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         return 0;
1504
1505  fail:
1506         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1507                b, mdname(mddev));
1508         return err;
1509 }
1510
1511 static void md_delayed_delete(struct work_struct *ws)
1512 {
1513         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1514         kobject_del(&rdev->kobj);
1515         kobject_put(&rdev->kobj);
1516 }
1517
1518 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1519 {
1520         char b[BDEVNAME_SIZE];
1521         if (!rdev->mddev) {
1522                 MD_BUG();
1523                 return;
1524         }
1525         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1526         list_del_rcu(&rdev->same_set);
1527         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1528         rdev->mddev = NULL;
1529         sysfs_remove_link(&rdev->kobj, "block");
1530         sysfs_put(rdev->sysfs_state);
1531         rdev->sysfs_state = NULL;
1532         /* We need to delay this, otherwise we can deadlock when
1533          * writing to 'remove' to "dev/state".  We also need
1534          * to delay it due to rcu usage.
1535          */
1536         synchronize_rcu();
1537         INIT_WORK(&rdev->del_work, md_delayed_delete);
1538         kobject_get(&rdev->kobj);
1539         schedule_work(&rdev->del_work);
1540 }
1541
1542 /*
1543  * prevent the device from being mounted, repartitioned or
1544  * otherwise reused by a RAID array (or any other kernel
1545  * subsystem), by bd_claiming the device.
1546  */
1547 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1548 {
1549         int err = 0;
1550         struct block_device *bdev;
1551         char b[BDEVNAME_SIZE];
1552
1553         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1554         if (IS_ERR(bdev)) {
1555                 printk(KERN_ERR "md: could not open %s.\n",
1556                         __bdevname(dev, b));
1557                 return PTR_ERR(bdev);
1558         }
1559         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1560         if (err) {
1561                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1562                         bdevname(bdev, b));
1563                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1564                 return err;
1565         }
1566         if (!shared)
1567                 set_bit(AllReserved, &rdev->flags);
1568         rdev->bdev = bdev;
1569         return err;
1570 }
1571
1572 static void unlock_rdev(mdk_rdev_t *rdev)
1573 {
1574         struct block_device *bdev = rdev->bdev;
1575         rdev->bdev = NULL;
1576         if (!bdev)
1577                 MD_BUG();
1578         bd_release(bdev);
1579         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1580 }
1581
1582 void md_autodetect_dev(dev_t dev);
1583
1584 static void export_rdev(mdk_rdev_t * rdev)
1585 {
1586         char b[BDEVNAME_SIZE];
1587         printk(KERN_INFO "md: export_rdev(%s)\n",
1588                 bdevname(rdev->bdev,b));
1589         if (rdev->mddev)
1590                 MD_BUG();
1591         free_disk_sb(rdev);
1592 #ifndef MODULE
1593         if (test_bit(AutoDetected, &rdev->flags))
1594                 md_autodetect_dev(rdev->bdev->bd_dev);
1595 #endif
1596         unlock_rdev(rdev);
1597         kobject_put(&rdev->kobj);
1598 }
1599
1600 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1601 {
1602         unbind_rdev_from_array(rdev);
1603         export_rdev(rdev);
1604 }
1605
1606 static void export_array(mddev_t *mddev)
1607 {
1608         mdk_rdev_t *rdev, *tmp;
1609
1610         rdev_for_each(rdev, tmp, mddev) {
1611                 if (!rdev->mddev) {
1612                         MD_BUG();
1613                         continue;
1614                 }
1615                 kick_rdev_from_array(rdev);
1616         }
1617         if (!list_empty(&mddev->disks))
1618                 MD_BUG();
1619         mddev->raid_disks = 0;
1620         mddev->major_version = 0;
1621 }
1622
1623 static void print_desc(mdp_disk_t *desc)
1624 {
1625         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1626                 desc->major,desc->minor,desc->raid_disk,desc->state);
1627 }
1628
1629 static void print_sb_90(mdp_super_t *sb)
1630 {
1631         int i;
1632
1633         printk(KERN_INFO
1634                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1635                 sb->major_version, sb->minor_version, sb->patch_version,
1636                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1637                 sb->ctime);
1638         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1639                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1640                 sb->md_minor, sb->layout, sb->chunk_size);
1641         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1642                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1643                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1644                 sb->failed_disks, sb->spare_disks,
1645                 sb->sb_csum, (unsigned long)sb->events_lo);
1646
1647         printk(KERN_INFO);
1648         for (i = 0; i < MD_SB_DISKS; i++) {
1649                 mdp_disk_t *desc;
1650
1651                 desc = sb->disks + i;
1652                 if (desc->number || desc->major || desc->minor ||
1653                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1654                         printk("     D %2d: ", i);
1655                         print_desc(desc);
1656                 }
1657         }
1658         printk(KERN_INFO "md:     THIS: ");
1659         print_desc(&sb->this_disk);
1660 }
1661
1662 static void print_sb_1(struct mdp_superblock_1 *sb)
1663 {
1664         __u8 *uuid;
1665
1666         uuid = sb->set_uuid;
1667         printk(KERN_INFO "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1668                         ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1669                KERN_INFO "md:    Name: \"%s\" CT:%llu\n",
1670                 le32_to_cpu(sb->major_version),
1671                 le32_to_cpu(sb->feature_map),
1672                 uuid[0], uuid[1], uuid[2], uuid[3],
1673                 uuid[4], uuid[5], uuid[6], uuid[7],
1674                 uuid[8], uuid[9], uuid[10], uuid[11],
1675                 uuid[12], uuid[13], uuid[14], uuid[15],
1676                 sb->set_name,
1677                 (unsigned long long)le64_to_cpu(sb->ctime)
1678                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1679
1680         uuid = sb->device_uuid;
1681         printk(KERN_INFO "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1682                         " RO:%llu\n"
1683                KERN_INFO "md:     Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1684                         ":%02x%02x%02x%02x%02x%02x\n"
1685                KERN_INFO "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1686                KERN_INFO "md:         (MaxDev:%u) \n",
1687                 le32_to_cpu(sb->level),
1688                 (unsigned long long)le64_to_cpu(sb->size),
1689                 le32_to_cpu(sb->raid_disks),
1690                 le32_to_cpu(sb->layout),
1691                 le32_to_cpu(sb->chunksize),
1692                 (unsigned long long)le64_to_cpu(sb->data_offset),
1693                 (unsigned long long)le64_to_cpu(sb->data_size),
1694                 (unsigned long long)le64_to_cpu(sb->super_offset),
1695                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1696                 le32_to_cpu(sb->dev_number),
1697                 uuid[0], uuid[1], uuid[2], uuid[3],
1698                 uuid[4], uuid[5], uuid[6], uuid[7],
1699                 uuid[8], uuid[9], uuid[10], uuid[11],
1700                 uuid[12], uuid[13], uuid[14], uuid[15],
1701                 sb->devflags,
1702                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1703                 (unsigned long long)le64_to_cpu(sb->events),
1704                 (unsigned long long)le64_to_cpu(sb->resync_offset),
1705                 le32_to_cpu(sb->sb_csum),
1706                 le32_to_cpu(sb->max_dev)
1707                 );
1708 }
1709
1710 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1711 {
1712         char b[BDEVNAME_SIZE];
1713         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1714                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1715                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1716                 rdev->desc_nr);
1717         if (rdev->sb_loaded) {
1718                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1719                 switch (major_version) {
1720                 case 0:
1721                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1722                         break;
1723                 case 1:
1724                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1725                         break;
1726                 }
1727         } else
1728                 printk(KERN_INFO "md: no rdev superblock!\n");
1729 }
1730
1731 static void md_print_devices(void)
1732 {
1733         struct list_head *tmp;
1734         mdk_rdev_t *rdev;
1735         mddev_t *mddev;
1736         char b[BDEVNAME_SIZE];
1737
1738         printk("\n");
1739         printk("md:     **********************************\n");
1740         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1741         printk("md:     **********************************\n");
1742         for_each_mddev(mddev, tmp) {
1743
1744                 if (mddev->bitmap)
1745                         bitmap_print_sb(mddev->bitmap);
1746                 else
1747                         printk("%s: ", mdname(mddev));
1748                 list_for_each_entry(rdev, &mddev->disks, same_set)
1749                         printk("<%s>", bdevname(rdev->bdev,b));
1750                 printk("\n");
1751
1752                 list_for_each_entry(rdev, &mddev->disks, same_set)
1753                         print_rdev(rdev, mddev->major_version);
1754         }
1755         printk("md:     **********************************\n");
1756         printk("\n");
1757 }
1758
1759
1760 static void sync_sbs(mddev_t * mddev, int nospares)
1761 {
1762         /* Update each superblock (in-memory image), but
1763          * if we are allowed to, skip spares which already
1764          * have the right event counter, or have one earlier
1765          * (which would mean they aren't being marked as dirty
1766          * with the rest of the array)
1767          */
1768         mdk_rdev_t *rdev;
1769
1770         list_for_each_entry(rdev, &mddev->disks, same_set) {
1771                 if (rdev->sb_events == mddev->events ||
1772                     (nospares &&
1773                      rdev->raid_disk < 0 &&
1774                      (rdev->sb_events&1)==0 &&
1775                      rdev->sb_events+1 == mddev->events)) {
1776                         /* Don't update this superblock */
1777                         rdev->sb_loaded = 2;
1778                 } else {
1779                         super_types[mddev->major_version].
1780                                 sync_super(mddev, rdev);
1781                         rdev->sb_loaded = 1;
1782                 }
1783         }
1784 }
1785
1786 static void md_update_sb(mddev_t * mddev, int force_change)
1787 {
1788         mdk_rdev_t *rdev;
1789         int sync_req;
1790         int nospares = 0;
1791
1792         if (mddev->external)
1793                 return;
1794 repeat:
1795         spin_lock_irq(&mddev->write_lock);
1796
1797         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1798         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1799                 force_change = 1;
1800         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1801                 /* just a clean<-> dirty transition, possibly leave spares alone,
1802                  * though if events isn't the right even/odd, we will have to do
1803                  * spares after all
1804                  */
1805                 nospares = 1;
1806         if (force_change)
1807                 nospares = 0;
1808         if (mddev->degraded)
1809                 /* If the array is degraded, then skipping spares is both
1810                  * dangerous and fairly pointless.
1811                  * Dangerous because a device that was removed from the array
1812                  * might have a event_count that still looks up-to-date,
1813                  * so it can be re-added without a resync.
1814                  * Pointless because if there are any spares to skip,
1815                  * then a recovery will happen and soon that array won't
1816                  * be degraded any more and the spare can go back to sleep then.
1817                  */
1818                 nospares = 0;
1819
1820         sync_req = mddev->in_sync;
1821         mddev->utime = get_seconds();
1822
1823         /* If this is just a dirty<->clean transition, and the array is clean
1824          * and 'events' is odd, we can roll back to the previous clean state */
1825         if (nospares
1826             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1827             && (mddev->events & 1)
1828             && mddev->events != 1)
1829                 mddev->events--;
1830         else {
1831                 /* otherwise we have to go forward and ... */
1832                 mddev->events ++;
1833                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1834                         /* .. if the array isn't clean, insist on an odd 'events' */
1835                         if ((mddev->events&1)==0) {
1836                                 mddev->events++;
1837                                 nospares = 0;
1838                         }
1839                 } else {
1840                         /* otherwise insist on an even 'events' (for clean states) */
1841                         if ((mddev->events&1)) {
1842                                 mddev->events++;
1843                                 nospares = 0;
1844                         }
1845                 }
1846         }
1847
1848         if (!mddev->events) {
1849                 /*
1850                  * oops, this 64-bit counter should never wrap.
1851                  * Either we are in around ~1 trillion A.C., assuming
1852                  * 1 reboot per second, or we have a bug:
1853                  */
1854                 MD_BUG();
1855                 mddev->events --;
1856         }
1857
1858         /*
1859          * do not write anything to disk if using
1860          * nonpersistent superblocks
1861          */
1862         if (!mddev->persistent) {
1863                 if (!mddev->external)
1864                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1865
1866                 spin_unlock_irq(&mddev->write_lock);
1867                 wake_up(&mddev->sb_wait);
1868                 return;
1869         }
1870         sync_sbs(mddev, nospares);
1871         spin_unlock_irq(&mddev->write_lock);
1872
1873         dprintk(KERN_INFO
1874                 "md: updating %s RAID superblock on device (in sync %d)\n",
1875                 mdname(mddev),mddev->in_sync);
1876
1877         bitmap_update_sb(mddev->bitmap);
1878         list_for_each_entry(rdev, &mddev->disks, same_set) {
1879                 char b[BDEVNAME_SIZE];
1880                 dprintk(KERN_INFO "md: ");
1881                 if (rdev->sb_loaded != 1)
1882                         continue; /* no noise on spare devices */
1883                 if (test_bit(Faulty, &rdev->flags))
1884                         dprintk("(skipping faulty ");
1885
1886                 dprintk("%s ", bdevname(rdev->bdev,b));
1887                 if (!test_bit(Faulty, &rdev->flags)) {
1888                         md_super_write(mddev,rdev,
1889                                        rdev->sb_start, rdev->sb_size,
1890                                        rdev->sb_page);
1891                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1892                                 bdevname(rdev->bdev,b),
1893                                 (unsigned long long)rdev->sb_start);
1894                         rdev->sb_events = mddev->events;
1895
1896                 } else
1897                         dprintk(")\n");
1898                 if (mddev->level == LEVEL_MULTIPATH)
1899                         /* only need to write one superblock... */
1900                         break;
1901         }
1902         md_super_wait(mddev);
1903         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1904
1905         spin_lock_irq(&mddev->write_lock);
1906         if (mddev->in_sync != sync_req ||
1907             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1908                 /* have to write it out again */
1909                 spin_unlock_irq(&mddev->write_lock);
1910                 goto repeat;
1911         }
1912         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1913         spin_unlock_irq(&mddev->write_lock);
1914         wake_up(&mddev->sb_wait);
1915
1916 }
1917
1918 /* words written to sysfs files may, or may not, be \n terminated.
1919  * We want to accept with case. For this we use cmd_match.
1920  */
1921 static int cmd_match(const char *cmd, const char *str)
1922 {
1923         /* See if cmd, written into a sysfs file, matches
1924          * str.  They must either be the same, or cmd can
1925          * have a trailing newline
1926          */
1927         while (*cmd && *str && *cmd == *str) {
1928                 cmd++;
1929                 str++;
1930         }
1931         if (*cmd == '\n')
1932                 cmd++;
1933         if (*str || *cmd)
1934                 return 0;
1935         return 1;
1936 }
1937
1938 struct rdev_sysfs_entry {
1939         struct attribute attr;
1940         ssize_t (*show)(mdk_rdev_t *, char *);
1941         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1942 };
1943
1944 static ssize_t
1945 state_show(mdk_rdev_t *rdev, char *page)
1946 {
1947         char *sep = "";
1948         size_t len = 0;
1949
1950         if (test_bit(Faulty, &rdev->flags)) {
1951                 len+= sprintf(page+len, "%sfaulty",sep);
1952                 sep = ",";
1953         }
1954         if (test_bit(In_sync, &rdev->flags)) {
1955                 len += sprintf(page+len, "%sin_sync",sep);
1956                 sep = ",";
1957         }
1958         if (test_bit(WriteMostly, &rdev->flags)) {
1959                 len += sprintf(page+len, "%swrite_mostly",sep);
1960                 sep = ",";
1961         }
1962         if (test_bit(Blocked, &rdev->flags)) {
1963                 len += sprintf(page+len, "%sblocked", sep);
1964                 sep = ",";
1965         }
1966         if (!test_bit(Faulty, &rdev->flags) &&
1967             !test_bit(In_sync, &rdev->flags)) {
1968                 len += sprintf(page+len, "%sspare", sep);
1969                 sep = ",";
1970         }
1971         return len+sprintf(page+len, "\n");
1972 }
1973
1974 static ssize_t
1975 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1976 {
1977         /* can write
1978          *  faulty  - simulates and error
1979          *  remove  - disconnects the device
1980          *  writemostly - sets write_mostly
1981          *  -writemostly - clears write_mostly
1982          *  blocked - sets the Blocked flag
1983          *  -blocked - clears the Blocked flag
1984          */
1985         int err = -EINVAL;
1986         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1987                 md_error(rdev->mddev, rdev);
1988                 err = 0;
1989         } else if (cmd_match(buf, "remove")) {
1990                 if (rdev->raid_disk >= 0)
1991                         err = -EBUSY;
1992                 else {
1993                         mddev_t *mddev = rdev->mddev;
1994                         kick_rdev_from_array(rdev);
1995                         if (mddev->pers)
1996                                 md_update_sb(mddev, 1);
1997                         md_new_event(mddev);
1998                         err = 0;
1999                 }
2000         } else if (cmd_match(buf, "writemostly")) {
2001                 set_bit(WriteMostly, &rdev->flags);
2002                 err = 0;
2003         } else if (cmd_match(buf, "-writemostly")) {
2004                 clear_bit(WriteMostly, &rdev->flags);
2005                 err = 0;
2006         } else if (cmd_match(buf, "blocked")) {
2007                 set_bit(Blocked, &rdev->flags);
2008                 err = 0;
2009         } else if (cmd_match(buf, "-blocked")) {
2010                 clear_bit(Blocked, &rdev->flags);
2011                 wake_up(&rdev->blocked_wait);
2012                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2013                 md_wakeup_thread(rdev->mddev->thread);
2014
2015                 err = 0;
2016         }
2017         if (!err && rdev->sysfs_state)
2018                 sysfs_notify_dirent(rdev->sysfs_state);
2019         return err ? err : len;
2020 }
2021 static struct rdev_sysfs_entry rdev_state =
2022 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2023
2024 static ssize_t
2025 errors_show(mdk_rdev_t *rdev, char *page)
2026 {
2027         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2028 }
2029
2030 static ssize_t
2031 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2032 {
2033         char *e;
2034         unsigned long n = simple_strtoul(buf, &e, 10);
2035         if (*buf && (*e == 0 || *e == '\n')) {
2036                 atomic_set(&rdev->corrected_errors, n);
2037                 return len;
2038         }
2039         return -EINVAL;
2040 }
2041 static struct rdev_sysfs_entry rdev_errors =
2042 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2043
2044 static ssize_t
2045 slot_show(mdk_rdev_t *rdev, char *page)
2046 {
2047         if (rdev->raid_disk < 0)
2048                 return sprintf(page, "none\n");
2049         else
2050                 return sprintf(page, "%d\n", rdev->raid_disk);
2051 }
2052
2053 static ssize_t
2054 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2055 {
2056         char *e;
2057         int err;
2058         char nm[20];
2059         int slot = simple_strtoul(buf, &e, 10);
2060         if (strncmp(buf, "none", 4)==0)
2061                 slot = -1;
2062         else if (e==buf || (*e && *e!= '\n'))
2063                 return -EINVAL;
2064         if (rdev->mddev->pers && slot == -1) {
2065                 /* Setting 'slot' on an active array requires also
2066                  * updating the 'rd%d' link, and communicating
2067                  * with the personality with ->hot_*_disk.
2068                  * For now we only support removing
2069                  * failed/spare devices.  This normally happens automatically,
2070                  * but not when the metadata is externally managed.
2071                  */
2072                 if (rdev->raid_disk == -1)
2073                         return -EEXIST;
2074                 /* personality does all needed checks */
2075                 if (rdev->mddev->pers->hot_add_disk == NULL)
2076                         return -EINVAL;
2077                 err = rdev->mddev->pers->
2078                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2079                 if (err)
2080                         return err;
2081                 sprintf(nm, "rd%d", rdev->raid_disk);
2082                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2083                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2084                 md_wakeup_thread(rdev->mddev->thread);
2085         } else if (rdev->mddev->pers) {
2086                 mdk_rdev_t *rdev2;
2087                 /* Activating a spare .. or possibly reactivating
2088                  * if we every get bitmaps working here.
2089                  */
2090
2091                 if (rdev->raid_disk != -1)
2092                         return -EBUSY;
2093
2094                 if (rdev->mddev->pers->hot_add_disk == NULL)
2095                         return -EINVAL;
2096
2097                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2098                         if (rdev2->raid_disk == slot)
2099                                 return -EEXIST;
2100
2101                 rdev->raid_disk = slot;
2102                 if (test_bit(In_sync, &rdev->flags))
2103                         rdev->saved_raid_disk = slot;
2104                 else
2105                         rdev->saved_raid_disk = -1;
2106                 err = rdev->mddev->pers->
2107                         hot_add_disk(rdev->mddev, rdev);
2108                 if (err) {
2109                         rdev->raid_disk = -1;
2110                         return err;
2111                 } else
2112                         sysfs_notify_dirent(rdev->sysfs_state);
2113                 sprintf(nm, "rd%d", rdev->raid_disk);
2114                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2115                         printk(KERN_WARNING
2116                                "md: cannot register "
2117                                "%s for %s\n",
2118                                nm, mdname(rdev->mddev));
2119
2120                 /* don't wakeup anyone, leave that to userspace. */
2121         } else {
2122                 if (slot >= rdev->mddev->raid_disks)
2123                         return -ENOSPC;
2124                 rdev->raid_disk = slot;
2125                 /* assume it is working */
2126                 clear_bit(Faulty, &rdev->flags);
2127                 clear_bit(WriteMostly, &rdev->flags);
2128                 set_bit(In_sync, &rdev->flags);
2129                 sysfs_notify_dirent(rdev->sysfs_state);
2130         }
2131         return len;
2132 }
2133
2134
2135 static struct rdev_sysfs_entry rdev_slot =
2136 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2137
2138 static ssize_t
2139 offset_show(mdk_rdev_t *rdev, char *page)
2140 {
2141         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2142 }
2143
2144 static ssize_t
2145 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2146 {
2147         char *e;
2148         unsigned long long offset = simple_strtoull(buf, &e, 10);
2149         if (e==buf || (*e && *e != '\n'))
2150                 return -EINVAL;
2151         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2152                 return -EBUSY;
2153         if (rdev->size && rdev->mddev->external)
2154                 /* Must set offset before size, so overlap checks
2155                  * can be sane */
2156                 return -EBUSY;
2157         rdev->data_offset = offset;
2158         return len;
2159 }
2160
2161 static struct rdev_sysfs_entry rdev_offset =
2162 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2163
2164 static ssize_t
2165 rdev_size_show(mdk_rdev_t *rdev, char *page)
2166 {
2167         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2168 }
2169
2170 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2171 {
2172         /* check if two start/length pairs overlap */
2173         if (s1+l1 <= s2)
2174                 return 0;
2175         if (s2+l2 <= s1)
2176                 return 0;
2177         return 1;
2178 }
2179
2180 static ssize_t
2181 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2182 {
2183         unsigned long long size;
2184         unsigned long long oldsize = rdev->size;
2185         mddev_t *my_mddev = rdev->mddev;
2186
2187         if (strict_strtoull(buf, 10, &size) < 0)
2188                 return -EINVAL;
2189         if (my_mddev->pers && rdev->raid_disk >= 0) {
2190                 if (my_mddev->persistent) {
2191                         size = super_types[my_mddev->major_version].
2192                                 rdev_size_change(rdev, size * 2);
2193                         if (!size)
2194                                 return -EBUSY;
2195                 } else if (!size) {
2196                         size = (rdev->bdev->bd_inode->i_size >> 10);
2197                         size -= rdev->data_offset/2;
2198                 }
2199         }
2200         if (size < my_mddev->size)
2201                 return -EINVAL; /* component must fit device */
2202
2203         rdev->size = size;
2204         if (size > oldsize && my_mddev->external) {
2205                 /* need to check that all other rdevs with the same ->bdev
2206                  * do not overlap.  We need to unlock the mddev to avoid
2207                  * a deadlock.  We have already changed rdev->size, and if
2208                  * we have to change it back, we will have the lock again.
2209                  */
2210                 mddev_t *mddev;
2211                 int overlap = 0;
2212                 struct list_head *tmp;
2213
2214                 mddev_unlock(my_mddev);
2215                 for_each_mddev(mddev, tmp) {
2216                         mdk_rdev_t *rdev2;
2217
2218                         mddev_lock(mddev);
2219                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2220                                 if (test_bit(AllReserved, &rdev2->flags) ||
2221                                     (rdev->bdev == rdev2->bdev &&
2222                                      rdev != rdev2 &&
2223                                      overlaps(rdev->data_offset, rdev->size * 2,
2224                                               rdev2->data_offset,
2225                                               rdev2->size * 2))) {
2226                                         overlap = 1;
2227                                         break;
2228                                 }
2229                         mddev_unlock(mddev);
2230                         if (overlap) {
2231                                 mddev_put(mddev);
2232                                 break;
2233                         }
2234                 }
2235                 mddev_lock(my_mddev);
2236                 if (overlap) {
2237                         /* Someone else could have slipped in a size
2238                          * change here, but doing so is just silly.
2239                          * We put oldsize back because we *know* it is
2240                          * safe, and trust userspace not to race with
2241                          * itself
2242                          */
2243                         rdev->size = oldsize;
2244                         return -EBUSY;
2245                 }
2246         }
2247         return len;
2248 }
2249
2250 static struct rdev_sysfs_entry rdev_size =
2251 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2252
2253 static struct attribute *rdev_default_attrs[] = {
2254         &rdev_state.attr,
2255         &rdev_errors.attr,
2256         &rdev_slot.attr,
2257         &rdev_offset.attr,
2258         &rdev_size.attr,
2259         NULL,
2260 };
2261 static ssize_t
2262 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2263 {
2264         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2265         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2266         mddev_t *mddev = rdev->mddev;
2267         ssize_t rv;
2268
2269         if (!entry->show)
2270                 return -EIO;
2271
2272         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2273         if (!rv) {
2274                 if (rdev->mddev == NULL)
2275                         rv = -EBUSY;
2276                 else
2277                         rv = entry->show(rdev, page);
2278                 mddev_unlock(mddev);
2279         }
2280         return rv;
2281 }
2282
2283 static ssize_t
2284 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2285               const char *page, size_t length)
2286 {
2287         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2288         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2289         ssize_t rv;
2290         mddev_t *mddev = rdev->mddev;
2291
2292         if (!entry->store)
2293                 return -EIO;
2294         if (!capable(CAP_SYS_ADMIN))
2295                 return -EACCES;
2296         rv = mddev ? mddev_lock(mddev): -EBUSY;
2297         if (!rv) {
2298                 if (rdev->mddev == NULL)
2299                         rv = -EBUSY;
2300                 else
2301                         rv = entry->store(rdev, page, length);
2302                 mddev_unlock(mddev);
2303         }
2304         return rv;
2305 }
2306
2307 static void rdev_free(struct kobject *ko)
2308 {
2309         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2310         kfree(rdev);
2311 }
2312 static struct sysfs_ops rdev_sysfs_ops = {
2313         .show           = rdev_attr_show,
2314         .store          = rdev_attr_store,
2315 };
2316 static struct kobj_type rdev_ktype = {
2317         .release        = rdev_free,
2318         .sysfs_ops      = &rdev_sysfs_ops,
2319         .default_attrs  = rdev_default_attrs,
2320 };
2321
2322 /*
2323  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2324  *
2325  * mark the device faulty if:
2326  *
2327  *   - the device is nonexistent (zero size)
2328  *   - the device has no valid superblock
2329  *
2330  * a faulty rdev _never_ has rdev->sb set.
2331  */
2332 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2333 {
2334         char b[BDEVNAME_SIZE];
2335         int err;
2336         mdk_rdev_t *rdev;
2337         sector_t size;
2338
2339         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2340         if (!rdev) {
2341                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2342                 return ERR_PTR(-ENOMEM);
2343         }
2344
2345         if ((err = alloc_disk_sb(rdev)))
2346                 goto abort_free;
2347
2348         err = lock_rdev(rdev, newdev, super_format == -2);
2349         if (err)
2350                 goto abort_free;
2351
2352         kobject_init(&rdev->kobj, &rdev_ktype);
2353
2354         rdev->desc_nr = -1;
2355         rdev->saved_raid_disk = -1;
2356         rdev->raid_disk = -1;
2357         rdev->flags = 0;
2358         rdev->data_offset = 0;
2359         rdev->sb_events = 0;
2360         atomic_set(&rdev->nr_pending, 0);
2361         atomic_set(&rdev->read_errors, 0);
2362         atomic_set(&rdev->corrected_errors, 0);
2363
2364         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2365         if (!size) {
2366                 printk(KERN_WARNING
2367                         "md: %s has zero or unknown size, marking faulty!\n",
2368                         bdevname(rdev->bdev,b));
2369                 err = -EINVAL;
2370                 goto abort_free;
2371         }
2372
2373         if (super_format >= 0) {
2374                 err = super_types[super_format].
2375                         load_super(rdev, NULL, super_minor);
2376                 if (err == -EINVAL) {
2377                         printk(KERN_WARNING
2378                                 "md: %s does not have a valid v%d.%d "
2379                                "superblock, not importing!\n",
2380                                 bdevname(rdev->bdev,b),
2381                                super_format, super_minor);
2382                         goto abort_free;
2383                 }
2384                 if (err < 0) {
2385                         printk(KERN_WARNING
2386                                 "md: could not read %s's sb, not importing!\n",
2387                                 bdevname(rdev->bdev,b));
2388                         goto abort_free;
2389                 }
2390         }
2391
2392         INIT_LIST_HEAD(&rdev->same_set);
2393         init_waitqueue_head(&rdev->blocked_wait);
2394
2395         return rdev;
2396
2397 abort_free:
2398         if (rdev->sb_page) {
2399                 if (rdev->bdev)
2400                         unlock_rdev(rdev);
2401                 free_disk_sb(rdev);
2402         }
2403         kfree(rdev);
2404         return ERR_PTR(err);
2405 }
2406
2407 /*
2408  * Check a full RAID array for plausibility
2409  */
2410
2411
2412 static void analyze_sbs(mddev_t * mddev)
2413 {
2414         int i;
2415         mdk_rdev_t *rdev, *freshest, *tmp;
2416         char b[BDEVNAME_SIZE];
2417
2418         freshest = NULL;
2419         rdev_for_each(rdev, tmp, mddev)
2420                 switch (super_types[mddev->major_version].
2421                         load_super(rdev, freshest, mddev->minor_version)) {
2422                 case 1:
2423                         freshest = rdev;
2424                         break;
2425                 case 0:
2426                         break;
2427                 default:
2428                         printk( KERN_ERR \
2429                                 "md: fatal superblock inconsistency in %s"
2430                                 " -- removing from array\n",
2431                                 bdevname(rdev->bdev,b));
2432                         kick_rdev_from_array(rdev);
2433                 }
2434
2435
2436         super_types[mddev->major_version].
2437                 validate_super(mddev, freshest);
2438
2439         i = 0;
2440         rdev_for_each(rdev, tmp, mddev) {
2441                 if (rdev != freshest)
2442                         if (super_types[mddev->major_version].
2443                             validate_super(mddev, rdev)) {
2444                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2445                                         " from array!\n",
2446                                         bdevname(rdev->bdev,b));
2447                                 kick_rdev_from_array(rdev);
2448                                 continue;
2449                         }
2450                 if (mddev->level == LEVEL_MULTIPATH) {
2451                         rdev->desc_nr = i++;
2452                         rdev->raid_disk = rdev->desc_nr;
2453                         set_bit(In_sync, &rdev->flags);
2454                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2455                         rdev->raid_disk = -1;
2456                         clear_bit(In_sync, &rdev->flags);
2457                 }
2458         }
2459
2460
2461
2462         if (mddev->recovery_cp != MaxSector &&
2463             mddev->level >= 1)
2464                 printk(KERN_ERR "md: %s: raid array is not clean"
2465                        " -- starting background reconstruction\n",
2466                        mdname(mddev));
2467
2468 }
2469
2470 static void md_safemode_timeout(unsigned long data);
2471
2472 static ssize_t
2473 safe_delay_show(mddev_t *mddev, char *page)
2474 {
2475         int msec = (mddev->safemode_delay*1000)/HZ;
2476         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2477 }
2478 static ssize_t
2479 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2480 {
2481         int scale=1;
2482         int dot=0;
2483         int i;
2484         unsigned long msec;
2485         char buf[30];
2486
2487         /* remove a period, and count digits after it */
2488         if (len >= sizeof(buf))
2489                 return -EINVAL;
2490         strlcpy(buf, cbuf, sizeof(buf));
2491         for (i=0; i<len; i++) {
2492                 if (dot) {
2493                         if (isdigit(buf[i])) {
2494                                 buf[i-1] = buf[i];
2495                                 scale *= 10;
2496                         }
2497                         buf[i] = 0;
2498                 } else if (buf[i] == '.') {
2499                         dot=1;
2500                         buf[i] = 0;
2501                 }
2502         }
2503         if (strict_strtoul(buf, 10, &msec) < 0)
2504                 return -EINVAL;
2505         msec = (msec * 1000) / scale;
2506         if (msec == 0)
2507                 mddev->safemode_delay = 0;
2508         else {
2509                 unsigned long old_delay = mddev->safemode_delay;
2510                 mddev->safemode_delay = (msec*HZ)/1000;
2511                 if (mddev->safemode_delay == 0)
2512                         mddev->safemode_delay = 1;
2513                 if (mddev->safemode_delay < old_delay)
2514                         md_safemode_timeout((unsigned long)mddev);
2515         }
2516         return len;
2517 }
2518 static struct md_sysfs_entry md_safe_delay =
2519 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2520
2521 static ssize_t
2522 level_show(mddev_t *mddev, char *page)
2523 {
2524         struct mdk_personality *p = mddev->pers;
2525         if (p)
2526                 return sprintf(page, "%s\n", p->name);
2527         else if (mddev->clevel[0])
2528                 return sprintf(page, "%s\n", mddev->clevel);
2529         else if (mddev->level != LEVEL_NONE)
2530                 return sprintf(page, "%d\n", mddev->level);
2531         else
2532                 return 0;
2533 }
2534
2535 static ssize_t
2536 level_store(mddev_t *mddev, const char *buf, size_t len)
2537 {
2538         ssize_t rv = len;
2539         if (mddev->pers)
2540                 return -EBUSY;
2541         if (len == 0)
2542                 return 0;
2543         if (len >= sizeof(mddev->clevel))
2544                 return -ENOSPC;
2545         strncpy(mddev->clevel, buf, len);
2546         if (mddev->clevel[len-1] == '\n')
2547                 len--;
2548         mddev->clevel[len] = 0;
2549         mddev->level = LEVEL_NONE;
2550         return rv;
2551 }
2552
2553 static struct md_sysfs_entry md_level =
2554 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2555
2556
2557 static ssize_t
2558 layout_show(mddev_t *mddev, char *page)
2559 {
2560         /* just a number, not meaningful for all levels */
2561         if (mddev->reshape_position != MaxSector &&
2562             mddev->layout != mddev->new_layout)
2563                 return sprintf(page, "%d (%d)\n",
2564                                mddev->new_layout, mddev->layout);
2565         return sprintf(page, "%d\n", mddev->layout);
2566 }
2567
2568 static ssize_t
2569 layout_store(mddev_t *mddev, const char *buf, size_t len)
2570 {
2571         char *e;
2572         unsigned long n = simple_strtoul(buf, &e, 10);
2573
2574         if (!*buf || (*e && *e != '\n'))
2575                 return -EINVAL;
2576
2577         if (mddev->pers)
2578                 return -EBUSY;
2579         if (mddev->reshape_position != MaxSector)
2580                 mddev->new_layout = n;
2581         else
2582                 mddev->layout = n;
2583         return len;
2584 }
2585 static struct md_sysfs_entry md_layout =
2586 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2587
2588
2589 static ssize_t
2590 raid_disks_show(mddev_t *mddev, char *page)
2591 {
2592         if (mddev->raid_disks == 0)
2593                 return 0;
2594         if (mddev->reshape_position != MaxSector &&
2595             mddev->delta_disks != 0)
2596                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2597                                mddev->raid_disks - mddev->delta_disks);
2598         return sprintf(page, "%d\n", mddev->raid_disks);
2599 }
2600
2601 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2602
2603 static ssize_t
2604 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2605 {
2606         char *e;
2607         int rv = 0;
2608         unsigned long n = simple_strtoul(buf, &e, 10);
2609
2610         if (!*buf || (*e && *e != '\n'))
2611                 return -EINVAL;
2612
2613         if (mddev->pers)
2614                 rv = update_raid_disks(mddev, n);
2615         else if (mddev->reshape_position != MaxSector) {
2616                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2617                 mddev->delta_disks = n - olddisks;
2618                 mddev->raid_disks = n;
2619         } else
2620                 mddev->raid_disks = n;
2621         return rv ? rv : len;
2622 }
2623 static struct md_sysfs_entry md_raid_disks =
2624 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2625
2626 static ssize_t
2627 chunk_size_show(mddev_t *mddev, char *page)
2628 {
2629         if (mddev->reshape_position != MaxSector &&
2630             mddev->chunk_size != mddev->new_chunk)
2631                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2632                                mddev->chunk_size);
2633         return sprintf(page, "%d\n", mddev->chunk_size);
2634 }
2635
2636 static ssize_t
2637 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2638 {
2639         /* can only set chunk_size if array is not yet active */
2640         char *e;
2641         unsigned long n = simple_strtoul(buf, &e, 10);
2642
2643         if (!*buf || (*e && *e != '\n'))
2644                 return -EINVAL;
2645
2646         if (mddev->pers)
2647                 return -EBUSY;
2648         else if (mddev->reshape_position != MaxSector)
2649                 mddev->new_chunk = n;
2650         else
2651                 mddev->chunk_size = n;
2652         return len;
2653 }
2654 static struct md_sysfs_entry md_chunk_size =
2655 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2656
2657 static ssize_t
2658 resync_start_show(mddev_t *mddev, char *page)
2659 {
2660         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2661 }
2662
2663 static ssize_t
2664 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2665 {
2666         char *e;
2667         unsigned long long n = simple_strtoull(buf, &e, 10);
2668
2669         if (mddev->pers)
2670                 return -EBUSY;
2671         if (!*buf || (*e && *e != '\n'))
2672                 return -EINVAL;
2673
2674         mddev->recovery_cp = n;
2675         return len;
2676 }
2677 static struct md_sysfs_entry md_resync_start =
2678 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2679
2680 /*
2681  * The array state can be:
2682  *
2683  * clear
2684  *     No devices, no size, no level
2685  *     Equivalent to STOP_ARRAY ioctl
2686  * inactive
2687  *     May have some settings, but array is not active
2688  *        all IO results in error
2689  *     When written, doesn't tear down array, but just stops it
2690  * suspended (not supported yet)
2691  *     All IO requests will block. The array can be reconfigured.
2692  *     Writing this, if accepted, will block until array is quiescent
2693  * readonly
2694  *     no resync can happen.  no superblocks get written.
2695  *     write requests fail
2696  * read-auto
2697  *     like readonly, but behaves like 'clean' on a write request.
2698  *
2699  * clean - no pending writes, but otherwise active.
2700  *     When written to inactive array, starts without resync
2701  *     If a write request arrives then
2702  *       if metadata is known, mark 'dirty' and switch to 'active'.
2703  *       if not known, block and switch to write-pending
2704  *     If written to an active array that has pending writes, then fails.
2705  * active
2706  *     fully active: IO and resync can be happening.
2707  *     When written to inactive array, starts with resync
2708  *
2709  * write-pending
2710  *     clean, but writes are blocked waiting for 'active' to be written.
2711  *
2712  * active-idle
2713  *     like active, but no writes have been seen for a while (100msec).
2714  *
2715  */
2716 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2717                    write_pending, active_idle, bad_word};
2718 static char *array_states[] = {
2719         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2720         "write-pending", "active-idle", NULL };
2721
2722 static int match_word(const char *word, char **list)
2723 {
2724         int n;
2725         for (n=0; list[n]; n++)
2726                 if (cmd_match(word, list[n]))
2727                         break;
2728         return n;
2729 }
2730
2731 static ssize_t
2732 array_state_show(mddev_t *mddev, char *page)
2733 {
2734         enum array_state st = inactive;
2735
2736         if (mddev->pers)
2737                 switch(mddev->ro) {
2738                 case 1:
2739                         st = readonly;
2740                         break;
2741                 case 2:
2742                         st = read_auto;
2743                         break;
2744                 case 0:
2745                         if (mddev->in_sync)
2746                                 st = clean;
2747                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2748                                 st = write_pending;
2749                         else if (mddev->safemode)
2750                                 st = active_idle;
2751                         else
2752                                 st = active;
2753                 }
2754         else {
2755                 if (list_empty(&mddev->disks) &&
2756                     mddev->raid_disks == 0 &&
2757                     mddev->size == 0)
2758                         st = clear;
2759                 else
2760                         st = inactive;
2761         }
2762         return sprintf(page, "%s\n", array_states[st]);
2763 }
2764
2765 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2766 static int do_md_run(mddev_t * mddev);
2767 static int restart_array(mddev_t *mddev);
2768
2769 static ssize_t
2770 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2771 {
2772         int err = -EINVAL;
2773         enum array_state st = match_word(buf, array_states);
2774         switch(st) {
2775         case bad_word:
2776                 break;
2777         case clear:
2778                 /* stopping an active array */
2779                 if (atomic_read(&mddev->openers) > 0)
2780                         return -EBUSY;
2781                 err = do_md_stop(mddev, 0, 0);
2782                 break;
2783         case inactive:
2784                 /* stopping an active array */
2785                 if (mddev->pers) {
2786                         if (atomic_read(&mddev->openers) > 0)
2787                                 return -EBUSY;
2788                         err = do_md_stop(mddev, 2, 0);
2789                 } else
2790                         err = 0; /* already inactive */
2791                 break;
2792         case suspended:
2793                 break; /* not supported yet */
2794         case readonly:
2795                 if (mddev->pers)
2796                         err = do_md_stop(mddev, 1, 0);
2797                 else {
2798                         mddev->ro = 1;
2799                         set_disk_ro(mddev->gendisk, 1);
2800                         err = do_md_run(mddev);
2801                 }
2802                 break;
2803         case read_auto:
2804                 if (mddev->pers) {
2805                         if (mddev->ro == 0)
2806                                 err = do_md_stop(mddev, 1, 0);
2807                         else if (mddev->ro == 1)
2808                                 err = restart_array(mddev);
2809                         if (err == 0) {
2810                                 mddev->ro = 2;
2811                                 set_disk_ro(mddev->gendisk, 0);
2812                         }
2813                 } else {
2814                         mddev->ro = 2;
2815                         err = do_md_run(mddev);
2816                 }
2817                 break;
2818         case clean:
2819                 if (mddev->pers) {
2820                         restart_array(mddev);
2821                         spin_lock_irq(&mddev->write_lock);
2822                         if (atomic_read(&mddev->writes_pending) == 0) {
2823                                 if (mddev->in_sync == 0) {
2824                                         mddev->in_sync = 1;
2825                                         if (mddev->safemode == 1)
2826                                                 mddev->safemode = 0;
2827                                         if (mddev->persistent)
2828                                                 set_bit(MD_CHANGE_CLEAN,
2829                                                         &mddev->flags);
2830                                 }
2831                                 err = 0;
2832                         } else
2833                                 err = -EBUSY;
2834                         spin_unlock_irq(&mddev->write_lock);
2835                 } else {
2836                         mddev->ro = 0;
2837                         mddev->recovery_cp = MaxSector;
2838                         err = do_md_run(mddev);
2839                 }
2840                 break;
2841         case active:
2842                 if (mddev->pers) {
2843                         restart_array(mddev);
2844                         if (mddev->external)
2845                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2846                         wake_up(&mddev->sb_wait);
2847                         err = 0;
2848                 } else {
2849                         mddev->ro = 0;
2850                         set_disk_ro(mddev->gendisk, 0);
2851                         err = do_md_run(mddev);
2852                 }
2853                 break;
2854         case write_pending:
2855         case active_idle:
2856                 /* these cannot be set */
2857                 break;
2858         }
2859         if (err)
2860                 return err;
2861         else {
2862                 sysfs_notify_dirent(mddev->sysfs_state);
2863                 return len;
2864         }
2865 }
2866 static struct md_sysfs_entry md_array_state =
2867 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2868
2869 static ssize_t
2870 null_show(mddev_t *mddev, char *page)
2871 {
2872         return -EINVAL;
2873 }
2874
2875 static ssize_t
2876 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2877 {
2878         /* buf must be %d:%d\n? giving major and minor numbers */
2879         /* The new device is added to the array.
2880          * If the array has a persistent superblock, we read the
2881          * superblock to initialise info and check validity.
2882          * Otherwise, only checking done is that in bind_rdev_to_array,
2883          * which mainly checks size.
2884          */
2885         char *e;
2886         int major = simple_strtoul(buf, &e, 10);
2887         int minor;
2888         dev_t dev;
2889         mdk_rdev_t *rdev;
2890         int err;
2891
2892         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2893                 return -EINVAL;
2894         minor = simple_strtoul(e+1, &e, 10);
2895         if (*e && *e != '\n')
2896                 return -EINVAL;
2897         dev = MKDEV(major, minor);
2898         if (major != MAJOR(dev) ||
2899             minor != MINOR(dev))
2900                 return -EOVERFLOW;
2901
2902
2903         if (mddev->persistent) {
2904                 rdev = md_import_device(dev, mddev->major_version,
2905                                         mddev->minor_version);
2906                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2907                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2908                                                        mdk_rdev_t, same_set);
2909                         err = super_types[mddev->major_version]
2910                                 .load_super(rdev, rdev0, mddev->minor_version);
2911                         if (err < 0)
2912                                 goto out;
2913                 }
2914         } else if (mddev->external)
2915                 rdev = md_import_device(dev, -2, -1);
2916         else
2917                 rdev = md_import_device(dev, -1, -1);
2918
2919         if (IS_ERR(rdev))
2920                 return PTR_ERR(rdev);
2921         err = bind_rdev_to_array(rdev, mddev);
2922  out:
2923         if (err)
2924                 export_rdev(rdev);
2925         return err ? err : len;
2926 }
2927
2928 static struct md_sysfs_entry md_new_device =
2929 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2930
2931 static ssize_t
2932 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2933 {
2934         char *end;
2935         unsigned long chunk, end_chunk;
2936
2937         if (!mddev->bitmap)
2938                 goto out;
2939         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2940         while (*buf) {
2941                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2942                 if (buf == end) break;
2943                 if (*end == '-') { /* range */
2944                         buf = end + 1;
2945                         end_chunk = simple_strtoul(buf, &end, 0);
2946                         if (buf == end) break;
2947                 }
2948                 if (*end && !isspace(*end)) break;
2949                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2950                 buf = end;
2951                 while (isspace(*buf)) buf++;
2952         }
2953         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2954 out:
2955         return len;
2956 }
2957
2958 static struct md_sysfs_entry md_bitmap =
2959 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2960
2961 static ssize_t
2962 size_show(mddev_t *mddev, char *page)
2963 {
2964         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2965 }
2966
2967 static int update_size(mddev_t *mddev, sector_t num_sectors);
2968
2969 static ssize_t
2970 size_store(mddev_t *mddev, const char *buf, size_t len)
2971 {
2972         /* If array is inactive, we can reduce the component size, but
2973          * not increase it (except from 0).
2974          * If array is active, we can try an on-line resize
2975          */
2976         char *e;
2977         int err = 0;
2978         unsigned long long size = simple_strtoull(buf, &e, 10);
2979         if (!*buf || *buf == '\n' ||
2980             (*e && *e != '\n'))
2981                 return -EINVAL;
2982
2983         if (mddev->pers) {
2984                 err = update_size(mddev, size * 2);
2985                 md_update_sb(mddev, 1);
2986         } else {
2987                 if (mddev->size == 0 ||
2988                     mddev->size > size)
2989                         mddev->size = size;
2990                 else
2991                         err = -ENOSPC;
2992         }
2993         return err ? err : len;
2994 }
2995
2996 static struct md_sysfs_entry md_size =
2997 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2998
2999
3000 /* Metdata version.
3001  * This is one of
3002  *   'none' for arrays with no metadata (good luck...)
3003  *   'external' for arrays with externally managed metadata,
3004  * or N.M for internally known formats
3005  */
3006 static ssize_t
3007 metadata_show(mddev_t *mddev, char *page)
3008 {
3009         if (mddev->persistent)
3010                 return sprintf(page, "%d.%d\n",
3011                                mddev->major_version, mddev->minor_version);
3012         else if (mddev->external)
3013                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3014         else
3015                 return sprintf(page, "none\n");
3016 }
3017
3018 static ssize_t
3019 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3020 {
3021         int major, minor;
3022         char *e;
3023         /* Changing the details of 'external' metadata is
3024          * always permitted.  Otherwise there must be
3025          * no devices attached to the array.
3026          */
3027         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3028                 ;
3029         else if (!list_empty(&mddev->disks))
3030                 return -EBUSY;
3031
3032         if (cmd_match(buf, "none")) {
3033                 mddev->persistent = 0;
3034                 mddev->external = 0;
3035                 mddev->major_version = 0;
3036                 mddev->minor_version = 90;
3037                 return len;
3038         }
3039         if (strncmp(buf, "external:", 9) == 0) {
3040                 size_t namelen = len-9;
3041                 if (namelen >= sizeof(mddev->metadata_type))
3042                         namelen = sizeof(mddev->metadata_type)-1;
3043                 strncpy(mddev->metadata_type, buf+9, namelen);
3044                 mddev->metadata_type[namelen] = 0;
3045                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3046                         mddev->metadata_type[--namelen] = 0;
3047                 mddev->persistent = 0;
3048                 mddev->external = 1;
3049                 mddev->major_version = 0;
3050                 mddev->minor_version = 90;
3051                 return len;
3052         }
3053         major = simple_strtoul(buf, &e, 10);
3054         if (e==buf || *e != '.')
3055                 return -EINVAL;
3056         buf = e+1;
3057         minor = simple_strtoul(buf, &e, 10);
3058         if (e==buf || (*e && *e != '\n') )
3059                 return -EINVAL;
3060         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3061                 return -ENOENT;
3062         mddev->major_version = major;
3063         mddev->minor_version = minor;
3064         mddev->persistent = 1;
3065         mddev->external = 0;
3066         return len;
3067 }
3068
3069 static struct md_sysfs_entry md_metadata =
3070 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3071
3072 static ssize_t
3073 action_show(mddev_t *mddev, char *page)
3074 {
3075         char *type = "idle";
3076         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3077             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3078                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3079                         type = "reshape";
3080                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3081                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3082                                 type = "resync";
3083                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3084                                 type = "check";
3085                         else
3086                                 type = "repair";
3087                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3088                         type = "recover";
3089         }
3090         return sprintf(page, "%s\n", type);
3091 }
3092
3093 static ssize_t
3094 action_store(mddev_t *mddev, const char *page, size_t len)
3095 {
3096         if (!mddev->pers || !mddev->pers->sync_request)
3097                 return -EINVAL;
3098
3099         if (cmd_match(page, "idle")) {
3100                 if (mddev->sync_thread) {
3101                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3102                         md_unregister_thread(mddev->sync_thread);
3103                         mddev->sync_thread = NULL;
3104                         mddev->recovery = 0;
3105                 }
3106         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3107                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3108                 return -EBUSY;
3109         else if (cmd_match(page, "resync"))
3110                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3111         else if (cmd_match(page, "recover")) {
3112                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3113                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3114         } else if (cmd_match(page, "reshape")) {
3115                 int err;
3116                 if (mddev->pers->start_reshape == NULL)
3117                         return -EINVAL;
3118                 err = mddev->pers->start_reshape(mddev);
3119                 if (err)
3120                         return err;
3121                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3122         } else {
3123                 if (cmd_match(page, "check"))
3124                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3125                 else if (!cmd_match(page, "repair"))
3126                         return -EINVAL;
3127                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3128                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3129         }
3130         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3131         md_wakeup_thread(mddev->thread);
3132         sysfs_notify_dirent(mddev->sysfs_action);
3133         return len;
3134 }
3135
3136 static ssize_t
3137 mismatch_cnt_show(mddev_t *mddev, char *page)
3138 {
3139         return sprintf(page, "%llu\n",
3140                        (unsigned long long) mddev->resync_mismatches);
3141 }
3142
3143 static struct md_sysfs_entry md_scan_mode =
3144 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3145
3146
3147 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3148
3149 static ssize_t
3150 sync_min_show(mddev_t *mddev, char *page)
3151 {
3152         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3153                        mddev->sync_speed_min ? "local": "system");
3154 }
3155
3156 static ssize_t
3157 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3158 {
3159         int min;
3160         char *e;
3161         if (strncmp(buf, "system", 6)==0) {
3162                 mddev->sync_speed_min = 0;
3163                 return len;
3164         }
3165         min = simple_strtoul(buf, &e, 10);
3166         if (buf == e || (*e && *e != '\n') || min <= 0)
3167                 return -EINVAL;
3168         mddev->sync_speed_min = min;
3169         return len;
3170 }
3171
3172 static struct md_sysfs_entry md_sync_min =
3173 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3174
3175 static ssize_t
3176 sync_max_show(mddev_t *mddev, char *page)
3177 {
3178         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3179                        mddev->sync_speed_max ? "local": "system");
3180 }
3181
3182 static ssize_t
3183 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3184 {
3185         int max;
3186         char *e;
3187         if (strncmp(buf, "system", 6)==0) {
3188                 mddev->sync_speed_max = 0;
3189                 return len;
3190         }
3191         max = simple_strtoul(buf, &e, 10);
3192         if (buf == e || (*e && *e != '\n') || max <= 0)
3193                 return -EINVAL;
3194         mddev->sync_speed_max = max;
3195         return len;
3196 }
3197
3198 static struct md_sysfs_entry md_sync_max =
3199 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3200
3201 static ssize_t
3202 degraded_show(mddev_t *mddev, char *page)
3203 {
3204         return sprintf(page, "%d\n", mddev->degraded);
3205 }
3206 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3207
3208 static ssize_t
3209 sync_force_parallel_show(mddev_t *mddev, char *page)
3210 {
3211         return sprintf(page, "%d\n", mddev->parallel_resync);
3212 }
3213
3214 static ssize_t
3215 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3216 {
3217         long n;
3218
3219         if (strict_strtol(buf, 10, &n))
3220                 return -EINVAL;
3221
3222         if (n != 0 && n != 1)
3223                 return -EINVAL;
3224
3225         mddev->parallel_resync = n;
3226
3227         if (mddev->sync_thread)
3228                 wake_up(&resync_wait);
3229
3230         return len;
3231 }
3232
3233 /* force parallel resync, even with shared block devices */
3234 static struct md_sysfs_entry md_sync_force_parallel =
3235 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3236        sync_force_parallel_show, sync_force_parallel_store);
3237
3238 static ssize_t
3239 sync_speed_show(mddev_t *mddev, char *page)
3240 {
3241         unsigned long resync, dt, db;
3242         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3243         dt = (jiffies - mddev->resync_mark) / HZ;
3244         if (!dt) dt++;
3245         db = resync - mddev->resync_mark_cnt;
3246         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3247 }
3248
3249 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3250
3251 static ssize_t
3252 sync_completed_show(mddev_t *mddev, char *page)
3253 {
3254         unsigned long max_blocks, resync;
3255
3256         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3257                 max_blocks = mddev->resync_max_sectors;
3258         else
3259                 max_blocks = mddev->size << 1;
3260
3261         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3262         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3263 }
3264
3265 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3266
3267 static ssize_t
3268 min_sync_show(mddev_t *mddev, char *page)
3269 {
3270         return sprintf(page, "%llu\n",
3271                        (unsigned long long)mddev->resync_min);
3272 }
3273 static ssize_t
3274 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3275 {
3276         unsigned long long min;
3277         if (strict_strtoull(buf, 10, &min))
3278                 return -EINVAL;
3279         if (min > mddev->resync_max)
3280                 return -EINVAL;
3281         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3282                 return -EBUSY;
3283
3284         /* Must be a multiple of chunk_size */
3285         if (mddev->chunk_size) {
3286                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3287                         return -EINVAL;
3288         }
3289         mddev->resync_min = min;
3290
3291         return len;
3292 }
3293
3294 static struct md_sysfs_entry md_min_sync =
3295 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3296
3297 static ssize_t
3298 max_sync_show(mddev_t *mddev, char *page)
3299 {
3300         if (mddev->resync_max == MaxSector)
3301                 return sprintf(page, "max\n");
3302         else
3303                 return sprintf(page, "%llu\n",
3304                                (unsigned long long)mddev->resync_max);
3305 }
3306 static ssize_t
3307 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3308 {
3309         if (strncmp(buf, "max", 3) == 0)
3310                 mddev->resync_max = MaxSector;
3311         else {
3312                 unsigned long long max;
3313                 if (strict_strtoull(buf, 10, &max))
3314                         return -EINVAL;
3315                 if (max < mddev->resync_min)
3316                         return -EINVAL;
3317                 if (max < mddev->resync_max &&
3318                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3319                         return -EBUSY;
3320
3321                 /* Must be a multiple of chunk_size */
3322                 if (mddev->chunk_size) {
3323                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3324                                 return -EINVAL;
3325                 }
3326                 mddev->resync_max = max;
3327         }
3328         wake_up(&mddev->recovery_wait);
3329         return len;
3330 }
3331
3332 static struct md_sysfs_entry md_max_sync =
3333 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3334
3335 static ssize_t
3336 suspend_lo_show(mddev_t *mddev, char *page)
3337 {
3338         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3339 }
3340
3341 static ssize_t
3342 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3343 {
3344         char *e;
3345         unsigned long long new = simple_strtoull(buf, &e, 10);
3346
3347         if (mddev->pers->quiesce == NULL)
3348                 return -EINVAL;
3349         if (buf == e || (*e && *e != '\n'))
3350                 return -EINVAL;
3351         if (new >= mddev->suspend_hi ||
3352             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3353                 mddev->suspend_lo = new;
3354                 mddev->pers->quiesce(mddev, 2);
3355                 return len;
3356         } else
3357                 return -EINVAL;
3358 }
3359 static struct md_sysfs_entry md_suspend_lo =
3360 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3361
3362
3363 static ssize_t
3364 suspend_hi_show(mddev_t *mddev, char *page)
3365 {
3366         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3367 }
3368
3369 static ssize_t
3370 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3371 {
3372         char *e;
3373         unsigned long long new = simple_strtoull(buf, &e, 10);
3374
3375         if (mddev->pers->quiesce == NULL)
3376                 return -EINVAL;
3377         if (buf == e || (*e && *e != '\n'))
3378                 return -EINVAL;
3379         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3380             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3381                 mddev->suspend_hi = new;
3382                 mddev->pers->quiesce(mddev, 1);
3383                 mddev->pers->quiesce(mddev, 0);
3384                 return len;
3385         } else
3386                 return -EINVAL;
3387 }
3388 static struct md_sysfs_entry md_suspend_hi =
3389 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3390
3391 static ssize_t
3392 reshape_position_show(mddev_t *mddev, char *page)
3393 {
3394         if (mddev->reshape_position != MaxSector)
3395                 return sprintf(page, "%llu\n",
3396                                (unsigned long long)mddev->reshape_position);
3397         strcpy(page, "none\n");
3398         return 5;
3399 }
3400
3401 static ssize_t
3402 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3403 {
3404         char *e;
3405         unsigned long long new = simple_strtoull(buf, &e, 10);
3406         if (mddev->pers)
3407                 return -EBUSY;
3408         if (buf == e || (*e && *e != '\n'))
3409                 return -EINVAL;
3410         mddev->reshape_position = new;
3411         mddev->delta_disks = 0;
3412         mddev->new_level = mddev->level;
3413         mddev->new_layout = mddev->layout;
3414         mddev->new_chunk = mddev->chunk_size;
3415         return len;
3416 }
3417
3418 static struct md_sysfs_entry md_reshape_position =
3419 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3420        reshape_position_store);
3421
3422
3423 static struct attribute *md_default_attrs[] = {
3424         &md_level.attr,
3425         &md_layout.attr,
3426         &md_raid_disks.attr,
3427         &md_chunk_size.attr,
3428         &md_size.attr,
3429         &md_resync_start.attr,
3430         &md_metadata.attr,
3431         &md_new_device.attr,
3432         &md_safe_delay.attr,
3433         &md_array_state.attr,
3434         &md_reshape_position.attr,
3435         NULL,
3436 };
3437
3438 static struct attribute *md_redundancy_attrs[] = {
3439         &md_scan_mode.attr,
3440         &md_mismatches.attr,
3441         &md_sync_min.attr,
3442         &md_sync_max.attr,
3443         &md_sync_speed.attr,
3444         &md_sync_force_parallel.attr,
3445         &md_sync_completed.attr,
3446         &md_min_sync.attr,
3447         &md_max_sync.attr,
3448         &md_suspend_lo.attr,
3449         &md_suspend_hi.attr,
3450         &md_bitmap.attr,
3451         &md_degraded.attr,
3452         NULL,
3453 };
3454 static struct attribute_group md_redundancy_group = {
3455         .name = NULL,
3456         .attrs = md_redundancy_attrs,
3457 };
3458
3459
3460 static ssize_t
3461 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3462 {
3463         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3464         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3465         ssize_t rv;
3466
3467         if (!entry->show)
3468                 return -EIO;
3469         rv = mddev_lock(mddev);
3470         if (!rv) {
3471                 rv = entry->show(mddev, page);
3472                 mddev_unlock(mddev);
3473         }
3474         return rv;
3475 }
3476
3477 static ssize_t
3478 md_attr_store(struct kobject *kobj, struct attribute *attr,
3479               const char *page, size_t length)
3480 {
3481         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3482         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3483         ssize_t rv;
3484
3485         if (!entry->store)
3486                 return -EIO;
3487         if (!capable(CAP_SYS_ADMIN))
3488                 return -EACCES;
3489         rv = mddev_lock(mddev);
3490         if (mddev->hold_active == UNTIL_IOCTL)
3491                 mddev->hold_active = 0;
3492         if (!rv) {
3493                 rv = entry->store(mddev, page, length);
3494                 mddev_unlock(mddev);
3495         }
3496         return rv;
3497 }
3498
3499 static void md_free(struct kobject *ko)
3500 {
3501         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3502
3503         if (mddev->sysfs_state)
3504                 sysfs_put(mddev->sysfs_state);
3505
3506         if (mddev->gendisk) {
3507                 del_gendisk(mddev->gendisk);
3508                 put_disk(mddev->gendisk);
3509         }
3510         if (mddev->queue)
3511                 blk_cleanup_queue(mddev->queue);
3512
3513         kfree(mddev);
3514 }
3515
3516 static struct sysfs_ops md_sysfs_ops = {
3517         .show   = md_attr_show,
3518         .store  = md_attr_store,
3519 };
3520 static struct kobj_type md_ktype = {
3521         .release        = md_free,
3522         .sysfs_ops      = &md_sysfs_ops,
3523         .default_attrs  = md_default_attrs,
3524 };
3525
3526 int mdp_major = 0;
3527
3528 static int md_alloc(dev_t dev, char *name)
3529 {
3530         static DEFINE_MUTEX(disks_mutex);
3531         mddev_t *mddev = mddev_find(dev);
3532         struct gendisk *disk;
3533         int partitioned;
3534         int shift;
3535         int unit;
3536         int error;
3537
3538         if (!mddev)
3539                 return -ENODEV;
3540
3541         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3542         shift = partitioned ? MdpMinorShift : 0;
3543         unit = MINOR(mddev->unit) >> shift;
3544
3545         /* wait for any previous instance if this device
3546          * to be completed removed (mddev_delayed_delete).
3547          */
3548         flush_scheduled_work();
3549
3550         mutex_lock(&disks_mutex);
3551         if (mddev->gendisk) {
3552                 mutex_unlock(&disks_mutex);
3553                 mddev_put(mddev);
3554                 return -EEXIST;
3555         }
3556
3557         if (name) {
3558                 /* Need to ensure that 'name' is not a duplicate.
3559                  */
3560                 mddev_t *mddev2;
3561                 spin_lock(&all_mddevs_lock);
3562
3563                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3564                         if (mddev2->gendisk &&
3565                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
3566                                 spin_unlock(&all_mddevs_lock);
3567                                 return -EEXIST;
3568                         }
3569                 spin_unlock(&all_mddevs_lock);
3570         }
3571
3572         mddev->queue = blk_alloc_queue(GFP_KERNEL);
3573         if (!mddev->queue) {
3574                 mutex_unlock(&disks_mutex);
3575                 mddev_put(mddev);
3576                 return -ENOMEM;
3577         }
3578         /* Can be unlocked because the queue is new: no concurrency */
3579         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3580
3581         blk_queue_make_request(mddev->queue, md_fail_request);
3582
3583         disk = alloc_disk(1 << shift);
3584         if (!disk) {
3585                 mutex_unlock(&disks_mutex);
3586                 blk_cleanup_queue(mddev->queue);
3587                 mddev->queue = NULL;
3588                 mddev_put(mddev);
3589                 return -ENOMEM;
3590         }
3591         disk->major = MAJOR(mddev->unit);
3592         disk->first_minor = unit << shift;
3593         if (name)
3594                 strcpy(disk->disk_name, name);
3595         else if (partitioned)
3596                 sprintf(disk->disk_name, "md_d%d", unit);
3597         else
3598                 sprintf(disk->disk_name, "md%d", unit);
3599         disk->fops = &md_fops;
3600         disk->private_data = mddev;
3601         disk->queue = mddev->queue;
3602         /* Allow extended partitions.  This makes the
3603          * 'mdp' device redundant, but we can't really
3604          * remove it now.
3605          */
3606         disk->flags |= GENHD_FL_EXT_DEVT;
3607         add_disk(disk);
3608         mddev->gendisk = disk;
3609         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3610                                      &disk_to_dev(disk)->kobj, "%s", "md");
3611         mutex_unlock(&disks_mutex);
3612         if (error)
3613                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3614                        disk->disk_name);
3615         else {
3616                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3617                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3618         }
3619         mddev_put(mddev);
3620         return 0;
3621 }
3622
3623 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3624 {
3625         md_alloc(dev, NULL);
3626         return NULL;
3627 }
3628
3629 static int add_named_array(const char *val, struct kernel_param *kp)
3630 {
3631         /* val must be "md_*" where * is not all digits.
3632          * We allocate an array with a large free minor number, and
3633          * set the name to val.  val must not already be an active name.
3634          */
3635         int len = strlen(val);
3636         char buf[DISK_NAME_LEN];
3637
3638         while (len && val[len-1] == '\n')
3639                 len--;
3640         if (len >= DISK_NAME_LEN)
3641                 return -E2BIG;
3642         strlcpy(buf, val, len+1);
3643         if (strncmp(buf, "md_", 3) != 0)
3644                 return -EINVAL;
3645         return md_alloc(0, buf);
3646 }
3647
3648 static void md_safemode_timeout(unsigned long data)
3649 {
3650         mddev_t *mddev = (mddev_t *) data;
3651
3652         if (!atomic_read(&mddev->writes_pending)) {
3653                 mddev->safemode = 1;
3654                 if (mddev->external)
3655                         sysfs_notify_dirent(mddev->sysfs_state);
3656         }
3657         md_wakeup_thread(mddev->thread);
3658 }
3659
3660 static int start_dirty_degraded;
3661
3662 static int do_md_run(mddev_t * mddev)
3663 {
3664         int err;
3665         int chunk_size;
3666         mdk_rdev_t *rdev;
3667         struct gendisk *disk;
3668         struct mdk_personality *pers;
3669         char b[BDEVNAME_SIZE];
3670
3671         if (list_empty(&mddev->disks))
3672                 /* cannot run an array with no devices.. */
3673                 return -EINVAL;
3674
3675         if (mddev->pers)
3676                 return -EBUSY;
3677
3678         /*
3679          * Analyze all RAID superblock(s)
3680          */
3681         if (!mddev->raid_disks) {
3682                 if (!mddev->persistent)
3683                         return -EINVAL;
3684                 analyze_sbs(mddev);
3685         }
3686
3687         chunk_size = mddev->chunk_size;
3688
3689         if (chunk_size) {
3690                 if (chunk_size > MAX_CHUNK_SIZE) {
3691                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3692                                 chunk_size, MAX_CHUNK_SIZE);
3693                         return -EINVAL;
3694                 }
3695                 /*
3696                  * chunk-size has to be a power of 2
3697                  */
3698                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3699                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3700                         return -EINVAL;
3701                 }
3702
3703                 /* devices must have minimum size of one chunk */
3704                 list_for_each_entry(rdev, &mddev->disks, same_set) {
3705                         if (test_bit(Faulty, &rdev->flags))
3706                                 continue;
3707                         if (rdev->size < chunk_size / 1024) {
3708                                 printk(KERN_WARNING
3709                                         "md: Dev %s smaller than chunk_size:"
3710                                         " %lluk < %dk\n",
3711                                         bdevname(rdev->bdev,b),
3712                                         (unsigned long long)rdev->size,
3713                                         chunk_size / 1024);
3714                                 return -EINVAL;
3715                         }
3716                 }
3717         }
3718
3719         if (mddev->level != LEVEL_NONE)
3720                 request_module("md-level-%d", mddev->level);
3721         else if (mddev->clevel[0])
3722                 request_module("md-%s", mddev->clevel);
3723
3724         /*
3725          * Drop all container device buffers, from now on
3726          * the only valid external interface is through the md
3727          * device.
3728          */
3729         list_for_each_entry(rdev, &mddev->disks, same_set) {
3730                 if (test_bit(Faulty, &rdev->flags))
3731                         continue;
3732                 sync_blockdev(rdev->bdev);
3733                 invalidate_bdev(rdev->bdev);
3734
3735                 /* perform some consistency tests on the device.
3736                  * We don't want the data to overlap the metadata,
3737                  * Internal Bitmap issues has handled elsewhere.
3738                  */
3739                 if (rdev->data_offset < rdev->sb_start) {
3740                         if (mddev->size &&
3741                             rdev->data_offset + mddev->size*2
3742                             > rdev->sb_start) {
3743                                 printk("md: %s: data overlaps metadata\n",
3744                                        mdname(mddev));
3745                                 return -EINVAL;
3746                         }
3747                 } else {
3748                         if (rdev->sb_start + rdev->sb_size/512
3749                             > rdev->data_offset) {
3750                                 printk("md: %s: metadata overlaps data\n",
3751                                        mdname(mddev));
3752                                 return -EINVAL;
3753                         }
3754                 }
3755                 sysfs_notify_dirent(rdev->sysfs_state);
3756         }
3757
3758         md_probe(mddev->unit, NULL, NULL);
3759         disk = mddev->gendisk;
3760         if (!disk)
3761                 return -ENOMEM;
3762
3763         spin_lock(&pers_lock);
3764         pers = find_pers(mddev->level, mddev->clevel);
3765         if (!pers || !try_module_get(pers->owner)) {
3766                 spin_unlock(&pers_lock);
3767                 if (mddev->level != LEVEL_NONE)
3768                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3769                                mddev->level);
3770                 else
3771                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3772                                mddev->clevel);
3773                 return -EINVAL;
3774         }
3775         mddev->pers = pers;
3776         spin_unlock(&pers_lock);
3777         mddev->level = pers->level;
3778         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3779
3780         if (mddev->reshape_position != MaxSector &&
3781             pers->start_reshape == NULL) {
3782                 /* This personality cannot handle reshaping... */
3783                 mddev->pers = NULL;
3784                 module_put(pers->owner);
3785                 return -EINVAL;
3786         }
3787
3788         if (pers->sync_request) {
3789                 /* Warn if this is a potentially silly
3790                  * configuration.
3791                  */
3792                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3793                 mdk_rdev_t *rdev2;
3794                 int warned = 0;
3795
3796                 list_for_each_entry(rdev, &mddev->disks, same_set)
3797                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
3798                                 if (rdev < rdev2 &&
3799                                     rdev->bdev->bd_contains ==
3800                                     rdev2->bdev->bd_contains) {
3801                                         printk(KERN_WARNING
3802                                                "%s: WARNING: %s appears to be"
3803                                                " on the same physical disk as"
3804                                                " %s.\n",
3805                                                mdname(mddev),
3806                                                bdevname(rdev->bdev,b),
3807                                                bdevname(rdev2->bdev,b2));
3808                                         warned = 1;
3809                                 }
3810                         }
3811
3812                 if (warned)
3813                         printk(KERN_WARNING
3814                                "True protection against single-disk"
3815                                " failure might be compromised.\n");
3816         }
3817
3818         mddev->recovery = 0;
3819         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3820         mddev->barriers_work = 1;
3821         mddev->ok_start_degraded = start_dirty_degraded;
3822
3823         if (start_readonly)
3824                 mddev->ro = 2; /* read-only, but switch on first write */
3825
3826         err = mddev->pers->run(mddev);
3827         if (err)
3828                 printk(KERN_ERR "md: pers->run() failed ...\n");
3829         else if (mddev->pers->sync_request) {
3830                 err = bitmap_create(mddev);
3831                 if (err) {
3832                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3833                                mdname(mddev), err);
3834                         mddev->pers->stop(mddev);
3835                 }
3836         }
3837         if (err) {
3838                 module_put(mddev->pers->owner);
3839                 mddev->pers = NULL;
3840                 bitmap_destroy(mddev);
3841                 return err;
3842         }
3843         if (mddev->pers->sync_request) {
3844                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3845                         printk(KERN_WARNING
3846                                "md: cannot register extra attributes for %s\n",
3847                                mdname(mddev));
3848                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3849         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3850                 mddev->ro = 0;
3851
3852         atomic_set(&mddev->writes_pending,0);
3853         mddev->safemode = 0;
3854         mddev->safemode_timer.function = md_safemode_timeout;
3855         mddev->safemode_timer.data = (unsigned long) mddev;
3856         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3857         mddev->in_sync = 1;
3858
3859         list_for_each_entry(rdev, &mddev->disks, same_set)
3860                 if (rdev->raid_disk >= 0) {
3861                         char nm[20];
3862                         sprintf(nm, "rd%d", rdev->raid_disk);
3863                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3864                                 printk("md: cannot register %s for %s\n",
3865                                        nm, mdname(mddev));
3866                 }
3867
3868         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3869
3870         if (mddev->flags)
3871                 md_update_sb(mddev, 0);
3872
3873         set_capacity(disk, mddev->array_sectors);
3874
3875         /* If we call blk_queue_make_request here, it will
3876          * re-initialise max_sectors etc which may have been
3877          * refined inside -> run.  So just set the bits we need to set.
3878          * Most initialisation happended when we called
3879          * blk_queue_make_request(..., md_fail_request)
3880          * earlier.
3881          */
3882         mddev->queue->queuedata = mddev;
3883         mddev->queue->make_request_fn = mddev->pers->make_request;
3884
3885         /* If there is a partially-recovered drive we need to
3886          * start recovery here.  If we leave it to md_check_recovery,
3887          * it will remove the drives and not do the right thing
3888          */
3889         if (mddev->degraded && !mddev->sync_thread) {
3890                 int spares = 0;
3891                 list_for_each_entry(rdev, &mddev->disks, same_set)
3892                         if (rdev->raid_disk >= 0 &&
3893                             !test_bit(In_sync, &rdev->flags) &&
3894                             !test_bit(Faulty, &rdev->flags))
3895                                 /* complete an interrupted recovery */
3896                                 spares++;
3897                 if (spares && mddev->pers->sync_request) {
3898                         mddev->recovery = 0;
3899                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3900                         mddev->sync_thread = md_register_thread(md_do_sync,
3901                                                                 mddev,
3902                                                                 "%s_resync");
3903                         if (!mddev->sync_thread) {
3904                                 printk(KERN_ERR "%s: could not start resync"
3905                                        " thread...\n",
3906                                        mdname(mddev));
3907                                 /* leave the spares where they are, it shouldn't hurt */
3908                                 mddev->recovery = 0;
3909                         }
3910                 }
3911         }
3912         md_wakeup_thread(mddev->thread);
3913         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3914
3915         mddev->changed = 1;
3916         md_new_event(mddev);
3917         sysfs_notify_dirent(mddev->sysfs_state);
3918         if (mddev->sysfs_action)
3919                 sysfs_notify_dirent(mddev->sysfs_action);
3920         sysfs_notify(&mddev->kobj, NULL, "degraded");
3921         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
3922         return 0;
3923 }
3924
3925 static int restart_array(mddev_t *mddev)
3926 {
3927         struct gendisk *disk = mddev->gendisk;
3928
3929         /* Complain if it has no devices */
3930         if (list_empty(&mddev->disks))
3931                 return -ENXIO;
3932         if (!mddev->pers)
3933                 return -EINVAL;
3934         if (!mddev->ro)
3935                 return -EBUSY;
3936         mddev->safemode = 0;
3937         mddev->ro = 0;
3938         set_disk_ro(disk, 0);
3939         printk(KERN_INFO "md: %s switched to read-write mode.\n",
3940                 mdname(mddev));
3941         /* Kick recovery or resync if necessary */
3942         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3943         md_wakeup_thread(mddev->thread);
3944         md_wakeup_thread(mddev->sync_thread);
3945         sysfs_notify_dirent(mddev->sysfs_state);
3946         return 0;
3947 }
3948
3949 /* similar to deny_write_access, but accounts for our holding a reference
3950  * to the file ourselves */
3951 static int deny_bitmap_write_access(struct file * file)
3952 {
3953         struct inode *inode = file->f_mapping->host;
3954
3955         spin_lock(&inode->i_lock);
3956         if (atomic_read(&inode->i_writecount) > 1) {
3957                 spin_unlock(&inode->i_lock);
3958                 return -ETXTBSY;
3959         }
3960         atomic_set(&inode->i_writecount, -1);
3961         spin_unlock(&inode->i_lock);
3962
3963         return 0;
3964 }
3965
3966 static void restore_bitmap_write_access(struct file *file)
3967 {
3968         struct inode *inode = file->f_mapping->host;
3969
3970         spin_lock(&inode->i_lock);
3971         atomic_set(&inode->i_writecount, 1);
3972         spin_unlock(&inode->i_lock);
3973 }
3974
3975 /* mode:
3976  *   0 - completely stop and dis-assemble array
3977  *   1 - switch to readonly
3978  *   2 - stop but do not disassemble array
3979  */
3980 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3981 {
3982         int err = 0;
3983         struct gendisk *disk = mddev->gendisk;
3984
3985         if (atomic_read(&mddev->openers) > is_open) {
3986                 printk("md: %s still in use.\n",mdname(mddev));
3987                 return -EBUSY;
3988         }
3989
3990         if (mddev->pers) {
3991
3992                 if (mddev->sync_thread) {
3993                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3994                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3995                         md_unregister_thread(mddev->sync_thread);
3996                         mddev->sync_thread = NULL;
3997                 }
3998
3999                 del_timer_sync(&mddev->safemode_timer);
4000
4001                 switch(mode) {
4002                 case 1: /* readonly */
4003                         err  = -ENXIO;
4004                         if (mddev->ro==1)
4005                                 goto out;
4006                         mddev->ro = 1;
4007                         break;
4008                 case 0: /* disassemble */
4009                 case 2: /* stop */
4010                         bitmap_flush(mddev);
4011                         md_super_wait(mddev);
4012                         if (mddev->ro)
4013                                 set_disk_ro(disk, 0);
4014                         blk_queue_make_request(mddev->queue, md_fail_request);
4015                         mddev->pers->stop(mddev);
4016                         mddev->queue->merge_bvec_fn = NULL;
4017                         mddev->queue->unplug_fn = NULL;
4018                         mddev->queue->backing_dev_info.congested_fn = NULL;
4019                         if (mddev->pers->sync_request) {
4020                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4021                                 if (mddev->sysfs_action)
4022                                         sysfs_put(mddev->sysfs_action);
4023                                 mddev->sysfs_action = NULL;
4024                         }
4025                         module_put(mddev->pers->owner);
4026                         mddev->pers = NULL;
4027                         /* tell userspace to handle 'inactive' */
4028                         sysfs_notify_dirent(mddev->sysfs_state);
4029
4030                         set_capacity(disk, 0);
4031                         mddev->changed = 1;
4032
4033                         if (mddev->ro)
4034                                 mddev->ro = 0;
4035                 }
4036                 if (!mddev->in_sync || mddev->flags) {
4037                         /* mark array as shutdown cleanly */
4038                         mddev->in_sync = 1;
4039                         md_update_sb(mddev, 1);
4040                 }
4041                 if (mode == 1)
4042                         set_disk_ro(disk, 1);
4043                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4044         }
4045
4046         /*
4047          * Free resources if final stop
4048          */
4049         if (mode == 0) {
4050                 mdk_rdev_t *rdev;
4051
4052                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4053
4054                 bitmap_destroy(mddev);
4055                 if (mddev->bitmap_file) {
4056                         restore_bitmap_write_access(mddev->bitmap_file);
4057                         fput(mddev->bitmap_file);
4058                         mddev->bitmap_file = NULL;
4059                 }
4060                 mddev->bitmap_offset = 0;
4061
4062                 list_for_each_entry(rdev, &mddev->disks, same_set)
4063                         if (rdev->raid_disk >= 0) {
4064                                 char nm[20];
4065                                 sprintf(nm, "rd%d", rdev->raid_disk);
4066                                 sysfs_remove_link(&mddev->kobj, nm);
4067                         }
4068
4069                 /* make sure all md_delayed_delete calls have finished */
4070                 flush_scheduled_work();
4071
4072                 export_array(mddev);
4073
4074                 mddev->array_sectors = 0;
4075                 mddev->size = 0;
4076                 mddev->raid_disks = 0;
4077                 mddev->recovery_cp = 0;
4078                 mddev->resync_min = 0;
4079                 mddev->resync_max = MaxSector;
4080                 mddev->reshape_position = MaxSector;
4081                 mddev->external = 0;
4082                 mddev->persistent = 0;
4083                 mddev->level = LEVEL_NONE;
4084                 mddev->clevel[0] = 0;
4085                 mddev->flags = 0;
4086                 mddev->ro = 0;
4087                 mddev->metadata_type[0] = 0;
4088                 mddev->chunk_size = 0;
4089                 mddev->ctime = mddev->utime = 0;
4090                 mddev->layout = 0;
4091                 mddev->max_disks = 0;
4092                 mddev->events = 0;
4093                 mddev->delta_disks = 0;
4094                 mddev->new_level = LEVEL_NONE;
4095                 mddev->new_layout = 0;
4096                 mddev->new_chunk = 0;
4097                 mddev->curr_resync = 0;
4098                 mddev->resync_mismatches = 0;
4099                 mddev->suspend_lo = mddev->suspend_hi = 0;
4100                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4101                 mddev->recovery = 0;
4102                 mddev->in_sync = 0;
4103                 mddev->changed = 0;
4104                 mddev->degraded = 0;
4105                 mddev->barriers_work = 0;
4106                 mddev->safemode = 0;
4107                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4108                 if (mddev->hold_active == UNTIL_STOP)
4109                         mddev->hold_active = 0;
4110
4111         } else if (mddev->pers)
4112                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4113                         mdname(mddev));
4114         err = 0;
4115         md_new_event(mddev);
4116         sysfs_notify_dirent(mddev->sysfs_state);
4117 out:
4118         return err;
4119 }
4120
4121 #ifndef MODULE
4122 static void autorun_array(mddev_t *mddev)
4123 {
4124         mdk_rdev_t *rdev;
4125         int err;
4126
4127         if (list_empty(&mddev->disks))
4128                 return;
4129
4130         printk(KERN_INFO "md: running: ");
4131
4132         list_for_each_entry(rdev, &mddev->disks, same_set) {
4133                 char b[BDEVNAME_SIZE];
4134                 printk("<%s>", bdevname(rdev->bdev,b));
4135         }
4136         printk("\n");
4137
4138         err = do_md_run(mddev);
4139         if (err) {
4140                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4141                 do_md_stop(mddev, 0, 0);
4142         }
4143 }
4144
4145 /*
4146  * lets try to run arrays based on all disks that have arrived
4147  * until now. (those are in pending_raid_disks)
4148  *
4149  * the method: pick the first pending disk, collect all disks with
4150  * the same UUID, remove all from the pending list and put them into
4151  * the 'same_array' list. Then order this list based on superblock
4152  * update time (freshest comes first), kick out 'old' disks and
4153  * compare superblocks. If everything's fine then run it.
4154  *
4155  * If "unit" is allocated, then bump its reference count
4156  */
4157 static void autorun_devices(int part)
4158 {
4159         mdk_rdev_t *rdev0, *rdev, *tmp;
4160         mddev_t *mddev;
4161         char b[BDEVNAME_SIZE];
4162
4163         printk(KERN_INFO "md: autorun ...\n");
4164         while (!list_empty(&pending_raid_disks)) {
4165                 int unit;
4166                 dev_t dev;
4167                 LIST_HEAD(candidates);
4168                 rdev0 = list_entry(pending_raid_disks.next,
4169                                          mdk_rdev_t, same_set);
4170
4171                 printk(KERN_INFO "md: considering %s ...\n",
4172                         bdevname(rdev0->bdev,b));
4173                 INIT_LIST_HEAD(&candidates);
4174                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4175                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4176                                 printk(KERN_INFO "md:  adding %s ...\n",
4177                                         bdevname(rdev->bdev,b));
4178                                 list_move(&rdev->same_set, &candidates);
4179                         }
4180                 /*
4181                  * now we have a set of devices, with all of them having
4182                  * mostly sane superblocks. It's time to allocate the
4183                  * mddev.
4184                  */
4185                 if (part) {
4186                         dev = MKDEV(mdp_major,
4187                                     rdev0->preferred_minor << MdpMinorShift);
4188                         unit = MINOR(dev) >> MdpMinorShift;
4189                 } else {
4190                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4191                         unit = MINOR(dev);
4192                 }
4193                 if (rdev0->preferred_minor != unit) {
4194                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4195                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4196                         break;
4197                 }
4198
4199                 md_probe(dev, NULL, NULL);
4200                 mddev = mddev_find(dev);
4201                 if (!mddev || !mddev->gendisk) {
4202                         if (mddev)
4203                                 mddev_put(mddev);
4204                         printk(KERN_ERR
4205                                 "md: cannot allocate memory for md drive.\n");
4206                         break;
4207                 }
4208                 if (mddev_lock(mddev))
4209                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4210                                mdname(mddev));
4211                 else if (mddev->raid_disks || mddev->major_version
4212                          || !list_empty(&mddev->disks)) {
4213                         printk(KERN_WARNING
4214                                 "md: %s already running, cannot run %s\n",
4215                                 mdname(mddev), bdevname(rdev0->bdev,b));
4216                         mddev_unlock(mddev);
4217                 } else {
4218                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4219                         mddev->persistent = 1;
4220                         rdev_for_each_list(rdev, tmp, &candidates) {
4221                                 list_del_init(&rdev->same_set);
4222                                 if (bind_rdev_to_array(rdev, mddev))
4223                                         export_rdev(rdev);
4224                         }
4225                         autorun_array(mddev);
4226                         mddev_unlock(mddev);
4227                 }
4228                 /* on success, candidates will be empty, on error
4229                  * it won't...
4230                  */
4231                 rdev_for_each_list(rdev, tmp, &candidates) {
4232                         list_del_init(&rdev->same_set);
4233                         export_rdev(rdev);
4234                 }
4235                 mddev_put(mddev);
4236         }
4237         printk(KERN_INFO "md: ... autorun DONE.\n");
4238 }
4239 #endif /* !MODULE */
4240
4241 static int get_version(void __user * arg)
4242 {
4243         mdu_version_t ver;
4244
4245         ver.major = MD_MAJOR_VERSION;
4246         ver.minor = MD_MINOR_VERSION;
4247         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4248
4249         if (copy_to_user(arg, &ver, sizeof(ver)))
4250                 return -EFAULT;
4251
4252         return 0;
4253 }
4254
4255 static int get_array_info(mddev_t * mddev, void __user * arg)
4256 {
4257         mdu_array_info_t info;
4258         int nr,working,active,failed,spare;
4259         mdk_rdev_t *rdev;
4260
4261         nr=working=active=failed=spare=0;
4262         list_for_each_entry(rdev, &mddev->disks, same_set) {
4263                 nr++;
4264                 if (test_bit(Faulty, &rdev->flags))
4265                         failed++;
4266                 else {
4267                         working++;
4268                         if (test_bit(In_sync, &rdev->flags))
4269                                 active++;
4270                         else
4271                                 spare++;
4272                 }
4273         }
4274
4275         info.major_version = mddev->major_version;
4276         info.minor_version = mddev->minor_version;
4277         info.patch_version = MD_PATCHLEVEL_VERSION;
4278         info.ctime         = mddev->ctime;
4279         info.level         = mddev->level;
4280         info.size          = mddev->size;
4281         if (info.size != mddev->size) /* overflow */
4282                 info.size = -1;
4283         info.nr_disks      = nr;
4284         info.raid_disks    = mddev->raid_disks;
4285         info.md_minor      = mddev->md_minor;
4286         info.not_persistent= !mddev->persistent;
4287
4288         info.utime         = mddev->utime;
4289         info.state         = 0;
4290         if (mddev->in_sync)
4291                 info.state = (1<<MD_SB_CLEAN);
4292         if (mddev->bitmap && mddev->bitmap_offset)
4293                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4294         info.active_disks  = active;
4295         info.working_disks = working;
4296         info.failed_disks  = failed;
4297         info.spare_disks   = spare;
4298
4299         info.layout        = mddev->layout;
4300         info.chunk_size    = mddev->chunk_size;
4301
4302         if (copy_to_user(arg, &info, sizeof(info)))
4303                 return -EFAULT;
4304
4305         return 0;
4306 }
4307
4308 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4309 {
4310         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4311         char *ptr, *buf = NULL;
4312         int err = -ENOMEM;
4313
4314         if (md_allow_write(mddev))
4315                 file = kmalloc(sizeof(*file), GFP_NOIO);
4316         else
4317                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4318
4319         if (!file)
4320                 goto out;
4321
4322         /* bitmap disabled, zero the first byte and copy out */
4323         if (!mddev->bitmap || !mddev->bitmap->file) {
4324                 file->pathname[0] = '\0';
4325                 goto copy_out;
4326         }
4327
4328         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4329         if (!buf)
4330                 goto out;
4331
4332         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4333         if (IS_ERR(ptr))
4334                 goto out;
4335
4336         strcpy(file->pathname, ptr);
4337
4338 copy_out:
4339         err = 0;
4340         if (copy_to_user(arg, file, sizeof(*file)))
4341                 err = -EFAULT;
4342 out:
4343         kfree(buf);
4344         kfree(file);
4345         return err;
4346 }
4347
4348 static int get_disk_info(mddev_t * mddev, void __user * arg)
4349 {
4350         mdu_disk_info_t info;
4351         mdk_rdev_t *rdev;
4352
4353         if (copy_from_user(&info, arg, sizeof(info)))
4354                 return -EFAULT;
4355
4356         rdev = find_rdev_nr(mddev, info.number);
4357         if (rdev) {
4358                 info.major = MAJOR(rdev->bdev->bd_dev);
4359                 info.minor = MINOR(rdev->bdev->bd_dev);
4360                 info.raid_disk = rdev->raid_disk;
4361                 info.state = 0;
4362                 if (test_bit(Faulty, &rdev->flags))
4363                         info.state |= (1<<MD_DISK_FAULTY);
4364                 else if (test_bit(In_sync, &rdev->flags)) {
4365                         info.state |= (1<<MD_DISK_ACTIVE);
4366                         info.state |= (1<<MD_DISK_SYNC);
4367                 }
4368                 if (test_bit(WriteMostly, &rdev->flags))
4369                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4370         } else {
4371                 info.major = info.minor = 0;
4372                 info.raid_disk = -1;
4373                 info.state = (1<<MD_DISK_REMOVED);
4374         }
4375
4376         if (copy_to_user(arg, &info, sizeof(info)))
4377                 return -EFAULT;
4378
4379         return 0;
4380 }
4381
4382 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4383 {
4384         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4385         mdk_rdev_t *rdev;
4386         dev_t dev = MKDEV(info->major,info->minor);
4387
4388         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4389                 return -EOVERFLOW;
4390
4391         if (!mddev->raid_disks) {
4392                 int err;
4393                 /* expecting a device which has a superblock */
4394                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4395                 if (IS_ERR(rdev)) {
4396                         printk(KERN_WARNING
4397                                 "md: md_import_device returned %ld\n",
4398                                 PTR_ERR(rdev));
4399                         return PTR_ERR(rdev);
4400                 }
4401                 if (!list_empty(&mddev->disks)) {
4402                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4403                                                         mdk_rdev_t, same_set);
4404                         int err = super_types[mddev->major_version]
4405                                 .load_super(rdev, rdev0, mddev->minor_version);
4406                         if (err < 0) {
4407                                 printk(KERN_WARNING
4408                                         "md: %s has different UUID to %s\n",
4409                                         bdevname(rdev->bdev,b),
4410                                         bdevname(rdev0->bdev,b2));
4411                                 export_rdev(rdev);
4412                                 return -EINVAL;
4413                         }
4414                 }
4415                 err = bind_rdev_to_array(rdev, mddev);
4416                 if (err)
4417                         export_rdev(rdev);
4418                 return err;
4419         }
4420
4421         /*
4422          * add_new_disk can be used once the array is assembled
4423          * to add "hot spares".  They must already have a superblock
4424          * written
4425          */
4426         if (mddev->pers) {
4427                 int err;
4428                 if (!mddev->pers->hot_add_disk) {
4429                         printk(KERN_WARNING
4430                                 "%s: personality does not support diskops!\n",
4431                                mdname(mddev));
4432                         return -EINVAL;
4433                 }
4434                 if (mddev->persistent)
4435                         rdev = md_import_device(dev, mddev->major_version,
4436                                                 mddev->minor_version);
4437                 else
4438                         rdev = md_import_device(dev, -1, -1);
4439                 if (IS_ERR(rdev)) {
4440                         printk(KERN_WARNING
4441                                 "md: md_import_device returned %ld\n",
4442                                 PTR_ERR(rdev));
4443                         return PTR_ERR(rdev);
4444                 }
4445                 /* set save_raid_disk if appropriate */
4446                 if (!mddev->persistent) {
4447                         if (info->state & (1<<MD_DISK_SYNC)  &&
4448                             info->raid_disk < mddev->raid_disks)
4449                                 rdev->raid_disk = info->raid_disk;
4450                         else
4451                                 rdev->raid_disk = -1;
4452                 } else
4453                         super_types[mddev->major_version].
4454                                 validate_super(mddev, rdev);
4455                 rdev->saved_raid_disk = rdev->raid_disk;
4456
4457                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4458                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4459                         set_bit(WriteMostly, &rdev->flags);
4460
4461                 rdev->raid_disk = -1;
4462                 err = bind_rdev_to_array(rdev, mddev);
4463                 if (!err && !mddev->pers->hot_remove_disk) {
4464                         /* If there is hot_add_disk but no hot_remove_disk
4465                          * then added disks for geometry changes,
4466                          * and should be added immediately.
4467                          */
4468                         super_types[mddev->major_version].
4469                                 validate_super(mddev, rdev);
4470                         err = mddev->pers->hot_add_disk(mddev, rdev);
4471                         if (err)
4472                                 unbind_rdev_from_array(rdev);
4473                 }
4474                 if (err)
4475                         export_rdev(rdev);
4476                 else
4477                         sysfs_notify_dirent(rdev->sysfs_state);
4478
4479                 md_update_sb(mddev, 1);
4480                 if (mddev->degraded)
4481                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4482                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4483                 md_wakeup_thread(mddev->thread);
4484                 return err;
4485         }
4486
4487         /* otherwise, add_new_disk is only allowed
4488          * for major_version==0 superblocks
4489          */
4490         if (mddev->major_version != 0) {
4491                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4492                        mdname(mddev));
4493                 return -EINVAL;
4494         }
4495
4496         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4497                 int err;
4498                 rdev = md_import_device(dev, -1, 0);
4499                 if (IS_ERR(rdev)) {
4500                         printk(KERN_WARNING
4501                                 "md: error, md_import_device() returned %ld\n",
4502                                 PTR_ERR(rdev));
4503                         return PTR_ERR(rdev);
4504                 }
4505                 rdev->desc_nr = info->number;
4506                 if (info->raid_disk < mddev->raid_disks)
4507                         rdev->raid_disk = info->raid_disk;
4508                 else
4509                         rdev->raid_disk = -1;
4510
4511                 if (rdev->raid_disk < mddev->raid_disks)
4512                         if (info->state & (1<<MD_DISK_SYNC))
4513                                 set_bit(In_sync, &rdev->flags);
4514
4515                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4516                         set_bit(WriteMostly, &rdev->flags);
4517
4518                 if (!mddev->persistent) {
4519                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4520                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4521                 } else
4522                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4523                 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4524
4525                 err = bind_rdev_to_array(rdev, mddev);
4526                 if (err) {
4527                         export_rdev(rdev);
4528                         return err;
4529                 }
4530         }
4531
4532         return 0;
4533 }
4534
4535 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4536 {
4537         char b[BDEVNAME_SIZE];
4538         mdk_rdev_t *rdev;
4539
4540         rdev = find_rdev(mddev, dev);
4541         if (!rdev)
4542                 return -ENXIO;
4543
4544         if (rdev->raid_disk >= 0)
4545                 goto busy;
4546
4547         kick_rdev_from_array(rdev);
4548         md_update_sb(mddev, 1);
4549         md_new_event(mddev);
4550
4551         return 0;
4552 busy:
4553         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4554                 bdevname(rdev->bdev,b), mdname(mddev));
4555         return -EBUSY;
4556 }
4557
4558 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4559 {
4560         char b[BDEVNAME_SIZE];
4561         int err;
4562         mdk_rdev_t *rdev;
4563
4564         if (!mddev->pers)
4565                 return -ENODEV;
4566
4567         if (mddev->major_version != 0) {
4568                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4569                         " version-0 superblocks.\n",
4570                         mdname(mddev));
4571                 return -EINVAL;
4572         }
4573         if (!mddev->pers->hot_add_disk) {
4574                 printk(KERN_WARNING
4575                         "%s: personality does not support diskops!\n",
4576                         mdname(mddev));
4577                 return -EINVAL;
4578         }
4579
4580         rdev = md_import_device(dev, -1, 0);
4581         if (IS_ERR(rdev)) {
4582                 printk(KERN_WARNING
4583                         "md: error, md_import_device() returned %ld\n",
4584                         PTR_ERR(rdev));
4585                 return -EINVAL;
4586         }
4587
4588         if (mddev->persistent)
4589                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4590         else
4591                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4592
4593         rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4594
4595         if (test_bit(Faulty, &rdev->flags)) {
4596                 printk(KERN_WARNING
4597                         "md: can not hot-add faulty %s disk to %s!\n",
4598                         bdevname(rdev->bdev,b), mdname(mddev));
4599                 err = -EINVAL;
4600                 goto abort_export;
4601         }
4602         clear_bit(In_sync, &rdev->flags);
4603         rdev->desc_nr = -1;
4604         rdev->saved_raid_disk = -1;
4605         err = bind_rdev_to_array(rdev, mddev);
4606         if (err)
4607                 goto abort_export;
4608
4609         /*
4610          * The rest should better be atomic, we can have disk failures
4611          * noticed in interrupt contexts ...
4612          */
4613
4614         if (rdev->desc_nr == mddev->max_disks) {
4615                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4616                         mdname(mddev));
4617                 err = -EBUSY;
4618                 goto abort_unbind_export;
4619         }
4620
4621         rdev->raid_disk = -1;
4622
4623         md_update_sb(mddev, 1);
4624
4625         /*
4626          * Kick recovery, maybe this spare has to be added to the
4627          * array immediately.
4628          */
4629         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4630         md_wakeup_thread(mddev->thread);
4631         md_new_event(mddev);
4632         return 0;
4633
4634 abort_unbind_export:
4635         unbind_rdev_from_array(rdev);
4636
4637 abort_export:
4638         export_rdev(rdev);
4639         return err;
4640 }
4641
4642 static int set_bitmap_file(mddev_t *mddev, int fd)
4643 {
4644         int err;
4645
4646         if (mddev->pers) {
4647                 if (!mddev->pers->quiesce)
4648                         return -EBUSY;
4649                 if (mddev->recovery || mddev->sync_thread)
4650                         return -EBUSY;
4651                 /* we should be able to change the bitmap.. */
4652         }
4653
4654
4655         if (fd >= 0) {
4656                 if (mddev->bitmap)
4657                         return -EEXIST; /* cannot add when bitmap is present */
4658                 mddev->bitmap_file = fget(fd);
4659
4660                 if (mddev->bitmap_file == NULL) {
4661                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4662                                mdname(mddev));
4663                         return -EBADF;
4664                 }
4665
4666                 err = deny_bitmap_write_access(mddev->bitmap_file);
4667                 if (err) {
4668                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4669                                mdname(mddev));
4670                         fput(mddev->bitmap_file);
4671                         mddev->bitmap_file = NULL;
4672                         return err;
4673                 }
4674                 mddev->bitmap_offset = 0; /* file overrides offset */
4675         } else if (mddev->bitmap == NULL)
4676                 return -ENOENT; /* cannot remove what isn't there */
4677         err = 0;
4678         if (mddev->pers) {
4679                 mddev->pers->quiesce(mddev, 1);
4680                 if (fd >= 0)
4681                         err = bitmap_create(mddev);
4682                 if (fd < 0 || err) {
4683                         bitmap_destroy(mddev);
4684                         fd = -1; /* make sure to put the file */
4685                 }
4686                 mddev->pers->quiesce(mddev, 0);
4687         }
4688         if (fd < 0) {
4689                 if (mddev->bitmap_file) {
4690                         restore_bitmap_write_access(mddev->bitmap_file);
4691                         fput(mddev->bitmap_file);
4692                 }
4693                 mddev->bitmap_file = NULL;
4694         }
4695
4696         return err;
4697 }
4698
4699 /*
4700  * set_array_info is used two different ways
4701  * The original usage is when creating a new array.
4702  * In this usage, raid_disks is > 0 and it together with
4703  *  level, size, not_persistent,layout,chunksize determine the
4704  *  shape of the array.
4705  *  This will always create an array with a type-0.90.0 superblock.
4706  * The newer usage is when assembling an array.
4707  *  In this case raid_disks will be 0, and the major_version field is
4708  *  use to determine which style super-blocks are to be found on the devices.
4709  *  The minor and patch _version numbers are also kept incase the
4710  *  super_block handler wishes to interpret them.
4711  */
4712 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4713 {
4714
4715         if (info->raid_disks == 0) {
4716                 /* just setting version number for superblock loading */
4717                 if (info->major_version < 0 ||
4718                     info->major_version >= ARRAY_SIZE(super_types) ||
4719                     super_types[info->major_version].name == NULL) {
4720                         /* maybe try to auto-load a module? */
4721                         printk(KERN_INFO
4722                                 "md: superblock version %d not known\n",
4723                                 info->major_version);
4724                         return -EINVAL;
4725                 }
4726                 mddev->major_version = info->major_version;
4727                 mddev->minor_version = info->minor_version;
4728                 mddev->patch_version = info->patch_version;
4729                 mddev->persistent = !info->not_persistent;
4730                 return 0;
4731         }
4732         mddev->major_version = MD_MAJOR_VERSION;
4733         mddev->minor_version = MD_MINOR_VERSION;
4734         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4735         mddev->ctime         = get_seconds();
4736
4737         mddev->level         = info->level;
4738         mddev->clevel[0]     = 0;
4739         mddev->size          = info->size;
4740         mddev->raid_disks    = info->raid_disks;
4741         /* don't set md_minor, it is determined by which /dev/md* was
4742          * openned
4743          */
4744         if (info->state & (1<<MD_SB_CLEAN))
4745                 mddev->recovery_cp = MaxSector;
4746         else
4747                 mddev->recovery_cp = 0;
4748         mddev->persistent    = ! info->not_persistent;
4749         mddev->external      = 0;
4750
4751         mddev->layout        = info->layout;
4752         mddev->chunk_size    = info->chunk_size;
4753
4754         mddev->max_disks     = MD_SB_DISKS;
4755
4756         if (mddev->persistent)
4757                 mddev->flags         = 0;
4758         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4759
4760         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4761         mddev->bitmap_offset = 0;
4762
4763         mddev->reshape_position = MaxSector;
4764
4765         /*
4766          * Generate a 128 bit UUID
4767          */
4768         get_random_bytes(mddev->uuid, 16);
4769
4770         mddev->new_level = mddev->level;
4771         mddev->new_chunk = mddev->chunk_size;
4772         mddev->new_layout = mddev->layout;
4773         mddev->delta_disks = 0;
4774
4775         return 0;
4776 }
4777
4778 static int update_size(mddev_t *mddev, sector_t num_sectors)
4779 {
4780         mdk_rdev_t *rdev;
4781         int rv;
4782         int fit = (num_sectors == 0);
4783
4784         if (mddev->pers->resize == NULL)
4785                 return -EINVAL;
4786         /* The "num_sectors" is the number of sectors of each device that
4787          * is used.  This can only make sense for arrays with redundancy.
4788          * linear and raid0 always use whatever space is available. We can only
4789          * consider changing this number if no resync or reconstruction is
4790          * happening, and if the new size is acceptable. It must fit before the
4791          * sb_start or, if that is <data_offset, it must fit before the size
4792          * of each device.  If num_sectors is zero, we find the largest size
4793          * that fits.
4794
4795          */
4796         if (mddev->sync_thread)
4797                 return -EBUSY;
4798         if (mddev->bitmap)
4799                 /* Sorry, cannot grow a bitmap yet, just remove it,
4800                  * grow, and re-add.
4801                  */
4802                 return -EBUSY;
4803         list_for_each_entry(rdev, &mddev->disks, same_set) {
4804                 sector_t avail;
4805                 avail = rdev->size * 2;
4806
4807                 if (fit && (num_sectors == 0 || num_sectors > avail))
4808                         num_sectors = avail;
4809                 if (avail < num_sectors)
4810                         return -ENOSPC;
4811         }
4812         rv = mddev->pers->resize(mddev, num_sectors);
4813         if (!rv) {
4814                 struct block_device *bdev;
4815
4816                 bdev = bdget_disk(mddev->gendisk, 0);
4817                 if (bdev) {
4818                         mutex_lock(&bdev->bd_inode->i_mutex);
4819                         i_size_write(bdev->bd_inode,
4820                                      (loff_t)mddev->array_sectors << 9);
4821                         mutex_unlock(&bdev->bd_inode->i_mutex);
4822                         bdput(bdev);
4823                 }
4824         }
4825         return rv;
4826 }
4827
4828 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4829 {
4830         int rv;
4831         /* change the number of raid disks */
4832         if (mddev->pers->check_reshape == NULL)
4833                 return -EINVAL;
4834         if (raid_disks <= 0 ||
4835             raid_disks >= mddev->max_disks)
4836                 return -EINVAL;
4837         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4838                 return -EBUSY;
4839         mddev->delta_disks = raid_disks - mddev->raid_disks;
4840
4841         rv = mddev->pers->check_reshape(mddev);
4842         return rv;
4843 }
4844
4845
4846 /*
4847  * update_array_info is used to change the configuration of an
4848  * on-line array.
4849  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4850  * fields in the info are checked against the array.
4851  * Any differences that cannot be handled will cause an error.
4852  * Normally, only one change can be managed at a time.
4853  */
4854 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4855 {
4856         int rv = 0;
4857         int cnt = 0;
4858         int state = 0;
4859
4860         /* calculate expected state,ignoring low bits */
4861         if (mddev->bitmap && mddev->bitmap_offset)
4862                 state |= (1 << MD_SB_BITMAP_PRESENT);
4863
4864         if (mddev->major_version != info->major_version ||
4865             mddev->minor_version != info->minor_version ||
4866 /*          mddev->patch_version != info->patch_version || */
4867             mddev->ctime         != info->ctime         ||
4868             mddev->level         != info->level         ||
4869 /*          mddev->layout        != info->layout        || */
4870             !mddev->persistent   != info->not_persistent||
4871             mddev->chunk_size    != info->chunk_size    ||
4872             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4873             ((state^info->state) & 0xfffffe00)
4874                 )
4875                 return -EINVAL;
4876         /* Check there is only one change */
4877         if (info->size >= 0 && mddev->size != info->size) cnt++;
4878         if (mddev->raid_disks != info->raid_disks) cnt++;
4879         if (mddev->layout != info->layout) cnt++;
4880         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4881         if (cnt == 0) return 0;
4882         if (cnt > 1) return -EINVAL;
4883
4884         if (mddev->layout != info->layout) {
4885                 /* Change layout
4886                  * we don't need to do anything at the md level, the
4887                  * personality will take care of it all.
4888                  */
4889                 if (mddev->pers->reconfig == NULL)
4890                         return -EINVAL;
4891                 else
4892                         return mddev->pers->reconfig(mddev, info->layout, -1);
4893         }
4894         if (info->size >= 0 && mddev->size != info->size)
4895                 rv = update_size(mddev, (sector_t)info->size * 2);
4896
4897         if (mddev->raid_disks    != info->raid_disks)
4898                 rv = update_raid_disks(mddev, info->raid_disks);
4899
4900         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4901                 if (mddev->pers->quiesce == NULL)
4902                         return -EINVAL;
4903                 if (mddev->recovery || mddev->sync_thread)
4904                         return -EBUSY;
4905                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4906                         /* add the bitmap */
4907                         if (mddev->bitmap)
4908                                 return -EEXIST;
4909                         if (mddev->default_bitmap_offset == 0)
4910                                 return -EINVAL;
4911                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4912                         mddev->pers->quiesce(mddev, 1);
4913                         rv = bitmap_create(mddev);
4914                         if (rv)
4915                                 bitmap_destroy(mddev);
4916                         mddev->pers->quiesce(mddev, 0);
4917                 } else {
4918                         /* remove the bitmap */
4919                         if (!mddev->bitmap)
4920                                 return -ENOENT;
4921                         if (mddev->bitmap->file)
4922                                 return -EINVAL;
4923                         mddev->pers->quiesce(mddev, 1);
4924                         bitmap_destroy(mddev);
4925                         mddev->pers->quiesce(mddev, 0);
4926                         mddev->bitmap_offset = 0;
4927                 }
4928         }
4929         md_update_sb(mddev, 1);
4930         return rv;
4931 }
4932
4933 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4934 {
4935         mdk_rdev_t *rdev;
4936
4937         if (mddev->pers == NULL)
4938                 return -ENODEV;
4939
4940         rdev = find_rdev(mddev, dev);
4941         if (!rdev)
4942                 return -ENODEV;
4943
4944         md_error(mddev, rdev);
4945         return 0;
4946 }
4947
4948 /*
4949  * We have a problem here : there is no easy way to give a CHS
4950  * virtual geometry. We currently pretend that we have a 2 heads
4951  * 4 sectors (with a BIG number of cylinders...). This drives
4952  * dosfs just mad... ;-)
4953  */
4954 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4955 {
4956         mddev_t *mddev = bdev->bd_disk->private_data;
4957
4958         geo->heads = 2;
4959         geo->sectors = 4;
4960         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4961         return 0;
4962 }
4963
4964 static int md_ioctl(struct block_device *bdev, fmode_t mode,
4965                         unsigned int cmd, unsigned long arg)
4966 {
4967         int err = 0;
4968         void __user *argp = (void __user *)arg;
4969         mddev_t *mddev = NULL;
4970
4971         if (!capable(CAP_SYS_ADMIN))
4972                 return -EACCES;
4973
4974         /*
4975          * Commands dealing with the RAID driver but not any
4976          * particular array:
4977          */
4978         switch (cmd)
4979         {
4980                 case RAID_VERSION:
4981                         err = get_version(argp);
4982                         goto done;
4983
4984                 case PRINT_RAID_DEBUG:
4985                         err = 0;
4986                         md_print_devices();
4987                         goto done;
4988
4989 #ifndef MODULE
4990                 case RAID_AUTORUN:
4991                         err = 0;
4992                         autostart_arrays(arg);
4993                         goto done;
4994 #endif
4995                 default:;
4996         }
4997
4998         /*
4999          * Commands creating/starting a new array:
5000          */
5001
5002         mddev = bdev->bd_disk->private_data;
5003
5004         if (!mddev) {
5005                 BUG();
5006                 goto abort;
5007         }
5008
5009         err = mddev_lock(mddev);
5010         if (err) {
5011                 printk(KERN_INFO
5012                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5013                         err, cmd);
5014                 goto abort;
5015         }
5016
5017         switch (cmd)
5018         {
5019                 case SET_ARRAY_INFO:
5020                         {
5021                                 mdu_array_info_t info;
5022                                 if (!arg)
5023                                         memset(&info, 0, sizeof(info));
5024                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5025                                         err = -EFAULT;
5026                                         goto abort_unlock;
5027                                 }
5028                                 if (mddev->pers) {
5029                                         err = update_array_info(mddev, &info);
5030                                         if (err) {
5031                                                 printk(KERN_WARNING "md: couldn't update"
5032                                                        " array info. %d\n", err);
5033                                                 goto abort_unlock;
5034                                         }
5035                                         goto done_unlock;
5036                                 }
5037                                 if (!list_empty(&mddev->disks)) {
5038                                         printk(KERN_WARNING
5039                                                "md: array %s already has disks!\n",
5040                                                mdname(mddev));
5041                                         err = -EBUSY;
5042                                         goto abort_unlock;
5043                                 }
5044                                 if (mddev->raid_disks) {
5045                                         printk(KERN_WARNING
5046                                                "md: array %s already initialised!\n",
5047                                                mdname(mddev));
5048                                         err = -EBUSY;
5049                                         goto abort_unlock;
5050                                 }
5051                                 err = set_array_info(mddev, &info);
5052                                 if (err) {
5053                                         printk(KERN_WARNING "md: couldn't set"
5054                                                " array info. %d\n", err);
5055                                         goto abort_unlock;
5056                                 }
5057                         }
5058                         goto done_unlock;
5059
5060                 default:;
5061         }
5062
5063         /*
5064          * Commands querying/configuring an existing array:
5065          */
5066         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5067          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5068         if ((!mddev->raid_disks && !mddev->external)
5069             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5070             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5071             && cmd != GET_BITMAP_FILE) {
5072                 err = -ENODEV;
5073                 goto abort_unlock;
5074         }
5075
5076         /*
5077          * Commands even a read-only array can execute:
5078          */
5079         switch (cmd)
5080         {
5081                 case GET_ARRAY_INFO:
5082                         err = get_array_info(mddev, argp);
5083                         goto done_unlock;
5084
5085                 case GET_BITMAP_FILE:
5086                         err = get_bitmap_file(mddev, argp);
5087                         goto done_unlock;
5088
5089                 case GET_DISK_INFO:
5090                         err = get_disk_info(mddev, argp);
5091                         goto done_unlock;
5092
5093                 case RESTART_ARRAY_RW:
5094                         err = restart_array(mddev);
5095                         goto done_unlock;
5096
5097                 case STOP_ARRAY:
5098                         err = do_md_stop(mddev, 0, 1);
5099                         goto done_unlock;
5100
5101                 case STOP_ARRAY_RO:
5102                         err = do_md_stop(mddev, 1, 1);
5103                         goto done_unlock;
5104
5105         }
5106
5107         /*
5108          * The remaining ioctls are changing the state of the
5109          * superblock, so we do not allow them on read-only arrays.
5110          * However non-MD ioctls (e.g. get-size) will still come through
5111          * here and hit the 'default' below, so only disallow
5112          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5113          */
5114         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5115                 if (mddev->ro == 2) {
5116                         mddev->ro = 0;
5117                         sysfs_notify_dirent(mddev->sysfs_state);
5118                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5119                         md_wakeup_thread(mddev->thread);
5120                 } else {
5121                         err = -EROFS;
5122                         goto abort_unlock;
5123                 }
5124         }
5125
5126         switch (cmd)
5127         {
5128                 case ADD_NEW_DISK:
5129                 {
5130                         mdu_disk_info_t info;
5131                         if (copy_from_user(&info, argp, sizeof(info)))
5132                                 err = -EFAULT;
5133                         else
5134                                 err = add_new_disk(mddev, &info);
5135                         goto done_unlock;
5136                 }
5137
5138                 case HOT_REMOVE_DISK:
5139                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5140                         goto done_unlock;
5141
5142                 case HOT_ADD_DISK:
5143                         err = hot_add_disk(mddev, new_decode_dev(arg));
5144                         goto done_unlock;
5145
5146                 case SET_DISK_FAULTY:
5147                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5148                         goto done_unlock;
5149
5150                 case RUN_ARRAY:
5151                         err = do_md_run(mddev);
5152                         goto done_unlock;
5153
5154                 case SET_BITMAP_FILE:
5155                         err = set_bitmap_file(mddev, (int)arg);
5156                         goto done_unlock;
5157
5158                 default:
5159                         err = -EINVAL;
5160                         goto abort_unlock;
5161         }
5162
5163 done_unlock:
5164 abort_unlock:
5165         if (mddev->hold_active == UNTIL_IOCTL &&
5166             err != -EINVAL)
5167                 mddev->hold_active = 0;
5168         mddev_unlock(mddev);
5169
5170         return err;
5171 done:
5172         if (err)
5173                 MD_BUG();
5174 abort:
5175         return err;
5176 }
5177
5178 static int md_open(struct block_device *bdev, fmode_t mode)
5179 {
5180         /*
5181          * Succeed if we can lock the mddev, which confirms that
5182          * it isn't being stopped right now.
5183          */
5184         mddev_t *mddev = mddev_find(bdev->bd_dev);
5185         int err;
5186
5187         if (mddev->gendisk != bdev->bd_disk) {
5188                 /* we are racing with mddev_put which is discarding this
5189                  * bd_disk.
5190                  */
5191                 mddev_put(mddev);
5192                 /* Wait until bdev->bd_disk is definitely gone */
5193                 flush_scheduled_work();
5194                 /* Then retry the open from the top */
5195                 return -ERESTARTSYS;
5196         }
5197         BUG_ON(mddev != bdev->bd_disk->private_data);
5198
5199         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5200                 goto out;
5201
5202         err = 0;
5203         atomic_inc(&mddev->openers);
5204         mddev_unlock(mddev);
5205
5206         check_disk_change(bdev);
5207  out:
5208         return err;
5209 }
5210
5211 static int md_release(struct gendisk *disk, fmode_t mode)
5212 {
5213         mddev_t *mddev = disk->private_data;
5214
5215         BUG_ON(!mddev);
5216         atomic_dec(&mddev->openers);
5217         mddev_put(mddev);
5218
5219         return 0;
5220 }
5221
5222 static int md_media_changed(struct gendisk *disk)
5223 {
5224         mddev_t *mddev = disk->private_data;
5225
5226         return mddev->changed;
5227 }
5228
5229 static int md_revalidate(struct gendisk *disk)
5230 {
5231         mddev_t *mddev = disk->private_data;
5232
5233         mddev->changed = 0;
5234         return 0;
5235 }
5236 static struct block_device_operations md_fops =
5237 {
5238         .owner          = THIS_MODULE,
5239         .open           = md_open,
5240         .release        = md_release,
5241         .locked_ioctl   = md_ioctl,
5242         .getgeo         = md_getgeo,
5243         .media_changed  = md_media_changed,
5244         .revalidate_disk= md_revalidate,
5245 };
5246
5247 static int md_thread(void * arg)
5248 {
5249         mdk_thread_t *thread = arg;
5250
5251         /*
5252          * md_thread is a 'system-thread', it's priority should be very
5253          * high. We avoid resource deadlocks individually in each
5254          * raid personality. (RAID5 does preallocation) We also use RR and
5255          * the very same RT priority as kswapd, thus we will never get
5256          * into a priority inversion deadlock.
5257          *
5258          * we definitely have to have equal or higher priority than
5259          * bdflush, otherwise bdflush will deadlock if there are too
5260          * many dirty RAID5 blocks.
5261          */
5262
5263         allow_signal(SIGKILL);
5264         while (!kthread_should_stop()) {
5265
5266                 /* We need to wait INTERRUPTIBLE so that
5267                  * we don't add to the load-average.
5268                  * That means we need to be sure no signals are
5269                  * pending
5270                  */
5271                 if (signal_pending(current))
5272                         flush_signals(current);
5273
5274                 wait_event_interruptible_timeout
5275                         (thread->wqueue,
5276                          test_bit(THREAD_WAKEUP, &thread->flags)
5277                          || kthread_should_stop(),
5278                          thread->timeout);
5279
5280                 clear_bit(THREAD_WAKEUP, &thread->flags);
5281
5282                 thread->run(thread->mddev);
5283         }
5284
5285         return 0;
5286 }
5287
5288 void md_wakeup_thread(mdk_thread_t *thread)
5289 {
5290         if (thread) {
5291                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5292                 set_bit(THREAD_WAKEUP, &thread->flags);
5293                 wake_up(&thread->wqueue);
5294         }
5295 }
5296
5297 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5298                                  const char *name)
5299 {
5300         mdk_thread_t *thread;
5301
5302         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5303         if (!thread)
5304                 return NULL;
5305
5306         init_waitqueue_head(&thread->wqueue);
5307
5308         thread->run = run;
5309         thread->mddev = mddev;
5310         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5311         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5312         if (IS_ERR(thread->tsk)) {
5313                 kfree(thread);
5314                 return NULL;
5315         }
5316         return thread;
5317 }
5318
5319 void md_unregister_thread(mdk_thread_t *thread)
5320 {
5321         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5322
5323         kthread_stop(thread->tsk);
5324         kfree(thread);
5325 }
5326
5327 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5328 {
5329         if (!mddev) {
5330                 MD_BUG();
5331                 return;
5332         }
5333
5334         if (!rdev || test_bit(Faulty, &rdev->flags))
5335                 return;
5336
5337         if (mddev->external)
5338                 set_bit(Blocked, &rdev->flags);
5339 /*
5340         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5341                 mdname(mddev),
5342                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5343                 __builtin_return_address(0),__builtin_return_address(1),
5344                 __builtin_return_address(2),__builtin_return_address(3));
5345 */
5346         if (!mddev->pers)
5347                 return;
5348         if (!mddev->pers->error_handler)
5349                 return;
5350         mddev->pers->error_handler(mddev,rdev);
5351         if (mddev->degraded)
5352                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5353         set_bit(StateChanged, &rdev->flags);
5354         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5355         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5356         md_wakeup_thread(mddev->thread);
5357         md_new_event_inintr(mddev);
5358 }
5359
5360 /* seq_file implementation /proc/mdstat */
5361
5362 static void status_unused(struct seq_file *seq)
5363 {
5364         int i = 0;
5365         mdk_rdev_t *rdev;
5366
5367         seq_printf(seq, "unused devices: ");
5368
5369         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5370                 char b[BDEVNAME_SIZE];
5371                 i++;
5372                 seq_printf(seq, "%s ",
5373                               bdevname(rdev->bdev,b));
5374         }
5375         if (!i)
5376                 seq_printf(seq, "<none>");
5377
5378         seq_printf(seq, "\n");
5379 }
5380
5381
5382 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5383 {
5384         sector_t max_blocks, resync, res;
5385         unsigned long dt, db, rt;
5386         int scale;
5387         unsigned int per_milli;
5388
5389         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5390
5391         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5392                 max_blocks = mddev->resync_max_sectors >> 1;
5393         else
5394                 max_blocks = mddev->size;
5395
5396         /*
5397          * Should not happen.
5398          */
5399         if (!max_blocks) {
5400                 MD_BUG();
5401                 return;
5402         }
5403         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5404          * in a sector_t, and (max_blocks>>scale) will fit in a
5405          * u32, as those are the requirements for sector_div.
5406          * Thus 'scale' must be at least 10
5407          */
5408         scale = 10;
5409         if (sizeof(sector_t) > sizeof(unsigned long)) {
5410                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5411                         scale++;
5412         }
5413         res = (resync>>scale)*1000;
5414         sector_div(res, (u32)((max_blocks>>scale)+1));
5415
5416         per_milli = res;
5417         {
5418                 int i, x = per_milli/50, y = 20-x;
5419                 seq_printf(seq, "[");
5420                 for (i = 0; i < x; i++)
5421                         seq_printf(seq, "=");
5422                 seq_printf(seq, ">");
5423                 for (i = 0; i < y; i++)
5424                         seq_printf(seq, ".");
5425                 seq_printf(seq, "] ");
5426         }
5427         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5428                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5429                     "reshape" :
5430                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5431                      "check" :
5432                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5433                       "resync" : "recovery"))),
5434                    per_milli/10, per_milli % 10,
5435                    (unsigned long long) resync,
5436                    (unsigned long long) max_blocks);
5437
5438         /*
5439          * We do not want to overflow, so the order of operands and
5440          * the * 100 / 100 trick are important. We do a +1 to be
5441          * safe against division by zero. We only estimate anyway.
5442          *
5443          * dt: time from mark until now
5444          * db: blocks written from mark until now
5445          * rt: remaining time
5446          */
5447         dt = ((jiffies - mddev->resync_mark) / HZ);
5448         if (!dt) dt++;
5449         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5450                 - mddev->resync_mark_cnt;
5451         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5452
5453         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5454
5455         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5456 }
5457
5458 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5459 {
5460         struct list_head *tmp;
5461         loff_t l = *pos;
5462         mddev_t *mddev;
5463
5464         if (l >= 0x10000)
5465                 return NULL;
5466         if (!l--)
5467                 /* header */
5468                 return (void*)1;
5469
5470         spin_lock(&all_mddevs_lock);
5471         list_for_each(tmp,&all_mddevs)
5472                 if (!l--) {
5473                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5474                         mddev_get(mddev);
5475                         spin_unlock(&all_mddevs_lock);
5476                         return mddev;
5477                 }
5478         spin_unlock(&all_mddevs_lock);
5479         if (!l--)
5480                 return (void*)2;/* tail */
5481         return NULL;
5482 }
5483
5484 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5485 {
5486         struct list_head *tmp;
5487         mddev_t *next_mddev, *mddev = v;
5488
5489         ++*pos;
5490         if (v == (void*)2)
5491                 return NULL;
5492
5493         spin_lock(&all_mddevs_lock);
5494         if (v == (void*)1)
5495                 tmp = all_mddevs.next;
5496         else
5497                 tmp = mddev->all_mddevs.next;
5498         if (tmp != &all_mddevs)
5499                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5500         else {
5501                 next_mddev = (void*)2;
5502                 *pos = 0x10000;
5503         }
5504         spin_unlock(&all_mddevs_lock);
5505
5506         if (v != (void*)1)
5507                 mddev_put(mddev);
5508         return next_mddev;
5509
5510 }
5511
5512 static void md_seq_stop(struct seq_file *seq, void *v)
5513 {
5514         mddev_t *mddev = v;
5515
5516         if (mddev && v != (void*)1 && v != (void*)2)
5517                 mddev_put(mddev);
5518 }
5519
5520 struct mdstat_info {
5521         int event;
5522 };
5523
5524 static int md_seq_show(struct seq_file *seq, void *v)
5525 {
5526         mddev_t *mddev = v;
5527         sector_t size;
5528         mdk_rdev_t *rdev;
5529         struct mdstat_info *mi = seq->private;
5530         struct bitmap *bitmap;
5531
5532         if (v == (void*)1) {
5533                 struct mdk_personality *pers;
5534                 seq_printf(seq, "Personalities : ");
5535                 spin_lock(&pers_lock);
5536                 list_for_each_entry(pers, &pers_list, list)
5537                         seq_printf(seq, "[%s] ", pers->name);
5538
5539                 spin_unlock(&pers_lock);
5540                 seq_printf(seq, "\n");
5541                 mi->event = atomic_read(&md_event_count);
5542                 return 0;
5543         }
5544         if (v == (void*)2) {
5545                 status_unused(seq);
5546                 return 0;
5547         }
5548
5549         if (mddev_lock(mddev) < 0)
5550                 return -EINTR;
5551
5552         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5553                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5554                                                 mddev->pers ? "" : "in");
5555                 if (mddev->pers) {
5556                         if (mddev->ro==1)
5557                                 seq_printf(seq, " (read-only)");
5558                         if (mddev->ro==2)
5559                                 seq_printf(seq, " (auto-read-only)");
5560                         seq_printf(seq, " %s", mddev->pers->name);
5561                 }
5562
5563                 size = 0;
5564                 list_for_each_entry(rdev, &mddev->disks, same_set) {
5565                         char b[BDEVNAME_SIZE];
5566                         seq_printf(seq, " %s[%d]",
5567                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5568                         if (test_bit(WriteMostly, &rdev->flags))
5569                                 seq_printf(seq, "(W)");
5570                         if (test_bit(Faulty, &rdev->flags)) {
5571                                 seq_printf(seq, "(F)");
5572                                 continue;
5573                         } else if (rdev->raid_disk < 0)
5574                                 seq_printf(seq, "(S)"); /* spare */
5575                         size += rdev->size;
5576                 }
5577
5578                 if (!list_empty(&mddev->disks)) {
5579                         if (mddev->pers)
5580                                 seq_printf(seq, "\n      %llu blocks",
5581                                            (unsigned long long)
5582                                            mddev->array_sectors / 2);
5583                         else
5584                                 seq_printf(seq, "\n      %llu blocks",
5585                                            (unsigned long long)size);
5586                 }
5587                 if (mddev->persistent) {
5588                         if (mddev->major_version != 0 ||
5589                             mddev->minor_version != 90) {
5590                                 seq_printf(seq," super %d.%d",
5591                                            mddev->major_version,
5592                                            mddev->minor_version);
5593                         }
5594                 } else if (mddev->external)
5595                         seq_printf(seq, " super external:%s",
5596                                    mddev->metadata_type);
5597                 else
5598                         seq_printf(seq, " super non-persistent");
5599
5600                 if (mddev->pers) {
5601                         mddev->pers->status(seq, mddev);
5602                         seq_printf(seq, "\n      ");
5603                         if (mddev->pers->sync_request) {
5604                                 if (mddev->curr_resync > 2) {
5605                                         status_resync(seq, mddev);
5606                                         seq_printf(seq, "\n      ");
5607                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5608                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5609                                 else if (mddev->recovery_cp < MaxSector)
5610                                         seq_printf(seq, "\tresync=PENDING\n      ");
5611                         }
5612                 } else
5613                         seq_printf(seq, "\n       ");
5614
5615                 if ((bitmap = mddev->bitmap)) {
5616                         unsigned long chunk_kb;
5617                         unsigned long flags;
5618                         spin_lock_irqsave(&bitmap->lock, flags);
5619                         chunk_kb = bitmap->chunksize >> 10;
5620                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5621                                 "%lu%s chunk",
5622                                 bitmap->pages - bitmap->missing_pages,
5623                                 bitmap->pages,
5624                                 (bitmap->pages - bitmap->missing_pages)
5625                                         << (PAGE_SHIFT - 10),
5626                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5627                                 chunk_kb ? "KB" : "B");
5628                         if (bitmap->file) {
5629                                 seq_printf(seq, ", file: ");
5630                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5631                         }
5632
5633                         seq_printf(seq, "\n");
5634                         spin_unlock_irqrestore(&bitmap->lock, flags);
5635                 }
5636
5637                 seq_printf(seq, "\n");
5638         }
5639         mddev_unlock(mddev);
5640
5641         return 0;
5642 }
5643
5644 static struct seq_operations md_seq_ops = {
5645         .start  = md_seq_start,
5646         .next   = md_seq_next,
5647         .stop   = md_seq_stop,
5648         .show   = md_seq_show,
5649 };
5650
5651 static int md_seq_open(struct inode *inode, struct file *file)
5652 {
5653         int error;
5654         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5655         if (mi == NULL)
5656                 return -ENOMEM;
5657
5658         error = seq_open(file, &md_seq_ops);
5659         if (error)
5660                 kfree(mi);
5661         else {
5662                 struct seq_file *p = file->private_data;
5663                 p->private = mi;
5664                 mi->event = atomic_read(&md_event_count);
5665         }
5666         return error;
5667 }
5668
5669 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5670 {
5671         struct seq_file *m = filp->private_data;
5672         struct mdstat_info *mi = m->private;
5673         int mask;
5674
5675         poll_wait(filp, &md_event_waiters, wait);
5676
5677         /* always allow read */
5678         mask = POLLIN | POLLRDNORM;
5679
5680         if (mi->event != atomic_read(&md_event_count))
5681                 mask |= POLLERR | POLLPRI;
5682         return mask;
5683 }
5684
5685 static const struct file_operations md_seq_fops = {
5686         .owner          = THIS_MODULE,
5687         .open           = md_seq_open,
5688         .read           = seq_read,
5689         .llseek         = seq_lseek,
5690         .release        = seq_release_private,
5691         .poll           = mdstat_poll,
5692 };
5693
5694 int register_md_personality(struct mdk_personality *p)
5695 {
5696         spin_lock(&pers_lock);
5697         list_add_tail(&p->list, &pers_list);
5698         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5699         spin_unlock(&pers_lock);
5700         return 0;
5701 }
5702
5703 int unregister_md_personality(struct mdk_personality *p)
5704 {
5705         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5706         spin_lock(&pers_lock);
5707         list_del_init(&p->list);
5708         spin_unlock(&pers_lock);
5709         return 0;
5710 }
5711
5712 static int is_mddev_idle(mddev_t *mddev)
5713 {
5714         mdk_rdev_t * rdev;
5715         int idle;
5716         long curr_events;
5717
5718         idle = 1;
5719         rcu_read_lock();
5720         rdev_for_each_rcu(rdev, mddev) {
5721                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5722                 curr_events = part_stat_read(&disk->part0, sectors[0]) +
5723                                 part_stat_read(&disk->part0, sectors[1]) -
5724                                 atomic_read(&disk->sync_io);
5725                 /* sync IO will cause sync_io to increase before the disk_stats
5726                  * as sync_io is counted when a request starts, and
5727                  * disk_stats is counted when it completes.
5728                  * So resync activity will cause curr_events to be smaller than
5729                  * when there was no such activity.
5730                  * non-sync IO will cause disk_stat to increase without
5731                  * increasing sync_io so curr_events will (eventually)
5732                  * be larger than it was before.  Once it becomes
5733                  * substantially larger, the test below will cause
5734                  * the array to appear non-idle, and resync will slow
5735                  * down.
5736                  * If there is a lot of outstanding resync activity when
5737                  * we set last_event to curr_events, then all that activity
5738                  * completing might cause the array to appear non-idle
5739                  * and resync will be slowed down even though there might
5740                  * not have been non-resync activity.  This will only
5741                  * happen once though.  'last_events' will soon reflect
5742                  * the state where there is little or no outstanding
5743                  * resync requests, and further resync activity will
5744                  * always make curr_events less than last_events.
5745                  *
5746                  */
5747                 if (curr_events - rdev->last_events > 4096) {
5748                         rdev->last_events = curr_events;
5749                         idle = 0;
5750                 }
5751         }
5752         rcu_read_unlock();
5753         return idle;
5754 }
5755
5756 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5757 {
5758         /* another "blocks" (512byte) blocks have been synced */
5759         atomic_sub(blocks, &mddev->recovery_active);
5760         wake_up(&mddev->recovery_wait);
5761         if (!ok) {
5762                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5763                 md_wakeup_thread(mddev->thread);
5764                 // stop recovery, signal do_sync ....
5765         }
5766 }
5767
5768
5769 /* md_write_start(mddev, bi)
5770  * If we need to update some array metadata (e.g. 'active' flag
5771  * in superblock) before writing, schedule a superblock update
5772  * and wait for it to complete.
5773  */
5774 void md_write_start(mddev_t *mddev, struct bio *bi)
5775 {
5776         int did_change = 0;
5777         if (bio_data_dir(bi) != WRITE)
5778                 return;
5779
5780         BUG_ON(mddev->ro == 1);
5781         if (mddev->ro == 2) {
5782                 /* need to switch to read/write */
5783                 mddev->ro = 0;
5784                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5785                 md_wakeup_thread(mddev->thread);
5786                 md_wakeup_thread(mddev->sync_thread);
5787                 did_change = 1;
5788         }
5789         atomic_inc(&mddev->writes_pending);
5790         if (mddev->safemode == 1)
5791                 mddev->safemode = 0;
5792         if (mddev->in_sync) {
5793                 spin_lock_irq(&mddev->write_lock);
5794                 if (mddev->in_sync) {
5795                         mddev->in_sync = 0;
5796                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5797                         md_wakeup_thread(mddev->thread);
5798                         did_change = 1;
5799                 }
5800                 spin_unlock_irq(&mddev->write_lock);
5801         }
5802         if (did_change)
5803                 sysfs_notify_dirent(mddev->sysfs_state);
5804         wait_event(mddev->sb_wait,
5805                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5806                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5807 }
5808
5809 void md_write_end(mddev_t *mddev)
5810 {
5811         if (atomic_dec_and_test(&mddev->writes_pending)) {
5812                 if (mddev->safemode == 2)
5813                         md_wakeup_thread(mddev->thread);
5814                 else if (mddev->safemode_delay)
5815                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5816         }
5817 }
5818
5819 /* md_allow_write(mddev)
5820  * Calling this ensures that the array is marked 'active' so that writes
5821  * may proceed without blocking.  It is important to call this before
5822  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5823  * Must be called with mddev_lock held.
5824  *
5825  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5826  * is dropped, so return -EAGAIN after notifying userspace.
5827  */
5828 int md_allow_write(mddev_t *mddev)
5829 {
5830         if (!mddev->pers)
5831                 return 0;
5832         if (mddev->ro)
5833                 return 0;
5834         if (!mddev->pers->sync_request)
5835                 return 0;
5836
5837         spin_lock_irq(&mddev->write_lock);
5838         if (mddev->in_sync) {
5839                 mddev->in_sync = 0;
5840                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5841                 if (mddev->safemode_delay &&
5842                     mddev->safemode == 0)
5843                         mddev->safemode = 1;
5844                 spin_unlock_irq(&mddev->write_lock);
5845                 md_update_sb(mddev, 0);
5846                 sysfs_notify_dirent(mddev->sysfs_state);
5847         } else
5848                 spin_unlock_irq(&mddev->write_lock);
5849
5850         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5851                 return -EAGAIN;
5852         else
5853                 return 0;
5854 }
5855 EXPORT_SYMBOL_GPL(md_allow_write);
5856
5857 #define SYNC_MARKS      10
5858 #define SYNC_MARK_STEP  (3*HZ)
5859 void md_do_sync(mddev_t *mddev)
5860 {
5861         mddev_t *mddev2;
5862         unsigned int currspeed = 0,
5863                  window;
5864         sector_t max_sectors,j, io_sectors;
5865         unsigned long mark[SYNC_MARKS];
5866         sector_t mark_cnt[SYNC_MARKS];
5867         int last_mark,m;
5868         struct list_head *tmp;
5869         sector_t last_check;
5870         int skipped = 0;
5871         mdk_rdev_t *rdev;
5872         char *desc;
5873
5874         /* just incase thread restarts... */
5875         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5876                 return;
5877         if (mddev->ro) /* never try to sync a read-only array */
5878                 return;
5879
5880         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5881                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5882                         desc = "data-check";
5883                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5884                         desc = "requested-resync";
5885                 else
5886                         desc = "resync";
5887         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5888                 desc = "reshape";
5889         else
5890                 desc = "recovery";
5891
5892         /* we overload curr_resync somewhat here.
5893          * 0 == not engaged in resync at all
5894          * 2 == checking that there is no conflict with another sync
5895          * 1 == like 2, but have yielded to allow conflicting resync to
5896          *              commense
5897          * other == active in resync - this many blocks
5898          *
5899          * Before starting a resync we must have set curr_resync to
5900          * 2, and then checked that every "conflicting" array has curr_resync
5901          * less than ours.  When we find one that is the same or higher
5902          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5903          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5904          * This will mean we have to start checking from the beginning again.
5905          *
5906          */
5907
5908         do {
5909                 mddev->curr_resync = 2;
5910
5911         try_again:
5912                 if (kthread_should_stop()) {
5913                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5914                         goto skip;
5915                 }
5916                 for_each_mddev(mddev2, tmp) {
5917                         if (mddev2 == mddev)
5918                                 continue;
5919                         if (!mddev->parallel_resync
5920                         &&  mddev2->curr_resync
5921                         &&  match_mddev_units(mddev, mddev2)) {
5922                                 DEFINE_WAIT(wq);
5923                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5924                                         /* arbitrarily yield */
5925                                         mddev->curr_resync = 1;
5926                                         wake_up(&resync_wait);
5927                                 }
5928                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5929                                         /* no need to wait here, we can wait the next
5930                                          * time 'round when curr_resync == 2
5931                                          */
5932                                         continue;
5933                                 /* We need to wait 'interruptible' so as not to
5934                                  * contribute to the load average, and not to
5935                                  * be caught by 'softlockup'
5936                                  */
5937                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
5938                                 if (!kthread_should_stop() &&
5939                                     mddev2->curr_resync >= mddev->curr_resync) {
5940                                         printk(KERN_INFO "md: delaying %s of %s"
5941                                                " until %s has finished (they"
5942                                                " share one or more physical units)\n",
5943                                                desc, mdname(mddev), mdname(mddev2));
5944                                         mddev_put(mddev2);
5945                                         if (signal_pending(current))
5946                                                 flush_signals(current);
5947                                         schedule();
5948                                         finish_wait(&resync_wait, &wq);
5949                                         goto try_again;
5950                                 }
5951                                 finish_wait(&resync_wait, &wq);
5952                         }
5953                 }
5954         } while (mddev->curr_resync < 2);
5955
5956         j = 0;
5957         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5958                 /* resync follows the size requested by the personality,
5959                  * which defaults to physical size, but can be virtual size
5960                  */
5961                 max_sectors = mddev->resync_max_sectors;
5962                 mddev->resync_mismatches = 0;
5963                 /* we don't use the checkpoint if there's a bitmap */
5964                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5965                         j = mddev->resync_min;
5966                 else if (!mddev->bitmap)
5967                         j = mddev->recovery_cp;
5968
5969         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5970                 max_sectors = mddev->size << 1;
5971         else {
5972                 /* recovery follows the physical size of devices */
5973                 max_sectors = mddev->size << 1;
5974                 j = MaxSector;
5975                 list_for_each_entry(rdev, &mddev->disks, same_set)
5976                         if (rdev->raid_disk >= 0 &&
5977                             !test_bit(Faulty, &rdev->flags) &&
5978                             !test_bit(In_sync, &rdev->flags) &&
5979                             rdev->recovery_offset < j)
5980                                 j = rdev->recovery_offset;
5981         }
5982
5983         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5984         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5985                 " %d KB/sec/disk.\n", speed_min(mddev));
5986         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5987                "(but not more than %d KB/sec) for %s.\n",
5988                speed_max(mddev), desc);
5989
5990         is_mddev_idle(mddev); /* this also initializes IO event counters */
5991
5992         io_sectors = 0;
5993         for (m = 0; m < SYNC_MARKS; m++) {
5994                 mark[m] = jiffies;
5995                 mark_cnt[m] = io_sectors;
5996         }
5997         last_mark = 0;
5998         mddev->resync_mark = mark[last_mark];
5999         mddev->resync_mark_cnt = mark_cnt[last_mark];
6000
6001         /*
6002          * Tune reconstruction:
6003          */
6004         window = 32*(PAGE_SIZE/512);
6005         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6006                 window/2,(unsigned long long) max_sectors/2);
6007
6008         atomic_set(&mddev->recovery_active, 0);
6009         last_check = 0;
6010
6011         if (j>2) {
6012                 printk(KERN_INFO
6013                        "md: resuming %s of %s from checkpoint.\n",
6014                        desc, mdname(mddev));
6015                 mddev->curr_resync = j;
6016         }
6017
6018         while (j < max_sectors) {
6019                 sector_t sectors;
6020
6021                 skipped = 0;
6022                 if (j >= mddev->resync_max) {
6023                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6024                         wait_event(mddev->recovery_wait,
6025                                    mddev->resync_max > j
6026                                    || kthread_should_stop());
6027                 }
6028                 if (kthread_should_stop())
6029                         goto interrupted;
6030                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6031                                                   currspeed < speed_min(mddev));
6032                 if (sectors == 0) {
6033                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6034                         goto out;
6035                 }
6036
6037                 if (!skipped) { /* actual IO requested */
6038                         io_sectors += sectors;
6039                         atomic_add(sectors, &mddev->recovery_active);
6040                 }
6041
6042                 j += sectors;
6043                 if (j>1) mddev->curr_resync = j;
6044                 mddev->curr_mark_cnt = io_sectors;
6045                 if (last_check == 0)
6046                         /* this is the earliers that rebuilt will be
6047                          * visible in /proc/mdstat
6048                          */
6049                         md_new_event(mddev);
6050
6051                 if (last_check + window > io_sectors || j == max_sectors)
6052                         continue;
6053
6054                 last_check = io_sectors;
6055
6056                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6057                         break;
6058
6059         repeat:
6060                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6061                         /* step marks */
6062                         int next = (last_mark+1) % SYNC_MARKS;
6063
6064                         mddev->resync_mark = mark[next];
6065                         mddev->resync_mark_cnt = mark_cnt[next];
6066                         mark[next] = jiffies;
6067                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6068                         last_mark = next;
6069                 }
6070
6071
6072                 if (kthread_should_stop())
6073                         goto interrupted;
6074
6075
6076                 /*
6077                  * this loop exits only if either when we are slower than
6078                  * the 'hard' speed limit, or the system was IO-idle for
6079                  * a jiffy.
6080                  * the system might be non-idle CPU-wise, but we only care
6081                  * about not overloading the IO subsystem. (things like an
6082                  * e2fsck being done on the RAID array should execute fast)
6083                  */
6084                 blk_unplug(mddev->queue);
6085                 cond_resched();
6086
6087                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6088                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6089
6090                 if (currspeed > speed_min(mddev)) {
6091                         if ((currspeed > speed_max(mddev)) ||
6092                                         !is_mddev_idle(mddev)) {
6093                                 msleep(500);
6094                                 goto repeat;
6095                         }
6096                 }
6097         }
6098         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6099         /*
6100          * this also signals 'finished resyncing' to md_stop
6101          */
6102  out:
6103         blk_unplug(mddev->queue);
6104
6105         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6106
6107         /* tell personality that we are finished */
6108         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6109
6110         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6111             mddev->curr_resync > 2) {
6112                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6113                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6114                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6115                                         printk(KERN_INFO
6116                                                "md: checkpointing %s of %s.\n",
6117                                                desc, mdname(mddev));
6118                                         mddev->recovery_cp = mddev->curr_resync;
6119                                 }
6120                         } else
6121                                 mddev->recovery_cp = MaxSector;
6122                 } else {
6123                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6124                                 mddev->curr_resync = MaxSector;
6125                         list_for_each_entry(rdev, &mddev->disks, same_set)
6126                                 if (rdev->raid_disk >= 0 &&
6127                                     !test_bit(Faulty, &rdev->flags) &&
6128                                     !test_bit(In_sync, &rdev->flags) &&
6129                                     rdev->recovery_offset < mddev->curr_resync)
6130                                         rdev->recovery_offset = mddev->curr_resync;
6131                 }
6132         }
6133         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6134
6135  skip:
6136         mddev->curr_resync = 0;
6137         mddev->resync_min = 0;
6138         mddev->resync_max = MaxSector;
6139         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6140         wake_up(&resync_wait);
6141         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6142         md_wakeup_thread(mddev->thread);
6143         return;
6144
6145  interrupted:
6146         /*
6147          * got a signal, exit.
6148          */
6149         printk(KERN_INFO
6150                "md: md_do_sync() got signal ... exiting\n");
6151         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6152         goto out;
6153
6154 }
6155 EXPORT_SYMBOL_GPL(md_do_sync);
6156
6157
6158 static int remove_and_add_spares(mddev_t *mddev)
6159 {
6160         mdk_rdev_t *rdev;
6161         int spares = 0;
6162
6163         list_for_each_entry(rdev, &mddev->disks, same_set)
6164                 if (rdev->raid_disk >= 0 &&
6165                     !test_bit(Blocked, &rdev->flags) &&
6166                     (test_bit(Faulty, &rdev->flags) ||
6167                      ! test_bit(In_sync, &rdev->flags)) &&
6168                     atomic_read(&rdev->nr_pending)==0) {
6169                         if (mddev->pers->hot_remove_disk(
6170                                     mddev, rdev->raid_disk)==0) {
6171                                 char nm[20];
6172                                 sprintf(nm,"rd%d", rdev->raid_disk);
6173                                 sysfs_remove_link(&mddev->kobj, nm);
6174                                 rdev->raid_disk = -1;
6175                         }
6176                 }
6177
6178         if (mddev->degraded && ! mddev->ro) {
6179                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6180                         if (rdev->raid_disk >= 0 &&
6181                             !test_bit(In_sync, &rdev->flags) &&
6182                             !test_bit(Blocked, &rdev->flags))
6183                                 spares++;
6184                         if (rdev->raid_disk < 0
6185                             && !test_bit(Faulty, &rdev->flags)) {
6186                                 rdev->recovery_offset = 0;
6187                                 if (mddev->pers->
6188                                     hot_add_disk(mddev, rdev) == 0) {
6189                                         char nm[20];
6190                                         sprintf(nm, "rd%d", rdev->raid_disk);
6191                                         if (sysfs_create_link(&mddev->kobj,
6192                                                               &rdev->kobj, nm))
6193                                                 printk(KERN_WARNING
6194                                                        "md: cannot register "
6195                                                        "%s for %s\n",
6196                                                        nm, mdname(mddev));
6197                                         spares++;
6198                                         md_new_event(mddev);
6199                                 } else
6200                                         break;
6201                         }
6202                 }
6203         }
6204         return spares;
6205 }
6206 /*
6207  * This routine is regularly called by all per-raid-array threads to
6208  * deal with generic issues like resync and super-block update.
6209  * Raid personalities that don't have a thread (linear/raid0) do not
6210  * need this as they never do any recovery or update the superblock.
6211  *
6212  * It does not do any resync itself, but rather "forks" off other threads
6213  * to do that as needed.
6214  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6215  * "->recovery" and create a thread at ->sync_thread.
6216  * When the thread finishes it sets MD_RECOVERY_DONE
6217  * and wakeups up this thread which will reap the thread and finish up.
6218  * This thread also removes any faulty devices (with nr_pending == 0).
6219  *
6220  * The overall approach is:
6221  *  1/ if the superblock needs updating, update it.
6222  *  2/ If a recovery thread is running, don't do anything else.
6223  *  3/ If recovery has finished, clean up, possibly marking spares active.
6224  *  4/ If there are any faulty devices, remove them.
6225  *  5/ If array is degraded, try to add spares devices
6226  *  6/ If array has spares or is not in-sync, start a resync thread.
6227  */
6228 void md_check_recovery(mddev_t *mddev)
6229 {
6230         mdk_rdev_t *rdev;
6231
6232
6233         if (mddev->bitmap)
6234                 bitmap_daemon_work(mddev->bitmap);
6235
6236         if (mddev->ro)
6237                 return;
6238
6239         if (signal_pending(current)) {
6240                 if (mddev->pers->sync_request && !mddev->external) {
6241                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6242                                mdname(mddev));
6243                         mddev->safemode = 2;
6244                 }
6245                 flush_signals(current);
6246         }
6247
6248         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6249                 return;
6250         if ( ! (
6251                 (mddev->flags && !mddev->external) ||
6252                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6253                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6254                 (mddev->external == 0 && mddev->safemode == 1) ||
6255                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6256                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6257                 ))
6258                 return;
6259
6260         if (mddev_trylock(mddev)) {
6261                 int spares = 0;
6262
6263                 if (mddev->ro) {
6264                         /* Only thing we do on a ro array is remove
6265                          * failed devices.
6266                          */
6267                         remove_and_add_spares(mddev);
6268                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6269                         goto unlock;
6270                 }
6271
6272                 if (!mddev->external) {
6273                         int did_change = 0;
6274                         spin_lock_irq(&mddev->write_lock);
6275                         if (mddev->safemode &&
6276                             !atomic_read(&mddev->writes_pending) &&
6277                             !mddev->in_sync &&
6278                             mddev->recovery_cp == MaxSector) {
6279                                 mddev->in_sync = 1;
6280                                 did_change = 1;
6281                                 if (mddev->persistent)
6282                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6283                         }
6284                         if (mddev->safemode == 1)
6285                                 mddev->safemode = 0;
6286                         spin_unlock_irq(&mddev->write_lock);
6287                         if (did_change)
6288                                 sysfs_notify_dirent(mddev->sysfs_state);
6289                 }
6290
6291                 if (mddev->flags)
6292                         md_update_sb(mddev, 0);
6293
6294                 list_for_each_entry(rdev, &mddev->disks, same_set)
6295                         if (test_and_clear_bit(StateChanged, &rdev->flags))
6296                                 sysfs_notify_dirent(rdev->sysfs_state);
6297
6298
6299                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6300                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6301                         /* resync/recovery still happening */
6302                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6303                         goto unlock;
6304                 }
6305                 if (mddev->sync_thread) {
6306                         /* resync has finished, collect result */
6307                         md_unregister_thread(mddev->sync_thread);
6308                         mddev->sync_thread = NULL;
6309                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6310                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6311                                 /* success...*/
6312                                 /* activate any spares */
6313                                 if (mddev->pers->spare_active(mddev))
6314                                         sysfs_notify(&mddev->kobj, NULL,
6315                                                      "degraded");
6316                         }
6317                         md_update_sb(mddev, 1);
6318
6319                         /* if array is no-longer degraded, then any saved_raid_disk
6320                          * information must be scrapped
6321                          */
6322                         if (!mddev->degraded)
6323                                 list_for_each_entry(rdev, &mddev->disks, same_set)
6324                                         rdev->saved_raid_disk = -1;
6325
6326                         mddev->recovery = 0;
6327                         /* flag recovery needed just to double check */
6328                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6329                         sysfs_notify_dirent(mddev->sysfs_action);
6330                         md_new_event(mddev);
6331                         goto unlock;
6332                 }
6333                 /* Set RUNNING before clearing NEEDED to avoid
6334                  * any transients in the value of "sync_action".
6335                  */
6336                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6337                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6338                 /* Clear some bits that don't mean anything, but
6339                  * might be left set
6340                  */
6341                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6342                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6343
6344                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6345                         goto unlock;
6346                 /* no recovery is running.
6347                  * remove any failed drives, then
6348                  * add spares if possible.
6349                  * Spare are also removed and re-added, to allow
6350                  * the personality to fail the re-add.
6351                  */
6352
6353                 if (mddev->reshape_position != MaxSector) {
6354                         if (mddev->pers->check_reshape(mddev) != 0)
6355                                 /* Cannot proceed */
6356                                 goto unlock;
6357                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6358                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6359                 } else if ((spares = remove_and_add_spares(mddev))) {
6360                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6361                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6362                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6363                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6364                 } else if (mddev->recovery_cp < MaxSector) {
6365                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6366                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6367                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6368                         /* nothing to be done ... */
6369                         goto unlock;
6370
6371                 if (mddev->pers->sync_request) {
6372                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6373                                 /* We are adding a device or devices to an array
6374                                  * which has the bitmap stored on all devices.
6375                                  * So make sure all bitmap pages get written
6376                                  */
6377                                 bitmap_write_all(mddev->bitmap);
6378                         }
6379                         mddev->sync_thread = md_register_thread(md_do_sync,
6380                                                                 mddev,
6381                                                                 "%s_resync");
6382                         if (!mddev->sync_thread) {
6383                                 printk(KERN_ERR "%s: could not start resync"
6384                                         " thread...\n",
6385                                         mdname(mddev));
6386                                 /* leave the spares where they are, it shouldn't hurt */
6387                                 mddev->recovery = 0;
6388                         } else
6389                                 md_wakeup_thread(mddev->sync_thread);
6390                         sysfs_notify_dirent(mddev->sysfs_action);
6391                         md_new_event(mddev);
6392                 }
6393         unlock:
6394                 if (!mddev->sync_thread) {
6395                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6396                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6397                                                &mddev->recovery))
6398                                 if (mddev->sysfs_action)
6399                                         sysfs_notify_dirent(mddev->sysfs_action);
6400                 }
6401                 mddev_unlock(mddev);
6402         }
6403 }
6404
6405 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6406 {
6407         sysfs_notify_dirent(rdev->sysfs_state);
6408         wait_event_timeout(rdev->blocked_wait,
6409                            !test_bit(Blocked, &rdev->flags),
6410                            msecs_to_jiffies(5000));
6411         rdev_dec_pending(rdev, mddev);
6412 }
6413 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6414
6415 static int md_notify_reboot(struct notifier_block *this,
6416                             unsigned long code, void *x)
6417 {
6418         struct list_head *tmp;
6419         mddev_t *mddev;
6420
6421         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6422
6423                 printk(KERN_INFO "md: stopping all md devices.\n");
6424
6425                 for_each_mddev(mddev, tmp)
6426                         if (mddev_trylock(mddev)) {
6427                                 /* Force a switch to readonly even array
6428                                  * appears to still be in use.  Hence
6429                                  * the '100'.
6430                                  */
6431                                 do_md_stop(mddev, 1, 100);
6432                                 mddev_unlock(mddev);
6433                         }
6434                 /*
6435                  * certain more exotic SCSI devices are known to be
6436                  * volatile wrt too early system reboots. While the
6437                  * right place to handle this issue is the given
6438                  * driver, we do want to have a safe RAID driver ...
6439                  */
6440                 mdelay(1000*1);
6441         }
6442         return NOTIFY_DONE;
6443 }
6444
6445 static struct notifier_block md_notifier = {
6446         .notifier_call  = md_notify_reboot,
6447         .next           = NULL,
6448         .priority       = INT_MAX, /* before any real devices */
6449 };
6450
6451 static void md_geninit(void)
6452 {
6453         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6454
6455         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6456 }
6457
6458 static int __init md_init(void)
6459 {
6460         if (register_blkdev(MAJOR_NR, "md"))
6461                 return -1;
6462         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6463                 unregister_blkdev(MAJOR_NR, "md");
6464                 return -1;
6465         }
6466         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6467                             md_probe, NULL, NULL);
6468         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6469                             md_probe, NULL, NULL);
6470
6471         register_reboot_notifier(&md_notifier);
6472         raid_table_header = register_sysctl_table(raid_root_table);
6473
6474         md_geninit();
6475         return 0;
6476 }
6477
6478
6479 #ifndef MODULE
6480
6481 /*
6482  * Searches all registered partitions for autorun RAID arrays
6483  * at boot time.
6484  */
6485
6486 static LIST_HEAD(all_detected_devices);
6487 struct detected_devices_node {
6488         struct list_head list;
6489         dev_t dev;
6490 };
6491
6492 void md_autodetect_dev(dev_t dev)
6493 {
6494         struct detected_devices_node *node_detected_dev;
6495
6496         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6497         if (node_detected_dev) {
6498                 node_detected_dev->dev = dev;
6499                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6500         } else {
6501                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6502                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6503         }
6504 }
6505
6506
6507 static void autostart_arrays(int part)
6508 {
6509         mdk_rdev_t *rdev;
6510         struct detected_devices_node *node_detected_dev;
6511         dev_t dev;
6512         int i_scanned, i_passed;
6513
6514         i_scanned = 0;
6515         i_passed = 0;
6516
6517         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6518
6519         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6520                 i_scanned++;
6521                 node_detected_dev = list_entry(all_detected_devices.next,
6522                                         struct detected_devices_node, list);
6523                 list_del(&node_detected_dev->list);
6524                 dev = node_detected_dev->dev;
6525                 kfree(node_detected_dev);
6526                 rdev = md_import_device(dev,0, 90);
6527                 if (IS_ERR(rdev))
6528                         continue;
6529
6530                 if (test_bit(Faulty, &rdev->flags)) {
6531                         MD_BUG();
6532                         continue;
6533                 }
6534                 set_bit(AutoDetected, &rdev->flags);
6535                 list_add(&rdev->same_set, &pending_raid_disks);
6536                 i_passed++;
6537         }
6538
6539         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6540                                                 i_scanned, i_passed);
6541
6542         autorun_devices(part);
6543 }
6544
6545 #endif /* !MODULE */
6546
6547 static __exit void md_exit(void)
6548 {
6549         mddev_t *mddev;
6550         struct list_head *tmp;
6551
6552         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6553         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6554
6555         unregister_blkdev(MAJOR_NR,"md");
6556         unregister_blkdev(mdp_major, "mdp");
6557         unregister_reboot_notifier(&md_notifier);
6558         unregister_sysctl_table(raid_table_header);
6559         remove_proc_entry("mdstat", NULL);
6560         for_each_mddev(mddev, tmp) {
6561                 export_array(mddev);
6562                 mddev->hold_active = 0;
6563         }
6564 }
6565
6566 subsys_initcall(md_init);
6567 module_exit(md_exit)
6568
6569 static int get_ro(char *buffer, struct kernel_param *kp)
6570 {
6571         return sprintf(buffer, "%d", start_readonly);
6572 }
6573 static int set_ro(const char *val, struct kernel_param *kp)
6574 {
6575         char *e;
6576         int num = simple_strtoul(val, &e, 10);
6577         if (*val && (*e == '\0' || *e == '\n')) {
6578                 start_readonly = num;
6579                 return 0;
6580         }
6581         return -EINVAL;
6582 }
6583
6584 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6585 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6586
6587 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6588
6589 EXPORT_SYMBOL(register_md_personality);
6590 EXPORT_SYMBOL(unregister_md_personality);
6591 EXPORT_SYMBOL(md_error);
6592 EXPORT_SYMBOL(md_done_sync);
6593 EXPORT_SYMBOL(md_write_start);
6594 EXPORT_SYMBOL(md_write_end);
6595 EXPORT_SYMBOL(md_register_thread);
6596 EXPORT_SYMBOL(md_unregister_thread);
6597 EXPORT_SYMBOL(md_wakeup_thread);
6598 EXPORT_SYMBOL(md_check_recovery);
6599 MODULE_LICENSE("GPL");
6600 MODULE_ALIAS("md");
6601 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);