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_SYNCIO) | (1<<BIO_RW_UNPLUG);
 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_SYNCIO) | (1 << BIO_RW_UNPLUG);
 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         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1485                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1486                        mdname(mddev), mddev->max_disks);
1487                 return -EBUSY;
1488         }
1489         bdevname(rdev->bdev,b);
1490         while ( (s=strchr(b, '/')) != NULL)
1491                 *s = '!';
1492
1493         rdev->mddev = mddev;
1494         printk(KERN_INFO "md: bind<%s>\n", b);
1495
1496         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1497                 goto fail;
1498
1499         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1500         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1501                 kobject_del(&rdev->kobj);
1502                 goto fail;
1503         }
1504         rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1505
1506         list_add_rcu(&rdev->same_set, &mddev->disks);
1507         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1508
1509         /* May as well allow recovery to be retried once */
1510         mddev->recovery_disabled = 0;
1511         return 0;
1512
1513  fail:
1514         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1515                b, mdname(mddev));
1516         return err;
1517 }
1518
1519 static void md_delayed_delete(struct work_struct *ws)
1520 {
1521         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1522         kobject_del(&rdev->kobj);
1523         kobject_put(&rdev->kobj);
1524 }
1525
1526 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1527 {
1528         char b[BDEVNAME_SIZE];
1529         if (!rdev->mddev) {
1530                 MD_BUG();
1531                 return;
1532         }
1533         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1534         list_del_rcu(&rdev->same_set);
1535         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1536         rdev->mddev = NULL;
1537         sysfs_remove_link(&rdev->kobj, "block");
1538         sysfs_put(rdev->sysfs_state);
1539         rdev->sysfs_state = NULL;
1540         /* We need to delay this, otherwise we can deadlock when
1541          * writing to 'remove' to "dev/state".  We also need
1542          * to delay it due to rcu usage.
1543          */
1544         synchronize_rcu();
1545         INIT_WORK(&rdev->del_work, md_delayed_delete);
1546         kobject_get(&rdev->kobj);
1547         schedule_work(&rdev->del_work);
1548 }
1549
1550 /*
1551  * prevent the device from being mounted, repartitioned or
1552  * otherwise reused by a RAID array (or any other kernel
1553  * subsystem), by bd_claiming the device.
1554  */
1555 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1556 {
1557         int err = 0;
1558         struct block_device *bdev;
1559         char b[BDEVNAME_SIZE];
1560
1561         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1562         if (IS_ERR(bdev)) {
1563                 printk(KERN_ERR "md: could not open %s.\n",
1564                         __bdevname(dev, b));
1565                 return PTR_ERR(bdev);
1566         }
1567         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1568         if (err) {
1569                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1570                         bdevname(bdev, b));
1571                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1572                 return err;
1573         }
1574         if (!shared)
1575                 set_bit(AllReserved, &rdev->flags);
1576         rdev->bdev = bdev;
1577         return err;
1578 }
1579
1580 static void unlock_rdev(mdk_rdev_t *rdev)
1581 {
1582         struct block_device *bdev = rdev->bdev;
1583         rdev->bdev = NULL;
1584         if (!bdev)
1585                 MD_BUG();
1586         bd_release(bdev);
1587         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1588 }
1589
1590 void md_autodetect_dev(dev_t dev);
1591
1592 static void export_rdev(mdk_rdev_t * rdev)
1593 {
1594         char b[BDEVNAME_SIZE];
1595         printk(KERN_INFO "md: export_rdev(%s)\n",
1596                 bdevname(rdev->bdev,b));
1597         if (rdev->mddev)
1598                 MD_BUG();
1599         free_disk_sb(rdev);
1600 #ifndef MODULE
1601         if (test_bit(AutoDetected, &rdev->flags))
1602                 md_autodetect_dev(rdev->bdev->bd_dev);
1603 #endif
1604         unlock_rdev(rdev);
1605         kobject_put(&rdev->kobj);
1606 }
1607
1608 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1609 {
1610         unbind_rdev_from_array(rdev);
1611         export_rdev(rdev);
1612 }
1613
1614 static void export_array(mddev_t *mddev)
1615 {
1616         mdk_rdev_t *rdev, *tmp;
1617
1618         rdev_for_each(rdev, tmp, mddev) {
1619                 if (!rdev->mddev) {
1620                         MD_BUG();
1621                         continue;
1622                 }
1623                 kick_rdev_from_array(rdev);
1624         }
1625         if (!list_empty(&mddev->disks))
1626                 MD_BUG();
1627         mddev->raid_disks = 0;
1628         mddev->major_version = 0;
1629 }
1630
1631 static void print_desc(mdp_disk_t *desc)
1632 {
1633         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1634                 desc->major,desc->minor,desc->raid_disk,desc->state);
1635 }
1636
1637 static void print_sb_90(mdp_super_t *sb)
1638 {
1639         int i;
1640
1641         printk(KERN_INFO
1642                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1643                 sb->major_version, sb->minor_version, sb->patch_version,
1644                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1645                 sb->ctime);
1646         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1647                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1648                 sb->md_minor, sb->layout, sb->chunk_size);
1649         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1650                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1651                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1652                 sb->failed_disks, sb->spare_disks,
1653                 sb->sb_csum, (unsigned long)sb->events_lo);
1654
1655         printk(KERN_INFO);
1656         for (i = 0; i < MD_SB_DISKS; i++) {
1657                 mdp_disk_t *desc;
1658
1659                 desc = sb->disks + i;
1660                 if (desc->number || desc->major || desc->minor ||
1661                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1662                         printk("     D %2d: ", i);
1663                         print_desc(desc);
1664                 }
1665         }
1666         printk(KERN_INFO "md:     THIS: ");
1667         print_desc(&sb->this_disk);
1668 }
1669
1670 static void print_sb_1(struct mdp_superblock_1 *sb)
1671 {
1672         __u8 *uuid;
1673
1674         uuid = sb->set_uuid;
1675         printk(KERN_INFO "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1676                         ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1677                KERN_INFO "md:    Name: \"%s\" CT:%llu\n",
1678                 le32_to_cpu(sb->major_version),
1679                 le32_to_cpu(sb->feature_map),
1680                 uuid[0], uuid[1], uuid[2], uuid[3],
1681                 uuid[4], uuid[5], uuid[6], uuid[7],
1682                 uuid[8], uuid[9], uuid[10], uuid[11],
1683                 uuid[12], uuid[13], uuid[14], uuid[15],
1684                 sb->set_name,
1685                 (unsigned long long)le64_to_cpu(sb->ctime)
1686                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1687
1688         uuid = sb->device_uuid;
1689         printk(KERN_INFO "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1690                         " RO:%llu\n"
1691                KERN_INFO "md:     Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1692                         ":%02x%02x%02x%02x%02x%02x\n"
1693                KERN_INFO "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1694                KERN_INFO "md:         (MaxDev:%u) \n",
1695                 le32_to_cpu(sb->level),
1696                 (unsigned long long)le64_to_cpu(sb->size),
1697                 le32_to_cpu(sb->raid_disks),
1698                 le32_to_cpu(sb->layout),
1699                 le32_to_cpu(sb->chunksize),
1700                 (unsigned long long)le64_to_cpu(sb->data_offset),
1701                 (unsigned long long)le64_to_cpu(sb->data_size),
1702                 (unsigned long long)le64_to_cpu(sb->super_offset),
1703                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1704                 le32_to_cpu(sb->dev_number),
1705                 uuid[0], uuid[1], uuid[2], uuid[3],
1706                 uuid[4], uuid[5], uuid[6], uuid[7],
1707                 uuid[8], uuid[9], uuid[10], uuid[11],
1708                 uuid[12], uuid[13], uuid[14], uuid[15],
1709                 sb->devflags,
1710                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1711                 (unsigned long long)le64_to_cpu(sb->events),
1712                 (unsigned long long)le64_to_cpu(sb->resync_offset),
1713                 le32_to_cpu(sb->sb_csum),
1714                 le32_to_cpu(sb->max_dev)
1715                 );
1716 }
1717
1718 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1719 {
1720         char b[BDEVNAME_SIZE];
1721         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1722                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1723                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1724                 rdev->desc_nr);
1725         if (rdev->sb_loaded) {
1726                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1727                 switch (major_version) {
1728                 case 0:
1729                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1730                         break;
1731                 case 1:
1732                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1733                         break;
1734                 }
1735         } else
1736                 printk(KERN_INFO "md: no rdev superblock!\n");
1737 }
1738
1739 static void md_print_devices(void)
1740 {
1741         struct list_head *tmp;
1742         mdk_rdev_t *rdev;
1743         mddev_t *mddev;
1744         char b[BDEVNAME_SIZE];
1745
1746         printk("\n");
1747         printk("md:     **********************************\n");
1748         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1749         printk("md:     **********************************\n");
1750         for_each_mddev(mddev, tmp) {
1751
1752                 if (mddev->bitmap)
1753                         bitmap_print_sb(mddev->bitmap);
1754                 else
1755                         printk("%s: ", mdname(mddev));
1756                 list_for_each_entry(rdev, &mddev->disks, same_set)
1757                         printk("<%s>", bdevname(rdev->bdev,b));
1758                 printk("\n");
1759
1760                 list_for_each_entry(rdev, &mddev->disks, same_set)
1761                         print_rdev(rdev, mddev->major_version);
1762         }
1763         printk("md:     **********************************\n");
1764         printk("\n");
1765 }
1766
1767
1768 static void sync_sbs(mddev_t * mddev, int nospares)
1769 {
1770         /* Update each superblock (in-memory image), but
1771          * if we are allowed to, skip spares which already
1772          * have the right event counter, or have one earlier
1773          * (which would mean they aren't being marked as dirty
1774          * with the rest of the array)
1775          */
1776         mdk_rdev_t *rdev;
1777
1778         list_for_each_entry(rdev, &mddev->disks, same_set) {
1779                 if (rdev->sb_events == mddev->events ||
1780                     (nospares &&
1781                      rdev->raid_disk < 0 &&
1782                      (rdev->sb_events&1)==0 &&
1783                      rdev->sb_events+1 == mddev->events)) {
1784                         /* Don't update this superblock */
1785                         rdev->sb_loaded = 2;
1786                 } else {
1787                         super_types[mddev->major_version].
1788                                 sync_super(mddev, rdev);
1789                         rdev->sb_loaded = 1;
1790                 }
1791         }
1792 }
1793
1794 static void md_update_sb(mddev_t * mddev, int force_change)
1795 {
1796         mdk_rdev_t *rdev;
1797         int sync_req;
1798         int nospares = 0;
1799
1800         if (mddev->external)
1801                 return;
1802 repeat:
1803         spin_lock_irq(&mddev->write_lock);
1804
1805         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1806         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1807                 force_change = 1;
1808         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1809                 /* just a clean<-> dirty transition, possibly leave spares alone,
1810                  * though if events isn't the right even/odd, we will have to do
1811                  * spares after all
1812                  */
1813                 nospares = 1;
1814         if (force_change)
1815                 nospares = 0;
1816         if (mddev->degraded)
1817                 /* If the array is degraded, then skipping spares is both
1818                  * dangerous and fairly pointless.
1819                  * Dangerous because a device that was removed from the array
1820                  * might have a event_count that still looks up-to-date,
1821                  * so it can be re-added without a resync.
1822                  * Pointless because if there are any spares to skip,
1823                  * then a recovery will happen and soon that array won't
1824                  * be degraded any more and the spare can go back to sleep then.
1825                  */
1826                 nospares = 0;
1827
1828         sync_req = mddev->in_sync;
1829         mddev->utime = get_seconds();
1830
1831         /* If this is just a dirty<->clean transition, and the array is clean
1832          * and 'events' is odd, we can roll back to the previous clean state */
1833         if (nospares
1834             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1835             && (mddev->events & 1)
1836             && mddev->events != 1)
1837                 mddev->events--;
1838         else {
1839                 /* otherwise we have to go forward and ... */
1840                 mddev->events ++;
1841                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1842                         /* .. if the array isn't clean, insist on an odd 'events' */
1843                         if ((mddev->events&1)==0) {
1844                                 mddev->events++;
1845                                 nospares = 0;
1846                         }
1847                 } else {
1848                         /* otherwise insist on an even 'events' (for clean states) */
1849                         if ((mddev->events&1)) {
1850                                 mddev->events++;
1851                                 nospares = 0;
1852                         }
1853                 }
1854         }
1855
1856         if (!mddev->events) {
1857                 /*
1858                  * oops, this 64-bit counter should never wrap.
1859                  * Either we are in around ~1 trillion A.C., assuming
1860                  * 1 reboot per second, or we have a bug:
1861                  */
1862                 MD_BUG();
1863                 mddev->events --;
1864         }
1865
1866         /*
1867          * do not write anything to disk if using
1868          * nonpersistent superblocks
1869          */
1870         if (!mddev->persistent) {
1871                 if (!mddev->external)
1872                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1873
1874                 spin_unlock_irq(&mddev->write_lock);
1875                 wake_up(&mddev->sb_wait);
1876                 return;
1877         }
1878         sync_sbs(mddev, nospares);
1879         spin_unlock_irq(&mddev->write_lock);
1880
1881         dprintk(KERN_INFO
1882                 "md: updating %s RAID superblock on device (in sync %d)\n",
1883                 mdname(mddev),mddev->in_sync);
1884
1885         bitmap_update_sb(mddev->bitmap);
1886         list_for_each_entry(rdev, &mddev->disks, same_set) {
1887                 char b[BDEVNAME_SIZE];
1888                 dprintk(KERN_INFO "md: ");
1889                 if (rdev->sb_loaded != 1)
1890                         continue; /* no noise on spare devices */
1891                 if (test_bit(Faulty, &rdev->flags))
1892                         dprintk("(skipping faulty ");
1893
1894                 dprintk("%s ", bdevname(rdev->bdev,b));
1895                 if (!test_bit(Faulty, &rdev->flags)) {
1896                         md_super_write(mddev,rdev,
1897                                        rdev->sb_start, rdev->sb_size,
1898                                        rdev->sb_page);
1899                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1900                                 bdevname(rdev->bdev,b),
1901                                 (unsigned long long)rdev->sb_start);
1902                         rdev->sb_events = mddev->events;
1903
1904                 } else
1905                         dprintk(")\n");
1906                 if (mddev->level == LEVEL_MULTIPATH)
1907                         /* only need to write one superblock... */
1908                         break;
1909         }
1910         md_super_wait(mddev);
1911         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1912
1913         spin_lock_irq(&mddev->write_lock);
1914         if (mddev->in_sync != sync_req ||
1915             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1916                 /* have to write it out again */
1917                 spin_unlock_irq(&mddev->write_lock);
1918                 goto repeat;
1919         }
1920         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1921         spin_unlock_irq(&mddev->write_lock);
1922         wake_up(&mddev->sb_wait);
1923
1924 }
1925
1926 /* words written to sysfs files may, or may not, be \n terminated.
1927  * We want to accept with case. For this we use cmd_match.
1928  */
1929 static int cmd_match(const char *cmd, const char *str)
1930 {
1931         /* See if cmd, written into a sysfs file, matches
1932          * str.  They must either be the same, or cmd can
1933          * have a trailing newline
1934          */
1935         while (*cmd && *str && *cmd == *str) {
1936                 cmd++;
1937                 str++;
1938         }
1939         if (*cmd == '\n')
1940                 cmd++;
1941         if (*str || *cmd)
1942                 return 0;
1943         return 1;
1944 }
1945
1946 struct rdev_sysfs_entry {
1947         struct attribute attr;
1948         ssize_t (*show)(mdk_rdev_t *, char *);
1949         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1950 };
1951
1952 static ssize_t
1953 state_show(mdk_rdev_t *rdev, char *page)
1954 {
1955         char *sep = "";
1956         size_t len = 0;
1957
1958         if (test_bit(Faulty, &rdev->flags)) {
1959                 len+= sprintf(page+len, "%sfaulty",sep);
1960                 sep = ",";
1961         }
1962         if (test_bit(In_sync, &rdev->flags)) {
1963                 len += sprintf(page+len, "%sin_sync",sep);
1964                 sep = ",";
1965         }
1966         if (test_bit(WriteMostly, &rdev->flags)) {
1967                 len += sprintf(page+len, "%swrite_mostly",sep);
1968                 sep = ",";
1969         }
1970         if (test_bit(Blocked, &rdev->flags)) {
1971                 len += sprintf(page+len, "%sblocked", sep);
1972                 sep = ",";
1973         }
1974         if (!test_bit(Faulty, &rdev->flags) &&
1975             !test_bit(In_sync, &rdev->flags)) {
1976                 len += sprintf(page+len, "%sspare", sep);
1977                 sep = ",";
1978         }
1979         return len+sprintf(page+len, "\n");
1980 }
1981
1982 static ssize_t
1983 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1984 {
1985         /* can write
1986          *  faulty  - simulates and error
1987          *  remove  - disconnects the device
1988          *  writemostly - sets write_mostly
1989          *  -writemostly - clears write_mostly
1990          *  blocked - sets the Blocked flag
1991          *  -blocked - clears the Blocked flag
1992          */
1993         int err = -EINVAL;
1994         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1995                 md_error(rdev->mddev, rdev);
1996                 err = 0;
1997         } else if (cmd_match(buf, "remove")) {
1998                 if (rdev->raid_disk >= 0)
1999                         err = -EBUSY;
2000                 else {
2001                         mddev_t *mddev = rdev->mddev;
2002                         kick_rdev_from_array(rdev);
2003                         if (mddev->pers)
2004                                 md_update_sb(mddev, 1);
2005                         md_new_event(mddev);
2006                         err = 0;
2007                 }
2008         } else if (cmd_match(buf, "writemostly")) {
2009                 set_bit(WriteMostly, &rdev->flags);
2010                 err = 0;
2011         } else if (cmd_match(buf, "-writemostly")) {
2012                 clear_bit(WriteMostly, &rdev->flags);
2013                 err = 0;
2014         } else if (cmd_match(buf, "blocked")) {
2015                 set_bit(Blocked, &rdev->flags);
2016                 err = 0;
2017         } else if (cmd_match(buf, "-blocked")) {
2018                 clear_bit(Blocked, &rdev->flags);
2019                 wake_up(&rdev->blocked_wait);
2020                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2021                 md_wakeup_thread(rdev->mddev->thread);
2022
2023                 err = 0;
2024         }
2025         if (!err && rdev->sysfs_state)
2026                 sysfs_notify_dirent(rdev->sysfs_state);
2027         return err ? err : len;
2028 }
2029 static struct rdev_sysfs_entry rdev_state =
2030 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2031
2032 static ssize_t
2033 errors_show(mdk_rdev_t *rdev, char *page)
2034 {
2035         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2036 }
2037
2038 static ssize_t
2039 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2040 {
2041         char *e;
2042         unsigned long n = simple_strtoul(buf, &e, 10);
2043         if (*buf && (*e == 0 || *e == '\n')) {
2044                 atomic_set(&rdev->corrected_errors, n);
2045                 return len;
2046         }
2047         return -EINVAL;
2048 }
2049 static struct rdev_sysfs_entry rdev_errors =
2050 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2051
2052 static ssize_t
2053 slot_show(mdk_rdev_t *rdev, char *page)
2054 {
2055         if (rdev->raid_disk < 0)
2056                 return sprintf(page, "none\n");
2057         else
2058                 return sprintf(page, "%d\n", rdev->raid_disk);
2059 }
2060
2061 static ssize_t
2062 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2063 {
2064         char *e;
2065         int err;
2066         char nm[20];
2067         int slot = simple_strtoul(buf, &e, 10);
2068         if (strncmp(buf, "none", 4)==0)
2069                 slot = -1;
2070         else if (e==buf || (*e && *e!= '\n'))
2071                 return -EINVAL;
2072         if (rdev->mddev->pers && slot == -1) {
2073                 /* Setting 'slot' on an active array requires also
2074                  * updating the 'rd%d' link, and communicating
2075                  * with the personality with ->hot_*_disk.
2076                  * For now we only support removing
2077                  * failed/spare devices.  This normally happens automatically,
2078                  * but not when the metadata is externally managed.
2079                  */
2080                 if (rdev->raid_disk == -1)
2081                         return -EEXIST;
2082                 /* personality does all needed checks */
2083                 if (rdev->mddev->pers->hot_add_disk == NULL)
2084                         return -EINVAL;
2085                 err = rdev->mddev->pers->
2086                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2087                 if (err)
2088                         return err;
2089                 sprintf(nm, "rd%d", rdev->raid_disk);
2090                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2091                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2092                 md_wakeup_thread(rdev->mddev->thread);
2093         } else if (rdev->mddev->pers) {
2094                 mdk_rdev_t *rdev2;
2095                 /* Activating a spare .. or possibly reactivating
2096                  * if we every get bitmaps working here.
2097                  */
2098
2099                 if (rdev->raid_disk != -1)
2100                         return -EBUSY;
2101
2102                 if (rdev->mddev->pers->hot_add_disk == NULL)
2103                         return -EINVAL;
2104
2105                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2106                         if (rdev2->raid_disk == slot)
2107                                 return -EEXIST;
2108
2109                 rdev->raid_disk = slot;
2110                 if (test_bit(In_sync, &rdev->flags))
2111                         rdev->saved_raid_disk = slot;
2112                 else
2113                         rdev->saved_raid_disk = -1;
2114                 err = rdev->mddev->pers->
2115                         hot_add_disk(rdev->mddev, rdev);
2116                 if (err) {
2117                         rdev->raid_disk = -1;
2118                         return err;
2119                 } else
2120                         sysfs_notify_dirent(rdev->sysfs_state);
2121                 sprintf(nm, "rd%d", rdev->raid_disk);
2122                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2123                         printk(KERN_WARNING
2124                                "md: cannot register "
2125                                "%s for %s\n",
2126                                nm, mdname(rdev->mddev));
2127
2128                 /* don't wakeup anyone, leave that to userspace. */
2129         } else {
2130                 if (slot >= rdev->mddev->raid_disks)
2131                         return -ENOSPC;
2132                 rdev->raid_disk = slot;
2133                 /* assume it is working */
2134                 clear_bit(Faulty, &rdev->flags);
2135                 clear_bit(WriteMostly, &rdev->flags);
2136                 set_bit(In_sync, &rdev->flags);
2137                 sysfs_notify_dirent(rdev->sysfs_state);
2138         }
2139         return len;
2140 }
2141
2142
2143 static struct rdev_sysfs_entry rdev_slot =
2144 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2145
2146 static ssize_t
2147 offset_show(mdk_rdev_t *rdev, char *page)
2148 {
2149         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2150 }
2151
2152 static ssize_t
2153 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2154 {
2155         char *e;
2156         unsigned long long offset = simple_strtoull(buf, &e, 10);
2157         if (e==buf || (*e && *e != '\n'))
2158                 return -EINVAL;
2159         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2160                 return -EBUSY;
2161         if (rdev->size && rdev->mddev->external)
2162                 /* Must set offset before size, so overlap checks
2163                  * can be sane */
2164                 return -EBUSY;
2165         rdev->data_offset = offset;
2166         return len;
2167 }
2168
2169 static struct rdev_sysfs_entry rdev_offset =
2170 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2171
2172 static ssize_t
2173 rdev_size_show(mdk_rdev_t *rdev, char *page)
2174 {
2175         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2176 }
2177
2178 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2179 {
2180         /* check if two start/length pairs overlap */
2181         if (s1+l1 <= s2)
2182                 return 0;
2183         if (s2+l2 <= s1)
2184                 return 0;
2185         return 1;
2186 }
2187
2188 static ssize_t
2189 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2190 {
2191         unsigned long long size;
2192         unsigned long long oldsize = rdev->size;
2193         mddev_t *my_mddev = rdev->mddev;
2194
2195         if (strict_strtoull(buf, 10, &size) < 0)
2196                 return -EINVAL;
2197         if (my_mddev->pers && rdev->raid_disk >= 0) {
2198                 if (my_mddev->persistent) {
2199                         size = super_types[my_mddev->major_version].
2200                                 rdev_size_change(rdev, size * 2);
2201                         if (!size)
2202                                 return -EBUSY;
2203                 } else if (!size) {
2204                         size = (rdev->bdev->bd_inode->i_size >> 10);
2205                         size -= rdev->data_offset/2;
2206                 }
2207         }
2208         if (size < my_mddev->size)
2209                 return -EINVAL; /* component must fit device */
2210
2211         rdev->size = size;
2212         if (size > oldsize && my_mddev->external) {
2213                 /* need to check that all other rdevs with the same ->bdev
2214                  * do not overlap.  We need to unlock the mddev to avoid
2215                  * a deadlock.  We have already changed rdev->size, and if
2216                  * we have to change it back, we will have the lock again.
2217                  */
2218                 mddev_t *mddev;
2219                 int overlap = 0;
2220                 struct list_head *tmp;
2221
2222                 mddev_unlock(my_mddev);
2223                 for_each_mddev(mddev, tmp) {
2224                         mdk_rdev_t *rdev2;
2225
2226                         mddev_lock(mddev);
2227                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2228                                 if (test_bit(AllReserved, &rdev2->flags) ||
2229                                     (rdev->bdev == rdev2->bdev &&
2230                                      rdev != rdev2 &&
2231                                      overlaps(rdev->data_offset, rdev->size * 2,
2232                                               rdev2->data_offset,
2233                                               rdev2->size * 2))) {
2234                                         overlap = 1;
2235                                         break;
2236                                 }
2237                         mddev_unlock(mddev);
2238                         if (overlap) {
2239                                 mddev_put(mddev);
2240                                 break;
2241                         }
2242                 }
2243                 mddev_lock(my_mddev);
2244                 if (overlap) {
2245                         /* Someone else could have slipped in a size
2246                          * change here, but doing so is just silly.
2247                          * We put oldsize back because we *know* it is
2248                          * safe, and trust userspace not to race with
2249                          * itself
2250                          */
2251                         rdev->size = oldsize;
2252                         return -EBUSY;
2253                 }
2254         }
2255         return len;
2256 }
2257
2258 static struct rdev_sysfs_entry rdev_size =
2259 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2260
2261 static struct attribute *rdev_default_attrs[] = {
2262         &rdev_state.attr,
2263         &rdev_errors.attr,
2264         &rdev_slot.attr,
2265         &rdev_offset.attr,
2266         &rdev_size.attr,
2267         NULL,
2268 };
2269 static ssize_t
2270 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2271 {
2272         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2273         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2274         mddev_t *mddev = rdev->mddev;
2275         ssize_t rv;
2276
2277         if (!entry->show)
2278                 return -EIO;
2279
2280         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2281         if (!rv) {
2282                 if (rdev->mddev == NULL)
2283                         rv = -EBUSY;
2284                 else
2285                         rv = entry->show(rdev, page);
2286                 mddev_unlock(mddev);
2287         }
2288         return rv;
2289 }
2290
2291 static ssize_t
2292 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2293               const char *page, size_t length)
2294 {
2295         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2296         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2297         ssize_t rv;
2298         mddev_t *mddev = rdev->mddev;
2299
2300         if (!entry->store)
2301                 return -EIO;
2302         if (!capable(CAP_SYS_ADMIN))
2303                 return -EACCES;
2304         rv = mddev ? mddev_lock(mddev): -EBUSY;
2305         if (!rv) {
2306                 if (rdev->mddev == NULL)
2307                         rv = -EBUSY;
2308                 else
2309                         rv = entry->store(rdev, page, length);
2310                 mddev_unlock(mddev);
2311         }
2312         return rv;
2313 }
2314
2315 static void rdev_free(struct kobject *ko)
2316 {
2317         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2318         kfree(rdev);
2319 }
2320 static struct sysfs_ops rdev_sysfs_ops = {
2321         .show           = rdev_attr_show,
2322         .store          = rdev_attr_store,
2323 };
2324 static struct kobj_type rdev_ktype = {
2325         .release        = rdev_free,
2326         .sysfs_ops      = &rdev_sysfs_ops,
2327         .default_attrs  = rdev_default_attrs,
2328 };
2329
2330 /*
2331  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2332  *
2333  * mark the device faulty if:
2334  *
2335  *   - the device is nonexistent (zero size)
2336  *   - the device has no valid superblock
2337  *
2338  * a faulty rdev _never_ has rdev->sb set.
2339  */
2340 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2341 {
2342         char b[BDEVNAME_SIZE];
2343         int err;
2344         mdk_rdev_t *rdev;
2345         sector_t size;
2346
2347         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2348         if (!rdev) {
2349                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2350                 return ERR_PTR(-ENOMEM);
2351         }
2352
2353         if ((err = alloc_disk_sb(rdev)))
2354                 goto abort_free;
2355
2356         err = lock_rdev(rdev, newdev, super_format == -2);
2357         if (err)
2358                 goto abort_free;
2359
2360         kobject_init(&rdev->kobj, &rdev_ktype);
2361
2362         rdev->desc_nr = -1;
2363         rdev->saved_raid_disk = -1;
2364         rdev->raid_disk = -1;
2365         rdev->flags = 0;
2366         rdev->data_offset = 0;
2367         rdev->sb_events = 0;
2368         atomic_set(&rdev->nr_pending, 0);
2369         atomic_set(&rdev->read_errors, 0);
2370         atomic_set(&rdev->corrected_errors, 0);
2371
2372         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2373         if (!size) {
2374                 printk(KERN_WARNING
2375                         "md: %s has zero or unknown size, marking faulty!\n",
2376                         bdevname(rdev->bdev,b));
2377                 err = -EINVAL;
2378                 goto abort_free;
2379         }
2380
2381         if (super_format >= 0) {
2382                 err = super_types[super_format].
2383                         load_super(rdev, NULL, super_minor);
2384                 if (err == -EINVAL) {
2385                         printk(KERN_WARNING
2386                                 "md: %s does not have a valid v%d.%d "
2387                                "superblock, not importing!\n",
2388                                 bdevname(rdev->bdev,b),
2389                                super_format, super_minor);
2390                         goto abort_free;
2391                 }
2392                 if (err < 0) {
2393                         printk(KERN_WARNING
2394                                 "md: could not read %s's sb, not importing!\n",
2395                                 bdevname(rdev->bdev,b));
2396                         goto abort_free;
2397                 }
2398         }
2399
2400         INIT_LIST_HEAD(&rdev->same_set);
2401         init_waitqueue_head(&rdev->blocked_wait);
2402
2403         return rdev;
2404
2405 abort_free:
2406         if (rdev->sb_page) {
2407                 if (rdev->bdev)
2408                         unlock_rdev(rdev);
2409                 free_disk_sb(rdev);
2410         }
2411         kfree(rdev);
2412         return ERR_PTR(err);
2413 }
2414
2415 /*
2416  * Check a full RAID array for plausibility
2417  */
2418
2419
2420 static void analyze_sbs(mddev_t * mddev)
2421 {
2422         int i;
2423         mdk_rdev_t *rdev, *freshest, *tmp;
2424         char b[BDEVNAME_SIZE];
2425
2426         freshest = NULL;
2427         rdev_for_each(rdev, tmp, mddev)
2428                 switch (super_types[mddev->major_version].
2429                         load_super(rdev, freshest, mddev->minor_version)) {
2430                 case 1:
2431                         freshest = rdev;
2432                         break;
2433                 case 0:
2434                         break;
2435                 default:
2436                         printk( KERN_ERR \
2437                                 "md: fatal superblock inconsistency in %s"
2438                                 " -- removing from array\n",
2439                                 bdevname(rdev->bdev,b));
2440                         kick_rdev_from_array(rdev);
2441                 }
2442
2443
2444         super_types[mddev->major_version].
2445                 validate_super(mddev, freshest);
2446
2447         i = 0;
2448         rdev_for_each(rdev, tmp, mddev) {
2449                 if (rdev->desc_nr >= mddev->max_disks ||
2450                     i > mddev->max_disks) {
2451                         printk(KERN_WARNING
2452                                "md: %s: %s: only %d devices permitted\n",
2453                                mdname(mddev), bdevname(rdev->bdev, b),
2454                                mddev->max_disks);
2455                         kick_rdev_from_array(rdev);
2456                         continue;
2457                 }
2458                 if (rdev != freshest)
2459                         if (super_types[mddev->major_version].
2460                             validate_super(mddev, rdev)) {
2461                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2462                                         " from array!\n",
2463                                         bdevname(rdev->bdev,b));
2464                                 kick_rdev_from_array(rdev);
2465                                 continue;
2466                         }
2467                 if (mddev->level == LEVEL_MULTIPATH) {
2468                         rdev->desc_nr = i++;
2469                         rdev->raid_disk = rdev->desc_nr;
2470                         set_bit(In_sync, &rdev->flags);
2471                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2472                         rdev->raid_disk = -1;
2473                         clear_bit(In_sync, &rdev->flags);
2474                 }
2475         }
2476
2477
2478
2479         if (mddev->recovery_cp != MaxSector &&
2480             mddev->level >= 1)
2481                 printk(KERN_ERR "md: %s: raid array is not clean"
2482                        " -- starting background reconstruction\n",
2483                        mdname(mddev));
2484
2485 }
2486
2487 static void md_safemode_timeout(unsigned long data);
2488
2489 static ssize_t
2490 safe_delay_show(mddev_t *mddev, char *page)
2491 {
2492         int msec = (mddev->safemode_delay*1000)/HZ;
2493         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2494 }
2495 static ssize_t
2496 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2497 {
2498         int scale=1;
2499         int dot=0;
2500         int i;
2501         unsigned long msec;
2502         char buf[30];
2503
2504         /* remove a period, and count digits after it */
2505         if (len >= sizeof(buf))
2506                 return -EINVAL;
2507         strlcpy(buf, cbuf, sizeof(buf));
2508         for (i=0; i<len; i++) {
2509                 if (dot) {
2510                         if (isdigit(buf[i])) {
2511                                 buf[i-1] = buf[i];
2512                                 scale *= 10;
2513                         }
2514                         buf[i] = 0;
2515                 } else if (buf[i] == '.') {
2516                         dot=1;
2517                         buf[i] = 0;
2518                 }
2519         }
2520         if (strict_strtoul(buf, 10, &msec) < 0)
2521                 return -EINVAL;
2522         msec = (msec * 1000) / scale;
2523         if (msec == 0)
2524                 mddev->safemode_delay = 0;
2525         else {
2526                 unsigned long old_delay = mddev->safemode_delay;
2527                 mddev->safemode_delay = (msec*HZ)/1000;
2528                 if (mddev->safemode_delay == 0)
2529                         mddev->safemode_delay = 1;
2530                 if (mddev->safemode_delay < old_delay)
2531                         md_safemode_timeout((unsigned long)mddev);
2532         }
2533         return len;
2534 }
2535 static struct md_sysfs_entry md_safe_delay =
2536 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2537
2538 static ssize_t
2539 level_show(mddev_t *mddev, char *page)
2540 {
2541         struct mdk_personality *p = mddev->pers;
2542         if (p)
2543                 return sprintf(page, "%s\n", p->name);
2544         else if (mddev->clevel[0])
2545                 return sprintf(page, "%s\n", mddev->clevel);
2546         else if (mddev->level != LEVEL_NONE)
2547                 return sprintf(page, "%d\n", mddev->level);
2548         else
2549                 return 0;
2550 }
2551
2552 static ssize_t
2553 level_store(mddev_t *mddev, const char *buf, size_t len)
2554 {
2555         ssize_t rv = len;
2556         if (mddev->pers)
2557                 return -EBUSY;
2558         if (len == 0)
2559                 return 0;
2560         if (len >= sizeof(mddev->clevel))
2561                 return -ENOSPC;
2562         strncpy(mddev->clevel, buf, len);
2563         if (mddev->clevel[len-1] == '\n')
2564                 len--;
2565         mddev->clevel[len] = 0;
2566         mddev->level = LEVEL_NONE;
2567         return rv;
2568 }
2569
2570 static struct md_sysfs_entry md_level =
2571 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2572
2573
2574 static ssize_t
2575 layout_show(mddev_t *mddev, char *page)
2576 {
2577         /* just a number, not meaningful for all levels */
2578         if (mddev->reshape_position != MaxSector &&
2579             mddev->layout != mddev->new_layout)
2580                 return sprintf(page, "%d (%d)\n",
2581                                mddev->new_layout, mddev->layout);
2582         return sprintf(page, "%d\n", mddev->layout);
2583 }
2584
2585 static ssize_t
2586 layout_store(mddev_t *mddev, const char *buf, size_t len)
2587 {
2588         char *e;
2589         unsigned long n = simple_strtoul(buf, &e, 10);
2590
2591         if (!*buf || (*e && *e != '\n'))
2592                 return -EINVAL;
2593
2594         if (mddev->pers)
2595                 return -EBUSY;
2596         if (mddev->reshape_position != MaxSector)
2597                 mddev->new_layout = n;
2598         else
2599                 mddev->layout = n;
2600         return len;
2601 }
2602 static struct md_sysfs_entry md_layout =
2603 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2604
2605
2606 static ssize_t
2607 raid_disks_show(mddev_t *mddev, char *page)
2608 {
2609         if (mddev->raid_disks == 0)
2610                 return 0;
2611         if (mddev->reshape_position != MaxSector &&
2612             mddev->delta_disks != 0)
2613                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2614                                mddev->raid_disks - mddev->delta_disks);
2615         return sprintf(page, "%d\n", mddev->raid_disks);
2616 }
2617
2618 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2619
2620 static ssize_t
2621 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2622 {
2623         char *e;
2624         int rv = 0;
2625         unsigned long n = simple_strtoul(buf, &e, 10);
2626
2627         if (!*buf || (*e && *e != '\n'))
2628                 return -EINVAL;
2629
2630         if (mddev->pers)
2631                 rv = update_raid_disks(mddev, n);
2632         else if (mddev->reshape_position != MaxSector) {
2633                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2634                 mddev->delta_disks = n - olddisks;
2635                 mddev->raid_disks = n;
2636         } else
2637                 mddev->raid_disks = n;
2638         return rv ? rv : len;
2639 }
2640 static struct md_sysfs_entry md_raid_disks =
2641 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2642
2643 static ssize_t
2644 chunk_size_show(mddev_t *mddev, char *page)
2645 {
2646         if (mddev->reshape_position != MaxSector &&
2647             mddev->chunk_size != mddev->new_chunk)
2648                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2649                                mddev->chunk_size);
2650         return sprintf(page, "%d\n", mddev->chunk_size);
2651 }
2652
2653 static ssize_t
2654 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2655 {
2656         /* can only set chunk_size if array is not yet active */
2657         char *e;
2658         unsigned long n = simple_strtoul(buf, &e, 10);
2659
2660         if (!*buf || (*e && *e != '\n'))
2661                 return -EINVAL;
2662
2663         if (mddev->pers)
2664                 return -EBUSY;
2665         else if (mddev->reshape_position != MaxSector)
2666                 mddev->new_chunk = n;
2667         else
2668                 mddev->chunk_size = n;
2669         return len;
2670 }
2671 static struct md_sysfs_entry md_chunk_size =
2672 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2673
2674 static ssize_t
2675 resync_start_show(mddev_t *mddev, char *page)
2676 {
2677         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2678 }
2679
2680 static ssize_t
2681 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2682 {
2683         char *e;
2684         unsigned long long n = simple_strtoull(buf, &e, 10);
2685
2686         if (mddev->pers)
2687                 return -EBUSY;
2688         if (!*buf || (*e && *e != '\n'))
2689                 return -EINVAL;
2690
2691         mddev->recovery_cp = n;
2692         return len;
2693 }
2694 static struct md_sysfs_entry md_resync_start =
2695 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2696
2697 /*
2698  * The array state can be:
2699  *
2700  * clear
2701  *     No devices, no size, no level
2702  *     Equivalent to STOP_ARRAY ioctl
2703  * inactive
2704  *     May have some settings, but array is not active
2705  *        all IO results in error
2706  *     When written, doesn't tear down array, but just stops it
2707  * suspended (not supported yet)
2708  *     All IO requests will block. The array can be reconfigured.
2709  *     Writing this, if accepted, will block until array is quiescent
2710  * readonly
2711  *     no resync can happen.  no superblocks get written.
2712  *     write requests fail
2713  * read-auto
2714  *     like readonly, but behaves like 'clean' on a write request.
2715  *
2716  * clean - no pending writes, but otherwise active.
2717  *     When written to inactive array, starts without resync
2718  *     If a write request arrives then
2719  *       if metadata is known, mark 'dirty' and switch to 'active'.
2720  *       if not known, block and switch to write-pending
2721  *     If written to an active array that has pending writes, then fails.
2722  * active
2723  *     fully active: IO and resync can be happening.
2724  *     When written to inactive array, starts with resync
2725  *
2726  * write-pending
2727  *     clean, but writes are blocked waiting for 'active' to be written.
2728  *
2729  * active-idle
2730  *     like active, but no writes have been seen for a while (100msec).
2731  *
2732  */
2733 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2734                    write_pending, active_idle, bad_word};
2735 static char *array_states[] = {
2736         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2737         "write-pending", "active-idle", NULL };
2738
2739 static int match_word(const char *word, char **list)
2740 {
2741         int n;
2742         for (n=0; list[n]; n++)
2743                 if (cmd_match(word, list[n]))
2744                         break;
2745         return n;
2746 }
2747
2748 static ssize_t
2749 array_state_show(mddev_t *mddev, char *page)
2750 {
2751         enum array_state st = inactive;
2752
2753         if (mddev->pers)
2754                 switch(mddev->ro) {
2755                 case 1:
2756                         st = readonly;
2757                         break;
2758                 case 2:
2759                         st = read_auto;
2760                         break;
2761                 case 0:
2762                         if (mddev->in_sync)
2763                                 st = clean;
2764                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2765                                 st = write_pending;
2766                         else if (mddev->safemode)
2767                                 st = active_idle;
2768                         else
2769                                 st = active;
2770                 }
2771         else {
2772                 if (list_empty(&mddev->disks) &&
2773                     mddev->raid_disks == 0 &&
2774                     mddev->size == 0)
2775                         st = clear;
2776                 else
2777                         st = inactive;
2778         }
2779         return sprintf(page, "%s\n", array_states[st]);
2780 }
2781
2782 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2783 static int do_md_run(mddev_t * mddev);
2784 static int restart_array(mddev_t *mddev);
2785
2786 static ssize_t
2787 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2788 {
2789         int err = -EINVAL;
2790         enum array_state st = match_word(buf, array_states);
2791         switch(st) {
2792         case bad_word:
2793                 break;
2794         case clear:
2795                 /* stopping an active array */
2796                 if (atomic_read(&mddev->openers) > 0)
2797                         return -EBUSY;
2798                 err = do_md_stop(mddev, 0, 0);
2799                 break;
2800         case inactive:
2801                 /* stopping an active array */
2802                 if (mddev->pers) {
2803                         if (atomic_read(&mddev->openers) > 0)
2804                                 return -EBUSY;
2805                         err = do_md_stop(mddev, 2, 0);
2806                 } else
2807                         err = 0; /* already inactive */
2808                 break;
2809         case suspended:
2810                 break; /* not supported yet */
2811         case readonly:
2812                 if (mddev->pers)
2813                         err = do_md_stop(mddev, 1, 0);
2814                 else {
2815                         mddev->ro = 1;
2816                         set_disk_ro(mddev->gendisk, 1);
2817                         err = do_md_run(mddev);
2818                 }
2819                 break;
2820         case read_auto:
2821                 if (mddev->pers) {
2822                         if (mddev->ro == 0)
2823                                 err = do_md_stop(mddev, 1, 0);
2824                         else if (mddev->ro == 1)
2825                                 err = restart_array(mddev);
2826                         if (err == 0) {
2827                                 mddev->ro = 2;
2828                                 set_disk_ro(mddev->gendisk, 0);
2829                         }
2830                 } else {
2831                         mddev->ro = 2;
2832                         err = do_md_run(mddev);
2833                 }
2834                 break;
2835         case clean:
2836                 if (mddev->pers) {
2837                         restart_array(mddev);
2838                         spin_lock_irq(&mddev->write_lock);
2839                         if (atomic_read(&mddev->writes_pending) == 0) {
2840                                 if (mddev->in_sync == 0) {
2841                                         mddev->in_sync = 1;
2842                                         if (mddev->safemode == 1)
2843                                                 mddev->safemode = 0;
2844                                         if (mddev->persistent)
2845                                                 set_bit(MD_CHANGE_CLEAN,
2846                                                         &mddev->flags);
2847                                 }
2848                                 err = 0;
2849                         } else
2850                                 err = -EBUSY;
2851                         spin_unlock_irq(&mddev->write_lock);
2852                 } else {
2853                         mddev->ro = 0;
2854                         mddev->recovery_cp = MaxSector;
2855                         err = do_md_run(mddev);
2856                 }
2857                 break;
2858         case active:
2859                 if (mddev->pers) {
2860                         restart_array(mddev);
2861                         if (mddev->external)
2862                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2863                         wake_up(&mddev->sb_wait);
2864                         err = 0;
2865                 } else {
2866                         mddev->ro = 0;
2867                         set_disk_ro(mddev->gendisk, 0);
2868                         err = do_md_run(mddev);
2869                 }
2870                 break;
2871         case write_pending:
2872         case active_idle:
2873                 /* these cannot be set */
2874                 break;
2875         }
2876         if (err)
2877                 return err;
2878         else {
2879                 sysfs_notify_dirent(mddev->sysfs_state);
2880                 return len;
2881         }
2882 }
2883 static struct md_sysfs_entry md_array_state =
2884 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2885
2886 static ssize_t
2887 null_show(mddev_t *mddev, char *page)
2888 {
2889         return -EINVAL;
2890 }
2891
2892 static ssize_t
2893 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2894 {
2895         /* buf must be %d:%d\n? giving major and minor numbers */
2896         /* The new device is added to the array.
2897          * If the array has a persistent superblock, we read the
2898          * superblock to initialise info and check validity.
2899          * Otherwise, only checking done is that in bind_rdev_to_array,
2900          * which mainly checks size.
2901          */
2902         char *e;
2903         int major = simple_strtoul(buf, &e, 10);
2904         int minor;
2905         dev_t dev;
2906         mdk_rdev_t *rdev;
2907         int err;
2908
2909         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2910                 return -EINVAL;
2911         minor = simple_strtoul(e+1, &e, 10);
2912         if (*e && *e != '\n')
2913                 return -EINVAL;
2914         dev = MKDEV(major, minor);
2915         if (major != MAJOR(dev) ||
2916             minor != MINOR(dev))
2917                 return -EOVERFLOW;
2918
2919
2920         if (mddev->persistent) {
2921                 rdev = md_import_device(dev, mddev->major_version,
2922                                         mddev->minor_version);
2923                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2924                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2925                                                        mdk_rdev_t, same_set);
2926                         err = super_types[mddev->major_version]
2927                                 .load_super(rdev, rdev0, mddev->minor_version);
2928                         if (err < 0)
2929                                 goto out;
2930                 }
2931         } else if (mddev->external)
2932                 rdev = md_import_device(dev, -2, -1);
2933         else
2934                 rdev = md_import_device(dev, -1, -1);
2935
2936         if (IS_ERR(rdev))
2937                 return PTR_ERR(rdev);
2938         err = bind_rdev_to_array(rdev, mddev);
2939  out:
2940         if (err)
2941                 export_rdev(rdev);
2942         return err ? err : len;
2943 }
2944
2945 static struct md_sysfs_entry md_new_device =
2946 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2947
2948 static ssize_t
2949 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2950 {
2951         char *end;
2952         unsigned long chunk, end_chunk;
2953
2954         if (!mddev->bitmap)
2955                 goto out;
2956         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2957         while (*buf) {
2958                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2959                 if (buf == end) break;
2960                 if (*end == '-') { /* range */
2961                         buf = end + 1;
2962                         end_chunk = simple_strtoul(buf, &end, 0);
2963                         if (buf == end) break;
2964                 }
2965                 if (*end && !isspace(*end)) break;
2966                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2967                 buf = end;
2968                 while (isspace(*buf)) buf++;
2969         }
2970         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2971 out:
2972         return len;
2973 }
2974
2975 static struct md_sysfs_entry md_bitmap =
2976 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2977
2978 static ssize_t
2979 size_show(mddev_t *mddev, char *page)
2980 {
2981         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2982 }
2983
2984 static int update_size(mddev_t *mddev, sector_t num_sectors);
2985
2986 static ssize_t
2987 size_store(mddev_t *mddev, const char *buf, size_t len)
2988 {
2989         /* If array is inactive, we can reduce the component size, but
2990          * not increase it (except from 0).
2991          * If array is active, we can try an on-line resize
2992          */
2993         char *e;
2994         int err = 0;
2995         unsigned long long size = simple_strtoull(buf, &e, 10);
2996         if (!*buf || *buf == '\n' ||
2997             (*e && *e != '\n'))
2998                 return -EINVAL;
2999
3000         if (mddev->pers) {
3001                 err = update_size(mddev, size * 2);
3002                 md_update_sb(mddev, 1);
3003         } else {
3004                 if (mddev->size == 0 ||
3005                     mddev->size > size)
3006                         mddev->size = size;
3007                 else
3008                         err = -ENOSPC;
3009         }
3010         return err ? err : len;
3011 }
3012
3013 static struct md_sysfs_entry md_size =
3014 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3015
3016
3017 /* Metdata version.
3018  * This is one of
3019  *   'none' for arrays with no metadata (good luck...)
3020  *   'external' for arrays with externally managed metadata,
3021  * or N.M for internally known formats
3022  */
3023 static ssize_t
3024 metadata_show(mddev_t *mddev, char *page)
3025 {
3026         if (mddev->persistent)
3027                 return sprintf(page, "%d.%d\n",
3028                                mddev->major_version, mddev->minor_version);
3029         else if (mddev->external)
3030                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3031         else
3032                 return sprintf(page, "none\n");
3033 }
3034
3035 static ssize_t
3036 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3037 {
3038         int major, minor;
3039         char *e;
3040         /* Changing the details of 'external' metadata is
3041          * always permitted.  Otherwise there must be
3042          * no devices attached to the array.
3043          */
3044         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3045                 ;
3046         else if (!list_empty(&mddev->disks))
3047                 return -EBUSY;
3048
3049         if (cmd_match(buf, "none")) {
3050                 mddev->persistent = 0;
3051                 mddev->external = 0;
3052                 mddev->major_version = 0;
3053                 mddev->minor_version = 90;
3054                 return len;
3055         }
3056         if (strncmp(buf, "external:", 9) == 0) {
3057                 size_t namelen = len-9;
3058                 if (namelen >= sizeof(mddev->metadata_type))
3059                         namelen = sizeof(mddev->metadata_type)-1;
3060                 strncpy(mddev->metadata_type, buf+9, namelen);
3061                 mddev->metadata_type[namelen] = 0;
3062                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3063                         mddev->metadata_type[--namelen] = 0;
3064                 mddev->persistent = 0;
3065                 mddev->external = 1;
3066                 mddev->major_version = 0;
3067                 mddev->minor_version = 90;
3068                 return len;
3069         }
3070         major = simple_strtoul(buf, &e, 10);
3071         if (e==buf || *e != '.')
3072                 return -EINVAL;
3073         buf = e+1;
3074         minor = simple_strtoul(buf, &e, 10);
3075         if (e==buf || (*e && *e != '\n') )
3076                 return -EINVAL;
3077         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3078                 return -ENOENT;
3079         mddev->major_version = major;
3080         mddev->minor_version = minor;
3081         mddev->persistent = 1;
3082         mddev->external = 0;
3083         return len;
3084 }
3085
3086 static struct md_sysfs_entry md_metadata =
3087 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3088
3089 static ssize_t
3090 action_show(mddev_t *mddev, char *page)
3091 {
3092         char *type = "idle";
3093         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3094             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3095                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3096                         type = "reshape";
3097                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3098                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3099                                 type = "resync";
3100                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3101                                 type = "check";
3102                         else
3103                                 type = "repair";
3104                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3105                         type = "recover";
3106         }
3107         return sprintf(page, "%s\n", type);
3108 }
3109
3110 static ssize_t
3111 action_store(mddev_t *mddev, const char *page, size_t len)
3112 {
3113         if (!mddev->pers || !mddev->pers->sync_request)
3114                 return -EINVAL;
3115
3116         if (cmd_match(page, "idle")) {
3117                 if (mddev->sync_thread) {
3118                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3119                         md_unregister_thread(mddev->sync_thread);
3120                         mddev->sync_thread = NULL;
3121                         mddev->recovery = 0;
3122                 }
3123         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3124                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3125                 return -EBUSY;
3126         else if (cmd_match(page, "resync"))
3127                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3128         else if (cmd_match(page, "recover")) {
3129                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3130                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3131         } else if (cmd_match(page, "reshape")) {
3132                 int err;
3133                 if (mddev->pers->start_reshape == NULL)
3134                         return -EINVAL;
3135                 err = mddev->pers->start_reshape(mddev);
3136                 if (err)
3137                         return err;
3138                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3139         } else {
3140                 if (cmd_match(page, "check"))
3141                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3142                 else if (!cmd_match(page, "repair"))
3143                         return -EINVAL;
3144                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3145                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3146         }
3147         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3148         md_wakeup_thread(mddev->thread);
3149         sysfs_notify_dirent(mddev->sysfs_action);
3150         return len;
3151 }
3152
3153 static ssize_t
3154 mismatch_cnt_show(mddev_t *mddev, char *page)
3155 {
3156         return sprintf(page, "%llu\n",
3157                        (unsigned long long) mddev->resync_mismatches);
3158 }
3159
3160 static struct md_sysfs_entry md_scan_mode =
3161 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3162
3163
3164 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3165
3166 static ssize_t
3167 sync_min_show(mddev_t *mddev, char *page)
3168 {
3169         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3170                        mddev->sync_speed_min ? "local": "system");
3171 }
3172
3173 static ssize_t
3174 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3175 {
3176         int min;
3177         char *e;
3178         if (strncmp(buf, "system", 6)==0) {
3179                 mddev->sync_speed_min = 0;
3180                 return len;
3181         }
3182         min = simple_strtoul(buf, &e, 10);
3183         if (buf == e || (*e && *e != '\n') || min <= 0)
3184                 return -EINVAL;
3185         mddev->sync_speed_min = min;
3186         return len;
3187 }
3188
3189 static struct md_sysfs_entry md_sync_min =
3190 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3191
3192 static ssize_t
3193 sync_max_show(mddev_t *mddev, char *page)
3194 {
3195         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3196                        mddev->sync_speed_max ? "local": "system");
3197 }
3198
3199 static ssize_t
3200 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3201 {
3202         int max;
3203         char *e;
3204         if (strncmp(buf, "system", 6)==0) {
3205                 mddev->sync_speed_max = 0;
3206                 return len;
3207         }
3208         max = simple_strtoul(buf, &e, 10);
3209         if (buf == e || (*e && *e != '\n') || max <= 0)
3210                 return -EINVAL;
3211         mddev->sync_speed_max = max;
3212         return len;
3213 }
3214
3215 static struct md_sysfs_entry md_sync_max =
3216 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3217
3218 static ssize_t
3219 degraded_show(mddev_t *mddev, char *page)
3220 {
3221         return sprintf(page, "%d\n", mddev->degraded);
3222 }
3223 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3224
3225 static ssize_t
3226 sync_force_parallel_show(mddev_t *mddev, char *page)
3227 {
3228         return sprintf(page, "%d\n", mddev->parallel_resync);
3229 }
3230
3231 static ssize_t
3232 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3233 {
3234         long n;
3235
3236         if (strict_strtol(buf, 10, &n))
3237                 return -EINVAL;
3238
3239         if (n != 0 && n != 1)
3240                 return -EINVAL;
3241
3242         mddev->parallel_resync = n;
3243
3244         if (mddev->sync_thread)
3245                 wake_up(&resync_wait);
3246
3247         return len;
3248 }
3249
3250 /* force parallel resync, even with shared block devices */
3251 static struct md_sysfs_entry md_sync_force_parallel =
3252 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3253        sync_force_parallel_show, sync_force_parallel_store);
3254
3255 static ssize_t
3256 sync_speed_show(mddev_t *mddev, char *page)
3257 {
3258         unsigned long resync, dt, db;
3259         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3260         dt = (jiffies - mddev->resync_mark) / HZ;
3261         if (!dt) dt++;
3262         db = resync - mddev->resync_mark_cnt;
3263         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3264 }
3265
3266 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3267
3268 static ssize_t
3269 sync_completed_show(mddev_t *mddev, char *page)
3270 {
3271         unsigned long max_blocks, resync;
3272
3273         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3274                 max_blocks = mddev->resync_max_sectors;
3275         else
3276                 max_blocks = mddev->size << 1;
3277
3278         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3279         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3280 }
3281
3282 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3283
3284 static ssize_t
3285 min_sync_show(mddev_t *mddev, char *page)
3286 {
3287         return sprintf(page, "%llu\n",
3288                        (unsigned long long)mddev->resync_min);
3289 }
3290 static ssize_t
3291 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3292 {
3293         unsigned long long min;
3294         if (strict_strtoull(buf, 10, &min))
3295                 return -EINVAL;
3296         if (min > mddev->resync_max)
3297                 return -EINVAL;
3298         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3299                 return -EBUSY;
3300
3301         /* Must be a multiple of chunk_size */
3302         if (mddev->chunk_size) {
3303                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3304                         return -EINVAL;
3305         }
3306         mddev->resync_min = min;
3307
3308         return len;
3309 }
3310
3311 static struct md_sysfs_entry md_min_sync =
3312 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3313
3314 static ssize_t
3315 max_sync_show(mddev_t *mddev, char *page)
3316 {
3317         if (mddev->resync_max == MaxSector)
3318                 return sprintf(page, "max\n");
3319         else
3320                 return sprintf(page, "%llu\n",
3321                                (unsigned long long)mddev->resync_max);
3322 }
3323 static ssize_t
3324 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3325 {
3326         if (strncmp(buf, "max", 3) == 0)
3327                 mddev->resync_max = MaxSector;
3328         else {
3329                 unsigned long long max;
3330                 if (strict_strtoull(buf, 10, &max))
3331                         return -EINVAL;
3332                 if (max < mddev->resync_min)
3333                         return -EINVAL;
3334                 if (max < mddev->resync_max &&
3335                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3336                         return -EBUSY;
3337
3338                 /* Must be a multiple of chunk_size */
3339                 if (mddev->chunk_size) {
3340                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3341                                 return -EINVAL;
3342                 }
3343                 mddev->resync_max = max;
3344         }
3345         wake_up(&mddev->recovery_wait);
3346         return len;
3347 }
3348
3349 static struct md_sysfs_entry md_max_sync =
3350 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3351
3352 static ssize_t
3353 suspend_lo_show(mddev_t *mddev, char *page)
3354 {
3355         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3356 }
3357
3358 static ssize_t
3359 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3360 {
3361         char *e;
3362         unsigned long long new = simple_strtoull(buf, &e, 10);
3363
3364         if (mddev->pers->quiesce == NULL)
3365                 return -EINVAL;
3366         if (buf == e || (*e && *e != '\n'))
3367                 return -EINVAL;
3368         if (new >= mddev->suspend_hi ||
3369             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3370                 mddev->suspend_lo = new;
3371                 mddev->pers->quiesce(mddev, 2);
3372                 return len;
3373         } else
3374                 return -EINVAL;
3375 }
3376 static struct md_sysfs_entry md_suspend_lo =
3377 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3378
3379
3380 static ssize_t
3381 suspend_hi_show(mddev_t *mddev, char *page)
3382 {
3383         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3384 }
3385
3386 static ssize_t
3387 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3388 {
3389         char *e;
3390         unsigned long long new = simple_strtoull(buf, &e, 10);
3391
3392         if (mddev->pers->quiesce == NULL)
3393                 return -EINVAL;
3394         if (buf == e || (*e && *e != '\n'))
3395                 return -EINVAL;
3396         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3397             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3398                 mddev->suspend_hi = new;
3399                 mddev->pers->quiesce(mddev, 1);
3400                 mddev->pers->quiesce(mddev, 0);
3401                 return len;
3402         } else
3403                 return -EINVAL;
3404 }
3405 static struct md_sysfs_entry md_suspend_hi =
3406 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3407
3408 static ssize_t
3409 reshape_position_show(mddev_t *mddev, char *page)
3410 {
3411         if (mddev->reshape_position != MaxSector)
3412                 return sprintf(page, "%llu\n",
3413                                (unsigned long long)mddev->reshape_position);
3414         strcpy(page, "none\n");
3415         return 5;
3416 }
3417
3418 static ssize_t
3419 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3420 {
3421         char *e;
3422         unsigned long long new = simple_strtoull(buf, &e, 10);
3423         if (mddev->pers)
3424                 return -EBUSY;
3425         if (buf == e || (*e && *e != '\n'))
3426                 return -EINVAL;
3427         mddev->reshape_position = new;
3428         mddev->delta_disks = 0;
3429         mddev->new_level = mddev->level;
3430         mddev->new_layout = mddev->layout;
3431         mddev->new_chunk = mddev->chunk_size;
3432         return len;
3433 }
3434
3435 static struct md_sysfs_entry md_reshape_position =
3436 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3437        reshape_position_store);
3438
3439
3440 static struct attribute *md_default_attrs[] = {
3441         &md_level.attr,
3442         &md_layout.attr,
3443         &md_raid_disks.attr,
3444         &md_chunk_size.attr,
3445         &md_size.attr,
3446         &md_resync_start.attr,
3447         &md_metadata.attr,
3448         &md_new_device.attr,
3449         &md_safe_delay.attr,
3450         &md_array_state.attr,
3451         &md_reshape_position.attr,
3452         NULL,
3453 };
3454
3455 static struct attribute *md_redundancy_attrs[] = {
3456         &md_scan_mode.attr,
3457         &md_mismatches.attr,
3458         &md_sync_min.attr,
3459         &md_sync_max.attr,
3460         &md_sync_speed.attr,
3461         &md_sync_force_parallel.attr,
3462         &md_sync_completed.attr,
3463         &md_min_sync.attr,
3464         &md_max_sync.attr,
3465         &md_suspend_lo.attr,
3466         &md_suspend_hi.attr,
3467         &md_bitmap.attr,
3468         &md_degraded.attr,
3469         NULL,
3470 };
3471 static struct attribute_group md_redundancy_group = {
3472         .name = NULL,
3473         .attrs = md_redundancy_attrs,
3474 };
3475
3476
3477 static ssize_t
3478 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3479 {
3480         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3481         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3482         ssize_t rv;
3483
3484         if (!entry->show)
3485                 return -EIO;
3486         rv = mddev_lock(mddev);
3487         if (!rv) {
3488                 rv = entry->show(mddev, page);
3489                 mddev_unlock(mddev);
3490         }
3491         return rv;
3492 }
3493
3494 static ssize_t
3495 md_attr_store(struct kobject *kobj, struct attribute *attr,
3496               const char *page, size_t length)
3497 {
3498         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3499         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3500         ssize_t rv;
3501
3502         if (!entry->store)
3503                 return -EIO;
3504         if (!capable(CAP_SYS_ADMIN))
3505                 return -EACCES;
3506         rv = mddev_lock(mddev);
3507         if (mddev->hold_active == UNTIL_IOCTL)
3508                 mddev->hold_active = 0;
3509         if (!rv) {
3510                 rv = entry->store(mddev, page, length);
3511                 mddev_unlock(mddev);
3512         }
3513         return rv;
3514 }
3515
3516 static void md_free(struct kobject *ko)
3517 {
3518         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3519
3520         if (mddev->sysfs_state)
3521                 sysfs_put(mddev->sysfs_state);
3522
3523         if (mddev->gendisk) {
3524                 del_gendisk(mddev->gendisk);
3525                 put_disk(mddev->gendisk);
3526         }
3527         if (mddev->queue)
3528                 blk_cleanup_queue(mddev->queue);
3529
3530         kfree(mddev);
3531 }
3532
3533 static struct sysfs_ops md_sysfs_ops = {
3534         .show   = md_attr_show,
3535         .store  = md_attr_store,
3536 };
3537 static struct kobj_type md_ktype = {
3538         .release        = md_free,
3539         .sysfs_ops      = &md_sysfs_ops,
3540         .default_attrs  = md_default_attrs,
3541 };
3542
3543 int mdp_major = 0;
3544
3545 static int md_alloc(dev_t dev, char *name)
3546 {
3547         static DEFINE_MUTEX(disks_mutex);
3548         mddev_t *mddev = mddev_find(dev);
3549         struct gendisk *disk;
3550         int partitioned;
3551         int shift;
3552         int unit;
3553         int error;
3554
3555         if (!mddev)
3556                 return -ENODEV;
3557
3558         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3559         shift = partitioned ? MdpMinorShift : 0;
3560         unit = MINOR(mddev->unit) >> shift;
3561
3562         /* wait for any previous instance if this device
3563          * to be completed removed (mddev_delayed_delete).
3564          */
3565         flush_scheduled_work();
3566
3567         mutex_lock(&disks_mutex);
3568         if (mddev->gendisk) {
3569                 mutex_unlock(&disks_mutex);
3570                 mddev_put(mddev);
3571                 return -EEXIST;
3572         }
3573
3574         if (name) {
3575                 /* Need to ensure that 'name' is not a duplicate.
3576                  */
3577                 mddev_t *mddev2;
3578                 spin_lock(&all_mddevs_lock);
3579
3580                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3581                         if (mddev2->gendisk &&
3582                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
3583                                 spin_unlock(&all_mddevs_lock);
3584                                 return -EEXIST;
3585                         }
3586                 spin_unlock(&all_mddevs_lock);
3587         }
3588
3589         mddev->queue = blk_alloc_queue(GFP_KERNEL);
3590         if (!mddev->queue) {
3591                 mutex_unlock(&disks_mutex);
3592                 mddev_put(mddev);
3593                 return -ENOMEM;
3594         }
3595         /* Can be unlocked because the queue is new: no concurrency */
3596         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3597
3598         blk_queue_make_request(mddev->queue, md_fail_request);
3599
3600         disk = alloc_disk(1 << shift);
3601         if (!disk) {
3602                 mutex_unlock(&disks_mutex);
3603                 blk_cleanup_queue(mddev->queue);
3604                 mddev->queue = NULL;
3605                 mddev_put(mddev);
3606                 return -ENOMEM;
3607         }
3608         disk->major = MAJOR(mddev->unit);
3609         disk->first_minor = unit << shift;
3610         if (name)
3611                 strcpy(disk->disk_name, name);
3612         else if (partitioned)
3613                 sprintf(disk->disk_name, "md_d%d", unit);
3614         else
3615                 sprintf(disk->disk_name, "md%d", unit);
3616         disk->fops = &md_fops;
3617         disk->private_data = mddev;
3618         disk->queue = mddev->queue;
3619         /* Allow extended partitions.  This makes the
3620          * 'mdp' device redundant, but we can't really
3621          * remove it now.
3622          */
3623         disk->flags |= GENHD_FL_EXT_DEVT;
3624         add_disk(disk);
3625         mddev->gendisk = disk;
3626         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3627                                      &disk_to_dev(disk)->kobj, "%s", "md");
3628         mutex_unlock(&disks_mutex);
3629         if (error)
3630                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3631                        disk->disk_name);
3632         else {
3633                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3634                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3635         }
3636         mddev_put(mddev);
3637         return 0;
3638 }
3639
3640 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3641 {
3642         md_alloc(dev, NULL);
3643         return NULL;
3644 }
3645
3646 static int add_named_array(const char *val, struct kernel_param *kp)
3647 {
3648         /* val must be "md_*" where * is not all digits.
3649          * We allocate an array with a large free minor number, and
3650          * set the name to val.  val must not already be an active name.
3651          */
3652         int len = strlen(val);
3653         char buf[DISK_NAME_LEN];
3654
3655         while (len && val[len-1] == '\n')
3656                 len--;
3657         if (len >= DISK_NAME_LEN)
3658                 return -E2BIG;
3659         strlcpy(buf, val, len+1);
3660         if (strncmp(buf, "md_", 3) != 0)
3661                 return -EINVAL;
3662         return md_alloc(0, buf);
3663 }
3664
3665 static void md_safemode_timeout(unsigned long data)
3666 {
3667         mddev_t *mddev = (mddev_t *) data;
3668
3669         if (!atomic_read(&mddev->writes_pending)) {
3670                 mddev->safemode = 1;
3671                 if (mddev->external)
3672                         sysfs_notify_dirent(mddev->sysfs_state);
3673         }
3674         md_wakeup_thread(mddev->thread);
3675 }
3676
3677 static int start_dirty_degraded;
3678
3679 static int do_md_run(mddev_t * mddev)
3680 {
3681         int err;
3682         int chunk_size;
3683         mdk_rdev_t *rdev;
3684         struct gendisk *disk;
3685         struct mdk_personality *pers;
3686         char b[BDEVNAME_SIZE];
3687
3688         if (list_empty(&mddev->disks))
3689                 /* cannot run an array with no devices.. */
3690                 return -EINVAL;
3691
3692         if (mddev->pers)
3693                 return -EBUSY;
3694
3695         /*
3696          * Analyze all RAID superblock(s)
3697          */
3698         if (!mddev->raid_disks) {
3699                 if (!mddev->persistent)
3700                         return -EINVAL;
3701                 analyze_sbs(mddev);
3702         }
3703
3704         chunk_size = mddev->chunk_size;
3705
3706         if (chunk_size) {
3707                 if (chunk_size > MAX_CHUNK_SIZE) {
3708                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3709                                 chunk_size, MAX_CHUNK_SIZE);
3710                         return -EINVAL;
3711                 }
3712                 /*
3713                  * chunk-size has to be a power of 2
3714                  */
3715                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3716                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3717                         return -EINVAL;
3718                 }
3719
3720                 /* devices must have minimum size of one chunk */
3721                 list_for_each_entry(rdev, &mddev->disks, same_set) {
3722                         if (test_bit(Faulty, &rdev->flags))
3723                                 continue;
3724                         if (rdev->size < chunk_size / 1024) {
3725                                 printk(KERN_WARNING
3726                                         "md: Dev %s smaller than chunk_size:"
3727                                         " %lluk < %dk\n",
3728                                         bdevname(rdev->bdev,b),
3729                                         (unsigned long long)rdev->size,
3730                                         chunk_size / 1024);
3731                                 return -EINVAL;
3732                         }
3733                 }
3734         }
3735
3736         if (mddev->level != LEVEL_NONE)
3737                 request_module("md-level-%d", mddev->level);
3738         else if (mddev->clevel[0])
3739                 request_module("md-%s", mddev->clevel);
3740
3741         /*
3742          * Drop all container device buffers, from now on
3743          * the only valid external interface is through the md
3744          * device.
3745          */
3746         list_for_each_entry(rdev, &mddev->disks, same_set) {
3747                 if (test_bit(Faulty, &rdev->flags))
3748                         continue;
3749                 sync_blockdev(rdev->bdev);
3750                 invalidate_bdev(rdev->bdev);
3751
3752                 /* perform some consistency tests on the device.
3753                  * We don't want the data to overlap the metadata,
3754                  * Internal Bitmap issues has handled elsewhere.
3755                  */
3756                 if (rdev->data_offset < rdev->sb_start) {
3757                         if (mddev->size &&
3758                             rdev->data_offset + mddev->size*2
3759                             > rdev->sb_start) {
3760                                 printk("md: %s: data overlaps metadata\n",
3761                                        mdname(mddev));
3762                                 return -EINVAL;
3763                         }
3764                 } else {
3765                         if (rdev->sb_start + rdev->sb_size/512
3766                             > rdev->data_offset) {
3767                                 printk("md: %s: metadata overlaps data\n",
3768                                        mdname(mddev));
3769                                 return -EINVAL;
3770                         }
3771                 }
3772                 sysfs_notify_dirent(rdev->sysfs_state);
3773         }
3774
3775         md_probe(mddev->unit, NULL, NULL);
3776         disk = mddev->gendisk;
3777         if (!disk)
3778                 return -ENOMEM;
3779
3780         spin_lock(&pers_lock);
3781         pers = find_pers(mddev->level, mddev->clevel);
3782         if (!pers || !try_module_get(pers->owner)) {
3783                 spin_unlock(&pers_lock);
3784                 if (mddev->level != LEVEL_NONE)
3785                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3786                                mddev->level);
3787                 else
3788                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3789                                mddev->clevel);
3790                 return -EINVAL;
3791         }
3792         mddev->pers = pers;
3793         spin_unlock(&pers_lock);
3794         mddev->level = pers->level;
3795         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3796
3797         if (mddev->reshape_position != MaxSector &&
3798             pers->start_reshape == NULL) {
3799                 /* This personality cannot handle reshaping... */
3800                 mddev->pers = NULL;
3801                 module_put(pers->owner);
3802                 return -EINVAL;
3803         }
3804
3805         if (pers->sync_request) {
3806                 /* Warn if this is a potentially silly
3807                  * configuration.
3808                  */
3809                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3810                 mdk_rdev_t *rdev2;
3811                 int warned = 0;
3812
3813                 list_for_each_entry(rdev, &mddev->disks, same_set)
3814                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
3815                                 if (rdev < rdev2 &&
3816                                     rdev->bdev->bd_contains ==
3817                                     rdev2->bdev->bd_contains) {
3818                                         printk(KERN_WARNING
3819                                                "%s: WARNING: %s appears to be"
3820                                                " on the same physical disk as"
3821                                                " %s.\n",
3822                                                mdname(mddev),
3823                                                bdevname(rdev->bdev,b),
3824                                                bdevname(rdev2->bdev,b2));
3825                                         warned = 1;
3826                                 }
3827                         }
3828
3829                 if (warned)
3830                         printk(KERN_WARNING
3831                                "True protection against single-disk"
3832                                " failure might be compromised.\n");
3833         }
3834
3835         mddev->recovery = 0;
3836         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3837         mddev->barriers_work = 1;
3838         mddev->ok_start_degraded = start_dirty_degraded;
3839
3840         if (start_readonly)
3841                 mddev->ro = 2; /* read-only, but switch on first write */
3842
3843         err = mddev->pers->run(mddev);
3844         if (err)
3845                 printk(KERN_ERR "md: pers->run() failed ...\n");
3846         else if (mddev->pers->sync_request) {
3847                 err = bitmap_create(mddev);
3848                 if (err) {
3849                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3850                                mdname(mddev), err);
3851                         mddev->pers->stop(mddev);
3852                 }
3853         }
3854         if (err) {
3855                 module_put(mddev->pers->owner);
3856                 mddev->pers = NULL;
3857                 bitmap_destroy(mddev);
3858                 return err;
3859         }
3860         if (mddev->pers->sync_request) {
3861                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3862                         printk(KERN_WARNING
3863                                "md: cannot register extra attributes for %s\n",
3864                                mdname(mddev));
3865                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3866         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3867                 mddev->ro = 0;
3868
3869         atomic_set(&mddev->writes_pending,0);
3870         mddev->safemode = 0;
3871         mddev->safemode_timer.function = md_safemode_timeout;
3872         mddev->safemode_timer.data = (unsigned long) mddev;
3873         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3874         mddev->in_sync = 1;
3875
3876         list_for_each_entry(rdev, &mddev->disks, same_set)
3877                 if (rdev->raid_disk >= 0) {
3878                         char nm[20];
3879                         sprintf(nm, "rd%d", rdev->raid_disk);
3880                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3881                                 printk("md: cannot register %s for %s\n",
3882                                        nm, mdname(mddev));
3883                 }
3884
3885         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3886
3887         if (mddev->flags)
3888                 md_update_sb(mddev, 0);
3889
3890         set_capacity(disk, mddev->array_sectors);
3891
3892         /* If we call blk_queue_make_request here, it will
3893          * re-initialise max_sectors etc which may have been
3894          * refined inside -> run.  So just set the bits we need to set.
3895          * Most initialisation happended when we called
3896          * blk_queue_make_request(..., md_fail_request)
3897          * earlier.
3898          */
3899         mddev->queue->queuedata = mddev;
3900         mddev->queue->make_request_fn = mddev->pers->make_request;
3901
3902         /* If there is a partially-recovered drive we need to
3903          * start recovery here.  If we leave it to md_check_recovery,
3904          * it will remove the drives and not do the right thing
3905          */
3906         if (mddev->degraded && !mddev->sync_thread) {
3907                 int spares = 0;
3908                 list_for_each_entry(rdev, &mddev->disks, same_set)
3909                         if (rdev->raid_disk >= 0 &&
3910                             !test_bit(In_sync, &rdev->flags) &&
3911                             !test_bit(Faulty, &rdev->flags))
3912                                 /* complete an interrupted recovery */
3913                                 spares++;
3914                 if (spares && mddev->pers->sync_request) {
3915                         mddev->recovery = 0;
3916                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3917                         mddev->sync_thread = md_register_thread(md_do_sync,
3918                                                                 mddev,
3919                                                                 "%s_resync");
3920                         if (!mddev->sync_thread) {
3921                                 printk(KERN_ERR "%s: could not start resync"
3922                                        " thread...\n",
3923                                        mdname(mddev));
3924                                 /* leave the spares where they are, it shouldn't hurt */
3925                                 mddev->recovery = 0;
3926                         }
3927                 }
3928         }
3929         md_wakeup_thread(mddev->thread);
3930         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3931
3932         mddev->changed = 1;
3933         md_new_event(mddev);
3934         sysfs_notify_dirent(mddev->sysfs_state);
3935         if (mddev->sysfs_action)
3936                 sysfs_notify_dirent(mddev->sysfs_action);
3937         sysfs_notify(&mddev->kobj, NULL, "degraded");
3938         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
3939         return 0;
3940 }
3941
3942 static int restart_array(mddev_t *mddev)
3943 {
3944         struct gendisk *disk = mddev->gendisk;
3945
3946         /* Complain if it has no devices */
3947         if (list_empty(&mddev->disks))
3948                 return -ENXIO;
3949         if (!mddev->pers)
3950                 return -EINVAL;
3951         if (!mddev->ro)
3952                 return -EBUSY;
3953         mddev->safemode = 0;
3954         mddev->ro = 0;
3955         set_disk_ro(disk, 0);
3956         printk(KERN_INFO "md: %s switched to read-write mode.\n",
3957                 mdname(mddev));
3958         /* Kick recovery or resync if necessary */
3959         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3960         md_wakeup_thread(mddev->thread);
3961         md_wakeup_thread(mddev->sync_thread);
3962         sysfs_notify_dirent(mddev->sysfs_state);
3963         return 0;
3964 }
3965
3966 /* similar to deny_write_access, but accounts for our holding a reference
3967  * to the file ourselves */
3968 static int deny_bitmap_write_access(struct file * file)
3969 {
3970         struct inode *inode = file->f_mapping->host;
3971
3972         spin_lock(&inode->i_lock);
3973         if (atomic_read(&inode->i_writecount) > 1) {
3974                 spin_unlock(&inode->i_lock);
3975                 return -ETXTBSY;
3976         }
3977         atomic_set(&inode->i_writecount, -1);
3978         spin_unlock(&inode->i_lock);
3979
3980         return 0;
3981 }
3982
3983 static void restore_bitmap_write_access(struct file *file)
3984 {
3985         struct inode *inode = file->f_mapping->host;
3986
3987         spin_lock(&inode->i_lock);
3988         atomic_set(&inode->i_writecount, 1);
3989         spin_unlock(&inode->i_lock);
3990 }
3991
3992 /* mode:
3993  *   0 - completely stop and dis-assemble array
3994  *   1 - switch to readonly
3995  *   2 - stop but do not disassemble array
3996  */
3997 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3998 {
3999         int err = 0;
4000         struct gendisk *disk = mddev->gendisk;
4001
4002         if (atomic_read(&mddev->openers) > is_open) {
4003                 printk("md: %s still in use.\n",mdname(mddev));
4004                 return -EBUSY;
4005         }
4006
4007         if (mddev->pers) {
4008
4009                 if (mddev->sync_thread) {
4010                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4011                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4012                         md_unregister_thread(mddev->sync_thread);
4013                         mddev->sync_thread = NULL;
4014                 }
4015
4016                 del_timer_sync(&mddev->safemode_timer);
4017
4018                 switch(mode) {
4019                 case 1: /* readonly */
4020                         err  = -ENXIO;
4021                         if (mddev->ro==1)
4022                                 goto out;
4023                         mddev->ro = 1;
4024                         break;
4025                 case 0: /* disassemble */
4026                 case 2: /* stop */
4027                         bitmap_flush(mddev);
4028                         md_super_wait(mddev);
4029                         if (mddev->ro)
4030                                 set_disk_ro(disk, 0);
4031                         blk_queue_make_request(mddev->queue, md_fail_request);
4032                         mddev->pers->stop(mddev);
4033                         mddev->queue->merge_bvec_fn = NULL;
4034                         mddev->queue->unplug_fn = NULL;
4035                         mddev->queue->backing_dev_info.congested_fn = NULL;
4036                         if (mddev->pers->sync_request) {
4037                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4038                                 if (mddev->sysfs_action)
4039                                         sysfs_put(mddev->sysfs_action);
4040                                 mddev->sysfs_action = NULL;
4041                         }
4042                         module_put(mddev->pers->owner);
4043                         mddev->pers = NULL;
4044                         /* tell userspace to handle 'inactive' */
4045                         sysfs_notify_dirent(mddev->sysfs_state);
4046
4047                         set_capacity(disk, 0);
4048                         mddev->changed = 1;
4049
4050                         if (mddev->ro)
4051                                 mddev->ro = 0;
4052                 }
4053                 if (!mddev->in_sync || mddev->flags) {
4054                         /* mark array as shutdown cleanly */
4055                         mddev->in_sync = 1;
4056                         md_update_sb(mddev, 1);
4057                 }
4058                 if (mode == 1)
4059                         set_disk_ro(disk, 1);
4060                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4061         }
4062
4063         /*
4064          * Free resources if final stop
4065          */
4066         if (mode == 0) {
4067                 mdk_rdev_t *rdev;
4068
4069                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4070
4071                 bitmap_destroy(mddev);
4072                 if (mddev->bitmap_file) {
4073                         restore_bitmap_write_access(mddev->bitmap_file);
4074                         fput(mddev->bitmap_file);
4075                         mddev->bitmap_file = NULL;
4076                 }
4077                 mddev->bitmap_offset = 0;
4078
4079                 list_for_each_entry(rdev, &mddev->disks, same_set)
4080                         if (rdev->raid_disk >= 0) {
4081                                 char nm[20];
4082                                 sprintf(nm, "rd%d", rdev->raid_disk);
4083                                 sysfs_remove_link(&mddev->kobj, nm);
4084                         }
4085
4086                 /* make sure all md_delayed_delete calls have finished */
4087                 flush_scheduled_work();
4088
4089                 export_array(mddev);
4090
4091                 mddev->array_sectors = 0;
4092                 mddev->size = 0;
4093                 mddev->raid_disks = 0;
4094                 mddev->recovery_cp = 0;
4095                 mddev->resync_min = 0;
4096                 mddev->resync_max = MaxSector;
4097                 mddev->reshape_position = MaxSector;
4098                 mddev->external = 0;
4099                 mddev->persistent = 0;
4100                 mddev->level = LEVEL_NONE;
4101                 mddev->clevel[0] = 0;
4102                 mddev->flags = 0;
4103                 mddev->ro = 0;
4104                 mddev->metadata_type[0] = 0;
4105                 mddev->chunk_size = 0;
4106                 mddev->ctime = mddev->utime = 0;
4107                 mddev->layout = 0;
4108                 mddev->max_disks = 0;
4109                 mddev->events = 0;
4110                 mddev->delta_disks = 0;
4111                 mddev->new_level = LEVEL_NONE;
4112                 mddev->new_layout = 0;
4113                 mddev->new_chunk = 0;
4114                 mddev->curr_resync = 0;
4115                 mddev->resync_mismatches = 0;
4116                 mddev->suspend_lo = mddev->suspend_hi = 0;
4117                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4118                 mddev->recovery = 0;
4119                 mddev->in_sync = 0;
4120                 mddev->changed = 0;
4121                 mddev->degraded = 0;
4122                 mddev->barriers_work = 0;
4123                 mddev->safemode = 0;
4124                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4125                 if (mddev->hold_active == UNTIL_STOP)
4126                         mddev->hold_active = 0;
4127
4128         } else if (mddev->pers)
4129                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4130                         mdname(mddev));
4131         err = 0;
4132         md_new_event(mddev);
4133         sysfs_notify_dirent(mddev->sysfs_state);
4134 out:
4135         return err;
4136 }
4137
4138 #ifndef MODULE
4139 static void autorun_array(mddev_t *mddev)
4140 {
4141         mdk_rdev_t *rdev;
4142         int err;
4143
4144         if (list_empty(&mddev->disks))
4145                 return;
4146
4147         printk(KERN_INFO "md: running: ");
4148
4149         list_for_each_entry(rdev, &mddev->disks, same_set) {
4150                 char b[BDEVNAME_SIZE];
4151                 printk("<%s>", bdevname(rdev->bdev,b));
4152         }
4153         printk("\n");
4154
4155         err = do_md_run(mddev);
4156         if (err) {
4157                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4158                 do_md_stop(mddev, 0, 0);
4159         }
4160 }
4161
4162 /*
4163  * lets try to run arrays based on all disks that have arrived
4164  * until now. (those are in pending_raid_disks)
4165  *
4166  * the method: pick the first pending disk, collect all disks with
4167  * the same UUID, remove all from the pending list and put them into
4168  * the 'same_array' list. Then order this list based on superblock
4169  * update time (freshest comes first), kick out 'old' disks and
4170  * compare superblocks. If everything's fine then run it.
4171  *
4172  * If "unit" is allocated, then bump its reference count
4173  */
4174 static void autorun_devices(int part)
4175 {
4176         mdk_rdev_t *rdev0, *rdev, *tmp;
4177         mddev_t *mddev;
4178         char b[BDEVNAME_SIZE];
4179
4180         printk(KERN_INFO "md: autorun ...\n");
4181         while (!list_empty(&pending_raid_disks)) {
4182                 int unit;
4183                 dev_t dev;
4184                 LIST_HEAD(candidates);
4185                 rdev0 = list_entry(pending_raid_disks.next,
4186                                          mdk_rdev_t, same_set);
4187
4188                 printk(KERN_INFO "md: considering %s ...\n",
4189                         bdevname(rdev0->bdev,b));
4190                 INIT_LIST_HEAD(&candidates);
4191                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4192                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4193                                 printk(KERN_INFO "md:  adding %s ...\n",
4194                                         bdevname(rdev->bdev,b));
4195                                 list_move(&rdev->same_set, &candidates);
4196                         }
4197                 /*
4198                  * now we have a set of devices, with all of them having
4199                  * mostly sane superblocks. It's time to allocate the
4200                  * mddev.
4201                  */
4202                 if (part) {
4203                         dev = MKDEV(mdp_major,
4204                                     rdev0->preferred_minor << MdpMinorShift);
4205                         unit = MINOR(dev) >> MdpMinorShift;
4206                 } else {
4207                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4208                         unit = MINOR(dev);
4209                 }
4210                 if (rdev0->preferred_minor != unit) {
4211                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4212                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4213                         break;
4214                 }
4215
4216                 md_probe(dev, NULL, NULL);
4217                 mddev = mddev_find(dev);
4218                 if (!mddev || !mddev->gendisk) {
4219                         if (mddev)
4220                                 mddev_put(mddev);
4221                         printk(KERN_ERR
4222                                 "md: cannot allocate memory for md drive.\n");
4223                         break;
4224                 }
4225                 if (mddev_lock(mddev))
4226                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4227                                mdname(mddev));
4228                 else if (mddev->raid_disks || mddev->major_version
4229                          || !list_empty(&mddev->disks)) {
4230                         printk(KERN_WARNING
4231                                 "md: %s already running, cannot run %s\n",
4232                                 mdname(mddev), bdevname(rdev0->bdev,b));
4233                         mddev_unlock(mddev);
4234                 } else {
4235                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4236                         mddev->persistent = 1;
4237                         rdev_for_each_list(rdev, tmp, &candidates) {
4238                                 list_del_init(&rdev->same_set);
4239                                 if (bind_rdev_to_array(rdev, mddev))
4240                                         export_rdev(rdev);
4241                         }
4242                         autorun_array(mddev);
4243                         mddev_unlock(mddev);
4244                 }
4245                 /* on success, candidates will be empty, on error
4246                  * it won't...
4247                  */
4248                 rdev_for_each_list(rdev, tmp, &candidates) {
4249                         list_del_init(&rdev->same_set);
4250                         export_rdev(rdev);
4251                 }
4252                 mddev_put(mddev);
4253         }
4254         printk(KERN_INFO "md: ... autorun DONE.\n");
4255 }
4256 #endif /* !MODULE */
4257
4258 static int get_version(void __user * arg)
4259 {
4260         mdu_version_t ver;
4261
4262         ver.major = MD_MAJOR_VERSION;
4263         ver.minor = MD_MINOR_VERSION;
4264         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4265
4266         if (copy_to_user(arg, &ver, sizeof(ver)))
4267                 return -EFAULT;
4268
4269         return 0;
4270 }
4271
4272 static int get_array_info(mddev_t * mddev, void __user * arg)
4273 {
4274         mdu_array_info_t info;
4275         int nr,working,active,failed,spare;
4276         mdk_rdev_t *rdev;
4277
4278         nr=working=active=failed=spare=0;
4279         list_for_each_entry(rdev, &mddev->disks, same_set) {
4280                 nr++;
4281                 if (test_bit(Faulty, &rdev->flags))
4282                         failed++;
4283                 else {
4284                         working++;
4285                         if (test_bit(In_sync, &rdev->flags))
4286                                 active++;
4287                         else
4288                                 spare++;
4289                 }
4290         }
4291
4292         info.major_version = mddev->major_version;
4293         info.minor_version = mddev->minor_version;
4294         info.patch_version = MD_PATCHLEVEL_VERSION;
4295         info.ctime         = mddev->ctime;
4296         info.level         = mddev->level;
4297         info.size          = mddev->size;
4298         if (info.size != mddev->size) /* overflow */
4299                 info.size = -1;
4300         info.nr_disks      = nr;
4301         info.raid_disks    = mddev->raid_disks;
4302         info.md_minor      = mddev->md_minor;
4303         info.not_persistent= !mddev->persistent;
4304
4305         info.utime         = mddev->utime;
4306         info.state         = 0;
4307         if (mddev->in_sync)
4308                 info.state = (1<<MD_SB_CLEAN);
4309         if (mddev->bitmap && mddev->bitmap_offset)
4310                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4311         info.active_disks  = active;
4312         info.working_disks = working;
4313         info.failed_disks  = failed;
4314         info.spare_disks   = spare;
4315
4316         info.layout        = mddev->layout;
4317         info.chunk_size    = mddev->chunk_size;
4318
4319         if (copy_to_user(arg, &info, sizeof(info)))
4320                 return -EFAULT;
4321
4322         return 0;
4323 }
4324
4325 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4326 {
4327         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4328         char *ptr, *buf = NULL;
4329         int err = -ENOMEM;
4330
4331         if (md_allow_write(mddev))
4332                 file = kmalloc(sizeof(*file), GFP_NOIO);
4333         else
4334                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4335
4336         if (!file)
4337                 goto out;
4338
4339         /* bitmap disabled, zero the first byte and copy out */
4340         if (!mddev->bitmap || !mddev->bitmap->file) {
4341                 file->pathname[0] = '\0';
4342                 goto copy_out;
4343         }
4344
4345         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4346         if (!buf)
4347                 goto out;
4348
4349         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4350         if (IS_ERR(ptr))
4351                 goto out;
4352
4353         strcpy(file->pathname, ptr);
4354
4355 copy_out:
4356         err = 0;
4357         if (copy_to_user(arg, file, sizeof(*file)))
4358                 err = -EFAULT;
4359 out:
4360         kfree(buf);
4361         kfree(file);
4362         return err;
4363 }
4364
4365 static int get_disk_info(mddev_t * mddev, void __user * arg)
4366 {
4367         mdu_disk_info_t info;
4368         mdk_rdev_t *rdev;
4369
4370         if (copy_from_user(&info, arg, sizeof(info)))
4371                 return -EFAULT;
4372
4373         rdev = find_rdev_nr(mddev, info.number);
4374         if (rdev) {
4375                 info.major = MAJOR(rdev->bdev->bd_dev);
4376                 info.minor = MINOR(rdev->bdev->bd_dev);
4377                 info.raid_disk = rdev->raid_disk;
4378                 info.state = 0;
4379                 if (test_bit(Faulty, &rdev->flags))
4380                         info.state |= (1<<MD_DISK_FAULTY);
4381                 else if (test_bit(In_sync, &rdev->flags)) {
4382                         info.state |= (1<<MD_DISK_ACTIVE);
4383                         info.state |= (1<<MD_DISK_SYNC);
4384                 }
4385                 if (test_bit(WriteMostly, &rdev->flags))
4386                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4387         } else {
4388                 info.major = info.minor = 0;
4389                 info.raid_disk = -1;
4390                 info.state = (1<<MD_DISK_REMOVED);
4391         }
4392
4393         if (copy_to_user(arg, &info, sizeof(info)))
4394                 return -EFAULT;
4395
4396         return 0;
4397 }
4398
4399 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4400 {
4401         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4402         mdk_rdev_t *rdev;
4403         dev_t dev = MKDEV(info->major,info->minor);
4404
4405         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4406                 return -EOVERFLOW;
4407
4408         if (!mddev->raid_disks) {
4409                 int err;
4410                 /* expecting a device which has a superblock */
4411                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4412                 if (IS_ERR(rdev)) {
4413                         printk(KERN_WARNING
4414                                 "md: md_import_device returned %ld\n",
4415                                 PTR_ERR(rdev));
4416                         return PTR_ERR(rdev);
4417                 }
4418                 if (!list_empty(&mddev->disks)) {
4419                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4420                                                         mdk_rdev_t, same_set);
4421                         int err = super_types[mddev->major_version]
4422                                 .load_super(rdev, rdev0, mddev->minor_version);
4423                         if (err < 0) {
4424                                 printk(KERN_WARNING
4425                                         "md: %s has different UUID to %s\n",
4426                                         bdevname(rdev->bdev,b),
4427                                         bdevname(rdev0->bdev,b2));
4428                                 export_rdev(rdev);
4429                                 return -EINVAL;
4430                         }
4431                 }
4432                 err = bind_rdev_to_array(rdev, mddev);
4433                 if (err)
4434                         export_rdev(rdev);
4435                 return err;
4436         }
4437
4438         /*
4439          * add_new_disk can be used once the array is assembled
4440          * to add "hot spares".  They must already have a superblock
4441          * written
4442          */
4443         if (mddev->pers) {
4444                 int err;
4445                 if (!mddev->pers->hot_add_disk) {
4446                         printk(KERN_WARNING
4447                                 "%s: personality does not support diskops!\n",
4448                                mdname(mddev));
4449                         return -EINVAL;
4450                 }
4451                 if (mddev->persistent)
4452                         rdev = md_import_device(dev, mddev->major_version,
4453                                                 mddev->minor_version);
4454                 else
4455                         rdev = md_import_device(dev, -1, -1);
4456                 if (IS_ERR(rdev)) {
4457                         printk(KERN_WARNING
4458                                 "md: md_import_device returned %ld\n",
4459                                 PTR_ERR(rdev));
4460                         return PTR_ERR(rdev);
4461                 }
4462                 /* set save_raid_disk if appropriate */
4463                 if (!mddev->persistent) {
4464                         if (info->state & (1<<MD_DISK_SYNC)  &&
4465                             info->raid_disk < mddev->raid_disks)
4466                                 rdev->raid_disk = info->raid_disk;
4467                         else
4468                                 rdev->raid_disk = -1;
4469                 } else
4470                         super_types[mddev->major_version].
4471                                 validate_super(mddev, rdev);
4472                 rdev->saved_raid_disk = rdev->raid_disk;
4473
4474                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4475                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4476                         set_bit(WriteMostly, &rdev->flags);
4477
4478                 rdev->raid_disk = -1;
4479                 err = bind_rdev_to_array(rdev, mddev);
4480                 if (!err && !mddev->pers->hot_remove_disk) {
4481                         /* If there is hot_add_disk but no hot_remove_disk
4482                          * then added disks for geometry changes,
4483                          * and should be added immediately.
4484                          */
4485                         super_types[mddev->major_version].
4486                                 validate_super(mddev, rdev);
4487                         err = mddev->pers->hot_add_disk(mddev, rdev);
4488                         if (err)
4489                                 unbind_rdev_from_array(rdev);
4490                 }
4491                 if (err)
4492                         export_rdev(rdev);
4493                 else
4494                         sysfs_notify_dirent(rdev->sysfs_state);
4495
4496                 md_update_sb(mddev, 1);
4497                 if (mddev->degraded)
4498                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4499                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4500                 md_wakeup_thread(mddev->thread);
4501                 return err;
4502         }
4503
4504         /* otherwise, add_new_disk is only allowed
4505          * for major_version==0 superblocks
4506          */
4507         if (mddev->major_version != 0) {
4508                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4509                        mdname(mddev));
4510                 return -EINVAL;
4511         }
4512
4513         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4514                 int err;
4515                 rdev = md_import_device(dev, -1, 0);
4516                 if (IS_ERR(rdev)) {
4517                         printk(KERN_WARNING
4518                                 "md: error, md_import_device() returned %ld\n",
4519                                 PTR_ERR(rdev));
4520                         return PTR_ERR(rdev);
4521                 }
4522                 rdev->desc_nr = info->number;
4523                 if (info->raid_disk < mddev->raid_disks)
4524                         rdev->raid_disk = info->raid_disk;
4525                 else
4526                         rdev->raid_disk = -1;
4527
4528                 if (rdev->raid_disk < mddev->raid_disks)
4529                         if (info->state & (1<<MD_DISK_SYNC))
4530                                 set_bit(In_sync, &rdev->flags);
4531
4532                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4533                         set_bit(WriteMostly, &rdev->flags);
4534
4535                 if (!mddev->persistent) {
4536                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4537                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4538                 } else
4539                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4540                 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4541
4542                 err = bind_rdev_to_array(rdev, mddev);
4543                 if (err) {
4544                         export_rdev(rdev);
4545                         return err;
4546                 }
4547         }
4548
4549         return 0;
4550 }
4551
4552 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4553 {
4554         char b[BDEVNAME_SIZE];
4555         mdk_rdev_t *rdev;
4556
4557         rdev = find_rdev(mddev, dev);
4558         if (!rdev)
4559                 return -ENXIO;
4560
4561         if (rdev->raid_disk >= 0)
4562                 goto busy;
4563
4564         kick_rdev_from_array(rdev);
4565         md_update_sb(mddev, 1);
4566         md_new_event(mddev);
4567
4568         return 0;
4569 busy:
4570         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4571                 bdevname(rdev->bdev,b), mdname(mddev));
4572         return -EBUSY;
4573 }
4574
4575 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4576 {
4577         char b[BDEVNAME_SIZE];
4578         int err;
4579         mdk_rdev_t *rdev;
4580
4581         if (!mddev->pers)
4582                 return -ENODEV;
4583
4584         if (mddev->major_version != 0) {
4585                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4586                         " version-0 superblocks.\n",
4587                         mdname(mddev));
4588                 return -EINVAL;
4589         }
4590         if (!mddev->pers->hot_add_disk) {
4591                 printk(KERN_WARNING
4592                         "%s: personality does not support diskops!\n",
4593                         mdname(mddev));
4594                 return -EINVAL;
4595         }
4596
4597         rdev = md_import_device(dev, -1, 0);
4598         if (IS_ERR(rdev)) {
4599                 printk(KERN_WARNING
4600                         "md: error, md_import_device() returned %ld\n",
4601                         PTR_ERR(rdev));
4602                 return -EINVAL;
4603         }
4604
4605         if (mddev->persistent)
4606                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4607         else
4608                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4609
4610         rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4611
4612         if (test_bit(Faulty, &rdev->flags)) {
4613                 printk(KERN_WARNING
4614                         "md: can not hot-add faulty %s disk to %s!\n",
4615                         bdevname(rdev->bdev,b), mdname(mddev));
4616                 err = -EINVAL;
4617                 goto abort_export;
4618         }
4619         clear_bit(In_sync, &rdev->flags);
4620         rdev->desc_nr = -1;
4621         rdev->saved_raid_disk = -1;
4622         err = bind_rdev_to_array(rdev, mddev);
4623         if (err)
4624                 goto abort_export;
4625
4626         /*
4627          * The rest should better be atomic, we can have disk failures
4628          * noticed in interrupt contexts ...
4629          */
4630
4631         rdev->raid_disk = -1;
4632
4633         md_update_sb(mddev, 1);
4634
4635         /*
4636          * Kick recovery, maybe this spare has to be added to the
4637          * array immediately.
4638          */
4639         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4640         md_wakeup_thread(mddev->thread);
4641         md_new_event(mddev);
4642         return 0;
4643
4644 abort_export:
4645         export_rdev(rdev);
4646         return err;
4647 }
4648
4649 static int set_bitmap_file(mddev_t *mddev, int fd)
4650 {
4651         int err;
4652
4653         if (mddev->pers) {
4654                 if (!mddev->pers->quiesce)
4655                         return -EBUSY;
4656                 if (mddev->recovery || mddev->sync_thread)
4657                         return -EBUSY;
4658                 /* we should be able to change the bitmap.. */
4659         }
4660
4661
4662         if (fd >= 0) {
4663                 if (mddev->bitmap)
4664                         return -EEXIST; /* cannot add when bitmap is present */
4665                 mddev->bitmap_file = fget(fd);
4666
4667                 if (mddev->bitmap_file == NULL) {
4668                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4669                                mdname(mddev));
4670                         return -EBADF;
4671                 }
4672
4673                 err = deny_bitmap_write_access(mddev->bitmap_file);
4674                 if (err) {
4675                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4676                                mdname(mddev));
4677                         fput(mddev->bitmap_file);
4678                         mddev->bitmap_file = NULL;
4679                         return err;
4680                 }
4681                 mddev->bitmap_offset = 0; /* file overrides offset */
4682         } else if (mddev->bitmap == NULL)
4683                 return -ENOENT; /* cannot remove what isn't there */
4684         err = 0;
4685         if (mddev->pers) {
4686                 mddev->pers->quiesce(mddev, 1);
4687                 if (fd >= 0)
4688                         err = bitmap_create(mddev);
4689                 if (fd < 0 || err) {
4690                         bitmap_destroy(mddev);
4691                         fd = -1; /* make sure to put the file */
4692                 }
4693                 mddev->pers->quiesce(mddev, 0);
4694         }
4695         if (fd < 0) {
4696                 if (mddev->bitmap_file) {
4697                         restore_bitmap_write_access(mddev->bitmap_file);
4698                         fput(mddev->bitmap_file);
4699                 }
4700                 mddev->bitmap_file = NULL;
4701         }
4702
4703         return err;
4704 }
4705
4706 /*
4707  * set_array_info is used two different ways
4708  * The original usage is when creating a new array.
4709  * In this usage, raid_disks is > 0 and it together with
4710  *  level, size, not_persistent,layout,chunksize determine the
4711  *  shape of the array.
4712  *  This will always create an array with a type-0.90.0 superblock.
4713  * The newer usage is when assembling an array.
4714  *  In this case raid_disks will be 0, and the major_version field is
4715  *  use to determine which style super-blocks are to be found on the devices.
4716  *  The minor and patch _version numbers are also kept incase the
4717  *  super_block handler wishes to interpret them.
4718  */
4719 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4720 {
4721
4722         if (info->raid_disks == 0) {
4723                 /* just setting version number for superblock loading */
4724                 if (info->major_version < 0 ||
4725                     info->major_version >= ARRAY_SIZE(super_types) ||
4726                     super_types[info->major_version].name == NULL) {
4727                         /* maybe try to auto-load a module? */
4728                         printk(KERN_INFO
4729                                 "md: superblock version %d not known\n",
4730                                 info->major_version);
4731                         return -EINVAL;
4732                 }
4733                 mddev->major_version = info->major_version;
4734                 mddev->minor_version = info->minor_version;
4735                 mddev->patch_version = info->patch_version;
4736                 mddev->persistent = !info->not_persistent;
4737                 return 0;
4738         }
4739         mddev->major_version = MD_MAJOR_VERSION;
4740         mddev->minor_version = MD_MINOR_VERSION;
4741         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4742         mddev->ctime         = get_seconds();
4743
4744         mddev->level         = info->level;
4745         mddev->clevel[0]     = 0;
4746         mddev->size          = info->size;
4747         mddev->raid_disks    = info->raid_disks;
4748         /* don't set md_minor, it is determined by which /dev/md* was
4749          * openned
4750          */
4751         if (info->state & (1<<MD_SB_CLEAN))
4752                 mddev->recovery_cp = MaxSector;
4753         else
4754                 mddev->recovery_cp = 0;
4755         mddev->persistent    = ! info->not_persistent;
4756         mddev->external      = 0;
4757
4758         mddev->layout        = info->layout;
4759         mddev->chunk_size    = info->chunk_size;
4760
4761         mddev->max_disks     = MD_SB_DISKS;
4762
4763         if (mddev->persistent)
4764                 mddev->flags         = 0;
4765         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4766
4767         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4768         mddev->bitmap_offset = 0;
4769
4770         mddev->reshape_position = MaxSector;
4771
4772         /*
4773          * Generate a 128 bit UUID
4774          */
4775         get_random_bytes(mddev->uuid, 16);
4776
4777         mddev->new_level = mddev->level;
4778         mddev->new_chunk = mddev->chunk_size;
4779         mddev->new_layout = mddev->layout;
4780         mddev->delta_disks = 0;
4781
4782         return 0;
4783 }
4784
4785 static int update_size(mddev_t *mddev, sector_t num_sectors)
4786 {
4787         mdk_rdev_t *rdev;
4788         int rv;
4789         int fit = (num_sectors == 0);
4790
4791         if (mddev->pers->resize == NULL)
4792                 return -EINVAL;
4793         /* The "num_sectors" is the number of sectors of each device that
4794          * is used.  This can only make sense for arrays with redundancy.
4795          * linear and raid0 always use whatever space is available. We can only
4796          * consider changing this number if no resync or reconstruction is
4797          * happening, and if the new size is acceptable. It must fit before the
4798          * sb_start or, if that is <data_offset, it must fit before the size
4799          * of each device.  If num_sectors is zero, we find the largest size
4800          * that fits.
4801
4802          */
4803         if (mddev->sync_thread)
4804                 return -EBUSY;
4805         if (mddev->bitmap)
4806                 /* Sorry, cannot grow a bitmap yet, just remove it,
4807                  * grow, and re-add.
4808                  */
4809                 return -EBUSY;
4810         list_for_each_entry(rdev, &mddev->disks, same_set) {
4811                 sector_t avail;
4812                 avail = rdev->size * 2;
4813
4814                 if (fit && (num_sectors == 0 || num_sectors > avail))
4815                         num_sectors = avail;
4816                 if (avail < num_sectors)
4817                         return -ENOSPC;
4818         }
4819         rv = mddev->pers->resize(mddev, num_sectors);
4820         if (!rv) {
4821                 struct block_device *bdev;
4822
4823                 bdev = bdget_disk(mddev->gendisk, 0);
4824                 if (bdev) {
4825                         mutex_lock(&bdev->bd_inode->i_mutex);
4826                         i_size_write(bdev->bd_inode,
4827                                      (loff_t)mddev->array_sectors << 9);
4828                         mutex_unlock(&bdev->bd_inode->i_mutex);
4829                         bdput(bdev);
4830                 }
4831         }
4832         return rv;
4833 }
4834
4835 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4836 {
4837         int rv;
4838         /* change the number of raid disks */
4839         if (mddev->pers->check_reshape == NULL)
4840                 return -EINVAL;
4841         if (raid_disks <= 0 ||
4842             raid_disks >= mddev->max_disks)
4843                 return -EINVAL;
4844         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4845                 return -EBUSY;
4846         mddev->delta_disks = raid_disks - mddev->raid_disks;
4847
4848         rv = mddev->pers->check_reshape(mddev);
4849         return rv;
4850 }
4851
4852
4853 /*
4854  * update_array_info is used to change the configuration of an
4855  * on-line array.
4856  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4857  * fields in the info are checked against the array.
4858  * Any differences that cannot be handled will cause an error.
4859  * Normally, only one change can be managed at a time.
4860  */
4861 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4862 {
4863         int rv = 0;
4864         int cnt = 0;
4865         int state = 0;
4866
4867         /* calculate expected state,ignoring low bits */
4868         if (mddev->bitmap && mddev->bitmap_offset)
4869                 state |= (1 << MD_SB_BITMAP_PRESENT);
4870
4871         if (mddev->major_version != info->major_version ||
4872             mddev->minor_version != info->minor_version ||
4873 /*          mddev->patch_version != info->patch_version || */
4874             mddev->ctime         != info->ctime         ||
4875             mddev->level         != info->level         ||
4876 /*          mddev->layout        != info->layout        || */
4877             !mddev->persistent   != info->not_persistent||
4878             mddev->chunk_size    != info->chunk_size    ||
4879             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4880             ((state^info->state) & 0xfffffe00)
4881                 )
4882                 return -EINVAL;
4883         /* Check there is only one change */
4884         if (info->size >= 0 && mddev->size != info->size) cnt++;
4885         if (mddev->raid_disks != info->raid_disks) cnt++;
4886         if (mddev->layout != info->layout) cnt++;
4887         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4888         if (cnt == 0) return 0;
4889         if (cnt > 1) return -EINVAL;
4890
4891         if (mddev->layout != info->layout) {
4892                 /* Change layout
4893                  * we don't need to do anything at the md level, the
4894                  * personality will take care of it all.
4895                  */
4896                 if (mddev->pers->reconfig == NULL)
4897                         return -EINVAL;
4898                 else
4899                         return mddev->pers->reconfig(mddev, info->layout, -1);
4900         }
4901         if (info->size >= 0 && mddev->size != info->size)
4902                 rv = update_size(mddev, (sector_t)info->size * 2);
4903
4904         if (mddev->raid_disks    != info->raid_disks)
4905                 rv = update_raid_disks(mddev, info->raid_disks);
4906
4907         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4908                 if (mddev->pers->quiesce == NULL)
4909                         return -EINVAL;
4910                 if (mddev->recovery || mddev->sync_thread)
4911                         return -EBUSY;
4912                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4913                         /* add the bitmap */
4914                         if (mddev->bitmap)
4915                                 return -EEXIST;
4916                         if (mddev->default_bitmap_offset == 0)
4917                                 return -EINVAL;
4918                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4919                         mddev->pers->quiesce(mddev, 1);
4920                         rv = bitmap_create(mddev);
4921                         if (rv)
4922                                 bitmap_destroy(mddev);
4923                         mddev->pers->quiesce(mddev, 0);
4924                 } else {
4925                         /* remove the bitmap */
4926                         if (!mddev->bitmap)
4927                                 return -ENOENT;
4928                         if (mddev->bitmap->file)
4929                                 return -EINVAL;
4930                         mddev->pers->quiesce(mddev, 1);
4931                         bitmap_destroy(mddev);
4932                         mddev->pers->quiesce(mddev, 0);
4933                         mddev->bitmap_offset = 0;
4934                 }
4935         }
4936         md_update_sb(mddev, 1);
4937         return rv;
4938 }
4939
4940 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4941 {
4942         mdk_rdev_t *rdev;
4943
4944         if (mddev->pers == NULL)
4945                 return -ENODEV;
4946
4947         rdev = find_rdev(mddev, dev);
4948         if (!rdev)
4949                 return -ENODEV;
4950
4951         md_error(mddev, rdev);
4952         return 0;
4953 }
4954
4955 /*
4956  * We have a problem here : there is no easy way to give a CHS
4957  * virtual geometry. We currently pretend that we have a 2 heads
4958  * 4 sectors (with a BIG number of cylinders...). This drives
4959  * dosfs just mad... ;-)
4960  */
4961 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4962 {
4963         mddev_t *mddev = bdev->bd_disk->private_data;
4964
4965         geo->heads = 2;
4966         geo->sectors = 4;
4967         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4968         return 0;
4969 }
4970
4971 static int md_ioctl(struct block_device *bdev, fmode_t mode,
4972                         unsigned int cmd, unsigned long arg)
4973 {
4974         int err = 0;
4975         void __user *argp = (void __user *)arg;
4976         mddev_t *mddev = NULL;
4977
4978         if (!capable(CAP_SYS_ADMIN))
4979                 return -EACCES;
4980
4981         /*
4982          * Commands dealing with the RAID driver but not any
4983          * particular array:
4984          */
4985         switch (cmd)
4986         {
4987                 case RAID_VERSION:
4988                         err = get_version(argp);
4989                         goto done;
4990
4991                 case PRINT_RAID_DEBUG:
4992                         err = 0;
4993                         md_print_devices();
4994                         goto done;
4995
4996 #ifndef MODULE
4997                 case RAID_AUTORUN:
4998                         err = 0;
4999                         autostart_arrays(arg);
5000                         goto done;
5001 #endif
5002                 default:;
5003         }
5004
5005         /*
5006          * Commands creating/starting a new array:
5007          */
5008
5009         mddev = bdev->bd_disk->private_data;
5010
5011         if (!mddev) {
5012                 BUG();
5013                 goto abort;
5014         }
5015
5016         err = mddev_lock(mddev);
5017         if (err) {
5018                 printk(KERN_INFO
5019                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5020                         err, cmd);
5021                 goto abort;
5022         }
5023
5024         switch (cmd)
5025         {
5026                 case SET_ARRAY_INFO:
5027                         {
5028                                 mdu_array_info_t info;
5029                                 if (!arg)
5030                                         memset(&info, 0, sizeof(info));
5031                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5032                                         err = -EFAULT;
5033                                         goto abort_unlock;
5034                                 }
5035                                 if (mddev->pers) {
5036                                         err = update_array_info(mddev, &info);
5037                                         if (err) {
5038                                                 printk(KERN_WARNING "md: couldn't update"
5039                                                        " array info. %d\n", err);
5040                                                 goto abort_unlock;
5041                                         }
5042                                         goto done_unlock;
5043                                 }
5044                                 if (!list_empty(&mddev->disks)) {
5045                                         printk(KERN_WARNING
5046                                                "md: array %s already has disks!\n",
5047                                                mdname(mddev));
5048                                         err = -EBUSY;
5049                                         goto abort_unlock;
5050                                 }
5051                                 if (mddev->raid_disks) {
5052                                         printk(KERN_WARNING
5053                                                "md: array %s already initialised!\n",
5054                                                mdname(mddev));
5055                                         err = -EBUSY;
5056                                         goto abort_unlock;
5057                                 }
5058                                 err = set_array_info(mddev, &info);
5059                                 if (err) {
5060                                         printk(KERN_WARNING "md: couldn't set"
5061                                                " array info. %d\n", err);
5062                                         goto abort_unlock;
5063                                 }
5064                         }
5065                         goto done_unlock;
5066
5067                 default:;
5068         }
5069
5070         /*
5071          * Commands querying/configuring an existing array:
5072          */
5073         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5074          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5075         if ((!mddev->raid_disks && !mddev->external)
5076             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5077             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5078             && cmd != GET_BITMAP_FILE) {
5079                 err = -ENODEV;
5080                 goto abort_unlock;
5081         }
5082
5083         /*
5084          * Commands even a read-only array can execute:
5085          */
5086         switch (cmd)
5087         {
5088                 case GET_ARRAY_INFO:
5089                         err = get_array_info(mddev, argp);
5090                         goto done_unlock;
5091
5092                 case GET_BITMAP_FILE:
5093                         err = get_bitmap_file(mddev, argp);
5094                         goto done_unlock;
5095
5096                 case GET_DISK_INFO:
5097                         err = get_disk_info(mddev, argp);
5098                         goto done_unlock;
5099
5100                 case RESTART_ARRAY_RW:
5101                         err = restart_array(mddev);
5102                         goto done_unlock;
5103
5104                 case STOP_ARRAY:
5105                         err = do_md_stop(mddev, 0, 1);
5106                         goto done_unlock;
5107
5108                 case STOP_ARRAY_RO:
5109                         err = do_md_stop(mddev, 1, 1);
5110                         goto done_unlock;
5111
5112         }
5113
5114         /*
5115          * The remaining ioctls are changing the state of the
5116          * superblock, so we do not allow them on read-only arrays.
5117          * However non-MD ioctls (e.g. get-size) will still come through
5118          * here and hit the 'default' below, so only disallow
5119          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5120          */
5121         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5122                 if (mddev->ro == 2) {
5123                         mddev->ro = 0;
5124                         sysfs_notify_dirent(mddev->sysfs_state);
5125                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5126                         md_wakeup_thread(mddev->thread);
5127                 } else {
5128                         err = -EROFS;
5129                         goto abort_unlock;
5130                 }
5131         }
5132
5133         switch (cmd)
5134         {
5135                 case ADD_NEW_DISK:
5136                 {
5137                         mdu_disk_info_t info;
5138                         if (copy_from_user(&info, argp, sizeof(info)))
5139                                 err = -EFAULT;
5140                         else
5141                                 err = add_new_disk(mddev, &info);
5142                         goto done_unlock;
5143                 }
5144
5145                 case HOT_REMOVE_DISK:
5146                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5147                         goto done_unlock;
5148
5149                 case HOT_ADD_DISK:
5150                         err = hot_add_disk(mddev, new_decode_dev(arg));
5151                         goto done_unlock;
5152
5153                 case SET_DISK_FAULTY:
5154                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5155                         goto done_unlock;
5156
5157                 case RUN_ARRAY:
5158                         err = do_md_run(mddev);
5159                         goto done_unlock;
5160
5161                 case SET_BITMAP_FILE:
5162                         err = set_bitmap_file(mddev, (int)arg);
5163                         goto done_unlock;
5164
5165                 default:
5166                         err = -EINVAL;
5167                         goto abort_unlock;
5168         }
5169
5170 done_unlock:
5171 abort_unlock:
5172         if (mddev->hold_active == UNTIL_IOCTL &&
5173             err != -EINVAL)
5174                 mddev->hold_active = 0;
5175         mddev_unlock(mddev);
5176
5177         return err;
5178 done:
5179         if (err)
5180                 MD_BUG();
5181 abort:
5182         return err;
5183 }
5184
5185 static int md_open(struct block_device *bdev, fmode_t mode)
5186 {
5187         /*
5188          * Succeed if we can lock the mddev, which confirms that
5189          * it isn't being stopped right now.
5190          */
5191         mddev_t *mddev = mddev_find(bdev->bd_dev);
5192         int err;
5193
5194         if (mddev->gendisk != bdev->bd_disk) {
5195                 /* we are racing with mddev_put which is discarding this
5196                  * bd_disk.
5197                  */
5198                 mddev_put(mddev);
5199                 /* Wait until bdev->bd_disk is definitely gone */
5200                 flush_scheduled_work();
5201                 /* Then retry the open from the top */
5202                 return -ERESTARTSYS;
5203         }
5204         BUG_ON(mddev != bdev->bd_disk->private_data);
5205
5206         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5207                 goto out;
5208
5209         err = 0;
5210         atomic_inc(&mddev->openers);
5211         mddev_unlock(mddev);
5212
5213         check_disk_change(bdev);
5214  out:
5215         return err;
5216 }
5217
5218 static int md_release(struct gendisk *disk, fmode_t mode)
5219 {
5220         mddev_t *mddev = disk->private_data;
5221
5222         BUG_ON(!mddev);
5223         atomic_dec(&mddev->openers);
5224         mddev_put(mddev);
5225
5226         return 0;
5227 }
5228
5229 static int md_media_changed(struct gendisk *disk)
5230 {
5231         mddev_t *mddev = disk->private_data;
5232
5233         return mddev->changed;
5234 }
5235
5236 static int md_revalidate(struct gendisk *disk)
5237 {
5238         mddev_t *mddev = disk->private_data;
5239
5240         mddev->changed = 0;
5241         return 0;
5242 }
5243 static struct block_device_operations md_fops =
5244 {
5245         .owner          = THIS_MODULE,
5246         .open           = md_open,
5247         .release        = md_release,
5248         .locked_ioctl   = md_ioctl,
5249         .getgeo         = md_getgeo,
5250         .media_changed  = md_media_changed,
5251         .revalidate_disk= md_revalidate,
5252 };
5253
5254 static int md_thread(void * arg)
5255 {
5256         mdk_thread_t *thread = arg;
5257
5258         /*
5259          * md_thread is a 'system-thread', it's priority should be very
5260          * high. We avoid resource deadlocks individually in each
5261          * raid personality. (RAID5 does preallocation) We also use RR and
5262          * the very same RT priority as kswapd, thus we will never get
5263          * into a priority inversion deadlock.
5264          *
5265          * we definitely have to have equal or higher priority than
5266          * bdflush, otherwise bdflush will deadlock if there are too
5267          * many dirty RAID5 blocks.
5268          */
5269
5270         allow_signal(SIGKILL);
5271         while (!kthread_should_stop()) {
5272
5273                 /* We need to wait INTERRUPTIBLE so that
5274                  * we don't add to the load-average.
5275                  * That means we need to be sure no signals are
5276                  * pending
5277                  */
5278                 if (signal_pending(current))
5279                         flush_signals(current);
5280
5281                 wait_event_interruptible_timeout
5282                         (thread->wqueue,
5283                          test_bit(THREAD_WAKEUP, &thread->flags)
5284                          || kthread_should_stop(),
5285                          thread->timeout);
5286
5287                 clear_bit(THREAD_WAKEUP, &thread->flags);
5288
5289                 thread->run(thread->mddev);
5290         }
5291
5292         return 0;
5293 }
5294
5295 void md_wakeup_thread(mdk_thread_t *thread)
5296 {
5297         if (thread) {
5298                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5299                 set_bit(THREAD_WAKEUP, &thread->flags);
5300                 wake_up(&thread->wqueue);
5301         }
5302 }
5303
5304 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5305                                  const char *name)
5306 {
5307         mdk_thread_t *thread;
5308
5309         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5310         if (!thread)
5311                 return NULL;
5312
5313         init_waitqueue_head(&thread->wqueue);
5314
5315         thread->run = run;
5316         thread->mddev = mddev;
5317         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5318         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5319         if (IS_ERR(thread->tsk)) {
5320                 kfree(thread);
5321                 return NULL;
5322         }
5323         return thread;
5324 }
5325
5326 void md_unregister_thread(mdk_thread_t *thread)
5327 {
5328         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5329
5330         kthread_stop(thread->tsk);
5331         kfree(thread);
5332 }
5333
5334 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5335 {
5336         if (!mddev) {
5337                 MD_BUG();
5338                 return;
5339         }
5340
5341         if (!rdev || test_bit(Faulty, &rdev->flags))
5342                 return;
5343
5344         if (mddev->external)
5345                 set_bit(Blocked, &rdev->flags);
5346 /*
5347         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5348                 mdname(mddev),
5349                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5350                 __builtin_return_address(0),__builtin_return_address(1),
5351                 __builtin_return_address(2),__builtin_return_address(3));
5352 */
5353         if (!mddev->pers)
5354                 return;
5355         if (!mddev->pers->error_handler)
5356                 return;
5357         mddev->pers->error_handler(mddev,rdev);
5358         if (mddev->degraded)
5359                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5360         set_bit(StateChanged, &rdev->flags);
5361         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5362         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5363         md_wakeup_thread(mddev->thread);
5364         md_new_event_inintr(mddev);
5365 }
5366
5367 /* seq_file implementation /proc/mdstat */
5368
5369 static void status_unused(struct seq_file *seq)
5370 {
5371         int i = 0;
5372         mdk_rdev_t *rdev;
5373
5374         seq_printf(seq, "unused devices: ");
5375
5376         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5377                 char b[BDEVNAME_SIZE];
5378                 i++;
5379                 seq_printf(seq, "%s ",
5380                               bdevname(rdev->bdev,b));
5381         }
5382         if (!i)
5383                 seq_printf(seq, "<none>");
5384
5385         seq_printf(seq, "\n");
5386 }
5387
5388
5389 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5390 {
5391         sector_t max_blocks, resync, res;
5392         unsigned long dt, db, rt;
5393         int scale;
5394         unsigned int per_milli;
5395
5396         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5397
5398         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5399                 max_blocks = mddev->resync_max_sectors >> 1;
5400         else
5401                 max_blocks = mddev->size;
5402
5403         /*
5404          * Should not happen.
5405          */
5406         if (!max_blocks) {
5407                 MD_BUG();
5408                 return;
5409         }
5410         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5411          * in a sector_t, and (max_blocks>>scale) will fit in a
5412          * u32, as those are the requirements for sector_div.
5413          * Thus 'scale' must be at least 10
5414          */
5415         scale = 10;
5416         if (sizeof(sector_t) > sizeof(unsigned long)) {
5417                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5418                         scale++;
5419         }
5420         res = (resync>>scale)*1000;
5421         sector_div(res, (u32)((max_blocks>>scale)+1));
5422
5423         per_milli = res;
5424         {
5425                 int i, x = per_milli/50, y = 20-x;
5426                 seq_printf(seq, "[");
5427                 for (i = 0; i < x; i++)
5428                         seq_printf(seq, "=");
5429                 seq_printf(seq, ">");
5430                 for (i = 0; i < y; i++)
5431                         seq_printf(seq, ".");
5432                 seq_printf(seq, "] ");
5433         }
5434         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5435                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5436                     "reshape" :
5437                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5438                      "check" :
5439                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5440                       "resync" : "recovery"))),
5441                    per_milli/10, per_milli % 10,
5442                    (unsigned long long) resync,
5443                    (unsigned long long) max_blocks);
5444
5445         /*
5446          * We do not want to overflow, so the order of operands and
5447          * the * 100 / 100 trick are important. We do a +1 to be
5448          * safe against division by zero. We only estimate anyway.
5449          *
5450          * dt: time from mark until now
5451          * db: blocks written from mark until now
5452          * rt: remaining time
5453          */
5454         dt = ((jiffies - mddev->resync_mark) / HZ);
5455         if (!dt) dt++;
5456         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5457                 - mddev->resync_mark_cnt;
5458         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5459
5460         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5461
5462         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5463 }
5464
5465 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5466 {
5467         struct list_head *tmp;
5468         loff_t l = *pos;
5469         mddev_t *mddev;
5470
5471         if (l >= 0x10000)
5472                 return NULL;
5473         if (!l--)
5474                 /* header */
5475                 return (void*)1;
5476
5477         spin_lock(&all_mddevs_lock);
5478         list_for_each(tmp,&all_mddevs)
5479                 if (!l--) {
5480                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5481                         mddev_get(mddev);
5482                         spin_unlock(&all_mddevs_lock);
5483                         return mddev;
5484                 }
5485         spin_unlock(&all_mddevs_lock);
5486         if (!l--)
5487                 return (void*)2;/* tail */
5488         return NULL;
5489 }
5490
5491 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5492 {
5493         struct list_head *tmp;
5494         mddev_t *next_mddev, *mddev = v;
5495
5496         ++*pos;
5497         if (v == (void*)2)
5498                 return NULL;
5499
5500         spin_lock(&all_mddevs_lock);
5501         if (v == (void*)1)
5502                 tmp = all_mddevs.next;
5503         else
5504                 tmp = mddev->all_mddevs.next;
5505         if (tmp != &all_mddevs)
5506                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5507         else {
5508                 next_mddev = (void*)2;
5509                 *pos = 0x10000;
5510         }
5511         spin_unlock(&all_mddevs_lock);
5512
5513         if (v != (void*)1)
5514                 mddev_put(mddev);
5515         return next_mddev;
5516
5517 }
5518
5519 static void md_seq_stop(struct seq_file *seq, void *v)
5520 {
5521         mddev_t *mddev = v;
5522
5523         if (mddev && v != (void*)1 && v != (void*)2)
5524                 mddev_put(mddev);
5525 }
5526
5527 struct mdstat_info {
5528         int event;
5529 };
5530
5531 static int md_seq_show(struct seq_file *seq, void *v)
5532 {
5533         mddev_t *mddev = v;
5534         sector_t size;
5535         mdk_rdev_t *rdev;
5536         struct mdstat_info *mi = seq->private;
5537         struct bitmap *bitmap;
5538
5539         if (v == (void*)1) {
5540                 struct mdk_personality *pers;
5541                 seq_printf(seq, "Personalities : ");
5542                 spin_lock(&pers_lock);
5543                 list_for_each_entry(pers, &pers_list, list)
5544                         seq_printf(seq, "[%s] ", pers->name);
5545
5546                 spin_unlock(&pers_lock);
5547                 seq_printf(seq, "\n");
5548                 mi->event = atomic_read(&md_event_count);
5549                 return 0;
5550         }
5551         if (v == (void*)2) {
5552                 status_unused(seq);
5553                 return 0;
5554         }
5555
5556         if (mddev_lock(mddev) < 0)
5557                 return -EINTR;
5558
5559         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5560                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5561                                                 mddev->pers ? "" : "in");
5562                 if (mddev->pers) {
5563                         if (mddev->ro==1)
5564                                 seq_printf(seq, " (read-only)");
5565                         if (mddev->ro==2)
5566                                 seq_printf(seq, " (auto-read-only)");
5567                         seq_printf(seq, " %s", mddev->pers->name);
5568                 }
5569
5570                 size = 0;
5571                 list_for_each_entry(rdev, &mddev->disks, same_set) {
5572                         char b[BDEVNAME_SIZE];
5573                         seq_printf(seq, " %s[%d]",
5574                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5575                         if (test_bit(WriteMostly, &rdev->flags))
5576                                 seq_printf(seq, "(W)");
5577                         if (test_bit(Faulty, &rdev->flags)) {
5578                                 seq_printf(seq, "(F)");
5579                                 continue;
5580                         } else if (rdev->raid_disk < 0)
5581                                 seq_printf(seq, "(S)"); /* spare */
5582                         size += rdev->size;
5583                 }
5584
5585                 if (!list_empty(&mddev->disks)) {
5586                         if (mddev->pers)
5587                                 seq_printf(seq, "\n      %llu blocks",
5588                                            (unsigned long long)
5589                                            mddev->array_sectors / 2);
5590                         else
5591                                 seq_printf(seq, "\n      %llu blocks",
5592                                            (unsigned long long)size);
5593                 }
5594                 if (mddev->persistent) {
5595                         if (mddev->major_version != 0 ||
5596                             mddev->minor_version != 90) {
5597                                 seq_printf(seq," super %d.%d",
5598                                            mddev->major_version,
5599                                            mddev->minor_version);
5600                         }
5601                 } else if (mddev->external)
5602                         seq_printf(seq, " super external:%s",
5603                                    mddev->metadata_type);
5604                 else
5605                         seq_printf(seq, " super non-persistent");
5606
5607                 if (mddev->pers) {
5608                         mddev->pers->status(seq, mddev);
5609                         seq_printf(seq, "\n      ");
5610                         if (mddev->pers->sync_request) {
5611                                 if (mddev->curr_resync > 2) {
5612                                         status_resync(seq, mddev);
5613                                         seq_printf(seq, "\n      ");
5614                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5615                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5616                                 else if (mddev->recovery_cp < MaxSector)
5617                                         seq_printf(seq, "\tresync=PENDING\n      ");
5618                         }
5619                 } else
5620                         seq_printf(seq, "\n       ");
5621
5622                 if ((bitmap = mddev->bitmap)) {
5623                         unsigned long chunk_kb;
5624                         unsigned long flags;
5625                         spin_lock_irqsave(&bitmap->lock, flags);
5626                         chunk_kb = bitmap->chunksize >> 10;
5627                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5628                                 "%lu%s chunk",
5629                                 bitmap->pages - bitmap->missing_pages,
5630                                 bitmap->pages,
5631                                 (bitmap->pages - bitmap->missing_pages)
5632                                         << (PAGE_SHIFT - 10),
5633                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5634                                 chunk_kb ? "KB" : "B");
5635                         if (bitmap->file) {
5636                                 seq_printf(seq, ", file: ");
5637                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5638                         }
5639
5640                         seq_printf(seq, "\n");
5641                         spin_unlock_irqrestore(&bitmap->lock, flags);
5642                 }
5643
5644                 seq_printf(seq, "\n");
5645         }
5646         mddev_unlock(mddev);
5647
5648         return 0;
5649 }
5650
5651 static struct seq_operations md_seq_ops = {
5652         .start  = md_seq_start,
5653         .next   = md_seq_next,
5654         .stop   = md_seq_stop,
5655         .show   = md_seq_show,
5656 };
5657
5658 static int md_seq_open(struct inode *inode, struct file *file)
5659 {
5660         int error;
5661         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5662         if (mi == NULL)
5663                 return -ENOMEM;
5664
5665         error = seq_open(file, &md_seq_ops);
5666         if (error)
5667                 kfree(mi);
5668         else {
5669                 struct seq_file *p = file->private_data;
5670                 p->private = mi;
5671                 mi->event = atomic_read(&md_event_count);
5672         }
5673         return error;
5674 }
5675
5676 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5677 {
5678         struct seq_file *m = filp->private_data;
5679         struct mdstat_info *mi = m->private;
5680         int mask;
5681
5682         poll_wait(filp, &md_event_waiters, wait);
5683
5684         /* always allow read */
5685         mask = POLLIN | POLLRDNORM;
5686
5687         if (mi->event != atomic_read(&md_event_count))
5688                 mask |= POLLERR | POLLPRI;
5689         return mask;
5690 }
5691
5692 static const struct file_operations md_seq_fops = {
5693         .owner          = THIS_MODULE,
5694         .open           = md_seq_open,
5695         .read           = seq_read,
5696         .llseek         = seq_lseek,
5697         .release        = seq_release_private,
5698         .poll           = mdstat_poll,
5699 };
5700
5701 int register_md_personality(struct mdk_personality *p)
5702 {
5703         spin_lock(&pers_lock);
5704         list_add_tail(&p->list, &pers_list);
5705         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5706         spin_unlock(&pers_lock);
5707         return 0;
5708 }
5709
5710 int unregister_md_personality(struct mdk_personality *p)
5711 {
5712         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5713         spin_lock(&pers_lock);
5714         list_del_init(&p->list);
5715         spin_unlock(&pers_lock);
5716         return 0;
5717 }
5718
5719 static int is_mddev_idle(mddev_t *mddev)
5720 {
5721         mdk_rdev_t * rdev;
5722         int idle;
5723         long curr_events;
5724
5725         idle = 1;
5726         rcu_read_lock();
5727         rdev_for_each_rcu(rdev, mddev) {
5728                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5729                 curr_events = part_stat_read(&disk->part0, sectors[0]) +
5730                                 part_stat_read(&disk->part0, sectors[1]) -
5731                                 atomic_read(&disk->sync_io);
5732                 /* sync IO will cause sync_io to increase before the disk_stats
5733                  * as sync_io is counted when a request starts, and
5734                  * disk_stats is counted when it completes.
5735                  * So resync activity will cause curr_events to be smaller than
5736                  * when there was no such activity.
5737                  * non-sync IO will cause disk_stat to increase without
5738                  * increasing sync_io so curr_events will (eventually)
5739                  * be larger than it was before.  Once it becomes
5740                  * substantially larger, the test below will cause
5741                  * the array to appear non-idle, and resync will slow
5742                  * down.
5743                  * If there is a lot of outstanding resync activity when
5744                  * we set last_event to curr_events, then all that activity
5745                  * completing might cause the array to appear non-idle
5746                  * and resync will be slowed down even though there might
5747                  * not have been non-resync activity.  This will only
5748                  * happen once though.  'last_events' will soon reflect
5749                  * the state where there is little or no outstanding
5750                  * resync requests, and further resync activity will
5751                  * always make curr_events less than last_events.
5752                  *
5753                  */
5754                 if (curr_events - rdev->last_events > 4096) {
5755                         rdev->last_events = curr_events;
5756                         idle = 0;
5757                 }
5758         }
5759         rcu_read_unlock();
5760         return idle;
5761 }
5762
5763 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5764 {
5765         /* another "blocks" (512byte) blocks have been synced */
5766         atomic_sub(blocks, &mddev->recovery_active);
5767         wake_up(&mddev->recovery_wait);
5768         if (!ok) {
5769                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5770                 md_wakeup_thread(mddev->thread);
5771                 // stop recovery, signal do_sync ....
5772         }
5773 }
5774
5775
5776 /* md_write_start(mddev, bi)
5777  * If we need to update some array metadata (e.g. 'active' flag
5778  * in superblock) before writing, schedule a superblock update
5779  * and wait for it to complete.
5780  */
5781 void md_write_start(mddev_t *mddev, struct bio *bi)
5782 {
5783         int did_change = 0;
5784         if (bio_data_dir(bi) != WRITE)
5785                 return;
5786
5787         BUG_ON(mddev->ro == 1);
5788         if (mddev->ro == 2) {
5789                 /* need to switch to read/write */
5790                 mddev->ro = 0;
5791                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5792                 md_wakeup_thread(mddev->thread);
5793                 md_wakeup_thread(mddev->sync_thread);
5794                 did_change = 1;
5795         }
5796         atomic_inc(&mddev->writes_pending);
5797         if (mddev->safemode == 1)
5798                 mddev->safemode = 0;
5799         if (mddev->in_sync) {
5800                 spin_lock_irq(&mddev->write_lock);
5801                 if (mddev->in_sync) {
5802                         mddev->in_sync = 0;
5803                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5804                         md_wakeup_thread(mddev->thread);
5805                         did_change = 1;
5806                 }
5807                 spin_unlock_irq(&mddev->write_lock);
5808         }
5809         if (did_change)
5810                 sysfs_notify_dirent(mddev->sysfs_state);
5811         wait_event(mddev->sb_wait,
5812                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5813                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5814 }
5815
5816 void md_write_end(mddev_t *mddev)
5817 {
5818         if (atomic_dec_and_test(&mddev->writes_pending)) {
5819                 if (mddev->safemode == 2)
5820                         md_wakeup_thread(mddev->thread);
5821                 else if (mddev->safemode_delay)
5822                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5823         }
5824 }
5825
5826 /* md_allow_write(mddev)
5827  * Calling this ensures that the array is marked 'active' so that writes
5828  * may proceed without blocking.  It is important to call this before
5829  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5830  * Must be called with mddev_lock held.
5831  *
5832  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5833  * is dropped, so return -EAGAIN after notifying userspace.
5834  */
5835 int md_allow_write(mddev_t *mddev)
5836 {
5837         if (!mddev->pers)
5838                 return 0;
5839         if (mddev->ro)
5840                 return 0;
5841         if (!mddev->pers->sync_request)
5842                 return 0;
5843
5844         spin_lock_irq(&mddev->write_lock);
5845         if (mddev->in_sync) {
5846                 mddev->in_sync = 0;
5847                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5848                 if (mddev->safemode_delay &&
5849                     mddev->safemode == 0)
5850                         mddev->safemode = 1;
5851                 spin_unlock_irq(&mddev->write_lock);
5852                 md_update_sb(mddev, 0);
5853                 sysfs_notify_dirent(mddev->sysfs_state);
5854         } else
5855                 spin_unlock_irq(&mddev->write_lock);
5856
5857         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5858                 return -EAGAIN;
5859         else
5860                 return 0;
5861 }
5862 EXPORT_SYMBOL_GPL(md_allow_write);
5863
5864 #define SYNC_MARKS      10
5865 #define SYNC_MARK_STEP  (3*HZ)
5866 void md_do_sync(mddev_t *mddev)
5867 {
5868         mddev_t *mddev2;
5869         unsigned int currspeed = 0,
5870                  window;
5871         sector_t max_sectors,j, io_sectors;
5872         unsigned long mark[SYNC_MARKS];
5873         sector_t mark_cnt[SYNC_MARKS];
5874         int last_mark,m;
5875         struct list_head *tmp;
5876         sector_t last_check;
5877         int skipped = 0;
5878         mdk_rdev_t *rdev;
5879         char *desc;
5880
5881         /* just incase thread restarts... */
5882         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5883                 return;
5884         if (mddev->ro) /* never try to sync a read-only array */
5885                 return;
5886
5887         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5888                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5889                         desc = "data-check";
5890                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5891                         desc = "requested-resync";
5892                 else
5893                         desc = "resync";
5894         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5895                 desc = "reshape";
5896         else
5897                 desc = "recovery";
5898
5899         /* we overload curr_resync somewhat here.
5900          * 0 == not engaged in resync at all
5901          * 2 == checking that there is no conflict with another sync
5902          * 1 == like 2, but have yielded to allow conflicting resync to
5903          *              commense
5904          * other == active in resync - this many blocks
5905          *
5906          * Before starting a resync we must have set curr_resync to
5907          * 2, and then checked that every "conflicting" array has curr_resync
5908          * less than ours.  When we find one that is the same or higher
5909          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5910          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5911          * This will mean we have to start checking from the beginning again.
5912          *
5913          */
5914
5915         do {
5916                 mddev->curr_resync = 2;
5917
5918         try_again:
5919                 if (kthread_should_stop()) {
5920                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5921                         goto skip;
5922                 }
5923                 for_each_mddev(mddev2, tmp) {
5924                         if (mddev2 == mddev)
5925                                 continue;
5926                         if (!mddev->parallel_resync
5927                         &&  mddev2->curr_resync
5928                         &&  match_mddev_units(mddev, mddev2)) {
5929                                 DEFINE_WAIT(wq);
5930                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5931                                         /* arbitrarily yield */
5932                                         mddev->curr_resync = 1;
5933                                         wake_up(&resync_wait);
5934                                 }
5935                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5936                                         /* no need to wait here, we can wait the next
5937                                          * time 'round when curr_resync == 2
5938                                          */
5939                                         continue;
5940                                 /* We need to wait 'interruptible' so as not to
5941                                  * contribute to the load average, and not to
5942                                  * be caught by 'softlockup'
5943                                  */
5944                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
5945                                 if (!kthread_should_stop() &&
5946                                     mddev2->curr_resync >= mddev->curr_resync) {
5947                                         printk(KERN_INFO "md: delaying %s of %s"
5948                                                " until %s has finished (they"
5949                                                " share one or more physical units)\n",
5950                                                desc, mdname(mddev), mdname(mddev2));
5951                                         mddev_put(mddev2);
5952                                         if (signal_pending(current))
5953                                                 flush_signals(current);
5954                                         schedule();
5955                                         finish_wait(&resync_wait, &wq);
5956                                         goto try_again;
5957                                 }
5958                                 finish_wait(&resync_wait, &wq);
5959                         }
5960                 }
5961         } while (mddev->curr_resync < 2);
5962
5963         j = 0;
5964         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5965                 /* resync follows the size requested by the personality,
5966                  * which defaults to physical size, but can be virtual size
5967                  */
5968                 max_sectors = mddev->resync_max_sectors;
5969                 mddev->resync_mismatches = 0;
5970                 /* we don't use the checkpoint if there's a bitmap */
5971                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5972                         j = mddev->resync_min;
5973                 else if (!mddev->bitmap)
5974                         j = mddev->recovery_cp;
5975
5976         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5977                 max_sectors = mddev->size << 1;
5978         else {
5979                 /* recovery follows the physical size of devices */
5980                 max_sectors = mddev->size << 1;
5981                 j = MaxSector;
5982                 list_for_each_entry(rdev, &mddev->disks, same_set)
5983                         if (rdev->raid_disk >= 0 &&
5984                             !test_bit(Faulty, &rdev->flags) &&
5985                             !test_bit(In_sync, &rdev->flags) &&
5986                             rdev->recovery_offset < j)
5987                                 j = rdev->recovery_offset;
5988         }
5989
5990         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5991         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5992                 " %d KB/sec/disk.\n", speed_min(mddev));
5993         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5994                "(but not more than %d KB/sec) for %s.\n",
5995                speed_max(mddev), desc);
5996
5997         is_mddev_idle(mddev); /* this also initializes IO event counters */
5998
5999         io_sectors = 0;
6000         for (m = 0; m < SYNC_MARKS; m++) {
6001                 mark[m] = jiffies;
6002                 mark_cnt[m] = io_sectors;
6003         }
6004         last_mark = 0;
6005         mddev->resync_mark = mark[last_mark];
6006         mddev->resync_mark_cnt = mark_cnt[last_mark];
6007
6008         /*
6009          * Tune reconstruction:
6010          */
6011         window = 32*(PAGE_SIZE/512);
6012         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6013                 window/2,(unsigned long long) max_sectors/2);
6014
6015         atomic_set(&mddev->recovery_active, 0);
6016         last_check = 0;
6017
6018         if (j>2) {
6019                 printk(KERN_INFO
6020                        "md: resuming %s of %s from checkpoint.\n",
6021                        desc, mdname(mddev));
6022                 mddev->curr_resync = j;
6023         }
6024
6025         while (j < max_sectors) {
6026                 sector_t sectors;
6027
6028                 skipped = 0;
6029                 if (j >= mddev->resync_max) {
6030                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6031                         wait_event(mddev->recovery_wait,
6032                                    mddev->resync_max > j
6033                                    || kthread_should_stop());
6034                 }
6035                 if (kthread_should_stop())
6036                         goto interrupted;
6037                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6038                                                   currspeed < speed_min(mddev));
6039                 if (sectors == 0) {
6040                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6041                         goto out;
6042                 }
6043
6044                 if (!skipped) { /* actual IO requested */
6045                         io_sectors += sectors;
6046                         atomic_add(sectors, &mddev->recovery_active);
6047                 }
6048
6049                 j += sectors;
6050                 if (j>1) mddev->curr_resync = j;
6051                 mddev->curr_mark_cnt = io_sectors;
6052                 if (last_check == 0)
6053                         /* this is the earliers that rebuilt will be
6054                          * visible in /proc/mdstat
6055                          */
6056                         md_new_event(mddev);
6057
6058                 if (last_check + window > io_sectors || j == max_sectors)
6059                         continue;
6060
6061                 last_check = io_sectors;
6062
6063                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6064                         break;
6065
6066         repeat:
6067                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6068                         /* step marks */
6069                         int next = (last_mark+1) % SYNC_MARKS;
6070
6071                         mddev->resync_mark = mark[next];
6072                         mddev->resync_mark_cnt = mark_cnt[next];
6073                         mark[next] = jiffies;
6074                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6075                         last_mark = next;
6076                 }
6077
6078
6079                 if (kthread_should_stop())
6080                         goto interrupted;
6081
6082
6083                 /*
6084                  * this loop exits only if either when we are slower than
6085                  * the 'hard' speed limit, or the system was IO-idle for
6086                  * a jiffy.
6087                  * the system might be non-idle CPU-wise, but we only care
6088                  * about not overloading the IO subsystem. (things like an
6089                  * e2fsck being done on the RAID array should execute fast)
6090                  */
6091                 blk_unplug(mddev->queue);
6092                 cond_resched();
6093
6094                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6095                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6096
6097                 if (currspeed > speed_min(mddev)) {
6098                         if ((currspeed > speed_max(mddev)) ||
6099                                         !is_mddev_idle(mddev)) {
6100                                 msleep(500);
6101                                 goto repeat;
6102                         }
6103                 }
6104         }
6105         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6106         /*
6107          * this also signals 'finished resyncing' to md_stop
6108          */
6109  out:
6110         blk_unplug(mddev->queue);
6111
6112         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6113
6114         /* tell personality that we are finished */
6115         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6116
6117         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6118             mddev->curr_resync > 2) {
6119                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6120                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6121                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6122                                         printk(KERN_INFO
6123                                                "md: checkpointing %s of %s.\n",
6124                                                desc, mdname(mddev));
6125                                         mddev->recovery_cp = mddev->curr_resync;
6126                                 }
6127                         } else
6128                                 mddev->recovery_cp = MaxSector;
6129                 } else {
6130                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6131                                 mddev->curr_resync = MaxSector;
6132                         list_for_each_entry(rdev, &mddev->disks, same_set)
6133                                 if (rdev->raid_disk >= 0 &&
6134                                     !test_bit(Faulty, &rdev->flags) &&
6135                                     !test_bit(In_sync, &rdev->flags) &&
6136                                     rdev->recovery_offset < mddev->curr_resync)
6137                                         rdev->recovery_offset = mddev->curr_resync;
6138                 }
6139         }
6140         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6141
6142  skip:
6143         mddev->curr_resync = 0;
6144         mddev->resync_min = 0;
6145         mddev->resync_max = MaxSector;
6146         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6147         wake_up(&resync_wait);
6148         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6149         md_wakeup_thread(mddev->thread);
6150         return;
6151
6152  interrupted:
6153         /*
6154          * got a signal, exit.
6155          */
6156         printk(KERN_INFO
6157                "md: md_do_sync() got signal ... exiting\n");
6158         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6159         goto out;
6160
6161 }
6162 EXPORT_SYMBOL_GPL(md_do_sync);
6163
6164
6165 static int remove_and_add_spares(mddev_t *mddev)
6166 {
6167         mdk_rdev_t *rdev;
6168         int spares = 0;
6169
6170         list_for_each_entry(rdev, &mddev->disks, same_set)
6171                 if (rdev->raid_disk >= 0 &&
6172                     !test_bit(Blocked, &rdev->flags) &&
6173                     (test_bit(Faulty, &rdev->flags) ||
6174                      ! test_bit(In_sync, &rdev->flags)) &&
6175                     atomic_read(&rdev->nr_pending)==0) {
6176                         if (mddev->pers->hot_remove_disk(
6177                                     mddev, rdev->raid_disk)==0) {
6178                                 char nm[20];
6179                                 sprintf(nm,"rd%d", rdev->raid_disk);
6180                                 sysfs_remove_link(&mddev->kobj, nm);
6181                                 rdev->raid_disk = -1;
6182                         }
6183                 }
6184
6185         if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6186                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6187                         if (rdev->raid_disk >= 0 &&
6188                             !test_bit(In_sync, &rdev->flags) &&
6189                             !test_bit(Blocked, &rdev->flags))
6190                                 spares++;
6191                         if (rdev->raid_disk < 0
6192                             && !test_bit(Faulty, &rdev->flags)) {
6193                                 rdev->recovery_offset = 0;
6194                                 if (mddev->pers->
6195                                     hot_add_disk(mddev, rdev) == 0) {
6196                                         char nm[20];
6197                                         sprintf(nm, "rd%d", rdev->raid_disk);
6198                                         if (sysfs_create_link(&mddev->kobj,
6199                                                               &rdev->kobj, nm))
6200                                                 printk(KERN_WARNING
6201                                                        "md: cannot register "
6202                                                        "%s for %s\n",
6203                                                        nm, mdname(mddev));
6204                                         spares++;
6205                                         md_new_event(mddev);
6206                                 } else
6207                                         break;
6208                         }
6209                 }
6210         }
6211         return spares;
6212 }
6213 /*
6214  * This routine is regularly called by all per-raid-array threads to
6215  * deal with generic issues like resync and super-block update.
6216  * Raid personalities that don't have a thread (linear/raid0) do not
6217  * need this as they never do any recovery or update the superblock.
6218  *
6219  * It does not do any resync itself, but rather "forks" off other threads
6220  * to do that as needed.
6221  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6222  * "->recovery" and create a thread at ->sync_thread.
6223  * When the thread finishes it sets MD_RECOVERY_DONE
6224  * and wakeups up this thread which will reap the thread and finish up.
6225  * This thread also removes any faulty devices (with nr_pending == 0).
6226  *
6227  * The overall approach is:
6228  *  1/ if the superblock needs updating, update it.
6229  *  2/ If a recovery thread is running, don't do anything else.
6230  *  3/ If recovery has finished, clean up, possibly marking spares active.
6231  *  4/ If there are any faulty devices, remove them.
6232  *  5/ If array is degraded, try to add spares devices
6233  *  6/ If array has spares or is not in-sync, start a resync thread.
6234  */
6235 void md_check_recovery(mddev_t *mddev)
6236 {
6237         mdk_rdev_t *rdev;
6238
6239
6240         if (mddev->bitmap)
6241                 bitmap_daemon_work(mddev->bitmap);
6242
6243         if (mddev->ro)
6244                 return;
6245
6246         if (signal_pending(current)) {
6247                 if (mddev->pers->sync_request && !mddev->external) {
6248                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6249                                mdname(mddev));
6250                         mddev->safemode = 2;
6251                 }
6252                 flush_signals(current);
6253         }
6254
6255         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6256                 return;
6257         if ( ! (
6258                 (mddev->flags && !mddev->external) ||
6259                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6260                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6261                 (mddev->external == 0 && mddev->safemode == 1) ||
6262                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6263                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6264                 ))
6265                 return;
6266
6267         if (mddev_trylock(mddev)) {
6268                 int spares = 0;
6269
6270                 if (mddev->ro) {
6271                         /* Only thing we do on a ro array is remove
6272                          * failed devices.
6273                          */
6274                         remove_and_add_spares(mddev);
6275                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6276                         goto unlock;
6277                 }
6278
6279                 if (!mddev->external) {
6280                         int did_change = 0;
6281                         spin_lock_irq(&mddev->write_lock);
6282                         if (mddev->safemode &&
6283                             !atomic_read(&mddev->writes_pending) &&
6284                             !mddev->in_sync &&
6285                             mddev->recovery_cp == MaxSector) {
6286                                 mddev->in_sync = 1;
6287                                 did_change = 1;
6288                                 if (mddev->persistent)
6289                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6290                         }
6291                         if (mddev->safemode == 1)
6292                                 mddev->safemode = 0;
6293                         spin_unlock_irq(&mddev->write_lock);
6294                         if (did_change)
6295                                 sysfs_notify_dirent(mddev->sysfs_state);
6296                 }
6297
6298                 if (mddev->flags)
6299                         md_update_sb(mddev, 0);
6300
6301                 list_for_each_entry(rdev, &mddev->disks, same_set)
6302                         if (test_and_clear_bit(StateChanged, &rdev->flags))
6303                                 sysfs_notify_dirent(rdev->sysfs_state);
6304
6305
6306                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6307                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6308                         /* resync/recovery still happening */
6309                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6310                         goto unlock;
6311                 }
6312                 if (mddev->sync_thread) {
6313                         /* resync has finished, collect result */
6314                         md_unregister_thread(mddev->sync_thread);
6315                         mddev->sync_thread = NULL;
6316                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6317                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6318                                 /* success...*/
6319                                 /* activate any spares */
6320                                 if (mddev->pers->spare_active(mddev))
6321                                         sysfs_notify(&mddev->kobj, NULL,
6322                                                      "degraded");
6323                         }
6324                         md_update_sb(mddev, 1);
6325
6326                         /* if array is no-longer degraded, then any saved_raid_disk
6327                          * information must be scrapped
6328                          */
6329                         if (!mddev->degraded)
6330                                 list_for_each_entry(rdev, &mddev->disks, same_set)
6331                                         rdev->saved_raid_disk = -1;
6332
6333                         mddev->recovery = 0;
6334                         /* flag recovery needed just to double check */
6335                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6336                         sysfs_notify_dirent(mddev->sysfs_action);
6337                         md_new_event(mddev);
6338                         goto unlock;
6339                 }
6340                 /* Set RUNNING before clearing NEEDED to avoid
6341                  * any transients in the value of "sync_action".
6342                  */
6343                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6344                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6345                 /* Clear some bits that don't mean anything, but
6346                  * might be left set
6347                  */
6348                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6349                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6350
6351                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6352                         goto unlock;
6353                 /* no recovery is running.
6354                  * remove any failed drives, then
6355                  * add spares if possible.
6356                  * Spare are also removed and re-added, to allow
6357                  * the personality to fail the re-add.
6358                  */
6359
6360                 if (mddev->reshape_position != MaxSector) {
6361                         if (mddev->pers->check_reshape(mddev) != 0)
6362                                 /* Cannot proceed */
6363                                 goto unlock;
6364                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6365                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6366                 } else if ((spares = remove_and_add_spares(mddev))) {
6367                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6368                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6369                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6370                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6371                 } else if (mddev->recovery_cp < MaxSector) {
6372                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6373                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6374                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6375                         /* nothing to be done ... */
6376                         goto unlock;
6377
6378                 if (mddev->pers->sync_request) {
6379                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6380                                 /* We are adding a device or devices to an array
6381                                  * which has the bitmap stored on all devices.
6382                                  * So make sure all bitmap pages get written
6383                                  */
6384                                 bitmap_write_all(mddev->bitmap);
6385                         }
6386                         mddev->sync_thread = md_register_thread(md_do_sync,
6387                                                                 mddev,
6388                                                                 "%s_resync");
6389                         if (!mddev->sync_thread) {
6390                                 printk(KERN_ERR "%s: could not start resync"
6391                                         " thread...\n",
6392                                         mdname(mddev));
6393                                 /* leave the spares where they are, it shouldn't hurt */
6394                                 mddev->recovery = 0;
6395                         } else
6396                                 md_wakeup_thread(mddev->sync_thread);
6397                         sysfs_notify_dirent(mddev->sysfs_action);
6398                         md_new_event(mddev);
6399                 }
6400         unlock:
6401                 if (!mddev->sync_thread) {
6402                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6403                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6404                                                &mddev->recovery))
6405                                 if (mddev->sysfs_action)
6406                                         sysfs_notify_dirent(mddev->sysfs_action);
6407                 }
6408                 mddev_unlock(mddev);
6409         }
6410 }
6411
6412 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6413 {
6414         sysfs_notify_dirent(rdev->sysfs_state);
6415         wait_event_timeout(rdev->blocked_wait,
6416                            !test_bit(Blocked, &rdev->flags),
6417                            msecs_to_jiffies(5000));
6418         rdev_dec_pending(rdev, mddev);
6419 }
6420 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6421
6422 static int md_notify_reboot(struct notifier_block *this,
6423                             unsigned long code, void *x)
6424 {
6425         struct list_head *tmp;
6426         mddev_t *mddev;
6427
6428         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6429
6430                 printk(KERN_INFO "md: stopping all md devices.\n");
6431
6432                 for_each_mddev(mddev, tmp)
6433                         if (mddev_trylock(mddev)) {
6434                                 /* Force a switch to readonly even array
6435                                  * appears to still be in use.  Hence
6436                                  * the '100'.
6437                                  */
6438                                 do_md_stop(mddev, 1, 100);
6439                                 mddev_unlock(mddev);
6440                         }
6441                 /*
6442                  * certain more exotic SCSI devices are known to be
6443                  * volatile wrt too early system reboots. While the
6444                  * right place to handle this issue is the given
6445                  * driver, we do want to have a safe RAID driver ...
6446                  */
6447                 mdelay(1000*1);
6448         }
6449         return NOTIFY_DONE;
6450 }
6451
6452 static struct notifier_block md_notifier = {
6453         .notifier_call  = md_notify_reboot,
6454         .next           = NULL,
6455         .priority       = INT_MAX, /* before any real devices */
6456 };
6457
6458 static void md_geninit(void)
6459 {
6460         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6461
6462         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6463 }
6464
6465 static int __init md_init(void)
6466 {
6467         if (register_blkdev(MAJOR_NR, "md"))
6468                 return -1;
6469         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6470                 unregister_blkdev(MAJOR_NR, "md");
6471                 return -1;
6472         }
6473         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6474                             md_probe, NULL, NULL);
6475         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6476                             md_probe, NULL, NULL);
6477
6478         register_reboot_notifier(&md_notifier);
6479         raid_table_header = register_sysctl_table(raid_root_table);
6480
6481         md_geninit();
6482         return 0;
6483 }
6484
6485
6486 #ifndef MODULE
6487
6488 /*
6489  * Searches all registered partitions for autorun RAID arrays
6490  * at boot time.
6491  */
6492
6493 static LIST_HEAD(all_detected_devices);
6494 struct detected_devices_node {
6495         struct list_head list;
6496         dev_t dev;
6497 };
6498
6499 void md_autodetect_dev(dev_t dev)
6500 {
6501         struct detected_devices_node *node_detected_dev;
6502
6503         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6504         if (node_detected_dev) {
6505                 node_detected_dev->dev = dev;
6506                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6507         } else {
6508                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6509                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6510         }
6511 }
6512
6513
6514 static void autostart_arrays(int part)
6515 {
6516         mdk_rdev_t *rdev;
6517         struct detected_devices_node *node_detected_dev;
6518         dev_t dev;
6519         int i_scanned, i_passed;
6520
6521         i_scanned = 0;
6522         i_passed = 0;
6523
6524         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6525
6526         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6527                 i_scanned++;
6528                 node_detected_dev = list_entry(all_detected_devices.next,
6529                                         struct detected_devices_node, list);
6530                 list_del(&node_detected_dev->list);
6531                 dev = node_detected_dev->dev;
6532                 kfree(node_detected_dev);
6533                 rdev = md_import_device(dev,0, 90);
6534                 if (IS_ERR(rdev))
6535                         continue;
6536
6537                 if (test_bit(Faulty, &rdev->flags)) {
6538                         MD_BUG();
6539                         continue;
6540                 }
6541                 set_bit(AutoDetected, &rdev->flags);
6542                 list_add(&rdev->same_set, &pending_raid_disks);
6543                 i_passed++;
6544         }
6545
6546         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6547                                                 i_scanned, i_passed);
6548
6549         autorun_devices(part);
6550 }
6551
6552 #endif /* !MODULE */
6553
6554 static __exit void md_exit(void)
6555 {
6556         mddev_t *mddev;
6557         struct list_head *tmp;
6558
6559         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6560         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6561
6562         unregister_blkdev(MAJOR_NR,"md");
6563         unregister_blkdev(mdp_major, "mdp");
6564         unregister_reboot_notifier(&md_notifier);
6565         unregister_sysctl_table(raid_table_header);
6566         remove_proc_entry("mdstat", NULL);
6567         for_each_mddev(mddev, tmp) {
6568                 export_array(mddev);
6569                 mddev->hold_active = 0;
6570         }
6571 }
6572
6573 subsys_initcall(md_init);
6574 module_exit(md_exit)
6575
6576 static int get_ro(char *buffer, struct kernel_param *kp)
6577 {
6578         return sprintf(buffer, "%d", start_readonly);
6579 }
6580 static int set_ro(const char *val, struct kernel_param *kp)
6581 {
6582         char *e;
6583         int num = simple_strtoul(val, &e, 10);
6584         if (*val && (*e == '\0' || *e == '\n')) {
6585                 start_readonly = num;
6586                 return 0;
6587         }
6588         return -EINVAL;
6589 }
6590
6591 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6592 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6593
6594 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6595
6596 EXPORT_SYMBOL(register_md_personality);
6597 EXPORT_SYMBOL(unregister_md_personality);
6598 EXPORT_SYMBOL(md_error);
6599 EXPORT_SYMBOL(md_done_sync);
6600 EXPORT_SYMBOL(md_write_start);
6601 EXPORT_SYMBOL(md_write_end);
6602 EXPORT_SYMBOL(md_register_thread);
6603 EXPORT_SYMBOL(md_unregister_thread);
6604 EXPORT_SYMBOL(md_wakeup_thread);
6605 EXPORT_SYMBOL(md_check_recovery);
6606 MODULE_LICENSE("GPL");
6607 MODULE_ALIAS("md");
6608 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);