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/kthread.h>
  36 #include <linux/raid/md.h>
  37 #include <linux/raid/bitmap.h>
  38 #include <linux/sysctl.h>
  39 #include <linux/buffer_head.h> /* for invalidate_bdev */
  40 #include <linux/poll.h>
  41 #include <linux/ctype.h>
  42 #include <linux/hdreg.h>
  43 #include <linux/proc_fs.h>
  44 #include <linux/random.h>
  45 #include <linux/reboot.h>
  46 #include <linux/file.h>
  47 #include <linux/delay.h>
  48
  49 #define MAJOR_NR MD_MAJOR
  50
  51 /* 63 partitions with the alternate major number (mdp) */
  52 #define MdpMinorShift 6
  53
  54 #define DEBUG 0
  55 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  56
  57
  58 #ifndef MODULE
  59 static void autostart_arrays(int part);
  60 #endif
  61
  62 static LIST_HEAD(pers_list);
  63 static DEFINE_SPINLOCK(pers_lock);
  64
  65 static void md_print_devices(void);
  66
  67 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  68
  69 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  70
  71 /*
  72  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  73  * is 1000 KB/sec, so the extra system load does not show up that much.
  74  * Increase it if you want to have more _guaranteed_ speed. Note that
  75  * the RAID driver will use the maximum available bandwidth if the IO
  76  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  77  * speed limit - in case reconstruction slows down your system despite
  78  * idle IO detection.
  79  *
  80  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  81  * or /sys/block/mdX/md/sync_speed_{min,max}
  82  */
  83
  84 static int sysctl_speed_limit_min = 1000;
  85 static int sysctl_speed_limit_max = 200000;
  86 static inline int speed_min(mddev_t *mddev)
  87 {
  88         return mddev->sync_speed_min ?
  89                 mddev->sync_speed_min : sysctl_speed_limit_min;
  90 }
  91
  92 static inline int speed_max(mddev_t *mddev)
  93 {
  94         return mddev->sync_speed_max ?
  95                 mddev->sync_speed_max : sysctl_speed_limit_max;
  96 }
  97
  98 static struct ctl_table_header *raid_table_header;
  99
 100 static ctl_table raid_table[] = {
 101         {
 102                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 103                 .procname       = "speed_limit_min",
 104                 .data           = &sysctl_speed_limit_min,
 105                 .maxlen         = sizeof(int),
 106                 .mode           = S_IRUGO|S_IWUSR,
 107                 .proc_handler   = &proc_dointvec,
 108         },
 109         {
 110                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 111                 .procname       = "speed_limit_max",
 112                 .data           = &sysctl_speed_limit_max,
 113                 .maxlen         = sizeof(int),
 114                 .mode           = S_IRUGO|S_IWUSR,
 115                 .proc_handler   = &proc_dointvec,
 116         },
 117         { .ctl_name = 0 }
 118 };
 119
 120 static ctl_table raid_dir_table[] = {
 121         {
 122                 .ctl_name       = DEV_RAID,
 123                 .procname       = "raid",
 124                 .maxlen         = 0,
 125                 .mode           = S_IRUGO|S_IXUGO,
 126                 .child          = raid_table,
 127         },
 128         { .ctl_name = 0 }
 129 };
 130
 131 static ctl_table raid_root_table[] = {
 132         {
 133                 .ctl_name       = CTL_DEV,
 134                 .procname       = "dev",
 135                 .maxlen         = 0,
 136                 .mode           = 0555,
 137                 .child          = raid_dir_table,
 138         },
 139         { .ctl_name = 0 }
 140 };
 141
 142 static struct block_device_operations md_fops;
 143
 144 static int start_readonly;
 145
 146 /*
 147  * We have a system wide 'event count' that is incremented
 148  * on any 'interesting' event, and readers of /proc/mdstat
 149  * can use 'poll' or 'select' to find out when the event
 150  * count increases.
 151  *
 152  * Events are:
 153  *  start array, stop array, error, add device, remove device,
 154  *  start build, activate spare
 155  */
 156 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 157 static atomic_t md_event_count;
 158 void md_new_event(mddev_t *mddev)
 159 {
 160         atomic_inc(&md_event_count);
 161         wake_up(&md_event_waiters);
 162 }
 163 EXPORT_SYMBOL_GPL(md_new_event);
 164
 165 /* Alternate version that can be called from interrupts
 166  * when calling sysfs_notify isn't needed.
 167  */
 168 static void md_new_event_inintr(mddev_t *mddev)
 169 {
 170         atomic_inc(&md_event_count);
 171         wake_up(&md_event_waiters);
 172 }
 173
 174 /*
 175  * Enables to iterate over all existing md arrays
 176  * all_mddevs_lock protects this list.
 177  */
 178 static LIST_HEAD(all_mddevs);
 179 static DEFINE_SPINLOCK(all_mddevs_lock);
 180
 181
 182 /*
 183  * iterates through all used mddevs in the system.
 184  * We take care to grab the all_mddevs_lock whenever navigating
 185  * the list, and to always hold a refcount when unlocked.
 186  * Any code which breaks out of this loop while own
 187  * a reference to the current mddev and must mddev_put it.
 188  */
 189 #define for_each_mddev(mddev,tmp)                                       \
 190                                                                         \
 191         for (({ spin_lock(&all_mddevs_lock);                            \
 192                 tmp = all_mddevs.next;                                  \
 193                 mddev = NULL;});                                        \
 194              ({ if (tmp != &all_mddevs)                                 \
 195                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 196                 spin_unlock(&all_mddevs_lock);                          \
 197                 if (mddev) mddev_put(mddev);                            \
 198                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 199                 tmp != &all_mddevs;});                                  \
 200              ({ spin_lock(&all_mddevs_lock);                            \
 201                 tmp = tmp->next;})                                      \
 202                 )
 203
 204
 205 static int md_fail_request(struct request_queue *q, struct bio *bio)
 206 {
 207         bio_io_error(bio);
 208         return 0;
 209 }
 210
 211 static inline mddev_t *mddev_get(mddev_t *mddev)
 212 {
 213         atomic_inc(&mddev->active);
 214         return mddev;
 215 }
 216
 217 static void mddev_delayed_delete(struct work_struct *ws)
 218 {
 219         mddev_t *mddev = container_of(ws, mddev_t, del_work);
 220         kobject_del(&mddev->kobj);
 221         kobject_put(&mddev->kobj);
 222 }
 223
 224 static void mddev_put(mddev_t *mddev)
 225 {
 226         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 227                 return;
 228         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 229             !mddev->hold_active) {
 230                 list_del(&mddev->all_mddevs);
 231                 if (mddev->gendisk) {
 232                         /* we did a probe so need to clean up.
 233                          * Call schedule_work inside the spinlock
 234                          * so that flush_scheduled_work() after
 235                          * mddev_find will succeed in waiting for the
 236                          * work to be done.
 237                          */
 238                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 239                         schedule_work(&mddev->del_work);
 240                 } else
 241                         kfree(mddev);
 242         }
 243         spin_unlock(&all_mddevs_lock);
 244 }
 245
 246 static mddev_t * mddev_find(dev_t unit)
 247 {
 248         mddev_t *mddev, *new = NULL;
 249
 250  retry:
 251         spin_lock(&all_mddevs_lock);
 252
 253         if (unit) {
 254                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 255                         if (mddev->unit == unit) {
 256                                 mddev_get(mddev);
 257                                 spin_unlock(&all_mddevs_lock);
 258                                 kfree(new);
 259                                 return mddev;
 260                         }
 261
 262                 if (new) {
 263                         list_add(&new->all_mddevs, &all_mddevs);
 264                         spin_unlock(&all_mddevs_lock);
 265                         new->hold_active = UNTIL_IOCTL;
 266                         return new;
 267                 }
 268         } else if (new) {
 269                 /* find an unused unit number */
 270                 static int next_minor = 512;
 271                 int start = next_minor;
 272                 int is_free = 0;
 273                 int dev = 0;
 274                 while (!is_free) {
 275                         dev = MKDEV(MD_MAJOR, next_minor);
 276                         next_minor++;
 277                         if (next_minor > MINORMASK)
 278                                 next_minor = 0;
 279                         if (next_minor == start) {
 280                                 /* Oh dear, all in use. */
 281                                 spin_unlock(&all_mddevs_lock);
 282                                 kfree(new);
 283                                 return NULL;
 284                         }
 285
 286                         is_free = 1;
 287                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 288                                 if (mddev->unit == dev) {
 289                                         is_free = 0;
 290                                         break;
 291                                 }
 292                 }
 293                 new->unit = dev;
 294                 new->md_minor = MINOR(dev);
 295                 new->hold_active = UNTIL_STOP;
 296                 list_add(&new->all_mddevs, &all_mddevs);
 297                 spin_unlock(&all_mddevs_lock);
 298                 return new;
 299         }
 300         spin_unlock(&all_mddevs_lock);
 301
 302         new = kzalloc(sizeof(*new), GFP_KERNEL);
 303         if (!new)
 304                 return NULL;
 305
 306         new->unit = unit;
 307         if (MAJOR(unit) == MD_MAJOR)
 308                 new->md_minor = MINOR(unit);
 309         else
 310                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 311
 312         mutex_init(&new->reconfig_mutex);
 313         INIT_LIST_HEAD(&new->disks);
 314         INIT_LIST_HEAD(&new->all_mddevs);
 315         init_timer(&new->safemode_timer);
 316         atomic_set(&new->active, 1);
 317         atomic_set(&new->openers, 0);
 318         spin_lock_init(&new->write_lock);
 319         init_waitqueue_head(&new->sb_wait);
 320         init_waitqueue_head(&new->recovery_wait);
 321         new->reshape_position = MaxSector;
 322         new->resync_min = 0;
 323         new->resync_max = MaxSector;
 324         new->level = LEVEL_NONE;
 325
 326         goto retry;
 327 }
 328
 329 static inline int mddev_lock(mddev_t * mddev)
 330 {
 331         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 332 }
 333
 334 static inline int mddev_trylock(mddev_t * mddev)
 335 {
 336         return mutex_trylock(&mddev->reconfig_mutex);
 337 }
 338
 339 static inline void mddev_unlock(mddev_t * mddev)
 340 {
 341         mutex_unlock(&mddev->reconfig_mutex);
 342
 343         md_wakeup_thread(mddev->thread);
 344 }
 345
 346 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 347 {
 348         mdk_rdev_t *rdev;
 349
 350         list_for_each_entry(rdev, &mddev->disks, same_set)
 351                 if (rdev->desc_nr == nr)
 352                         return rdev;
 353
 354         return NULL;
 355 }
 356
 357 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 358 {
 359         mdk_rdev_t *rdev;
 360
 361         list_for_each_entry(rdev, &mddev->disks, same_set)
 362                 if (rdev->bdev->bd_dev == dev)
 363                         return rdev;
 364
 365         return NULL;
 366 }
 367
 368 static struct mdk_personality *find_pers(int level, char *clevel)
 369 {
 370         struct mdk_personality *pers;
 371         list_for_each_entry(pers, &pers_list, list) {
 372                 if (level != LEVEL_NONE && pers->level == level)
 373                         return pers;
 374                 if (strcmp(pers->name, clevel)==0)
 375                         return pers;
 376         }
 377         return NULL;
 378 }
 379
 380 /* return the offset of the super block in 512byte sectors */
 381 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 382 {
 383         sector_t num_sectors = bdev->bd_inode->i_size / 512;
 384         return MD_NEW_SIZE_SECTORS(num_sectors);
 385 }
 386
 387 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
 388 {
 389         sector_t num_sectors = rdev->sb_start;
 390
 391         if (chunk_size)
 392                 num_sectors &= ~((sector_t)chunk_size/512 - 1);
 393         return num_sectors;
 394 }
 395
 396 static int alloc_disk_sb(mdk_rdev_t * rdev)
 397 {
 398         if (rdev->sb_page)
 399                 MD_BUG();
 400
 401         rdev->sb_page = alloc_page(GFP_KERNEL);
 402         if (!rdev->sb_page) {
 403                 printk(KERN_ALERT "md: out of memory.\n");
 404                 return -ENOMEM;
 405         }
 406
 407         return 0;
 408 }
 409
 410 static void free_disk_sb(mdk_rdev_t * rdev)
 411 {
 412         if (rdev->sb_page) {
 413                 put_page(rdev->sb_page);
 414                 rdev->sb_loaded = 0;
 415                 rdev->sb_page = NULL;
 416                 rdev->sb_start = 0;
 417                 rdev->size = 0;
 418         }
 419 }
 420
 421
 422 static void super_written(struct bio *bio, int error)
 423 {
 424         mdk_rdev_t *rdev = bio->bi_private;
 425         mddev_t *mddev = rdev->mddev;
 426
 427         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 428                 printk("md: super_written gets error=%d, uptodate=%d\n",
 429                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 430                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 431                 md_error(mddev, rdev);
 432         }
 433
 434         if (atomic_dec_and_test(&mddev->pending_writes))
 435                 wake_up(&mddev->sb_wait);
 436         bio_put(bio);
 437 }
 438
 439 static void super_written_barrier(struct bio *bio, int error)
 440 {
 441         struct bio *bio2 = bio->bi_private;
 442         mdk_rdev_t *rdev = bio2->bi_private;
 443         mddev_t *mddev = rdev->mddev;
 444
 445         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 446             error == -EOPNOTSUPP) {
 447                 unsigned long flags;
 448                 /* barriers don't appear to be supported :-( */
 449                 set_bit(BarriersNotsupp, &rdev->flags);
 450                 mddev->barriers_work = 0;
 451                 spin_lock_irqsave(&mddev->write_lock, flags);
 452                 bio2->bi_next = mddev->biolist;
 453                 mddev->biolist = bio2;
 454                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 455                 wake_up(&mddev->sb_wait);
 456                 bio_put(bio);
 457         } else {
 458                 bio_put(bio2);
 459                 bio->bi_private = rdev;
 460                 super_written(bio, error);
 461         }
 462 }
 463
 464 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 465                    sector_t sector, int size, struct page *page)
 466 {
 467         /* write first size bytes of page to sector of rdev
 468          * Increment mddev->pending_writes before returning
 469          * and decrement it on completion, waking up sb_wait
 470          * if zero is reached.
 471          * If an error occurred, call md_error
 472          *
 473          * As we might need to resubmit the request if BIO_RW_BARRIER
 474          * causes ENOTSUPP, we allocate a spare bio...
 475          */
 476         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 477         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 478
 479         bio->bi_bdev = rdev->bdev;
 480         bio->bi_sector = sector;
 481         bio_add_page(bio, page, size, 0);
 482         bio->bi_private = rdev;
 483         bio->bi_end_io = super_written;
 484         bio->bi_rw = rw;
 485
 486         atomic_inc(&mddev->pending_writes);
 487         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 488                 struct bio *rbio;
 489                 rw |= (1<<BIO_RW_BARRIER);
 490                 rbio = bio_clone(bio, GFP_NOIO);
 491                 rbio->bi_private = bio;
 492                 rbio->bi_end_io = super_written_barrier;
 493                 submit_bio(rw, rbio);
 494         } else
 495                 submit_bio(rw, bio);
 496 }
 497
 498 void md_super_wait(mddev_t *mddev)
 499 {
 500         /* wait for all superblock writes that were scheduled to complete.
 501          * if any had to be retried (due to BARRIER problems), retry them
 502          */
 503         DEFINE_WAIT(wq);
 504         for(;;) {
 505                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 506                 if (atomic_read(&mddev->pending_writes)==0)
 507                         break;
 508                 while (mddev->biolist) {
 509                         struct bio *bio;
 510                         spin_lock_irq(&mddev->write_lock);
 511                         bio = mddev->biolist;
 512                         mddev->biolist = bio->bi_next ;
 513                         bio->bi_next = NULL;
 514                         spin_unlock_irq(&mddev->write_lock);
 515                         submit_bio(bio->bi_rw, bio);
 516                 }
 517                 schedule();
 518         }
 519         finish_wait(&mddev->sb_wait, &wq);
 520 }
 521
 522 static void bi_complete(struct bio *bio, int error)
 523 {
 524         complete((struct completion*)bio->bi_private);
 525 }
 526
 527 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 528                    struct page *page, int rw)
 529 {
 530         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 531         struct completion event;
 532         int ret;
 533
 534         rw |= (1 << BIO_RW_SYNC);
 535
 536         bio->bi_bdev = bdev;
 537         bio->bi_sector = sector;
 538         bio_add_page(bio, page, size, 0);
 539         init_completion(&event);
 540         bio->bi_private = &event;
 541         bio->bi_end_io = bi_complete;
 542         submit_bio(rw, bio);
 543         wait_for_completion(&event);
 544
 545         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 546         bio_put(bio);
 547         return ret;
 548 }
 549 EXPORT_SYMBOL_GPL(sync_page_io);
 550
 551 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 552 {
 553         char b[BDEVNAME_SIZE];
 554         if (!rdev->sb_page) {
 555                 MD_BUG();
 556                 return -EINVAL;
 557         }
 558         if (rdev->sb_loaded)
 559                 return 0;
 560
 561
 562         if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
 563                 goto fail;
 564         rdev->sb_loaded = 1;
 565         return 0;
 566
 567 fail:
 568         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 569                 bdevname(rdev->bdev,b));
 570         return -EINVAL;
 571 }
 572
 573 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 574 {
 575         return  sb1->set_uuid0 == sb2->set_uuid0 &&
 576                 sb1->set_uuid1 == sb2->set_uuid1 &&
 577                 sb1->set_uuid2 == sb2->set_uuid2 &&
 578                 sb1->set_uuid3 == sb2->set_uuid3;
 579 }
 580
 581 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 582 {
 583         int ret;
 584         mdp_super_t *tmp1, *tmp2;
 585
 586         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 587         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 588
 589         if (!tmp1 || !tmp2) {
 590                 ret = 0;
 591                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 592                 goto abort;
 593         }
 594
 595         *tmp1 = *sb1;
 596         *tmp2 = *sb2;
 597
 598         /*
 599          * nr_disks is not constant
 600          */
 601         tmp1->nr_disks = 0;
 602         tmp2->nr_disks = 0;
 603
 604         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 605 abort:
 606         kfree(tmp1);
 607         kfree(tmp2);
 608         return ret;
 609 }
 610
 611
 612 static u32 md_csum_fold(u32 csum)
 613 {
 614         csum = (csum & 0xffff) + (csum >> 16);
 615         return (csum & 0xffff) + (csum >> 16);
 616 }
 617
 618 static unsigned int calc_sb_csum(mdp_super_t * sb)
 619 {
 620         u64 newcsum = 0;
 621         u32 *sb32 = (u32*)sb;
 622         int i;
 623         unsigned int disk_csum, csum;
 624
 625         disk_csum = sb->sb_csum;
 626         sb->sb_csum = 0;
 627
 628         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 629                 newcsum += sb32[i];
 630         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 631
 632
 633 #ifdef CONFIG_ALPHA
 634         /* This used to use csum_partial, which was wrong for several
 635          * reasons including that different results are returned on
 636          * different architectures.  It isn't critical that we get exactly
 637          * the same return value as before (we always csum_fold before
 638          * testing, and that removes any differences).  However as we
 639          * know that csum_partial always returned a 16bit value on
 640          * alphas, do a fold to maximise conformity to previous behaviour.
 641          */
 642         sb->sb_csum = md_csum_fold(disk_csum);
 643 #else
 644         sb->sb_csum = disk_csum;
 645 #endif
 646         return csum;
 647 }
 648
 649
 650 /*
 651  * Handle superblock details.
 652  * We want to be able to handle multiple superblock formats
 653  * so we have a common interface to them all, and an array of
 654  * different handlers.
 655  * We rely on user-space to write the initial superblock, and support
 656  * reading and updating of superblocks.
 657  * Interface methods are:
 658  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 659  *      loads and validates a superblock on dev.
 660  *      if refdev != NULL, compare superblocks on both devices
 661  *    Return:
 662  *      0 - dev has a superblock that is compatible with refdev
 663  *      1 - dev has a superblock that is compatible and newer than refdev
 664  *          so dev should be used as the refdev in future
 665  *     -EINVAL superblock incompatible or invalid
 666  *     -othererror e.g. -EIO
 667  *
 668  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 669  *      Verify that dev is acceptable into mddev.
 670  *       The first time, mddev->raid_disks will be 0, and data from
 671  *       dev should be merged in.  Subsequent calls check that dev
 672  *       is new enough.  Return 0 or -EINVAL
 673  *
 674  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 675  *     Update the superblock for rdev with data in mddev
 676  *     This does not write to disc.
 677  *
 678  */
 679
 680 struct super_type  {
 681         char                *name;
 682         struct module       *owner;
 683         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
 684                                           int minor_version);
 685         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 686         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 687         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
 688                                                 sector_t num_sectors);
 689 };
 690
 691 /*
 692  * load_super for 0.90.0
 693  */
 694 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 695 {
 696         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 697         mdp_super_t *sb;
 698         int ret;
 699
 700         /*
 701          * Calculate the position of the superblock (512byte sectors),
 702          * it's at the end of the disk.
 703          *
 704          * It also happens to be a multiple of 4Kb.
 705          */
 706         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
 707
 708         ret = read_disk_sb(rdev, MD_SB_BYTES);
 709         if (ret) return ret;
 710
 711         ret = -EINVAL;
 712
 713         bdevname(rdev->bdev, b);
 714         sb = (mdp_super_t*)page_address(rdev->sb_page);
 715
 716         if (sb->md_magic != MD_SB_MAGIC) {
 717                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 718                        b);
 719                 goto abort;
 720         }
 721
 722         if (sb->major_version != 0 ||
 723             sb->minor_version < 90 ||
 724             sb->minor_version > 91) {
 725                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 726                         sb->major_version, sb->minor_version,
 727                         b);
 728                 goto abort;
 729         }
 730
 731         if (sb->raid_disks <= 0)
 732                 goto abort;
 733
 734         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 735                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 736                         b);
 737                 goto abort;
 738         }
 739
 740         rdev->preferred_minor = sb->md_minor;
 741         rdev->data_offset = 0;
 742         rdev->sb_size = MD_SB_BYTES;
 743
 744         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 745                 if (sb->level != 1 && sb->level != 4
 746                     && sb->level != 5 && sb->level != 6
 747                     && sb->level != 10) {
 748                         /* FIXME use a better test */
 749                         printk(KERN_WARNING
 750                                "md: bitmaps not supported for this level.\n");
 751                         goto abort;
 752                 }
 753         }
 754
 755         if (sb->level == LEVEL_MULTIPATH)
 756                 rdev->desc_nr = -1;
 757         else
 758                 rdev->desc_nr = sb->this_disk.number;
 759
 760         if (!refdev) {
 761                 ret = 1;
 762         } else {
 763                 __u64 ev1, ev2;
 764                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 765                 if (!uuid_equal(refsb, sb)) {
 766                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 767                                 b, bdevname(refdev->bdev,b2));
 768                         goto abort;
 769                 }
 770                 if (!sb_equal(refsb, sb)) {
 771                         printk(KERN_WARNING "md: %s has same UUID"
 772                                " but different superblock to %s\n",
 773                                b, bdevname(refdev->bdev, b2));
 774                         goto abort;
 775                 }
 776                 ev1 = md_event(sb);
 777                 ev2 = md_event(refsb);
 778                 if (ev1 > ev2)
 779                         ret = 1;
 780                 else
 781                         ret = 0;
 782         }
 783         rdev->size = calc_num_sectors(rdev, sb->chunk_size) / 2;
 784
 785         if (rdev->size < sb->size && sb->level > 1)
 786                 /* "this cannot possibly happen" ... */
 787                 ret = -EINVAL;
 788
 789  abort:
 790         return ret;
 791 }
 792
 793 /*
 794  * validate_super for 0.90.0
 795  */
 796 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 797 {
 798         mdp_disk_t *desc;
 799         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 800         __u64 ev1 = md_event(sb);
 801
 802         rdev->raid_disk = -1;
 803         clear_bit(Faulty, &rdev->flags);
 804         clear_bit(In_sync, &rdev->flags);
 805         clear_bit(WriteMostly, &rdev->flags);
 806         clear_bit(BarriersNotsupp, &rdev->flags);
 807
 808         if (mddev->raid_disks == 0) {
 809                 mddev->major_version = 0;
 810                 mddev->minor_version = sb->minor_version;
 811                 mddev->patch_version = sb->patch_version;
 812                 mddev->external = 0;
 813                 mddev->chunk_size = sb->chunk_size;
 814                 mddev->ctime = sb->ctime;
 815                 mddev->utime = sb->utime;
 816                 mddev->level = sb->level;
 817                 mddev->clevel[0] = 0;
 818                 mddev->layout = sb->layout;
 819                 mddev->raid_disks = sb->raid_disks;
 820                 mddev->size = sb->size;
 821                 mddev->events = ev1;
 822                 mddev->bitmap_offset = 0;
 823                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 824
 825                 if (mddev->minor_version >= 91) {
 826                         mddev->reshape_position = sb->reshape_position;
 827                         mddev->delta_disks = sb->delta_disks;
 828                         mddev->new_level = sb->new_level;
 829                         mddev->new_layout = sb->new_layout;
 830                         mddev->new_chunk = sb->new_chunk;
 831                 } else {
 832                         mddev->reshape_position = MaxSector;
 833                         mddev->delta_disks = 0;
 834                         mddev->new_level = mddev->level;
 835                         mddev->new_layout = mddev->layout;
 836                         mddev->new_chunk = mddev->chunk_size;
 837                 }
 838
 839                 if (sb->state & (1<<MD_SB_CLEAN))
 840                         mddev->recovery_cp = MaxSector;
 841                 else {
 842                         if (sb->events_hi == sb->cp_events_hi &&
 843                                 sb->events_lo == sb->cp_events_lo) {
 844                                 mddev->recovery_cp = sb->recovery_cp;
 845                         } else
 846                                 mddev->recovery_cp = 0;
 847                 }
 848
 849                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 850                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 851                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 852                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 853
 854                 mddev->max_disks = MD_SB_DISKS;
 855
 856                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 857                     mddev->bitmap_file == NULL)
 858                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 859
 860         } else if (mddev->pers == NULL) {
 861                 /* Insist on good event counter while assembling */
 862                 ++ev1;
 863                 if (ev1 < mddev->events)
 864                         return -EINVAL;
 865         } else if (mddev->bitmap) {
 866                 /* if adding to array with a bitmap, then we can accept an
 867                  * older device ... but not too old.
 868                  */
 869                 if (ev1 < mddev->bitmap->events_cleared)
 870                         return 0;
 871         } else {
 872                 if (ev1 < mddev->events)
 873                         /* just a hot-add of a new device, leave raid_disk at -1 */
 874                         return 0;
 875         }
 876
 877         if (mddev->level != LEVEL_MULTIPATH) {
 878                 desc = sb->disks + rdev->desc_nr;
 879
 880                 if (desc->state & (1<<MD_DISK_FAULTY))
 881                         set_bit(Faulty, &rdev->flags);
 882                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
 883                             desc->raid_disk < mddev->raid_disks */) {
 884                         set_bit(In_sync, &rdev->flags);
 885                         rdev->raid_disk = desc->raid_disk;
 886                 }
 887                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 888                         set_bit(WriteMostly, &rdev->flags);
 889         } else /* MULTIPATH are always insync */
 890                 set_bit(In_sync, &rdev->flags);
 891         return 0;
 892 }
 893
 894 /*
 895  * sync_super for 0.90.0
 896  */
 897 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 898 {
 899         mdp_super_t *sb;
 900         mdk_rdev_t *rdev2;
 901         int next_spare = mddev->raid_disks;
 902
 903
 904         /* make rdev->sb match mddev data..
 905          *
 906          * 1/ zero out disks
 907          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 908          * 3/ any empty disks < next_spare become removed
 909          *
 910          * disks[0] gets initialised to REMOVED because
 911          * we cannot be sure from other fields if it has
 912          * been initialised or not.
 913          */
 914         int i;
 915         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 916
 917         rdev->sb_size = MD_SB_BYTES;
 918
 919         sb = (mdp_super_t*)page_address(rdev->sb_page);
 920
 921         memset(sb, 0, sizeof(*sb));
 922
 923         sb->md_magic = MD_SB_MAGIC;
 924         sb->major_version = mddev->major_version;
 925         sb->patch_version = mddev->patch_version;
 926         sb->gvalid_words  = 0; /* ignored */
 927         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 928         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 929         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 930         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 931
 932         sb->ctime = mddev->ctime;
 933         sb->level = mddev->level;
 934         sb->size  = mddev->size;
 935         sb->raid_disks = mddev->raid_disks;
 936         sb->md_minor = mddev->md_minor;
 937         sb->not_persistent = 0;
 938         sb->utime = mddev->utime;
 939         sb->state = 0;
 940         sb->events_hi = (mddev->events>>32);
 941         sb->events_lo = (u32)mddev->events;
 942
 943         if (mddev->reshape_position == MaxSector)
 944                 sb->minor_version = 90;
 945         else {
 946                 sb->minor_version = 91;
 947                 sb->reshape_position = mddev->reshape_position;
 948                 sb->new_level = mddev->new_level;
 949                 sb->delta_disks = mddev->delta_disks;
 950                 sb->new_layout = mddev->new_layout;
 951                 sb->new_chunk = mddev->new_chunk;
 952         }
 953         mddev->minor_version = sb->minor_version;
 954         if (mddev->in_sync)
 955         {
 956                 sb->recovery_cp = mddev->recovery_cp;
 957                 sb->cp_events_hi = (mddev->events>>32);
 958                 sb->cp_events_lo = (u32)mddev->events;
 959                 if (mddev->recovery_cp == MaxSector)
 960                         sb->state = (1<< MD_SB_CLEAN);
 961         } else
 962                 sb->recovery_cp = 0;
 963
 964         sb->layout = mddev->layout;
 965         sb->chunk_size = mddev->chunk_size;
 966
 967         if (mddev->bitmap && mddev->bitmap_file == NULL)
 968                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 969
 970         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 971         list_for_each_entry(rdev2, &mddev->disks, same_set) {
 972                 mdp_disk_t *d;
 973                 int desc_nr;
 974                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 975                     && !test_bit(Faulty, &rdev2->flags))
 976                         desc_nr = rdev2->raid_disk;
 977                 else
 978                         desc_nr = next_spare++;
 979                 rdev2->desc_nr = desc_nr;
 980                 d = &sb->disks[rdev2->desc_nr];
 981                 nr_disks++;
 982                 d->number = rdev2->desc_nr;
 983                 d->major = MAJOR(rdev2->bdev->bd_dev);
 984                 d->minor = MINOR(rdev2->bdev->bd_dev);
 985                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 986                     && !test_bit(Faulty, &rdev2->flags))
 987                         d->raid_disk = rdev2->raid_disk;
 988                 else
 989                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 990                 if (test_bit(Faulty, &rdev2->flags))
 991                         d->state = (1<<MD_DISK_FAULTY);
 992                 else if (test_bit(In_sync, &rdev2->flags)) {
 993                         d->state = (1<<MD_DISK_ACTIVE);
 994                         d->state |= (1<<MD_DISK_SYNC);
 995                         active++;
 996                         working++;
 997                 } else {
 998                         d->state = 0;
 999                         spare++;
1000                         working++;
1001                 }
1002                 if (test_bit(WriteMostly, &rdev2->flags))
1003                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1004         }
1005         /* now set the "removed" and "faulty" bits on any missing devices */
1006         for (i=0 ; i < mddev->raid_disks ; i++) {
1007                 mdp_disk_t *d = &sb->disks[i];
1008                 if (d->state == 0 && d->number == 0) {
1009                         d->number = i;
1010                         d->raid_disk = i;
1011                         d->state = (1<<MD_DISK_REMOVED);
1012                         d->state |= (1<<MD_DISK_FAULTY);
1013                         failed++;
1014                 }
1015         }
1016         sb->nr_disks = nr_disks;
1017         sb->active_disks = active;
1018         sb->working_disks = working;
1019         sb->failed_disks = failed;
1020         sb->spare_disks = spare;
1021
1022         sb->this_disk = sb->disks[rdev->desc_nr];
1023         sb->sb_csum = calc_sb_csum(sb);
1024 }
1025
1026 /*
1027  * rdev_size_change for 0.90.0
1028  */
1029 static unsigned long long
1030 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1031 {
1032         if (num_sectors && num_sectors < rdev->mddev->size * 2)
1033                 return 0; /* component must fit device */
1034         if (rdev->mddev->bitmap_offset)
1035                 return 0; /* can't move bitmap */
1036         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1037         if (!num_sectors || num_sectors > rdev->sb_start)
1038                 num_sectors = rdev->sb_start;
1039         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1040                        rdev->sb_page);
1041         md_super_wait(rdev->mddev);
1042         return num_sectors / 2; /* kB for sysfs */
1043 }
1044
1045
1046 /*
1047  * version 1 superblock
1048  */
1049
1050 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1051 {
1052         __le32 disk_csum;
1053         u32 csum;
1054         unsigned long long newcsum;
1055         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1056         __le32 *isuper = (__le32*)sb;
1057         int i;
1058
1059         disk_csum = sb->sb_csum;
1060         sb->sb_csum = 0;
1061         newcsum = 0;
1062         for (i=0; size>=4; size -= 4 )
1063                 newcsum += le32_to_cpu(*isuper++);
1064
1065         if (size == 2)
1066                 newcsum += le16_to_cpu(*(__le16*) isuper);
1067
1068         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1069         sb->sb_csum = disk_csum;
1070         return cpu_to_le32(csum);
1071 }
1072
1073 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1074 {
1075         struct mdp_superblock_1 *sb;
1076         int ret;
1077         sector_t sb_start;
1078         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1079         int bmask;
1080
1081         /*
1082          * Calculate the position of the superblock in 512byte sectors.
1083          * It is always aligned to a 4K boundary and
1084          * depeding on minor_version, it can be:
1085          * 0: At least 8K, but less than 12K, from end of device
1086          * 1: At start of device
1087          * 2: 4K from start of device.
1088          */
1089         switch(minor_version) {
1090         case 0:
1091                 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1092                 sb_start -= 8*2;
1093                 sb_start &= ~(sector_t)(4*2-1);
1094                 break;
1095         case 1:
1096                 sb_start = 0;
1097                 break;
1098         case 2:
1099                 sb_start = 8;
1100                 break;
1101         default:
1102                 return -EINVAL;
1103         }
1104         rdev->sb_start = sb_start;
1105
1106         /* superblock is rarely larger than 1K, but it can be larger,
1107          * and it is safe to read 4k, so we do that
1108          */
1109         ret = read_disk_sb(rdev, 4096);
1110         if (ret) return ret;
1111
1112
1113         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1114
1115         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1116             sb->major_version != cpu_to_le32(1) ||
1117             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1118             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1119             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1120                 return -EINVAL;
1121
1122         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1123                 printk("md: invalid superblock checksum on %s\n",
1124                         bdevname(rdev->bdev,b));
1125                 return -EINVAL;
1126         }
1127         if (le64_to_cpu(sb->data_size) < 10) {
1128                 printk("md: data_size too small on %s\n",
1129                        bdevname(rdev->bdev,b));
1130                 return -EINVAL;
1131         }
1132         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1133                 if (sb->level != cpu_to_le32(1) &&
1134                     sb->level != cpu_to_le32(4) &&
1135                     sb->level != cpu_to_le32(5) &&
1136                     sb->level != cpu_to_le32(6) &&
1137                     sb->level != cpu_to_le32(10)) {
1138                         printk(KERN_WARNING
1139                                "md: bitmaps not supported for this level.\n");
1140                         return -EINVAL;
1141                 }
1142         }
1143
1144         rdev->preferred_minor = 0xffff;
1145         rdev->data_offset = le64_to_cpu(sb->data_offset);
1146         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1147
1148         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1149         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1150         if (rdev->sb_size & bmask)
1151                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1152
1153         if (minor_version
1154             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1155                 return -EINVAL;
1156
1157         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1158                 rdev->desc_nr = -1;
1159         else
1160                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1161
1162         if (!refdev) {
1163                 ret = 1;
1164         } else {
1165                 __u64 ev1, ev2;
1166                 struct mdp_superblock_1 *refsb =
1167                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1168
1169                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1170                     sb->level != refsb->level ||
1171                     sb->layout != refsb->layout ||
1172                     sb->chunksize != refsb->chunksize) {
1173                         printk(KERN_WARNING "md: %s has strangely different"
1174                                 " superblock to %s\n",
1175                                 bdevname(rdev->bdev,b),
1176                                 bdevname(refdev->bdev,b2));
1177                         return -EINVAL;
1178                 }
1179                 ev1 = le64_to_cpu(sb->events);
1180                 ev2 = le64_to_cpu(refsb->events);
1181
1182                 if (ev1 > ev2)
1183                         ret = 1;
1184                 else
1185                         ret = 0;
1186         }
1187         if (minor_version)
1188                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1189         else
1190                 rdev->size = rdev->sb_start / 2;
1191         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1192                 return -EINVAL;
1193         rdev->size = le64_to_cpu(sb->data_size)/2;
1194         if (le32_to_cpu(sb->chunksize))
1195                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1196
1197         if (le64_to_cpu(sb->size) > rdev->size*2)
1198                 return -EINVAL;
1199         return ret;
1200 }
1201
1202 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1203 {
1204         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1205         __u64 ev1 = le64_to_cpu(sb->events);
1206
1207         rdev->raid_disk = -1;
1208         clear_bit(Faulty, &rdev->flags);
1209         clear_bit(In_sync, &rdev->flags);
1210         clear_bit(WriteMostly, &rdev->flags);
1211         clear_bit(BarriersNotsupp, &rdev->flags);
1212
1213         if (mddev->raid_disks == 0) {
1214                 mddev->major_version = 1;
1215                 mddev->patch_version = 0;
1216                 mddev->external = 0;
1217                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1218                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1219                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1220                 mddev->level = le32_to_cpu(sb->level);
1221                 mddev->clevel[0] = 0;
1222                 mddev->layout = le32_to_cpu(sb->layout);
1223                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1224                 mddev->size = le64_to_cpu(sb->size)/2;
1225                 mddev->events = ev1;
1226                 mddev->bitmap_offset = 0;
1227                 mddev->default_bitmap_offset = 1024 >> 9;
1228
1229                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1230                 memcpy(mddev->uuid, sb->set_uuid, 16);
1231
1232                 mddev->max_disks =  (4096-256)/2;
1233
1234                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1235                     mddev->bitmap_file == NULL )
1236                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1237
1238                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1239                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1240                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1241                         mddev->new_level = le32_to_cpu(sb->new_level);
1242                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1243                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1244                 } else {
1245                         mddev->reshape_position = MaxSector;
1246                         mddev->delta_disks = 0;
1247                         mddev->new_level = mddev->level;
1248                         mddev->new_layout = mddev->layout;
1249                         mddev->new_chunk = mddev->chunk_size;
1250                 }
1251
1252         } else if (mddev->pers == NULL) {
1253                 /* Insist of good event counter while assembling */
1254                 ++ev1;
1255                 if (ev1 < mddev->events)
1256                         return -EINVAL;
1257         } else if (mddev->bitmap) {
1258                 /* If adding to array with a bitmap, then we can accept an
1259                  * older device, but not too old.
1260                  */
1261                 if (ev1 < mddev->bitmap->events_cleared)
1262                         return 0;
1263         } else {
1264                 if (ev1 < mddev->events)
1265                         /* just a hot-add of a new device, leave raid_disk at -1 */
1266                         return 0;
1267         }
1268         if (mddev->level != LEVEL_MULTIPATH) {
1269                 int role;
1270                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1271                 switch(role) {
1272                 case 0xffff: /* spare */
1273                         break;
1274                 case 0xfffe: /* faulty */
1275                         set_bit(Faulty, &rdev->flags);
1276                         break;
1277                 default:
1278                         if ((le32_to_cpu(sb->feature_map) &
1279                              MD_FEATURE_RECOVERY_OFFSET))
1280                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1281                         else
1282                                 set_bit(In_sync, &rdev->flags);
1283                         rdev->raid_disk = role;
1284                         break;
1285                 }
1286                 if (sb->devflags & WriteMostly1)
1287                         set_bit(WriteMostly, &rdev->flags);
1288         } else /* MULTIPATH are always insync */
1289                 set_bit(In_sync, &rdev->flags);
1290
1291         return 0;
1292 }
1293
1294 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1295 {
1296         struct mdp_superblock_1 *sb;
1297         mdk_rdev_t *rdev2;
1298         int max_dev, i;
1299         /* make rdev->sb match mddev and rdev data. */
1300
1301         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1302
1303         sb->feature_map = 0;
1304         sb->pad0 = 0;
1305         sb->recovery_offset = cpu_to_le64(0);
1306         memset(sb->pad1, 0, sizeof(sb->pad1));
1307         memset(sb->pad2, 0, sizeof(sb->pad2));
1308         memset(sb->pad3, 0, sizeof(sb->pad3));
1309
1310         sb->utime = cpu_to_le64((__u64)mddev->utime);
1311         sb->events = cpu_to_le64(mddev->events);
1312         if (mddev->in_sync)
1313                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1314         else
1315                 sb->resync_offset = cpu_to_le64(0);
1316
1317         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1318
1319         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1320         sb->size = cpu_to_le64(mddev->size<<1);
1321
1322         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1323                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1324                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1325         }
1326
1327         if (rdev->raid_disk >= 0 &&
1328             !test_bit(In_sync, &rdev->flags) &&
1329             rdev->recovery_offset > 0) {
1330                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1331                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1332         }
1333
1334         if (mddev->reshape_position != MaxSector) {
1335                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1336                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1337                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1338                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1339                 sb->new_level = cpu_to_le32(mddev->new_level);
1340                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1341         }
1342
1343         max_dev = 0;
1344         list_for_each_entry(rdev2, &mddev->disks, same_set)
1345                 if (rdev2->desc_nr+1 > max_dev)
1346                         max_dev = rdev2->desc_nr+1;
1347
1348         if (max_dev > le32_to_cpu(sb->max_dev))
1349                 sb->max_dev = cpu_to_le32(max_dev);
1350         for (i=0; i<max_dev;i++)
1351                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1352
1353         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1354                 i = rdev2->desc_nr;
1355                 if (test_bit(Faulty, &rdev2->flags))
1356                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1357                 else if (test_bit(In_sync, &rdev2->flags))
1358                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1359                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1360                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1361                 else
1362                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1363         }
1364
1365         sb->sb_csum = calc_sb_1_csum(sb);
1366 }
1367
1368 static unsigned long long
1369 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1370 {
1371         struct mdp_superblock_1 *sb;
1372         sector_t max_sectors;
1373         if (num_sectors && num_sectors < rdev->mddev->size * 2)
1374                 return 0; /* component must fit device */
1375         if (rdev->sb_start < rdev->data_offset) {
1376                 /* minor versions 1 and 2; superblock before data */
1377                 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1378                 max_sectors -= rdev->data_offset;
1379                 if (!num_sectors || num_sectors > max_sectors)
1380                         num_sectors = max_sectors;
1381         } else if (rdev->mddev->bitmap_offset) {
1382                 /* minor version 0 with bitmap we can't move */
1383                 return 0;
1384         } else {
1385                 /* minor version 0; superblock after data */
1386                 sector_t sb_start;
1387                 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1388                 sb_start &= ~(sector_t)(4*2 - 1);
1389                 max_sectors = rdev->size * 2 + sb_start - rdev->sb_start;
1390                 if (!num_sectors || num_sectors > max_sectors)
1391                         num_sectors = max_sectors;
1392                 rdev->sb_start = sb_start;
1393         }
1394         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1395         sb->data_size = cpu_to_le64(num_sectors);
1396         sb->super_offset = rdev->sb_start;
1397         sb->sb_csum = calc_sb_1_csum(sb);
1398         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1399                        rdev->sb_page);
1400         md_super_wait(rdev->mddev);
1401         return num_sectors / 2; /* kB for sysfs */
1402 }
1403
1404 static struct super_type super_types[] = {
1405         [0] = {
1406                 .name   = "0.90.0",
1407                 .owner  = THIS_MODULE,
1408                 .load_super         = super_90_load,
1409                 .validate_super     = super_90_validate,
1410                 .sync_super         = super_90_sync,
1411                 .rdev_size_change   = super_90_rdev_size_change,
1412         },
1413         [1] = {
1414                 .name   = "md-1",
1415                 .owner  = THIS_MODULE,
1416                 .load_super         = super_1_load,
1417                 .validate_super     = super_1_validate,
1418                 .sync_super         = super_1_sync,
1419                 .rdev_size_change   = super_1_rdev_size_change,
1420         },
1421 };
1422
1423 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1424 {
1425         mdk_rdev_t *rdev, *rdev2;
1426
1427         rcu_read_lock();
1428         rdev_for_each_rcu(rdev, mddev1)
1429                 rdev_for_each_rcu(rdev2, mddev2)
1430                         if (rdev->bdev->bd_contains ==
1431                             rdev2->bdev->bd_contains) {
1432                                 rcu_read_unlock();
1433                                 return 1;
1434                         }
1435         rcu_read_unlock();
1436         return 0;
1437 }
1438
1439 static LIST_HEAD(pending_raid_disks);
1440
1441 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1442 {
1443         char b[BDEVNAME_SIZE];
1444         struct kobject *ko;
1445         char *s;
1446         int err;
1447
1448         if (rdev->mddev) {
1449                 MD_BUG();
1450                 return -EINVAL;
1451         }
1452
1453         /* prevent duplicates */
1454         if (find_rdev(mddev, rdev->bdev->bd_dev))
1455                 return -EEXIST;
1456
1457         /* make sure rdev->size exceeds mddev->size */
1458         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1459                 if (mddev->pers) {
1460                         /* Cannot change size, so fail
1461                          * If mddev->level <= 0, then we don't care
1462                          * about aligning sizes (e.g. linear)
1463                          */
1464                         if (mddev->level > 0)
1465                                 return -ENOSPC;
1466                 } else
1467                         mddev->size = rdev->size;
1468         }
1469
1470         /* Verify rdev->desc_nr is unique.
1471          * If it is -1, assign a free number, else
1472          * check number is not in use
1473          */
1474         if (rdev->desc_nr < 0) {
1475                 int choice = 0;
1476                 if (mddev->pers) choice = mddev->raid_disks;
1477                 while (find_rdev_nr(mddev, choice))
1478                         choice++;
1479                 rdev->desc_nr = choice;
1480         } else {
1481                 if (find_rdev_nr(mddev, rdev->desc_nr))
1482                         return -EBUSY;
1483         }
1484         bdevname(rdev->bdev,b);
1485         while ( (s=strchr(b, '/')) != NULL)
1486                 *s = '!';
1487
1488         rdev->mddev = mddev;
1489         printk(KERN_INFO "md: bind<%s>\n", b);
1490
1491         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1492                 goto fail;
1493
1494         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1495         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1496                 kobject_del(&rdev->kobj);
1497                 goto fail;
1498         }
1499         rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1500
1501         list_add_rcu(&rdev->same_set, &mddev->disks);
1502         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1503
1504         /* May as well allow recovery to be retried once */
1505         mddev->recovery_disabled = 0;
1506         return 0;
1507
1508  fail:
1509         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1510                b, mdname(mddev));
1511         return err;
1512 }
1513
1514 static void md_delayed_delete(struct work_struct *ws)
1515 {
1516         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1517         kobject_del(&rdev->kobj);
1518         kobject_put(&rdev->kobj);
1519 }
1520
1521 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1522 {
1523         char b[BDEVNAME_SIZE];
1524         if (!rdev->mddev) {
1525                 MD_BUG();
1526                 return;
1527         }
1528         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1529         list_del_rcu(&rdev->same_set);
1530         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1531         rdev->mddev = NULL;
1532         sysfs_remove_link(&rdev->kobj, "block");
1533         sysfs_put(rdev->sysfs_state);
1534         rdev->sysfs_state = NULL;
1535         /* We need to delay this, otherwise we can deadlock when
1536          * writing to 'remove' to "dev/state".  We also need
1537          * to delay it due to rcu usage.
1538          */
1539         synchronize_rcu();
1540         INIT_WORK(&rdev->del_work, md_delayed_delete);
1541         kobject_get(&rdev->kobj);
1542         schedule_work(&rdev->del_work);
1543 }
1544
1545 /*
1546  * prevent the device from being mounted, repartitioned or
1547  * otherwise reused by a RAID array (or any other kernel
1548  * subsystem), by bd_claiming the device.
1549  */
1550 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1551 {
1552         int err = 0;
1553         struct block_device *bdev;
1554         char b[BDEVNAME_SIZE];
1555
1556         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1557         if (IS_ERR(bdev)) {
1558                 printk(KERN_ERR "md: could not open %s.\n",
1559                         __bdevname(dev, b));
1560                 return PTR_ERR(bdev);
1561         }
1562         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1563         if (err) {
1564                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1565                         bdevname(bdev, b));
1566                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1567                 return err;
1568         }
1569         if (!shared)
1570                 set_bit(AllReserved, &rdev->flags);
1571         rdev->bdev = bdev;
1572         return err;
1573 }
1574
1575 static void unlock_rdev(mdk_rdev_t *rdev)
1576 {
1577         struct block_device *bdev = rdev->bdev;
1578         rdev->bdev = NULL;
1579         if (!bdev)
1580                 MD_BUG();
1581         bd_release(bdev);
1582         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1583 }
1584
1585 void md_autodetect_dev(dev_t dev);
1586
1587 static void export_rdev(mdk_rdev_t * rdev)
1588 {
1589         char b[BDEVNAME_SIZE];
1590         printk(KERN_INFO "md: export_rdev(%s)\n",
1591                 bdevname(rdev->bdev,b));
1592         if (rdev->mddev)
1593                 MD_BUG();
1594         free_disk_sb(rdev);
1595 #ifndef MODULE
1596         if (test_bit(AutoDetected, &rdev->flags))
1597                 md_autodetect_dev(rdev->bdev->bd_dev);
1598 #endif
1599         unlock_rdev(rdev);
1600         kobject_put(&rdev->kobj);
1601 }
1602
1603 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1604 {
1605         unbind_rdev_from_array(rdev);
1606         export_rdev(rdev);
1607 }
1608
1609 static void export_array(mddev_t *mddev)
1610 {
1611         mdk_rdev_t *rdev, *tmp;
1612
1613         rdev_for_each(rdev, tmp, mddev) {
1614                 if (!rdev->mddev) {
1615                         MD_BUG();
1616                         continue;
1617                 }
1618                 kick_rdev_from_array(rdev);
1619         }
1620         if (!list_empty(&mddev->disks))
1621                 MD_BUG();
1622         mddev->raid_disks = 0;
1623         mddev->major_version = 0;
1624 }
1625
1626 static void print_desc(mdp_disk_t *desc)
1627 {
1628         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1629                 desc->major,desc->minor,desc->raid_disk,desc->state);
1630 }
1631
1632 static void print_sb_90(mdp_super_t *sb)
1633 {
1634         int i;
1635
1636         printk(KERN_INFO
1637                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1638                 sb->major_version, sb->minor_version, sb->patch_version,
1639                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1640                 sb->ctime);
1641         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1642                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1643                 sb->md_minor, sb->layout, sb->chunk_size);
1644         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1645                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1646                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1647                 sb->failed_disks, sb->spare_disks,
1648                 sb->sb_csum, (unsigned long)sb->events_lo);
1649
1650         printk(KERN_INFO);
1651         for (i = 0; i < MD_SB_DISKS; i++) {
1652                 mdp_disk_t *desc;
1653
1654                 desc = sb->disks + i;
1655                 if (desc->number || desc->major || desc->minor ||
1656                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1657                         printk("     D %2d: ", i);
1658                         print_desc(desc);
1659                 }
1660         }
1661         printk(KERN_INFO "md:     THIS: ");
1662         print_desc(&sb->this_disk);
1663 }
1664
1665 static void print_sb_1(struct mdp_superblock_1 *sb)
1666 {
1667         __u8 *uuid;
1668
1669         uuid = sb->set_uuid;
1670         printk(KERN_INFO "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1671                         ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1672                KERN_INFO "md:    Name: \"%s\" CT:%llu\n",
1673                 le32_to_cpu(sb->major_version),
1674                 le32_to_cpu(sb->feature_map),
1675                 uuid[0], uuid[1], uuid[2], uuid[3],
1676                 uuid[4], uuid[5], uuid[6], uuid[7],
1677                 uuid[8], uuid[9], uuid[10], uuid[11],
1678                 uuid[12], uuid[13], uuid[14], uuid[15],
1679                 sb->set_name,
1680                 (unsigned long long)le64_to_cpu(sb->ctime)
1681                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1682
1683         uuid = sb->device_uuid;
1684         printk(KERN_INFO "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1685                         " RO:%llu\n"
1686                KERN_INFO "md:     Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1687                         ":%02x%02x%02x%02x%02x%02x\n"
1688                KERN_INFO "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1689                KERN_INFO "md:         (MaxDev:%u) \n",
1690                 le32_to_cpu(sb->level),
1691                 (unsigned long long)le64_to_cpu(sb->size),
1692                 le32_to_cpu(sb->raid_disks),
1693                 le32_to_cpu(sb->layout),
1694                 le32_to_cpu(sb->chunksize),
1695                 (unsigned long long)le64_to_cpu(sb->data_offset),
1696                 (unsigned long long)le64_to_cpu(sb->data_size),
1697                 (unsigned long long)le64_to_cpu(sb->super_offset),
1698                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1699                 le32_to_cpu(sb->dev_number),
1700                 uuid[0], uuid[1], uuid[2], uuid[3],
1701                 uuid[4], uuid[5], uuid[6], uuid[7],
1702                 uuid[8], uuid[9], uuid[10], uuid[11],
1703                 uuid[12], uuid[13], uuid[14], uuid[15],
1704                 sb->devflags,
1705                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1706                 (unsigned long long)le64_to_cpu(sb->events),
1707                 (unsigned long long)le64_to_cpu(sb->resync_offset),
1708                 le32_to_cpu(sb->sb_csum),
1709                 le32_to_cpu(sb->max_dev)
1710                 );
1711 }
1712
1713 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1714 {
1715         char b[BDEVNAME_SIZE];
1716         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1717                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1718                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1719                 rdev->desc_nr);
1720         if (rdev->sb_loaded) {
1721                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1722                 switch (major_version) {
1723                 case 0:
1724                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1725                         break;
1726                 case 1:
1727                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1728                         break;
1729                 }
1730         } else
1731                 printk(KERN_INFO "md: no rdev superblock!\n");
1732 }
1733
1734 static void md_print_devices(void)
1735 {
1736         struct list_head *tmp;
1737         mdk_rdev_t *rdev;
1738         mddev_t *mddev;
1739         char b[BDEVNAME_SIZE];
1740
1741         printk("\n");
1742         printk("md:     **********************************\n");
1743         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1744         printk("md:     **********************************\n");
1745         for_each_mddev(mddev, tmp) {
1746
1747                 if (mddev->bitmap)
1748                         bitmap_print_sb(mddev->bitmap);
1749                 else
1750                         printk("%s: ", mdname(mddev));
1751                 list_for_each_entry(rdev, &mddev->disks, same_set)
1752                         printk("<%s>", bdevname(rdev->bdev,b));
1753                 printk("\n");
1754
1755                 list_for_each_entry(rdev, &mddev->disks, same_set)
1756                         print_rdev(rdev, mddev->major_version);
1757         }
1758         printk("md:     **********************************\n");
1759         printk("\n");
1760 }
1761
1762
1763 static void sync_sbs(mddev_t * mddev, int nospares)
1764 {
1765         /* Update each superblock (in-memory image), but
1766          * if we are allowed to, skip spares which already
1767          * have the right event counter, or have one earlier
1768          * (which would mean they aren't being marked as dirty
1769          * with the rest of the array)
1770          */
1771         mdk_rdev_t *rdev;
1772
1773         list_for_each_entry(rdev, &mddev->disks, same_set) {
1774                 if (rdev->sb_events == mddev->events ||
1775                     (nospares &&
1776                      rdev->raid_disk < 0 &&
1777                      (rdev->sb_events&1)==0 &&
1778                      rdev->sb_events+1 == mddev->events)) {
1779                         /* Don't update this superblock */
1780                         rdev->sb_loaded = 2;
1781                 } else {
1782                         super_types[mddev->major_version].
1783                                 sync_super(mddev, rdev);
1784                         rdev->sb_loaded = 1;
1785                 }
1786         }
1787 }
1788
1789 static void md_update_sb(mddev_t * mddev, int force_change)
1790 {
1791         mdk_rdev_t *rdev;
1792         int sync_req;
1793         int nospares = 0;
1794
1795         if (mddev->external)
1796                 return;
1797 repeat:
1798         spin_lock_irq(&mddev->write_lock);
1799
1800         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1801         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1802                 force_change = 1;
1803         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1804                 /* just a clean<-> dirty transition, possibly leave spares alone,
1805                  * though if events isn't the right even/odd, we will have to do
1806                  * spares after all
1807                  */
1808                 nospares = 1;
1809         if (force_change)
1810                 nospares = 0;
1811         if (mddev->degraded)
1812                 /* If the array is degraded, then skipping spares is both
1813                  * dangerous and fairly pointless.
1814                  * Dangerous because a device that was removed from the array
1815                  * might have a event_count that still looks up-to-date,
1816                  * so it can be re-added without a resync.
1817                  * Pointless because if there are any spares to skip,
1818                  * then a recovery will happen and soon that array won't
1819                  * be degraded any more and the spare can go back to sleep then.
1820                  */
1821                 nospares = 0;
1822
1823         sync_req = mddev->in_sync;
1824         mddev->utime = get_seconds();
1825
1826         /* If this is just a dirty<->clean transition, and the array is clean
1827          * and 'events' is odd, we can roll back to the previous clean state */
1828         if (nospares
1829             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1830             && (mddev->events & 1)
1831             && mddev->events != 1)
1832                 mddev->events--;
1833         else {
1834                 /* otherwise we have to go forward and ... */
1835                 mddev->events ++;
1836                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1837                         /* .. if the array isn't clean, insist on an odd 'events' */
1838                         if ((mddev->events&1)==0) {
1839                                 mddev->events++;
1840                                 nospares = 0;
1841                         }
1842                 } else {
1843                         /* otherwise insist on an even 'events' (for clean states) */
1844                         if ((mddev->events&1)) {
1845                                 mddev->events++;
1846                                 nospares = 0;
1847                         }
1848                 }
1849         }
1850
1851         if (!mddev->events) {
1852                 /*
1853                  * oops, this 64-bit counter should never wrap.
1854                  * Either we are in around ~1 trillion A.C., assuming
1855                  * 1 reboot per second, or we have a bug:
1856                  */
1857                 MD_BUG();
1858                 mddev->events --;
1859         }
1860
1861         /*
1862          * do not write anything to disk if using
1863          * nonpersistent superblocks
1864          */
1865         if (!mddev->persistent) {
1866                 if (!mddev->external)
1867                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1868
1869                 spin_unlock_irq(&mddev->write_lock);
1870                 wake_up(&mddev->sb_wait);
1871                 return;
1872         }
1873         sync_sbs(mddev, nospares);
1874         spin_unlock_irq(&mddev->write_lock);
1875
1876         dprintk(KERN_INFO
1877                 "md: updating %s RAID superblock on device (in sync %d)\n",
1878                 mdname(mddev),mddev->in_sync);
1879
1880         bitmap_update_sb(mddev->bitmap);
1881         list_for_each_entry(rdev, &mddev->disks, same_set) {
1882                 char b[BDEVNAME_SIZE];
1883                 dprintk(KERN_INFO "md: ");
1884                 if (rdev->sb_loaded != 1)
1885                         continue; /* no noise on spare devices */
1886                 if (test_bit(Faulty, &rdev->flags))
1887                         dprintk("(skipping faulty ");
1888
1889                 dprintk("%s ", bdevname(rdev->bdev,b));
1890                 if (!test_bit(Faulty, &rdev->flags)) {
1891                         md_super_write(mddev,rdev,
1892                                        rdev->sb_start, rdev->sb_size,
1893                                        rdev->sb_page);
1894                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1895                                 bdevname(rdev->bdev,b),
1896                                 (unsigned long long)rdev->sb_start);
1897                         rdev->sb_events = mddev->events;
1898
1899                 } else
1900                         dprintk(")\n");
1901                 if (mddev->level == LEVEL_MULTIPATH)
1902                         /* only need to write one superblock... */
1903                         break;
1904         }
1905         md_super_wait(mddev);
1906         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1907
1908         spin_lock_irq(&mddev->write_lock);
1909         if (mddev->in_sync != sync_req ||
1910             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1911                 /* have to write it out again */
1912                 spin_unlock_irq(&mddev->write_lock);
1913                 goto repeat;
1914         }
1915         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1916         spin_unlock_irq(&mddev->write_lock);
1917         wake_up(&mddev->sb_wait);
1918
1919 }
1920
1921 /* words written to sysfs files may, or may not, be \n terminated.
1922  * We want to accept with case. For this we use cmd_match.
1923  */
1924 static int cmd_match(const char *cmd, const char *str)
1925 {
1926         /* See if cmd, written into a sysfs file, matches
1927          * str.  They must either be the same, or cmd can
1928          * have a trailing newline
1929          */
1930         while (*cmd && *str && *cmd == *str) {
1931                 cmd++;
1932                 str++;
1933         }
1934         if (*cmd == '\n')
1935                 cmd++;
1936         if (*str || *cmd)
1937                 return 0;
1938         return 1;
1939 }
1940
1941 struct rdev_sysfs_entry {
1942         struct attribute attr;
1943         ssize_t (*show)(mdk_rdev_t *, char *);
1944         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1945 };
1946
1947 static ssize_t
1948 state_show(mdk_rdev_t *rdev, char *page)
1949 {
1950         char *sep = "";
1951         size_t len = 0;
1952
1953         if (test_bit(Faulty, &rdev->flags)) {
1954                 len+= sprintf(page+len, "%sfaulty",sep);
1955                 sep = ",";
1956         }
1957         if (test_bit(In_sync, &rdev->flags)) {
1958                 len += sprintf(page+len, "%sin_sync",sep);
1959                 sep = ",";
1960         }
1961         if (test_bit(WriteMostly, &rdev->flags)) {
1962                 len += sprintf(page+len, "%swrite_mostly",sep);
1963                 sep = ",";
1964         }
1965         if (test_bit(Blocked, &rdev->flags)) {
1966                 len += sprintf(page+len, "%sblocked", sep);
1967                 sep = ",";
1968         }
1969         if (!test_bit(Faulty, &rdev->flags) &&
1970             !test_bit(In_sync, &rdev->flags)) {
1971                 len += sprintf(page+len, "%sspare", sep);
1972                 sep = ",";
1973         }
1974         return len+sprintf(page+len, "\n");
1975 }
1976
1977 static ssize_t
1978 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1979 {
1980         /* can write
1981          *  faulty  - simulates and error
1982          *  remove  - disconnects the device
1983          *  writemostly - sets write_mostly
1984          *  -writemostly - clears write_mostly
1985          *  blocked - sets the Blocked flag
1986          *  -blocked - clears the Blocked flag
1987          */
1988         int err = -EINVAL;
1989         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1990                 md_error(rdev->mddev, rdev);
1991                 err = 0;
1992         } else if (cmd_match(buf, "remove")) {
1993                 if (rdev->raid_disk >= 0)
1994                         err = -EBUSY;
1995                 else {
1996                         mddev_t *mddev = rdev->mddev;
1997                         kick_rdev_from_array(rdev);
1998                         if (mddev->pers)
1999                                 md_update_sb(mddev, 1);
2000                         md_new_event(mddev);
2001                         err = 0;
2002                 }
2003         } else if (cmd_match(buf, "writemostly")) {
2004                 set_bit(WriteMostly, &rdev->flags);
2005                 err = 0;
2006         } else if (cmd_match(buf, "-writemostly")) {
2007                 clear_bit(WriteMostly, &rdev->flags);
2008                 err = 0;
2009         } else if (cmd_match(buf, "blocked")) {
2010                 set_bit(Blocked, &rdev->flags);
2011                 err = 0;
2012         } else if (cmd_match(buf, "-blocked")) {
2013                 clear_bit(Blocked, &rdev->flags);
2014                 wake_up(&rdev->blocked_wait);
2015                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2016                 md_wakeup_thread(rdev->mddev->thread);
2017
2018                 err = 0;
2019         }
2020         if (!err && rdev->sysfs_state)
2021                 sysfs_notify_dirent(rdev->sysfs_state);
2022         return err ? err : len;
2023 }
2024 static struct rdev_sysfs_entry rdev_state =
2025 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2026
2027 static ssize_t
2028 errors_show(mdk_rdev_t *rdev, char *page)
2029 {
2030         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2031 }
2032
2033 static ssize_t
2034 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2035 {
2036         char *e;
2037         unsigned long n = simple_strtoul(buf, &e, 10);
2038         if (*buf && (*e == 0 || *e == '\n')) {
2039                 atomic_set(&rdev->corrected_errors, n);
2040                 return len;
2041         }
2042         return -EINVAL;
2043 }
2044 static struct rdev_sysfs_entry rdev_errors =
2045 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2046
2047 static ssize_t
2048 slot_show(mdk_rdev_t *rdev, char *page)
2049 {
2050         if (rdev->raid_disk < 0)
2051                 return sprintf(page, "none\n");
2052         else
2053                 return sprintf(page, "%d\n", rdev->raid_disk);
2054 }
2055
2056 static ssize_t
2057 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2058 {
2059         char *e;
2060         int err;
2061         char nm[20];
2062         int slot = simple_strtoul(buf, &e, 10);
2063         if (strncmp(buf, "none", 4)==0)
2064                 slot = -1;
2065         else if (e==buf || (*e && *e!= '\n'))
2066                 return -EINVAL;
2067         if (rdev->mddev->pers && slot == -1) {
2068                 /* Setting 'slot' on an active array requires also
2069                  * updating the 'rd%d' link, and communicating
2070                  * with the personality with ->hot_*_disk.
2071                  * For now we only support removing
2072                  * failed/spare devices.  This normally happens automatically,
2073                  * but not when the metadata is externally managed.
2074                  */
2075                 if (rdev->raid_disk == -1)
2076                         return -EEXIST;
2077                 /* personality does all needed checks */
2078                 if (rdev->mddev->pers->hot_add_disk == NULL)
2079                         return -EINVAL;
2080                 err = rdev->mddev->pers->
2081                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2082                 if (err)
2083                         return err;
2084                 sprintf(nm, "rd%d", rdev->raid_disk);
2085                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2086                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2087                 md_wakeup_thread(rdev->mddev->thread);
2088         } else if (rdev->mddev->pers) {
2089                 mdk_rdev_t *rdev2;
2090                 /* Activating a spare .. or possibly reactivating
2091                  * if we every get bitmaps working here.
2092                  */
2093
2094                 if (rdev->raid_disk != -1)
2095                         return -EBUSY;
2096
2097                 if (rdev->mddev->pers->hot_add_disk == NULL)
2098                         return -EINVAL;
2099
2100                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2101                         if (rdev2->raid_disk == slot)
2102                                 return -EEXIST;
2103
2104                 rdev->raid_disk = slot;
2105                 if (test_bit(In_sync, &rdev->flags))
2106                         rdev->saved_raid_disk = slot;
2107                 else
2108                         rdev->saved_raid_disk = -1;
2109                 err = rdev->mddev->pers->
2110                         hot_add_disk(rdev->mddev, rdev);
2111                 if (err) {
2112                         rdev->raid_disk = -1;
2113                         return err;
2114                 } else
2115                         sysfs_notify_dirent(rdev->sysfs_state);
2116                 sprintf(nm, "rd%d", rdev->raid_disk);
2117                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2118                         printk(KERN_WARNING
2119                                "md: cannot register "
2120                                "%s for %s\n",
2121                                nm, mdname(rdev->mddev));
2122
2123                 /* don't wakeup anyone, leave that to userspace. */
2124         } else {
2125                 if (slot >= rdev->mddev->raid_disks)
2126                         return -ENOSPC;
2127                 rdev->raid_disk = slot;
2128                 /* assume it is working */
2129                 clear_bit(Faulty, &rdev->flags);
2130                 clear_bit(WriteMostly, &rdev->flags);
2131                 set_bit(In_sync, &rdev->flags);
2132                 sysfs_notify_dirent(rdev->sysfs_state);
2133         }
2134         return len;
2135 }
2136
2137
2138 static struct rdev_sysfs_entry rdev_slot =
2139 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2140
2141 static ssize_t
2142 offset_show(mdk_rdev_t *rdev, char *page)
2143 {
2144         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2145 }
2146
2147 static ssize_t
2148 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2149 {
2150         char *e;
2151         unsigned long long offset = simple_strtoull(buf, &e, 10);
2152         if (e==buf || (*e && *e != '\n'))
2153                 return -EINVAL;
2154         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2155                 return -EBUSY;
2156         if (rdev->size && rdev->mddev->external)
2157                 /* Must set offset before size, so overlap checks
2158                  * can be sane */
2159                 return -EBUSY;
2160         rdev->data_offset = offset;
2161         return len;
2162 }
2163
2164 static struct rdev_sysfs_entry rdev_offset =
2165 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2166
2167 static ssize_t
2168 rdev_size_show(mdk_rdev_t *rdev, char *page)
2169 {
2170         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2171 }
2172
2173 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2174 {
2175         /* check if two start/length pairs overlap */
2176         if (s1+l1 <= s2)
2177                 return 0;
2178         if (s2+l2 <= s1)
2179                 return 0;
2180         return 1;
2181 }
2182
2183 static ssize_t
2184 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2185 {
2186         unsigned long long size;
2187         unsigned long long oldsize = rdev->size;
2188         mddev_t *my_mddev = rdev->mddev;
2189
2190         if (strict_strtoull(buf, 10, &size) < 0)
2191                 return -EINVAL;
2192         if (my_mddev->pers && rdev->raid_disk >= 0) {
2193                 if (my_mddev->persistent) {
2194                         size = super_types[my_mddev->major_version].
2195                                 rdev_size_change(rdev, size * 2);
2196                         if (!size)
2197                                 return -EBUSY;
2198                 } else if (!size) {
2199                         size = (rdev->bdev->bd_inode->i_size >> 10);
2200                         size -= rdev->data_offset/2;
2201                 }
2202         }
2203         if (size < my_mddev->size)
2204                 return -EINVAL; /* component must fit device */
2205
2206         rdev->size = size;
2207         if (size > oldsize && my_mddev->external) {
2208                 /* need to check that all other rdevs with the same ->bdev
2209                  * do not overlap.  We need to unlock the mddev to avoid
2210                  * a deadlock.  We have already changed rdev->size, and if
2211                  * we have to change it back, we will have the lock again.
2212                  */
2213                 mddev_t *mddev;
2214                 int overlap = 0;
2215                 struct list_head *tmp;
2216
2217                 mddev_unlock(my_mddev);
2218                 for_each_mddev(mddev, tmp) {
2219                         mdk_rdev_t *rdev2;
2220
2221                         mddev_lock(mddev);
2222                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2223                                 if (test_bit(AllReserved, &rdev2->flags) ||
2224                                     (rdev->bdev == rdev2->bdev &&
2225                                      rdev != rdev2 &&
2226                                      overlaps(rdev->data_offset, rdev->size * 2,
2227                                               rdev2->data_offset,
2228                                               rdev2->size * 2))) {
2229                                         overlap = 1;
2230                                         break;
2231                                 }
2232                         mddev_unlock(mddev);
2233                         if (overlap) {
2234                                 mddev_put(mddev);
2235                                 break;
2236                         }
2237                 }
2238                 mddev_lock(my_mddev);
2239                 if (overlap) {
2240                         /* Someone else could have slipped in a size
2241                          * change here, but doing so is just silly.
2242                          * We put oldsize back because we *know* it is
2243                          * safe, and trust userspace not to race with
2244                          * itself
2245                          */
2246                         rdev->size = oldsize;
2247                         return -EBUSY;
2248                 }
2249         }
2250         return len;
2251 }
2252
2253 static struct rdev_sysfs_entry rdev_size =
2254 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2255
2256 static struct attribute *rdev_default_attrs[] = {
2257         &rdev_state.attr,
2258         &rdev_errors.attr,
2259         &rdev_slot.attr,
2260         &rdev_offset.attr,
2261         &rdev_size.attr,
2262         NULL,
2263 };
2264 static ssize_t
2265 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2266 {
2267         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2268         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2269         mddev_t *mddev = rdev->mddev;
2270         ssize_t rv;
2271
2272         if (!entry->show)
2273                 return -EIO;
2274
2275         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2276         if (!rv) {
2277                 if (rdev->mddev == NULL)
2278                         rv = -EBUSY;
2279                 else
2280                         rv = entry->show(rdev, page);
2281                 mddev_unlock(mddev);
2282         }
2283         return rv;
2284 }
2285
2286 static ssize_t
2287 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2288               const char *page, size_t length)
2289 {
2290         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2291         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2292         ssize_t rv;
2293         mddev_t *mddev = rdev->mddev;
2294
2295         if (!entry->store)
2296                 return -EIO;
2297         if (!capable(CAP_SYS_ADMIN))
2298                 return -EACCES;
2299         rv = mddev ? mddev_lock(mddev): -EBUSY;
2300         if (!rv) {
2301                 if (rdev->mddev == NULL)
2302                         rv = -EBUSY;
2303                 else
2304                         rv = entry->store(rdev, page, length);
2305                 mddev_unlock(mddev);
2306         }
2307         return rv;
2308 }
2309
2310 static void rdev_free(struct kobject *ko)
2311 {
2312         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2313         kfree(rdev);
2314 }
2315 static struct sysfs_ops rdev_sysfs_ops = {
2316         .show           = rdev_attr_show,
2317         .store          = rdev_attr_store,
2318 };
2319 static struct kobj_type rdev_ktype = {
2320         .release        = rdev_free,
2321         .sysfs_ops      = &rdev_sysfs_ops,
2322         .default_attrs  = rdev_default_attrs,
2323 };
2324
2325 /*
2326  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2327  *
2328  * mark the device faulty if:
2329  *
2330  *   - the device is nonexistent (zero size)
2331  *   - the device has no valid superblock
2332  *
2333  * a faulty rdev _never_ has rdev->sb set.
2334  */
2335 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2336 {
2337         char b[BDEVNAME_SIZE];
2338         int err;
2339         mdk_rdev_t *rdev;
2340         sector_t size;
2341
2342         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2343         if (!rdev) {
2344                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2345                 return ERR_PTR(-ENOMEM);
2346         }
2347
2348         if ((err = alloc_disk_sb(rdev)))
2349                 goto abort_free;
2350
2351         err = lock_rdev(rdev, newdev, super_format == -2);
2352         if (err)
2353                 goto abort_free;
2354
2355         kobject_init(&rdev->kobj, &rdev_ktype);
2356
2357         rdev->desc_nr = -1;
2358         rdev->saved_raid_disk = -1;
2359         rdev->raid_disk = -1;
2360         rdev->flags = 0;
2361         rdev->data_offset = 0;
2362         rdev->sb_events = 0;
2363         atomic_set(&rdev->nr_pending, 0);
2364         atomic_set(&rdev->read_errors, 0);
2365         atomic_set(&rdev->corrected_errors, 0);
2366
2367         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2368         if (!size) {
2369                 printk(KERN_WARNING
2370                         "md: %s has zero or unknown size, marking faulty!\n",
2371                         bdevname(rdev->bdev,b));
2372                 err = -EINVAL;
2373                 goto abort_free;
2374         }
2375
2376         if (super_format >= 0) {
2377                 err = super_types[super_format].
2378                         load_super(rdev, NULL, super_minor);
2379                 if (err == -EINVAL) {
2380                         printk(KERN_WARNING
2381                                 "md: %s does not have a valid v%d.%d "
2382                                "superblock, not importing!\n",
2383                                 bdevname(rdev->bdev,b),
2384                                super_format, super_minor);
2385                         goto abort_free;
2386                 }
2387                 if (err < 0) {
2388                         printk(KERN_WARNING
2389                                 "md: could not read %s's sb, not importing!\n",
2390                                 bdevname(rdev->bdev,b));
2391                         goto abort_free;
2392                 }
2393         }
2394
2395         INIT_LIST_HEAD(&rdev->same_set);
2396         init_waitqueue_head(&rdev->blocked_wait);
2397
2398         return rdev;
2399
2400 abort_free:
2401         if (rdev->sb_page) {
2402                 if (rdev->bdev)
2403                         unlock_rdev(rdev);
2404                 free_disk_sb(rdev);
2405         }
2406         kfree(rdev);
2407         return ERR_PTR(err);
2408 }
2409
2410 /*
2411  * Check a full RAID array for plausibility
2412  */
2413
2414
2415 static void analyze_sbs(mddev_t * mddev)
2416 {
2417         int i;
2418         mdk_rdev_t *rdev, *freshest, *tmp;
2419         char b[BDEVNAME_SIZE];
2420
2421         freshest = NULL;
2422         rdev_for_each(rdev, tmp, mddev)
2423                 switch (super_types[mddev->major_version].
2424                         load_super(rdev, freshest, mddev->minor_version)) {
2425                 case 1:
2426                         freshest = rdev;
2427                         break;
2428                 case 0:
2429                         break;
2430                 default:
2431                         printk( KERN_ERR \
2432                                 "md: fatal superblock inconsistency in %s"
2433                                 " -- removing from array\n",
2434                                 bdevname(rdev->bdev,b));
2435                         kick_rdev_from_array(rdev);
2436                 }
2437
2438
2439         super_types[mddev->major_version].
2440                 validate_super(mddev, freshest);
2441
2442         i = 0;
2443         rdev_for_each(rdev, tmp, mddev) {
2444                 if (rdev != freshest)
2445                         if (super_types[mddev->major_version].
2446                             validate_super(mddev, rdev)) {
2447                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2448                                         " from array!\n",
2449                                         bdevname(rdev->bdev,b));
2450                                 kick_rdev_from_array(rdev);
2451                                 continue;
2452                         }
2453                 if (mddev->level == LEVEL_MULTIPATH) {
2454                         rdev->desc_nr = i++;
2455                         rdev->raid_disk = rdev->desc_nr;
2456                         set_bit(In_sync, &rdev->flags);
2457                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2458                         rdev->raid_disk = -1;
2459                         clear_bit(In_sync, &rdev->flags);
2460                 }
2461         }
2462
2463
2464
2465         if (mddev->recovery_cp != MaxSector &&
2466             mddev->level >= 1)
2467                 printk(KERN_ERR "md: %s: raid array is not clean"
2468                        " -- starting background reconstruction\n",
2469                        mdname(mddev));
2470
2471 }
2472
2473 static void md_safemode_timeout(unsigned long data);
2474
2475 static ssize_t
2476 safe_delay_show(mddev_t *mddev, char *page)
2477 {
2478         int msec = (mddev->safemode_delay*1000)/HZ;
2479         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2480 }
2481 static ssize_t
2482 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2483 {
2484         int scale=1;
2485         int dot=0;
2486         int i;
2487         unsigned long msec;
2488         char buf[30];
2489
2490         /* remove a period, and count digits after it */
2491         if (len >= sizeof(buf))
2492                 return -EINVAL;
2493         strlcpy(buf, cbuf, sizeof(buf));
2494         for (i=0; i<len; i++) {
2495                 if (dot) {
2496                         if (isdigit(buf[i])) {
2497                                 buf[i-1] = buf[i];
2498                                 scale *= 10;
2499                         }
2500                         buf[i] = 0;
2501                 } else if (buf[i] == '.') {
2502                         dot=1;
2503                         buf[i] = 0;
2504                 }
2505         }
2506         if (strict_strtoul(buf, 10, &msec) < 0)
2507                 return -EINVAL;
2508         msec = (msec * 1000) / scale;
2509         if (msec == 0)
2510                 mddev->safemode_delay = 0;
2511         else {
2512                 unsigned long old_delay = mddev->safemode_delay;
2513                 mddev->safemode_delay = (msec*HZ)/1000;
2514                 if (mddev->safemode_delay == 0)
2515                         mddev->safemode_delay = 1;
2516                 if (mddev->safemode_delay < old_delay)
2517                         md_safemode_timeout((unsigned long)mddev);
2518         }
2519         return len;
2520 }
2521 static struct md_sysfs_entry md_safe_delay =
2522 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2523
2524 static ssize_t
2525 level_show(mddev_t *mddev, char *page)
2526 {
2527         struct mdk_personality *p = mddev->pers;
2528         if (p)
2529                 return sprintf(page, "%s\n", p->name);
2530         else if (mddev->clevel[0])
2531                 return sprintf(page, "%s\n", mddev->clevel);
2532         else if (mddev->level != LEVEL_NONE)
2533                 return sprintf(page, "%d\n", mddev->level);
2534         else
2535                 return 0;
2536 }
2537
2538 static ssize_t
2539 level_store(mddev_t *mddev, const char *buf, size_t len)
2540 {
2541         ssize_t rv = len;
2542         if (mddev->pers)
2543                 return -EBUSY;
2544         if (len == 0)
2545                 return 0;
2546         if (len >= sizeof(mddev->clevel))
2547                 return -ENOSPC;
2548         strncpy(mddev->clevel, buf, len);
2549         if (mddev->clevel[len-1] == '\n')
2550                 len--;
2551         mddev->clevel[len] = 0;
2552         mddev->level = LEVEL_NONE;
2553         return rv;
2554 }
2555
2556 static struct md_sysfs_entry md_level =
2557 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2558
2559
2560 static ssize_t
2561 layout_show(mddev_t *mddev, char *page)
2562 {
2563         /* just a number, not meaningful for all levels */
2564         if (mddev->reshape_position != MaxSector &&
2565             mddev->layout != mddev->new_layout)
2566                 return sprintf(page, "%d (%d)\n",
2567                                mddev->new_layout, mddev->layout);
2568         return sprintf(page, "%d\n", mddev->layout);
2569 }
2570
2571 static ssize_t
2572 layout_store(mddev_t *mddev, const char *buf, size_t len)
2573 {
2574         char *e;
2575         unsigned long n = simple_strtoul(buf, &e, 10);
2576
2577         if (!*buf || (*e && *e != '\n'))
2578                 return -EINVAL;
2579
2580         if (mddev->pers)
2581                 return -EBUSY;
2582         if (mddev->reshape_position != MaxSector)
2583                 mddev->new_layout = n;
2584         else
2585                 mddev->layout = n;
2586         return len;
2587 }
2588 static struct md_sysfs_entry md_layout =
2589 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2590
2591
2592 static ssize_t
2593 raid_disks_show(mddev_t *mddev, char *page)
2594 {
2595         if (mddev->raid_disks == 0)
2596                 return 0;
2597         if (mddev->reshape_position != MaxSector &&
2598             mddev->delta_disks != 0)
2599                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2600                                mddev->raid_disks - mddev->delta_disks);
2601         return sprintf(page, "%d\n", mddev->raid_disks);
2602 }
2603
2604 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2605
2606 static ssize_t
2607 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2608 {
2609         char *e;
2610         int rv = 0;
2611         unsigned long n = simple_strtoul(buf, &e, 10);
2612
2613         if (!*buf || (*e && *e != '\n'))
2614                 return -EINVAL;
2615
2616         if (mddev->pers)
2617                 rv = update_raid_disks(mddev, n);
2618         else if (mddev->reshape_position != MaxSector) {
2619                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2620                 mddev->delta_disks = n - olddisks;
2621                 mddev->raid_disks = n;
2622         } else
2623                 mddev->raid_disks = n;
2624         return rv ? rv : len;
2625 }
2626 static struct md_sysfs_entry md_raid_disks =
2627 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2628
2629 static ssize_t
2630 chunk_size_show(mddev_t *mddev, char *page)
2631 {
2632         if (mddev->reshape_position != MaxSector &&
2633             mddev->chunk_size != mddev->new_chunk)
2634                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2635                                mddev->chunk_size);
2636         return sprintf(page, "%d\n", mddev->chunk_size);
2637 }
2638
2639 static ssize_t
2640 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2641 {
2642         /* can only set chunk_size if array is not yet active */
2643         char *e;
2644         unsigned long n = simple_strtoul(buf, &e, 10);
2645
2646         if (!*buf || (*e && *e != '\n'))
2647                 return -EINVAL;
2648
2649         if (mddev->pers)
2650                 return -EBUSY;
2651         else if (mddev->reshape_position != MaxSector)
2652                 mddev->new_chunk = n;
2653         else
2654                 mddev->chunk_size = n;
2655         return len;
2656 }
2657 static struct md_sysfs_entry md_chunk_size =
2658 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2659
2660 static ssize_t
2661 resync_start_show(mddev_t *mddev, char *page)
2662 {
2663         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2664 }
2665
2666 static ssize_t
2667 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2668 {
2669         char *e;
2670         unsigned long long n = simple_strtoull(buf, &e, 10);
2671
2672         if (mddev->pers)
2673                 return -EBUSY;
2674         if (!*buf || (*e && *e != '\n'))
2675                 return -EINVAL;
2676
2677         mddev->recovery_cp = n;
2678         return len;
2679 }
2680 static struct md_sysfs_entry md_resync_start =
2681 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2682
2683 /*
2684  * The array state can be:
2685  *
2686  * clear
2687  *     No devices, no size, no level
2688  *     Equivalent to STOP_ARRAY ioctl
2689  * inactive
2690  *     May have some settings, but array is not active
2691  *        all IO results in error
2692  *     When written, doesn't tear down array, but just stops it
2693  * suspended (not supported yet)
2694  *     All IO requests will block. The array can be reconfigured.
2695  *     Writing this, if accepted, will block until array is quiescent
2696  * readonly
2697  *     no resync can happen.  no superblocks get written.
2698  *     write requests fail
2699  * read-auto
2700  *     like readonly, but behaves like 'clean' on a write request.
2701  *
2702  * clean - no pending writes, but otherwise active.
2703  *     When written to inactive array, starts without resync
2704  *     If a write request arrives then
2705  *       if metadata is known, mark 'dirty' and switch to 'active'.
2706  *       if not known, block and switch to write-pending
2707  *     If written to an active array that has pending writes, then fails.
2708  * active
2709  *     fully active: IO and resync can be happening.
2710  *     When written to inactive array, starts with resync
2711  *
2712  * write-pending
2713  *     clean, but writes are blocked waiting for 'active' to be written.
2714  *
2715  * active-idle
2716  *     like active, but no writes have been seen for a while (100msec).
2717  *
2718  */
2719 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2720                    write_pending, active_idle, bad_word};
2721 static char *array_states[] = {
2722         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2723         "write-pending", "active-idle", NULL };
2724
2725 static int match_word(const char *word, char **list)
2726 {
2727         int n;
2728         for (n=0; list[n]; n++)
2729                 if (cmd_match(word, list[n]))
2730                         break;
2731         return n;
2732 }
2733
2734 static ssize_t
2735 array_state_show(mddev_t *mddev, char *page)
2736 {
2737         enum array_state st = inactive;
2738
2739         if (mddev->pers)
2740                 switch(mddev->ro) {
2741                 case 1:
2742                         st = readonly;
2743                         break;
2744                 case 2:
2745                         st = read_auto;
2746                         break;
2747                 case 0:
2748                         if (mddev->in_sync)
2749                                 st = clean;
2750                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2751                                 st = write_pending;
2752                         else if (mddev->safemode)
2753                                 st = active_idle;
2754                         else
2755                                 st = active;
2756                 }
2757         else {
2758                 if (list_empty(&mddev->disks) &&
2759                     mddev->raid_disks == 0 &&
2760                     mddev->size == 0)
2761                         st = clear;
2762                 else
2763                         st = inactive;
2764         }
2765         return sprintf(page, "%s\n", array_states[st]);
2766 }
2767
2768 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2769 static int do_md_run(mddev_t * mddev);
2770 static int restart_array(mddev_t *mddev);
2771
2772 static ssize_t
2773 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2774 {
2775         int err = -EINVAL;
2776         enum array_state st = match_word(buf, array_states);
2777         switch(st) {
2778         case bad_word:
2779                 break;
2780         case clear:
2781                 /* stopping an active array */
2782                 if (atomic_read(&mddev->openers) > 0)
2783                         return -EBUSY;
2784                 err = do_md_stop(mddev, 0, 0);
2785                 break;
2786         case inactive:
2787                 /* stopping an active array */
2788                 if (mddev->pers) {
2789                         if (atomic_read(&mddev->openers) > 0)
2790                                 return -EBUSY;
2791                         err = do_md_stop(mddev, 2, 0);
2792                 } else
2793                         err = 0; /* already inactive */
2794                 break;
2795         case suspended:
2796                 break; /* not supported yet */
2797         case readonly:
2798                 if (mddev->pers)
2799                         err = do_md_stop(mddev, 1, 0);
2800                 else {
2801                         mddev->ro = 1;
2802                         set_disk_ro(mddev->gendisk, 1);
2803                         err = do_md_run(mddev);
2804                 }
2805                 break;
2806         case read_auto:
2807                 if (mddev->pers) {
2808                         if (mddev->ro == 0)
2809                                 err = do_md_stop(mddev, 1, 0);
2810                         else if (mddev->ro == 1)
2811                                 err = restart_array(mddev);
2812                         if (err == 0) {
2813                                 mddev->ro = 2;
2814                                 set_disk_ro(mddev->gendisk, 0);
2815                         }
2816                 } else {
2817                         mddev->ro = 2;
2818                         err = do_md_run(mddev);
2819                 }
2820                 break;
2821         case clean:
2822                 if (mddev->pers) {
2823                         restart_array(mddev);
2824                         spin_lock_irq(&mddev->write_lock);
2825                         if (atomic_read(&mddev->writes_pending) == 0) {
2826                                 if (mddev->in_sync == 0) {
2827                                         mddev->in_sync = 1;
2828                                         if (mddev->safemode == 1)
2829                                                 mddev->safemode = 0;
2830                                         if (mddev->persistent)
2831                                                 set_bit(MD_CHANGE_CLEAN,
2832                                                         &mddev->flags);
2833                                 }
2834                                 err = 0;
2835                         } else
2836                                 err = -EBUSY;
2837                         spin_unlock_irq(&mddev->write_lock);
2838                 } else {
2839                         mddev->ro = 0;
2840                         mddev->recovery_cp = MaxSector;
2841                         err = do_md_run(mddev);
2842                 }
2843                 break;
2844         case active:
2845                 if (mddev->pers) {
2846                         restart_array(mddev);
2847                         if (mddev->external)
2848                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2849                         wake_up(&mddev->sb_wait);
2850                         err = 0;
2851                 } else {
2852                         mddev->ro = 0;
2853                         set_disk_ro(mddev->gendisk, 0);
2854                         err = do_md_run(mddev);
2855                 }
2856                 break;
2857         case write_pending:
2858         case active_idle:
2859                 /* these cannot be set */
2860                 break;
2861         }
2862         if (err)
2863                 return err;
2864         else {
2865                 sysfs_notify_dirent(mddev->sysfs_state);
2866                 return len;
2867         }
2868 }
2869 static struct md_sysfs_entry md_array_state =
2870 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2871
2872 static ssize_t
2873 null_show(mddev_t *mddev, char *page)
2874 {
2875         return -EINVAL;
2876 }
2877
2878 static ssize_t
2879 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2880 {
2881         /* buf must be %d:%d\n? giving major and minor numbers */
2882         /* The new device is added to the array.
2883          * If the array has a persistent superblock, we read the
2884          * superblock to initialise info and check validity.
2885          * Otherwise, only checking done is that in bind_rdev_to_array,
2886          * which mainly checks size.
2887          */
2888         char *e;
2889         int major = simple_strtoul(buf, &e, 10);
2890         int minor;
2891         dev_t dev;
2892         mdk_rdev_t *rdev;
2893         int err;
2894
2895         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2896                 return -EINVAL;
2897         minor = simple_strtoul(e+1, &e, 10);
2898         if (*e && *e != '\n')
2899                 return -EINVAL;
2900         dev = MKDEV(major, minor);
2901         if (major != MAJOR(dev) ||
2902             minor != MINOR(dev))
2903                 return -EOVERFLOW;
2904
2905
2906         if (mddev->persistent) {
2907                 rdev = md_import_device(dev, mddev->major_version,
2908                                         mddev->minor_version);
2909                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2910                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2911                                                        mdk_rdev_t, same_set);
2912                         err = super_types[mddev->major_version]
2913                                 .load_super(rdev, rdev0, mddev->minor_version);
2914                         if (err < 0)
2915                                 goto out;
2916                 }
2917         } else if (mddev->external)
2918                 rdev = md_import_device(dev, -2, -1);
2919         else
2920                 rdev = md_import_device(dev, -1, -1);
2921
2922         if (IS_ERR(rdev))
2923                 return PTR_ERR(rdev);
2924         err = bind_rdev_to_array(rdev, mddev);
2925  out:
2926         if (err)
2927                 export_rdev(rdev);
2928         return err ? err : len;
2929 }
2930
2931 static struct md_sysfs_entry md_new_device =
2932 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2933
2934 static ssize_t
2935 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2936 {
2937         char *end;
2938         unsigned long chunk, end_chunk;
2939
2940         if (!mddev->bitmap)
2941                 goto out;
2942         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2943         while (*buf) {
2944                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2945                 if (buf == end) break;
2946                 if (*end == '-') { /* range */
2947                         buf = end + 1;
2948                         end_chunk = simple_strtoul(buf, &end, 0);
2949                         if (buf == end) break;
2950                 }
2951                 if (*end && !isspace(*end)) break;
2952                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2953                 buf = end;
2954                 while (isspace(*buf)) buf++;
2955         }
2956         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2957 out:
2958         return len;
2959 }
2960
2961 static struct md_sysfs_entry md_bitmap =
2962 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2963
2964 static ssize_t
2965 size_show(mddev_t *mddev, char *page)
2966 {
2967         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2968 }
2969
2970 static int update_size(mddev_t *mddev, sector_t num_sectors);
2971
2972 static ssize_t
2973 size_store(mddev_t *mddev, const char *buf, size_t len)
2974 {
2975         /* If array is inactive, we can reduce the component size, but
2976          * not increase it (except from 0).
2977          * If array is active, we can try an on-line resize
2978          */
2979         char *e;
2980         int err = 0;
2981         unsigned long long size = simple_strtoull(buf, &e, 10);
2982         if (!*buf || *buf == '\n' ||
2983             (*e && *e != '\n'))
2984                 return -EINVAL;
2985
2986         if (mddev->pers) {
2987                 err = update_size(mddev, size * 2);
2988                 md_update_sb(mddev, 1);
2989         } else {
2990                 if (mddev->size == 0 ||
2991                     mddev->size > size)
2992                         mddev->size = size;
2993                 else
2994                         err = -ENOSPC;
2995         }
2996         return err ? err : len;
2997 }
2998
2999 static struct md_sysfs_entry md_size =
3000 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3001
3002
3003 /* Metdata version.
3004  * This is one of
3005  *   'none' for arrays with no metadata (good luck...)
3006  *   'external' for arrays with externally managed metadata,
3007  * or N.M for internally known formats
3008  */
3009 static ssize_t
3010 metadata_show(mddev_t *mddev, char *page)
3011 {
3012         if (mddev->persistent)
3013                 return sprintf(page, "%d.%d\n",
3014                                mddev->major_version, mddev->minor_version);
3015         else if (mddev->external)
3016                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3017         else
3018                 return sprintf(page, "none\n");
3019 }
3020
3021 static ssize_t
3022 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3023 {
3024         int major, minor;
3025         char *e;
3026         /* Changing the details of 'external' metadata is
3027          * always permitted.  Otherwise there must be
3028          * no devices attached to the array.
3029          */
3030         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3031                 ;
3032         else if (!list_empty(&mddev->disks))
3033                 return -EBUSY;
3034
3035         if (cmd_match(buf, "none")) {
3036                 mddev->persistent = 0;
3037                 mddev->external = 0;
3038                 mddev->major_version = 0;
3039                 mddev->minor_version = 90;
3040                 return len;
3041         }
3042         if (strncmp(buf, "external:", 9) == 0) {
3043                 size_t namelen = len-9;
3044                 if (namelen >= sizeof(mddev->metadata_type))
3045                         namelen = sizeof(mddev->metadata_type)-1;
3046                 strncpy(mddev->metadata_type, buf+9, namelen);
3047                 mddev->metadata_type[namelen] = 0;
3048                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3049                         mddev->metadata_type[--namelen] = 0;
3050                 mddev->persistent = 0;
3051                 mddev->external = 1;
3052                 mddev->major_version = 0;
3053                 mddev->minor_version = 90;
3054                 return len;
3055         }
3056         major = simple_strtoul(buf, &e, 10);
3057         if (e==buf || *e != '.')
3058                 return -EINVAL;
3059         buf = e+1;
3060         minor = simple_strtoul(buf, &e, 10);
3061         if (e==buf || (*e && *e != '\n') )
3062                 return -EINVAL;
3063         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3064                 return -ENOENT;
3065         mddev->major_version = major;
3066         mddev->minor_version = minor;
3067         mddev->persistent = 1;
3068         mddev->external = 0;
3069         return len;
3070 }
3071
3072 static struct md_sysfs_entry md_metadata =
3073 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3074
3075 static ssize_t
3076 action_show(mddev_t *mddev, char *page)
3077 {
3078         char *type = "idle";
3079         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3080             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3081                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3082                         type = "reshape";
3083                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3084                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3085                                 type = "resync";
3086                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3087                                 type = "check";
3088                         else
3089                                 type = "repair";
3090                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3091                         type = "recover";
3092         }
3093         return sprintf(page, "%s\n", type);
3094 }
3095
3096 static ssize_t
3097 action_store(mddev_t *mddev, const char *page, size_t len)
3098 {
3099         if (!mddev->pers || !mddev->pers->sync_request)
3100                 return -EINVAL;
3101
3102         if (cmd_match(page, "idle")) {
3103                 if (mddev->sync_thread) {
3104                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3105                         md_unregister_thread(mddev->sync_thread);
3106                         mddev->sync_thread = NULL;
3107                         mddev->recovery = 0;
3108                 }
3109         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3110                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3111                 return -EBUSY;
3112         else if (cmd_match(page, "resync"))
3113                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3114         else if (cmd_match(page, "recover")) {
3115                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3116                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3117         } else if (cmd_match(page, "reshape")) {
3118                 int err;
3119                 if (mddev->pers->start_reshape == NULL)
3120                         return -EINVAL;
3121                 err = mddev->pers->start_reshape(mddev);
3122                 if (err)
3123                         return err;
3124                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3125         } else {
3126                 if (cmd_match(page, "check"))
3127                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3128                 else if (!cmd_match(page, "repair"))
3129                         return -EINVAL;
3130                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3131                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3132         }
3133         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3134         md_wakeup_thread(mddev->thread);
3135         sysfs_notify_dirent(mddev->sysfs_action);
3136         return len;
3137 }
3138
3139 static ssize_t
3140 mismatch_cnt_show(mddev_t *mddev, char *page)
3141 {
3142         return sprintf(page, "%llu\n",
3143                        (unsigned long long) mddev->resync_mismatches);
3144 }
3145
3146 static struct md_sysfs_entry md_scan_mode =
3147 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3148
3149
3150 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3151
3152 static ssize_t
3153 sync_min_show(mddev_t *mddev, char *page)
3154 {
3155         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3156                        mddev->sync_speed_min ? "local": "system");
3157 }
3158
3159 static ssize_t
3160 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3161 {
3162         int min;
3163         char *e;
3164         if (strncmp(buf, "system", 6)==0) {
3165                 mddev->sync_speed_min = 0;
3166                 return len;
3167         }
3168         min = simple_strtoul(buf, &e, 10);
3169         if (buf == e || (*e && *e != '\n') || min <= 0)
3170                 return -EINVAL;
3171         mddev->sync_speed_min = min;
3172         return len;
3173 }
3174
3175 static struct md_sysfs_entry md_sync_min =
3176 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3177
3178 static ssize_t
3179 sync_max_show(mddev_t *mddev, char *page)
3180 {
3181         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3182                        mddev->sync_speed_max ? "local": "system");
3183 }
3184
3185 static ssize_t
3186 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3187 {
3188         int max;
3189         char *e;
3190         if (strncmp(buf, "system", 6)==0) {
3191                 mddev->sync_speed_max = 0;
3192                 return len;
3193         }
3194         max = simple_strtoul(buf, &e, 10);
3195         if (buf == e || (*e && *e != '\n') || max <= 0)
3196                 return -EINVAL;
3197         mddev->sync_speed_max = max;
3198         return len;
3199 }
3200
3201 static struct md_sysfs_entry md_sync_max =
3202 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3203
3204 static ssize_t
3205 degraded_show(mddev_t *mddev, char *page)
3206 {
3207         return sprintf(page, "%d\n", mddev->degraded);
3208 }
3209 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3210
3211 static ssize_t
3212 sync_force_parallel_show(mddev_t *mddev, char *page)
3213 {
3214         return sprintf(page, "%d\n", mddev->parallel_resync);
3215 }
3216
3217 static ssize_t
3218 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3219 {
3220         long n;
3221
3222         if (strict_strtol(buf, 10, &n))
3223                 return -EINVAL;
3224
3225         if (n != 0 && n != 1)
3226                 return -EINVAL;
3227
3228         mddev->parallel_resync = n;
3229
3230         if (mddev->sync_thread)
3231                 wake_up(&resync_wait);
3232
3233         return len;
3234 }
3235
3236 /* force parallel resync, even with shared block devices */
3237 static struct md_sysfs_entry md_sync_force_parallel =
3238 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3239        sync_force_parallel_show, sync_force_parallel_store);
3240
3241 static ssize_t
3242 sync_speed_show(mddev_t *mddev, char *page)
3243 {
3244         unsigned long resync, dt, db;
3245         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3246         dt = (jiffies - mddev->resync_mark) / HZ;
3247         if (!dt) dt++;
3248         db = resync - mddev->resync_mark_cnt;
3249         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3250 }
3251
3252 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3253
3254 static ssize_t
3255 sync_completed_show(mddev_t *mddev, char *page)
3256 {
3257         unsigned long max_blocks, resync;
3258
3259         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3260                 max_blocks = mddev->resync_max_sectors;
3261         else
3262                 max_blocks = mddev->size << 1;
3263
3264         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3265         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3266 }
3267
3268 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3269
3270 static ssize_t
3271 min_sync_show(mddev_t *mddev, char *page)
3272 {
3273         return sprintf(page, "%llu\n",
3274                        (unsigned long long)mddev->resync_min);
3275 }
3276 static ssize_t
3277 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3278 {
3279         unsigned long long min;
3280         if (strict_strtoull(buf, 10, &min))
3281                 return -EINVAL;
3282         if (min > mddev->resync_max)
3283                 return -EINVAL;
3284         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3285                 return -EBUSY;
3286
3287         /* Must be a multiple of chunk_size */
3288         if (mddev->chunk_size) {
3289                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3290                         return -EINVAL;
3291         }
3292         mddev->resync_min = min;
3293
3294         return len;
3295 }
3296
3297 static struct md_sysfs_entry md_min_sync =
3298 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3299
3300 static ssize_t
3301 max_sync_show(mddev_t *mddev, char *page)
3302 {
3303         if (mddev->resync_max == MaxSector)
3304                 return sprintf(page, "max\n");
3305         else
3306                 return sprintf(page, "%llu\n",
3307                                (unsigned long long)mddev->resync_max);
3308 }
3309 static ssize_t
3310 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3311 {
3312         if (strncmp(buf, "max", 3) == 0)
3313                 mddev->resync_max = MaxSector;
3314         else {
3315                 unsigned long long max;
3316                 if (strict_strtoull(buf, 10, &max))
3317                         return -EINVAL;
3318                 if (max < mddev->resync_min)
3319                         return -EINVAL;
3320                 if (max < mddev->resync_max &&
3321                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3322                         return -EBUSY;
3323
3324                 /* Must be a multiple of chunk_size */
3325                 if (mddev->chunk_size) {
3326                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3327                                 return -EINVAL;
3328                 }
3329                 mddev->resync_max = max;
3330         }
3331         wake_up(&mddev->recovery_wait);
3332         return len;
3333 }
3334
3335 static struct md_sysfs_entry md_max_sync =
3336 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3337
3338 static ssize_t
3339 suspend_lo_show(mddev_t *mddev, char *page)
3340 {
3341         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3342 }
3343
3344 static ssize_t
3345 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3346 {
3347         char *e;
3348         unsigned long long new = simple_strtoull(buf, &e, 10);
3349
3350         if (mddev->pers->quiesce == NULL)
3351                 return -EINVAL;
3352         if (buf == e || (*e && *e != '\n'))
3353                 return -EINVAL;
3354         if (new >= mddev->suspend_hi ||
3355             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3356                 mddev->suspend_lo = new;
3357                 mddev->pers->quiesce(mddev, 2);
3358                 return len;
3359         } else
3360                 return -EINVAL;
3361 }
3362 static struct md_sysfs_entry md_suspend_lo =
3363 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3364
3365
3366 static ssize_t
3367 suspend_hi_show(mddev_t *mddev, char *page)
3368 {
3369         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3370 }
3371
3372 static ssize_t
3373 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3374 {
3375         char *e;
3376         unsigned long long new = simple_strtoull(buf, &e, 10);
3377
3378         if (mddev->pers->quiesce == NULL)
3379                 return -EINVAL;
3380         if (buf == e || (*e && *e != '\n'))
3381                 return -EINVAL;
3382         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3383             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3384                 mddev->suspend_hi = new;
3385                 mddev->pers->quiesce(mddev, 1);
3386                 mddev->pers->quiesce(mddev, 0);
3387                 return len;
3388         } else
3389                 return -EINVAL;
3390 }
3391 static struct md_sysfs_entry md_suspend_hi =
3392 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3393
3394 static ssize_t
3395 reshape_position_show(mddev_t *mddev, char *page)
3396 {
3397         if (mddev->reshape_position != MaxSector)
3398                 return sprintf(page, "%llu\n",
3399                                (unsigned long long)mddev->reshape_position);
3400         strcpy(page, "none\n");
3401         return 5;
3402 }
3403
3404 static ssize_t
3405 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3406 {
3407         char *e;
3408         unsigned long long new = simple_strtoull(buf, &e, 10);
3409         if (mddev->pers)
3410                 return -EBUSY;
3411         if (buf == e || (*e && *e != '\n'))
3412                 return -EINVAL;
3413         mddev->reshape_position = new;
3414         mddev->delta_disks = 0;
3415         mddev->new_level = mddev->level;
3416         mddev->new_layout = mddev->layout;
3417         mddev->new_chunk = mddev->chunk_size;
3418         return len;
3419 }
3420
3421 static struct md_sysfs_entry md_reshape_position =
3422 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3423        reshape_position_store);
3424
3425
3426 static struct attribute *md_default_attrs[] = {
3427         &md_level.attr,
3428         &md_layout.attr,
3429         &md_raid_disks.attr,
3430         &md_chunk_size.attr,
3431         &md_size.attr,
3432         &md_resync_start.attr,
3433         &md_metadata.attr,
3434         &md_new_device.attr,
3435         &md_safe_delay.attr,
3436         &md_array_state.attr,
3437         &md_reshape_position.attr,
3438         NULL,
3439 };
3440
3441 static struct attribute *md_redundancy_attrs[] = {
3442         &md_scan_mode.attr,
3443         &md_mismatches.attr,
3444         &md_sync_min.attr,
3445         &md_sync_max.attr,
3446         &md_sync_speed.attr,
3447         &md_sync_force_parallel.attr,
3448         &md_sync_completed.attr,
3449         &md_min_sync.attr,
3450         &md_max_sync.attr,
3451         &md_suspend_lo.attr,
3452         &md_suspend_hi.attr,
3453         &md_bitmap.attr,
3454         &md_degraded.attr,
3455         NULL,
3456 };
3457 static struct attribute_group md_redundancy_group = {
3458         .name = NULL,
3459         .attrs = md_redundancy_attrs,
3460 };
3461
3462
3463 static ssize_t
3464 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3465 {
3466         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3467         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3468         ssize_t rv;
3469
3470         if (!entry->show)
3471                 return -EIO;
3472         rv = mddev_lock(mddev);
3473         if (!rv) {
3474                 rv = entry->show(mddev, page);
3475                 mddev_unlock(mddev);
3476         }
3477         return rv;
3478 }
3479
3480 static ssize_t
3481 md_attr_store(struct kobject *kobj, struct attribute *attr,
3482               const char *page, size_t length)
3483 {
3484         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3485         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3486         ssize_t rv;
3487
3488         if (!entry->store)
3489                 return -EIO;
3490         if (!capable(CAP_SYS_ADMIN))
3491                 return -EACCES;
3492         rv = mddev_lock(mddev);
3493         if (mddev->hold_active == UNTIL_IOCTL)
3494                 mddev->hold_active = 0;
3495         if (!rv) {
3496                 rv = entry->store(mddev, page, length);
3497                 mddev_unlock(mddev);
3498         }
3499         return rv;
3500 }
3501
3502 static void md_free(struct kobject *ko)
3503 {
3504         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3505
3506         if (mddev->sysfs_state)
3507                 sysfs_put(mddev->sysfs_state);
3508
3509         if (mddev->gendisk) {
3510                 del_gendisk(mddev->gendisk);
3511                 put_disk(mddev->gendisk);
3512         }
3513         if (mddev->queue)
3514                 blk_cleanup_queue(mddev->queue);
3515
3516         kfree(mddev);
3517 }
3518
3519 static struct sysfs_ops md_sysfs_ops = {
3520         .show   = md_attr_show,
3521         .store  = md_attr_store,
3522 };
3523 static struct kobj_type md_ktype = {
3524         .release        = md_free,
3525         .sysfs_ops      = &md_sysfs_ops,
3526         .default_attrs  = md_default_attrs,
3527 };
3528
3529 int mdp_major = 0;
3530
3531 static int md_alloc(dev_t dev, char *name)
3532 {
3533         static DEFINE_MUTEX(disks_mutex);
3534         mddev_t *mddev = mddev_find(dev);
3535         struct gendisk *disk;
3536         int partitioned;
3537         int shift;
3538         int unit;
3539         int error;
3540
3541         if (!mddev)
3542                 return -ENODEV;
3543
3544         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3545         shift = partitioned ? MdpMinorShift : 0;
3546         unit = MINOR(mddev->unit) >> shift;
3547
3548         /* wait for any previous instance if this device
3549          * to be completed removed (mddev_delayed_delete).
3550          */
3551         flush_scheduled_work();
3552
3553         mutex_lock(&disks_mutex);
3554         if (mddev->gendisk) {
3555                 mutex_unlock(&disks_mutex);
3556                 mddev_put(mddev);
3557                 return -EEXIST;
3558         }
3559
3560         if (name) {
3561                 /* Need to ensure that 'name' is not a duplicate.
3562                  */
3563                 mddev_t *mddev2;
3564                 spin_lock(&all_mddevs_lock);
3565
3566                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3567                         if (mddev2->gendisk &&
3568                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
3569                                 spin_unlock(&all_mddevs_lock);
3570                                 return -EEXIST;
3571                         }
3572                 spin_unlock(&all_mddevs_lock);
3573         }
3574
3575         mddev->queue = blk_alloc_queue(GFP_KERNEL);
3576         if (!mddev->queue) {
3577                 mutex_unlock(&disks_mutex);
3578                 mddev_put(mddev);
3579                 return -ENOMEM;
3580         }
3581         /* Can be unlocked because the queue is new: no concurrency */
3582         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3583
3584         blk_queue_make_request(mddev->queue, md_fail_request);
3585
3586         disk = alloc_disk(1 << shift);
3587         if (!disk) {
3588                 mutex_unlock(&disks_mutex);
3589                 blk_cleanup_queue(mddev->queue);
3590                 mddev->queue = NULL;
3591                 mddev_put(mddev);
3592                 return -ENOMEM;
3593         }
3594         disk->major = MAJOR(mddev->unit);
3595         disk->first_minor = unit << shift;
3596         if (name)
3597                 strcpy(disk->disk_name, name);
3598         else if (partitioned)
3599                 sprintf(disk->disk_name, "md_d%d", unit);
3600         else
3601                 sprintf(disk->disk_name, "md%d", unit);
3602         disk->fops = &md_fops;
3603         disk->private_data = mddev;
3604         disk->queue = mddev->queue;
3605         /* Allow extended partitions.  This makes the
3606          * 'mdp' device redundant, but we can't really
3607          * remove it now.
3608          */
3609         disk->flags |= GENHD_FL_EXT_DEVT;
3610         add_disk(disk);
3611         mddev->gendisk = disk;
3612         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3613                                      &disk_to_dev(disk)->kobj, "%s", "md");
3614         mutex_unlock(&disks_mutex);
3615         if (error)
3616                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3617                        disk->disk_name);
3618         else {
3619                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3620                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3621         }
3622         mddev_put(mddev);
3623         return 0;
3624 }
3625
3626 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3627 {
3628         md_alloc(dev, NULL);
3629         return NULL;
3630 }
3631
3632 static int add_named_array(const char *val, struct kernel_param *kp)
3633 {
3634         /* val must be "md_*" where * is not all digits.
3635          * We allocate an array with a large free minor number, and
3636          * set the name to val.  val must not already be an active name.
3637          */
3638         int len = strlen(val);
3639         char buf[DISK_NAME_LEN];
3640
3641         while (len && val[len-1] == '\n')
3642                 len--;
3643         if (len >= DISK_NAME_LEN)
3644                 return -E2BIG;
3645         strlcpy(buf, val, len+1);
3646         if (strncmp(buf, "md_", 3) != 0)
3647                 return -EINVAL;
3648         return md_alloc(0, buf);
3649 }
3650
3651 static void md_safemode_timeout(unsigned long data)
3652 {
3653         mddev_t *mddev = (mddev_t *) data;
3654
3655         if (!atomic_read(&mddev->writes_pending)) {
3656                 mddev->safemode = 1;
3657                 if (mddev->external)
3658                         sysfs_notify_dirent(mddev->sysfs_state);
3659         }
3660         md_wakeup_thread(mddev->thread);
3661 }
3662
3663 static int start_dirty_degraded;
3664
3665 static int do_md_run(mddev_t * mddev)
3666 {
3667         int err;
3668         int chunk_size;
3669         mdk_rdev_t *rdev;
3670         struct gendisk *disk;
3671         struct mdk_personality *pers;
3672         char b[BDEVNAME_SIZE];
3673
3674         if (list_empty(&mddev->disks))
3675                 /* cannot run an array with no devices.. */
3676                 return -EINVAL;
3677
3678         if (mddev->pers)
3679                 return -EBUSY;
3680
3681         /*
3682          * Analyze all RAID superblock(s)
3683          */
3684         if (!mddev->raid_disks) {
3685                 if (!mddev->persistent)
3686                         return -EINVAL;
3687                 analyze_sbs(mddev);
3688         }
3689
3690         chunk_size = mddev->chunk_size;
3691
3692         if (chunk_size) {
3693                 if (chunk_size > MAX_CHUNK_SIZE) {
3694                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3695                                 chunk_size, MAX_CHUNK_SIZE);
3696                         return -EINVAL;
3697                 }
3698                 /*
3699                  * chunk-size has to be a power of 2
3700                  */
3701                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3702                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3703                         return -EINVAL;
3704                 }
3705
3706                 /* devices must have minimum size of one chunk */
3707                 list_for_each_entry(rdev, &mddev->disks, same_set) {
3708                         if (test_bit(Faulty, &rdev->flags))
3709                                 continue;
3710                         if (rdev->size < chunk_size / 1024) {
3711                                 printk(KERN_WARNING
3712                                         "md: Dev %s smaller than chunk_size:"
3713                                         " %lluk < %dk\n",
3714                                         bdevname(rdev->bdev,b),
3715                                         (unsigned long long)rdev->size,
3716                                         chunk_size / 1024);
3717                                 return -EINVAL;
3718                         }
3719                 }
3720         }
3721
3722         if (mddev->level != LEVEL_NONE)
3723                 request_module("md-level-%d", mddev->level);
3724         else if (mddev->clevel[0])
3725                 request_module("md-%s", mddev->clevel);
3726
3727         /*
3728          * Drop all container device buffers, from now on
3729          * the only valid external interface is through the md
3730          * device.
3731          */
3732         list_for_each_entry(rdev, &mddev->disks, same_set) {
3733                 if (test_bit(Faulty, &rdev->flags))
3734                         continue;
3735                 sync_blockdev(rdev->bdev);
3736                 invalidate_bdev(rdev->bdev);
3737
3738                 /* perform some consistency tests on the device.
3739                  * We don't want the data to overlap the metadata,
3740                  * Internal Bitmap issues has handled elsewhere.
3741                  */
3742                 if (rdev->data_offset < rdev->sb_start) {
3743                         if (mddev->size &&
3744                             rdev->data_offset + mddev->size*2
3745                             > rdev->sb_start) {
3746                                 printk("md: %s: data overlaps metadata\n",
3747                                        mdname(mddev));
3748                                 return -EINVAL;
3749                         }
3750                 } else {
3751                         if (rdev->sb_start + rdev->sb_size/512
3752                             > rdev->data_offset) {
3753                                 printk("md: %s: metadata overlaps data\n",
3754                                        mdname(mddev));
3755                                 return -EINVAL;
3756                         }
3757                 }
3758                 sysfs_notify_dirent(rdev->sysfs_state);
3759         }
3760
3761         md_probe(mddev->unit, NULL, NULL);
3762         disk = mddev->gendisk;
3763         if (!disk)
3764                 return -ENOMEM;
3765
3766         spin_lock(&pers_lock);
3767         pers = find_pers(mddev->level, mddev->clevel);
3768         if (!pers || !try_module_get(pers->owner)) {
3769                 spin_unlock(&pers_lock);
3770                 if (mddev->level != LEVEL_NONE)
3771                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3772                                mddev->level);
3773                 else
3774                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3775                                mddev->clevel);
3776                 return -EINVAL;
3777         }
3778         mddev->pers = pers;
3779         spin_unlock(&pers_lock);
3780         mddev->level = pers->level;
3781         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3782
3783         if (mddev->reshape_position != MaxSector &&
3784             pers->start_reshape == NULL) {
3785                 /* This personality cannot handle reshaping... */
3786                 mddev->pers = NULL;
3787                 module_put(pers->owner);
3788                 return -EINVAL;
3789         }
3790
3791         if (pers->sync_request) {
3792                 /* Warn if this is a potentially silly
3793                  * configuration.
3794                  */
3795                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3796                 mdk_rdev_t *rdev2;
3797                 int warned = 0;
3798
3799                 list_for_each_entry(rdev, &mddev->disks, same_set)
3800                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
3801                                 if (rdev < rdev2 &&
3802                                     rdev->bdev->bd_contains ==
3803                                     rdev2->bdev->bd_contains) {
3804                                         printk(KERN_WARNING
3805                                                "%s: WARNING: %s appears to be"
3806                                                " on the same physical disk as"
3807                                                " %s.\n",
3808                                                mdname(mddev),
3809                                                bdevname(rdev->bdev,b),
3810                                                bdevname(rdev2->bdev,b2));
3811                                         warned = 1;
3812                                 }
3813                         }
3814
3815                 if (warned)
3816                         printk(KERN_WARNING
3817                                "True protection against single-disk"
3818                                " failure might be compromised.\n");
3819         }
3820
3821         mddev->recovery = 0;
3822         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3823         mddev->barriers_work = 1;
3824         mddev->ok_start_degraded = start_dirty_degraded;
3825
3826         if (start_readonly)
3827                 mddev->ro = 2; /* read-only, but switch on first write */
3828
3829         err = mddev->pers->run(mddev);
3830         if (err)
3831                 printk(KERN_ERR "md: pers->run() failed ...\n");
3832         else if (mddev->pers->sync_request) {
3833                 err = bitmap_create(mddev);
3834                 if (err) {
3835                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3836                                mdname(mddev), err);
3837                         mddev->pers->stop(mddev);
3838                 }
3839         }
3840         if (err) {
3841                 module_put(mddev->pers->owner);
3842                 mddev->pers = NULL;
3843                 bitmap_destroy(mddev);
3844                 return err;
3845         }
3846         if (mddev->pers->sync_request) {
3847                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3848                         printk(KERN_WARNING
3849                                "md: cannot register extra attributes for %s\n",
3850                                mdname(mddev));
3851                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3852         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3853                 mddev->ro = 0;
3854
3855         atomic_set(&mddev->writes_pending,0);
3856         mddev->safemode = 0;
3857         mddev->safemode_timer.function = md_safemode_timeout;
3858         mddev->safemode_timer.data = (unsigned long) mddev;
3859         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3860         mddev->in_sync = 1;
3861
3862         list_for_each_entry(rdev, &mddev->disks, same_set)
3863                 if (rdev->raid_disk >= 0) {
3864                         char nm[20];
3865                         sprintf(nm, "rd%d", rdev->raid_disk);
3866                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3867                                 printk("md: cannot register %s for %s\n",
3868                                        nm, mdname(mddev));
3869                 }
3870
3871         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3872
3873         if (mddev->flags)
3874                 md_update_sb(mddev, 0);
3875
3876         set_capacity(disk, mddev->array_sectors);
3877
3878         /* If we call blk_queue_make_request here, it will
3879          * re-initialise max_sectors etc which may have been
3880          * refined inside -> run.  So just set the bits we need to set.
3881          * Most initialisation happended when we called
3882          * blk_queue_make_request(..., md_fail_request)
3883          * earlier.
3884          */
3885         mddev->queue->queuedata = mddev;
3886         mddev->queue->make_request_fn = mddev->pers->make_request;
3887
3888         /* If there is a partially-recovered drive we need to
3889          * start recovery here.  If we leave it to md_check_recovery,
3890          * it will remove the drives and not do the right thing
3891          */
3892         if (mddev->degraded && !mddev->sync_thread) {
3893                 int spares = 0;
3894                 list_for_each_entry(rdev, &mddev->disks, same_set)
3895                         if (rdev->raid_disk >= 0 &&
3896                             !test_bit(In_sync, &rdev->flags) &&
3897                             !test_bit(Faulty, &rdev->flags))
3898                                 /* complete an interrupted recovery */
3899                                 spares++;
3900                 if (spares && mddev->pers->sync_request) {
3901                         mddev->recovery = 0;
3902                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3903                         mddev->sync_thread = md_register_thread(md_do_sync,
3904                                                                 mddev,
3905                                                                 "%s_resync");
3906                         if (!mddev->sync_thread) {
3907                                 printk(KERN_ERR "%s: could not start resync"
3908                                        " thread...\n",
3909                                        mdname(mddev));
3910                                 /* leave the spares where they are, it shouldn't hurt */
3911                                 mddev->recovery = 0;
3912                         }
3913                 }
3914         }
3915         md_wakeup_thread(mddev->thread);
3916         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3917
3918         mddev->changed = 1;
3919         md_new_event(mddev);
3920         sysfs_notify_dirent(mddev->sysfs_state);
3921         if (mddev->sysfs_action)
3922                 sysfs_notify_dirent(mddev->sysfs_action);
3923         sysfs_notify(&mddev->kobj, NULL, "degraded");
3924         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
3925         return 0;
3926 }
3927
3928 static int restart_array(mddev_t *mddev)
3929 {
3930         struct gendisk *disk = mddev->gendisk;
3931
3932         /* Complain if it has no devices */
3933         if (list_empty(&mddev->disks))
3934                 return -ENXIO;
3935         if (!mddev->pers)
3936                 return -EINVAL;
3937         if (!mddev->ro)
3938                 return -EBUSY;
3939         mddev->safemode = 0;
3940         mddev->ro = 0;
3941         set_disk_ro(disk, 0);
3942         printk(KERN_INFO "md: %s switched to read-write mode.\n",
3943                 mdname(mddev));
3944         /* Kick recovery or resync if necessary */
3945         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3946         md_wakeup_thread(mddev->thread);
3947         md_wakeup_thread(mddev->sync_thread);
3948         sysfs_notify_dirent(mddev->sysfs_state);
3949         return 0;
3950 }
3951
3952 /* similar to deny_write_access, but accounts for our holding a reference
3953  * to the file ourselves */
3954 static int deny_bitmap_write_access(struct file * file)
3955 {
3956         struct inode *inode = file->f_mapping->host;
3957
3958         spin_lock(&inode->i_lock);
3959         if (atomic_read(&inode->i_writecount) > 1) {
3960                 spin_unlock(&inode->i_lock);
3961                 return -ETXTBSY;
3962         }
3963         atomic_set(&inode->i_writecount, -1);
3964         spin_unlock(&inode->i_lock);
3965
3966         return 0;
3967 }
3968
3969 static void restore_bitmap_write_access(struct file *file)
3970 {
3971         struct inode *inode = file->f_mapping->host;
3972
3973         spin_lock(&inode->i_lock);
3974         atomic_set(&inode->i_writecount, 1);
3975         spin_unlock(&inode->i_lock);
3976 }
3977
3978 /* mode:
3979  *   0 - completely stop and dis-assemble array
3980  *   1 - switch to readonly
3981  *   2 - stop but do not disassemble array
3982  */
3983 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3984 {
3985         int err = 0;
3986         struct gendisk *disk = mddev->gendisk;
3987
3988         if (atomic_read(&mddev->openers) > is_open) {
3989                 printk("md: %s still in use.\n",mdname(mddev));
3990                 return -EBUSY;
3991         }
3992
3993         if (mddev->pers) {
3994
3995                 if (mddev->sync_thread) {
3996                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3997                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3998                         md_unregister_thread(mddev->sync_thread);
3999                         mddev->sync_thread = NULL;
4000                 }
4001
4002                 del_timer_sync(&mddev->safemode_timer);
4003
4004                 switch(mode) {
4005                 case 1: /* readonly */
4006                         err  = -ENXIO;
4007                         if (mddev->ro==1)
4008                                 goto out;
4009                         mddev->ro = 1;
4010                         break;
4011                 case 0: /* disassemble */
4012                 case 2: /* stop */
4013                         bitmap_flush(mddev);
4014                         md_super_wait(mddev);
4015                         if (mddev->ro)
4016                                 set_disk_ro(disk, 0);
4017                         blk_queue_make_request(mddev->queue, md_fail_request);
4018                         mddev->pers->stop(mddev);
4019                         mddev->queue->merge_bvec_fn = NULL;
4020                         mddev->queue->unplug_fn = NULL;
4021                         mddev->queue->backing_dev_info.congested_fn = NULL;
4022                         if (mddev->pers->sync_request) {
4023                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4024                                 if (mddev->sysfs_action)
4025                                         sysfs_put(mddev->sysfs_action);
4026                                 mddev->sysfs_action = NULL;
4027                         }
4028                         module_put(mddev->pers->owner);
4029                         mddev->pers = NULL;
4030                         /* tell userspace to handle 'inactive' */
4031                         sysfs_notify_dirent(mddev->sysfs_state);
4032
4033                         set_capacity(disk, 0);
4034                         mddev->changed = 1;
4035
4036                         if (mddev->ro)
4037                                 mddev->ro = 0;
4038                 }
4039                 if (!mddev->in_sync || mddev->flags) {
4040                         /* mark array as shutdown cleanly */
4041                         mddev->in_sync = 1;
4042                         md_update_sb(mddev, 1);
4043                 }
4044                 if (mode == 1)
4045                         set_disk_ro(disk, 1);
4046                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4047         }
4048
4049         /*
4050          * Free resources if final stop
4051          */
4052         if (mode == 0) {
4053                 mdk_rdev_t *rdev;
4054
4055                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4056
4057                 bitmap_destroy(mddev);
4058                 if (mddev->bitmap_file) {
4059                         restore_bitmap_write_access(mddev->bitmap_file);
4060                         fput(mddev->bitmap_file);
4061                         mddev->bitmap_file = NULL;
4062                 }
4063                 mddev->bitmap_offset = 0;
4064
4065                 list_for_each_entry(rdev, &mddev->disks, same_set)
4066                         if (rdev->raid_disk >= 0) {
4067                                 char nm[20];
4068                                 sprintf(nm, "rd%d", rdev->raid_disk);
4069                                 sysfs_remove_link(&mddev->kobj, nm);
4070                         }
4071
4072                 /* make sure all md_delayed_delete calls have finished */
4073                 flush_scheduled_work();
4074
4075                 export_array(mddev);
4076
4077                 mddev->array_sectors = 0;
4078                 mddev->size = 0;
4079                 mddev->raid_disks = 0;
4080                 mddev->recovery_cp = 0;
4081                 mddev->resync_min = 0;
4082                 mddev->resync_max = MaxSector;
4083                 mddev->reshape_position = MaxSector;
4084                 mddev->external = 0;
4085                 mddev->persistent = 0;
4086                 mddev->level = LEVEL_NONE;
4087                 mddev->clevel[0] = 0;
4088                 mddev->flags = 0;
4089                 mddev->ro = 0;
4090                 mddev->metadata_type[0] = 0;
4091                 mddev->chunk_size = 0;
4092                 mddev->ctime = mddev->utime = 0;
4093                 mddev->layout = 0;
4094                 mddev->max_disks = 0;
4095                 mddev->events = 0;
4096                 mddev->delta_disks = 0;
4097                 mddev->new_level = LEVEL_NONE;
4098                 mddev->new_layout = 0;
4099                 mddev->new_chunk = 0;
4100                 mddev->curr_resync = 0;
4101                 mddev->resync_mismatches = 0;
4102                 mddev->suspend_lo = mddev->suspend_hi = 0;
4103                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4104                 mddev->recovery = 0;
4105                 mddev->in_sync = 0;
4106                 mddev->changed = 0;
4107                 mddev->degraded = 0;
4108                 mddev->barriers_work = 0;
4109                 mddev->safemode = 0;
4110                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4111                 if (mddev->hold_active == UNTIL_STOP)
4112                         mddev->hold_active = 0;
4113
4114         } else if (mddev->pers)
4115                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4116                         mdname(mddev));
4117         err = 0;
4118         md_new_event(mddev);
4119         sysfs_notify_dirent(mddev->sysfs_state);
4120 out:
4121         return err;
4122 }
4123
4124 #ifndef MODULE
4125 static void autorun_array(mddev_t *mddev)
4126 {
4127         mdk_rdev_t *rdev;
4128         int err;
4129
4130         if (list_empty(&mddev->disks))
4131                 return;
4132
4133         printk(KERN_INFO "md: running: ");
4134
4135         list_for_each_entry(rdev, &mddev->disks, same_set) {
4136                 char b[BDEVNAME_SIZE];
4137                 printk("<%s>", bdevname(rdev->bdev,b));
4138         }
4139         printk("\n");
4140
4141         err = do_md_run(mddev);
4142         if (err) {
4143                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4144                 do_md_stop(mddev, 0, 0);
4145         }
4146 }
4147
4148 /*
4149  * lets try to run arrays based on all disks that have arrived
4150  * until now. (those are in pending_raid_disks)
4151  *
4152  * the method: pick the first pending disk, collect all disks with
4153  * the same UUID, remove all from the pending list and put them into
4154  * the 'same_array' list. Then order this list based on superblock
4155  * update time (freshest comes first), kick out 'old' disks and
4156  * compare superblocks. If everything's fine then run it.
4157  *
4158  * If "unit" is allocated, then bump its reference count
4159  */
4160 static void autorun_devices(int part)
4161 {
4162         mdk_rdev_t *rdev0, *rdev, *tmp;
4163         mddev_t *mddev;
4164         char b[BDEVNAME_SIZE];
4165
4166         printk(KERN_INFO "md: autorun ...\n");
4167         while (!list_empty(&pending_raid_disks)) {
4168                 int unit;
4169                 dev_t dev;
4170                 LIST_HEAD(candidates);
4171                 rdev0 = list_entry(pending_raid_disks.next,
4172                                          mdk_rdev_t, same_set);
4173
4174                 printk(KERN_INFO "md: considering %s ...\n",
4175                         bdevname(rdev0->bdev,b));
4176                 INIT_LIST_HEAD(&candidates);
4177                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4178                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4179                                 printk(KERN_INFO "md:  adding %s ...\n",
4180                                         bdevname(rdev->bdev,b));
4181                                 list_move(&rdev->same_set, &candidates);
4182                         }
4183                 /*
4184                  * now we have a set of devices, with all of them having
4185                  * mostly sane superblocks. It's time to allocate the
4186                  * mddev.
4187                  */
4188                 if (part) {
4189                         dev = MKDEV(mdp_major,
4190                                     rdev0->preferred_minor << MdpMinorShift);
4191                         unit = MINOR(dev) >> MdpMinorShift;
4192                 } else {
4193                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4194                         unit = MINOR(dev);
4195                 }
4196                 if (rdev0->preferred_minor != unit) {
4197                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4198                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4199                         break;
4200                 }
4201
4202                 md_probe(dev, NULL, NULL);
4203                 mddev = mddev_find(dev);
4204                 if (!mddev || !mddev->gendisk) {
4205                         if (mddev)
4206                                 mddev_put(mddev);
4207                         printk(KERN_ERR
4208                                 "md: cannot allocate memory for md drive.\n");
4209                         break;
4210                 }
4211                 if (mddev_lock(mddev))
4212                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4213                                mdname(mddev));
4214                 else if (mddev->raid_disks || mddev->major_version
4215                          || !list_empty(&mddev->disks)) {
4216                         printk(KERN_WARNING
4217                                 "md: %s already running, cannot run %s\n",
4218                                 mdname(mddev), bdevname(rdev0->bdev,b));
4219                         mddev_unlock(mddev);
4220                 } else {
4221                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4222                         mddev->persistent = 1;
4223                         rdev_for_each_list(rdev, tmp, &candidates) {
4224                                 list_del_init(&rdev->same_set);
4225                                 if (bind_rdev_to_array(rdev, mddev))
4226                                         export_rdev(rdev);
4227                         }
4228                         autorun_array(mddev);
4229                         mddev_unlock(mddev);
4230                 }
4231                 /* on success, candidates will be empty, on error
4232                  * it won't...
4233                  */
4234                 rdev_for_each_list(rdev, tmp, &candidates) {
4235                         list_del_init(&rdev->same_set);
4236                         export_rdev(rdev);
4237                 }
4238                 mddev_put(mddev);
4239         }
4240         printk(KERN_INFO "md: ... autorun DONE.\n");
4241 }
4242 #endif /* !MODULE */
4243
4244 static int get_version(void __user * arg)
4245 {
4246         mdu_version_t ver;
4247
4248         ver.major = MD_MAJOR_VERSION;
4249         ver.minor = MD_MINOR_VERSION;
4250         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4251
4252         if (copy_to_user(arg, &ver, sizeof(ver)))
4253                 return -EFAULT;
4254
4255         return 0;
4256 }
4257
4258 static int get_array_info(mddev_t * mddev, void __user * arg)
4259 {
4260         mdu_array_info_t info;
4261         int nr,working,active,failed,spare;
4262         mdk_rdev_t *rdev;
4263
4264         nr=working=active=failed=spare=0;
4265         list_for_each_entry(rdev, &mddev->disks, same_set) {
4266                 nr++;
4267                 if (test_bit(Faulty, &rdev->flags))
4268                         failed++;
4269                 else {
4270                         working++;
4271                         if (test_bit(In_sync, &rdev->flags))
4272                                 active++;
4273                         else
4274                                 spare++;
4275                 }
4276         }
4277
4278         info.major_version = mddev->major_version;
4279         info.minor_version = mddev->minor_version;
4280         info.patch_version = MD_PATCHLEVEL_VERSION;
4281         info.ctime         = mddev->ctime;
4282         info.level         = mddev->level;
4283         info.size          = mddev->size;
4284         if (info.size != mddev->size) /* overflow */
4285                 info.size = -1;
4286         info.nr_disks      = nr;
4287         info.raid_disks    = mddev->raid_disks;
4288         info.md_minor      = mddev->md_minor;
4289         info.not_persistent= !mddev->persistent;
4290
4291         info.utime         = mddev->utime;
4292         info.state         = 0;
4293         if (mddev->in_sync)
4294                 info.state = (1<<MD_SB_CLEAN);
4295         if (mddev->bitmap && mddev->bitmap_offset)
4296                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4297         info.active_disks  = active;
4298         info.working_disks = working;
4299         info.failed_disks  = failed;
4300         info.spare_disks   = spare;
4301
4302         info.layout        = mddev->layout;
4303         info.chunk_size    = mddev->chunk_size;
4304
4305         if (copy_to_user(arg, &info, sizeof(info)))
4306                 return -EFAULT;
4307
4308         return 0;
4309 }
4310
4311 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4312 {
4313         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4314         char *ptr, *buf = NULL;
4315         int err = -ENOMEM;
4316
4317         if (md_allow_write(mddev))
4318                 file = kmalloc(sizeof(*file), GFP_NOIO);
4319         else
4320                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4321
4322         if (!file)
4323                 goto out;
4324
4325         /* bitmap disabled, zero the first byte and copy out */
4326         if (!mddev->bitmap || !mddev->bitmap->file) {
4327                 file->pathname[0] = '\0';
4328                 goto copy_out;
4329         }
4330
4331         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4332         if (!buf)
4333                 goto out;
4334
4335         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4336         if (IS_ERR(ptr))
4337                 goto out;
4338
4339         strcpy(file->pathname, ptr);
4340
4341 copy_out:
4342         err = 0;
4343         if (copy_to_user(arg, file, sizeof(*file)))
4344                 err = -EFAULT;
4345 out:
4346         kfree(buf);
4347         kfree(file);
4348         return err;
4349 }
4350
4351 static int get_disk_info(mddev_t * mddev, void __user * arg)
4352 {
4353         mdu_disk_info_t info;
4354         mdk_rdev_t *rdev;
4355
4356         if (copy_from_user(&info, arg, sizeof(info)))
4357                 return -EFAULT;
4358
4359         rdev = find_rdev_nr(mddev, info.number);
4360         if (rdev) {
4361                 info.major = MAJOR(rdev->bdev->bd_dev);
4362                 info.minor = MINOR(rdev->bdev->bd_dev);
4363                 info.raid_disk = rdev->raid_disk;
4364                 info.state = 0;
4365                 if (test_bit(Faulty, &rdev->flags))
4366                         info.state |= (1<<MD_DISK_FAULTY);
4367                 else if (test_bit(In_sync, &rdev->flags)) {
4368                         info.state |= (1<<MD_DISK_ACTIVE);
4369                         info.state |= (1<<MD_DISK_SYNC);
4370                 }
4371                 if (test_bit(WriteMostly, &rdev->flags))
4372                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4373         } else {
4374                 info.major = info.minor = 0;
4375                 info.raid_disk = -1;
4376                 info.state = (1<<MD_DISK_REMOVED);
4377         }
4378
4379         if (copy_to_user(arg, &info, sizeof(info)))
4380                 return -EFAULT;
4381
4382         return 0;
4383 }
4384
4385 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4386 {
4387         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4388         mdk_rdev_t *rdev;
4389         dev_t dev = MKDEV(info->major,info->minor);
4390
4391         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4392                 return -EOVERFLOW;
4393
4394         if (!mddev->raid_disks) {
4395                 int err;
4396                 /* expecting a device which has a superblock */
4397                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4398                 if (IS_ERR(rdev)) {
4399                         printk(KERN_WARNING
4400                                 "md: md_import_device returned %ld\n",
4401                                 PTR_ERR(rdev));
4402                         return PTR_ERR(rdev);
4403                 }
4404                 if (!list_empty(&mddev->disks)) {
4405                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4406                                                         mdk_rdev_t, same_set);
4407                         int err = super_types[mddev->major_version]
4408                                 .load_super(rdev, rdev0, mddev->minor_version);
4409                         if (err < 0) {
4410                                 printk(KERN_WARNING
4411                                         "md: %s has different UUID to %s\n",
4412                                         bdevname(rdev->bdev,b),
4413                                         bdevname(rdev0->bdev,b2));
4414                                 export_rdev(rdev);
4415                                 return -EINVAL;
4416                         }
4417                 }
4418                 err = bind_rdev_to_array(rdev, mddev);
4419                 if (err)
4420                         export_rdev(rdev);
4421                 return err;
4422         }
4423
4424         /*
4425          * add_new_disk can be used once the array is assembled
4426          * to add "hot spares".  They must already have a superblock
4427          * written
4428          */
4429         if (mddev->pers) {
4430                 int err;
4431                 if (!mddev->pers->hot_add_disk) {
4432                         printk(KERN_WARNING
4433                                 "%s: personality does not support diskops!\n",
4434                                mdname(mddev));
4435                         return -EINVAL;
4436                 }
4437                 if (mddev->persistent)
4438                         rdev = md_import_device(dev, mddev->major_version,
4439                                                 mddev->minor_version);
4440                 else
4441                         rdev = md_import_device(dev, -1, -1);
4442                 if (IS_ERR(rdev)) {
4443                         printk(KERN_WARNING
4444                                 "md: md_import_device returned %ld\n",
4445                                 PTR_ERR(rdev));
4446                         return PTR_ERR(rdev);
4447                 }
4448                 /* set save_raid_disk if appropriate */
4449                 if (!mddev->persistent) {
4450                         if (info->state & (1<<MD_DISK_SYNC)  &&
4451                             info->raid_disk < mddev->raid_disks)
4452                                 rdev->raid_disk = info->raid_disk;
4453                         else
4454                                 rdev->raid_disk = -1;
4455                 } else
4456                         super_types[mddev->major_version].
4457                                 validate_super(mddev, rdev);
4458                 rdev->saved_raid_disk = rdev->raid_disk;
4459
4460                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4461                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4462                         set_bit(WriteMostly, &rdev->flags);
4463
4464                 rdev->raid_disk = -1;
4465                 err = bind_rdev_to_array(rdev, mddev);
4466                 if (!err && !mddev->pers->hot_remove_disk) {
4467                         /* If there is hot_add_disk but no hot_remove_disk
4468                          * then added disks for geometry changes,
4469                          * and should be added immediately.
4470                          */
4471                         super_types[mddev->major_version].
4472                                 validate_super(mddev, rdev);
4473                         err = mddev->pers->hot_add_disk(mddev, rdev);
4474                         if (err)
4475                                 unbind_rdev_from_array(rdev);
4476                 }
4477                 if (err)
4478                         export_rdev(rdev);
4479                 else
4480                         sysfs_notify_dirent(rdev->sysfs_state);
4481
4482                 md_update_sb(mddev, 1);
4483                 if (mddev->degraded)
4484                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4485                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4486                 md_wakeup_thread(mddev->thread);
4487                 return err;
4488         }
4489
4490         /* otherwise, add_new_disk is only allowed
4491          * for major_version==0 superblocks
4492          */
4493         if (mddev->major_version != 0) {
4494                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4495                        mdname(mddev));
4496                 return -EINVAL;
4497         }
4498
4499         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4500                 int err;
4501                 rdev = md_import_device(dev, -1, 0);
4502                 if (IS_ERR(rdev)) {
4503                         printk(KERN_WARNING
4504                                 "md: error, md_import_device() returned %ld\n",
4505                                 PTR_ERR(rdev));
4506                         return PTR_ERR(rdev);
4507                 }
4508                 rdev->desc_nr = info->number;
4509                 if (info->raid_disk < mddev->raid_disks)
4510                         rdev->raid_disk = info->raid_disk;
4511                 else
4512                         rdev->raid_disk = -1;
4513
4514                 if (rdev->raid_disk < mddev->raid_disks)
4515                         if (info->state & (1<<MD_DISK_SYNC))
4516                                 set_bit(In_sync, &rdev->flags);
4517
4518                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4519                         set_bit(WriteMostly, &rdev->flags);
4520
4521                 if (!mddev->persistent) {
4522                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4523                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4524                 } else
4525                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4526                 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4527
4528                 err = bind_rdev_to_array(rdev, mddev);
4529                 if (err) {
4530                         export_rdev(rdev);
4531                         return err;
4532                 }
4533         }
4534
4535         return 0;
4536 }
4537
4538 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4539 {
4540         char b[BDEVNAME_SIZE];
4541         mdk_rdev_t *rdev;
4542
4543         rdev = find_rdev(mddev, dev);
4544         if (!rdev)
4545                 return -ENXIO;
4546
4547         if (rdev->raid_disk >= 0)
4548                 goto busy;
4549
4550         kick_rdev_from_array(rdev);
4551         md_update_sb(mddev, 1);
4552         md_new_event(mddev);
4553
4554         return 0;
4555 busy:
4556         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4557                 bdevname(rdev->bdev,b), mdname(mddev));
4558         return -EBUSY;
4559 }
4560
4561 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4562 {
4563         char b[BDEVNAME_SIZE];
4564         int err;
4565         mdk_rdev_t *rdev;
4566
4567         if (!mddev->pers)
4568                 return -ENODEV;
4569
4570         if (mddev->major_version != 0) {
4571                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4572                         " version-0 superblocks.\n",
4573                         mdname(mddev));
4574                 return -EINVAL;
4575         }
4576         if (!mddev->pers->hot_add_disk) {
4577                 printk(KERN_WARNING
4578                         "%s: personality does not support diskops!\n",
4579                         mdname(mddev));
4580                 return -EINVAL;
4581         }
4582
4583         rdev = md_import_device(dev, -1, 0);
4584         if (IS_ERR(rdev)) {
4585                 printk(KERN_WARNING
4586                         "md: error, md_import_device() returned %ld\n",
4587                         PTR_ERR(rdev));
4588                 return -EINVAL;
4589         }
4590
4591         if (mddev->persistent)
4592                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4593         else
4594                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4595
4596         rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4597
4598         if (test_bit(Faulty, &rdev->flags)) {
4599                 printk(KERN_WARNING
4600                         "md: can not hot-add faulty %s disk to %s!\n",
4601                         bdevname(rdev->bdev,b), mdname(mddev));
4602                 err = -EINVAL;
4603                 goto abort_export;
4604         }
4605         clear_bit(In_sync, &rdev->flags);
4606         rdev->desc_nr = -1;
4607         rdev->saved_raid_disk = -1;
4608         err = bind_rdev_to_array(rdev, mddev);
4609         if (err)
4610                 goto abort_export;
4611
4612         /*
4613          * The rest should better be atomic, we can have disk failures
4614          * noticed in interrupt contexts ...
4615          */
4616
4617         if (rdev->desc_nr == mddev->max_disks) {
4618                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4619                         mdname(mddev));
4620                 err = -EBUSY;
4621                 goto abort_unbind_export;
4622         }
4623
4624         rdev->raid_disk = -1;
4625
4626         md_update_sb(mddev, 1);
4627
4628         /*
4629          * Kick recovery, maybe this spare has to be added to the
4630          * array immediately.
4631          */
4632         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4633         md_wakeup_thread(mddev->thread);
4634         md_new_event(mddev);
4635         return 0;
4636
4637 abort_unbind_export:
4638         unbind_rdev_from_array(rdev);
4639
4640 abort_export:
4641         export_rdev(rdev);
4642         return err;
4643 }
4644
4645 static int set_bitmap_file(mddev_t *mddev, int fd)
4646 {
4647         int err;
4648
4649         if (mddev->pers) {
4650                 if (!mddev->pers->quiesce)
4651                         return -EBUSY;
4652                 if (mddev->recovery || mddev->sync_thread)
4653                         return -EBUSY;
4654                 /* we should be able to change the bitmap.. */
4655         }
4656
4657
4658         if (fd >= 0) {
4659                 if (mddev->bitmap)
4660                         return -EEXIST; /* cannot add when bitmap is present */
4661                 mddev->bitmap_file = fget(fd);
4662
4663                 if (mddev->bitmap_file == NULL) {
4664                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4665                                mdname(mddev));
4666                         return -EBADF;
4667                 }
4668
4669                 err = deny_bitmap_write_access(mddev->bitmap_file);
4670                 if (err) {
4671                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4672                                mdname(mddev));
4673                         fput(mddev->bitmap_file);
4674                         mddev->bitmap_file = NULL;
4675                         return err;
4676                 }
4677                 mddev->bitmap_offset = 0; /* file overrides offset */
4678         } else if (mddev->bitmap == NULL)
4679                 return -ENOENT; /* cannot remove what isn't there */
4680         err = 0;
4681         if (mddev->pers) {
4682                 mddev->pers->quiesce(mddev, 1);
4683                 if (fd >= 0)
4684                         err = bitmap_create(mddev);
4685                 if (fd < 0 || err) {
4686                         bitmap_destroy(mddev);
4687                         fd = -1; /* make sure to put the file */
4688                 }
4689                 mddev->pers->quiesce(mddev, 0);
4690         }
4691         if (fd < 0) {
4692                 if (mddev->bitmap_file) {
4693                         restore_bitmap_write_access(mddev->bitmap_file);
4694                         fput(mddev->bitmap_file);
4695                 }
4696                 mddev->bitmap_file = NULL;
4697         }
4698
4699         return err;
4700 }
4701
4702 /*
4703  * set_array_info is used two different ways
4704  * The original usage is when creating a new array.
4705  * In this usage, raid_disks is > 0 and it together with
4706  *  level, size, not_persistent,layout,chunksize determine the
4707  *  shape of the array.
4708  *  This will always create an array with a type-0.90.0 superblock.
4709  * The newer usage is when assembling an array.
4710  *  In this case raid_disks will be 0, and the major_version field is
4711  *  use to determine which style super-blocks are to be found on the devices.
4712  *  The minor and patch _version numbers are also kept incase the
4713  *  super_block handler wishes to interpret them.
4714  */
4715 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4716 {
4717
4718         if (info->raid_disks == 0) {
4719                 /* just setting version number for superblock loading */
4720                 if (info->major_version < 0 ||
4721                     info->major_version >= ARRAY_SIZE(super_types) ||
4722                     super_types[info->major_version].name == NULL) {
4723                         /* maybe try to auto-load a module? */
4724                         printk(KERN_INFO
4725                                 "md: superblock version %d not known\n",
4726                                 info->major_version);
4727                         return -EINVAL;
4728                 }
4729                 mddev->major_version = info->major_version;
4730                 mddev->minor_version = info->minor_version;
4731                 mddev->patch_version = info->patch_version;
4732                 mddev->persistent = !info->not_persistent;
4733                 return 0;
4734         }
4735         mddev->major_version = MD_MAJOR_VERSION;
4736         mddev->minor_version = MD_MINOR_VERSION;
4737         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4738         mddev->ctime         = get_seconds();
4739
4740         mddev->level         = info->level;
4741         mddev->clevel[0]     = 0;
4742         mddev->size          = info->size;
4743         mddev->raid_disks    = info->raid_disks;
4744         /* don't set md_minor, it is determined by which /dev/md* was
4745          * openned
4746          */
4747         if (info->state & (1<<MD_SB_CLEAN))
4748                 mddev->recovery_cp = MaxSector;
4749         else
4750                 mddev->recovery_cp = 0;
4751         mddev->persistent    = ! info->not_persistent;
4752         mddev->external      = 0;
4753
4754         mddev->layout        = info->layout;
4755         mddev->chunk_size    = info->chunk_size;
4756
4757         mddev->max_disks     = MD_SB_DISKS;
4758
4759         if (mddev->persistent)
4760                 mddev->flags         = 0;
4761         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4762
4763         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4764         mddev->bitmap_offset = 0;
4765
4766         mddev->reshape_position = MaxSector;
4767
4768         /*
4769          * Generate a 128 bit UUID
4770          */
4771         get_random_bytes(mddev->uuid, 16);
4772
4773         mddev->new_level = mddev->level;
4774         mddev->new_chunk = mddev->chunk_size;
4775         mddev->new_layout = mddev->layout;
4776         mddev->delta_disks = 0;
4777
4778         return 0;
4779 }
4780
4781 static int update_size(mddev_t *mddev, sector_t num_sectors)
4782 {
4783         mdk_rdev_t *rdev;
4784         int rv;
4785         int fit = (num_sectors == 0);
4786
4787         if (mddev->pers->resize == NULL)
4788                 return -EINVAL;
4789         /* The "num_sectors" is the number of sectors of each device that
4790          * is used.  This can only make sense for arrays with redundancy.
4791          * linear and raid0 always use whatever space is available. We can only
4792          * consider changing this number if no resync or reconstruction is
4793          * happening, and if the new size is acceptable. It must fit before the
4794          * sb_start or, if that is <data_offset, it must fit before the size
4795          * of each device.  If num_sectors is zero, we find the largest size
4796          * that fits.
4797
4798          */
4799         if (mddev->sync_thread)
4800                 return -EBUSY;
4801         if (mddev->bitmap)
4802                 /* Sorry, cannot grow a bitmap yet, just remove it,
4803                  * grow, and re-add.
4804                  */
4805                 return -EBUSY;
4806         list_for_each_entry(rdev, &mddev->disks, same_set) {
4807                 sector_t avail;
4808                 avail = rdev->size * 2;
4809
4810                 if (fit && (num_sectors == 0 || num_sectors > avail))
4811                         num_sectors = avail;
4812                 if (avail < num_sectors)
4813                         return -ENOSPC;
4814         }
4815         rv = mddev->pers->resize(mddev, num_sectors);
4816         if (!rv) {
4817                 struct block_device *bdev;
4818
4819                 bdev = bdget_disk(mddev->gendisk, 0);
4820                 if (bdev) {
4821                         mutex_lock(&bdev->bd_inode->i_mutex);
4822                         i_size_write(bdev->bd_inode,
4823                                      (loff_t)mddev->array_sectors << 9);
4824                         mutex_unlock(&bdev->bd_inode->i_mutex);
4825                         bdput(bdev);
4826                 }
4827         }
4828         return rv;
4829 }
4830
4831 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4832 {
4833         int rv;
4834         /* change the number of raid disks */
4835         if (mddev->pers->check_reshape == NULL)
4836                 return -EINVAL;
4837         if (raid_disks <= 0 ||
4838             raid_disks >= mddev->max_disks)
4839                 return -EINVAL;
4840         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4841                 return -EBUSY;
4842         mddev->delta_disks = raid_disks - mddev->raid_disks;
4843
4844         rv = mddev->pers->check_reshape(mddev);
4845         return rv;
4846 }
4847
4848
4849 /*
4850  * update_array_info is used to change the configuration of an
4851  * on-line array.
4852  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4853  * fields in the info are checked against the array.
4854  * Any differences that cannot be handled will cause an error.
4855  * Normally, only one change can be managed at a time.
4856  */
4857 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4858 {
4859         int rv = 0;
4860         int cnt = 0;
4861         int state = 0;
4862
4863         /* calculate expected state,ignoring low bits */
4864         if (mddev->bitmap && mddev->bitmap_offset)
4865                 state |= (1 << MD_SB_BITMAP_PRESENT);
4866
4867         if (mddev->major_version != info->major_version ||
4868             mddev->minor_version != info->minor_version ||
4869 /*          mddev->patch_version != info->patch_version || */
4870             mddev->ctime         != info->ctime         ||
4871             mddev->level         != info->level         ||
4872 /*          mddev->layout        != info->layout        || */
4873             !mddev->persistent   != info->not_persistent||
4874             mddev->chunk_size    != info->chunk_size    ||
4875             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4876             ((state^info->state) & 0xfffffe00)
4877                 )
4878                 return -EINVAL;
4879         /* Check there is only one change */
4880         if (info->size >= 0 && mddev->size != info->size) cnt++;
4881         if (mddev->raid_disks != info->raid_disks) cnt++;
4882         if (mddev->layout != info->layout) cnt++;
4883         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4884         if (cnt == 0) return 0;
4885         if (cnt > 1) return -EINVAL;
4886
4887         if (mddev->layout != info->layout) {
4888                 /* Change layout
4889                  * we don't need to do anything at the md level, the
4890                  * personality will take care of it all.
4891                  */
4892                 if (mddev->pers->reconfig == NULL)
4893                         return -EINVAL;
4894                 else
4895                         return mddev->pers->reconfig(mddev, info->layout, -1);
4896         }
4897         if (info->size >= 0 && mddev->size != info->size)
4898                 rv = update_size(mddev, (sector_t)info->size * 2);
4899
4900         if (mddev->raid_disks    != info->raid_disks)
4901                 rv = update_raid_disks(mddev, info->raid_disks);
4902
4903         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4904                 if (mddev->pers->quiesce == NULL)
4905                         return -EINVAL;
4906                 if (mddev->recovery || mddev->sync_thread)
4907                         return -EBUSY;
4908                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4909                         /* add the bitmap */
4910                         if (mddev->bitmap)
4911                                 return -EEXIST;
4912                         if (mddev->default_bitmap_offset == 0)
4913                                 return -EINVAL;
4914                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4915                         mddev->pers->quiesce(mddev, 1);
4916                         rv = bitmap_create(mddev);
4917                         if (rv)
4918                                 bitmap_destroy(mddev);
4919                         mddev->pers->quiesce(mddev, 0);
4920                 } else {
4921                         /* remove the bitmap */
4922                         if (!mddev->bitmap)
4923                                 return -ENOENT;
4924                         if (mddev->bitmap->file)
4925                                 return -EINVAL;
4926                         mddev->pers->quiesce(mddev, 1);
4927                         bitmap_destroy(mddev);
4928                         mddev->pers->quiesce(mddev, 0);
4929                         mddev->bitmap_offset = 0;
4930                 }
4931         }
4932         md_update_sb(mddev, 1);
4933         return rv;
4934 }
4935
4936 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4937 {
4938         mdk_rdev_t *rdev;
4939
4940         if (mddev->pers == NULL)
4941                 return -ENODEV;
4942
4943         rdev = find_rdev(mddev, dev);
4944         if (!rdev)
4945                 return -ENODEV;
4946
4947         md_error(mddev, rdev);
4948         return 0;
4949 }
4950
4951 /*
4952  * We have a problem here : there is no easy way to give a CHS
4953  * virtual geometry. We currently pretend that we have a 2 heads
4954  * 4 sectors (with a BIG number of cylinders...). This drives
4955  * dosfs just mad... ;-)
4956  */
4957 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4958 {
4959         mddev_t *mddev = bdev->bd_disk->private_data;
4960
4961         geo->heads = 2;
4962         geo->sectors = 4;
4963         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4964         return 0;
4965 }
4966
4967 static int md_ioctl(struct block_device *bdev, fmode_t mode,
4968                         unsigned int cmd, unsigned long arg)
4969 {
4970         int err = 0;
4971         void __user *argp = (void __user *)arg;
4972         mddev_t *mddev = NULL;
4973
4974         if (!capable(CAP_SYS_ADMIN))
4975                 return -EACCES;
4976
4977         /*
4978          * Commands dealing with the RAID driver but not any
4979          * particular array:
4980          */
4981         switch (cmd)
4982         {
4983                 case RAID_VERSION:
4984                         err = get_version(argp);
4985                         goto done;
4986
4987                 case PRINT_RAID_DEBUG:
4988                         err = 0;
4989                         md_print_devices();
4990                         goto done;
4991
4992 #ifndef MODULE
4993                 case RAID_AUTORUN:
4994                         err = 0;
4995                         autostart_arrays(arg);
4996                         goto done;
4997 #endif
4998                 default:;
4999         }
5000
5001         /*
5002          * Commands creating/starting a new array:
5003          */
5004
5005         mddev = bdev->bd_disk->private_data;
5006
5007         if (!mddev) {
5008                 BUG();
5009                 goto abort;
5010         }
5011
5012         err = mddev_lock(mddev);
5013         if (err) {
5014                 printk(KERN_INFO
5015                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5016                         err, cmd);
5017                 goto abort;
5018         }
5019
5020         switch (cmd)
5021         {
5022                 case SET_ARRAY_INFO:
5023                         {
5024                                 mdu_array_info_t info;
5025                                 if (!arg)
5026                                         memset(&info, 0, sizeof(info));
5027                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5028                                         err = -EFAULT;
5029                                         goto abort_unlock;
5030                                 }
5031                                 if (mddev->pers) {
5032                                         err = update_array_info(mddev, &info);
5033                                         if (err) {
5034                                                 printk(KERN_WARNING "md: couldn't update"
5035                                                        " array info. %d\n", err);
5036                                                 goto abort_unlock;
5037                                         }
5038                                         goto done_unlock;
5039                                 }
5040                                 if (!list_empty(&mddev->disks)) {
5041                                         printk(KERN_WARNING
5042                                                "md: array %s already has disks!\n",
5043                                                mdname(mddev));
5044                                         err = -EBUSY;
5045                                         goto abort_unlock;
5046                                 }
5047                                 if (mddev->raid_disks) {
5048                                         printk(KERN_WARNING
5049                                                "md: array %s already initialised!\n",
5050                                                mdname(mddev));
5051                                         err = -EBUSY;
5052                                         goto abort_unlock;
5053                                 }
5054                                 err = set_array_info(mddev, &info);
5055                                 if (err) {
5056                                         printk(KERN_WARNING "md: couldn't set"
5057                                                " array info. %d\n", err);
5058                                         goto abort_unlock;
5059                                 }
5060                         }
5061                         goto done_unlock;
5062
5063                 default:;
5064         }
5065
5066         /*
5067          * Commands querying/configuring an existing array:
5068          */
5069         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5070          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5071         if ((!mddev->raid_disks && !mddev->external)
5072             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5073             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5074             && cmd != GET_BITMAP_FILE) {
5075                 err = -ENODEV;
5076                 goto abort_unlock;
5077         }
5078
5079         /*
5080          * Commands even a read-only array can execute:
5081          */
5082         switch (cmd)
5083         {
5084                 case GET_ARRAY_INFO:
5085                         err = get_array_info(mddev, argp);
5086                         goto done_unlock;
5087
5088                 case GET_BITMAP_FILE:
5089                         err = get_bitmap_file(mddev, argp);
5090                         goto done_unlock;
5091
5092                 case GET_DISK_INFO:
5093                         err = get_disk_info(mddev, argp);
5094                         goto done_unlock;
5095
5096                 case RESTART_ARRAY_RW:
5097                         err = restart_array(mddev);
5098                         goto done_unlock;
5099
5100                 case STOP_ARRAY:
5101                         err = do_md_stop(mddev, 0, 1);
5102                         goto done_unlock;
5103
5104                 case STOP_ARRAY_RO:
5105                         err = do_md_stop(mddev, 1, 1);
5106                         goto done_unlock;
5107
5108         }
5109
5110         /*
5111          * The remaining ioctls are changing the state of the
5112          * superblock, so we do not allow them on read-only arrays.
5113          * However non-MD ioctls (e.g. get-size) will still come through
5114          * here and hit the 'default' below, so only disallow
5115          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5116          */
5117         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5118                 if (mddev->ro == 2) {
5119                         mddev->ro = 0;
5120                         sysfs_notify_dirent(mddev->sysfs_state);
5121                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5122                         md_wakeup_thread(mddev->thread);
5123                 } else {
5124                         err = -EROFS;
5125                         goto abort_unlock;
5126                 }
5127         }
5128
5129         switch (cmd)
5130         {
5131                 case ADD_NEW_DISK:
5132                 {
5133                         mdu_disk_info_t info;
5134                         if (copy_from_user(&info, argp, sizeof(info)))
5135                                 err = -EFAULT;
5136                         else
5137                                 err = add_new_disk(mddev, &info);
5138                         goto done_unlock;
5139                 }
5140
5141                 case HOT_REMOVE_DISK:
5142                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5143                         goto done_unlock;
5144
5145                 case HOT_ADD_DISK:
5146                         err = hot_add_disk(mddev, new_decode_dev(arg));
5147                         goto done_unlock;
5148
5149                 case SET_DISK_FAULTY:
5150                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5151                         goto done_unlock;
5152
5153                 case RUN_ARRAY:
5154                         err = do_md_run(mddev);
5155                         goto done_unlock;
5156
5157                 case SET_BITMAP_FILE:
5158                         err = set_bitmap_file(mddev, (int)arg);
5159                         goto done_unlock;
5160
5161                 default:
5162                         err = -EINVAL;
5163                         goto abort_unlock;
5164         }
5165
5166 done_unlock:
5167 abort_unlock:
5168         if (mddev->hold_active == UNTIL_IOCTL &&
5169             err != -EINVAL)
5170                 mddev->hold_active = 0;
5171         mddev_unlock(mddev);
5172
5173         return err;
5174 done:
5175         if (err)
5176                 MD_BUG();
5177 abort:
5178         return err;
5179 }
5180
5181 static int md_open(struct block_device *bdev, fmode_t mode)
5182 {
5183         /*
5184          * Succeed if we can lock the mddev, which confirms that
5185          * it isn't being stopped right now.
5186          */
5187         mddev_t *mddev = mddev_find(bdev->bd_dev);
5188         int err;
5189
5190         if (mddev->gendisk != bdev->bd_disk) {
5191                 /* we are racing with mddev_put which is discarding this
5192                  * bd_disk.
5193                  */
5194                 mddev_put(mddev);
5195                 /* Wait until bdev->bd_disk is definitely gone */
5196                 flush_scheduled_work();
5197                 /* Then retry the open from the top */
5198                 return -ERESTARTSYS;
5199         }
5200         BUG_ON(mddev != bdev->bd_disk->private_data);
5201
5202         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5203                 goto out;
5204
5205         err = 0;
5206         atomic_inc(&mddev->openers);
5207         mddev_unlock(mddev);
5208
5209         check_disk_change(bdev);
5210  out:
5211         return err;
5212 }
5213
5214 static int md_release(struct gendisk *disk, fmode_t mode)
5215 {
5216         mddev_t *mddev = disk->private_data;
5217
5218         BUG_ON(!mddev);
5219         atomic_dec(&mddev->openers);
5220         mddev_put(mddev);
5221
5222         return 0;
5223 }
5224
5225 static int md_media_changed(struct gendisk *disk)
5226 {
5227         mddev_t *mddev = disk->private_data;
5228
5229         return mddev->changed;
5230 }
5231
5232 static int md_revalidate(struct gendisk *disk)
5233 {
5234         mddev_t *mddev = disk->private_data;
5235
5236         mddev->changed = 0;
5237         return 0;
5238 }
5239 static struct block_device_operations md_fops =
5240 {
5241         .owner          = THIS_MODULE,
5242         .open           = md_open,
5243         .release        = md_release,
5244         .locked_ioctl   = md_ioctl,
5245         .getgeo         = md_getgeo,
5246         .media_changed  = md_media_changed,
5247         .revalidate_disk= md_revalidate,
5248 };
5249
5250 static int md_thread(void * arg)
5251 {
5252         mdk_thread_t *thread = arg;
5253
5254         /*
5255          * md_thread is a 'system-thread', it's priority should be very
5256          * high. We avoid resource deadlocks individually in each
5257          * raid personality. (RAID5 does preallocation) We also use RR and
5258          * the very same RT priority as kswapd, thus we will never get
5259          * into a priority inversion deadlock.
5260          *
5261          * we definitely have to have equal or higher priority than
5262          * bdflush, otherwise bdflush will deadlock if there are too
5263          * many dirty RAID5 blocks.
5264          */
5265
5266         allow_signal(SIGKILL);
5267         while (!kthread_should_stop()) {
5268
5269                 /* We need to wait INTERRUPTIBLE so that
5270                  * we don't add to the load-average.
5271                  * That means we need to be sure no signals are
5272                  * pending
5273                  */
5274                 if (signal_pending(current))
5275                         flush_signals(current);
5276
5277                 wait_event_interruptible_timeout
5278                         (thread->wqueue,
5279                          test_bit(THREAD_WAKEUP, &thread->flags)
5280                          || kthread_should_stop(),
5281                          thread->timeout);
5282
5283                 clear_bit(THREAD_WAKEUP, &thread->flags);
5284
5285                 thread->run(thread->mddev);
5286         }
5287
5288         return 0;
5289 }
5290
5291 void md_wakeup_thread(mdk_thread_t *thread)
5292 {
5293         if (thread) {
5294                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5295                 set_bit(THREAD_WAKEUP, &thread->flags);
5296                 wake_up(&thread->wqueue);
5297         }
5298 }
5299
5300 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5301                                  const char *name)
5302 {
5303         mdk_thread_t *thread;
5304
5305         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5306         if (!thread)
5307                 return NULL;
5308
5309         init_waitqueue_head(&thread->wqueue);
5310
5311         thread->run = run;
5312         thread->mddev = mddev;
5313         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5314         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5315         if (IS_ERR(thread->tsk)) {
5316                 kfree(thread);
5317                 return NULL;
5318         }
5319         return thread;
5320 }
5321
5322 void md_unregister_thread(mdk_thread_t *thread)
5323 {
5324         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5325
5326         kthread_stop(thread->tsk);
5327         kfree(thread);
5328 }
5329
5330 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5331 {
5332         if (!mddev) {
5333                 MD_BUG();
5334                 return;
5335         }
5336
5337         if (!rdev || test_bit(Faulty, &rdev->flags))
5338                 return;
5339
5340         if (mddev->external)
5341                 set_bit(Blocked, &rdev->flags);
5342 /*
5343         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5344                 mdname(mddev),
5345                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5346                 __builtin_return_address(0),__builtin_return_address(1),
5347                 __builtin_return_address(2),__builtin_return_address(3));
5348 */
5349         if (!mddev->pers)
5350                 return;
5351         if (!mddev->pers->error_handler)
5352                 return;
5353         mddev->pers->error_handler(mddev,rdev);
5354         if (mddev->degraded)
5355                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5356         set_bit(StateChanged, &rdev->flags);
5357         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5358         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5359         md_wakeup_thread(mddev->thread);
5360         md_new_event_inintr(mddev);
5361 }
5362
5363 /* seq_file implementation /proc/mdstat */
5364
5365 static void status_unused(struct seq_file *seq)
5366 {
5367         int i = 0;
5368         mdk_rdev_t *rdev;
5369
5370         seq_printf(seq, "unused devices: ");
5371
5372         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5373                 char b[BDEVNAME_SIZE];
5374                 i++;
5375                 seq_printf(seq, "%s ",
5376                               bdevname(rdev->bdev,b));
5377         }
5378         if (!i)
5379                 seq_printf(seq, "<none>");
5380
5381         seq_printf(seq, "\n");
5382 }
5383
5384
5385 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5386 {
5387         sector_t max_blocks, resync, res;
5388         unsigned long dt, db, rt;
5389         int scale;
5390         unsigned int per_milli;
5391
5392         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5393
5394         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5395                 max_blocks = mddev->resync_max_sectors >> 1;
5396         else
5397                 max_blocks = mddev->size;
5398
5399         /*
5400          * Should not happen.
5401          */
5402         if (!max_blocks) {
5403                 MD_BUG();
5404                 return;
5405         }
5406         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5407          * in a sector_t, and (max_blocks>>scale) will fit in a
5408          * u32, as those are the requirements for sector_div.
5409          * Thus 'scale' must be at least 10
5410          */
5411         scale = 10;
5412         if (sizeof(sector_t) > sizeof(unsigned long)) {
5413                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5414                         scale++;
5415         }
5416         res = (resync>>scale)*1000;
5417         sector_div(res, (u32)((max_blocks>>scale)+1));
5418
5419         per_milli = res;
5420         {
5421                 int i, x = per_milli/50, y = 20-x;
5422                 seq_printf(seq, "[");
5423                 for (i = 0; i < x; i++)
5424                         seq_printf(seq, "=");
5425                 seq_printf(seq, ">");
5426                 for (i = 0; i < y; i++)
5427                         seq_printf(seq, ".");
5428                 seq_printf(seq, "] ");
5429         }
5430         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5431                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5432                     "reshape" :
5433                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5434                      "check" :
5435                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5436                       "resync" : "recovery"))),
5437                    per_milli/10, per_milli % 10,
5438                    (unsigned long long) resync,
5439                    (unsigned long long) max_blocks);
5440
5441         /*
5442          * We do not want to overflow, so the order of operands and
5443          * the * 100 / 100 trick are important. We do a +1 to be
5444          * safe against division by zero. We only estimate anyway.
5445          *
5446          * dt: time from mark until now
5447          * db: blocks written from mark until now
5448          * rt: remaining time
5449          */
5450         dt = ((jiffies - mddev->resync_mark) / HZ);
5451         if (!dt) dt++;
5452         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5453                 - mddev->resync_mark_cnt;
5454         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5455
5456         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5457
5458         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5459 }
5460
5461 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5462 {
5463         struct list_head *tmp;
5464         loff_t l = *pos;
5465         mddev_t *mddev;
5466
5467         if (l >= 0x10000)
5468                 return NULL;
5469         if (!l--)
5470                 /* header */
5471                 return (void*)1;
5472
5473         spin_lock(&all_mddevs_lock);
5474         list_for_each(tmp,&all_mddevs)
5475                 if (!l--) {
5476                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5477                         mddev_get(mddev);
5478                         spin_unlock(&all_mddevs_lock);
5479                         return mddev;
5480                 }
5481         spin_unlock(&all_mddevs_lock);
5482         if (!l--)
5483                 return (void*)2;/* tail */
5484         return NULL;
5485 }
5486
5487 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5488 {
5489         struct list_head *tmp;
5490         mddev_t *next_mddev, *mddev = v;
5491
5492         ++*pos;
5493         if (v == (void*)2)
5494                 return NULL;
5495
5496         spin_lock(&all_mddevs_lock);
5497         if (v == (void*)1)
5498                 tmp = all_mddevs.next;
5499         else
5500                 tmp = mddev->all_mddevs.next;
5501         if (tmp != &all_mddevs)
5502                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5503         else {
5504                 next_mddev = (void*)2;
5505                 *pos = 0x10000;
5506         }
5507         spin_unlock(&all_mddevs_lock);
5508
5509         if (v != (void*)1)
5510                 mddev_put(mddev);
5511         return next_mddev;
5512
5513 }
5514
5515 static void md_seq_stop(struct seq_file *seq, void *v)
5516 {
5517         mddev_t *mddev = v;
5518
5519         if (mddev && v != (void*)1 && v != (void*)2)
5520                 mddev_put(mddev);
5521 }
5522
5523 struct mdstat_info {
5524         int event;
5525 };
5526
5527 static int md_seq_show(struct seq_file *seq, void *v)
5528 {
5529         mddev_t *mddev = v;
5530         sector_t size;
5531         mdk_rdev_t *rdev;
5532         struct mdstat_info *mi = seq->private;
5533         struct bitmap *bitmap;
5534
5535         if (v == (void*)1) {
5536                 struct mdk_personality *pers;
5537                 seq_printf(seq, "Personalities : ");
5538                 spin_lock(&pers_lock);
5539                 list_for_each_entry(pers, &pers_list, list)
5540                         seq_printf(seq, "[%s] ", pers->name);
5541
5542                 spin_unlock(&pers_lock);
5543                 seq_printf(seq, "\n");
5544                 mi->event = atomic_read(&md_event_count);
5545                 return 0;
5546         }
5547         if (v == (void*)2) {
5548                 status_unused(seq);
5549                 return 0;
5550         }
5551
5552         if (mddev_lock(mddev) < 0)
5553                 return -EINTR;
5554
5555         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5556                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5557                                                 mddev->pers ? "" : "in");
5558                 if (mddev->pers) {
5559                         if (mddev->ro==1)
5560                                 seq_printf(seq, " (read-only)");
5561                         if (mddev->ro==2)
5562                                 seq_printf(seq, " (auto-read-only)");
5563                         seq_printf(seq, " %s", mddev->pers->name);
5564                 }
5565
5566                 size = 0;
5567                 list_for_each_entry(rdev, &mddev->disks, same_set) {
5568                         char b[BDEVNAME_SIZE];
5569                         seq_printf(seq, " %s[%d]",
5570                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5571                         if (test_bit(WriteMostly, &rdev->flags))
5572                                 seq_printf(seq, "(W)");
5573                         if (test_bit(Faulty, &rdev->flags)) {
5574                                 seq_printf(seq, "(F)");
5575                                 continue;
5576                         } else if (rdev->raid_disk < 0)
5577                                 seq_printf(seq, "(S)"); /* spare */
5578                         size += rdev->size;
5579                 }
5580
5581                 if (!list_empty(&mddev->disks)) {
5582                         if (mddev->pers)
5583                                 seq_printf(seq, "\n      %llu blocks",
5584                                            (unsigned long long)
5585                                            mddev->array_sectors / 2);
5586                         else
5587                                 seq_printf(seq, "\n      %llu blocks",
5588                                            (unsigned long long)size);
5589                 }
5590                 if (mddev->persistent) {
5591                         if (mddev->major_version != 0 ||
5592                             mddev->minor_version != 90) {
5593                                 seq_printf(seq," super %d.%d",
5594                                            mddev->major_version,
5595                                            mddev->minor_version);
5596                         }
5597                 } else if (mddev->external)
5598                         seq_printf(seq, " super external:%s",
5599                                    mddev->metadata_type);
5600                 else
5601                         seq_printf(seq, " super non-persistent");
5602
5603                 if (mddev->pers) {
5604                         mddev->pers->status(seq, mddev);
5605                         seq_printf(seq, "\n      ");
5606                         if (mddev->pers->sync_request) {
5607                                 if (mddev->curr_resync > 2) {
5608                                         status_resync(seq, mddev);
5609                                         seq_printf(seq, "\n      ");
5610                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5611                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5612                                 else if (mddev->recovery_cp < MaxSector)
5613                                         seq_printf(seq, "\tresync=PENDING\n      ");
5614                         }
5615                 } else
5616                         seq_printf(seq, "\n       ");
5617
5618                 if ((bitmap = mddev->bitmap)) {
5619                         unsigned long chunk_kb;
5620                         unsigned long flags;
5621                         spin_lock_irqsave(&bitmap->lock, flags);
5622                         chunk_kb = bitmap->chunksize >> 10;
5623                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5624                                 "%lu%s chunk",
5625                                 bitmap->pages - bitmap->missing_pages,
5626                                 bitmap->pages,
5627                                 (bitmap->pages - bitmap->missing_pages)
5628                                         << (PAGE_SHIFT - 10),
5629                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5630                                 chunk_kb ? "KB" : "B");
5631                         if (bitmap->file) {
5632                                 seq_printf(seq, ", file: ");
5633                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5634                         }
5635
5636                         seq_printf(seq, "\n");
5637                         spin_unlock_irqrestore(&bitmap->lock, flags);
5638                 }
5639
5640                 seq_printf(seq, "\n");
5641         }
5642         mddev_unlock(mddev);
5643
5644         return 0;
5645 }
5646
5647 static struct seq_operations md_seq_ops = {
5648         .start  = md_seq_start,
5649         .next   = md_seq_next,
5650         .stop   = md_seq_stop,
5651         .show   = md_seq_show,
5652 };
5653
5654 static int md_seq_open(struct inode *inode, struct file *file)
5655 {
5656         int error;
5657         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5658         if (mi == NULL)
5659                 return -ENOMEM;
5660
5661         error = seq_open(file, &md_seq_ops);
5662         if (error)
5663                 kfree(mi);
5664         else {
5665                 struct seq_file *p = file->private_data;
5666                 p->private = mi;
5667                 mi->event = atomic_read(&md_event_count);
5668         }
5669         return error;
5670 }
5671
5672 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5673 {
5674         struct seq_file *m = filp->private_data;
5675         struct mdstat_info *mi = m->private;
5676         int mask;
5677
5678         poll_wait(filp, &md_event_waiters, wait);
5679
5680         /* always allow read */
5681         mask = POLLIN | POLLRDNORM;
5682
5683         if (mi->event != atomic_read(&md_event_count))
5684                 mask |= POLLERR | POLLPRI;
5685         return mask;
5686 }
5687
5688 static const struct file_operations md_seq_fops = {
5689         .owner          = THIS_MODULE,
5690         .open           = md_seq_open,
5691         .read           = seq_read,
5692         .llseek         = seq_lseek,
5693         .release        = seq_release_private,
5694         .poll           = mdstat_poll,
5695 };
5696
5697 int register_md_personality(struct mdk_personality *p)
5698 {
5699         spin_lock(&pers_lock);
5700         list_add_tail(&p->list, &pers_list);
5701         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5702         spin_unlock(&pers_lock);
5703         return 0;
5704 }
5705
5706 int unregister_md_personality(struct mdk_personality *p)
5707 {
5708         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5709         spin_lock(&pers_lock);
5710         list_del_init(&p->list);
5711         spin_unlock(&pers_lock);
5712         return 0;
5713 }
5714
5715 static int is_mddev_idle(mddev_t *mddev)
5716 {
5717         mdk_rdev_t * rdev;
5718         int idle;
5719         long curr_events;
5720
5721         idle = 1;
5722         rcu_read_lock();
5723         rdev_for_each_rcu(rdev, mddev) {
5724                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5725                 curr_events = part_stat_read(&disk->part0, sectors[0]) +
5726                                 part_stat_read(&disk->part0, sectors[1]) -
5727                                 atomic_read(&disk->sync_io);
5728                 /* sync IO will cause sync_io to increase before the disk_stats
5729                  * as sync_io is counted when a request starts, and
5730                  * disk_stats is counted when it completes.
5731                  * So resync activity will cause curr_events to be smaller than
5732                  * when there was no such activity.
5733                  * non-sync IO will cause disk_stat to increase without
5734                  * increasing sync_io so curr_events will (eventually)
5735                  * be larger than it was before.  Once it becomes
5736                  * substantially larger, the test below will cause
5737                  * the array to appear non-idle, and resync will slow
5738                  * down.
5739                  * If there is a lot of outstanding resync activity when
5740                  * we set last_event to curr_events, then all that activity
5741                  * completing might cause the array to appear non-idle
5742                  * and resync will be slowed down even though there might
5743                  * not have been non-resync activity.  This will only
5744                  * happen once though.  'last_events' will soon reflect
5745                  * the state where there is little or no outstanding
5746                  * resync requests, and further resync activity will
5747                  * always make curr_events less than last_events.
5748                  *
5749                  */
5750                 if (curr_events - rdev->last_events > 4096) {
5751                         rdev->last_events = curr_events;
5752                         idle = 0;
5753                 }
5754         }
5755         rcu_read_unlock();
5756         return idle;
5757 }
5758
5759 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5760 {
5761         /* another "blocks" (512byte) blocks have been synced */
5762         atomic_sub(blocks, &mddev->recovery_active);
5763         wake_up(&mddev->recovery_wait);
5764         if (!ok) {
5765                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5766                 md_wakeup_thread(mddev->thread);
5767                 // stop recovery, signal do_sync ....
5768         }
5769 }
5770
5771
5772 /* md_write_start(mddev, bi)
5773  * If we need to update some array metadata (e.g. 'active' flag
5774  * in superblock) before writing, schedule a superblock update
5775  * and wait for it to complete.
5776  */
5777 void md_write_start(mddev_t *mddev, struct bio *bi)
5778 {
5779         int did_change = 0;
5780         if (bio_data_dir(bi) != WRITE)
5781                 return;
5782
5783         BUG_ON(mddev->ro == 1);
5784         if (mddev->ro == 2) {
5785                 /* need to switch to read/write */
5786                 mddev->ro = 0;
5787                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5788                 md_wakeup_thread(mddev->thread);
5789                 md_wakeup_thread(mddev->sync_thread);
5790                 did_change = 1;
5791         }
5792         atomic_inc(&mddev->writes_pending);
5793         if (mddev->safemode == 1)
5794                 mddev->safemode = 0;
5795         if (mddev->in_sync) {
5796                 spin_lock_irq(&mddev->write_lock);
5797                 if (mddev->in_sync) {
5798                         mddev->in_sync = 0;
5799                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5800                         md_wakeup_thread(mddev->thread);
5801                         did_change = 1;
5802                 }
5803                 spin_unlock_irq(&mddev->write_lock);
5804         }
5805         if (did_change)
5806                 sysfs_notify_dirent(mddev->sysfs_state);
5807         wait_event(mddev->sb_wait,
5808                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5809                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5810 }
5811
5812 void md_write_end(mddev_t *mddev)
5813 {
5814         if (atomic_dec_and_test(&mddev->writes_pending)) {
5815                 if (mddev->safemode == 2)
5816                         md_wakeup_thread(mddev->thread);
5817                 else if (mddev->safemode_delay)
5818                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5819         }
5820 }
5821
5822 /* md_allow_write(mddev)
5823  * Calling this ensures that the array is marked 'active' so that writes
5824  * may proceed without blocking.  It is important to call this before
5825  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5826  * Must be called with mddev_lock held.
5827  *
5828  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5829  * is dropped, so return -EAGAIN after notifying userspace.
5830  */
5831 int md_allow_write(mddev_t *mddev)
5832 {
5833         if (!mddev->pers)
5834                 return 0;
5835         if (mddev->ro)
5836                 return 0;
5837         if (!mddev->pers->sync_request)
5838                 return 0;
5839
5840         spin_lock_irq(&mddev->write_lock);
5841         if (mddev->in_sync) {
5842                 mddev->in_sync = 0;
5843                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5844                 if (mddev->safemode_delay &&
5845                     mddev->safemode == 0)
5846                         mddev->safemode = 1;
5847                 spin_unlock_irq(&mddev->write_lock);
5848                 md_update_sb(mddev, 0);
5849                 sysfs_notify_dirent(mddev->sysfs_state);
5850         } else
5851                 spin_unlock_irq(&mddev->write_lock);
5852
5853         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5854                 return -EAGAIN;
5855         else
5856                 return 0;
5857 }
5858 EXPORT_SYMBOL_GPL(md_allow_write);
5859
5860 #define SYNC_MARKS      10
5861 #define SYNC_MARK_STEP  (3*HZ)
5862 void md_do_sync(mddev_t *mddev)
5863 {
5864         mddev_t *mddev2;
5865         unsigned int currspeed = 0,
5866                  window;
5867         sector_t max_sectors,j, io_sectors;
5868         unsigned long mark[SYNC_MARKS];
5869         sector_t mark_cnt[SYNC_MARKS];
5870         int last_mark,m;
5871         struct list_head *tmp;
5872         sector_t last_check;
5873         int skipped = 0;
5874         mdk_rdev_t *rdev;
5875         char *desc;
5876
5877         /* just incase thread restarts... */
5878         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5879                 return;
5880         if (mddev->ro) /* never try to sync a read-only array */
5881                 return;
5882
5883         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5884                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5885                         desc = "data-check";
5886                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5887                         desc = "requested-resync";
5888                 else
5889                         desc = "resync";
5890         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5891                 desc = "reshape";
5892         else
5893                 desc = "recovery";
5894
5895         /* we overload curr_resync somewhat here.
5896          * 0 == not engaged in resync at all
5897          * 2 == checking that there is no conflict with another sync
5898          * 1 == like 2, but have yielded to allow conflicting resync to
5899          *              commense
5900          * other == active in resync - this many blocks
5901          *
5902          * Before starting a resync we must have set curr_resync to
5903          * 2, and then checked that every "conflicting" array has curr_resync
5904          * less than ours.  When we find one that is the same or higher
5905          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5906          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5907          * This will mean we have to start checking from the beginning again.
5908          *
5909          */
5910
5911         do {
5912                 mddev->curr_resync = 2;
5913
5914         try_again:
5915                 if (kthread_should_stop()) {
5916                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5917                         goto skip;
5918                 }
5919                 for_each_mddev(mddev2, tmp) {
5920                         if (mddev2 == mddev)
5921                                 continue;
5922                         if (!mddev->parallel_resync
5923                         &&  mddev2->curr_resync
5924                         &&  match_mddev_units(mddev, mddev2)) {
5925                                 DEFINE_WAIT(wq);
5926                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5927                                         /* arbitrarily yield */
5928                                         mddev->curr_resync = 1;
5929                                         wake_up(&resync_wait);
5930                                 }
5931                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5932                                         /* no need to wait here, we can wait the next
5933                                          * time 'round when curr_resync == 2
5934                                          */
5935                                         continue;
5936                                 /* We need to wait 'interruptible' so as not to
5937                                  * contribute to the load average, and not to
5938                                  * be caught by 'softlockup'
5939                                  */
5940                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
5941                                 if (!kthread_should_stop() &&
5942                                     mddev2->curr_resync >= mddev->curr_resync) {
5943                                         printk(KERN_INFO "md: delaying %s of %s"
5944                                                " until %s has finished (they"
5945                                                " share one or more physical units)\n",
5946                                                desc, mdname(mddev), mdname(mddev2));
5947                                         mddev_put(mddev2);
5948                                         if (signal_pending(current))
5949                                                 flush_signals(current);
5950                                         schedule();
5951                                         finish_wait(&resync_wait, &wq);
5952                                         goto try_again;
5953                                 }
5954                                 finish_wait(&resync_wait, &wq);
5955                         }
5956                 }
5957         } while (mddev->curr_resync < 2);
5958
5959         j = 0;
5960         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5961                 /* resync follows the size requested by the personality,
5962                  * which defaults to physical size, but can be virtual size
5963                  */
5964                 max_sectors = mddev->resync_max_sectors;
5965                 mddev->resync_mismatches = 0;
5966                 /* we don't use the checkpoint if there's a bitmap */
5967                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5968                         j = mddev->resync_min;
5969                 else if (!mddev->bitmap)
5970                         j = mddev->recovery_cp;
5971
5972         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5973                 max_sectors = mddev->size << 1;
5974         else {
5975                 /* recovery follows the physical size of devices */
5976                 max_sectors = mddev->size << 1;
5977                 j = MaxSector;
5978                 list_for_each_entry(rdev, &mddev->disks, same_set)
5979                         if (rdev->raid_disk >= 0 &&
5980                             !test_bit(Faulty, &rdev->flags) &&
5981                             !test_bit(In_sync, &rdev->flags) &&
5982                             rdev->recovery_offset < j)
5983                                 j = rdev->recovery_offset;
5984         }
5985
5986         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5987         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5988                 " %d KB/sec/disk.\n", speed_min(mddev));
5989         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5990                "(but not more than %d KB/sec) for %s.\n",
5991                speed_max(mddev), desc);
5992
5993         is_mddev_idle(mddev); /* this also initializes IO event counters */
5994
5995         io_sectors = 0;
5996         for (m = 0; m < SYNC_MARKS; m++) {
5997                 mark[m] = jiffies;
5998                 mark_cnt[m] = io_sectors;
5999         }
6000         last_mark = 0;
6001         mddev->resync_mark = mark[last_mark];
6002         mddev->resync_mark_cnt = mark_cnt[last_mark];
6003
6004         /*
6005          * Tune reconstruction:
6006          */
6007         window = 32*(PAGE_SIZE/512);
6008         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6009                 window/2,(unsigned long long) max_sectors/2);
6010
6011         atomic_set(&mddev->recovery_active, 0);
6012         last_check = 0;
6013
6014         if (j>2) {
6015                 printk(KERN_INFO
6016                        "md: resuming %s of %s from checkpoint.\n",
6017                        desc, mdname(mddev));
6018                 mddev->curr_resync = j;
6019         }
6020
6021         while (j < max_sectors) {
6022                 sector_t sectors;
6023
6024                 skipped = 0;
6025                 if (j >= mddev->resync_max) {
6026                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6027                         wait_event(mddev->recovery_wait,
6028                                    mddev->resync_max > j
6029                                    || kthread_should_stop());
6030                 }
6031                 if (kthread_should_stop())
6032                         goto interrupted;
6033                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6034                                                   currspeed < speed_min(mddev));
6035                 if (sectors == 0) {
6036                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6037                         goto out;
6038                 }
6039
6040                 if (!skipped) { /* actual IO requested */
6041                         io_sectors += sectors;
6042                         atomic_add(sectors, &mddev->recovery_active);
6043                 }
6044
6045                 j += sectors;
6046                 if (j>1) mddev->curr_resync = j;
6047                 mddev->curr_mark_cnt = io_sectors;
6048                 if (last_check == 0)
6049                         /* this is the earliers that rebuilt will be
6050                          * visible in /proc/mdstat
6051                          */
6052                         md_new_event(mddev);
6053
6054                 if (last_check + window > io_sectors || j == max_sectors)
6055                         continue;
6056
6057                 last_check = io_sectors;
6058
6059                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6060                         break;
6061
6062         repeat:
6063                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6064                         /* step marks */
6065                         int next = (last_mark+1) % SYNC_MARKS;
6066
6067                         mddev->resync_mark = mark[next];
6068                         mddev->resync_mark_cnt = mark_cnt[next];
6069                         mark[next] = jiffies;
6070                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6071                         last_mark = next;
6072                 }
6073
6074
6075                 if (kthread_should_stop())
6076                         goto interrupted;
6077
6078
6079                 /*
6080                  * this loop exits only if either when we are slower than
6081                  * the 'hard' speed limit, or the system was IO-idle for
6082                  * a jiffy.
6083                  * the system might be non-idle CPU-wise, but we only care
6084                  * about not overloading the IO subsystem. (things like an
6085                  * e2fsck being done on the RAID array should execute fast)
6086                  */
6087                 blk_unplug(mddev->queue);
6088                 cond_resched();
6089
6090                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6091                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6092
6093                 if (currspeed > speed_min(mddev)) {
6094                         if ((currspeed > speed_max(mddev)) ||
6095                                         !is_mddev_idle(mddev)) {
6096                                 msleep(500);
6097                                 goto repeat;
6098                         }
6099                 }
6100         }
6101         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6102         /*
6103          * this also signals 'finished resyncing' to md_stop
6104          */
6105  out:
6106         blk_unplug(mddev->queue);
6107
6108         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6109
6110         /* tell personality that we are finished */
6111         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6112
6113         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6114             mddev->curr_resync > 2) {
6115                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6116                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6117                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6118                                         printk(KERN_INFO
6119                                                "md: checkpointing %s of %s.\n",
6120                                                desc, mdname(mddev));
6121                                         mddev->recovery_cp = mddev->curr_resync;
6122                                 }
6123                         } else
6124                                 mddev->recovery_cp = MaxSector;
6125                 } else {
6126                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6127                                 mddev->curr_resync = MaxSector;
6128                         list_for_each_entry(rdev, &mddev->disks, same_set)
6129                                 if (rdev->raid_disk >= 0 &&
6130                                     !test_bit(Faulty, &rdev->flags) &&
6131                                     !test_bit(In_sync, &rdev->flags) &&
6132                                     rdev->recovery_offset < mddev->curr_resync)
6133                                         rdev->recovery_offset = mddev->curr_resync;
6134                 }
6135         }
6136         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6137
6138  skip:
6139         mddev->curr_resync = 0;
6140         mddev->resync_min = 0;
6141         mddev->resync_max = MaxSector;
6142         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6143         wake_up(&resync_wait);
6144         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6145         md_wakeup_thread(mddev->thread);
6146         return;
6147
6148  interrupted:
6149         /*
6150          * got a signal, exit.
6151          */
6152         printk(KERN_INFO
6153                "md: md_do_sync() got signal ... exiting\n");
6154         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6155         goto out;
6156
6157 }
6158 EXPORT_SYMBOL_GPL(md_do_sync);
6159
6160
6161 static int remove_and_add_spares(mddev_t *mddev)
6162 {
6163         mdk_rdev_t *rdev;
6164         int spares = 0;
6165
6166         list_for_each_entry(rdev, &mddev->disks, same_set)
6167                 if (rdev->raid_disk >= 0 &&
6168                     !test_bit(Blocked, &rdev->flags) &&
6169                     (test_bit(Faulty, &rdev->flags) ||
6170                      ! test_bit(In_sync, &rdev->flags)) &&
6171                     atomic_read(&rdev->nr_pending)==0) {
6172                         if (mddev->pers->hot_remove_disk(
6173                                     mddev, rdev->raid_disk)==0) {
6174                                 char nm[20];
6175                                 sprintf(nm,"rd%d", rdev->raid_disk);
6176                                 sysfs_remove_link(&mddev->kobj, nm);
6177                                 rdev->raid_disk = -1;
6178                         }
6179                 }
6180
6181         if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6182                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6183                         if (rdev->raid_disk >= 0 &&
6184                             !test_bit(In_sync, &rdev->flags) &&
6185                             !test_bit(Blocked, &rdev->flags))
6186                                 spares++;
6187                         if (rdev->raid_disk < 0
6188                             && !test_bit(Faulty, &rdev->flags)) {
6189                                 rdev->recovery_offset = 0;
6190                                 if (mddev->pers->
6191                                     hot_add_disk(mddev, rdev) == 0) {
6192                                         char nm[20];
6193                                         sprintf(nm, "rd%d", rdev->raid_disk);
6194                                         if (sysfs_create_link(&mddev->kobj,
6195                                                               &rdev->kobj, nm))
6196                                                 printk(KERN_WARNING
6197                                                        "md: cannot register "
6198                                                        "%s for %s\n",
6199                                                        nm, mdname(mddev));
6200                                         spares++;
6201                                         md_new_event(mddev);
6202                                 } else
6203                                         break;
6204                         }
6205                 }
6206         }
6207         return spares;
6208 }
6209 /*
6210  * This routine is regularly called by all per-raid-array threads to
6211  * deal with generic issues like resync and super-block update.
6212  * Raid personalities that don't have a thread (linear/raid0) do not
6213  * need this as they never do any recovery or update the superblock.
6214  *
6215  * It does not do any resync itself, but rather "forks" off other threads
6216  * to do that as needed.
6217  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6218  * "->recovery" and create a thread at ->sync_thread.
6219  * When the thread finishes it sets MD_RECOVERY_DONE
6220  * and wakeups up this thread which will reap the thread and finish up.
6221  * This thread also removes any faulty devices (with nr_pending == 0).
6222  *
6223  * The overall approach is:
6224  *  1/ if the superblock needs updating, update it.
6225  *  2/ If a recovery thread is running, don't do anything else.
6226  *  3/ If recovery has finished, clean up, possibly marking spares active.
6227  *  4/ If there are any faulty devices, remove them.
6228  *  5/ If array is degraded, try to add spares devices
6229  *  6/ If array has spares or is not in-sync, start a resync thread.
6230  */
6231 void md_check_recovery(mddev_t *mddev)
6232 {
6233         mdk_rdev_t *rdev;
6234
6235
6236         if (mddev->bitmap)
6237                 bitmap_daemon_work(mddev->bitmap);
6238
6239         if (mddev->ro)
6240                 return;
6241
6242         if (signal_pending(current)) {
6243                 if (mddev->pers->sync_request && !mddev->external) {
6244                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6245                                mdname(mddev));
6246                         mddev->safemode = 2;
6247                 }
6248                 flush_signals(current);
6249         }
6250
6251         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6252                 return;
6253         if ( ! (
6254                 (mddev->flags && !mddev->external) ||
6255                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6256                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6257                 (mddev->external == 0 && mddev->safemode == 1) ||
6258                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6259                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6260                 ))
6261                 return;
6262
6263         if (mddev_trylock(mddev)) {
6264                 int spares = 0;
6265
6266                 if (mddev->ro) {
6267                         /* Only thing we do on a ro array is remove
6268                          * failed devices.
6269                          */
6270                         remove_and_add_spares(mddev);
6271                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6272                         goto unlock;
6273                 }
6274
6275                 if (!mddev->external) {
6276                         int did_change = 0;
6277                         spin_lock_irq(&mddev->write_lock);
6278                         if (mddev->safemode &&
6279                             !atomic_read(&mddev->writes_pending) &&
6280                             !mddev->in_sync &&
6281                             mddev->recovery_cp == MaxSector) {
6282                                 mddev->in_sync = 1;
6283                                 did_change = 1;
6284                                 if (mddev->persistent)
6285                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6286                         }
6287                         if (mddev->safemode == 1)
6288                                 mddev->safemode = 0;
6289                         spin_unlock_irq(&mddev->write_lock);
6290                         if (did_change)
6291                                 sysfs_notify_dirent(mddev->sysfs_state);
6292                 }
6293
6294                 if (mddev->flags)
6295                         md_update_sb(mddev, 0);
6296
6297                 list_for_each_entry(rdev, &mddev->disks, same_set)
6298                         if (test_and_clear_bit(StateChanged, &rdev->flags))
6299                                 sysfs_notify_dirent(rdev->sysfs_state);
6300
6301
6302                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6303                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6304                         /* resync/recovery still happening */
6305                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6306                         goto unlock;
6307                 }
6308                 if (mddev->sync_thread) {
6309                         /* resync has finished, collect result */
6310                         md_unregister_thread(mddev->sync_thread);
6311                         mddev->sync_thread = NULL;
6312                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6313                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6314                                 /* success...*/
6315                                 /* activate any spares */
6316                                 if (mddev->pers->spare_active(mddev))
6317                                         sysfs_notify(&mddev->kobj, NULL,
6318                                                      "degraded");
6319                         }
6320                         md_update_sb(mddev, 1);
6321
6322                         /* if array is no-longer degraded, then any saved_raid_disk
6323                          * information must be scrapped
6324                          */
6325                         if (!mddev->degraded)
6326                                 list_for_each_entry(rdev, &mddev->disks, same_set)
6327                                         rdev->saved_raid_disk = -1;
6328
6329                         mddev->recovery = 0;
6330                         /* flag recovery needed just to double check */
6331                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6332                         sysfs_notify_dirent(mddev->sysfs_action);
6333                         md_new_event(mddev);
6334                         goto unlock;
6335                 }
6336                 /* Set RUNNING before clearing NEEDED to avoid
6337                  * any transients in the value of "sync_action".
6338                  */
6339                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6340                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6341                 /* Clear some bits that don't mean anything, but
6342                  * might be left set
6343                  */
6344                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6345                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6346
6347                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6348                         goto unlock;
6349                 /* no recovery is running.
6350                  * remove any failed drives, then
6351                  * add spares if possible.
6352                  * Spare are also removed and re-added, to allow
6353                  * the personality to fail the re-add.
6354                  */
6355
6356                 if (mddev->reshape_position != MaxSector) {
6357                         if (mddev->pers->check_reshape(mddev) != 0)
6358                                 /* Cannot proceed */
6359                                 goto unlock;
6360                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6361                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6362                 } else if ((spares = remove_and_add_spares(mddev))) {
6363                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6364                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6365                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6366                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6367                 } else if (mddev->recovery_cp < MaxSector) {
6368                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6369                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6370                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6371                         /* nothing to be done ... */
6372                         goto unlock;
6373
6374                 if (mddev->pers->sync_request) {
6375                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6376                                 /* We are adding a device or devices to an array
6377                                  * which has the bitmap stored on all devices.
6378                                  * So make sure all bitmap pages get written
6379                                  */
6380                                 bitmap_write_all(mddev->bitmap);
6381                         }
6382                         mddev->sync_thread = md_register_thread(md_do_sync,
6383                                                                 mddev,
6384                                                                 "%s_resync");
6385                         if (!mddev->sync_thread) {
6386                                 printk(KERN_ERR "%s: could not start resync"
6387                                         " thread...\n",
6388                                         mdname(mddev));
6389                                 /* leave the spares where they are, it shouldn't hurt */
6390                                 mddev->recovery = 0;
6391                         } else
6392                                 md_wakeup_thread(mddev->sync_thread);
6393                         sysfs_notify_dirent(mddev->sysfs_action);
6394                         md_new_event(mddev);
6395                 }
6396         unlock:
6397                 if (!mddev->sync_thread) {
6398                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6399                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6400                                                &mddev->recovery))
6401                                 if (mddev->sysfs_action)
6402                                         sysfs_notify_dirent(mddev->sysfs_action);
6403                 }
6404                 mddev_unlock(mddev);
6405         }
6406 }
6407
6408 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6409 {
6410         sysfs_notify_dirent(rdev->sysfs_state);
6411         wait_event_timeout(rdev->blocked_wait,
6412                            !test_bit(Blocked, &rdev->flags),
6413                            msecs_to_jiffies(5000));
6414         rdev_dec_pending(rdev, mddev);
6415 }
6416 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6417
6418 static int md_notify_reboot(struct notifier_block *this,
6419                             unsigned long code, void *x)
6420 {
6421         struct list_head *tmp;
6422         mddev_t *mddev;
6423
6424         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6425
6426                 printk(KERN_INFO "md: stopping all md devices.\n");
6427
6428                 for_each_mddev(mddev, tmp)
6429                         if (mddev_trylock(mddev)) {
6430                                 /* Force a switch to readonly even array
6431                                  * appears to still be in use.  Hence
6432                                  * the '100'.
6433                                  */
6434                                 do_md_stop(mddev, 1, 100);
6435                                 mddev_unlock(mddev);
6436                         }
6437                 /*
6438                  * certain more exotic SCSI devices are known to be
6439                  * volatile wrt too early system reboots. While the
6440                  * right place to handle this issue is the given
6441                  * driver, we do want to have a safe RAID driver ...
6442                  */
6443                 mdelay(1000*1);
6444         }
6445         return NOTIFY_DONE;
6446 }
6447
6448 static struct notifier_block md_notifier = {
6449         .notifier_call  = md_notify_reboot,
6450         .next           = NULL,
6451         .priority       = INT_MAX, /* before any real devices */
6452 };
6453
6454 static void md_geninit(void)
6455 {
6456         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6457
6458         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6459 }
6460
6461 static int __init md_init(void)
6462 {
6463         if (register_blkdev(MAJOR_NR, "md"))
6464                 return -1;
6465         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6466                 unregister_blkdev(MAJOR_NR, "md");
6467                 return -1;
6468         }
6469         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6470                             md_probe, NULL, NULL);
6471         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6472                             md_probe, NULL, NULL);
6473
6474         register_reboot_notifier(&md_notifier);
6475         raid_table_header = register_sysctl_table(raid_root_table);
6476
6477         md_geninit();
6478         return 0;
6479 }
6480
6481
6482 #ifndef MODULE
6483
6484 /*
6485  * Searches all registered partitions for autorun RAID arrays
6486  * at boot time.
6487  */
6488
6489 static LIST_HEAD(all_detected_devices);
6490 struct detected_devices_node {
6491         struct list_head list;
6492         dev_t dev;
6493 };
6494
6495 void md_autodetect_dev(dev_t dev)
6496 {
6497         struct detected_devices_node *node_detected_dev;
6498
6499         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6500         if (node_detected_dev) {
6501                 node_detected_dev->dev = dev;
6502                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6503         } else {
6504                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6505                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6506         }
6507 }
6508
6509
6510 static void autostart_arrays(int part)
6511 {
6512         mdk_rdev_t *rdev;
6513         struct detected_devices_node *node_detected_dev;
6514         dev_t dev;
6515         int i_scanned, i_passed;
6516
6517         i_scanned = 0;
6518         i_passed = 0;
6519
6520         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6521
6522         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6523                 i_scanned++;
6524                 node_detected_dev = list_entry(all_detected_devices.next,
6525                                         struct detected_devices_node, list);
6526                 list_del(&node_detected_dev->list);
6527                 dev = node_detected_dev->dev;
6528                 kfree(node_detected_dev);
6529                 rdev = md_import_device(dev,0, 90);
6530                 if (IS_ERR(rdev))
6531                         continue;
6532
6533                 if (test_bit(Faulty, &rdev->flags)) {
6534                         MD_BUG();
6535                         continue;
6536                 }
6537                 set_bit(AutoDetected, &rdev->flags);
6538                 list_add(&rdev->same_set, &pending_raid_disks);
6539                 i_passed++;
6540         }
6541
6542         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6543                                                 i_scanned, i_passed);
6544
6545         autorun_devices(part);
6546 }
6547
6548 #endif /* !MODULE */
6549
6550 static __exit void md_exit(void)
6551 {
6552         mddev_t *mddev;
6553         struct list_head *tmp;
6554
6555         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6556         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6557
6558         unregister_blkdev(MAJOR_NR,"md");
6559         unregister_blkdev(mdp_major, "mdp");
6560         unregister_reboot_notifier(&md_notifier);
6561         unregister_sysctl_table(raid_table_header);
6562         remove_proc_entry("mdstat", NULL);
6563         for_each_mddev(mddev, tmp) {
6564                 export_array(mddev);
6565                 mddev->hold_active = 0;
6566         }
6567 }
6568
6569 subsys_initcall(md_init);
6570 module_exit(md_exit)
6571
6572 static int get_ro(char *buffer, struct kernel_param *kp)
6573 {
6574         return sprintf(buffer, "%d", start_readonly);
6575 }
6576 static int set_ro(const char *val, struct kernel_param *kp)
6577 {
6578         char *e;
6579         int num = simple_strtoul(val, &e, 10);
6580         if (*val && (*e == '\0' || *e == '\n')) {
6581                 start_readonly = num;
6582                 return 0;
6583         }
6584         return -EINVAL;
6585 }
6586
6587 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6588 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6589
6590 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6591
6592 EXPORT_SYMBOL(register_md_personality);
6593 EXPORT_SYMBOL(unregister_md_personality);
6594 EXPORT_SYMBOL(md_error);
6595 EXPORT_SYMBOL(md_done_sync);
6596 EXPORT_SYMBOL(md_write_start);
6597 EXPORT_SYMBOL(md_write_end);
6598 EXPORT_SYMBOL(md_register_thread);
6599 EXPORT_SYMBOL(md_unregister_thread);
6600 EXPORT_SYMBOL(md_wakeup_thread);
6601 EXPORT_SYMBOL(md_check_recovery);
6602 MODULE_LICENSE("GPL");
6603 MODULE_ALIAS("md");
6604 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);