 {
1992         return sprintf(page, "%d\n", mddev->chunk_size);
1993 }
1994
1995 static ssize_t
1996 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1997 {
1998         /* can only set chunk_size if array is not yet active */
1999         char *e;
2000         unsigned long n = simple_strtoul(buf, &e, 10);
2001
2002         if (mddev->pers)
2003                 return -EBUSY;
2004         if (!*buf || (*e && *e != '\n'))
2005                 return -EINVAL;
2006
2007         mddev->chunk_size = n;
2008         return len;
2009 }
2010 static struct md_sysfs_entry md_chunk_size =
2011 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2012
2013 static ssize_t
2014 null_show(mddev_t *mddev, char *page)
2015 {
2016         return -EINVAL;
2017 }
2018
2019 static ssize_t
2020 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2021 {
2022         /* buf must be %d:%d\n? giving major and minor numbers */
2023         /* The new device is added to the array.
2024          * If the array has a persistent superblock, we read the
2025          * superblock to initialise info and check validity.
2026          * Otherwise, only checking done is that in bind_rdev_to_array,
2027          * which mainly checks size.
2028          */
2029         char *e;
2030         int major = simple_strtoul(buf, &e, 10);
2031         int minor;
2032         dev_t dev;
2033         mdk_rdev_t *rdev;
2034         int err;
2035
2036         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2037                 return -EINVAL;
2038         minor = simple_strtoul(e+1, &e, 10);
2039         if (*e && *e != '\n')
2040                 return -EINVAL;
2041         dev = MKDEV(major, minor);
2042         if (major != MAJOR(dev) ||
2043             minor != MINOR(dev))
2044                 return -EOVERFLOW;
2045
2046
2047         if (mddev->persistent) {
2048                 rdev = md_import_device(dev, mddev->major_version,
2049                                         mddev->minor_version);
2050                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2051                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2052                                                        mdk_rdev_t, same_set);
2053                         err = super_types[mddev->major_version]
2054                                 .load_super(rdev, rdev0, mddev->minor_version);
2055                         if (err < 0)
2056                                 goto out;
2057                 }
2058         } else
2059                 rdev = md_import_device(dev, -1, -1);
2060
2061         if (IS_ERR(rdev))
2062                 return PTR_ERR(rdev);
2063         err = bind_rdev_to_array(rdev, mddev);
2064  out:
2065         if (err)
2066                 export_rdev(rdev);
2067         return err ? err : len;
2068 }
2069
2070 static struct md_sysfs_entry md_new_device =
2071 __ATTR(new_dev, 0200, null_show, new_dev_store);
2072
2073 static ssize_t
2074 size_show(mddev_t *mddev, char *page)
2075 {
2076         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2077 }
2078
2079 static int update_size(mddev_t *mddev, unsigned long size);
2080
2081 static ssize_t
2082 size_store(mddev_t *mddev, const char *buf, size_t len)
2083 {
2084         /* If array is inactive, we can reduce the component size, but
2085          * not increase it (except from 0).
2086          * If array is active, we can try an on-line resize
2087          */
2088         char *e;
2089         int err = 0;
2090         unsigned long long size = simple_strtoull(buf, &e, 10);
2091         if (!*buf || *buf == '\n' ||
2092             (*e && *e != '\n'))
2093                 return -EINVAL;
2094
2095         if (mddev->pers) {
2096                 err = update_size(mddev, size);
2097                 md_update_sb(mddev);
2098         } else {
2099                 if (mddev->size == 0 ||
2100                     mddev->size > size)
2101                         mddev->size = size;
2102                 else
2103                         err = -ENOSPC;
2104         }
2105         return err ? err : len;
2106 }
2107
2108 static struct md_sysfs_entry md_size =
2109 __ATTR(component_size, 0644, size_show, size_store);
2110
2111
2112 /* Metdata version.
2113  * This is either 'none' for arrays with externally managed metadata,
2114  * or N.M for internally known formats
2115  */
2116 static ssize_t
2117 metadata_show(mddev_t *mddev, char *page)
2118 {
2119         if (mddev->persistent)
2120                 return sprintf(page, "%d.%d\n",
2121                                mddev->major_version, mddev->minor_version);
2122         else
2123                 return sprintf(page, "none\n");
2124 }
2125
2126 static ssize_t
2127 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2128 {
2129         int major, minor;
2130         char *e;
2131         if (!list_empty(&mddev->disks))
2132                 return -EBUSY;
2133
2134         if (cmd_match(buf, "none")) {
2135                 mddev->persistent = 0;
2136                 mddev->major_version = 0;
2137                 mddev->minor_version = 90;
2138                 return len;
2139         }
2140         major = simple_strtoul(buf, &e, 10);
2141         if (e==buf || *e != '.')
2142                 return -EINVAL;
2143         buf = e+1;
2144         minor = simple_strtoul(buf, &e, 10);
2145         if (e==buf || *e != '\n')
2146                 return -EINVAL;
2147         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2148             super_types[major].name == NULL)
2149                 return -ENOENT;
2150         mddev->major_version = major;
2151         mddev->minor_version = minor;
2152         mddev->persistent = 1;
2153         return len;
2154 }
2155
2156 static struct md_sysfs_entry md_metadata =
2157 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
2158
2159 static ssize_t
2160 action_show(mddev_t *mddev, char *page)
2161 {
2162         char *type = "idle";
2163         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2164             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2165                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2166                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2167                                 type = "resync";
2168                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2169                                 type = "check";
2170                         else
2171                                 type = "repair";
2172                 } else
2173                         type = "recover";
2174         }
2175         return sprintf(page, "%s\n", type);
2176 }
2177
2178 static ssize_t
2179 action_store(mddev_t *mddev, const char *page, size_t len)
2180 {
2181         if (!mddev->pers || !mddev->pers->sync_request)
2182                 return -EINVAL;
2183
2184         if (cmd_match(page, "idle")) {
2185                 if (mddev->sync_thread) {
2186                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2187                         md_unregister_thread(mddev->sync_thread);
2188                         mddev->sync_thread = NULL;
2189                         mddev->recovery = 0;
2190                 }
2191         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2192                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2193                 return -EBUSY;
2194         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2195                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2196         else {
2197                 if (cmd_match(page, "check"))
2198                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2199                 else if (!cmd_match(page, "repair"))
2200                         return -EINVAL;
2201                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2202                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2203         }
2204         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2205         md_wakeup_thread(mddev->thread);
2206         return len;
2207 }
2208
2209 static ssize_t
2210 mismatch_cnt_show(mddev_t *mddev, char *page)
2211 {
2212         return sprintf(page, "%llu\n",
2213                        (unsigned long long) mddev->resync_mismatches);
2214 }
2215
2216 static struct md_sysfs_entry
2217 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2218
2219
2220 static struct md_sysfs_entry
2221 md_mismatches = __ATTR_RO(mismatch_cnt);
2222
2223 static ssize_t
2224 sync_min_show(mddev_t *mddev, char *page)
2225 {
2226         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2227                        mddev->sync_speed_min ? "local": "system");
2228 }
2229
2230 static ssize_t
2231 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2232 {
2233         int min;
2234         char *e;
2235         if (strncmp(buf, "system", 6)==0) {
2236                 mddev->sync_speed_min = 0;
2237                 return len;
2238         }
2239         min = simple_strtoul(buf, &e, 10);
2240         if (buf == e || (*e && *e != '\n') || min <= 0)
2241                 return -EINVAL;
2242         mddev->sync_speed_min = min;
2243         return len;
2244 }
2245
2246 static struct md_sysfs_entry md_sync_min =
2247 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2248
2249 static ssize_t
2250 sync_max_show(mddev_t *mddev, char *page)
2251 {
2252         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2253                        mddev->sync_speed_max ? "local": "system");
2254 }
2255
2256 static ssize_t
2257 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2258 {
2259         int max;
2260         char *e;
2261         if (strncmp(buf, "system", 6)==0) {
2262                 mddev->sync_speed_max = 0;
2263                 return len;
2264         }
2265         max = simple_strtoul(buf, &e, 10);
2266         if (buf == e || (*e && *e != '\n') || max <= 0)
2267                 return -EINVAL;
2268         mddev->sync_speed_max = max;
2269         return len;
2270 }
2271
2272 static struct md_sysfs_entry md_sync_max =
2273 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2274
2275
2276 static ssize_t
2277 sync_speed_show(mddev_t *mddev, char *page)
2278 {
2279         unsigned long resync, dt, db;
2280         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2281         dt = ((jiffies - mddev->resync_mark) / HZ);
2282         if (!dt) dt++;
2283         db = resync - (mddev->resync_mark_cnt);
2284         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2285 }
2286
2287 static struct md_sysfs_entry
2288 md_sync_speed = __ATTR_RO(sync_speed);
2289
2290 static ssize_t
2291 sync_completed_show(mddev_t *mddev, char *page)
2292 {
2293         unsigned long max_blocks, resync;
2294
2295         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2296                 max_blocks = mddev->resync_max_sectors;
2297         else
2298                 max_blocks = mddev->size << 1;
2299
2300         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2301         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2302 }
2303
2304 static struct md_sysfs_entry
2305 md_sync_completed = __ATTR_RO(sync_completed);
2306
2307 static struct attribute *md_default_attrs[] = {
2308         &md_level.attr,
2309         &md_raid_disks.attr,
2310         &md_chunk_size.attr,
2311         &md_size.attr,
2312         &md_metadata.attr,
2313         &md_new_device.attr,
2314         NULL,
2315 };
2316
2317 static struct attribute *md_redundancy_attrs[] = {
2318         &md_scan_mode.attr,
2319         &md_mismatches.attr,
2320         &md_sync_min.attr,
2321         &md_sync_max.attr,
2322         &md_sync_speed.attr,
2323         &md_sync_completed.attr,
2324         NULL,
2325 };
2326 static struct attribute_group md_redundancy_group = {
2327         .name = NULL,
2328         .attrs = md_redundancy_attrs,
2329 };
2330
2331
2332 static ssize_t
2333 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2334 {
2335         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2336         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2337         ssize_t rv;
2338
2339         if (!entry->show)
2340                 return -EIO;
2341         mddev_lock(mddev);
2342         rv = entry->show(mddev, page);
2343         mddev_unlock(mddev);
2344         return rv;
2345 }
2346
2347 static ssize_t
2348 md_attr_store(struct kobject *kobj, struct attribute *attr,
2349               const char *page, size_t length)
2350 {
2351         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2352         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2353         ssize_t rv;
2354
2355         if (!entry->store)
2356                 return -EIO;
2357         mddev_lock(mddev);
2358         rv = entry->store(mddev, page, length);
2359         mddev_unlock(mddev);
2360         return rv;
2361 }
2362
2363 static void md_free(struct kobject *ko)
2364 {
2365         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2366         kfree(mddev);
2367 }
2368
2369 static struct sysfs_ops md_sysfs_ops = {
2370         .show   = md_attr_show,
2371         .store  = md_attr_store,
2372 };
2373 static struct kobj_type md_ktype = {
2374         .release        = md_free,
2375         .sysfs_ops      = &md_sysfs_ops,
2376         .default_attrs  = md_default_attrs,
2377 };
2378
2379 int mdp_major = 0;
2380
2381 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2382 {
2383         static DECLARE_MUTEX(disks_sem);
2384         mddev_t *mddev = mddev_find(dev);
2385         struct gendisk *disk;
2386         int partitioned = (MAJOR(dev) != MD_MAJOR);
2387         int shift = partitioned ? MdpMinorShift : 0;
2388         int unit = MINOR(dev) >> shift;
2389
2390         if (!mddev)
2391                 return NULL;
2392
2393         down(&disks_sem);
2394         if (mddev->gendisk) {
2395                 up(&disks_sem);
2396                 mddev_put(mddev);
2397                 return NULL;
2398         }
2399         disk = alloc_disk(1 << shift);
2400         if (!disk) {
2401                 up(&disks_sem);
2402                 mddev_put(mddev);
2403                 return NULL;
2404         }
2405         disk->major = MAJOR(dev);
2406         disk->first_minor = unit << shift;
2407         if (partitioned) {
2408                 sprintf(disk->disk_name, "md_d%d", unit);
2409                 sprintf(disk->devfs_name, "md/d%d", unit);
2410         } else {
2411                 sprintf(disk->disk_name, "md%d", unit);
2412                 sprintf(disk->devfs_name, "md/%d", unit);
2413         }
2414         disk->fops = &md_fops;
2415         disk->private_data = mddev;
2416         disk->queue = mddev->queue;
2417         add_disk(disk);
2418         mddev->gendisk = disk;
2419         up(&disks_sem);
2420         mddev->kobj.parent = &disk->kobj;
2421         mddev->kobj.k_name = NULL;
2422         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2423         mddev->kobj.ktype = &md_ktype;
2424         kobject_register(&mddev->kobj);
2425         return NULL;
2426 }
2427
2428 void md_wakeup_thread(mdk_thread_t *thread);
2429
2430 static void md_safemode_timeout(unsigned long data)
2431 {
2432         mddev_t *mddev = (mddev_t *) data;
2433
2434         mddev->safemode = 1;
2435         md_wakeup_thread(mddev->thread);
2436 }
2437
2438 static int start_dirty_degraded;
2439
2440 static int do_md_run(mddev_t * mddev)
2441 {
2442         int err;
2443         int chunk_size;
2444         struct list_head *tmp;
2445         mdk_rdev_t *rdev;
2446         struct gendisk *disk;
2447         struct mdk_personality *pers;
2448         char b[BDEVNAME_SIZE];
2449
2450         if (list_empty(&mddev->disks))
2451                 /* cannot run an array with no devices.. */
2452                 return -EINVAL;
2453
2454         if (mddev->pers)
2455                 return -EBUSY;
2456
2457         /*
2458          * Analyze all RAID superblock(s)
2459          */
2460         if (!mddev->raid_disks)
2461                 analyze_sbs(mddev);
2462
2463         chunk_size = mddev->chunk_size;
2464
2465         if (chunk_size) {
2466                 if (chunk_size > MAX_CHUNK_SIZE) {
2467                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2468                                 chunk_size, MAX_CHUNK_SIZE);
2469                         return -EINVAL;
2470                 }
2471                 /*
2472                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2473                  */
2474                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2475                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2476                         return -EINVAL;
2477                 }
2478                 if (chunk_size < PAGE_SIZE) {
2479                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2480                                 chunk_size, PAGE_SIZE);
2481                         return -EINVAL;
2482                 }
2483
2484                 /* devices must have minimum size of one chunk */
2485                 ITERATE_RDEV(mddev,rdev,tmp) {
2486                         if (test_bit(Faulty, &rdev->flags))
2487                                 continue;
2488                         if (rdev->size < chunk_size / 1024) {
2489                                 printk(KERN_WARNING
2490                                         "md: Dev %s smaller than chunk_size:"
2491                                         " %lluk < %dk\n",
2492                                         bdevname(rdev->bdev,b),
2493                                         (unsigned long long)rdev->size,
2494                                         chunk_size / 1024);
2495                                 return -EINVAL;
2496                         }
2497                 }
2498         }
2499
2500 #ifdef CONFIG_KMOD
2501         if (mddev->level != LEVEL_NONE)
2502                 request_module("md-level-%d", mddev->level);
2503         else if (mddev->clevel[0])
2504                 request_module("md-%s", mddev->clevel);
2505 #endif
2506
2507         /*
2508          * Drop all container device buffers, from now on
2509          * the only valid external interface is through the md
2510          * device.
2511          * Also find largest hardsector size
2512          */
2513         ITERATE_RDEV(mddev,rdev,tmp) {
2514                 if (test_bit(Faulty, &rdev->flags))
2515                         continue;
2516                 sync_blockdev(rdev->bdev);
2517                 invalidate_bdev(rdev->bdev, 0);
2518         }
2519
2520         md_probe(mddev->unit, NULL, NULL);
2521         disk = mddev->gendisk;
2522         if (!disk)
2523                 return -ENOMEM;
2524
2525         spin_lock(&pers_lock);
2526         pers = find_pers(mddev->level, mddev->clevel);
2527         if (!pers || !try_module_get(pers->owner)) {
2528                 spin_unlock(&pers_lock);
2529                 if (mddev->level != LEVEL_NONE)
2530                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2531                                mddev->level);
2532                 else
2533                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2534                                mddev->clevel);
2535                 return -EINVAL;
2536         }
2537         mddev->pers = pers;
2538         spin_unlock(&pers_lock);
2539         mddev->level = pers->level;
2540         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2541
2542         mddev->recovery = 0;
2543         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2544         mddev->barriers_work = 1;
2545         mddev->ok_start_degraded = start_dirty_degraded;
2546
2547         if (start_readonly)
2548                 mddev->ro = 2; /* read-only, but switch on first write */
2549
2550         err = mddev->pers->run(mddev);
2551         if (!err && mddev->pers->sync_request) {
2552                 err = bitmap_create(mddev);
2553                 if (err) {
2554                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2555                                mdname(mddev), err);
2556                         mddev->pers->stop(mddev);
2557                 }
2558         }
2559         if (err) {
2560                 printk(KERN_ERR "md: pers->run() failed ...\n");
2561                 module_put(mddev->pers->owner);
2562                 mddev->pers = NULL;
2563                 bitmap_destroy(mddev);
2564                 return err;
2565         }
2566         if (mddev->pers->sync_request)
2567                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2568         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2569                 mddev->ro = 0;
2570
2571         atomic_set(&mddev->writes_pending,0);
2572         mddev->safemode = 0;
2573         mddev->safemode_timer.function = md_safemode_timeout;
2574         mddev->safemode_timer.data = (unsigned long) mddev;
2575         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2576         mddev->in_sync = 1;
2577
2578         ITERATE_RDEV(mddev,rdev,tmp)
2579                 if (rdev->raid_disk >= 0) {
2580                         char nm[20];
2581                         sprintf(nm, "rd%d", rdev->raid_disk);
2582                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2583                 }
2584
2585         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2586         md_wakeup_thread(mddev->thread);
2587
2588         if (mddev->sb_dirty)
2589                 md_update_sb(mddev);
2590
2591         set_capacity(disk, mddev->array_size<<1);
2592
2593         /* If we call blk_queue_make_request here, it will
2594          * re-initialise max_sectors etc which may have been
2595          * refined inside -> run.  So just set the bits we need to set.
2596          * Most initialisation happended when we called
2597          * blk_queue_make_request(..., md_fail_request)
2598          * earlier.
2599          */
2600         mddev->queue->queuedata = mddev;
2601         mddev->queue->make_request_fn = mddev->pers->make_request;
2602
2603         mddev->changed = 1;
2604         md_new_event(mddev);
2605         return 0;
2606 }
2607
2608 static int restart_array(mddev_t *mddev)
2609 {
2610         struct gendisk *disk = mddev->gendisk;
2611         int err;
2612
2613         /*
2614          * Complain if it has no devices
2615          */
2616         err = -ENXIO;
2617         if (list_empty(&mddev->disks))
2618                 goto out;
2619
2620         if (mddev->pers) {
2621                 err = -EBUSY;
2622                 if (!mddev->ro)
2623                         goto out;
2624
2625                 mddev->safemode = 0;
2626                 mddev->ro = 0;
2627                 set_disk_ro(disk, 0);
2628
2629                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2630                         mdname(mddev));
2631                 /*
2632                  * Kick recovery or resync if necessary
2633                  */
2634                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2635                 md_wakeup_thread(mddev->thread);
2636                 err = 0;
2637         } else {
2638                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2639                         mdname(mddev));
2640                 err = -EINVAL;
2641         }
2642
2643 out:
2644         return err;
2645 }
2646
2647 static int do_md_stop(mddev_t * mddev, int ro)
2648 {
2649         int err = 0;
2650         struct gendisk *disk = mddev->gendisk;
2651
2652         if (mddev->pers) {
2653                 if (atomic_read(&mddev->active)>2) {
2654                         printk("md: %s still in use.\n",mdname(mddev));
2655                         return -EBUSY;
2656                 }
2657
2658                 if (mddev->sync_thread) {
2659                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2660                         md_unregister_thread(mddev->sync_thread);
2661                         mddev->sync_thread = NULL;
2662                 }
2663
2664                 del_timer_sync(&mddev->safemode_timer);
2665
2666                 invalidate_partition(disk, 0);
2667
2668                 if (ro) {
2669                         err  = -ENXIO;
2670                         if (mddev->ro==1)
2671                                 goto out;
2672                         mddev->ro = 1;
2673                 } else {
2674                         bitmap_flush(mddev);
2675                         md_super_wait(mddev);
2676                         if (mddev->ro)
2677                                 set_disk_ro(disk, 0);
2678                         blk_queue_make_request(mddev->queue, md_fail_request);
2679                         mddev->pers->stop(mddev);
2680                         if (mddev->pers->sync_request)
2681                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2682
2683                         module_put(mddev->pers->owner);
2684                         mddev->pers = NULL;
2685                         if (mddev->ro)
2686                                 mddev->ro = 0;
2687                 }
2688                 if (!mddev->in_sync) {
2689                         /* mark array as shutdown cleanly */
2690                         mddev->in_sync = 1;
2691                         md_update_sb(mddev);
2692                 }
2693                 if (ro)
2694                         set_disk_ro(disk, 1);
2695         }
2696
2697         /*
2698          * Free resources if final stop
2699          */
2700         if (!ro) {
2701                 mdk_rdev_t *rdev;
2702                 struct list_head *tmp;
2703                 struct gendisk *disk;
2704                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2705
2706                 bitmap_destroy(mddev);
2707                 if (mddev->bitmap_file) {
2708                         atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2709                         fput(mddev->bitmap_file);
2710                         mddev->bitmap_file = NULL;
2711                 }
2712                 mddev->bitmap_offset = 0;
2713
2714                 ITERATE_RDEV(mddev,rdev,tmp)
2715                         if (rdev->raid_disk >= 0) {
2716                                 char nm[20];
2717                                 sprintf(nm, "rd%d", rdev->raid_disk);
2718                                 sysfs_remove_link(&mddev->kobj, nm);
2719                         }
2720
2721                 export_array(mddev);
2722
2723                 mddev->array_size = 0;
2724                 disk = mddev->gendisk;
2725                 if (disk)
2726                         set_capacity(disk, 0);
2727                 mddev->changed = 1;
2728         } else
2729                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2730                         mdname(mddev));
2731         err = 0;
2732         md_new_event(mddev);
2733 out:
2734         return err;
2735 }
2736
2737 static void autorun_array(mddev_t *mddev)
2738 {
2739         mdk_rdev_t *rdev;
2740         struct list_head *tmp;
2741         int err;
2742
2743         if (list_empty(&mddev->disks))
2744                 return;
2745
2746         printk(KERN_INFO "md: running: ");
2747
2748         ITERATE_RDEV(mddev,rdev,tmp) {
2749                 char b[BDEVNAME_SIZE];
2750                 printk("<%s>", bdevname(rdev->bdev,b));
2751         }
2752         printk("\n");
2753
2754         err = do_md_run (mddev);
2755         if (err) {
2756                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2757                 do_md_stop (mddev, 0);
2758         }
2759 }
2760
2761 /*
2762  * lets try to run arrays based on all disks that have arrived
2763  * until now. (those are in pending_raid_disks)
2764  *
2765  * the method: pick the first pending disk, collect all disks with
2766  * the same UUID, remove all from the pending list and put them into
2767  * the 'same_array' list. Then order this list based on superblock
2768  * update time (freshest comes first), kick out 'old' disks and
2769  * compare superblocks. If everything's fine then run it.
2770  *
2771  * If "unit" is allocated, then bump its reference count
2772  */
2773 static void autorun_devices(int part)
2774 {
2775         struct list_head candidates;
2776         struct list_head *tmp;
2777         mdk_rdev_t *rdev0, *rdev;
2778         mddev_t *mddev;
2779         char b[BDEVNAME_SIZE];
2780
2781         printk(KERN_INFO "md: autorun ...\n");
2782         while (!list_empty(&pending_raid_disks)) {
2783                 dev_t dev;
2784                 rdev0 = list_entry(pending_raid_disks.next,
2785                                          mdk_rdev_t, same_set);
2786
2787                 printk(KERN_INFO "md: considering %s ...\n",
2788                         bdevname(rdev0->bdev,b));
2789                 INIT_LIST_HEAD(&candidates);
2790                 ITERATE_RDEV_PENDING(rdev,tmp)
2791                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2792                                 printk(KERN_INFO "md:  adding %s ...\n",
2793                                         bdevname(rdev->bdev,b));
2794                                 list_move(&rdev->same_set, &candidates);
2795                         }
2796                 /*
2797                  * now we have a set of devices, with all of them having
2798                  * mostly sane superblocks. It's time to allocate the
2799                  * mddev.
2800                  */
2801                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2802                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2803                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2804                         break;
2805                 }
2806                 if (part)
2807                         dev = MKDEV(mdp_major,
2808                                     rdev0->preferred_minor << MdpMinorShift);
2809                 else
2810                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2811
2812                 md_probe(dev, NULL, NULL);
2813                 mddev = mddev_find(dev);
2814                 if (!mddev) {
2815                         printk(KERN_ERR
2816                                 "md: cannot allocate memory for md drive.\n");
2817                         break;
2818                 }
2819                 if (mddev_lock(mddev))
2820                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2821                                mdname(mddev));
2822                 else if (mddev->raid_disks || mddev->major_version
2823                          || !list_empty(&mddev->disks)) {
2824                         printk(KERN_WARNING
2825                                 "md: %s already running, cannot run %s\n",
2826                                 mdname(mddev), bdevname(rdev0->bdev,b));
2827                         mddev_unlock(mddev);
2828                 } else {
2829                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2830                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2831                                 list_del_init(&rdev->same_set);
2832                                 if (bind_rdev_to_array(rdev, mddev))
2833                                         export_rdev(rdev);
2834                         }
2835                         autorun_array(mddev);
2836                         mddev_unlock(mddev);
2837                 }
2838                 /* on success, candidates will be empty, on error
2839                  * it won't...
2840                  */
2841                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2842                         export_rdev(rdev);
2843                 mddev_put(mddev);
2844         }
2845         printk(KERN_INFO "md: ... autorun DONE.\n");
2846 }
2847
2848 /*
2849  * import RAID devices based on one partition
2850  * if possible, the array gets run as well.
2851  */
2852
2853 static int autostart_array(dev_t startdev)
2854 {
2855         char b[BDEVNAME_SIZE];
2856         int err = -EINVAL, i;
2857         mdp_super_t *sb = NULL;
2858         mdk_rdev_t *start_rdev = NULL, *rdev;
2859
2860         start_rdev = md_import_device(startdev, 0, 0);
2861         if (IS_ERR(start_rdev))
2862                 return err;
2863
2864
2865         /* NOTE: this can only work for 0.90.0 superblocks */
2866         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2867         if (sb->major_version != 0 ||
2868             sb->minor_version != 90 ) {
2869                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2870                 export_rdev(start_rdev);
2871                 return err;
2872         }
2873
2874         if (test_bit(Faulty, &start_rdev->flags)) {
2875                 printk(KERN_WARNING
2876                         "md: can not autostart based on faulty %s!\n",
2877                         bdevname(start_rdev->bdev,b));
2878                 export_rdev(start_rdev);
2879                 return err;
2880         }
2881         list_add(&start_rdev->same_set, &pending_raid_disks);
2882
2883         for (i = 0; i < MD_SB_DISKS; i++) {
2884                 mdp_disk_t *desc = sb->disks + i;
2885                 dev_t dev = MKDEV(desc->major, desc->minor);
2886
2887                 if (!dev)
2888                         continue;
2889                 if (dev == startdev)
2890                         continue;
2891                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2892                         continue;
2893                 rdev = md_import_device(dev, 0, 0);
2894                 if (IS_ERR(rdev))
2895                         continue;
2896
2897                 list_add(&rdev->same_set, &pending_raid_disks);
2898         }
2899
2900         /*
2901          * possibly return codes
2902          */
2903         autorun_devices(0);
2904         return 0;
2905
2906 }
2907
2908
2909 static int get_version(void __user * arg)
2910 {
2911         mdu_version_t ver;
2912
2913         ver.major = MD_MAJOR_VERSION;
2914         ver.minor = MD_MINOR_VERSION;
2915         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2916
2917         if (copy_to_user(arg, &ver, sizeof(ver)))
2918                 return -EFAULT;
2919
2920         return 0;
2921 }
2922
2923 static int get_array_info(mddev_t * mddev, void __user * arg)
2924 {
2925         mdu_array_info_t info;
2926         int nr,working,active,failed,spare;
2927         mdk_rdev_t *rdev;
2928         struct list_head *tmp;
2929
2930         nr=working=active=failed=spare=0;
2931         ITERATE_RDEV(mddev,rdev,tmp) {
2932                 nr++;
2933                 if (test_bit(Faulty, &rdev->flags))
2934                         failed++;
2935                 else {
2936                         working++;
2937                         if (test_bit(In_sync, &rdev->flags))
2938                                 active++;
2939                         else
2940                                 spare++;
2941                 }
2942         }
2943
2944         info.major_version = mddev->major_version;
2945         info.minor_version = mddev->minor_version;
2946         info.patch_version = MD_PATCHLEVEL_VERSION;
2947         info.ctime         = mddev->ctime;
2948         info.level         = mddev->level;
2949         info.size          = mddev->size;
2950         if (info.size != mddev->size) /* overflow */
2951                 info.size = -1;
2952         info.nr_disks      = nr;
2953         info.raid_disks    = mddev->raid_disks;
2954         info.md_minor      = mddev->md_minor;
2955         info.not_persistent= !mddev->persistent;
2956
2957         info.utime         = mddev->utime;
2958         info.state         = 0;
2959         if (mddev->in_sync)
2960                 info.state = (1<<MD_SB_CLEAN);
2961         if (mddev->bitmap && mddev->bitmap_offset)
2962                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2963         info.active_disks  = active;
2964         info.working_disks = working;
2965         info.failed_disks  = failed;
2966         info.spare_disks   = spare;
2967
2968         info.layout        = mddev->layout;
2969         info.chunk_size    = mddev->chunk_size;
2970
2971         if (copy_to_user(arg, &info, sizeof(info)))
2972                 return -EFAULT;
2973
2974         return 0;
2975 }
2976
2977 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2978 {
2979         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2980         char *ptr, *buf = NULL;
2981         int err = -ENOMEM;
2982
2983         file = kmalloc(sizeof(*file), GFP_KERNEL);
2984         if (!file)
2985                 goto out;
2986
2987         /* bitmap disabled, zero the first byte and copy out */
2988         if (!mddev->bitmap || !mddev->bitmap->file) {
2989                 file->pathname[0] = '\0';
2990                 goto copy_out;
2991         }
2992
2993         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2994         if (!buf)
2995                 goto out;
2996
2997         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2998         if (!ptr)
2999                 goto out;
3000
3001         strcpy(file->pathname, ptr);
3002
3003 copy_out:
3004         err = 0;
3005         if (copy_to_user(arg, file, sizeof(*file)))
3006                 err = -EFAULT;
3007 out:
3008         kfree(buf);
3009         kfree(file);
3010         return err;
3011 }
3012
3013 static int get_disk_info(mddev_t * mddev, void __user * arg)
3014 {
3015         mdu_disk_info_t info;
3016         unsigned int nr;
3017         mdk_rdev_t *rdev;
3018
3019         if (copy_from_user(&info, arg, sizeof(info)))
3020                 return -EFAULT;
3021
3022         nr = info.number;
3023
3024         rdev = find_rdev_nr(mddev, nr);
3025         if (rdev) {
3026                 info.major = MAJOR(rdev->bdev->bd_dev);
3027                 info.minor = MINOR(rdev->bdev->bd_dev);
3028                 info.raid_disk = rdev->raid_disk;
3029                 info.state = 0;
3030                 if (test_bit(Faulty, &rdev->flags))
3031                         info.state |= (1<<MD_DISK_FAULTY);
3032                 else if (test_bit(In_sync, &rdev->flags)) {
3033                         info.state |= (1<<MD_DISK_ACTIVE);
3034                         info.state |= (1<<MD_DISK_SYNC);
3035                 }
3036                 if (test_bit(WriteMostly, &rdev->flags))
3037                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3038         } else {
3039                 info.major = info.minor = 0;
3040                 info.raid_disk = -1;
3041                 info.state = (1<<MD_DISK_REMOVED);
3042         }
3043
3044         if (copy_to_user(arg, &info, sizeof(info)))
3045                 return -EFAULT;
3046
3047         return 0;
3048 }
3049
3050 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3051 {
3052         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3053         mdk_rdev_t *rdev;
3054         dev_t dev = MKDEV(info->major,info->minor);
3055
3056         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3057                 return -EOVERFLOW;
3058
3059         if (!mddev->raid_disks) {
3060                 int err;
3061                 /* expecting a device which has a superblock */
3062                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3063                 if (IS_ERR(rdev)) {
3064                         printk(KERN_WARNING
3065                                 "md: md_import_device returned %ld\n",
3066                                 PTR_ERR(rdev));
3067                         return PTR_ERR(rdev);
3068                 }
3069                 if (!list_empty(&mddev->disks)) {
3070                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3071                                                         mdk_rdev_t, same_set);
3072                         int err = super_types[mddev->major_version]
3073                                 .load_super(rdev, rdev0, mddev->minor_version);
3074                         if (err < 0) {
3075                                 printk(KERN_WARNING
3076                                         "md: %s has different UUID to %s\n",
3077                                         bdevname(rdev->bdev,b),
3078                                         bdevname(rdev0->bdev,b2));
3079                                 export_rdev(rdev);
3080                                 return -EINVAL;
3081                         }
3082                 }
3083                 err = bind_rdev_to_array(rdev, mddev);
3084                 if (err)
3085                         export_rdev(rdev);
3086                 return err;
3087         }
3088
3089         /*
3090          * add_new_disk can be used once the array is assembled
3091          * to add "hot spares".  They must already have a superblock
3092          * written
3093          */
3094         if (mddev->pers) {
3095                 int err;
3096                 if (!mddev->pers->hot_add_disk) {
3097                         printk(KERN_WARNING
3098                                 "%s: personality does not support diskops!\n",
3099                                mdname(mddev));
3100                         return -EINVAL;
3101                 }
3102                 if (mddev->persistent)
3103                         rdev = md_import_device(dev, mddev->major_version,
3104                                                 mddev->minor_version);
3105                 else
3106                         rdev = md_import_device(dev, -1, -1);
3107                 if (IS_ERR(rdev)) {
3108                         printk(KERN_WARNING
3109                                 "md: md_import_device returned %ld\n",
3110                                 PTR_ERR(rdev));
3111                         return PTR_ERR(rdev);
3112                 }
3113                 /* set save_raid_disk if appropriate */
3114                 if (!mddev->persistent) {
3115                         if (info->state & (1<<MD_DISK_SYNC)  &&
3116                             info->raid_disk < mddev->raid_disks)
3117                                 rdev->raid_disk = info->raid_disk;
3118                         else
3119                                 rdev->raid_disk = -1;
3120                 } else
3121                         super_types[mddev->major_version].
3122                                 validate_super(mddev, rdev);
3123                 rdev->saved_raid_disk = rdev->raid_disk;
3124
3125                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3126                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3127                         set_bit(WriteMostly, &rdev->flags);
3128
3129                 rdev->raid_disk = -1;
3130                 err = bind_rdev_to_array(rdev, mddev);
3131                 if (err)
3132                         export_rdev(rdev);
3133
3134                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3135                 md_wakeup_thread(mddev->thread);
3136                 return err;
3137         }
3138
3139         /* otherwise, add_new_disk is only allowed
3140          * for major_version==0 superblocks
3141          */
3142         if (mddev->major_version != 0) {
3143                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3144                        mdname(mddev));
3145                 return -EINVAL;
3146         }
3147
3148         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3149                 int err;
3150                 rdev = md_import_device (dev, -1, 0);
3151                 if (IS_ERR(rdev)) {
3152                         printk(KERN_WARNING
3153                                 "md: error, md_import_device() returned %ld\n",
3154                                 PTR_ERR(rdev));
3155                         return PTR_ERR(rdev);
3156                 }
3157                 rdev->desc_nr = info->number;
3158                 if (info->raid_disk < mddev->raid_disks)
3159                         rdev->raid_disk = info->raid_disk;
3160                 else
3161                         rdev->raid_disk = -1;
3162
3163                 rdev->flags = 0;
3164
3165                 if (rdev->raid_disk < mddev->raid_disks)
3166                         if (info->state & (1<<MD_DISK_SYNC))
3167                                 set_bit(In_sync, &rdev->flags);
3168
3169                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3170                         set_bit(WriteMostly, &rdev->flags);
3171
3172                 if (!mddev->persistent) {
3173                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3174                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3175                 } else
3176                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3177                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3178
3179                 err = bind_rdev_to_array(rdev, mddev);
3180                 if (err) {
3181                         export_rdev(rdev);
3182                         return err;
3183                 }
3184         }
3185
3186         return 0;
3187 }
3188
3189 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3190 {
3191         char b[BDEVNAME_SIZE];
3192         mdk_rdev_t *rdev;
3193
3194         if (!mddev->pers)
3195                 return -ENODEV;
3196
3197         rdev = find_rdev(mddev, dev);
3198         if (!rdev)
3199                 return -ENXIO;
3200
3201         if (rdev->raid_disk >= 0)
3202                 goto busy;
3203
3204         kick_rdev_from_array(rdev);
3205         md_update_sb(mddev);
3206         md_new_event(mddev);
3207
3208         return 0;
3209 busy:
3210         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3211                 bdevname(rdev->bdev,b), mdname(mddev));
3212         return -EBUSY;
3213 }
3214
3215 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3216 {
3217         char b[BDEVNAME_SIZE];
3218         int err;
3219         unsigned int size;
3220         mdk_rdev_t *rdev;
3221
3222         if (!mddev->pers)
3223                 return -ENODEV;
3224
3225         if (mddev->major_version != 0) {
3226                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3227                         " version-0 superblocks.\n",
3228                         mdname(mddev));
3229                 return -EINVAL;
3230         }
3231         if (!mddev->pers->hot_add_disk) {
3232                 printk(KERN_WARNING
3233                         "%s: personality does not support diskops!\n",
3234                         mdname(mddev));
3235                 return -EINVAL;
3236         }
3237
3238         rdev = md_import_device (dev, -1, 0);
3239         if (IS_ERR(rdev)) {
3240                 printk(KERN_WARNING
3241                         "md: error, md_import_device() returned %ld\n",
3242                         PTR_ERR(rdev));
3243                 return -EINVAL;
3244         }
3245
3246         if (mddev->persistent)
3247                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3248         else
3249                 rdev->sb_offset =
3250                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3251
3252         size = calc_dev_size(rdev, mddev->chunk_size);
3253         rdev->size = size;
3254
3255         if (test_bit(Faulty, &rdev->flags)) {
3256                 printk(KERN_WARNING
3257                         "md: can not hot-add faulty %s disk to %s!\n",
3258                         bdevname(rdev->bdev,b), mdname(mddev));
3259                 err = -EINVAL;
3260                 goto abort_export;
3261         }
3262         clear_bit(In_sync, &rdev->flags);
3263         rdev->desc_nr = -1;
3264         rdev->saved_raid_disk = -1;
3265         err = bind_rdev_to_array(rdev, mddev);
3266         if (err)
3267                 goto abort_export;
3268
3269         /*
3270          * The rest should better be atomic, we can have disk failures
3271          * noticed in interrupt contexts ...
3272          */
3273
3274         if (rdev->desc_nr == mddev->max_disks) {
3275                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3276                         mdname(mddev));
3277                 err = -EBUSY;
3278                 goto abort_unbind_export;
3279         }
3280
3281         rdev->raid_disk = -1;
3282
3283         md_update_sb(mddev);
3284
3285         /*
3286          * Kick recovery, maybe this spare has to be added to the
3287          * array immediately.
3288          */
3289         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3290         md_wakeup_thread(mddev->thread);
3291         md_new_event(mddev);
3292         return 0;
3293
3294 abort_unbind_export:
3295         unbind_rdev_from_array(rdev);
3296
3297 abort_export:
3298         export_rdev(rdev);
3299         return err;
3300 }
3301
3302 /* similar to deny_write_access, but accounts for our holding a reference
3303  * to the file ourselves */
3304 static int deny_bitmap_write_access(struct file * file)
3305 {
3306         struct inode *inode = file->f_mapping->host;
3307
3308         spin_lock(&inode->i_lock);
3309         if (atomic_read(&inode->i_writecount) > 1) {
3310                 spin_unlock(&inode->i_lock);
3311                 return -ETXTBSY;
3312         }
3313         atomic_set(&inode->i_writecount, -1);
3314         spin_unlock(&inode->i_lock);
3315
3316         return 0;
3317 }
3318
3319 static int set_bitmap_file(mddev_t *mddev, int fd)
3320 {
3321         int err;
3322
3323         if (mddev->pers) {
3324                 if (!mddev->pers->quiesce)
3325                         return -EBUSY;
3326                 if (mddev->recovery || mddev->sync_thread)
3327                         return -EBUSY;
3328                 /* we should be able to change the bitmap.. */
3329         }
3330
3331
3332         if (fd >= 0) {
3333                 if (mddev->bitmap)
3334                         return -EEXIST; /* cannot add when bitmap is present */
3335                 mddev->bitmap_file = fget(fd);
3336
3337                 if (mddev->bitmap_file == NULL) {
3338                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3339                                mdname(mddev));
3340                         return -EBADF;
3341                 }
3342
3343                 err = deny_bitmap_write_access(mddev->bitmap_file);
3344                 if (err) {
3345                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3346                                mdname(mddev));
3347                         fput(mddev->bitmap_file);
3348                         mddev->bitmap_file = NULL;
3349                         return err;
3350                 }
3351                 mddev->bitmap_offset = 0; /* file overrides offset */
3352         } else if (mddev->bitmap == NULL)
3353                 return -ENOENT; /* cannot remove what isn't there */
3354         err = 0;
3355         if (mddev->pers) {
3356                 mddev->pers->quiesce(mddev, 1);
3357                 if (fd >= 0)
3358                         err = bitmap_create(mddev);
3359                 if (fd < 0 || err)
3360                         bitmap_destroy(mddev);
3361                 mddev->pers->quiesce(mddev, 0);
3362         } else if (fd < 0) {
3363                 if (mddev->bitmap_file)
3364                         fput(mddev->bitmap_file);
3365                 mddev->bitmap_file = NULL;
3366         }
3367
3368         return err;
3369 }
3370
3371 /*
3372  * set_array_info is used two different ways
3373  * The original usage is when creating a new array.
3374  * In this usage, raid_disks is > 0 and it together with
3375  *  level, size, not_persistent,layout,chunksize determine the
3376  *  shape of the array.
3377  *  This will always create an array with a type-0.90.0 superblock.
3378  * The newer usage is when assembling an array.
3379  *  In this case raid_disks will be 0, and the major_version field is
3380  *  use to determine which style super-blocks are to be found on the devices.
3381  *  The minor and patch _version numbers are also kept incase the
3382  *  super_block handler wishes to interpret them.
3383  */
3384 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3385 {
3386
3387         if (info->raid_disks == 0) {
3388                 /* just setting version number for superblock loading */
3389                 if (info->major_version < 0 ||
3390                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3391                     super_types[info->major_version].name == NULL) {
3392                         /* maybe try to auto-load a module? */
3393                         printk(KERN_INFO
3394                                 "md: superblock version %d not known\n",
3395                                 info->major_version);
3396                         return -EINVAL;
3397                 }
3398                 mddev->major_version = info->major_version;
3399                 mddev->minor_version = info->minor_version;
3400                 mddev->patch_version = info->patch_version;
3401                 return 0;
3402         }
3403         mddev->major_version = MD_MAJOR_VERSION;
3404         mddev->minor_version = MD_MINOR_VERSION;
3405         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3406         mddev->ctime         = get_seconds();
3407
3408         mddev->level         = info->level;
3409         mddev->clevel[0]     = 0;
3410         mddev->size          = info->size;
3411         mddev->raid_disks    = info->raid_disks;
3412         /* don't set md_minor, it is determined by which /dev/md* was
3413          * openned
3414          */
3415         if (info->state & (1<<MD_SB_CLEAN))
3416                 mddev->recovery_cp = MaxSector;
3417         else
3418                 mddev->recovery_cp = 0;
3419         mddev->persistent    = ! info->not_persistent;
3420
3421         mddev->layout        = info->layout;
3422         mddev->chunk_size    = info->chunk_size;
3423
3424         mddev->max_disks     = MD_SB_DISKS;
3425
3426         mddev->sb_dirty      = 1;
3427
3428         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3429         mddev->bitmap_offset = 0;
3430
3431         /*
3432          * Generate a 128 bit UUID
3433          */
3434         get_random_bytes(mddev->uuid, 16);
3435
3436         return 0;
3437 }
3438
3439 static int update_size(mddev_t *mddev, unsigned long size)
3440 {
3441         mdk_rdev_t * rdev;
3442         int rv;
3443         struct list_head *tmp;
3444         int fit = (size == 0);
3445
3446         if (mddev->pers->resize == NULL)
3447                 return -EINVAL;
3448         /* The "size" is the amount of each device that is used.
3449          * This can only make sense for arrays with redundancy.
3450          * linear and raid0 always use whatever space is available
3451          * We can only consider changing the size if no resync
3452          * or reconstruction is happening, and if the new size
3453          * is acceptable. It must fit before the sb_offset or,
3454          * if that is <data_offset, it must fit before the
3455          * size of each device.
3456          * If size is zero, we find the largest size that fits.
3457          */
3458         if (mddev->sync_thread)
3459                 return -EBUSY;
3460         ITERATE_RDEV(mddev,rdev,tmp) {
3461                 sector_t avail;
3462                 if (rdev->sb_offset > rdev->data_offset)
3463                         avail = (rdev->sb_offset*2) - rdev->data_offset;
3464                 else
3465                         avail = get_capacity(rdev->bdev->bd_disk)
3466                                 - rdev->data_offset;
3467                 if (fit && (size == 0 || size > avail/2))
3468                         size = avail/2;
3469                 if (avail < ((sector_t)size << 1))
3470                         return -ENOSPC;
3471         }
3472         rv = mddev->pers->resize(mddev, (sector_t)size *2);
3473         if (!rv) {
3474                 struct block_device *bdev;
3475
3476                 bdev = bdget_disk(mddev->gendisk, 0);
3477                 if (bdev) {
3478                         mutex_lock(&bdev->bd_inode->i_mutex);
3479                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
3480                         mutex_unlock(&bdev->bd_inode->i_mutex);
3481                         bdput(bdev);
3482                 }
3483         }
3484         return rv;
3485 }
3486
3487 static int update_raid_disks(mddev_t *mddev, int raid_disks)
3488 {
3489         int rv;
3490         /* change the number of raid disks */
3491         if (mddev->pers->reshape == NULL)
3492                 return -EINVAL;
3493         if (raid_disks <= 0 ||
3494             raid_disks >= mddev->max_disks)
3495                 return -EINVAL;
3496         if (mddev->sync_thread)
3497                 return -EBUSY;
3498         rv = mddev->pers->reshape(mddev, raid_disks);
3499         return rv;
3500 }
3501
3502
3503 /*
3504  * update_array_info is used to change the configuration of an
3505  * on-line array.
3506  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3507  * fields in the info are checked against the array.
3508  * Any differences that cannot be handled will cause an error.
3509  * Normally, only one change can be managed at a time.
3510  */
3511 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3512 {
3513         int rv = 0;
3514         int cnt = 0;
3515         int state = 0;
3516
3517         /* calculate expected state,ignoring low bits */
3518         if (mddev->bitmap && mddev->bitmap_offset)
3519                 state |= (1 << MD_SB_BITMAP_PRESENT);
3520
3521         if (mddev->major_version != info->major_version ||
3522             mddev->minor_version != info->minor_version ||
3523 /*          mddev->patch_version != info->patch_version || */
3524             mddev->ctime         != info->ctime         ||
3525             mddev->level         != info->level         ||
3526 /*          mddev->layout        != info->layout        || */
3527             !mddev->persistent   != info->not_persistent||
3528             mddev->chunk_size    != info->chunk_size    ||
3529             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3530             ((state^info->state) & 0xfffffe00)
3531                 )
3532                 return -EINVAL;
3533         /* Check there is only one change */
3534         if (info->size >= 0 && mddev->size != info->size) cnt++;
3535         if (mddev->raid_disks != info->raid_disks) cnt++;
3536         if (mddev->layout != info->layout) cnt++;
3537         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3538         if (cnt == 0) return 0;
3539         if (cnt > 1) return -EINVAL;
3540
3541         if (mddev->layout != info->layout) {
3542                 /* Change layout
3543                  * we don't need to do anything at the md level, the
3544                  * personality will take care of it all.
3545                  */
3546                 if (mddev->pers->reconfig == NULL)
3547                         return -EINVAL;
3548                 else
3549                         return mddev->pers->reconfig(mddev, info->layout, -1);
3550         }
3551         if (info->size >= 0 && mddev->size != info->size)
3552                 rv = update_size(mddev, info->size);
3553
3554         if (mddev->raid_disks    != info->raid_disks)
3555                 rv = update_raid_disks(mddev, info->raid_disks);
3556
3557         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3558                 if (mddev->pers->quiesce == NULL)
3559                         return -EINVAL;
3560                 if (mddev->recovery || mddev->sync_thread)
3561                         return -EBUSY;
3562                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3563                         /* add the bitmap */
3564                         if (mddev->bitmap)
3565                                 return -EEXIST;
3566                         if (mddev->default_bitmap_offset == 0)
3567                                 return -EINVAL;
3568                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3569                         mddev->pers->quiesce(mddev, 1);
3570                         rv = bitmap_create(mddev);
3571                         if (rv)
3572                                 bitmap_destroy(mddev);
3573                         mddev->pers->quiesce(mddev, 0);
3574                 } else {
3575                         /* remove the bitmap */
3576                         if (!mddev->bitmap)
3577                                 return -ENOENT;
3578                         if (mddev->bitmap->file)
3579                                 return -EINVAL;
3580                         mddev->pers->quiesce(mddev, 1);
3581                         bitmap_destroy(mddev);
3582                         mddev->pers->quiesce(mddev, 0);
3583                         mddev->bitmap_offset = 0;
3584                 }
3585         }
3586         md_update_sb(mddev);
3587         return rv;
3588 }
3589
3590 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3591 {
3592         mdk_rdev_t *rdev;
3593
3594         if (mddev->pers == NULL)
3595                 return -ENODEV;
3596
3597         rdev = find_rdev(mddev, dev);
3598         if (!rdev)
3599                 return -ENODEV;
3600
3601         md_error(mddev, rdev);
3602         return 0;
3603 }
3604
3605 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3606 {
3607         mddev_t *mddev = bdev->bd_disk->private_data;
3608
3609         geo->heads = 2;
3610         geo->sectors = 4;
3611         geo->cylinders = get_capacity(mddev->gendisk) / 8;
3612         return 0;
3613 }
3614
3615 static int md_ioctl(struct inode *inode, struct file *file,
3616                         unsigned int cmd, unsigned long arg)
3617 {
3618         int err = 0;
3619         void __user *argp = (void __user *)arg;
3620         mddev_t *mddev = NULL;
3621
3622         if (!capable(CAP_SYS_ADMIN))
3623                 return -EACCES;
3624
3625         /*
3626          * Commands dealing with the RAID driver but not any
3627          * particular array:
3628          */
3629         switch (cmd)
3630         {
3631                 case RAID_VERSION:
3632                         err = get_version(argp);
3633                         goto done;
3634
3635                 case PRINT_RAID_DEBUG:
3636                         err = 0;
3637                         md_print_devices();
3638                         goto done;
3639
3640 #ifndef MODULE
3641                 case RAID_AUTORUN:
3642                         err = 0;
3643                         autostart_arrays(arg);
3644                         goto done;
3645 #endif
3646                 default:;
3647         }
3648
3649         /*
3650          * Commands creating/starting a new array:
3651          */
3652
3653         mddev = inode->i_bdev->bd_disk->private_data;
3654
3655         if (!mddev) {
3656                 BUG();
3657                 goto abort;
3658         }
3659
3660
3661         if (cmd == START_ARRAY) {
3662                 /* START_ARRAY doesn't need to lock the array as autostart_array
3663                  * does the locking, and it could even be a different array
3664                  */
3665                 static int cnt = 3;
3666                 if (cnt > 0 ) {
3667                         printk(KERN_WARNING
3668                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3669                                "START_ARRAY is removed in kernel 2.6.19 and above.\n",
3670                                current->comm, current->pid);
3671                         cnt--;
3672                 }
3673                 err = autostart_array(new_decode_dev(arg));
3674                 if (err) {
3675                         printk(KERN_WARNING "md: autostart failed!\n");
3676                         goto abort;
3677                 }
3678                 goto done;
3679         }
3680
3681         err = mddev_lock(mddev);
3682         if (err) {
3683                 printk(KERN_INFO
3684                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3685                         err, cmd);
3686                 goto abort;
3687         }
3688
3689         switch (cmd)
3690         {
3691                 case SET_ARRAY_INFO:
3692                         {
3693                                 mdu_array_info_t info;
3694                                 if (!arg)
3695                                         memset(&info, 0, sizeof(info));
3696                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3697                                         err = -EFAULT;
3698                                         goto abort_unlock;
3699                                 }
3700                                 if (mddev->pers) {
3701                                         err = update_array_info(mddev, &info);
3702                                         if (err) {
3703                                                 printk(KERN_WARNING "md: couldn't update"
3704                                                        " array info. %d\n", err);
3705                                                 goto abort_unlock;
3706                                         }
3707                                         goto done_unlock;
3708                                 }
3709                                 if (!list_empty(&mddev->disks)) {
3710                                         printk(KERN_WARNING
3711                                                "md: array %s already has disks!\n",
3712                                                mdname(mddev));
3713                                         err = -EBUSY;
3714                                         goto abort_unlock;
3715                                 }
3716                                 if (mddev->raid_disks) {
3717                                         printk(KERN_WARNING
3718                                                "md: array %s already initialised!\n",
3719                                                mdname(mddev));
3720                                         err = -EBUSY;
3721                                         goto abort_unlock;
3722                                 }
3723                                 err = set_array_info(mddev, &info);
3724                                 if (err) {
3725                                         printk(KERN_WARNING "md: couldn't set"
3726                                                " array info. %d\n", err);
3727                                         goto abort_unlock;
3728                                 }
3729                         }
3730                         goto done_unlock;
3731
3732                 default:;
3733         }
3734
3735         /*
3736          * Commands querying/configuring an existing array:
3737          */
3738         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3739          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3740         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3741                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3742                 err = -ENODEV;
3743                 goto abort_unlock;
3744         }
3745
3746         /*
3747          * Commands even a read-only array can execute:
3748          */
3749         switch (cmd)
3750         {
3751                 case GET_ARRAY_INFO:
3752                         err = get_array_info(mddev, argp);
3753                         goto done_unlock;
3754
3755                 case GET_BITMAP_FILE:
3756                         err = get_bitmap_file(mddev, argp);
3757                         goto done_unlock;
3758
3759                 case GET_DISK_INFO:
3760                         err = get_disk_info(mddev, argp);
3761                         goto done_unlock;
3762
3763                 case RESTART_ARRAY_RW:
3764                         err = restart_array(mddev);
3765                         goto done_unlock;
3766
3767                 case STOP_ARRAY:
3768                         err = do_md_stop (mddev, 0);
3769                         goto done_unlock;
3770
3771                 case STOP_ARRAY_RO:
3772                         err = do_md_stop (mddev, 1);
3773                         goto done_unlock;
3774
3775         /*
3776          * We have a problem here : there is no easy way to give a CHS
3777          * virtual geometry. We currently pretend that we have a 2 heads
3778          * 4 sectors (with a BIG number of cylinders...). This drives
3779          * dosfs just mad... ;-)
3780          */
3781         }
3782
3783         /*
3784          * The remaining ioctls are changing the state of the
3785          * superblock, so we do not allow them on read-only arrays.
3786          * However non-MD ioctls (e.g. get-size) will still come through
3787          * here and hit the 'default' below, so only disallow
3788          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3789          */
3790         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3791             mddev->ro && mddev->pers) {
3792                 if (mddev->ro == 2) {
3793                         mddev->ro = 0;
3794                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3795                 md_wakeup_thread(mddev->thread);
3796
3797                 } else {
3798                         err = -EROFS;
3799                         goto abort_unlock;
3800                 }
3801         }
3802
3803         switch (cmd)
3804         {
3805                 case ADD_NEW_DISK:
3806                 {
3807                         mdu_disk_info_t info;
3808                         if (copy_from_user(&info, argp, sizeof(info)))
3809                                 err = -EFAULT;
3810                         else
3811                                 err = add_new_disk(mddev, &info);
3812                         goto done_unlock;
3813                 }
3814
3815                 case HOT_REMOVE_DISK:
3816                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3817                         goto done_unlock;
3818
3819                 case HOT_ADD_DISK:
3820                         err = hot_add_disk(mddev, new_decode_dev(arg));
3821                         goto done_unlock;
3822
3823                 case SET_DISK_FAULTY:
3824                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3825                         goto done_unlock;
3826
3827                 case RUN_ARRAY:
3828                         err = do_md_run (mddev);
3829                         goto done_unlock;
3830
3831                 case SET_BITMAP_FILE:
3832                         err = set_bitmap_file(mddev, (int)arg);
3833                         goto done_unlock;
3834
3835                 default:
3836                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3837                                 printk(KERN_WARNING "md: %s(pid %d) used"
3838                                         " obsolete MD ioctl, upgrade your"
3839                                         " software to use new ictls.\n",
3840                                         current->comm, current->pid);
3841                         err = -EINVAL;
3842                         goto abort_unlock;
3843         }
3844
3845 done_unlock:
3846 abort_unlock:
3847         mddev_unlock(mddev);
3848
3849         return err;
3850 done:
3851         if (err)
3852                 MD_BUG();
3853 abort:
3854         return err;
3855 }
3856
3857 static int md_open(struct inode *inode, struct file *file)
3858 {
3859         /*
3860          * Succeed if we can lock the mddev, which confirms that
3861          * it isn't being stopped right now.
3862          */
3863         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3864         int err;
3865
3866         if ((err = mddev_lock(mddev)))
3867                 goto out;
3868
3869         err = 0;
3870         mddev_get(mddev);
3871         mddev_unlock(mddev);
3872
3873         check_disk_change(inode->i_bdev);
3874  out:
3875         return err;
3876 }
3877
3878 static int md_release(struct inode *inode, struct file * file)
3879 {
3880         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3881
3882         if (!mddev)
3883                 BUG();
3884         mddev_put(mddev);
3885
3886         return 0;
3887 }
3888
3889 static int md_media_changed(struct gendisk *disk)
3890 {
3891         mddev_t *mddev = disk->private_data;
3892
3893         return mddev->changed;
3894 }
3895
3896 static int md_revalidate(struct gendisk *disk)
3897 {
3898         mddev_t *mddev = disk->private_data;
3899
3900         mddev->changed = 0;
3901         return 0;
3902 }
3903 static struct block_device_operations md_fops =
3904 {
3905         .owner          = THIS_MODULE,
3906         .open           = md_open,
3907         .release        = md_release,
3908         .ioctl          = md_ioctl,
3909         .getgeo         = md_getgeo,
3910         .media_changed  = md_media_changed,
3911         .revalidate_disk= md_revalidate,
3912 };
3913
3914 static int md_thread(void * arg)
3915 {
3916         mdk_thread_t *thread = arg;
3917
3918         /*
3919          * md_thread is a 'system-thread', it's priority should be very
3920          * high. We avoid resource deadlocks individually in each
3921          * raid personality. (RAID5 does preallocation) We also use RR and
3922          * the very same RT priority as kswapd, thus we will never get
3923          * into a priority inversion deadlock.
3924          *
3925          * we definitely have to have equal or higher priority than
3926          * bdflush, otherwise bdflush will deadlock if there are too
3927          * many dirty RAID5 blocks.
3928          */
3929
3930         allow_signal(SIGKILL);
3931         while (!kthread_should_stop()) {
3932
3933                 /* We need to wait INTERRUPTIBLE so that
3934                  * we don't add to the load-average.
3935                  * That means we need to be sure no signals are
3936                  * pending
3937                  */
3938                 if (signal_pending(current))
3939                         flush_signals(current);
3940
3941                 wait_event_interruptible_timeout
3942                         (thread->wqueue,
3943                          test_bit(THREAD_WAKEUP, &thread->flags)
3944                          || kthread_should_stop(),
3945                          thread->timeout);
3946                 try_to_freeze();
3947
3948                 clear_bit(THREAD_WAKEUP, &thread->flags);
3949
3950                 thread->run(thread->mddev);
3951         }
3952
3953         return 0;
3954 }
3955
3956 void md_wakeup_thread(mdk_thread_t *thread)
3957 {
3958         if (thread) {
3959                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3960                 set_bit(THREAD_WAKEUP, &thread->flags);
3961                 wake_up(&thread->wqueue);
3962         }
3963 }
3964
3965 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3966                                  const char *name)
3967 {
3968         mdk_thread_t *thread;
3969
3970         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3971         if (!thread)
3972                 return NULL;
3973
3974         init_waitqueue_head(&thread->wqueue);
3975
3976         thread->run = run;
3977         thread->mddev = mddev;
3978         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3979         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3980         if (IS_ERR(thread->tsk)) {
3981                 kfree(thread);
3982                 return NULL;
3983         }
3984         return thread;
3985 }
3986
3987 void md_unregister_thread(mdk_thread_t *thread)
3988 {
3989         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3990
3991         kthread_stop(thread->tsk);
3992         kfree(thread);
3993 }
3994
3995 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3996 {
3997         if (!mddev) {
3998                 MD_BUG();
3999                 return;
4000         }
4001
4002         if (!rdev || test_bit(Faulty, &rdev->flags))
4003                 return;
4004 /*
4005         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4006                 mdname(mddev),
4007                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4008                 __builtin_return_address(0),__builtin_return_address(1),
4009                 __builtin_return_address(2),__builtin_return_address(3));
4010 */
4011         if (!mddev->pers->error_handler)
4012                 return;
4013         mddev->pers->error_handler(mddev,rdev);
4014         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4015         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4016         md_wakeup_thread(mddev->thread);
4017         md_new_event(mddev);
4018 }
4019
4020 /* seq_file implementation /proc/mdstat */
4021
4022 static void status_unused(struct seq_file *seq)
4023 {
4024         int i = 0;
4025         mdk_rdev_t *rdev;
4026         struct list_head *tmp;
4027
4028         seq_printf(seq, "unused devices: ");
4029
4030         ITERATE_RDEV_PENDING(rdev,tmp) {
4031                 char b[BDEVNAME_SIZE];
4032                 i++;
4033                 seq_printf(seq, "%s ",
4034                               bdevname(rdev->bdev,b));
4035         }
4036         if (!i)
4037                 seq_printf(seq, "<none>");
4038
4039         seq_printf(seq, "\n");
4040 }
4041
4042
4043 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4044 {
4045         unsigned long max_blocks, resync, res, dt, db, rt;
4046
4047         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4048
4049         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4050                 max_blocks = mddev->resync_max_sectors >> 1;
4051         else
4052                 max_blocks = mddev->size;
4053
4054         /*
4055          * Should not happen.
4056          */
4057         if (!max_blocks) {
4058                 MD_BUG();
4059                 return;
4060         }
4061         res = (resync/1024)*1000/(max_blocks/1024 + 1);
4062         {
4063                 int i, x = res/50, y = 20-x;
4064                 seq_printf(seq, "[");
4065                 for (i = 0; i < x; i++)
4066                         seq_printf(seq, "=");
4067                 seq_printf(seq, ">");
4068                 for (i = 0; i < y; i++)
4069                         seq_printf(seq, ".");
4070                 seq_printf(seq, "] ");
4071         }
4072         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
4073                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4074                        "resync" : "recovery"),
4075                       res/10, res % 10, resync, max_blocks);
4076
4077         /*
4078          * We do not want to overflow, so the order of operands and
4079          * the * 100 / 100 trick are important. We do a +1 to be
4080          * safe against division by zero. We only estimate anyway.
4081          *
4082          * dt: time from mark until now
4083          * db: blocks written from mark until now
4084          * rt: remaining time
4085          */
4086         dt = ((jiffies - mddev->resync_mark) / HZ);
4087         if (!dt) dt++;
4088         db = resync - (mddev->resync_mark_cnt/2);
4089         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
4090
4091         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4092
4093         seq_printf(seq, " speed=%ldK/sec", db/dt);
4094 }
4095
4096 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4097 {
4098         struct list_head *tmp;
4099         loff_t l = *pos;
4100         mddev_t *mddev;
4101
4102         if (l >= 0x10000)
4103                 return NULL;
4104         if (!l--)
4105                 /* header */
4106                 return (void*)1;
4107
4108         spin_lock(&all_mddevs_lock);
4109         list_for_each(tmp,&all_mddevs)
4110                 if (!l--) {
4111                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4112                         mddev_get(mddev);
4113                         spin_unlock(&all_mddevs_lock);
4114                         return mddev;
4115                 }
4116         spin_unlock(&all_mddevs_lock);
4117         if (!l--)
4118                 return (void*)2;/* tail */
4119         return NULL;
4120 }
4121
4122 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4123 {
4124         struct list_head *tmp;
4125         mddev_t *next_mddev, *mddev = v;
4126
4127         ++*pos;
4128         if (v == (void*)2)
4129                 return NULL;
4130
4131         spin_lock(&all_mddevs_lock);
4132         if (v == (void*)1)
4133                 tmp = all_mddevs.next;
4134         else
4135                 tmp = mddev->all_mddevs.next;
4136         if (tmp != &all_mddevs)
4137                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4138         else {
4139                 next_mddev = (void*)2;
4140                 *pos = 0x10000;
4141         }
4142         spin_unlock(&all_mddevs_lock);
4143
4144         if (v != (void*)1)
4145                 mddev_put(mddev);
4146         return next_mddev;
4147
4148 }
4149
4150 static void md_seq_stop(struct seq_file *seq, void *v)
4151 {
4152         mddev_t *mddev = v;
4153
4154         if (mddev && v != (void*)1 && v != (void*)2)
4155                 mddev_put(mddev);
4156 }
4157
4158 struct mdstat_info {
4159         int event;
4160 };
4161
4162 static int md_seq_show(struct seq_file *seq, void *v)
4163 {
4164         mddev_t *mddev = v;
4165         sector_t size;
4166         struct list_head *tmp2;
4167         mdk_rdev_t *rdev;
4168         struct mdstat_info *mi = seq->private;
4169         struct bitmap *bitmap;
4170
4171         if (v == (void*)1) {
4172                 struct mdk_personality *pers;
4173                 seq_printf(seq, "Personalities : ");
4174                 spin_lock(&pers_lock);
4175                 list_for_each_entry(pers, &pers_list, list)
4176                         seq_printf(seq, "[%s] ", pers->name);
4177
4178                 spin_unlock(&pers_lock);
4179                 seq_printf(seq, "\n");
4180                 mi->event = atomic_read(&md_event_count);
4181                 return 0;
4182         }
4183         if (v == (void*)2) {
4184                 status_unused(seq);
4185                 return 0;
4186         }
4187
4188         if (mddev_lock(mddev)!=0)
4189                 return -EINTR;
4190         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4191                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4192                                                 mddev->pers ? "" : "in");
4193                 if (mddev->pers) {
4194                         if (mddev->ro==1)
4195                                 seq_printf(seq, " (read-only)");
4196                         if (mddev->ro==2)
4197                                 seq_printf(seq, "(auto-read-only)");
4198                         seq_printf(seq, " %s", mddev->pers->name);
4199                 }
4200
4201                 size = 0;
4202                 ITERATE_RDEV(mddev,rdev,tmp2) {
4203                         char b[BDEVNAME_SIZE];
4204                         seq_printf(seq, " %s[%d]",
4205                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4206                         if (test_bit(WriteMostly, &rdev->flags))
4207                                 seq_printf(seq, "(W)");
4208                         if (test_bit(Faulty, &rdev->flags)) {
4209                                 seq_printf(seq, "(F)");
4210                                 continue;
4211                         } else if (rdev->raid_disk < 0)
4212                                 seq_printf(seq, "(S)"); /* spare */
4213                         size += rdev->size;
4214                 }
4215
4216                 if (!list_empty(&mddev->disks)) {
4217                         if (mddev->pers)
4218                                 seq_printf(seq, "\n      %llu blocks",
4219                                         (unsigned long long)mddev->array_size);
4220                         else
4221                                 seq_printf(seq, "\n      %llu blocks",
4222                                         (unsigned long long)size);
4223                 }
4224                 if (mddev->persistent) {
4225                         if (mddev->major_version != 0 ||
4226                             mddev->minor_version != 90) {
4227                                 seq_printf(seq," super %d.%d",
4228                                            mddev->major_version,
4229                                            mddev->minor_version);
4230                         }
4231                 } else
4232                         seq_printf(seq, " super non-persistent");
4233
4234                 if (mddev->pers) {
4235                         mddev->pers->status (seq, mddev);
4236                         seq_printf(seq, "\n      ");
4237                         if (mddev->pers->sync_request) {
4238                                 if (mddev->curr_resync > 2) {
4239                                         status_resync (seq, mddev);
4240                                         seq_printf(seq, "\n      ");
4241                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4242                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4243                                 else if (mddev->recovery_cp < MaxSector)
4244                                         seq_printf(seq, "\tresync=PENDING\n      ");
4245                         }
4246                 } else
4247                         seq_printf(seq, "\n       ");
4248
4249                 if ((bitmap = mddev->bitmap)) {
4250                         unsigned long chunk_kb;
4251                         unsigned long flags;
4252                         spin_lock_irqsave(&bitmap->lock, flags);
4253                         chunk_kb = bitmap->chunksize >> 10;
4254                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4255                                 "%lu%s chunk",
4256                                 bitmap->pages - bitmap->missing_pages,
4257                                 bitmap->pages,
4258                                 (bitmap->pages - bitmap->missing_pages)
4259                                         << (PAGE_SHIFT - 10),
4260                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4261                                 chunk_kb ? "KB" : "B");
4262                         if (bitmap->file) {
4263                                 seq_printf(seq, ", file: ");
4264                                 seq_path(seq, bitmap->file->f_vfsmnt,
4265                                          bitmap->file->f_dentry," \t\n");
4266                         }
4267
4268                         seq_printf(seq, "\n");
4269                         spin_unlock_irqrestore(&bitmap->lock, flags);
4270                 }
4271
4272                 seq_printf(seq, "\n");
4273         }
4274         mddev_unlock(mddev);
4275
4276         return 0;
4277 }
4278
4279 static struct seq_operations md_seq_ops = {
4280         .start  = md_seq_start,
4281         .next   = md_seq_next,
4282         .stop   = md_seq_stop,
4283         .show   = md_seq_show,
4284 };
4285
4286 static int md_seq_open(struct inode *inode, struct file *file)
4287 {
4288         int error;
4289         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4290         if (mi == NULL)
4291                 return -ENOMEM;
4292
4293         error = seq_open(file, &md_seq_ops);
4294         if (error)
4295                 kfree(mi);
4296         else {
4297                 struct seq_file *p = file->private_data;
4298                 p->private = mi;
4299                 mi->event = atomic_read(&md_event_count);
4300         }
4301         return error;
4302 }
4303
4304 static int md_seq_release(struct inode *inode, struct file *file)
4305 {
4306         struct seq_file *m = file->private_data;
4307         struct mdstat_info *mi = m->private;
4308         m->private = NULL;
4309         kfree(mi);
4310         return seq_release(inode, file);
4311 }
4312
4313 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4314 {
4315         struct seq_file *m = filp->private_data;
4316         struct mdstat_info *mi = m->private;
4317         int mask;
4318
4319         poll_wait(filp, &md_event_waiters, wait);
4320
4321         /* always allow read */
4322         mask = POLLIN | POLLRDNORM;
4323
4324         if (mi->event != atomic_read(&md_event_count))
4325                 mask |= POLLERR | POLLPRI;
4326         return mask;
4327 }
4328
4329 static struct file_operations md_seq_fops = {
4330         .owner          = THIS_MODULE,
4331         .open           = md_seq_open,
4332         .read           = seq_read,
4333         .llseek         = seq_lseek,
4334         .release        = md_seq_release,
4335         .poll           = mdstat_poll,
4336 };
4337
4338 int register_md_personality(struct mdk_personality *p)
4339 {
4340         spin_lock(&pers_lock);
4341         list_add_tail(&p->list, &pers_list);
4342         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4343         spin_unlock(&pers_lock);
4344         return 0;
4345 }
4346
4347 int unregister_md_personality(struct mdk_personality *p)
4348 {
4349         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4350         spin_lock(&pers_lock);
4351         list_del_init(&p->list);
4352         spin_unlock(&pers_lock);
4353         return 0;
4354 }
4355
4356 static int is_mddev_idle(mddev_t *mddev)
4357 {
4358         mdk_rdev_t * rdev;
4359         struct list_head *tmp;
4360         int idle;
4361         unsigned long curr_events;
4362
4363         idle = 1;
4364         ITERATE_RDEV(mddev,rdev,tmp) {
4365                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4366                 curr_events = disk_stat_read(disk, sectors[0]) +
4367                                 disk_stat_read(disk, sectors[1]) -
4368                                 atomic_read(&disk->sync_io);
4369                 /* The difference between curr_events and last_events
4370                  * will be affected by any new non-sync IO (making
4371                  * curr_events bigger) and any difference in the amount of
4372                  * in-flight syncio (making current_events bigger or smaller)
4373                  * The amount in-flight is currently limited to
4374                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4375                  * which is at most 4096 sectors.
4376                  * These numbers are fairly fragile and should be made
4377                  * more robust, probably by enforcing the
4378                  * 'window size' that md_do_sync sort-of uses.
4379                  *
4380                  * Note: the following is an unsigned comparison.
4381                  */
4382                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4383                         rdev->last_events = curr_events;
4384                         idle = 0;
4385                 }
4386         }
4387         return idle;
4388 }
4389
4390 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4391 {
4392         /* another "blocks" (512byte) blocks have been synced */
4393         atomic_sub(blocks, &mddev->recovery_active);
4394         wake_up(&mddev->recovery_wait);
4395         if (!ok) {
4396                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4397                 md_wakeup_thread(mddev->thread);
4398                 // stop recovery, signal do_sync ....
4399         }
4400 }
4401
4402
4403 /* md_write_start(mddev, bi)
4404  * If we need to update some array metadata (e.g. 'active' flag
4405  * in superblock) before writing, schedule a superblock update
4406  * and wait for it to complete.
4407  */
4408 void md_write_start(mddev_t *mddev, struct bio *bi)
4409 {
4410         if (bio_data_dir(bi) != WRITE)
4411                 return;
4412
4413         BUG_ON(mddev->ro == 1);
4414         if (mddev->ro == 2) {
4415                 /* need to switch to read/write */
4416                 mddev->ro = 0;
4417                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4418                 md_wakeup_thread(mddev->thread);
4419         }
4420         atomic_inc(&mddev->writes_pending);
4421         if (mddev->in_sync) {
4422                 spin_lock_irq(&mddev->write_lock);
4423                 if (mddev->in_sync) {
4424                         mddev->in_sync = 0;
4425                         mddev->sb_dirty = 1;
4426                         md_wakeup_thread(mddev->thread);
4427                 }
4428                 spin_unlock_irq(&mddev->write_lock);
4429         }
4430         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4431 }
4432
4433 void md_write_end(mddev_t *mddev)
4434 {
4435         if (atomic_dec_and_test(&mddev->writes_pending)) {
4436                 if (mddev->safemode == 2)
4437                         md_wakeup_thread(mddev->thread);
4438                 else
4439                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4440         }
4441 }
4442
4443 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4444
4445 #define SYNC_MARKS      10
4446 #define SYNC_MARK_STEP  (3*HZ)
4447 static void md_do_sync(mddev_t *mddev)
4448 {
4449         mddev_t *mddev2;
4450         unsigned int currspeed = 0,
4451                  window;
4452         sector_t max_sectors,j, io_sectors;
4453         unsigned long mark[SYNC_MARKS];
4454         sector_t mark_cnt[SYNC_MARKS];
4455         int last_mark,m;
4456         struct list_head *tmp;
4457         sector_t last_check;
4458         int skipped = 0;
4459
4460         /* just incase thread restarts... */
4461         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4462                 return;
4463
4464         /* we overload curr_resync somewhat here.
4465          * 0 == not engaged in resync at all
4466          * 2 == checking that there is no conflict with another sync
4467          * 1 == like 2, but have yielded to allow conflicting resync to
4468          *              commense
4469          * other == active in resync - this many blocks
4470          *
4471          * Before starting a resync we must have set curr_resync to
4472          * 2, and then checked that every "conflicting" array has curr_resync
4473          * less than ours.  When we find one that is the same or higher
4474          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4475          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4476          * This will mean we have to start checking from the beginning again.
4477          *
4478          */
4479
4480         do {
4481                 mddev->curr_resync = 2;
4482
4483         try_again:
4484                 if (kthread_should_stop()) {
4485                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4486                         goto skip;
4487                 }
4488                 ITERATE_MDDEV(mddev2,tmp) {
4489                         if (mddev2 == mddev)
4490                                 continue;
4491                         if (mddev2->curr_resync &&
4492                             match_mddev_units(mddev,mddev2)) {
4493                                 DEFINE_WAIT(wq);
4494                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4495                                         /* arbitrarily yield */
4496                                         mddev->curr_resync = 1;
4497                                         wake_up(&resync_wait);
4498                                 }
4499                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4500                                         /* no need to wait here, we can wait the next
4501                                          * time 'round when curr_resync == 2
4502                                          */
4503                                         continue;
4504                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4505                                 if (!kthread_should_stop() &&
4506                                     mddev2->curr_resync >= mddev->curr_resync) {
4507                                         printk(KERN_INFO "md: delaying resync of %s"
4508                                                " until %s has finished resync (they"
4509                                                " share one or more physical units)\n",
4510                                                mdname(mddev), mdname(mddev2));
4511                                         mddev_put(mddev2);
4512                                         schedule();
4513                                         finish_wait(&resync_wait, &wq);
4514                                         goto try_again;
4515                                 }
4516                                 finish_wait(&resync_wait, &wq);
4517                         }
4518                 }
4519         } while (mddev->curr_resync < 2);
4520
4521         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4522                 /* resync follows the size requested by the personality,
4523                  * which defaults to physical size, but can be virtual size
4524                  */
4525                 max_sectors = mddev->resync_max_sectors;
4526                 mddev->resync_mismatches = 0;
4527         } else
4528                 /* recovery follows the physical size of devices */
4529                 max_sectors = mddev->size << 1;
4530
4531         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4532         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4533                 " %d KB/sec/disc.\n", speed_min(mddev));
4534         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4535                "(but not more than %d KB/sec) for reconstruction.\n",
4536                speed_max(mddev));
4537
4538         is_mddev_idle(mddev); /* this also initializes IO event counters */
4539         /* we don't use the checkpoint if there's a bitmap */
4540         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4541             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4542                 j = mddev->recovery_cp;
4543         else
4544                 j = 0;
4545         io_sectors = 0;
4546         for (m = 0; m < SYNC_MARKS; m++) {
4547                 mark[m] = jiffies;
4548                 mark_cnt[m] = io_sectors;
4549         }
4550         last_mark = 0;
4551         mddev->resync_mark = mark[last_mark];
4552         mddev->resync_mark_cnt = mark_cnt[last_mark];
4553
4554         /*
4555          * Tune reconstruction:
4556          */
4557         window = 32*(PAGE_SIZE/512);
4558         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4559                 window/2,(unsigned long long) max_sectors/2);
4560
4561         atomic_set(&mddev->recovery_active, 0);
4562         init_waitqueue_head(&mddev->recovery_wait);
4563         last_check = 0;
4564
4565         if (j>2) {
4566                 printk(KERN_INFO
4567                         "md: resuming recovery of %s from checkpoint.\n",
4568                         mdname(mddev));
4569                 mddev->curr_resync = j;
4570         }
4571
4572         while (j < max_sectors) {
4573                 sector_t sectors;
4574
4575                 skipped = 0;
4576                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4577                                             currspeed < speed_min(mddev));
4578                 if (sectors == 0) {
4579                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4580                         goto out;
4581                 }
4582
4583                 if (!skipped) { /* actual IO requested */
4584                         io_sectors += sectors;
4585                         atomic_add(sectors, &mddev->recovery_active);
4586                 }
4587
4588                 j += sectors;
4589                 if (j>1) mddev->curr_resync = j;
4590                 if (last_check == 0)
4591                         /* this is the earliers that rebuilt will be
4592                          * visible in /proc/mdstat
4593                          */
4594                         md_new_event(mddev);
4595
4596                 if (last_check + window > io_sectors || j == max_sectors)
4597                         continue;
4598
4599                 last_check = io_sectors;
4600
4601                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4602                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4603                         break;
4604
4605         repeat:
4606                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4607                         /* step marks */
4608                         int next = (last_mark+1) % SYNC_MARKS;
4609
4610                         mddev->resync_mark = mark[next];
4611                         mddev->resync_mark_cnt = mark_cnt[next];
4612                         mark[next] = jiffies;
4613                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4614                         last_mark = next;
4615                 }
4616
4617
4618                 if (kthread_should_stop()) {
4619                         /*
4620                          * got a signal, exit.
4621                          */
4622                         printk(KERN_INFO
4623                                 "md: md_do_sync() got signal ... exiting\n");
4624                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4625                         goto out;
4626                 }
4627
4628                 /*
4629                  * this loop exits only if either when we are slower than
4630                  * the 'hard' speed limit, or the system was IO-idle for
4631                  * a jiffy.
4632                  * the system might be non-idle CPU-wise, but we only care
4633                  * about not overloading the IO subsystem. (things like an
4634                  * e2fsck being done on the RAID array should execute fast)
4635                  */
4636                 mddev->queue->unplug_fn(mddev->queue);
4637                 cond_resched();
4638
4639                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4640                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4641
4642                 if (currspeed > speed_min(mddev)) {
4643                         if ((currspeed > speed_max(mddev)) ||
4644                                         !is_mddev_idle(mddev)) {
4645                                 msleep(500);
4646                                 goto repeat;
4647                         }
4648                 }
4649         }
4650         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4651         /*
4652          * this also signals 'finished resyncing' to md_stop
4653          */
4654  out:
4655         mddev->queue->unplug_fn(mddev->queue);
4656
4657         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4658
4659         /* tell personality that we are finished */
4660         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4661
4662         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4663             mddev->curr_resync > 2 &&
4664             mddev->curr_resync >= mddev->recovery_cp) {
4665                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4666                         printk(KERN_INFO
4667                                 "md: checkpointing recovery of %s.\n",
4668                                 mdname(mddev));
4669                         mddev->recovery_cp = mddev->curr_resync;
4670                 } else
4671                         mddev->recovery_cp = MaxSector;
4672         }
4673
4674  skip:
4675         mddev->curr_resync = 0;
4676         wake_up(&resync_wait);
4677         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4678         md_wakeup_thread(mddev->thread);
4679 }
4680
4681
4682 /*
4683  * This routine is regularly called by all per-raid-array threads to
4684  * deal with generic issues like resync and super-block update.
4685  * Raid personalities that don't have a thread (linear/raid0) do not
4686  * need this as they never do any recovery or update the superblock.
4687  *
4688  * It does not do any resync itself, but rather "forks" off other threads
4689  * to do that as needed.
4690  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4691  * "->recovery" and create a thread at ->sync_thread.
4692  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4693  * and wakeups up this thread which will reap the thread and finish up.
4694  * This thread also removes any faulty devices (with nr_pending == 0).
4695  *
4696  * The overall approach is:
4697  *  1/ if the superblock needs updating, update it.
4698  *  2/ If a recovery thread is running, don't do anything else.
4699  *  3/ If recovery has finished, clean up, possibly marking spares active.
4700  *  4/ If there are any faulty devices, remove them.
4701  *  5/ If array is degraded, try to add spares devices
4702  *  6/ If array has spares or is not in-sync, start a resync thread.
4703  */
4704 void md_check_recovery(mddev_t *mddev)
4705 {
4706         mdk_rdev_t *rdev;
4707         struct list_head *rtmp;
4708
4709
4710         if (mddev->bitmap)
4711                 bitmap_daemon_work(mddev->bitmap);
4712
4713         if (mddev->ro)
4714                 return;
4715
4716         if (signal_pending(current)) {
4717                 if (mddev->pers->sync_request) {
4718                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4719                                mdname(mddev));
4720                         mddev->safemode = 2;
4721                 }
4722                 flush_signals(current);
4723         }
4724
4725         if ( ! (
4726                 mddev->sb_dirty ||
4727                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4728                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4729                 (mddev->safemode == 1) ||
4730                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4731                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4732                 ))
4733                 return;
4734
4735         if (mddev_trylock(mddev)==0) {
4736                 int spares =0;
4737
4738                 spin_lock_irq(&mddev->write_lock);
4739                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4740                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4741                         mddev->in_sync = 1;
4742                         mddev->sb_dirty = 1;
4743                 }
4744                 if (mddev->safemode == 1)
4745                         mddev->safemode = 0;
4746                 spin_unlock_irq(&mddev->write_lock);
4747
4748                 if (mddev->sb_dirty)
4749                         md_update_sb(mddev);
4750
4751
4752                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4753                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4754                         /* resync/recovery still happening */
4755                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4756                         goto unlock;
4757                 }
4758                 if (mddev->sync_thread) {
4759                         /* resync has finished, collect result */
4760                         md_unregister_thread(mddev->sync_thread);
4761                         mddev->sync_thread = NULL;
4762                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4763                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4764                                 /* success...*/
4765                                 /* activate any spares */
4766                                 mddev->pers->spare_active(mddev);
4767                         }
4768                         md_update_sb(mddev);
4769
4770                         /* if array is no-longer degraded, then any saved_raid_disk
4771                          * information must be scrapped
4772                          */
4773                         if (!mddev->degraded)
4774                                 ITERATE_RDEV(mddev,rdev,rtmp)
4775                                         rdev->saved_raid_disk = -1;
4776
4777                         mddev->recovery = 0;
4778                         /* flag recovery needed just to double check */
4779                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4780                         md_new_event(mddev);
4781                         goto unlock;
4782                 }
4783                 /* Clear some bits that don't mean anything, but
4784                  * might be left set
4785                  */
4786                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4787                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4788                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4789                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4790
4791                 /* no recovery is running.
4792                  * remove any failed drives, then
4793                  * add spares if possible.
4794                  * Spare are also removed and re-added, to allow
4795                  * the personality to fail the re-add.
4796                  */
4797                 ITERATE_RDEV(mddev,rdev,rtmp)
4798                         if (rdev->raid_disk >= 0 &&
4799                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4800                             atomic_read(&rdev->nr_pending)==0) {
4801                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4802                                         char nm[20];
4803                                         sprintf(nm,"rd%d", rdev->raid_disk);
4804                                         sysfs_remove_link(&mddev->kobj, nm);
4805                                         rdev->raid_disk = -1;
4806                                 }
4807                         }
4808
4809                 if (mddev->degraded) {
4810                         ITERATE_RDEV(mddev,rdev,rtmp)
4811                                 if (rdev->raid_disk < 0
4812                                     && !test_bit(Faulty, &rdev->flags)) {
4813                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4814                                                 char nm[20];
4815                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4816                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4817                                                 spares++;
4818                                                 md_new_event(mddev);
4819                                         } else
4820                                                 break;
4821                                 }
4822                 }
4823
4824                 if (spares) {
4825                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4826                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4827                 } else if (mddev->recovery_cp < MaxSector) {
4828                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4829                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4830                         /* nothing to be done ... */
4831                         goto unlock;
4832
4833                 if (mddev->pers->sync_request) {
4834                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4835                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4836                                 /* We are adding a device or devices to an array
4837                                  * which has the bitmap stored on all devices.
4838                                  * So make sure all bitmap pages get written
4839                                  */
4840                                 bitmap_write_all(mddev->bitmap);
4841                         }
4842                         mddev->sync_thread = md_register_thread(md_do_sync,
4843                                                                 mddev,
4844                                                                 "%s_resync");
4845                         if (!mddev->sync_thread) {
4846                                 printk(KERN_ERR "%s: could not start resync"
4847                                         " thread...\n",
4848                                         mdname(mddev));
4849                                 /* leave the spares where they are, it shouldn't hurt */
4850                                 mddev->recovery = 0;
4851                         } else
4852                                 md_wakeup_thread(mddev->sync_thread);
4853                         md_new_event(mddev);
4854                 }
4855         unlock:
4856                 mddev_unlock(mddev);
4857         }
4858 }
4859
4860 static int md_notify_reboot(struct notifier_block *this,
4861                             unsigned long code, void *x)
4862 {
4863         struct list_head *tmp;
4864         mddev_t *mddev;
4865
4866         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4867
4868                 printk(KERN_INFO "md: stopping all md devices.\n");
4869
4870                 ITERATE_MDDEV(mddev,tmp)
4871                         if (mddev_trylock(mddev)==0)
4872                                 do_md_stop (mddev, 1);
4873                 /*
4874                  * certain more exotic SCSI devices are known to be
4875                  * volatile wrt too early system reboots. While the
4876                  * right place to handle this issue is the given
4877                  * driver, we do want to have a safe RAID driver ...
4878                  */
4879                 mdelay(1000*1);
4880         }
4881         return NOTIFY_DONE;
4882 }
4883
4884 static struct notifier_block md_notifier = {
4885         .notifier_call  = md_notify_reboot,
4886         .next           = NULL,
4887         .priority       = INT_MAX, /* before any real devices */
4888 };
4889
4890 static void md_geninit(void)
4891 {
4892         struct proc_dir_entry *p;
4893
4894         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4895
4896         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4897         if (p)
4898                 p->proc_fops = &md_seq_fops;
4899 }
4900
4901 static int __init md_init(void)
4902 {
4903         int minor;
4904
4905         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4906                         " MD_SB_DISKS=%d\n",
4907                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4908                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4909         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4910                         BITMAP_MINOR);
4911
4912         if (register_blkdev(MAJOR_NR, "md"))
4913                 return -1;
4914         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4915                 unregister_blkdev(MAJOR_NR, "md");
4916                 return -1;
4917         }
4918         devfs_mk_dir("md");
4919         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4920                                 md_probe, NULL, NULL);
4921         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4922                             md_probe, NULL, NULL);
4923
4924         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4925                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4926                                 S_IFBLK|S_IRUSR|S_IWUSR,
4927                                 "md/%d", minor);
4928
4929         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4930                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4931                               S_IFBLK|S_IRUSR|S_IWUSR,
4932                               "md/mdp%d", minor);
4933
4934
4935         register_reboot_notifier(&md_notifier);
4936         raid_table_header = register_sysctl_table(raid_root_table, 1);
4937
4938         md_geninit();
4939         return (0);
4940 }
4941
4942
4943 #ifndef MODULE
4944
4945 /*
4946  * Searches all registered partitions for autorun RAID arrays
4947  * at boot time.
4948  */
4949 static dev_t detected_devices[128];
4950 static int dev_cnt;
4951
4952 void md_autodetect_dev(dev_t dev)
4953 {
4954         if (dev_cnt >= 0 && dev_cnt < 127)
4955                 detected_devices[dev_cnt++] = dev;
4956 }
4957
4958
4959 static void autostart_arrays(int part)
4960 {
4961         mdk_rdev_t *rdev;
4962         int i;
4963
4964         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4965
4966         for (i = 0; i < dev_cnt; i++) {
4967                 dev_t dev = detected_devices[i];
4968
4969                 rdev = md_import_device(dev,0, 0);
4970                 if (IS_ERR(rdev))
4971                         continue;
4972
4973                 if (test_bit(Faulty, &rdev->flags)) {
4974                         MD_BUG();
4975                         continue;
4976                 }
4977                 list_add(&rdev->same_set, &pending_raid_disks);
4978         }
4979         dev_cnt = 0;
4980
4981         autorun_devices(part);
4982 }
4983
4984 #endif
4985
4986 static __exit void md_exit(void)
4987 {
4988         mddev_t *mddev;
4989         struct list_head *tmp;
4990         int i;
4991         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4992         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4993         for (i=0; i < MAX_MD_DEVS; i++)
4994                 devfs_remove("md/%d", i);
4995         for (i=0; i < MAX_MD_DEVS; i++)
4996                 devfs_remove("md/d%d", i);
4997
4998         devfs_remove("md");
4999
5000         unregister_blkdev(MAJOR_NR,"md");
5001         unregister_blkdev(mdp_major, "mdp");
5002         unregister_reboot_notifier(&md_notifier);
5003         unregister_sysctl_table(raid_table_header);
5004         remove_proc_entry("mdstat", NULL);
5005         ITERATE_MDDEV(mddev,tmp) {
5006                 struct gendisk *disk = mddev->gendisk;
5007                 if (!disk)
5008                         continue;
5009                 export_array(mddev);
5010                 del_gendisk(disk);
5011                 put_disk(disk);
5012                 mddev->gendisk = NULL;
5013                 mddev_put(mddev);
5014         }
5015 }
5016
5017 module_init(md_init)
5018 module_exit(md_exit)
5019
5020 static int get_ro(char *buffer, struct kernel_param *kp)
5021 {
5022         return sprintf(buffer, "%d", start_readonly);
5023 }
5024 static int set_ro(const char *val, struct kernel_param *kp)
5025 {
5026         char *e;
5027         int num = simple_strtoul(val, &e, 10);
5028         if (*val && (*e == '\0' || *e == '\n')) {
5029                 start_readonly = num;
5030                 return 0;
5031         }
5032         return -EINVAL;
5033 }
5034
5035 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
5036 module_param(start_dirty_degraded, int, 0644);
5037
5038
5039 EXPORT_SYMBOL(register_md_personality);
5040 EXPORT_SYMBOL(unregister_md_personality);
5041 EXPORT_SYMBOL(md_error);
5042 EXPORT_SYMBOL(md_done_sync);
5043 EXPORT_SYMBOL(md_write_start);
5044 EXPORT_SYMBOL(md_write_end);
5045 EXPORT_SYMBOL(md_register_thread);
5046 EXPORT_SYMBOL(md_unregister_thread);
5047 EXPORT_SYMBOL(md_wakeup_thread);
5048 EXPORT_SYMBOL(md_print_devices);
5049 EXPORT_SYMBOL(md_check_recovery);
5050 MODULE_LICENSE("GPL");
5051 MODULE_ALIAS("md");
5052 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);