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/blkdev.h>
  37 #include <linux/sysctl.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/buffer_head.h> /* for invalidate_bdev */
  40 #include <linux/poll.h>
  41 #include <linux/ctype.h>
  42 #include <linux/string.h>
  43 #include <linux/hdreg.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/random.h>
  46 #include <linux/reboot.h>
  47 #include <linux/file.h>
  48 #include <linux/compat.h>
  49 #include <linux/delay.h>
  50 #include <linux/raid/md_p.h>
  51 #include <linux/raid/md_u.h>
  52 #include <linux/slab.h>
  53 #include "md.h"
  54 #include "bitmap.h"
  55
  56 #define DEBUG 0
  57 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  58
  59
  60 #ifndef MODULE
  61 static void autostart_arrays(int part);
  62 #endif
  63
  64 static LIST_HEAD(pers_list);
  65 static DEFINE_SPINLOCK(pers_lock);
  66
  67 static void md_print_devices(void);
  68
  69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  70
  71 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  72
  73 /*
  74  * Default number of read corrections we'll attempt on an rdev
  75  * before ejecting it from the array. We divide the read error
  76  * count by 2 for every hour elapsed between read errors.
  77  */
  78 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
  79 /*
  80  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  81  * is 1000 KB/sec, so the extra system load does not show up that much.
  82  * Increase it if you want to have more _guaranteed_ speed. Note that
  83  * the RAID driver will use the maximum available bandwidth if the IO
  84  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  85  * speed limit - in case reconstruction slows down your system despite
  86  * idle IO detection.
  87  *
  88  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  89  * or /sys/block/mdX/md/sync_speed_{min,max}
  90  */
  91
  92 static int sysctl_speed_limit_min = 1000;
  93 static int sysctl_speed_limit_max = 200000;
  94 static inline int speed_min(mddev_t *mddev)
  95 {
  96         return mddev->sync_speed_min ?
  97                 mddev->sync_speed_min : sysctl_speed_limit_min;
  98 }
  99
 100 static inline int speed_max(mddev_t *mddev)
 101 {
 102         return mddev->sync_speed_max ?
 103                 mddev->sync_speed_max : sysctl_speed_limit_max;
 104 }
 105
 106 static struct ctl_table_header *raid_table_header;
 107
 108 static ctl_table raid_table[] = {
 109         {
 110                 .procname       = "speed_limit_min",
 111                 .data           = &sysctl_speed_limit_min,
 112                 .maxlen         = sizeof(int),
 113                 .mode           = S_IRUGO|S_IWUSR,
 114                 .proc_handler   = proc_dointvec,
 115         },
 116         {
 117                 .procname       = "speed_limit_max",
 118                 .data           = &sysctl_speed_limit_max,
 119                 .maxlen         = sizeof(int),
 120                 .mode           = S_IRUGO|S_IWUSR,
 121                 .proc_handler   = proc_dointvec,
 122         },
 123         { }
 124 };
 125
 126 static ctl_table raid_dir_table[] = {
 127         {
 128                 .procname       = "raid",
 129                 .maxlen         = 0,
 130                 .mode           = S_IRUGO|S_IXUGO,
 131                 .child          = raid_table,
 132         },
 133         { }
 134 };
 135
 136 static ctl_table raid_root_table[] = {
 137         {
 138                 .procname       = "dev",
 139                 .maxlen         = 0,
 140                 .mode           = 0555,
 141                 .child          = raid_dir_table,
 142         },
 143         {  }
 144 };
 145
 146 static const struct block_device_operations md_fops;
 147
 148 static int start_readonly;
 149
 150 /*
 151  * We have a system wide 'event count' that is incremented
 152  * on any 'interesting' event, and readers of /proc/mdstat
 153  * can use 'poll' or 'select' to find out when the event
 154  * count increases.
 155  *
 156  * Events are:
 157  *  start array, stop array, error, add device, remove device,
 158  *  start build, activate spare
 159  */
 160 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 161 static atomic_t md_event_count;
 162 void md_new_event(mddev_t *mddev)
 163 {
 164         atomic_inc(&md_event_count);
 165         wake_up(&md_event_waiters);
 166 }
 167 EXPORT_SYMBOL_GPL(md_new_event);
 168
 169 /* Alternate version that can be called from interrupts
 170  * when calling sysfs_notify isn't needed.
 171  */
 172 static void md_new_event_inintr(mddev_t *mddev)
 173 {
 174         atomic_inc(&md_event_count);
 175         wake_up(&md_event_waiters);
 176 }
 177
 178 /*
 179  * Enables to iterate over all existing md arrays
 180  * all_mddevs_lock protects this list.
 181  */
 182 static LIST_HEAD(all_mddevs);
 183 static DEFINE_SPINLOCK(all_mddevs_lock);
 184
 185
 186 /*
 187  * iterates through all used mddevs in the system.
 188  * We take care to grab the all_mddevs_lock whenever navigating
 189  * the list, and to always hold a refcount when unlocked.
 190  * Any code which breaks out of this loop while own
 191  * a reference to the current mddev and must mddev_put it.
 192  */
 193 #define for_each_mddev(mddev,tmp)                                       \
 194                                                                         \
 195         for (({ spin_lock(&all_mddevs_lock);                            \
 196                 tmp = all_mddevs.next;                                  \
 197                 mddev = NULL;});                                        \
 198              ({ if (tmp != &all_mddevs)                                 \
 199                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 200                 spin_unlock(&all_mddevs_lock);                          \
 201                 if (mddev) mddev_put(mddev);                            \
 202                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 203                 tmp != &all_mddevs;});                                  \
 204              ({ spin_lock(&all_mddevs_lock);                            \
 205                 tmp = tmp->next;})                                      \
 206                 )
 207
 208
 209 /* Rather than calling directly into the personality make_request function,
 210  * IO requests come here first so that we can check if the device is
 211  * being suspended pending a reconfiguration.
 212  * We hold a refcount over the call to ->make_request.  By the time that
 213  * call has finished, the bio has been linked into some internal structure
 214  * and so is visible to ->quiesce(), so we don't need the refcount any more.
 215  */
 216 static int md_make_request(struct request_queue *q, struct bio *bio)
 217 {
 218         const int rw = bio_data_dir(bio);
 219         mddev_t *mddev = q->queuedata;
 220         int rv;
 221         int cpu;
 222
 223         if (mddev == NULL || mddev->pers == NULL) {
 224                 bio_io_error(bio);
 225                 return 0;
 226         }
 227         rcu_read_lock();
 228         if (mddev->suspended || mddev->barrier) {
 229                 DEFINE_WAIT(__wait);
 230                 for (;;) {
 231                         prepare_to_wait(&mddev->sb_wait, &__wait,
 232                                         TASK_UNINTERRUPTIBLE);
 233                         if (!mddev->suspended && !mddev->barrier)
 234                                 break;
 235                         rcu_read_unlock();
 236                         schedule();
 237                         rcu_read_lock();
 238                 }
 239                 finish_wait(&mddev->sb_wait, &__wait);
 240         }
 241         atomic_inc(&mddev->active_io);
 242         rcu_read_unlock();
 243
 244         rv = mddev->pers->make_request(mddev, bio);
 245
 246         cpu = part_stat_lock();
 247         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 248         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 249                       bio_sectors(bio));
 250         part_stat_unlock();
 251
 252         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 253                 wake_up(&mddev->sb_wait);
 254
 255         return rv;
 256 }
 257
 258 /* mddev_suspend makes sure no new requests are submitted
 259  * to the device, and that any requests that have been submitted
 260  * are completely handled.
 261  * Once ->stop is called and completes, the module will be completely
 262  * unused.
 263  */
 264 static void mddev_suspend(mddev_t *mddev)
 265 {
 266         BUG_ON(mddev->suspended);
 267         mddev->suspended = 1;
 268         synchronize_rcu();
 269         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 270         mddev->pers->quiesce(mddev, 1);
 271 }
 272
 273 static void mddev_resume(mddev_t *mddev)
 274 {
 275         mddev->suspended = 0;
 276         wake_up(&mddev->sb_wait);
 277         mddev->pers->quiesce(mddev, 0);
 278 }
 279
 280 int mddev_congested(mddev_t *mddev, int bits)
 281 {
 282         if (mddev->barrier)
 283                 return 1;
 284         return mddev->suspended;
 285 }
 286 EXPORT_SYMBOL(mddev_congested);
 287
 288 /*
 289  * Generic barrier handling for md
 290  */
 291
 292 #define POST_REQUEST_BARRIER ((void*)1)
 293
 294 static void md_end_barrier(struct bio *bio, int err)
 295 {
 296         mdk_rdev_t *rdev = bio->bi_private;
 297         mddev_t *mddev = rdev->mddev;
 298         if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
 299                 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
 300
 301         rdev_dec_pending(rdev, mddev);
 302
 303         if (atomic_dec_and_test(&mddev->flush_pending)) {
 304                 if (mddev->barrier == POST_REQUEST_BARRIER) {
 305                         /* This was a post-request barrier */
 306                         mddev->barrier = NULL;
 307                         wake_up(&mddev->sb_wait);
 308                 } else
 309                         /* The pre-request barrier has finished */
 310                         schedule_work(&mddev->barrier_work);
 311         }
 312         bio_put(bio);
 313 }
 314
 315 static void submit_barriers(mddev_t *mddev)
 316 {
 317         mdk_rdev_t *rdev;
 318
 319         rcu_read_lock();
 320         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
 321                 if (rdev->raid_disk >= 0 &&
 322                     !test_bit(Faulty, &rdev->flags)) {
 323                         /* Take two references, one is dropped
 324                          * when request finishes, one after
 325                          * we reclaim rcu_read_lock
 326                          */
 327                         struct bio *bi;
 328                         atomic_inc(&rdev->nr_pending);
 329                         atomic_inc(&rdev->nr_pending);
 330                         rcu_read_unlock();
 331                         bi = bio_alloc(GFP_KERNEL, 0);
 332                         bi->bi_end_io = md_end_barrier;
 333                         bi->bi_private = rdev;
 334                         bi->bi_bdev = rdev->bdev;
 335                         atomic_inc(&mddev->flush_pending);
 336                         submit_bio(WRITE_BARRIER, bi);
 337                         rcu_read_lock();
 338                         rdev_dec_pending(rdev, mddev);
 339                 }
 340         rcu_read_unlock();
 341 }
 342
 343 static void md_submit_barrier(struct work_struct *ws)
 344 {
 345         mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
 346         struct bio *bio = mddev->barrier;
 347
 348         atomic_set(&mddev->flush_pending, 1);
 349
 350         if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
 351                 bio_endio(bio, -EOPNOTSUPP);
 352         else if (bio->bi_size == 0)
 353                 /* an empty barrier - all done */
 354                 bio_endio(bio, 0);
 355         else {
 356                 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
 357                 if (mddev->pers->make_request(mddev, bio))
 358                         generic_make_request(bio);
 359                 mddev->barrier = POST_REQUEST_BARRIER;
 360                 submit_barriers(mddev);
 361         }
 362         if (atomic_dec_and_test(&mddev->flush_pending)) {
 363                 mddev->barrier = NULL;
 364                 wake_up(&mddev->sb_wait);
 365         }
 366 }
 367
 368 void md_barrier_request(mddev_t *mddev, struct bio *bio)
 369 {
 370         spin_lock_irq(&mddev->write_lock);
 371         wait_event_lock_irq(mddev->sb_wait,
 372                             !mddev->barrier,
 373                             mddev->write_lock, /*nothing*/);
 374         mddev->barrier = bio;
 375         spin_unlock_irq(&mddev->write_lock);
 376
 377         atomic_set(&mddev->flush_pending, 1);
 378         INIT_WORK(&mddev->barrier_work, md_submit_barrier);
 379
 380         submit_barriers(mddev);
 381
 382         if (atomic_dec_and_test(&mddev->flush_pending))
 383                 schedule_work(&mddev->barrier_work);
 384 }
 385 EXPORT_SYMBOL(md_barrier_request);
 386
 387 static inline mddev_t *mddev_get(mddev_t *mddev)
 388 {
 389         atomic_inc(&mddev->active);
 390         return mddev;
 391 }
 392
 393 static void mddev_delayed_delete(struct work_struct *ws);
 394
 395 static void mddev_put(mddev_t *mddev)
 396 {
 397         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 398                 return;
 399         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 400             mddev->ctime == 0 && !mddev->hold_active) {
 401                 /* Array is not configured at all, and not held active,
 402                  * so destroy it */
 403                 list_del(&mddev->all_mddevs);
 404                 if (mddev->gendisk) {
 405                         /* we did a probe so need to clean up.
 406                          * Call schedule_work inside the spinlock
 407                          * so that flush_scheduled_work() after
 408                          * mddev_find will succeed in waiting for the
 409                          * work to be done.
 410                          */
 411                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 412                         schedule_work(&mddev->del_work);
 413                 } else
 414                         kfree(mddev);
 415         }
 416         spin_unlock(&all_mddevs_lock);
 417 }
 418
 419 static void mddev_init(mddev_t *mddev)
 420 {
 421         mutex_init(&mddev->open_mutex);
 422         mutex_init(&mddev->reconfig_mutex);
 423         mutex_init(&mddev->bitmap_info.mutex);
 424         INIT_LIST_HEAD(&mddev->disks);
 425         INIT_LIST_HEAD(&mddev->all_mddevs);
 426         init_timer(&mddev->safemode_timer);
 427         atomic_set(&mddev->active, 1);
 428         atomic_set(&mddev->openers, 0);
 429         atomic_set(&mddev->active_io, 0);
 430         spin_lock_init(&mddev->write_lock);
 431         atomic_set(&mddev->flush_pending, 0);
 432         init_waitqueue_head(&mddev->sb_wait);
 433         init_waitqueue_head(&mddev->recovery_wait);
 434         mddev->reshape_position = MaxSector;
 435         mddev->resync_min = 0;
 436         mddev->resync_max = MaxSector;
 437         mddev->level = LEVEL_NONE;
 438 }
 439
 440 static mddev_t * mddev_find(dev_t unit)
 441 {
 442         mddev_t *mddev, *new = NULL;
 443
 444  retry:
 445         spin_lock(&all_mddevs_lock);
 446
 447         if (unit) {
 448                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 449                         if (mddev->unit == unit) {
 450                                 mddev_get(mddev);
 451                                 spin_unlock(&all_mddevs_lock);
 452                                 kfree(new);
 453                                 return mddev;
 454                         }
 455
 456                 if (new) {
 457                         list_add(&new->all_mddevs, &all_mddevs);
 458                         spin_unlock(&all_mddevs_lock);
 459                         new->hold_active = UNTIL_IOCTL;
 460                         return new;
 461                 }
 462         } else if (new) {
 463                 /* find an unused unit number */
 464                 static int next_minor = 512;
 465                 int start = next_minor;
 466                 int is_free = 0;
 467                 int dev = 0;
 468                 while (!is_free) {
 469                         dev = MKDEV(MD_MAJOR, next_minor);
 470                         next_minor++;
 471                         if (next_minor > MINORMASK)
 472                                 next_minor = 0;
 473                         if (next_minor == start) {
 474                                 /* Oh dear, all in use. */
 475                                 spin_unlock(&all_mddevs_lock);
 476                                 kfree(new);
 477                                 return NULL;
 478                         }
 479
 480                         is_free = 1;
 481                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 482                                 if (mddev->unit == dev) {
 483                                         is_free = 0;
 484                                         break;
 485                                 }
 486                 }
 487                 new->unit = dev;
 488                 new->md_minor = MINOR(dev);
 489                 new->hold_active = UNTIL_STOP;
 490                 list_add(&new->all_mddevs, &all_mddevs);
 491                 spin_unlock(&all_mddevs_lock);
 492                 return new;
 493         }
 494         spin_unlock(&all_mddevs_lock);
 495
 496         new = kzalloc(sizeof(*new), GFP_KERNEL);
 497         if (!new)
 498                 return NULL;
 499
 500         new->unit = unit;
 501         if (MAJOR(unit) == MD_MAJOR)
 502                 new->md_minor = MINOR(unit);
 503         else
 504                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 505
 506         mddev_init(new);
 507
 508         goto retry;
 509 }
 510
 511 static inline int mddev_lock(mddev_t * mddev)
 512 {
 513         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 514 }
 515
 516 static inline int mddev_is_locked(mddev_t *mddev)
 517 {
 518         return mutex_is_locked(&mddev->reconfig_mutex);
 519 }
 520
 521 static inline int mddev_trylock(mddev_t * mddev)
 522 {
 523         return mutex_trylock(&mddev->reconfig_mutex);
 524 }
 525
 526 static struct attribute_group md_redundancy_group;
 527
 528 static void mddev_unlock(mddev_t * mddev)
 529 {
 530         if (mddev->to_remove) {
 531                 /* These cannot be removed under reconfig_mutex as
 532                  * an access to the files will try to take reconfig_mutex
 533                  * while holding the file unremovable, which leads to
 534                  * a deadlock.
 535                  * So hold open_mutex instead - we are allowed to take
 536                  * it while holding reconfig_mutex, and md_run can
 537                  * use it to wait for the remove to complete.
 538                  */
 539                 struct attribute_group *to_remove = mddev->to_remove;
 540                 mddev->to_remove = NULL;
 541                 mutex_lock(&mddev->open_mutex);
 542                 mutex_unlock(&mddev->reconfig_mutex);
 543
 544                 if (mddev->kobj.sd) {
 545                         if (to_remove != &md_redundancy_group)
 546                                 sysfs_remove_group(&mddev->kobj, to_remove);
 547                         if (mddev->pers == NULL ||
 548                             mddev->pers->sync_request == NULL) {
 549                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 550                                 if (mddev->sysfs_action)
 551                                         sysfs_put(mddev->sysfs_action);
 552                                 mddev->sysfs_action = NULL;
 553                         }
 554                 }
 555                 mutex_unlock(&mddev->open_mutex);
 556         } else
 557                 mutex_unlock(&mddev->reconfig_mutex);
 558
 559         md_wakeup_thread(mddev->thread);
 560 }
 561
 562 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 563 {
 564         mdk_rdev_t *rdev;
 565
 566         list_for_each_entry(rdev, &mddev->disks, same_set)
 567                 if (rdev->desc_nr == nr)
 568                         return rdev;
 569
 570         return NULL;
 571 }
 572
 573 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 574 {
 575         mdk_rdev_t *rdev;
 576
 577         list_for_each_entry(rdev, &mddev->disks, same_set)
 578                 if (rdev->bdev->bd_dev == dev)
 579                         return rdev;
 580
 581         return NULL;
 582 }
 583
 584 static struct mdk_personality *find_pers(int level, char *clevel)
 585 {
 586         struct mdk_personality *pers;
 587         list_for_each_entry(pers, &pers_list, list) {
 588                 if (level != LEVEL_NONE && pers->level == level)
 589                         return pers;
 590                 if (strcmp(pers->name, clevel)==0)
 591                         return pers;
 592         }
 593         return NULL;
 594 }
 595
 596 /* return the offset of the super block in 512byte sectors */
 597 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 598 {
 599         sector_t num_sectors = bdev->bd_inode->i_size / 512;
 600         return MD_NEW_SIZE_SECTORS(num_sectors);
 601 }
 602
 603 static int alloc_disk_sb(mdk_rdev_t * rdev)
 604 {
 605         if (rdev->sb_page)
 606                 MD_BUG();
 607
 608         rdev->sb_page = alloc_page(GFP_KERNEL);
 609         if (!rdev->sb_page) {
 610                 printk(KERN_ALERT "md: out of memory.\n");
 611                 return -ENOMEM;
 612         }
 613
 614         return 0;
 615 }
 616
 617 static void free_disk_sb(mdk_rdev_t * rdev)
 618 {
 619         if (rdev->sb_page) {
 620                 put_page(rdev->sb_page);
 621                 rdev->sb_loaded = 0;
 622                 rdev->sb_page = NULL;
 623                 rdev->sb_start = 0;
 624                 rdev->sectors = 0;
 625         }
 626 }
 627
 628
 629 static void super_written(struct bio *bio, int error)
 630 {
 631         mdk_rdev_t *rdev = bio->bi_private;
 632         mddev_t *mddev = rdev->mddev;
 633
 634         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 635                 printk("md: super_written gets error=%d, uptodate=%d\n",
 636                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 637                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 638                 md_error(mddev, rdev);
 639         }
 640
 641         if (atomic_dec_and_test(&mddev->pending_writes))
 642                 wake_up(&mddev->sb_wait);
 643         bio_put(bio);
 644 }
 645
 646 static void super_written_barrier(struct bio *bio, int error)
 647 {
 648         struct bio *bio2 = bio->bi_private;
 649         mdk_rdev_t *rdev = bio2->bi_private;
 650         mddev_t *mddev = rdev->mddev;
 651
 652         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 653             error == -EOPNOTSUPP) {
 654                 unsigned long flags;
 655                 /* barriers don't appear to be supported :-( */
 656                 set_bit(BarriersNotsupp, &rdev->flags);
 657                 mddev->barriers_work = 0;
 658                 spin_lock_irqsave(&mddev->write_lock, flags);
 659                 bio2->bi_next = mddev->biolist;
 660                 mddev->biolist = bio2;
 661                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 662                 wake_up(&mddev->sb_wait);
 663                 bio_put(bio);
 664         } else {
 665                 bio_put(bio2);
 666                 bio->bi_private = rdev;
 667                 super_written(bio, error);
 668         }
 669 }
 670
 671 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 672                    sector_t sector, int size, struct page *page)
 673 {
 674         /* write first size bytes of page to sector of rdev
 675          * Increment mddev->pending_writes before returning
 676          * and decrement it on completion, waking up sb_wait
 677          * if zero is reached.
 678          * If an error occurred, call md_error
 679          *
 680          * As we might need to resubmit the request if BIO_RW_BARRIER
 681          * causes ENOTSUPP, we allocate a spare bio...
 682          */
 683         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 684         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
 685
 686         bio->bi_bdev = rdev->bdev;
 687         bio->bi_sector = sector;
 688         bio_add_page(bio, page, size, 0);
 689         bio->bi_private = rdev;
 690         bio->bi_end_io = super_written;
 691         bio->bi_rw = rw;
 692
 693         atomic_inc(&mddev->pending_writes);
 694         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 695                 struct bio *rbio;
 696                 rw |= (1<<BIO_RW_BARRIER);
 697                 rbio = bio_clone(bio, GFP_NOIO);
 698                 rbio->bi_private = bio;
 699                 rbio->bi_end_io = super_written_barrier;
 700                 submit_bio(rw, rbio);
 701         } else
 702                 submit_bio(rw, bio);
 703 }
 704
 705 void md_super_wait(mddev_t *mddev)
 706 {
 707         /* wait for all superblock writes that were scheduled to complete.
 708          * if any had to be retried (due to BARRIER problems), retry them
 709          */
 710         DEFINE_WAIT(wq);
 711         for(;;) {
 712                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 713                 if (atomic_read(&mddev->pending_writes)==0)
 714                         break;
 715                 while (mddev->biolist) {
 716                         struct bio *bio;
 717                         spin_lock_irq(&mddev->write_lock);
 718                         bio = mddev->biolist;
 719                         mddev->biolist = bio->bi_next ;
 720                         bio->bi_next = NULL;
 721                         spin_unlock_irq(&mddev->write_lock);
 722                         submit_bio(bio->bi_rw, bio);
 723                 }
 724                 schedule();
 725         }
 726         finish_wait(&mddev->sb_wait, &wq);
 727 }
 728
 729 static void bi_complete(struct bio *bio, int error)
 730 {
 731         complete((struct completion*)bio->bi_private);
 732 }
 733
 734 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 735                    struct page *page, int rw)
 736 {
 737         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 738         struct completion event;
 739         int ret;
 740
 741         rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
 742
 743         bio->bi_bdev = bdev;
 744         bio->bi_sector = sector;
 745         bio_add_page(bio, page, size, 0);
 746         init_completion(&event);
 747         bio->bi_private = &event;
 748         bio->bi_end_io = bi_complete;
 749         submit_bio(rw, bio);
 750         wait_for_completion(&event);
 751
 752         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 753         bio_put(bio);
 754         return ret;
 755 }
 756 EXPORT_SYMBOL_GPL(sync_page_io);
 757
 758 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 759 {
 760         char b[BDEVNAME_SIZE];
 761         if (!rdev->sb_page) {
 762                 MD_BUG();
 763                 return -EINVAL;
 764         }
 765         if (rdev->sb_loaded)
 766                 return 0;
 767
 768
 769         if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
 770                 goto fail;
 771         rdev->sb_loaded = 1;
 772         return 0;
 773
 774 fail:
 775         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 776                 bdevname(rdev->bdev,b));
 777         return -EINVAL;
 778 }
 779
 780 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 781 {
 782         return  sb1->set_uuid0 == sb2->set_uuid0 &&
 783                 sb1->set_uuid1 == sb2->set_uuid1 &&
 784                 sb1->set_uuid2 == sb2->set_uuid2 &&
 785                 sb1->set_uuid3 == sb2->set_uuid3;
 786 }
 787
 788 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 789 {
 790         int ret;
 791         mdp_super_t *tmp1, *tmp2;
 792
 793         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 794         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 795
 796         if (!tmp1 || !tmp2) {
 797                 ret = 0;
 798                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 799                 goto abort;
 800         }
 801
 802         *tmp1 = *sb1;
 803         *tmp2 = *sb2;
 804
 805         /*
 806          * nr_disks is not constant
 807          */
 808         tmp1->nr_disks = 0;
 809         tmp2->nr_disks = 0;
 810
 811         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 812 abort:
 813         kfree(tmp1);
 814         kfree(tmp2);
 815         return ret;
 816 }
 817
 818
 819 static u32 md_csum_fold(u32 csum)
 820 {
 821         csum = (csum & 0xffff) + (csum >> 16);
 822         return (csum & 0xffff) + (csum >> 16);
 823 }
 824
 825 static unsigned int calc_sb_csum(mdp_super_t * sb)
 826 {
 827         u64 newcsum = 0;
 828         u32 *sb32 = (u32*)sb;
 829         int i;
 830         unsigned int disk_csum, csum;
 831
 832         disk_csum = sb->sb_csum;
 833         sb->sb_csum = 0;
 834
 835         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 836                 newcsum += sb32[i];
 837         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 838
 839
 840 #ifdef CONFIG_ALPHA
 841         /* This used to use csum_partial, which was wrong for several
 842          * reasons including that different results are returned on
 843          * different architectures.  It isn't critical that we get exactly
 844          * the same return value as before (we always csum_fold before
 845          * testing, and that removes any differences).  However as we
 846          * know that csum_partial always returned a 16bit value on
 847          * alphas, do a fold to maximise conformity to previous behaviour.
 848          */
 849         sb->sb_csum = md_csum_fold(disk_csum);
 850 #else
 851         sb->sb_csum = disk_csum;
 852 #endif
 853         return csum;
 854 }
 855
 856
 857 /*
 858  * Handle superblock details.
 859  * We want to be able to handle multiple superblock formats
 860  * so we have a common interface to them all, and an array of
 861  * different handlers.
 862  * We rely on user-space to write the initial superblock, and support
 863  * reading and updating of superblocks.
 864  * Interface methods are:
 865  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 866  *      loads and validates a superblock on dev.
 867  *      if refdev != NULL, compare superblocks on both devices
 868  *    Return:
 869  *      0 - dev has a superblock that is compatible with refdev
 870  *      1 - dev has a superblock that is compatible and newer than refdev
 871  *          so dev should be used as the refdev in future
 872  *     -EINVAL superblock incompatible or invalid
 873  *     -othererror e.g. -EIO
 874  *
 875  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 876  *      Verify that dev is acceptable into mddev.
 877  *       The first time, mddev->raid_disks will be 0, and data from
 878  *       dev should be merged in.  Subsequent calls check that dev
 879  *       is new enough.  Return 0 or -EINVAL
 880  *
 881  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 882  *     Update the superblock for rdev with data in mddev
 883  *     This does not write to disc.
 884  *
 885  */
 886
 887 struct super_type  {
 888         char                *name;
 889         struct module       *owner;
 890         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
 891                                           int minor_version);
 892         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 893         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 894         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
 895                                                 sector_t num_sectors);
 896 };
 897
 898 /*
 899  * Check that the given mddev has no bitmap.
 900  *
 901  * This function is called from the run method of all personalities that do not
 902  * support bitmaps. It prints an error message and returns non-zero if mddev
 903  * has a bitmap. Otherwise, it returns 0.
 904  *
 905  */
 906 int md_check_no_bitmap(mddev_t *mddev)
 907 {
 908         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
 909                 return 0;
 910         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
 911                 mdname(mddev), mddev->pers->name);
 912         return 1;
 913 }
 914 EXPORT_SYMBOL(md_check_no_bitmap);
 915
 916 /*
 917  * load_super for 0.90.0
 918  */
 919 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 920 {
 921         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 922         mdp_super_t *sb;
 923         int ret;
 924
 925         /*
 926          * Calculate the position of the superblock (512byte sectors),
 927          * it's at the end of the disk.
 928          *
 929          * It also happens to be a multiple of 4Kb.
 930          */
 931         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
 932
 933         ret = read_disk_sb(rdev, MD_SB_BYTES);
 934         if (ret) return ret;
 935
 936         ret = -EINVAL;
 937
 938         bdevname(rdev->bdev, b);
 939         sb = (mdp_super_t*)page_address(rdev->sb_page);
 940
 941         if (sb->md_magic != MD_SB_MAGIC) {
 942                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 943                        b);
 944                 goto abort;
 945         }
 946
 947         if (sb->major_version != 0 ||
 948             sb->minor_version < 90 ||
 949             sb->minor_version > 91) {
 950                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 951                         sb->major_version, sb->minor_version,
 952                         b);
 953                 goto abort;
 954         }
 955
 956         if (sb->raid_disks <= 0)
 957                 goto abort;
 958
 959         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 960                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 961                         b);
 962                 goto abort;
 963         }
 964
 965         rdev->preferred_minor = sb->md_minor;
 966         rdev->data_offset = 0;
 967         rdev->sb_size = MD_SB_BYTES;
 968
 969         if (sb->level == LEVEL_MULTIPATH)
 970                 rdev->desc_nr = -1;
 971         else
 972                 rdev->desc_nr = sb->this_disk.number;
 973
 974         if (!refdev) {
 975                 ret = 1;
 976         } else {
 977                 __u64 ev1, ev2;
 978                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 979                 if (!uuid_equal(refsb, sb)) {
 980                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 981                                 b, bdevname(refdev->bdev,b2));
 982                         goto abort;
 983                 }
 984                 if (!sb_equal(refsb, sb)) {
 985                         printk(KERN_WARNING "md: %s has same UUID"
 986                                " but different superblock to %s\n",
 987                                b, bdevname(refdev->bdev, b2));
 988                         goto abort;
 989                 }
 990                 ev1 = md_event(sb);
 991                 ev2 = md_event(refsb);
 992                 if (ev1 > ev2)
 993                         ret = 1;
 994                 else
 995                         ret = 0;
 996         }
 997         rdev->sectors = rdev->sb_start;
 998
 999         if (rdev->sectors < sb->size * 2 && sb->level > 1)
1000                 /* "this cannot possibly happen" ... */
1001                 ret = -EINVAL;
1002
1003  abort:
1004         return ret;
1005 }
1006
1007 /*
1008  * validate_super for 0.90.0
1009  */
1010 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1011 {
1012         mdp_disk_t *desc;
1013         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1014         __u64 ev1 = md_event(sb);
1015
1016         rdev->raid_disk = -1;
1017         clear_bit(Faulty, &rdev->flags);
1018         clear_bit(In_sync, &rdev->flags);
1019         clear_bit(WriteMostly, &rdev->flags);
1020         clear_bit(BarriersNotsupp, &rdev->flags);
1021
1022         if (mddev->raid_disks == 0) {
1023                 mddev->major_version = 0;
1024                 mddev->minor_version = sb->minor_version;
1025                 mddev->patch_version = sb->patch_version;
1026                 mddev->external = 0;
1027                 mddev->chunk_sectors = sb->chunk_size >> 9;
1028                 mddev->ctime = sb->ctime;
1029                 mddev->utime = sb->utime;
1030                 mddev->level = sb->level;
1031                 mddev->clevel[0] = 0;
1032                 mddev->layout = sb->layout;
1033                 mddev->raid_disks = sb->raid_disks;
1034                 mddev->dev_sectors = sb->size * 2;
1035                 mddev->events = ev1;
1036                 mddev->bitmap_info.offset = 0;
1037                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1038
1039                 if (mddev->minor_version >= 91) {
1040                         mddev->reshape_position = sb->reshape_position;
1041                         mddev->delta_disks = sb->delta_disks;
1042                         mddev->new_level = sb->new_level;
1043                         mddev->new_layout = sb->new_layout;
1044                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1045                 } else {
1046                         mddev->reshape_position = MaxSector;
1047                         mddev->delta_disks = 0;
1048                         mddev->new_level = mddev->level;
1049                         mddev->new_layout = mddev->layout;
1050                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1051                 }
1052
1053                 if (sb->state & (1<<MD_SB_CLEAN))
1054                         mddev->recovery_cp = MaxSector;
1055                 else {
1056                         if (sb->events_hi == sb->cp_events_hi &&
1057                                 sb->events_lo == sb->cp_events_lo) {
1058                                 mddev->recovery_cp = sb->recovery_cp;
1059                         } else
1060                                 mddev->recovery_cp = 0;
1061                 }
1062
1063                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1064                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1065                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1066                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1067
1068                 mddev->max_disks = MD_SB_DISKS;
1069
1070                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1071                     mddev->bitmap_info.file == NULL)
1072                         mddev->bitmap_info.offset =
1073                                 mddev->bitmap_info.default_offset;
1074
1075         } else if (mddev->pers == NULL) {
1076                 /* Insist on good event counter while assembling, except
1077                  * for spares (which don't need an event count) */
1078                 ++ev1;
1079                 if (sb->disks[rdev->desc_nr].state & (
1080                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1081                         if (ev1 < mddev->events)
1082                                 return -EINVAL;
1083         } else if (mddev->bitmap) {
1084                 /* if adding to array with a bitmap, then we can accept an
1085                  * older device ... but not too old.
1086                  */
1087                 if (ev1 < mddev->bitmap->events_cleared)
1088                         return 0;
1089         } else {
1090                 if (ev1 < mddev->events)
1091                         /* just a hot-add of a new device, leave raid_disk at -1 */
1092                         return 0;
1093         }
1094
1095         if (mddev->level != LEVEL_MULTIPATH) {
1096                 desc = sb->disks + rdev->desc_nr;
1097
1098                 if (desc->state & (1<<MD_DISK_FAULTY))
1099                         set_bit(Faulty, &rdev->flags);
1100                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1101                             desc->raid_disk < mddev->raid_disks */) {
1102                         set_bit(In_sync, &rdev->flags);
1103                         rdev->raid_disk = desc->raid_disk;
1104                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1105                         /* active but not in sync implies recovery up to
1106                          * reshape position.  We don't know exactly where
1107                          * that is, so set to zero for now */
1108                         if (mddev->minor_version >= 91) {
1109                                 rdev->recovery_offset = 0;
1110                                 rdev->raid_disk = desc->raid_disk;
1111                         }
1112                 }
1113                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1114                         set_bit(WriteMostly, &rdev->flags);
1115         } else /* MULTIPATH are always insync */
1116                 set_bit(In_sync, &rdev->flags);
1117         return 0;
1118 }
1119
1120 /*
1121  * sync_super for 0.90.0
1122  */
1123 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1124 {
1125         mdp_super_t *sb;
1126         mdk_rdev_t *rdev2;
1127         int next_spare = mddev->raid_disks;
1128
1129
1130         /* make rdev->sb match mddev data..
1131          *
1132          * 1/ zero out disks
1133          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1134          * 3/ any empty disks < next_spare become removed
1135          *
1136          * disks[0] gets initialised to REMOVED because
1137          * we cannot be sure from other fields if it has
1138          * been initialised or not.
1139          */
1140         int i;
1141         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1142
1143         rdev->sb_size = MD_SB_BYTES;
1144
1145         sb = (mdp_super_t*)page_address(rdev->sb_page);
1146
1147         memset(sb, 0, sizeof(*sb));
1148
1149         sb->md_magic = MD_SB_MAGIC;
1150         sb->major_version = mddev->major_version;
1151         sb->patch_version = mddev->patch_version;
1152         sb->gvalid_words  = 0; /* ignored */
1153         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1154         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1155         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1156         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1157
1158         sb->ctime = mddev->ctime;
1159         sb->level = mddev->level;
1160         sb->size = mddev->dev_sectors / 2;
1161         sb->raid_disks = mddev->raid_disks;
1162         sb->md_minor = mddev->md_minor;
1163         sb->not_persistent = 0;
1164         sb->utime = mddev->utime;
1165         sb->state = 0;
1166         sb->events_hi = (mddev->events>>32);
1167         sb->events_lo = (u32)mddev->events;
1168
1169         if (mddev->reshape_position == MaxSector)
1170                 sb->minor_version = 90;
1171         else {
1172                 sb->minor_version = 91;
1173                 sb->reshape_position = mddev->reshape_position;
1174                 sb->new_level = mddev->new_level;
1175                 sb->delta_disks = mddev->delta_disks;
1176                 sb->new_layout = mddev->new_layout;
1177                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1178         }
1179         mddev->minor_version = sb->minor_version;
1180         if (mddev->in_sync)
1181         {
1182                 sb->recovery_cp = mddev->recovery_cp;
1183                 sb->cp_events_hi = (mddev->events>>32);
1184                 sb->cp_events_lo = (u32)mddev->events;
1185                 if (mddev->recovery_cp == MaxSector)
1186                         sb->state = (1<< MD_SB_CLEAN);
1187         } else
1188                 sb->recovery_cp = 0;
1189
1190         sb->layout = mddev->layout;
1191         sb->chunk_size = mddev->chunk_sectors << 9;
1192
1193         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1194                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1195
1196         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1197         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1198                 mdp_disk_t *d;
1199                 int desc_nr;
1200                 int is_active = test_bit(In_sync, &rdev2->flags);
1201
1202                 if (rdev2->raid_disk >= 0 &&
1203                     sb->minor_version >= 91)
1204                         /* we have nowhere to store the recovery_offset,
1205                          * but if it is not below the reshape_position,
1206                          * we can piggy-back on that.
1207                          */
1208                         is_active = 1;
1209                 if (rdev2->raid_disk < 0 ||
1210                     test_bit(Faulty, &rdev2->flags))
1211                         is_active = 0;
1212                 if (is_active)
1213                         desc_nr = rdev2->raid_disk;
1214                 else
1215                         desc_nr = next_spare++;
1216                 rdev2->desc_nr = desc_nr;
1217                 d = &sb->disks[rdev2->desc_nr];
1218                 nr_disks++;
1219                 d->number = rdev2->desc_nr;
1220                 d->major = MAJOR(rdev2->bdev->bd_dev);
1221                 d->minor = MINOR(rdev2->bdev->bd_dev);
1222                 if (is_active)
1223                         d->raid_disk = rdev2->raid_disk;
1224                 else
1225                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1226                 if (test_bit(Faulty, &rdev2->flags))
1227                         d->state = (1<<MD_DISK_FAULTY);
1228                 else if (is_active) {
1229                         d->state = (1<<MD_DISK_ACTIVE);
1230                         if (test_bit(In_sync, &rdev2->flags))
1231                                 d->state |= (1<<MD_DISK_SYNC);
1232                         active++;
1233                         working++;
1234                 } else {
1235                         d->state = 0;
1236                         spare++;
1237                         working++;
1238                 }
1239                 if (test_bit(WriteMostly, &rdev2->flags))
1240                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1241         }
1242         /* now set the "removed" and "faulty" bits on any missing devices */
1243         for (i=0 ; i < mddev->raid_disks ; i++) {
1244                 mdp_disk_t *d = &sb->disks[i];
1245                 if (d->state == 0 && d->number == 0) {
1246                         d->number = i;
1247                         d->raid_disk = i;
1248                         d->state = (1<<MD_DISK_REMOVED);
1249                         d->state |= (1<<MD_DISK_FAULTY);
1250                         failed++;
1251                 }
1252         }
1253         sb->nr_disks = nr_disks;
1254         sb->active_disks = active;
1255         sb->working_disks = working;
1256         sb->failed_disks = failed;
1257         sb->spare_disks = spare;
1258
1259         sb->this_disk = sb->disks[rdev->desc_nr];
1260         sb->sb_csum = calc_sb_csum(sb);
1261 }
1262
1263 /*
1264  * rdev_size_change for 0.90.0
1265  */
1266 static unsigned long long
1267 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1268 {
1269         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1270                 return 0; /* component must fit device */
1271         if (rdev->mddev->bitmap_info.offset)
1272                 return 0; /* can't move bitmap */
1273         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1274         if (!num_sectors || num_sectors > rdev->sb_start)
1275                 num_sectors = rdev->sb_start;
1276         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1277                        rdev->sb_page);
1278         md_super_wait(rdev->mddev);
1279         return num_sectors / 2; /* kB for sysfs */
1280 }
1281
1282
1283 /*
1284  * version 1 superblock
1285  */
1286
1287 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1288 {
1289         __le32 disk_csum;
1290         u32 csum;
1291         unsigned long long newcsum;
1292         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1293         __le32 *isuper = (__le32*)sb;
1294         int i;
1295
1296         disk_csum = sb->sb_csum;
1297         sb->sb_csum = 0;
1298         newcsum = 0;
1299         for (i=0; size>=4; size -= 4 )
1300                 newcsum += le32_to_cpu(*isuper++);
1301
1302         if (size == 2)
1303                 newcsum += le16_to_cpu(*(__le16*) isuper);
1304
1305         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1306         sb->sb_csum = disk_csum;
1307         return cpu_to_le32(csum);
1308 }
1309
1310 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1311 {
1312         struct mdp_superblock_1 *sb;
1313         int ret;
1314         sector_t sb_start;
1315         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1316         int bmask;
1317
1318         /*
1319          * Calculate the position of the superblock in 512byte sectors.
1320          * It is always aligned to a 4K boundary and
1321          * depeding on minor_version, it can be:
1322          * 0: At least 8K, but less than 12K, from end of device
1323          * 1: At start of device
1324          * 2: 4K from start of device.
1325          */
1326         switch(minor_version) {
1327         case 0:
1328                 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1329                 sb_start -= 8*2;
1330                 sb_start &= ~(sector_t)(4*2-1);
1331                 break;
1332         case 1:
1333                 sb_start = 0;
1334                 break;
1335         case 2:
1336                 sb_start = 8;
1337                 break;
1338         default:
1339                 return -EINVAL;
1340         }
1341         rdev->sb_start = sb_start;
1342
1343         /* superblock is rarely larger than 1K, but it can be larger,
1344          * and it is safe to read 4k, so we do that
1345          */
1346         ret = read_disk_sb(rdev, 4096);
1347         if (ret) return ret;
1348
1349
1350         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1351
1352         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1353             sb->major_version != cpu_to_le32(1) ||
1354             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1355             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1356             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1357                 return -EINVAL;
1358
1359         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1360                 printk("md: invalid superblock checksum on %s\n",
1361                         bdevname(rdev->bdev,b));
1362                 return -EINVAL;
1363         }
1364         if (le64_to_cpu(sb->data_size) < 10) {
1365                 printk("md: data_size too small on %s\n",
1366                        bdevname(rdev->bdev,b));
1367                 return -EINVAL;
1368         }
1369
1370         rdev->preferred_minor = 0xffff;
1371         rdev->data_offset = le64_to_cpu(sb->data_offset);
1372         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1373
1374         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1375         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1376         if (rdev->sb_size & bmask)
1377                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1378
1379         if (minor_version
1380             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1381                 return -EINVAL;
1382
1383         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1384                 rdev->desc_nr = -1;
1385         else
1386                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1387
1388         if (!refdev) {
1389                 ret = 1;
1390         } else {
1391                 __u64 ev1, ev2;
1392                 struct mdp_superblock_1 *refsb =
1393                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1394
1395                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1396                     sb->level != refsb->level ||
1397                     sb->layout != refsb->layout ||
1398                     sb->chunksize != refsb->chunksize) {
1399                         printk(KERN_WARNING "md: %s has strangely different"
1400                                 " superblock to %s\n",
1401                                 bdevname(rdev->bdev,b),
1402                                 bdevname(refdev->bdev,b2));
1403                         return -EINVAL;
1404                 }
1405                 ev1 = le64_to_cpu(sb->events);
1406                 ev2 = le64_to_cpu(refsb->events);
1407
1408                 if (ev1 > ev2)
1409                         ret = 1;
1410                 else
1411                         ret = 0;
1412         }
1413         if (minor_version)
1414                 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1415                         le64_to_cpu(sb->data_offset);
1416         else
1417                 rdev->sectors = rdev->sb_start;
1418         if (rdev->sectors < le64_to_cpu(sb->data_size))
1419                 return -EINVAL;
1420         rdev->sectors = le64_to_cpu(sb->data_size);
1421         if (le64_to_cpu(sb->size) > rdev->sectors)
1422                 return -EINVAL;
1423         return ret;
1424 }
1425
1426 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1427 {
1428         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1429         __u64 ev1 = le64_to_cpu(sb->events);
1430
1431         rdev->raid_disk = -1;
1432         clear_bit(Faulty, &rdev->flags);
1433         clear_bit(In_sync, &rdev->flags);
1434         clear_bit(WriteMostly, &rdev->flags);
1435         clear_bit(BarriersNotsupp, &rdev->flags);
1436
1437         if (mddev->raid_disks == 0) {
1438                 mddev->major_version = 1;
1439                 mddev->patch_version = 0;
1440                 mddev->external = 0;
1441                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1442                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1443                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1444                 mddev->level = le32_to_cpu(sb->level);
1445                 mddev->clevel[0] = 0;
1446                 mddev->layout = le32_to_cpu(sb->layout);
1447                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1448                 mddev->dev_sectors = le64_to_cpu(sb->size);
1449                 mddev->events = ev1;
1450                 mddev->bitmap_info.offset = 0;
1451                 mddev->bitmap_info.default_offset = 1024 >> 9;
1452
1453                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1454                 memcpy(mddev->uuid, sb->set_uuid, 16);
1455
1456                 mddev->max_disks =  (4096-256)/2;
1457
1458                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1459                     mddev->bitmap_info.file == NULL )
1460                         mddev->bitmap_info.offset =
1461                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1462
1463                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1464                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1465                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1466                         mddev->new_level = le32_to_cpu(sb->new_level);
1467                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1468                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1469                 } else {
1470                         mddev->reshape_position = MaxSector;
1471                         mddev->delta_disks = 0;
1472                         mddev->new_level = mddev->level;
1473                         mddev->new_layout = mddev->layout;
1474                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1475                 }
1476
1477         } else if (mddev->pers == NULL) {
1478                 /* Insist of good event counter while assembling, except for
1479                  * spares (which don't need an event count) */
1480                 ++ev1;
1481                 if (rdev->desc_nr >= 0 &&
1482                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1483                     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1484                         if (ev1 < mddev->events)
1485                                 return -EINVAL;
1486         } else if (mddev->bitmap) {
1487                 /* If adding to array with a bitmap, then we can accept an
1488                  * older device, but not too old.
1489                  */
1490                 if (ev1 < mddev->bitmap->events_cleared)
1491                         return 0;
1492         } else {
1493                 if (ev1 < mddev->events)
1494                         /* just a hot-add of a new device, leave raid_disk at -1 */
1495                         return 0;
1496         }
1497         if (mddev->level != LEVEL_MULTIPATH) {
1498                 int role;
1499                 if (rdev->desc_nr < 0 ||
1500                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1501                         role = 0xffff;
1502                         rdev->desc_nr = -1;
1503                 } else
1504                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1505                 switch(role) {
1506                 case 0xffff: /* spare */
1507                         break;
1508                 case 0xfffe: /* faulty */
1509                         set_bit(Faulty, &rdev->flags);
1510                         break;
1511                 default:
1512                         if ((le32_to_cpu(sb->feature_map) &
1513                              MD_FEATURE_RECOVERY_OFFSET))
1514                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1515                         else
1516                                 set_bit(In_sync, &rdev->flags);
1517                         rdev->raid_disk = role;
1518                         break;
1519                 }
1520                 if (sb->devflags & WriteMostly1)
1521                         set_bit(WriteMostly, &rdev->flags);
1522         } else /* MULTIPATH are always insync */
1523                 set_bit(In_sync, &rdev->flags);
1524
1525         return 0;
1526 }
1527
1528 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1529 {
1530         struct mdp_superblock_1 *sb;
1531         mdk_rdev_t *rdev2;
1532         int max_dev, i;
1533         /* make rdev->sb match mddev and rdev data. */
1534
1535         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1536
1537         sb->feature_map = 0;
1538         sb->pad0 = 0;
1539         sb->recovery_offset = cpu_to_le64(0);
1540         memset(sb->pad1, 0, sizeof(sb->pad1));
1541         memset(sb->pad2, 0, sizeof(sb->pad2));
1542         memset(sb->pad3, 0, sizeof(sb->pad3));
1543
1544         sb->utime = cpu_to_le64((__u64)mddev->utime);
1545         sb->events = cpu_to_le64(mddev->events);
1546         if (mddev->in_sync)
1547                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1548         else
1549                 sb->resync_offset = cpu_to_le64(0);
1550
1551         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1552
1553         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1554         sb->size = cpu_to_le64(mddev->dev_sectors);
1555         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1556         sb->level = cpu_to_le32(mddev->level);
1557         sb->layout = cpu_to_le32(mddev->layout);
1558
1559         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1560                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1561                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1562         }
1563
1564         if (rdev->raid_disk >= 0 &&
1565             !test_bit(In_sync, &rdev->flags)) {
1566                 sb->feature_map |=
1567                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1568                 sb->recovery_offset =
1569                         cpu_to_le64(rdev->recovery_offset);
1570         }
1571
1572         if (mddev->reshape_position != MaxSector) {
1573                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1574                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1575                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1576                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1577                 sb->new_level = cpu_to_le32(mddev->new_level);
1578                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1579         }
1580
1581         max_dev = 0;
1582         list_for_each_entry(rdev2, &mddev->disks, same_set)
1583                 if (rdev2->desc_nr+1 > max_dev)
1584                         max_dev = rdev2->desc_nr+1;
1585
1586         if (max_dev > le32_to_cpu(sb->max_dev)) {
1587                 int bmask;
1588                 sb->max_dev = cpu_to_le32(max_dev);
1589                 rdev->sb_size = max_dev * 2 + 256;
1590                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1591                 if (rdev->sb_size & bmask)
1592                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1593         }
1594         for (i=0; i<max_dev;i++)
1595                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1596
1597         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1598                 i = rdev2->desc_nr;
1599                 if (test_bit(Faulty, &rdev2->flags))
1600                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1601                 else if (test_bit(In_sync, &rdev2->flags))
1602                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1603                 else if (rdev2->raid_disk >= 0)
1604                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1605                 else
1606                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1607         }
1608
1609         sb->sb_csum = calc_sb_1_csum(sb);
1610 }
1611
1612 static unsigned long long
1613 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1614 {
1615         struct mdp_superblock_1 *sb;
1616         sector_t max_sectors;
1617         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1618                 return 0; /* component must fit device */
1619         if (rdev->sb_start < rdev->data_offset) {
1620                 /* minor versions 1 and 2; superblock before data */
1621                 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1622                 max_sectors -= rdev->data_offset;
1623                 if (!num_sectors || num_sectors > max_sectors)
1624                         num_sectors = max_sectors;
1625         } else if (rdev->mddev->bitmap_info.offset) {
1626                 /* minor version 0 with bitmap we can't move */
1627                 return 0;
1628         } else {
1629                 /* minor version 0; superblock after data */
1630                 sector_t sb_start;
1631                 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1632                 sb_start &= ~(sector_t)(4*2 - 1);
1633                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1634                 if (!num_sectors || num_sectors > max_sectors)
1635                         num_sectors = max_sectors;
1636                 rdev->sb_start = sb_start;
1637         }
1638         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1639         sb->data_size = cpu_to_le64(num_sectors);
1640         sb->super_offset = rdev->sb_start;
1641         sb->sb_csum = calc_sb_1_csum(sb);
1642         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1643                        rdev->sb_page);
1644         md_super_wait(rdev->mddev);
1645         return num_sectors / 2; /* kB for sysfs */
1646 }
1647
1648 static struct super_type super_types[] = {
1649         [0] = {
1650                 .name   = "0.90.0",
1651                 .owner  = THIS_MODULE,
1652                 .load_super         = super_90_load,
1653                 .validate_super     = super_90_validate,
1654                 .sync_super         = super_90_sync,
1655                 .rdev_size_change   = super_90_rdev_size_change,
1656         },
1657         [1] = {
1658                 .name   = "md-1",
1659                 .owner  = THIS_MODULE,
1660                 .load_super         = super_1_load,
1661                 .validate_super     = super_1_validate,
1662                 .sync_super         = super_1_sync,
1663                 .rdev_size_change   = super_1_rdev_size_change,
1664         },
1665 };
1666
1667 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1668 {
1669         mdk_rdev_t *rdev, *rdev2;
1670
1671         rcu_read_lock();
1672         rdev_for_each_rcu(rdev, mddev1)
1673                 rdev_for_each_rcu(rdev2, mddev2)
1674                         if (rdev->bdev->bd_contains ==
1675                             rdev2->bdev->bd_contains) {
1676                                 rcu_read_unlock();
1677                                 return 1;
1678                         }
1679         rcu_read_unlock();
1680         return 0;
1681 }
1682
1683 static LIST_HEAD(pending_raid_disks);
1684
1685 /*
1686  * Try to register data integrity profile for an mddev
1687  *
1688  * This is called when an array is started and after a disk has been kicked
1689  * from the array. It only succeeds if all working and active component devices
1690  * are integrity capable with matching profiles.
1691  */
1692 int md_integrity_register(mddev_t *mddev)
1693 {
1694         mdk_rdev_t *rdev, *reference = NULL;
1695
1696         if (list_empty(&mddev->disks))
1697                 return 0; /* nothing to do */
1698         if (blk_get_integrity(mddev->gendisk))
1699                 return 0; /* already registered */
1700         list_for_each_entry(rdev, &mddev->disks, same_set) {
1701                 /* skip spares and non-functional disks */
1702                 if (test_bit(Faulty, &rdev->flags))
1703                         continue;
1704                 if (rdev->raid_disk < 0)
1705                         continue;
1706                 /*
1707                  * If at least one rdev is not integrity capable, we can not
1708                  * enable data integrity for the md device.
1709                  */
1710                 if (!bdev_get_integrity(rdev->bdev))
1711                         return -EINVAL;
1712                 if (!reference) {
1713                         /* Use the first rdev as the reference */
1714                         reference = rdev;
1715                         continue;
1716                 }
1717                 /* does this rdev's profile match the reference profile? */
1718                 if (blk_integrity_compare(reference->bdev->bd_disk,
1719                                 rdev->bdev->bd_disk) < 0)
1720                         return -EINVAL;
1721         }
1722         /*
1723          * All component devices are integrity capable and have matching
1724          * profiles, register the common profile for the md device.
1725          */
1726         if (blk_integrity_register(mddev->gendisk,
1727                         bdev_get_integrity(reference->bdev)) != 0) {
1728                 printk(KERN_ERR "md: failed to register integrity for %s\n",
1729                         mdname(mddev));
1730                 return -EINVAL;
1731         }
1732         printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1733                 mdname(mddev));
1734         return 0;
1735 }
1736 EXPORT_SYMBOL(md_integrity_register);
1737
1738 /* Disable data integrity if non-capable/non-matching disk is being added */
1739 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1740 {
1741         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1742         struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1743
1744         if (!bi_mddev) /* nothing to do */
1745                 return;
1746         if (rdev->raid_disk < 0) /* skip spares */
1747                 return;
1748         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1749                                              rdev->bdev->bd_disk) >= 0)
1750                 return;
1751         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1752         blk_integrity_unregister(mddev->gendisk);
1753 }
1754 EXPORT_SYMBOL(md_integrity_add_rdev);
1755
1756 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1757 {
1758         char b[BDEVNAME_SIZE];
1759         struct kobject *ko;
1760         char *s;
1761         int err;
1762
1763         if (rdev->mddev) {
1764                 MD_BUG();
1765                 return -EINVAL;
1766         }
1767
1768         /* prevent duplicates */
1769         if (find_rdev(mddev, rdev->bdev->bd_dev))
1770                 return -EEXIST;
1771
1772         /* make sure rdev->sectors exceeds mddev->dev_sectors */
1773         if (rdev->sectors && (mddev->dev_sectors == 0 ||
1774                         rdev->sectors < mddev->dev_sectors)) {
1775                 if (mddev->pers) {
1776                         /* Cannot change size, so fail
1777                          * If mddev->level <= 0, then we don't care
1778                          * about aligning sizes (e.g. linear)
1779                          */
1780                         if (mddev->level > 0)
1781                                 return -ENOSPC;
1782                 } else
1783                         mddev->dev_sectors = rdev->sectors;
1784         }
1785
1786         /* Verify rdev->desc_nr is unique.
1787          * If it is -1, assign a free number, else
1788          * check number is not in use
1789          */
1790         if (rdev->desc_nr < 0) {
1791                 int choice = 0;
1792                 if (mddev->pers) choice = mddev->raid_disks;
1793                 while (find_rdev_nr(mddev, choice))
1794                         choice++;
1795                 rdev->desc_nr = choice;
1796         } else {
1797                 if (find_rdev_nr(mddev, rdev->desc_nr))
1798                         return -EBUSY;
1799         }
1800         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1801                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1802                        mdname(mddev), mddev->max_disks);
1803                 return -EBUSY;
1804         }
1805         bdevname(rdev->bdev,b);
1806         while ( (s=strchr(b, '/')) != NULL)
1807                 *s = '!';
1808
1809         rdev->mddev = mddev;
1810         printk(KERN_INFO "md: bind<%s>\n", b);
1811
1812         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1813                 goto fail;
1814
1815         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1816         if (sysfs_create_link(&rdev->kobj, ko, "block"))
1817                 /* failure here is OK */;
1818         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1819
1820         list_add_rcu(&rdev->same_set, &mddev->disks);
1821         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1822
1823         /* May as well allow recovery to be retried once */
1824         mddev->recovery_disabled = 0;
1825
1826         return 0;
1827
1828  fail:
1829         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1830                b, mdname(mddev));
1831         return err;
1832 }
1833
1834 static void md_delayed_delete(struct work_struct *ws)
1835 {
1836         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1837         kobject_del(&rdev->kobj);
1838         kobject_put(&rdev->kobj);
1839 }
1840
1841 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1842 {
1843         char b[BDEVNAME_SIZE];
1844         if (!rdev->mddev) {
1845                 MD_BUG();
1846                 return;
1847         }
1848         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1849         list_del_rcu(&rdev->same_set);
1850         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1851         rdev->mddev = NULL;
1852         sysfs_remove_link(&rdev->kobj, "block");
1853         sysfs_put(rdev->sysfs_state);
1854         rdev->sysfs_state = NULL;
1855         /* We need to delay this, otherwise we can deadlock when
1856          * writing to 'remove' to "dev/state".  We also need
1857          * to delay it due to rcu usage.
1858          */
1859         synchronize_rcu();
1860         INIT_WORK(&rdev->del_work, md_delayed_delete);
1861         kobject_get(&rdev->kobj);
1862         schedule_work(&rdev->del_work);
1863 }
1864
1865 /*
1866  * prevent the device from being mounted, repartitioned or
1867  * otherwise reused by a RAID array (or any other kernel
1868  * subsystem), by bd_claiming the device.
1869  */
1870 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1871 {
1872         int err = 0;
1873         struct block_device *bdev;
1874         char b[BDEVNAME_SIZE];
1875
1876         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1877         if (IS_ERR(bdev)) {
1878                 printk(KERN_ERR "md: could not open %s.\n",
1879                         __bdevname(dev, b));
1880                 return PTR_ERR(bdev);
1881         }
1882         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1883         if (err) {
1884                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1885                         bdevname(bdev, b));
1886                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1887                 return err;
1888         }
1889         if (!shared)
1890                 set_bit(AllReserved, &rdev->flags);
1891         rdev->bdev = bdev;
1892         return err;
1893 }
1894
1895 static void unlock_rdev(mdk_rdev_t *rdev)
1896 {
1897         struct block_device *bdev = rdev->bdev;
1898         rdev->bdev = NULL;
1899         if (!bdev)
1900                 MD_BUG();
1901         bd_release(bdev);
1902         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1903 }
1904
1905 void md_autodetect_dev(dev_t dev);
1906
1907 static void export_rdev(mdk_rdev_t * rdev)
1908 {
1909         char b[BDEVNAME_SIZE];
1910         printk(KERN_INFO "md: export_rdev(%s)\n",
1911                 bdevname(rdev->bdev,b));
1912         if (rdev->mddev)
1913                 MD_BUG();
1914         free_disk_sb(rdev);
1915 #ifndef MODULE
1916         if (test_bit(AutoDetected, &rdev->flags))
1917                 md_autodetect_dev(rdev->bdev->bd_dev);
1918 #endif
1919         unlock_rdev(rdev);
1920         kobject_put(&rdev->kobj);
1921 }
1922
1923 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1924 {
1925         unbind_rdev_from_array(rdev);
1926         export_rdev(rdev);
1927 }
1928
1929 static void export_array(mddev_t *mddev)
1930 {
1931         mdk_rdev_t *rdev, *tmp;
1932
1933         rdev_for_each(rdev, tmp, mddev) {
1934                 if (!rdev->mddev) {
1935                         MD_BUG();
1936                         continue;
1937                 }
1938                 kick_rdev_from_array(rdev);
1939         }
1940         if (!list_empty(&mddev->disks))
1941                 MD_BUG();
1942         mddev->raid_disks = 0;
1943         mddev->major_version = 0;
1944 }
1945
1946 static void print_desc(mdp_disk_t *desc)
1947 {
1948         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1949                 desc->major,desc->minor,desc->raid_disk,desc->state);
1950 }
1951
1952 static void print_sb_90(mdp_super_t *sb)
1953 {
1954         int i;
1955
1956         printk(KERN_INFO
1957                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1958                 sb->major_version, sb->minor_version, sb->patch_version,
1959                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1960                 sb->ctime);
1961         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1962                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1963                 sb->md_minor, sb->layout, sb->chunk_size);
1964         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1965                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1966                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1967                 sb->failed_disks, sb->spare_disks,
1968                 sb->sb_csum, (unsigned long)sb->events_lo);
1969
1970         printk(KERN_INFO);
1971         for (i = 0; i < MD_SB_DISKS; i++) {
1972                 mdp_disk_t *desc;
1973
1974                 desc = sb->disks + i;
1975                 if (desc->number || desc->major || desc->minor ||
1976                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1977                         printk("     D %2d: ", i);
1978                         print_desc(desc);
1979                 }
1980         }
1981         printk(KERN_INFO "md:     THIS: ");
1982         print_desc(&sb->this_disk);
1983 }
1984
1985 static void print_sb_1(struct mdp_superblock_1 *sb)
1986 {
1987         __u8 *uuid;
1988
1989         uuid = sb->set_uuid;
1990         printk(KERN_INFO
1991                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1992                "md:    Name: \"%s\" CT:%llu\n",
1993                 le32_to_cpu(sb->major_version),
1994                 le32_to_cpu(sb->feature_map),
1995                 uuid,
1996                 sb->set_name,
1997                 (unsigned long long)le64_to_cpu(sb->ctime)
1998                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1999
2000         uuid = sb->device_uuid;
2001         printk(KERN_INFO
2002                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2003                         " RO:%llu\n"
2004                "md:     Dev:%08x UUID: %pU\n"
2005                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2006                "md:         (MaxDev:%u) \n",
2007                 le32_to_cpu(sb->level),
2008                 (unsigned long long)le64_to_cpu(sb->size),
2009                 le32_to_cpu(sb->raid_disks),
2010                 le32_to_cpu(sb->layout),
2011                 le32_to_cpu(sb->chunksize),
2012                 (unsigned long long)le64_to_cpu(sb->data_offset),
2013                 (unsigned long long)le64_to_cpu(sb->data_size),
2014                 (unsigned long long)le64_to_cpu(sb->super_offset),
2015                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2016                 le32_to_cpu(sb->dev_number),
2017                 uuid,
2018                 sb->devflags,
2019                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2020                 (unsigned long long)le64_to_cpu(sb->events),
2021                 (unsigned long long)le64_to_cpu(sb->resync_offset),
2022                 le32_to_cpu(sb->sb_csum),
2023                 le32_to_cpu(sb->max_dev)
2024                 );
2025 }
2026
2027 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2028 {
2029         char b[BDEVNAME_SIZE];
2030         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2031                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2032                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2033                 rdev->desc_nr);
2034         if (rdev->sb_loaded) {
2035                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2036                 switch (major_version) {
2037                 case 0:
2038                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2039                         break;
2040                 case 1:
2041                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2042                         break;
2043                 }
2044         } else
2045                 printk(KERN_INFO "md: no rdev superblock!\n");
2046 }
2047
2048 static void md_print_devices(void)
2049 {
2050         struct list_head *tmp;
2051         mdk_rdev_t *rdev;
2052         mddev_t *mddev;
2053         char b[BDEVNAME_SIZE];
2054
2055         printk("\n");
2056         printk("md:     **********************************\n");
2057         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
2058         printk("md:     **********************************\n");
2059         for_each_mddev(mddev, tmp) {
2060
2061                 if (mddev->bitmap)
2062                         bitmap_print_sb(mddev->bitmap);
2063                 else
2064                         printk("%s: ", mdname(mddev));
2065                 list_for_each_entry(rdev, &mddev->disks, same_set)
2066                         printk("<%s>", bdevname(rdev->bdev,b));
2067                 printk("\n");
2068
2069                 list_for_each_entry(rdev, &mddev->disks, same_set)
2070                         print_rdev(rdev, mddev->major_version);
2071         }
2072         printk("md:     **********************************\n");
2073         printk("\n");
2074 }
2075
2076
2077 static void sync_sbs(mddev_t * mddev, int nospares)
2078 {
2079         /* Update each superblock (in-memory image), but
2080          * if we are allowed to, skip spares which already
2081          * have the right event counter, or have one earlier
2082          * (which would mean they aren't being marked as dirty
2083          * with the rest of the array)
2084          */
2085         mdk_rdev_t *rdev;
2086
2087         /* First make sure individual recovery_offsets are correct */
2088         list_for_each_entry(rdev, &mddev->disks, same_set) {
2089                 if (rdev->raid_disk >= 0 &&
2090                     mddev->delta_disks >= 0 &&
2091                     !test_bit(In_sync, &rdev->flags) &&
2092                     mddev->curr_resync_completed > rdev->recovery_offset)
2093                                 rdev->recovery_offset = mddev->curr_resync_completed;
2094
2095         }
2096         list_for_each_entry(rdev, &mddev->disks, same_set) {
2097                 if (rdev->sb_events == mddev->events ||
2098                     (nospares &&
2099                      rdev->raid_disk < 0 &&
2100                      rdev->sb_events+1 == mddev->events)) {
2101                         /* Don't update this superblock */
2102                         rdev->sb_loaded = 2;
2103                 } else {
2104                         super_types[mddev->major_version].
2105                                 sync_super(mddev, rdev);
2106                         rdev->sb_loaded = 1;
2107                 }
2108         }
2109 }
2110
2111 static void md_update_sb(mddev_t * mddev, int force_change)
2112 {
2113         mdk_rdev_t *rdev;
2114         int sync_req;
2115         int nospares = 0;
2116
2117         mddev->utime = get_seconds();
2118         if (mddev->external)
2119                 return;
2120 repeat:
2121         spin_lock_irq(&mddev->write_lock);
2122
2123         set_bit(MD_CHANGE_PENDING, &mddev->flags);
2124         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2125                 force_change = 1;
2126         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2127                 /* just a clean<-> dirty transition, possibly leave spares alone,
2128                  * though if events isn't the right even/odd, we will have to do
2129                  * spares after all
2130                  */
2131                 nospares = 1;
2132         if (force_change)
2133                 nospares = 0;
2134         if (mddev->degraded)
2135                 /* If the array is degraded, then skipping spares is both
2136                  * dangerous and fairly pointless.
2137                  * Dangerous because a device that was removed from the array
2138                  * might have a event_count that still looks up-to-date,
2139                  * so it can be re-added without a resync.
2140                  * Pointless because if there are any spares to skip,
2141                  * then a recovery will happen and soon that array won't
2142                  * be degraded any more and the spare can go back to sleep then.
2143                  */
2144                 nospares = 0;
2145
2146         sync_req = mddev->in_sync;
2147
2148         /* If this is just a dirty<->clean transition, and the array is clean
2149          * and 'events' is odd, we can roll back to the previous clean state */
2150         if (nospares
2151             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2152             && mddev->can_decrease_events
2153             && mddev->events != 1) {
2154                 mddev->events--;
2155                 mddev->can_decrease_events = 0;
2156         } else {
2157                 /* otherwise we have to go forward and ... */
2158                 mddev->events ++;
2159                 mddev->can_decrease_events = nospares;
2160         }
2161
2162         if (!mddev->events) {
2163                 /*
2164                  * oops, this 64-bit counter should never wrap.
2165                  * Either we are in around ~1 trillion A.C., assuming
2166                  * 1 reboot per second, or we have a bug:
2167                  */
2168                 MD_BUG();
2169                 mddev->events --;
2170         }
2171
2172         /*
2173          * do not write anything to disk if using
2174          * nonpersistent superblocks
2175          */
2176         if (!mddev->persistent) {
2177                 if (!mddev->external)
2178                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2179
2180                 spin_unlock_irq(&mddev->write_lock);
2181                 wake_up(&mddev->sb_wait);
2182                 return;
2183         }
2184         sync_sbs(mddev, nospares);
2185         spin_unlock_irq(&mddev->write_lock);
2186
2187         dprintk(KERN_INFO
2188                 "md: updating %s RAID superblock on device (in sync %d)\n",
2189                 mdname(mddev),mddev->in_sync);
2190
2191         bitmap_update_sb(mddev->bitmap);
2192         list_for_each_entry(rdev, &mddev->disks, same_set) {
2193                 char b[BDEVNAME_SIZE];
2194                 dprintk(KERN_INFO "md: ");
2195                 if (rdev->sb_loaded != 1)
2196                         continue; /* no noise on spare devices */
2197                 if (test_bit(Faulty, &rdev->flags))
2198                         dprintk("(skipping faulty ");
2199
2200                 dprintk("%s ", bdevname(rdev->bdev,b));
2201                 if (!test_bit(Faulty, &rdev->flags)) {
2202                         md_super_write(mddev,rdev,
2203                                        rdev->sb_start, rdev->sb_size,
2204                                        rdev->sb_page);
2205                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2206                                 bdevname(rdev->bdev,b),
2207                                 (unsigned long long)rdev->sb_start);
2208                         rdev->sb_events = mddev->events;
2209
2210                 } else
2211                         dprintk(")\n");
2212                 if (mddev->level == LEVEL_MULTIPATH)
2213                         /* only need to write one superblock... */
2214                         break;
2215         }
2216         md_super_wait(mddev);
2217         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2218
2219         spin_lock_irq(&mddev->write_lock);
2220         if (mddev->in_sync != sync_req ||
2221             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2222                 /* have to write it out again */
2223                 spin_unlock_irq(&mddev->write_lock);
2224                 goto repeat;
2225         }
2226         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2227         spin_unlock_irq(&mddev->write_lock);
2228         wake_up(&mddev->sb_wait);
2229         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2230                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2231
2232 }
2233
2234 /* words written to sysfs files may, or may not, be \n terminated.
2235  * We want to accept with case. For this we use cmd_match.
2236  */
2237 static int cmd_match(const char *cmd, const char *str)
2238 {
2239         /* See if cmd, written into a sysfs file, matches
2240          * str.  They must either be the same, or cmd can
2241          * have a trailing newline
2242          */
2243         while (*cmd && *str && *cmd == *str) {
2244                 cmd++;
2245                 str++;
2246         }
2247         if (*cmd == '\n')
2248                 cmd++;
2249         if (*str || *cmd)
2250                 return 0;
2251         return 1;
2252 }
2253
2254 struct rdev_sysfs_entry {
2255         struct attribute attr;
2256         ssize_t (*show)(mdk_rdev_t *, char *);
2257         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2258 };
2259
2260 static ssize_t
2261 state_show(mdk_rdev_t *rdev, char *page)
2262 {
2263         char *sep = "";
2264         size_t len = 0;
2265
2266         if (test_bit(Faulty, &rdev->flags)) {
2267                 len+= sprintf(page+len, "%sfaulty",sep);
2268                 sep = ",";
2269         }
2270         if (test_bit(In_sync, &rdev->flags)) {
2271                 len += sprintf(page+len, "%sin_sync",sep);
2272                 sep = ",";
2273         }
2274         if (test_bit(WriteMostly, &rdev->flags)) {
2275                 len += sprintf(page+len, "%swrite_mostly",sep);
2276                 sep = ",";
2277         }
2278         if (test_bit(Blocked, &rdev->flags)) {
2279                 len += sprintf(page+len, "%sblocked", sep);
2280                 sep = ",";
2281         }
2282         if (!test_bit(Faulty, &rdev->flags) &&
2283             !test_bit(In_sync, &rdev->flags)) {
2284                 len += sprintf(page+len, "%sspare", sep);
2285                 sep = ",";
2286         }
2287         return len+sprintf(page+len, "\n");
2288 }
2289
2290 static ssize_t
2291 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2292 {
2293         /* can write
2294          *  faulty  - simulates and error
2295          *  remove  - disconnects the device
2296          *  writemostly - sets write_mostly
2297          *  -writemostly - clears write_mostly
2298          *  blocked - sets the Blocked flag
2299          *  -blocked - clears the Blocked flag
2300          *  insync - sets Insync providing device isn't active
2301          */
2302         int err = -EINVAL;
2303         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2304                 md_error(rdev->mddev, rdev);
2305                 err = 0;
2306         } else if (cmd_match(buf, "remove")) {
2307                 if (rdev->raid_disk >= 0)
2308                         err = -EBUSY;
2309                 else {
2310                         mddev_t *mddev = rdev->mddev;
2311                         kick_rdev_from_array(rdev);
2312                         if (mddev->pers)
2313                                 md_update_sb(mddev, 1);
2314                         md_new_event(mddev);
2315                         err = 0;
2316                 }
2317         } else if (cmd_match(buf, "writemostly")) {
2318                 set_bit(WriteMostly, &rdev->flags);
2319                 err = 0;
2320         } else if (cmd_match(buf, "-writemostly")) {
2321                 clear_bit(WriteMostly, &rdev->flags);
2322                 err = 0;
2323         } else if (cmd_match(buf, "blocked")) {
2324                 set_bit(Blocked, &rdev->flags);
2325                 err = 0;
2326         } else if (cmd_match(buf, "-blocked")) {
2327                 clear_bit(Blocked, &rdev->flags);
2328                 wake_up(&rdev->blocked_wait);
2329                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2330                 md_wakeup_thread(rdev->mddev->thread);
2331
2332                 err = 0;
2333         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2334                 set_bit(In_sync, &rdev->flags);
2335                 err = 0;
2336         }
2337         if (!err)
2338                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2339         return err ? err : len;
2340 }
2341 static struct rdev_sysfs_entry rdev_state =
2342 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2343
2344 static ssize_t
2345 errors_show(mdk_rdev_t *rdev, char *page)
2346 {
2347         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2348 }
2349
2350 static ssize_t
2351 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2352 {
2353         char *e;
2354         unsigned long n = simple_strtoul(buf, &e, 10);
2355         if (*buf && (*e == 0 || *e == '\n')) {
2356                 atomic_set(&rdev->corrected_errors, n);
2357                 return len;
2358         }
2359         return -EINVAL;
2360 }
2361 static struct rdev_sysfs_entry rdev_errors =
2362 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2363
2364 static ssize_t
2365 slot_show(mdk_rdev_t *rdev, char *page)
2366 {
2367         if (rdev->raid_disk < 0)
2368                 return sprintf(page, "none\n");
2369         else
2370                 return sprintf(page, "%d\n", rdev->raid_disk);
2371 }
2372
2373 static ssize_t
2374 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2375 {
2376         char *e;
2377         int err;
2378         char nm[20];
2379         int slot = simple_strtoul(buf, &e, 10);
2380         if (strncmp(buf, "none", 4)==0)
2381                 slot = -1;
2382         else if (e==buf || (*e && *e!= '\n'))
2383                 return -EINVAL;
2384         if (rdev->mddev->pers && slot == -1) {
2385                 /* Setting 'slot' on an active array requires also
2386                  * updating the 'rd%d' link, and communicating
2387                  * with the personality with ->hot_*_disk.
2388                  * For now we only support removing
2389                  * failed/spare devices.  This normally happens automatically,
2390                  * but not when the metadata is externally managed.
2391                  */
2392                 if (rdev->raid_disk == -1)
2393                         return -EEXIST;
2394                 /* personality does all needed checks */
2395                 if (rdev->mddev->pers->hot_add_disk == NULL)
2396                         return -EINVAL;
2397                 err = rdev->mddev->pers->
2398                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2399                 if (err)
2400                         return err;
2401                 sprintf(nm, "rd%d", rdev->raid_disk);
2402                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2403                 rdev->raid_disk = -1;
2404                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2405                 md_wakeup_thread(rdev->mddev->thread);
2406         } else if (rdev->mddev->pers) {
2407                 mdk_rdev_t *rdev2;
2408                 /* Activating a spare .. or possibly reactivating
2409                  * if we ever get bitmaps working here.
2410                  */
2411
2412                 if (rdev->raid_disk != -1)
2413                         return -EBUSY;
2414
2415                 if (rdev->mddev->pers->hot_add_disk == NULL)
2416                         return -EINVAL;
2417
2418                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2419                         if (rdev2->raid_disk == slot)
2420                                 return -EEXIST;
2421
2422                 rdev->raid_disk = slot;
2423                 if (test_bit(In_sync, &rdev->flags))
2424                         rdev->saved_raid_disk = slot;
2425                 else
2426                         rdev->saved_raid_disk = -1;
2427                 err = rdev->mddev->pers->
2428                         hot_add_disk(rdev->mddev, rdev);
2429                 if (err) {
2430                         rdev->raid_disk = -1;
2431                         return err;
2432                 } else
2433                         sysfs_notify_dirent_safe(rdev->sysfs_state);
2434                 sprintf(nm, "rd%d", rdev->raid_disk);
2435                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2436                         /* failure here is OK */;
2437                 /* don't wakeup anyone, leave that to userspace. */
2438         } else {
2439                 if (slot >= rdev->mddev->raid_disks)
2440                         return -ENOSPC;
2441                 rdev->raid_disk = slot;
2442                 /* assume it is working */
2443                 clear_bit(Faulty, &rdev->flags);
2444                 clear_bit(WriteMostly, &rdev->flags);
2445                 set_bit(In_sync, &rdev->flags);
2446                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2447         }
2448         return len;
2449 }
2450
2451
2452 static struct rdev_sysfs_entry rdev_slot =
2453 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2454
2455 static ssize_t
2456 offset_show(mdk_rdev_t *rdev, char *page)
2457 {
2458         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2459 }
2460
2461 static ssize_t
2462 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2463 {
2464         char *e;
2465         unsigned long long offset = simple_strtoull(buf, &e, 10);
2466         if (e==buf || (*e && *e != '\n'))
2467                 return -EINVAL;
2468         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2469                 return -EBUSY;
2470         if (rdev->sectors && rdev->mddev->external)
2471                 /* Must set offset before size, so overlap checks
2472                  * can be sane */
2473                 return -EBUSY;
2474         rdev->data_offset = offset;
2475         return len;
2476 }
2477
2478 static struct rdev_sysfs_entry rdev_offset =
2479 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2480
2481 static ssize_t
2482 rdev_size_show(mdk_rdev_t *rdev, char *page)
2483 {
2484         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2485 }
2486
2487 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2488 {
2489         /* check if two start/length pairs overlap */
2490         if (s1+l1 <= s2)
2491                 return 0;
2492         if (s2+l2 <= s1)
2493                 return 0;
2494         return 1;
2495 }
2496
2497 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2498 {
2499         unsigned long long blocks;
2500         sector_t new;
2501
2502         if (strict_strtoull(buf, 10, &blocks) < 0)
2503                 return -EINVAL;
2504
2505         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2506                 return -EINVAL; /* sector conversion overflow */
2507
2508         new = blocks * 2;
2509         if (new != blocks * 2)
2510                 return -EINVAL; /* unsigned long long to sector_t overflow */
2511
2512         *sectors = new;
2513         return 0;
2514 }
2515
2516 static ssize_t
2517 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2518 {
2519         mddev_t *my_mddev = rdev->mddev;
2520         sector_t oldsectors = rdev->sectors;
2521         sector_t sectors;
2522
2523         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2524                 return -EINVAL;
2525         if (my_mddev->pers && rdev->raid_disk >= 0) {
2526                 if (my_mddev->persistent) {
2527                         sectors = super_types[my_mddev->major_version].
2528                                 rdev_size_change(rdev, sectors);
2529                         if (!sectors)
2530                                 return -EBUSY;
2531                 } else if (!sectors)
2532                         sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2533                                 rdev->data_offset;
2534         }
2535         if (sectors < my_mddev->dev_sectors)
2536                 return -EINVAL; /* component must fit device */
2537
2538         rdev->sectors = sectors;
2539         if (sectors > oldsectors && my_mddev->external) {
2540                 /* need to check that all other rdevs with the same ->bdev
2541                  * do not overlap.  We need to unlock the mddev to avoid
2542                  * a deadlock.  We have already changed rdev->sectors, and if
2543                  * we have to change it back, we will have the lock again.
2544                  */
2545                 mddev_t *mddev;
2546                 int overlap = 0;
2547                 struct list_head *tmp;
2548
2549                 mddev_unlock(my_mddev);
2550                 for_each_mddev(mddev, tmp) {
2551                         mdk_rdev_t *rdev2;
2552
2553                         mddev_lock(mddev);
2554                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2555                                 if (test_bit(AllReserved, &rdev2->flags) ||
2556                                     (rdev->bdev == rdev2->bdev &&
2557                                      rdev != rdev2 &&
2558                                      overlaps(rdev->data_offset, rdev->sectors,
2559                                               rdev2->data_offset,
2560                                               rdev2->sectors))) {
2561                                         overlap = 1;
2562                                         break;
2563                                 }
2564                         mddev_unlock(mddev);
2565                         if (overlap) {
2566                                 mddev_put(mddev);
2567                                 break;
2568                         }
2569                 }
2570                 mddev_lock(my_mddev);
2571                 if (overlap) {
2572                         /* Someone else could have slipped in a size
2573                          * change here, but doing so is just silly.
2574                          * We put oldsectors back because we *know* it is
2575                          * safe, and trust userspace not to race with
2576                          * itself
2577                          */
2578                         rdev->sectors = oldsectors;
2579                         return -EBUSY;
2580                 }
2581         }
2582         return len;
2583 }
2584
2585 static struct rdev_sysfs_entry rdev_size =
2586 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2587
2588
2589 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2590 {
2591         unsigned long long recovery_start = rdev->recovery_offset;
2592
2593         if (test_bit(In_sync, &rdev->flags) ||
2594             recovery_start == MaxSector)
2595                 return sprintf(page, "none\n");
2596
2597         return sprintf(page, "%llu\n", recovery_start);
2598 }
2599
2600 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2601 {
2602         unsigned long long recovery_start;
2603
2604         if (cmd_match(buf, "none"))
2605                 recovery_start = MaxSector;
2606         else if (strict_strtoull(buf, 10, &recovery_start))
2607                 return -EINVAL;
2608
2609         if (rdev->mddev->pers &&
2610             rdev->raid_disk >= 0)
2611                 return -EBUSY;
2612
2613         rdev->recovery_offset = recovery_start;
2614         if (recovery_start == MaxSector)
2615                 set_bit(In_sync, &rdev->flags);
2616         else
2617                 clear_bit(In_sync, &rdev->flags);
2618         return len;
2619 }
2620
2621 static struct rdev_sysfs_entry rdev_recovery_start =
2622 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2623
2624 static struct attribute *rdev_default_attrs[] = {
2625         &rdev_state.attr,
2626         &rdev_errors.attr,
2627         &rdev_slot.attr,
2628         &rdev_offset.attr,
2629         &rdev_size.attr,
2630         &rdev_recovery_start.attr,
2631         NULL,
2632 };
2633 static ssize_t
2634 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2635 {
2636         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2637         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2638         mddev_t *mddev = rdev->mddev;
2639         ssize_t rv;
2640
2641         if (!entry->show)
2642                 return -EIO;
2643
2644         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2645         if (!rv) {
2646                 if (rdev->mddev == NULL)
2647                         rv = -EBUSY;
2648                 else
2649                         rv = entry->show(rdev, page);
2650                 mddev_unlock(mddev);
2651         }
2652         return rv;
2653 }
2654
2655 static ssize_t
2656 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2657               const char *page, size_t length)
2658 {
2659         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2660         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2661         ssize_t rv;
2662         mddev_t *mddev = rdev->mddev;
2663
2664         if (!entry->store)
2665                 return -EIO;
2666         if (!capable(CAP_SYS_ADMIN))
2667                 return -EACCES;
2668         rv = mddev ? mddev_lock(mddev): -EBUSY;
2669         if (!rv) {
2670                 if (rdev->mddev == NULL)
2671                         rv = -EBUSY;
2672                 else
2673                         rv = entry->store(rdev, page, length);
2674                 mddev_unlock(mddev);
2675         }
2676         return rv;
2677 }
2678
2679 static void rdev_free(struct kobject *ko)
2680 {
2681         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2682         kfree(rdev);
2683 }
2684 static const struct sysfs_ops rdev_sysfs_ops = {
2685         .show           = rdev_attr_show,
2686         .store          = rdev_attr_store,
2687 };
2688 static struct kobj_type rdev_ktype = {
2689         .release        = rdev_free,
2690         .sysfs_ops      = &rdev_sysfs_ops,
2691         .default_attrs  = rdev_default_attrs,
2692 };
2693
2694 /*
2695  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2696  *
2697  * mark the device faulty if:
2698  *
2699  *   - the device is nonexistent (zero size)
2700  *   - the device has no valid superblock
2701  *
2702  * a faulty rdev _never_ has rdev->sb set.
2703  */
2704 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2705 {
2706         char b[BDEVNAME_SIZE];
2707         int err;
2708         mdk_rdev_t *rdev;
2709         sector_t size;
2710
2711         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2712         if (!rdev) {
2713                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2714                 return ERR_PTR(-ENOMEM);
2715         }
2716
2717         if ((err = alloc_disk_sb(rdev)))
2718                 goto abort_free;
2719
2720         err = lock_rdev(rdev, newdev, super_format == -2);
2721         if (err)
2722                 goto abort_free;
2723
2724         kobject_init(&rdev->kobj, &rdev_ktype);
2725
2726         rdev->desc_nr = -1;
2727         rdev->saved_raid_disk = -1;
2728         rdev->raid_disk = -1;
2729         rdev->flags = 0;
2730         rdev->data_offset = 0;
2731         rdev->sb_events = 0;
2732         rdev->last_read_error.tv_sec  = 0;
2733         rdev->last_read_error.tv_nsec = 0;
2734         atomic_set(&rdev->nr_pending, 0);
2735         atomic_set(&rdev->read_errors, 0);
2736         atomic_set(&rdev->corrected_errors, 0);
2737
2738         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2739         if (!size) {
2740                 printk(KERN_WARNING
2741                         "md: %s has zero or unknown size, marking faulty!\n",
2742                         bdevname(rdev->bdev,b));
2743                 err = -EINVAL;
2744                 goto abort_free;
2745         }
2746
2747         if (super_format >= 0) {
2748                 err = super_types[super_format].
2749                         load_super(rdev, NULL, super_minor);
2750                 if (err == -EINVAL) {
2751                         printk(KERN_WARNING
2752                                 "md: %s does not have a valid v%d.%d "
2753                                "superblock, not importing!\n",
2754                                 bdevname(rdev->bdev,b),
2755                                super_format, super_minor);
2756                         goto abort_free;
2757                 }
2758                 if (err < 0) {
2759                         printk(KERN_WARNING
2760                                 "md: could not read %s's sb, not importing!\n",
2761                                 bdevname(rdev->bdev,b));
2762                         goto abort_free;
2763                 }
2764         }
2765
2766         INIT_LIST_HEAD(&rdev->same_set);
2767         init_waitqueue_head(&rdev->blocked_wait);
2768
2769         return rdev;
2770
2771 abort_free:
2772         if (rdev->sb_page) {
2773                 if (rdev->bdev)
2774                         unlock_rdev(rdev);
2775                 free_disk_sb(rdev);
2776         }
2777         kfree(rdev);
2778         return ERR_PTR(err);
2779 }
2780
2781 /*
2782  * Check a full RAID array for plausibility
2783  */
2784
2785
2786 static void analyze_sbs(mddev_t * mddev)
2787 {
2788         int i;
2789         mdk_rdev_t *rdev, *freshest, *tmp;
2790         char b[BDEVNAME_SIZE];
2791
2792         freshest = NULL;
2793         rdev_for_each(rdev, tmp, mddev)
2794                 switch (super_types[mddev->major_version].
2795                         load_super(rdev, freshest, mddev->minor_version)) {
2796                 case 1:
2797                         freshest = rdev;
2798                         break;
2799                 case 0:
2800                         break;
2801                 default:
2802                         printk( KERN_ERR \
2803                                 "md: fatal superblock inconsistency in %s"
2804                                 " -- removing from array\n",
2805                                 bdevname(rdev->bdev,b));
2806                         kick_rdev_from_array(rdev);
2807                 }
2808
2809
2810         super_types[mddev->major_version].
2811                 validate_super(mddev, freshest);
2812
2813         i = 0;
2814         rdev_for_each(rdev, tmp, mddev) {
2815                 if (mddev->max_disks &&
2816                     (rdev->desc_nr >= mddev->max_disks ||
2817                      i > mddev->max_disks)) {
2818                         printk(KERN_WARNING
2819                                "md: %s: %s: only %d devices permitted\n",
2820                                mdname(mddev), bdevname(rdev->bdev, b),
2821                                mddev->max_disks);
2822                         kick_rdev_from_array(rdev);
2823                         continue;
2824                 }
2825                 if (rdev != freshest)
2826                         if (super_types[mddev->major_version].
2827                             validate_super(mddev, rdev)) {
2828                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2829                                         " from array!\n",
2830                                         bdevname(rdev->bdev,b));
2831                                 kick_rdev_from_array(rdev);
2832                                 continue;
2833                         }
2834                 if (mddev->level == LEVEL_MULTIPATH) {
2835                         rdev->desc_nr = i++;
2836                         rdev->raid_disk = rdev->desc_nr;
2837                         set_bit(In_sync, &rdev->flags);
2838                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2839                         rdev->raid_disk = -1;
2840                         clear_bit(In_sync, &rdev->flags);
2841                 }
2842         }
2843 }
2844
2845 /* Read a fixed-point number.
2846  * Numbers in sysfs attributes should be in "standard" units where
2847  * possible, so time should be in seconds.
2848  * However we internally use a a much smaller unit such as
2849  * milliseconds or jiffies.
2850  * This function takes a decimal number with a possible fractional
2851  * component, and produces an integer which is the result of
2852  * multiplying that number by 10^'scale'.
2853  * all without any floating-point arithmetic.
2854  */
2855 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2856 {
2857         unsigned long result = 0;
2858         long decimals = -1;
2859         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2860                 if (*cp == '.')
2861                         decimals = 0;
2862                 else if (decimals < scale) {
2863                         unsigned int value;
2864                         value = *cp - '0';
2865                         result = result * 10 + value;
2866                         if (decimals >= 0)
2867                                 decimals++;
2868                 }
2869                 cp++;
2870         }
2871         if (*cp == '\n')
2872                 cp++;
2873         if (*cp)
2874                 return -EINVAL;
2875         if (decimals < 0)
2876                 decimals = 0;
2877         while (decimals < scale) {
2878                 result *= 10;
2879                 decimals ++;
2880         }
2881         *res = result;
2882         return 0;
2883 }
2884
2885
2886 static void md_safemode_timeout(unsigned long data);
2887
2888 static ssize_t
2889 safe_delay_show(mddev_t *mddev, char *page)
2890 {
2891         int msec = (mddev->safemode_delay*1000)/HZ;
2892         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2893 }
2894 static ssize_t
2895 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2896 {
2897         unsigned long msec;
2898
2899         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2900                 return -EINVAL;
2901         if (msec == 0)
2902                 mddev->safemode_delay = 0;
2903         else {
2904                 unsigned long old_delay = mddev->safemode_delay;
2905                 mddev->safemode_delay = (msec*HZ)/1000;
2906                 if (mddev->safemode_delay == 0)
2907                         mddev->safemode_delay = 1;
2908                 if (mddev->safemode_delay < old_delay)
2909                         md_safemode_timeout((unsigned long)mddev);
2910         }
2911         return len;
2912 }
2913 static struct md_sysfs_entry md_safe_delay =
2914 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2915
2916 static ssize_t
2917 level_show(mddev_t *mddev, char *page)
2918 {
2919         struct mdk_personality *p = mddev->pers;
2920         if (p)
2921                 return sprintf(page, "%s\n", p->name);
2922         else if (mddev->clevel[0])
2923                 return sprintf(page, "%s\n", mddev->clevel);
2924         else if (mddev->level != LEVEL_NONE)
2925                 return sprintf(page, "%d\n", mddev->level);
2926         else
2927                 return 0;
2928 }
2929
2930 static ssize_t
2931 level_store(mddev_t *mddev, const char *buf, size_t len)
2932 {
2933         char clevel[16];
2934         ssize_t rv = len;
2935         struct mdk_personality *pers;
2936         long level;
2937         void *priv;
2938         mdk_rdev_t *rdev;
2939
2940         if (mddev->pers == NULL) {
2941                 if (len == 0)
2942                         return 0;
2943                 if (len >= sizeof(mddev->clevel))
2944                         return -ENOSPC;
2945                 strncpy(mddev->clevel, buf, len);
2946                 if (mddev->clevel[len-1] == '\n')
2947                         len--;
2948                 mddev->clevel[len] = 0;
2949                 mddev->level = LEVEL_NONE;
2950                 return rv;
2951         }
2952
2953         /* request to change the personality.  Need to ensure:
2954          *  - array is not engaged in resync/recovery/reshape
2955          *  - old personality can be suspended
2956          *  - new personality will access other array.
2957          */
2958
2959         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2960                 return -EBUSY;
2961
2962         if (!mddev->pers->quiesce) {
2963                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2964                        mdname(mddev), mddev->pers->name);
2965                 return -EINVAL;
2966         }
2967
2968         /* Now find the new personality */
2969         if (len == 0 || len >= sizeof(clevel))
2970                 return -EINVAL;
2971         strncpy(clevel, buf, len);
2972         if (clevel[len-1] == '\n')
2973                 len--;
2974         clevel[len] = 0;
2975         if (strict_strtol(clevel, 10, &level))
2976                 level = LEVEL_NONE;
2977
2978         if (request_module("md-%s", clevel) != 0)
2979                 request_module("md-level-%s", clevel);
2980         spin_lock(&pers_lock);
2981         pers = find_pers(level, clevel);
2982         if (!pers || !try_module_get(pers->owner)) {
2983                 spin_unlock(&pers_lock);
2984                 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
2985                 return -EINVAL;
2986         }
2987         spin_unlock(&pers_lock);
2988
2989         if (pers == mddev->pers) {
2990                 /* Nothing to do! */
2991                 module_put(pers->owner);
2992                 return rv;
2993         }
2994         if (!pers->takeover) {
2995                 module_put(pers->owner);
2996                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2997                        mdname(mddev), clevel);
2998                 return -EINVAL;
2999         }
3000
3001         list_for_each_entry(rdev, &mddev->disks, same_set)
3002                 rdev->new_raid_disk = rdev->raid_disk;
3003
3004         /* ->takeover must set new_* and/or delta_disks
3005          * if it succeeds, and may set them when it fails.
3006          */
3007         priv = pers->takeover(mddev);
3008         if (IS_ERR(priv)) {
3009                 mddev->new_level = mddev->level;
3010                 mddev->new_layout = mddev->layout;
3011                 mddev->new_chunk_sectors = mddev->chunk_sectors;
3012                 mddev->raid_disks -= mddev->delta_disks;
3013                 mddev->delta_disks = 0;
3014                 module_put(pers->owner);
3015                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3016                        mdname(mddev), clevel);
3017                 return PTR_ERR(priv);
3018         }
3019
3020         /* Looks like we have a winner */
3021         mddev_suspend(mddev);
3022         mddev->pers->stop(mddev);
3023
3024         if (mddev->pers->sync_request == NULL &&
3025             pers->sync_request != NULL) {
3026                 /* need to add the md_redundancy_group */
3027                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3028                         printk(KERN_WARNING
3029                                "md: cannot register extra attributes for %s\n",
3030                                mdname(mddev));
3031                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3032         }
3033         if (mddev->pers->sync_request != NULL &&
3034             pers->sync_request == NULL) {
3035                 /* need to remove the md_redundancy_group */
3036                 if (mddev->to_remove == NULL)
3037                         mddev->to_remove = &md_redundancy_group;
3038         }
3039
3040         if (mddev->pers->sync_request == NULL &&
3041             mddev->external) {
3042                 /* We are converting from a no-redundancy array
3043                  * to a redundancy array and metadata is managed
3044                  * externally so we need to be sure that writes
3045                  * won't block due to a need to transition
3046                  *      clean->dirty
3047                  * until external management is started.
3048                  */
3049                 mddev->in_sync = 0;
3050                 mddev->safemode_delay = 0;
3051                 mddev->safemode = 0;
3052         }
3053
3054         list_for_each_entry(rdev, &mddev->disks, same_set) {
3055                 char nm[20];
3056                 if (rdev->raid_disk < 0)
3057                         continue;
3058                 if (rdev->new_raid_disk > mddev->raid_disks)
3059                         rdev->new_raid_disk = -1;
3060                 if (rdev->new_raid_disk == rdev->raid_disk)
3061                         continue;
3062                 sprintf(nm, "rd%d", rdev->raid_disk);
3063                 sysfs_remove_link(&mddev->kobj, nm);
3064         }
3065         list_for_each_entry(rdev, &mddev->disks, same_set) {
3066                 if (rdev->raid_disk < 0)
3067                         continue;
3068                 if (rdev->new_raid_disk == rdev->raid_disk)
3069                         continue;
3070                 rdev->raid_disk = rdev->new_raid_disk;
3071                 if (rdev->raid_disk < 0)
3072                         clear_bit(In_sync, &rdev->flags);
3073                 else {
3074                         char nm[20];
3075                         sprintf(nm, "rd%d", rdev->raid_disk);
3076                         if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3077                                 printk("md: cannot register %s for %s after level change\n",
3078                                        nm, mdname(mddev));
3079                 }
3080         }
3081
3082         module_put(mddev->pers->owner);
3083         mddev->pers = pers;
3084         mddev->private = priv;
3085         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3086         mddev->level = mddev->new_level;
3087         mddev->layout = mddev->new_layout;
3088         mddev->chunk_sectors = mddev->new_chunk_sectors;
3089         mddev->delta_disks = 0;
3090         if (mddev->pers->sync_request == NULL) {
3091                 /* this is now an array without redundancy, so
3092                  * it must always be in_sync
3093                  */
3094                 mddev->in_sync = 1;
3095                 del_timer_sync(&mddev->safemode_timer);
3096         }
3097         pers->run(mddev);
3098         mddev_resume(mddev);
3099         set_bit(MD_CHANGE_DEVS, &mddev->flags);
3100         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3101         md_wakeup_thread(mddev->thread);
3102         sysfs_notify(&mddev->kobj, NULL, "level");
3103         md_new_event(mddev);
3104         return rv;
3105 }
3106
3107 static struct md_sysfs_entry md_level =
3108 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3109
3110
3111 static ssize_t
3112 layout_show(mddev_t *mddev, char *page)
3113 {
3114         /* just a number, not meaningful for all levels */
3115         if (mddev->reshape_position != MaxSector &&
3116             mddev->layout != mddev->new_layout)
3117                 return sprintf(page, "%d (%d)\n",
3118                                mddev->new_layout, mddev->layout);
3119         return sprintf(page, "%d\n", mddev->layout);
3120 }
3121
3122 static ssize_t
3123 layout_store(mddev_t *mddev, const char *buf, size_t len)
3124 {
3125         char *e;
3126         unsigned long n = simple_strtoul(buf, &e, 10);
3127
3128         if (!*buf || (*e && *e != '\n'))
3129                 return -EINVAL;
3130
3131         if (mddev->pers) {
3132                 int err;
3133                 if (mddev->pers->check_reshape == NULL)
3134                         return -EBUSY;
3135                 mddev->new_layout = n;
3136                 err = mddev->pers->check_reshape(mddev);
3137                 if (err) {
3138                         mddev->new_layout = mddev->layout;
3139                         return err;
3140                 }
3141         } else {
3142                 mddev->new_layout = n;
3143                 if (mddev->reshape_position == MaxSector)
3144                         mddev->layout = n;
3145         }
3146         return len;
3147 }
3148 static struct md_sysfs_entry md_layout =
3149 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3150
3151
3152 static ssize_t
3153 raid_disks_show(mddev_t *mddev, char *page)
3154 {
3155         if (mddev->raid_disks == 0)
3156                 return 0;
3157         if (mddev->reshape_position != MaxSector &&
3158             mddev->delta_disks != 0)
3159                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3160                                mddev->raid_disks - mddev->delta_disks);
3161         return sprintf(page, "%d\n", mddev->raid_disks);
3162 }
3163
3164 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3165
3166 static ssize_t
3167 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3168 {
3169         char *e;
3170         int rv = 0;
3171         unsigned long n = simple_strtoul(buf, &e, 10);
3172
3173         if (!*buf || (*e && *e != '\n'))
3174                 return -EINVAL;
3175
3176         if (mddev->pers)
3177                 rv = update_raid_disks(mddev, n);
3178         else if (mddev->reshape_position != MaxSector) {
3179                 int olddisks = mddev->raid_disks - mddev->delta_disks;
3180                 mddev->delta_disks = n - olddisks;
3181                 mddev->raid_disks = n;
3182         } else
3183                 mddev->raid_disks = n;
3184         return rv ? rv : len;
3185 }
3186 static struct md_sysfs_entry md_raid_disks =
3187 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3188
3189 static ssize_t
3190 chunk_size_show(mddev_t *mddev, char *page)
3191 {
3192         if (mddev->reshape_position != MaxSector &&
3193             mddev->chunk_sectors != mddev->new_chunk_sectors)
3194                 return sprintf(page, "%d (%d)\n",
3195                                mddev->new_chunk_sectors << 9,
3196                                mddev->chunk_sectors << 9);
3197         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3198 }
3199
3200 static ssize_t
3201 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3202 {
3203         char *e;
3204         unsigned long n = simple_strtoul(buf, &e, 10);
3205
3206         if (!*buf || (*e && *e != '\n'))
3207                 return -EINVAL;
3208
3209         if (mddev->pers) {
3210                 int err;
3211                 if (mddev->pers->check_reshape == NULL)
3212                         return -EBUSY;
3213                 mddev->new_chunk_sectors = n >> 9;
3214                 err = mddev->pers->check_reshape(mddev);
3215                 if (err) {
3216                         mddev->new_chunk_sectors = mddev->chunk_sectors;
3217                         return err;
3218                 }
3219         } else {
3220                 mddev->new_chunk_sectors = n >> 9;
3221                 if (mddev->reshape_position == MaxSector)
3222                         mddev->chunk_sectors = n >> 9;
3223         }
3224         return len;
3225 }
3226 static struct md_sysfs_entry md_chunk_size =
3227 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3228
3229 static ssize_t
3230 resync_start_show(mddev_t *mddev, char *page)
3231 {
3232         if (mddev->recovery_cp == MaxSector)
3233                 return sprintf(page, "none\n");
3234         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3235 }
3236
3237 static ssize_t
3238 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3239 {
3240         char *e;
3241         unsigned long long n = simple_strtoull(buf, &e, 10);
3242
3243         if (mddev->pers)
3244                 return -EBUSY;
3245         if (cmd_match(buf, "none"))
3246                 n = MaxSector;
3247         else if (!*buf || (*e && *e != '\n'))
3248                 return -EINVAL;
3249
3250         mddev->recovery_cp = n;
3251         return len;
3252 }
3253 static struct md_sysfs_entry md_resync_start =
3254 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3255
3256 /*
3257  * The array state can be:
3258  *
3259  * clear
3260  *     No devices, no size, no level
3261  *     Equivalent to STOP_ARRAY ioctl
3262  * inactive
3263  *     May have some settings, but array is not active
3264  *        all IO results in error
3265  *     When written, doesn't tear down array, but just stops it
3266  * suspended (not supported yet)
3267  *     All IO requests will block. The array can be reconfigured.
3268  *     Writing this, if accepted, will block until array is quiescent
3269  * readonly
3270  *     no resync can happen.  no superblocks get written.
3271  *     write requests fail
3272  * read-auto
3273  *     like readonly, but behaves like 'clean' on a write request.
3274  *
3275  * clean - no pending writes, but otherwise active.
3276  *     When written to inactive array, starts without resync
3277  *     If a write request arrives then
3278  *       if metadata is known, mark 'dirty' and switch to 'active'.
3279  *       if not known, block and switch to write-pending
3280  *     If written to an active array that has pending writes, then fails.
3281  * active
3282  *     fully active: IO and resync can be happening.
3283  *     When written to inactive array, starts with resync
3284  *
3285  * write-pending
3286  *     clean, but writes are blocked waiting for 'active' to be written.
3287  *
3288  * active-idle
3289  *     like active, but no writes have been seen for a while (100msec).
3290  *
3291  */
3292 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3293                    write_pending, active_idle, bad_word};
3294 static char *array_states[] = {
3295         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3296         "write-pending", "active-idle", NULL };
3297
3298 static int match_word(const char *word, char **list)
3299 {
3300         int n;
3301         for (n=0; list[n]; n++)
3302                 if (cmd_match(word, list[n]))
3303                         break;
3304         return n;
3305 }
3306
3307 static ssize_t
3308 array_state_show(mddev_t *mddev, char *page)
3309 {
3310         enum array_state st = inactive;
3311
3312         if (mddev->pers)
3313                 switch(mddev->ro) {
3314                 case 1:
3315                         st = readonly;
3316                         break;
3317                 case 2:
3318                         st = read_auto;
3319                         break;
3320                 case 0:
3321                         if (mddev->in_sync)
3322                                 st = clean;
3323                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3324                                 st = write_pending;
3325                         else if (mddev->safemode)
3326                                 st = active_idle;
3327                         else
3328                                 st = active;
3329                 }
3330         else {
3331                 if (list_empty(&mddev->disks) &&
3332                     mddev->raid_disks == 0 &&
3333                     mddev->dev_sectors == 0)
3334                         st = clear;
3335                 else
3336                         st = inactive;
3337         }
3338         return sprintf(page, "%s\n", array_states[st]);
3339 }
3340
3341 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3342 static int md_set_readonly(mddev_t * mddev, int is_open);
3343 static int do_md_run(mddev_t * mddev);
3344 static int restart_array(mddev_t *mddev);
3345
3346 static ssize_t
3347 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3348 {
3349         int err = -EINVAL;
3350         enum array_state st = match_word(buf, array_states);
3351         switch(st) {
3352         case bad_word:
3353                 break;
3354         case clear:
3355                 /* stopping an active array */
3356                 if (atomic_read(&mddev->openers) > 0)
3357                         return -EBUSY;
3358                 err = do_md_stop(mddev, 0, 0);
3359                 break;
3360         case inactive:
3361                 /* stopping an active array */
3362                 if (mddev->pers) {
3363                         if (atomic_read(&mddev->openers) > 0)
3364                                 return -EBUSY;
3365                         err = do_md_stop(mddev, 2, 0);
3366                 } else
3367                         err = 0; /* already inactive */
3368                 break;
3369         case suspended:
3370                 break; /* not supported yet */
3371         case readonly:
3372                 if (mddev->pers)
3373                         err = md_set_readonly(mddev, 0);
3374                 else {
3375                         mddev->ro = 1;
3376                         set_disk_ro(mddev->gendisk, 1);
3377                         err = do_md_run(mddev);
3378                 }
3379                 break;
3380         case read_auto:
3381                 if (mddev->pers) {
3382                         if (mddev->ro == 0)
3383                                 err = md_set_readonly(mddev, 0);
3384                         else if (mddev->ro == 1)
3385                                 err = restart_array(mddev);
3386                         if (err == 0) {
3387                                 mddev->ro = 2;
3388                                 set_disk_ro(mddev->gendisk, 0);
3389                         }
3390                 } else {
3391                         mddev->ro = 2;
3392                         err = do_md_run(mddev);
3393                 }
3394                 break;
3395         case clean:
3396                 if (mddev->pers) {
3397                         restart_array(mddev);
3398                         spin_lock_irq(&mddev->write_lock);
3399                         if (atomic_read(&mddev->writes_pending) == 0) {
3400                                 if (mddev->in_sync == 0) {
3401                                         mddev->in_sync = 1;
3402                                         if (mddev->safemode == 1)
3403                                                 mddev->safemode = 0;
3404                                         if (mddev->persistent)
3405                                                 set_bit(MD_CHANGE_CLEAN,
3406                                                         &mddev->flags);
3407                                 }
3408                                 err = 0;
3409                         } else
3410                                 err = -EBUSY;
3411                         spin_unlock_irq(&mddev->write_lock);
3412                 } else
3413                         err = -EINVAL;
3414                 break;
3415         case active:
3416                 if (mddev->pers) {
3417                         restart_array(mddev);
3418                         if (mddev->external)
3419                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3420                         wake_up(&mddev->sb_wait);
3421                         err = 0;
3422                 } else {
3423                         mddev->ro = 0;
3424                         set_disk_ro(mddev->gendisk, 0);
3425                         err = do_md_run(mddev);
3426                 }
3427                 break;
3428         case write_pending:
3429         case active_idle:
3430                 /* these cannot be set */
3431                 break;
3432         }
3433         if (err)
3434                 return err;
3435         else {
3436                 sysfs_notify_dirent_safe(mddev->sysfs_state);
3437                 return len;
3438         }
3439 }
3440 static struct md_sysfs_entry md_array_state =
3441 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3442
3443 static ssize_t
3444 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3445         return sprintf(page, "%d\n",
3446                        atomic_read(&mddev->max_corr_read_errors));
3447 }
3448
3449 static ssize_t
3450 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3451 {
3452         char *e;
3453         unsigned long n = simple_strtoul(buf, &e, 10);
3454
3455         if (*buf && (*e == 0 || *e == '\n')) {
3456                 atomic_set(&mddev->max_corr_read_errors, n);
3457                 return len;
3458         }
3459         return -EINVAL;
3460 }
3461
3462 static struct md_sysfs_entry max_corr_read_errors =
3463 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3464         max_corrected_read_errors_store);
3465
3466 static ssize_t
3467 null_show(mddev_t *mddev, char *page)
3468 {
3469         return -EINVAL;
3470 }
3471
3472 static ssize_t
3473 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3474 {
3475         /* buf must be %d:%d\n? giving major and minor numbers */
3476         /* The new device is added to the array.
3477          * If the array has a persistent superblock, we read the
3478          * superblock to initialise info and check validity.
3479          * Otherwise, only checking done is that in bind_rdev_to_array,
3480          * which mainly checks size.
3481          */
3482         char *e;
3483         int major = simple_strtoul(buf, &e, 10);
3484         int minor;
3485         dev_t dev;
3486         mdk_rdev_t *rdev;
3487         int err;
3488
3489         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3490                 return -EINVAL;
3491         minor = simple_strtoul(e+1, &e, 10);
3492         if (*e && *e != '\n')
3493                 return -EINVAL;
3494         dev = MKDEV(major, minor);
3495         if (major != MAJOR(dev) ||
3496             minor != MINOR(dev))
3497                 return -EOVERFLOW;
3498
3499
3500         if (mddev->persistent) {
3501                 rdev = md_import_device(dev, mddev->major_version,
3502                                         mddev->minor_version);
3503                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3504                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3505                                                        mdk_rdev_t, same_set);
3506                         err = super_types[mddev->major_version]
3507                                 .load_super(rdev, rdev0, mddev->minor_version);
3508                         if (err < 0)
3509                                 goto out;
3510                 }
3511         } else if (mddev->external)
3512                 rdev = md_import_device(dev, -2, -1);
3513         else
3514                 rdev = md_import_device(dev, -1, -1);
3515
3516         if (IS_ERR(rdev))
3517                 return PTR_ERR(rdev);
3518         err = bind_rdev_to_array(rdev, mddev);
3519  out:
3520         if (err)
3521                 export_rdev(rdev);
3522         return err ? err : len;
3523 }
3524
3525 static struct md_sysfs_entry md_new_device =
3526 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3527
3528 static ssize_t
3529 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3530 {
3531         char *end;
3532         unsigned long chunk, end_chunk;
3533
3534         if (!mddev->bitmap)
3535                 goto out;
3536         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3537         while (*buf) {
3538                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3539                 if (buf == end) break;
3540                 if (*end == '-') { /* range */
3541                         buf = end + 1;
3542                         end_chunk = simple_strtoul(buf, &end, 0);
3543                         if (buf == end) break;
3544                 }
3545                 if (*end && !isspace(*end)) break;
3546                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3547                 buf = skip_spaces(end);
3548         }
3549         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3550 out:
3551         return len;
3552 }
3553
3554 static struct md_sysfs_entry md_bitmap =
3555 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3556
3557 static ssize_t
3558 size_show(mddev_t *mddev, char *page)
3559 {
3560         return sprintf(page, "%llu\n",
3561                 (unsigned long long)mddev->dev_sectors / 2);
3562 }
3563
3564 static int update_size(mddev_t *mddev, sector_t num_sectors);
3565
3566 static ssize_t
3567 size_store(mddev_t *mddev, const char *buf, size_t len)
3568 {
3569         /* If array is inactive, we can reduce the component size, but
3570          * not increase it (except from 0).
3571          * If array is active, we can try an on-line resize
3572          */
3573         sector_t sectors;
3574         int err = strict_blocks_to_sectors(buf, &sectors);
3575
3576         if (err < 0)
3577                 return err;
3578         if (mddev->pers) {
3579                 err = update_size(mddev, sectors);
3580                 md_update_sb(mddev, 1);
3581         } else {
3582                 if (mddev->dev_sectors == 0 ||
3583                     mddev->dev_sectors > sectors)
3584                         mddev->dev_sectors = sectors;
3585                 else
3586                         err = -ENOSPC;
3587         }
3588         return err ? err : len;
3589 }
3590
3591 static struct md_sysfs_entry md_size =
3592 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3593
3594
3595 /* Metdata version.
3596  * This is one of
3597  *   'none' for arrays with no metadata (good luck...)
3598  *   'external' for arrays with externally managed metadata,
3599  * or N.M for internally known formats
3600  */
3601 static ssize_t
3602 metadata_show(mddev_t *mddev, char *page)
3603 {
3604         if (mddev->persistent)
3605                 return sprintf(page, "%d.%d\n",
3606                                mddev->major_version, mddev->minor_version);
3607         else if (mddev->external)
3608                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3609         else
3610                 return sprintf(page, "none\n");
3611 }
3612
3613 static ssize_t
3614 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3615 {
3616         int major, minor;
3617         char *e;
3618         /* Changing the details of 'external' metadata is
3619          * always permitted.  Otherwise there must be
3620          * no devices attached to the array.
3621          */
3622         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3623                 ;
3624         else if (!list_empty(&mddev->disks))
3625                 return -EBUSY;
3626
3627         if (cmd_match(buf, "none")) {
3628                 mddev->persistent = 0;
3629                 mddev->external = 0;
3630                 mddev->major_version = 0;
3631                 mddev->minor_version = 90;
3632                 return len;
3633         }
3634         if (strncmp(buf, "external:", 9) == 0) {
3635                 size_t namelen = len-9;
3636                 if (namelen >= sizeof(mddev->metadata_type))
3637                         namelen = sizeof(mddev->metadata_type)-1;
3638                 strncpy(mddev->metadata_type, buf+9, namelen);
3639                 mddev->metadata_type[namelen] = 0;
3640                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3641                         mddev->metadata_type[--namelen] = 0;
3642                 mddev->persistent = 0;
3643                 mddev->external = 1;
3644                 mddev->major_version = 0;
3645                 mddev->minor_version = 90;
3646                 return len;
3647         }
3648         major = simple_strtoul(buf, &e, 10);
3649         if (e==buf || *e != '.')
3650                 return -EINVAL;
3651         buf = e+1;
3652         minor = simple_strtoul(buf, &e, 10);
3653         if (e==buf || (*e && *e != '\n') )
3654                 return -EINVAL;
3655         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3656                 return -ENOENT;
3657         mddev->major_version = major;
3658         mddev->minor_version = minor;
3659         mddev->persistent = 1;
3660         mddev->external = 0;
3661         return len;
3662 }
3663
3664 static struct md_sysfs_entry md_metadata =
3665 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3666
3667 static ssize_t
3668 action_show(mddev_t *mddev, char *page)
3669 {
3670         char *type = "idle";
3671         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3672                 type = "frozen";
3673         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3674             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3675                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3676                         type = "reshape";
3677                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3678                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3679                                 type = "resync";
3680                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3681                                 type = "check";
3682                         else
3683                                 type = "repair";
3684                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3685                         type = "recover";
3686         }
3687         return sprintf(page, "%s\n", type);
3688 }
3689
3690 static ssize_t
3691 action_store(mddev_t *mddev, const char *page, size_t len)
3692 {
3693         if (!mddev->pers || !mddev->pers->sync_request)
3694                 return -EINVAL;
3695
3696         if (cmd_match(page, "frozen"))
3697                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3698         else
3699                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3700
3701         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3702                 if (mddev->sync_thread) {
3703                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3704                         md_unregister_thread(mddev->sync_thread);
3705                         mddev->sync_thread = NULL;
3706                         mddev->recovery = 0;
3707                 }
3708         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3709                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3710                 return -EBUSY;
3711         else if (cmd_match(page, "resync"))
3712                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3713         else if (cmd_match(page, "recover")) {
3714                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3715                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3716         } else if (cmd_match(page, "reshape")) {
3717                 int err;
3718                 if (mddev->pers->start_reshape == NULL)
3719                         return -EINVAL;
3720                 err = mddev->pers->start_reshape(mddev);
3721                 if (err)
3722                         return err;
3723                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3724         } else {
3725                 if (cmd_match(page, "check"))
3726                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3727                 else if (!cmd_match(page, "repair"))
3728                         return -EINVAL;
3729                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3730                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3731         }
3732         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3733         md_wakeup_thread(mddev->thread);
3734         sysfs_notify_dirent_safe(mddev->sysfs_action);
3735         return len;
3736 }
3737
3738 static ssize_t
3739 mismatch_cnt_show(mddev_t *mddev, char *page)
3740 {
3741         return sprintf(page, "%llu\n",
3742                        (unsigned long long) mddev->resync_mismatches);
3743 }
3744
3745 static struct md_sysfs_entry md_scan_mode =
3746 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3747
3748
3749 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3750
3751 static ssize_t
3752 sync_min_show(mddev_t *mddev, char *page)
3753 {
3754         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3755                        mddev->sync_speed_min ? "local": "system");
3756 }
3757
3758 static ssize_t
3759 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3760 {
3761         int min;
3762         char *e;
3763         if (strncmp(buf, "system", 6)==0) {
3764                 mddev->sync_speed_min = 0;
3765                 return len;
3766         }
3767         min = simple_strtoul(buf, &e, 10);
3768         if (buf == e || (*e && *e != '\n') || min <= 0)
3769                 return -EINVAL;
3770         mddev->sync_speed_min = min;
3771         return len;
3772 }
3773
3774 static struct md_sysfs_entry md_sync_min =
3775 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3776
3777 static ssize_t
3778 sync_max_show(mddev_t *mddev, char *page)
3779 {
3780         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3781                        mddev->sync_speed_max ? "local": "system");
3782 }
3783
3784 static ssize_t
3785 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3786 {
3787         int max;
3788         char *e;
3789         if (strncmp(buf, "system", 6)==0) {
3790                 mddev->sync_speed_max = 0;
3791                 return len;
3792         }
3793         max = simple_strtoul(buf, &e, 10);
3794         if (buf == e || (*e && *e != '\n') || max <= 0)
3795                 return -EINVAL;
3796         mddev->sync_speed_max = max;
3797         return len;
3798 }
3799
3800 static struct md_sysfs_entry md_sync_max =
3801 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3802
3803 static ssize_t
3804 degraded_show(mddev_t *mddev, char *page)
3805 {
3806         return sprintf(page, "%d\n", mddev->degraded);
3807 }
3808 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3809
3810 static ssize_t
3811 sync_force_parallel_show(mddev_t *mddev, char *page)
3812 {
3813         return sprintf(page, "%d\n", mddev->parallel_resync);
3814 }
3815
3816 static ssize_t
3817 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3818 {
3819         long n;
3820
3821         if (strict_strtol(buf, 10, &n))
3822                 return -EINVAL;
3823
3824         if (n != 0 && n != 1)
3825                 return -EINVAL;
3826
3827         mddev->parallel_resync = n;
3828
3829         if (mddev->sync_thread)
3830                 wake_up(&resync_wait);
3831
3832         return len;
3833 }
3834
3835 /* force parallel resync, even with shared block devices */
3836 static struct md_sysfs_entry md_sync_force_parallel =
3837 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3838        sync_force_parallel_show, sync_force_parallel_store);
3839
3840 static ssize_t
3841 sync_speed_show(mddev_t *mddev, char *page)
3842 {
3843         unsigned long resync, dt, db;
3844         if (mddev->curr_resync == 0)
3845                 return sprintf(page, "none\n");
3846         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3847         dt = (jiffies - mddev->resync_mark) / HZ;
3848         if (!dt) dt++;
3849         db = resync - mddev->resync_mark_cnt;
3850         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3851 }
3852
3853 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3854
3855 static ssize_t
3856 sync_completed_show(mddev_t *mddev, char *page)
3857 {
3858         unsigned long max_sectors, resync;
3859
3860         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3861                 return sprintf(page, "none\n");
3862
3863         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3864                 max_sectors = mddev->resync_max_sectors;
3865         else
3866                 max_sectors = mddev->dev_sectors;
3867
3868         resync = mddev->curr_resync_completed;
3869         return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3870 }
3871
3872 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3873
3874 static ssize_t
3875 min_sync_show(mddev_t *mddev, char *page)
3876 {
3877         return sprintf(page, "%llu\n",
3878                        (unsigned long long)mddev->resync_min);
3879 }
3880 static ssize_t
3881 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3882 {
3883         unsigned long long min;
3884         if (strict_strtoull(buf, 10, &min))
3885                 return -EINVAL;
3886         if (min > mddev->resync_max)
3887                 return -EINVAL;
3888         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3889                 return -EBUSY;
3890
3891         /* Must be a multiple of chunk_size */
3892         if (mddev->chunk_sectors) {
3893                 sector_t temp = min;
3894                 if (sector_div(temp, mddev->chunk_sectors))
3895                         return -EINVAL;
3896         }
3897         mddev->resync_min = min;
3898
3899         return len;
3900 }
3901
3902 static struct md_sysfs_entry md_min_sync =
3903 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3904
3905 static ssize_t
3906 max_sync_show(mddev_t *mddev, char *page)
3907 {
3908         if (mddev->resync_max == MaxSector)
3909                 return sprintf(page, "max\n");
3910         else
3911                 return sprintf(page, "%llu\n",
3912                                (unsigned long long)mddev->resync_max);
3913 }
3914 static ssize_t
3915 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3916 {
3917         if (strncmp(buf, "max", 3) == 0)
3918                 mddev->resync_max = MaxSector;
3919         else {
3920                 unsigned long long max;
3921                 if (strict_strtoull(buf, 10, &max))
3922                         return -EINVAL;
3923                 if (max < mddev->resync_min)
3924                         return -EINVAL;
3925                 if (max < mddev->resync_max &&
3926                     mddev->ro == 0 &&
3927                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3928                         return -EBUSY;
3929
3930                 /* Must be a multiple of chunk_size */
3931                 if (mddev->chunk_sectors) {
3932                         sector_t temp = max;
3933                         if (sector_div(temp, mddev->chunk_sectors))
3934                                 return -EINVAL;
3935                 }
3936                 mddev->resync_max = max;
3937         }
3938         wake_up(&mddev->recovery_wait);
3939         return len;
3940 }
3941
3942 static struct md_sysfs_entry md_max_sync =
3943 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3944
3945 static ssize_t
3946 suspend_lo_show(mddev_t *mddev, char *page)
3947 {
3948         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3949 }
3950
3951 static ssize_t
3952 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3953 {
3954         char *e;
3955         unsigned long long new = simple_strtoull(buf, &e, 10);
3956
3957         if (mddev->pers == NULL ||
3958             mddev->pers->quiesce == NULL)
3959                 return -EINVAL;
3960         if (buf == e || (*e && *e != '\n'))
3961                 return -EINVAL;
3962         if (new >= mddev->suspend_hi ||
3963             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3964                 mddev->suspend_lo = new;
3965                 mddev->pers->quiesce(mddev, 2);
3966                 return len;
3967         } else
3968                 return -EINVAL;
3969 }
3970 static struct md_sysfs_entry md_suspend_lo =
3971 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3972
3973
3974 static ssize_t
3975 suspend_hi_show(mddev_t *mddev, char *page)
3976 {
3977         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3978 }
3979
3980 static ssize_t
3981 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3982 {
3983         char *e;
3984         unsigned long long new = simple_strtoull(buf, &e, 10);
3985
3986         if (mddev->pers == NULL ||
3987             mddev->pers->quiesce == NULL)
3988                 return -EINVAL;
3989         if (buf == e || (*e && *e != '\n'))
3990                 return -EINVAL;
3991         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3992             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3993                 mddev->suspend_hi = new;
3994                 mddev->pers->quiesce(mddev, 1);
3995                 mddev->pers->quiesce(mddev, 0);
3996                 return len;
3997         } else
3998                 return -EINVAL;
3999 }
4000 static struct md_sysfs_entry md_suspend_hi =
4001 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4002
4003 static ssize_t
4004 reshape_position_show(mddev_t *mddev, char *page)
4005 {
4006         if (mddev->reshape_position != MaxSector)
4007                 return sprintf(page, "%llu\n",
4008                                (unsigned long long)mddev->reshape_position);
4009         strcpy(page, "none\n");
4010         return 5;
4011 }
4012
4013 static ssize_t
4014 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4015 {
4016         char *e;
4017         unsigned long long new = simple_strtoull(buf, &e, 10);
4018         if (mddev->pers)
4019                 return -EBUSY;
4020         if (buf == e || (*e && *e != '\n'))
4021                 return -EINVAL;
4022         mddev->reshape_position = new;
4023         mddev->delta_disks = 0;
4024         mddev->new_level = mddev->level;
4025         mddev->new_layout = mddev->layout;
4026         mddev->new_chunk_sectors = mddev->chunk_sectors;
4027         return len;
4028 }
4029
4030 static struct md_sysfs_entry md_reshape_position =
4031 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4032        reshape_position_store);
4033
4034 static ssize_t
4035 array_size_show(mddev_t *mddev, char *page)
4036 {
4037         if (mddev->external_size)
4038                 return sprintf(page, "%llu\n",
4039                                (unsigned long long)mddev->array_sectors/2);
4040         else
4041                 return sprintf(page, "default\n");
4042 }
4043
4044 static ssize_t
4045 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4046 {
4047         sector_t sectors;
4048
4049         if (strncmp(buf, "default", 7) == 0) {
4050                 if (mddev->pers)
4051                         sectors = mddev->pers->size(mddev, 0, 0);
4052                 else
4053                         sectors = mddev->array_sectors;
4054
4055                 mddev->external_size = 0;
4056         } else {
4057                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4058                         return -EINVAL;
4059                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4060                         return -E2BIG;
4061
4062                 mddev->external_size = 1;
4063         }
4064
4065         mddev->array_sectors = sectors;
4066         set_capacity(mddev->gendisk, mddev->array_sectors);
4067         if (mddev->pers)
4068                 revalidate_disk(mddev->gendisk);
4069
4070         return len;
4071 }
4072
4073 static struct md_sysfs_entry md_array_size =
4074 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4075        array_size_store);
4076
4077 static struct attribute *md_default_attrs[] = {
4078         &md_level.attr,
4079         &md_layout.attr,
4080         &md_raid_disks.attr,
4081         &md_chunk_size.attr,
4082         &md_size.attr,
4083         &md_resync_start.attr,
4084         &md_metadata.attr,
4085         &md_new_device.attr,
4086         &md_safe_delay.attr,
4087         &md_array_state.attr,
4088         &md_reshape_position.attr,
4089         &md_array_size.attr,
4090         &max_corr_read_errors.attr,
4091         NULL,
4092 };
4093
4094 static struct attribute *md_redundancy_attrs[] = {
4095         &md_scan_mode.attr,
4096         &md_mismatches.attr,
4097         &md_sync_min.attr,
4098         &md_sync_max.attr,
4099         &md_sync_speed.attr,
4100         &md_sync_force_parallel.attr,
4101         &md_sync_completed.attr,
4102         &md_min_sync.attr,
4103         &md_max_sync.attr,
4104         &md_suspend_lo.attr,
4105         &md_suspend_hi.attr,
4106         &md_bitmap.attr,
4107         &md_degraded.attr,
4108         NULL,
4109 };
4110 static struct attribute_group md_redundancy_group = {
4111         .name = NULL,
4112         .attrs = md_redundancy_attrs,
4113 };
4114
4115
4116 static ssize_t
4117 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4118 {
4119         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4120         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4121         ssize_t rv;
4122
4123         if (!entry->show)
4124                 return -EIO;
4125         rv = mddev_lock(mddev);
4126         if (!rv) {
4127                 rv = entry->show(mddev, page);
4128                 mddev_unlock(mddev);
4129         }
4130         return rv;
4131 }
4132
4133 static ssize_t
4134 md_attr_store(struct kobject *kobj, struct attribute *attr,
4135               const char *page, size_t length)
4136 {
4137         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4138         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4139         ssize_t rv;
4140
4141         if (!entry->store)
4142                 return -EIO;
4143         if (!capable(CAP_SYS_ADMIN))
4144                 return -EACCES;
4145         rv = mddev_lock(mddev);
4146         if (mddev->hold_active == UNTIL_IOCTL)
4147                 mddev->hold_active = 0;
4148         if (!rv) {
4149                 rv = entry->store(mddev, page, length);
4150                 mddev_unlock(mddev);
4151         }
4152         return rv;
4153 }
4154
4155 static void md_free(struct kobject *ko)
4156 {
4157         mddev_t *mddev = container_of(ko, mddev_t, kobj);
4158
4159         if (mddev->sysfs_state)
4160                 sysfs_put(mddev->sysfs_state);
4161
4162         if (mddev->gendisk) {
4163                 del_gendisk(mddev->gendisk);
4164                 put_disk(mddev->gendisk);
4165         }
4166         if (mddev->queue)
4167                 blk_cleanup_queue(mddev->queue);
4168
4169         kfree(mddev);
4170 }
4171
4172 static const struct sysfs_ops md_sysfs_ops = {
4173         .show   = md_attr_show,
4174         .store  = md_attr_store,
4175 };
4176 static struct kobj_type md_ktype = {
4177         .release        = md_free,
4178         .sysfs_ops      = &md_sysfs_ops,
4179         .default_attrs  = md_default_attrs,
4180 };
4181
4182 int mdp_major = 0;
4183
4184 static void mddev_delayed_delete(struct work_struct *ws)
4185 {
4186         mddev_t *mddev = container_of(ws, mddev_t, del_work);
4187
4188         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4189         kobject_del(&mddev->kobj);
4190         kobject_put(&mddev->kobj);
4191 }
4192
4193 static int md_alloc(dev_t dev, char *name)
4194 {
4195         static DEFINE_MUTEX(disks_mutex);
4196         mddev_t *mddev = mddev_find(dev);
4197         struct gendisk *disk;
4198         int partitioned;
4199         int shift;
4200         int unit;
4201         int error;
4202
4203         if (!mddev)
4204                 return -ENODEV;
4205
4206         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4207         shift = partitioned ? MdpMinorShift : 0;
4208         unit = MINOR(mddev->unit) >> shift;
4209
4210         /* wait for any previous instance if this device
4211          * to be completed removed (mddev_delayed_delete).
4212          */
4213         flush_scheduled_work();
4214
4215         mutex_lock(&disks_mutex);
4216         error = -EEXIST;
4217         if (mddev->gendisk)
4218                 goto abort;
4219
4220         if (name) {
4221                 /* Need to ensure that 'name' is not a duplicate.
4222                  */
4223                 mddev_t *mddev2;
4224                 spin_lock(&all_mddevs_lock);
4225
4226                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4227                         if (mddev2->gendisk &&
4228                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
4229                                 spin_unlock(&all_mddevs_lock);
4230                                 goto abort;
4231                         }
4232                 spin_unlock(&all_mddevs_lock);
4233         }
4234
4235         error = -ENOMEM;
4236         mddev->queue = blk_alloc_queue(GFP_KERNEL);
4237         if (!mddev->queue)
4238                 goto abort;
4239         mddev->queue->queuedata = mddev;
4240
4241         /* Can be unlocked because the queue is new: no concurrency */
4242         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4243
4244         blk_queue_make_request(mddev->queue, md_make_request);
4245
4246         disk = alloc_disk(1 << shift);
4247         if (!disk) {
4248                 blk_cleanup_queue(mddev->queue);
4249                 mddev->queue = NULL;
4250                 goto abort;
4251         }
4252         disk->major = MAJOR(mddev->unit);
4253         disk->first_minor = unit << shift;
4254         if (name)
4255                 strcpy(disk->disk_name, name);
4256         else if (partitioned)
4257                 sprintf(disk->disk_name, "md_d%d", unit);
4258         else
4259                 sprintf(disk->disk_name, "md%d", unit);
4260         disk->fops = &md_fops;
4261         disk->private_data = mddev;
4262         disk->queue = mddev->queue;
4263         /* Allow extended partitions.  This makes the
4264          * 'mdp' device redundant, but we can't really
4265          * remove it now.
4266          */
4267         disk->flags |= GENHD_FL_EXT_DEVT;
4268         add_disk(disk);
4269         mddev->gendisk = disk;
4270         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4271                                      &disk_to_dev(disk)->kobj, "%s", "md");
4272         if (error) {
4273                 /* This isn't possible, but as kobject_init_and_add is marked
4274                  * __must_check, we must do something with the result
4275                  */
4276                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4277                        disk->disk_name);
4278                 error = 0;
4279         }
4280         if (mddev->kobj.sd &&
4281             sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4282                 printk(KERN_DEBUG "pointless warning\n");
4283  abort:
4284         mutex_unlock(&disks_mutex);
4285         if (!error && mddev->kobj.sd) {
4286                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4287                 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4288         }
4289         mddev_put(mddev);
4290         return error;
4291 }
4292
4293 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4294 {
4295         md_alloc(dev, NULL);
4296         return NULL;
4297 }
4298
4299 static int add_named_array(const char *val, struct kernel_param *kp)
4300 {
4301         /* val must be "md_*" where * is not all digits.
4302          * We allocate an array with a large free minor number, and
4303          * set the name to val.  val must not already be an active name.
4304          */
4305         int len = strlen(val);
4306         char buf[DISK_NAME_LEN];
4307
4308         while (len && val[len-1] == '\n')
4309                 len--;
4310         if (len >= DISK_NAME_LEN)
4311                 return -E2BIG;
4312         strlcpy(buf, val, len+1);
4313         if (strncmp(buf, "md_", 3) != 0)
4314                 return -EINVAL;
4315         return md_alloc(0, buf);
4316 }
4317
4318 static void md_safemode_timeout(unsigned long data)
4319 {
4320         mddev_t *mddev = (mddev_t *) data;
4321
4322         if (!atomic_read(&mddev->writes_pending)) {
4323                 mddev->safemode = 1;
4324                 if (mddev->external)
4325                         sysfs_notify_dirent_safe(mddev->sysfs_state);
4326         }
4327         md_wakeup_thread(mddev->thread);
4328 }
4329
4330 static int start_dirty_degraded;
4331
4332 static int md_run(mddev_t *mddev)
4333 {
4334         int err;
4335         mdk_rdev_t *rdev;
4336         struct mdk_personality *pers;
4337
4338         if (list_empty(&mddev->disks))
4339                 /* cannot run an array with no devices.. */
4340                 return -EINVAL;
4341
4342         if (mddev->pers)
4343                 return -EBUSY;
4344
4345         /* These two calls synchronise us with the
4346          * sysfs_remove_group calls in mddev_unlock,
4347          * so they must have completed.
4348          */
4349         mutex_lock(&mddev->open_mutex);
4350         mutex_unlock(&mddev->open_mutex);
4351
4352         /*
4353          * Analyze all RAID superblock(s)
4354          */
4355         if (!mddev->raid_disks) {
4356                 if (!mddev->persistent)
4357                         return -EINVAL;
4358                 analyze_sbs(mddev);
4359         }
4360
4361         if (mddev->level != LEVEL_NONE)
4362                 request_module("md-level-%d", mddev->level);
4363         else if (mddev->clevel[0])
4364                 request_module("md-%s", mddev->clevel);
4365
4366         /*
4367          * Drop all container device buffers, from now on
4368          * the only valid external interface is through the md
4369          * device.
4370          */
4371         list_for_each_entry(rdev, &mddev->disks, same_set) {
4372                 if (test_bit(Faulty, &rdev->flags))
4373                         continue;
4374                 sync_blockdev(rdev->bdev);
4375                 invalidate_bdev(rdev->bdev);
4376
4377                 /* perform some consistency tests on the device.
4378                  * We don't want the data to overlap the metadata,
4379                  * Internal Bitmap issues have been handled elsewhere.
4380                  */
4381                 if (rdev->data_offset < rdev->sb_start) {
4382                         if (mddev->dev_sectors &&
4383                             rdev->data_offset + mddev->dev_sectors
4384                             > rdev->sb_start) {
4385                                 printk("md: %s: data overlaps metadata\n",
4386                                        mdname(mddev));
4387                                 return -EINVAL;
4388                         }
4389                 } else {
4390                         if (rdev->sb_start + rdev->sb_size/512
4391                             > rdev->data_offset) {
4392                                 printk("md: %s: metadata overlaps data\n",
4393                                        mdname(mddev));
4394                                 return -EINVAL;
4395                         }
4396                 }
4397                 sysfs_notify_dirent_safe(rdev->sysfs_state);
4398         }
4399
4400         spin_lock(&pers_lock);
4401         pers = find_pers(mddev->level, mddev->clevel);
4402         if (!pers || !try_module_get(pers->owner)) {
4403                 spin_unlock(&pers_lock);
4404                 if (mddev->level != LEVEL_NONE)
4405                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4406                                mddev->level);
4407                 else
4408                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4409                                mddev->clevel);
4410                 return -EINVAL;
4411         }
4412         mddev->pers = pers;
4413         spin_unlock(&pers_lock);
4414         if (mddev->level != pers->level) {
4415                 mddev->level = pers->level;
4416                 mddev->new_level = pers->level;
4417         }
4418         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4419
4420         if (mddev->reshape_position != MaxSector &&
4421             pers->start_reshape == NULL) {
4422                 /* This personality cannot handle reshaping... */
4423                 mddev->pers = NULL;
4424                 module_put(pers->owner);
4425                 return -EINVAL;
4426         }
4427
4428         if (pers->sync_request) {
4429                 /* Warn if this is a potentially silly
4430                  * configuration.
4431                  */
4432                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4433                 mdk_rdev_t *rdev2;
4434                 int warned = 0;
4435
4436                 list_for_each_entry(rdev, &mddev->disks, same_set)
4437                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
4438                                 if (rdev < rdev2 &&
4439                                     rdev->bdev->bd_contains ==
4440                                     rdev2->bdev->bd_contains) {
4441                                         printk(KERN_WARNING
4442                                                "%s: WARNING: %s appears to be"
4443                                                " on the same physical disk as"
4444                                                " %s.\n",
4445                                                mdname(mddev),
4446                                                bdevname(rdev->bdev,b),
4447                                                bdevname(rdev2->bdev,b2));
4448                                         warned = 1;
4449                                 }
4450                         }
4451
4452                 if (warned)
4453                         printk(KERN_WARNING
4454                                "True protection against single-disk"
4455                                " failure might be compromised.\n");
4456         }
4457
4458         mddev->recovery = 0;
4459         /* may be over-ridden by personality */
4460         mddev->resync_max_sectors = mddev->dev_sectors;
4461
4462         mddev->barriers_work = 1;
4463         mddev->ok_start_degraded = start_dirty_degraded;
4464
4465         if (start_readonly && mddev->ro == 0)
4466                 mddev->ro = 2; /* read-only, but switch on first write */
4467
4468         err = mddev->pers->run(mddev);
4469         if (err)
4470                 printk(KERN_ERR "md: pers->run() failed ...\n");
4471         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4472                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4473                           " but 'external_size' not in effect?\n", __func__);
4474                 printk(KERN_ERR
4475                        "md: invalid array_size %llu > default size %llu\n",
4476                        (unsigned long long)mddev->array_sectors / 2,
4477                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4478                 err = -EINVAL;
4479                 mddev->pers->stop(mddev);
4480         }
4481         if (err == 0 && mddev->pers->sync_request) {
4482                 err = bitmap_create(mddev);
4483                 if (err) {
4484                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4485                                mdname(mddev), err);
4486                         mddev->pers->stop(mddev);
4487                 }
4488         }
4489         if (err) {
4490                 module_put(mddev->pers->owner);
4491                 mddev->pers = NULL;
4492                 bitmap_destroy(mddev);
4493                 return err;
4494         }
4495         if (mddev->pers->sync_request) {
4496                 if (mddev->kobj.sd &&
4497                     sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4498                         printk(KERN_WARNING
4499                                "md: cannot register extra attributes for %s\n",
4500                                mdname(mddev));
4501                 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4502         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4503                 mddev->ro = 0;
4504
4505         atomic_set(&mddev->writes_pending,0);
4506         atomic_set(&mddev->max_corr_read_errors,
4507                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4508         mddev->safemode = 0;
4509         mddev->safemode_timer.function = md_safemode_timeout;
4510         mddev->safemode_timer.data = (unsigned long) mddev;
4511         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4512         mddev->in_sync = 1;
4513
4514         list_for_each_entry(rdev, &mddev->disks, same_set)
4515                 if (rdev->raid_disk >= 0) {
4516                         char nm[20];
4517                         sprintf(nm, "rd%d", rdev->raid_disk);
4518                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4519                                 /* failure here is OK */;
4520                 }
4521
4522         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4523
4524         if (mddev->flags)
4525                 md_update_sb(mddev, 0);
4526
4527         md_wakeup_thread(mddev->thread);
4528         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4529
4530         md_new_event(mddev);
4531         sysfs_notify_dirent_safe(mddev->sysfs_state);
4532         sysfs_notify_dirent_safe(mddev->sysfs_action);
4533         sysfs_notify(&mddev->kobj, NULL, "degraded");
4534         return 0;
4535 }
4536
4537 static int do_md_run(mddev_t *mddev)
4538 {
4539         int err;
4540
4541         err = md_run(mddev);
4542         if (err)
4543                 goto out;
4544
4545         set_capacity(mddev->gendisk, mddev->array_sectors);
4546         revalidate_disk(mddev->gendisk);
4547         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4548 out:
4549         return err;
4550 }
4551
4552 static int restart_array(mddev_t *mddev)
4553 {
4554         struct gendisk *disk = mddev->gendisk;
4555
4556         /* Complain if it has no devices */
4557         if (list_empty(&mddev->disks))
4558                 return -ENXIO;
4559         if (!mddev->pers)
4560                 return -EINVAL;
4561         if (!mddev->ro)
4562                 return -EBUSY;
4563         mddev->safemode = 0;
4564         mddev->ro = 0;
4565         set_disk_ro(disk, 0);
4566         printk(KERN_INFO "md: %s switched to read-write mode.\n",
4567                 mdname(mddev));
4568         /* Kick recovery or resync if necessary */
4569         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4570         md_wakeup_thread(mddev->thread);
4571         md_wakeup_thread(mddev->sync_thread);
4572         sysfs_notify_dirent_safe(mddev->sysfs_state);
4573         return 0;
4574 }
4575
4576 /* similar to deny_write_access, but accounts for our holding a reference
4577  * to the file ourselves */
4578 static int deny_bitmap_write_access(struct file * file)
4579 {
4580         struct inode *inode = file->f_mapping->host;
4581
4582         spin_lock(&inode->i_lock);
4583         if (atomic_read(&inode->i_writecount) > 1) {
4584                 spin_unlock(&inode->i_lock);
4585                 return -ETXTBSY;
4586         }
4587         atomic_set(&inode->i_writecount, -1);
4588         spin_unlock(&inode->i_lock);
4589
4590         return 0;
4591 }
4592
4593 void restore_bitmap_write_access(struct file *file)
4594 {
4595         struct inode *inode = file->f_mapping->host;
4596
4597         spin_lock(&inode->i_lock);
4598         atomic_set(&inode->i_writecount, 1);
4599         spin_unlock(&inode->i_lock);
4600 }
4601
4602 static void md_clean(mddev_t *mddev)
4603 {
4604         mddev->array_sectors = 0;
4605         mddev->external_size = 0;
4606         mddev->dev_sectors = 0;
4607         mddev->raid_disks = 0;
4608         mddev->recovery_cp = 0;
4609         mddev->resync_min = 0;
4610         mddev->resync_max = MaxSector;
4611         mddev->reshape_position = MaxSector;
4612         mddev->external = 0;
4613         mddev->persistent = 0;
4614         mddev->level = LEVEL_NONE;
4615         mddev->clevel[0] = 0;
4616         mddev->flags = 0;
4617         mddev->ro = 0;
4618         mddev->metadata_type[0] = 0;
4619         mddev->chunk_sectors = 0;
4620         mddev->ctime = mddev->utime = 0;
4621         mddev->layout = 0;
4622         mddev->max_disks = 0;
4623         mddev->events = 0;
4624         mddev->can_decrease_events = 0;
4625         mddev->delta_disks = 0;
4626         mddev->new_level = LEVEL_NONE;
4627         mddev->new_layout = 0;
4628         mddev->new_chunk_sectors = 0;
4629         mddev->curr_resync = 0;
4630         mddev->resync_mismatches = 0;
4631         mddev->suspend_lo = mddev->suspend_hi = 0;
4632         mddev->sync_speed_min = mddev->sync_speed_max = 0;
4633         mddev->recovery = 0;
4634         mddev->in_sync = 0;
4635         mddev->degraded = 0;
4636         mddev->barriers_work = 0;
4637         mddev->safemode = 0;
4638         mddev->bitmap_info.offset = 0;
4639         mddev->bitmap_info.default_offset = 0;
4640         mddev->bitmap_info.chunksize = 0;
4641         mddev->bitmap_info.daemon_sleep = 0;
4642         mddev->bitmap_info.max_write_behind = 0;
4643 }
4644
4645 static void md_stop_writes(mddev_t *mddev)
4646 {
4647         if (mddev->sync_thread) {
4648                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4649                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4650                 md_unregister_thread(mddev->sync_thread);
4651                 mddev->sync_thread = NULL;
4652         }
4653
4654         del_timer_sync(&mddev->safemode_timer);
4655
4656         bitmap_flush(mddev);
4657         md_super_wait(mddev);
4658
4659         if (!mddev->in_sync || mddev->flags) {
4660                 /* mark array as shutdown cleanly */
4661                 mddev->in_sync = 1;
4662                 md_update_sb(mddev, 1);
4663         }
4664 }
4665
4666 static void md_stop(mddev_t *mddev)
4667 {
4668         md_stop_writes(mddev);
4669
4670         mddev->pers->stop(mddev);
4671         if (mddev->pers->sync_request && mddev->to_remove == NULL)
4672                 mddev->to_remove = &md_redundancy_group;
4673         module_put(mddev->pers->owner);
4674         mddev->pers = NULL;
4675         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4676 }
4677
4678 static int md_set_readonly(mddev_t *mddev, int is_open)
4679 {
4680         int err = 0;
4681         mutex_lock(&mddev->open_mutex);
4682         if (atomic_read(&mddev->openers) > is_open) {
4683                 printk("md: %s still in use.\n",mdname(mddev));
4684                 err = -EBUSY;
4685                 goto out;
4686         }
4687         if (mddev->pers) {
4688                 md_stop_writes(mddev);
4689
4690                 err  = -ENXIO;
4691                 if (mddev->ro==1)
4692                         goto out;
4693                 mddev->ro = 1;
4694                 set_disk_ro(mddev->gendisk, 1);
4695                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4696                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4697                 err = 0;
4698         }
4699 out:
4700         mutex_unlock(&mddev->open_mutex);
4701         return err;
4702 }
4703
4704 /* mode:
4705  *   0 - completely stop and dis-assemble array
4706  *   2 - stop but do not disassemble array
4707  */
4708 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4709 {
4710         int err = 0;
4711         struct gendisk *disk = mddev->gendisk;
4712         mdk_rdev_t *rdev;
4713
4714         mutex_lock(&mddev->open_mutex);
4715         if (atomic_read(&mddev->openers) > is_open) {
4716                 printk("md: %s still in use.\n",mdname(mddev));
4717                 err = -EBUSY;
4718         } else if (mddev->pers) {
4719
4720                 if (mddev->ro)
4721                         set_disk_ro(disk, 0);
4722
4723                 md_stop(mddev);
4724                 mddev->queue->merge_bvec_fn = NULL;
4725                 mddev->queue->unplug_fn = NULL;
4726                 mddev->queue->backing_dev_info.congested_fn = NULL;
4727
4728                 /* tell userspace to handle 'inactive' */
4729                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4730
4731                 list_for_each_entry(rdev, &mddev->disks, same_set)
4732                         if (rdev->raid_disk >= 0) {
4733                                 char nm[20];
4734                                 sprintf(nm, "rd%d", rdev->raid_disk);
4735                                 sysfs_remove_link(&mddev->kobj, nm);
4736                         }
4737
4738                 set_capacity(disk, 0);
4739                 revalidate_disk(disk);
4740
4741                 if (mddev->ro)
4742                         mddev->ro = 0;
4743
4744                 err = 0;
4745         }
4746         mutex_unlock(&mddev->open_mutex);
4747         if (err)
4748                 return err;
4749         /*
4750          * Free resources if final stop
4751          */
4752         if (mode == 0) {
4753
4754                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4755
4756                 bitmap_destroy(mddev);
4757                 if (mddev->bitmap_info.file) {
4758                         restore_bitmap_write_access(mddev->bitmap_info.file);
4759                         fput(mddev->bitmap_info.file);
4760                         mddev->bitmap_info.file = NULL;
4761                 }
4762                 mddev->bitmap_info.offset = 0;
4763
4764                 export_array(mddev);
4765
4766                 md_clean(mddev);
4767                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4768                 if (mddev->hold_active == UNTIL_STOP)
4769                         mddev->hold_active = 0;
4770
4771         }
4772         err = 0;
4773         blk_integrity_unregister(disk);
4774         md_new_event(mddev);
4775         sysfs_notify_dirent_safe(mddev->sysfs_state);
4776         return err;
4777 }
4778
4779 #ifndef MODULE
4780 static void autorun_array(mddev_t *mddev)
4781 {
4782         mdk_rdev_t *rdev;
4783         int err;
4784
4785         if (list_empty(&mddev->disks))
4786                 return;
4787
4788         printk(KERN_INFO "md: running: ");
4789
4790         list_for_each_entry(rdev, &mddev->disks, same_set) {
4791                 char b[BDEVNAME_SIZE];
4792                 printk("<%s>", bdevname(rdev->bdev,b));
4793         }
4794         printk("\n");
4795
4796         err = do_md_run(mddev);
4797         if (err) {
4798                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4799                 do_md_stop(mddev, 0, 0);
4800         }
4801 }
4802
4803 /*
4804  * lets try to run arrays based on all disks that have arrived
4805  * until now. (those are in pending_raid_disks)
4806  *
4807  * the method: pick the first pending disk, collect all disks with
4808  * the same UUID, remove all from the pending list and put them into
4809  * the 'same_array' list. Then order this list based on superblock
4810  * update time (freshest comes first), kick out 'old' disks and
4811  * compare superblocks. If everything's fine then run it.
4812  *
4813  * If "unit" is allocated, then bump its reference count
4814  */
4815 static void autorun_devices(int part)
4816 {
4817         mdk_rdev_t *rdev0, *rdev, *tmp;
4818         mddev_t *mddev;
4819         char b[BDEVNAME_SIZE];
4820
4821         printk(KERN_INFO "md: autorun ...\n");
4822         while (!list_empty(&pending_raid_disks)) {
4823                 int unit;
4824                 dev_t dev;
4825                 LIST_HEAD(candidates);
4826                 rdev0 = list_entry(pending_raid_disks.next,
4827                                          mdk_rdev_t, same_set);
4828
4829                 printk(KERN_INFO "md: considering %s ...\n",
4830                         bdevname(rdev0->bdev,b));
4831                 INIT_LIST_HEAD(&candidates);
4832                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4833                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4834                                 printk(KERN_INFO "md:  adding %s ...\n",
4835                                         bdevname(rdev->bdev,b));
4836                                 list_move(&rdev->same_set, &candidates);
4837                         }
4838                 /*
4839                  * now we have a set of devices, with all of them having
4840                  * mostly sane superblocks. It's time to allocate the
4841                  * mddev.
4842                  */
4843                 if (part) {
4844                         dev = MKDEV(mdp_major,
4845                                     rdev0->preferred_minor << MdpMinorShift);
4846                         unit = MINOR(dev) >> MdpMinorShift;
4847                 } else {
4848                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4849                         unit = MINOR(dev);
4850                 }
4851                 if (rdev0->preferred_minor != unit) {
4852                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4853                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4854                         break;
4855                 }
4856
4857                 md_probe(dev, NULL, NULL);
4858                 mddev = mddev_find(dev);
4859                 if (!mddev || !mddev->gendisk) {
4860                         if (mddev)
4861                                 mddev_put(mddev);
4862                         printk(KERN_ERR
4863                                 "md: cannot allocate memory for md drive.\n");
4864                         break;
4865                 }
4866                 if (mddev_lock(mddev))
4867                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4868                                mdname(mddev));
4869                 else if (mddev->raid_disks || mddev->major_version
4870                          || !list_empty(&mddev->disks)) {
4871                         printk(KERN_WARNING
4872                                 "md: %s already running, cannot run %s\n",
4873                                 mdname(mddev), bdevname(rdev0->bdev,b));
4874                         mddev_unlock(mddev);
4875                 } else {
4876                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4877                         mddev->persistent = 1;
4878                         rdev_for_each_list(rdev, tmp, &candidates) {
4879                                 list_del_init(&rdev->same_set);
4880                                 if (bind_rdev_to_array(rdev, mddev))
4881                                         export_rdev(rdev);
4882                         }
4883                         autorun_array(mddev);
4884                         mddev_unlock(mddev);
4885                 }
4886                 /* on success, candidates will be empty, on error
4887                  * it won't...
4888                  */
4889                 rdev_for_each_list(rdev, tmp, &candidates) {
4890                         list_del_init(&rdev->same_set);
4891                         export_rdev(rdev);
4892                 }
4893                 mddev_put(mddev);
4894         }
4895         printk(KERN_INFO "md: ... autorun DONE.\n");
4896 }
4897 #endif /* !MODULE */
4898
4899 static int get_version(void __user * arg)
4900 {
4901         mdu_version_t ver;
4902
4903         ver.major = MD_MAJOR_VERSION;
4904         ver.minor = MD_MINOR_VERSION;
4905         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4906
4907         if (copy_to_user(arg, &ver, sizeof(ver)))
4908                 return -EFAULT;
4909
4910         return 0;
4911 }
4912
4913 static int get_array_info(mddev_t * mddev, void __user * arg)
4914 {
4915         mdu_array_info_t info;
4916         int nr,working,insync,failed,spare;
4917         mdk_rdev_t *rdev;
4918
4919         nr=working=insync=failed=spare=0;
4920         list_for_each_entry(rdev, &mddev->disks, same_set) {
4921                 nr++;
4922                 if (test_bit(Faulty, &rdev->flags))
4923                         failed++;
4924                 else {
4925                         working++;
4926                         if (test_bit(In_sync, &rdev->flags))
4927                                 insync++;
4928                         else
4929                                 spare++;
4930                 }
4931         }
4932
4933         info.major_version = mddev->major_version;
4934         info.minor_version = mddev->minor_version;
4935         info.patch_version = MD_PATCHLEVEL_VERSION;
4936         info.ctime         = mddev->ctime;
4937         info.level         = mddev->level;
4938         info.size          = mddev->dev_sectors / 2;
4939         if (info.size != mddev->dev_sectors / 2) /* overflow */
4940                 info.size = -1;
4941         info.nr_disks      = nr;
4942         info.raid_disks    = mddev->raid_disks;
4943         info.md_minor      = mddev->md_minor;
4944         info.not_persistent= !mddev->persistent;
4945
4946         info.utime         = mddev->utime;
4947         info.state         = 0;
4948         if (mddev->in_sync)
4949                 info.state = (1<<MD_SB_CLEAN);
4950         if (mddev->bitmap && mddev->bitmap_info.offset)
4951                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4952         info.active_disks  = insync;
4953         info.working_disks = working;
4954         info.failed_disks  = failed;
4955         info.spare_disks   = spare;
4956
4957         info.layout        = mddev->layout;
4958         info.chunk_size    = mddev->chunk_sectors << 9;
4959
4960         if (copy_to_user(arg, &info, sizeof(info)))
4961                 return -EFAULT;
4962
4963         return 0;
4964 }
4965
4966 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4967 {
4968         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4969         char *ptr, *buf = NULL;
4970         int err = -ENOMEM;
4971
4972         if (md_allow_write(mddev))
4973                 file = kmalloc(sizeof(*file), GFP_NOIO);
4974         else
4975                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4976
4977         if (!file)
4978                 goto out;
4979
4980         /* bitmap disabled, zero the first byte and copy out */
4981         if (!mddev->bitmap || !mddev->bitmap->file) {
4982                 file->pathname[0] = '\0';
4983                 goto copy_out;
4984         }
4985
4986         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4987         if (!buf)
4988                 goto out;
4989
4990         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4991         if (IS_ERR(ptr))
4992                 goto out;
4993
4994         strcpy(file->pathname, ptr);
4995
4996 copy_out:
4997         err = 0;
4998         if (copy_to_user(arg, file, sizeof(*file)))
4999                 err = -EFAULT;
5000 out:
5001         kfree(buf);
5002         kfree(file);
5003         return err;
5004 }
5005
5006 static int get_disk_info(mddev_t * mddev, void __user * arg)
5007 {
5008         mdu_disk_info_t info;
5009         mdk_rdev_t *rdev;
5010
5011         if (copy_from_user(&info, arg, sizeof(info)))
5012                 return -EFAULT;
5013
5014         rdev = find_rdev_nr(mddev, info.number);
5015         if (rdev) {
5016                 info.major = MAJOR(rdev->bdev->bd_dev);
5017                 info.minor = MINOR(rdev->bdev->bd_dev);
5018                 info.raid_disk = rdev->raid_disk;
5019                 info.state = 0;
5020                 if (test_bit(Faulty, &rdev->flags))
5021                         info.state |= (1<<MD_DISK_FAULTY);
5022                 else if (test_bit(In_sync, &rdev->flags)) {
5023                         info.state |= (1<<MD_DISK_ACTIVE);
5024                         info.state |= (1<<MD_DISK_SYNC);
5025                 }
5026                 if (test_bit(WriteMostly, &rdev->flags))
5027                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
5028         } else {
5029                 info.major = info.minor = 0;
5030                 info.raid_disk = -1;
5031                 info.state = (1<<MD_DISK_REMOVED);
5032         }
5033
5034         if (copy_to_user(arg, &info, sizeof(info)))
5035                 return -EFAULT;
5036
5037         return 0;
5038 }
5039
5040 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5041 {
5042         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5043         mdk_rdev_t *rdev;
5044         dev_t dev = MKDEV(info->major,info->minor);
5045
5046         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5047                 return -EOVERFLOW;
5048
5049         if (!mddev->raid_disks) {
5050                 int err;
5051                 /* expecting a device which has a superblock */
5052                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5053                 if (IS_ERR(rdev)) {
5054                         printk(KERN_WARNING
5055                                 "md: md_import_device returned %ld\n",
5056                                 PTR_ERR(rdev));
5057                         return PTR_ERR(rdev);
5058                 }
5059                 if (!list_empty(&mddev->disks)) {
5060                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5061                                                         mdk_rdev_t, same_set);
5062                         err = super_types[mddev->major_version]
5063                                 .load_super(rdev, rdev0, mddev->minor_version);
5064                         if (err < 0) {
5065                                 printk(KERN_WARNING
5066                                         "md: %s has different UUID to %s\n",
5067                                         bdevname(rdev->bdev,b),
5068                                         bdevname(rdev0->bdev,b2));
5069                                 export_rdev(rdev);
5070                                 return -EINVAL;
5071                         }
5072                 }
5073                 err = bind_rdev_to_array(rdev, mddev);
5074                 if (err)
5075                         export_rdev(rdev);
5076                 return err;
5077         }
5078
5079         /*
5080          * add_new_disk can be used once the array is assembled
5081          * to add "hot spares".  They must already have a superblock
5082          * written
5083          */
5084         if (mddev->pers) {
5085                 int err;
5086                 if (!mddev->pers->hot_add_disk) {
5087                         printk(KERN_WARNING
5088                                 "%s: personality does not support diskops!\n",
5089                                mdname(mddev));
5090                         return -EINVAL;
5091                 }
5092                 if (mddev->persistent)
5093                         rdev = md_import_device(dev, mddev->major_version,
5094                                                 mddev->minor_version);
5095                 else
5096                         rdev = md_import_device(dev, -1, -1);
5097                 if (IS_ERR(rdev)) {
5098                         printk(KERN_WARNING
5099                                 "md: md_import_device returned %ld\n",
5100                                 PTR_ERR(rdev));
5101                         return PTR_ERR(rdev);
5102                 }
5103                 /* set save_raid_disk if appropriate */
5104                 if (!mddev->persistent) {
5105                         if (info->state & (1<<MD_DISK_SYNC)  &&
5106                             info->raid_disk < mddev->raid_disks)
5107                                 rdev->raid_disk = info->raid_disk;
5108                         else
5109                                 rdev->raid_disk = -1;
5110                 } else
5111                         super_types[mddev->major_version].
5112                                 validate_super(mddev, rdev);
5113                 rdev->saved_raid_disk = rdev->raid_disk;
5114
5115                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5116                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5117                         set_bit(WriteMostly, &rdev->flags);
5118                 else
5119                         clear_bit(WriteMostly, &rdev->flags);
5120
5121                 rdev->raid_disk = -1;
5122                 err = bind_rdev_to_array(rdev, mddev);
5123                 if (!err && !mddev->pers->hot_remove_disk) {
5124                         /* If there is hot_add_disk but no hot_remove_disk
5125                          * then added disks for geometry changes,
5126                          * and should be added immediately.
5127                          */
5128                         super_types[mddev->major_version].
5129                                 validate_super(mddev, rdev);
5130                         err = mddev->pers->hot_add_disk(mddev, rdev);
5131                         if (err)
5132                                 unbind_rdev_from_array(rdev);
5133                 }
5134                 if (err)
5135                         export_rdev(rdev);
5136                 else
5137                         sysfs_notify_dirent_safe(rdev->sysfs_state);
5138
5139                 md_update_sb(mddev, 1);
5140                 if (mddev->degraded)
5141                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5142                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5143                 md_wakeup_thread(mddev->thread);
5144                 return err;
5145         }
5146
5147         /* otherwise, add_new_disk is only allowed
5148          * for major_version==0 superblocks
5149          */
5150         if (mddev->major_version != 0) {
5151                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5152                        mdname(mddev));
5153                 return -EINVAL;
5154         }
5155
5156         if (!(info->state & (1<<MD_DISK_FAULTY))) {
5157                 int err;
5158                 rdev = md_import_device(dev, -1, 0);
5159                 if (IS_ERR(rdev)) {
5160                         printk(KERN_WARNING
5161                                 "md: error, md_import_device() returned %ld\n",
5162                                 PTR_ERR(rdev));
5163                         return PTR_ERR(rdev);
5164                 }
5165                 rdev->desc_nr = info->number;
5166                 if (info->raid_disk < mddev->raid_disks)
5167                         rdev->raid_disk = info->raid_disk;
5168                 else
5169                         rdev->raid_disk = -1;
5170
5171                 if (rdev->raid_disk < mddev->raid_disks)
5172                         if (info->state & (1<<MD_DISK_SYNC))
5173                                 set_bit(In_sync, &rdev->flags);
5174
5175                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5176                         set_bit(WriteMostly, &rdev->flags);
5177
5178                 if (!mddev->persistent) {
5179                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
5180                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5181                 } else
5182                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5183                 rdev->sectors = rdev->sb_start;
5184
5185                 err = bind_rdev_to_array(rdev, mddev);
5186                 if (err) {
5187                         export_rdev(rdev);
5188                         return err;
5189                 }
5190         }
5191
5192         return 0;
5193 }
5194
5195 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5196 {
5197         char b[BDEVNAME_SIZE];
5198         mdk_rdev_t *rdev;
5199
5200         rdev = find_rdev(mddev, dev);
5201         if (!rdev)
5202                 return -ENXIO;
5203
5204         if (rdev->raid_disk >= 0)
5205                 goto busy;
5206
5207         kick_rdev_from_array(rdev);
5208         md_update_sb(mddev, 1);
5209         md_new_event(mddev);
5210
5211         return 0;
5212 busy:
5213         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5214                 bdevname(rdev->bdev,b), mdname(mddev));
5215         return -EBUSY;
5216 }
5217
5218 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5219 {
5220         char b[BDEVNAME_SIZE];
5221         int err;
5222         mdk_rdev_t *rdev;
5223
5224         if (!mddev->pers)
5225                 return -ENODEV;
5226
5227         if (mddev->major_version != 0) {
5228                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5229                         " version-0 superblocks.\n",
5230                         mdname(mddev));
5231                 return -EINVAL;
5232         }
5233         if (!mddev->pers->hot_add_disk) {
5234                 printk(KERN_WARNING
5235                         "%s: personality does not support diskops!\n",
5236                         mdname(mddev));
5237                 return -EINVAL;
5238         }
5239
5240         rdev = md_import_device(dev, -1, 0);
5241         if (IS_ERR(rdev)) {
5242                 printk(KERN_WARNING
5243                         "md: error, md_import_device() returned %ld\n",
5244                         PTR_ERR(rdev));
5245                 return -EINVAL;
5246         }
5247
5248         if (mddev->persistent)
5249                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5250         else
5251                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5252
5253         rdev->sectors = rdev->sb_start;
5254
5255         if (test_bit(Faulty, &rdev->flags)) {
5256                 printk(KERN_WARNING
5257                         "md: can not hot-add faulty %s disk to %s!\n",
5258                         bdevname(rdev->bdev,b), mdname(mddev));
5259                 err = -EINVAL;
5260                 goto abort_export;
5261         }
5262         clear_bit(In_sync, &rdev->flags);
5263         rdev->desc_nr = -1;
5264         rdev->saved_raid_disk = -1;
5265         err = bind_rdev_to_array(rdev, mddev);
5266         if (err)
5267                 goto abort_export;
5268
5269         /*
5270          * The rest should better be atomic, we can have disk failures
5271          * noticed in interrupt contexts ...
5272          */
5273
5274         rdev->raid_disk = -1;
5275
5276         md_update_sb(mddev, 1);
5277
5278         /*
5279          * Kick recovery, maybe this spare has to be added to the
5280          * array immediately.
5281          */
5282         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5283         md_wakeup_thread(mddev->thread);
5284         md_new_event(mddev);
5285         return 0;
5286
5287 abort_export:
5288         export_rdev(rdev);
5289         return err;
5290 }
5291
5292 static int set_bitmap_file(mddev_t *mddev, int fd)
5293 {
5294         int err;
5295
5296         if (mddev->pers) {
5297                 if (!mddev->pers->quiesce)
5298                         return -EBUSY;
5299                 if (mddev->recovery || mddev->sync_thread)
5300                         return -EBUSY;
5301                 /* we should be able to change the bitmap.. */
5302         }
5303
5304
5305         if (fd >= 0) {
5306                 if (mddev->bitmap)
5307                         return -EEXIST; /* cannot add when bitmap is present */
5308                 mddev->bitmap_info.file = fget(fd);
5309
5310                 if (mddev->bitmap_info.file == NULL) {
5311                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5312                                mdname(mddev));
5313                         return -EBADF;
5314                 }
5315
5316                 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5317                 if (err) {
5318                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5319                                mdname(mddev));
5320                         fput(mddev->bitmap_info.file);
5321                         mddev->bitmap_info.file = NULL;
5322                         return err;
5323                 }
5324                 mddev->bitmap_info.offset = 0; /* file overrides offset */
5325         } else if (mddev->bitmap == NULL)
5326                 return -ENOENT; /* cannot remove what isn't there */
5327         err = 0;
5328         if (mddev->pers) {
5329                 mddev->pers->quiesce(mddev, 1);
5330                 if (fd >= 0)
5331                         err = bitmap_create(mddev);
5332                 if (fd < 0 || err) {
5333                         bitmap_destroy(mddev);
5334                         fd = -1; /* make sure to put the file */
5335                 }
5336                 mddev->pers->quiesce(mddev, 0);
5337         }
5338         if (fd < 0) {
5339                 if (mddev->bitmap_info.file) {
5340                         restore_bitmap_write_access(mddev->bitmap_info.file);
5341                         fput(mddev->bitmap_info.file);
5342                 }
5343                 mddev->bitmap_info.file = NULL;
5344         }
5345
5346         return err;
5347 }
5348
5349 /*
5350  * set_array_info is used two different ways
5351  * The original usage is when creating a new array.
5352  * In this usage, raid_disks is > 0 and it together with
5353  *  level, size, not_persistent,layout,chunksize determine the
5354  *  shape of the array.
5355  *  This will always create an array with a type-0.90.0 superblock.
5356  * The newer usage is when assembling an array.
5357  *  In this case raid_disks will be 0, and the major_version field is
5358  *  use to determine which style super-blocks are to be found on the devices.
5359  *  The minor and patch _version numbers are also kept incase the
5360  *  super_block handler wishes to interpret them.
5361  */
5362 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5363 {
5364
5365         if (info->raid_disks == 0) {
5366                 /* just setting version number for superblock loading */
5367                 if (info->major_version < 0 ||
5368                     info->major_version >= ARRAY_SIZE(super_types) ||
5369                     super_types[info->major_version].name == NULL) {
5370                         /* maybe try to auto-load a module? */
5371                         printk(KERN_INFO
5372                                 "md: superblock version %d not known\n",
5373                                 info->major_version);
5374                         return -EINVAL;
5375                 }
5376                 mddev->major_version = info->major_version;
5377                 mddev->minor_version = info->minor_version;
5378                 mddev->patch_version = info->patch_version;
5379                 mddev->persistent = !info->not_persistent;
5380                 /* ensure mddev_put doesn't delete this now that there
5381                  * is some minimal configuration.
5382                  */
5383                 mddev->ctime         = get_seconds();
5384                 return 0;
5385         }
5386         mddev->major_version = MD_MAJOR_VERSION;
5387         mddev->minor_version = MD_MINOR_VERSION;
5388         mddev->patch_version = MD_PATCHLEVEL_VERSION;
5389         mddev->ctime         = get_seconds();
5390
5391         mddev->level         = info->level;
5392         mddev->clevel[0]     = 0;
5393         mddev->dev_sectors   = 2 * (sector_t)info->size;
5394         mddev->raid_disks    = info->raid_disks;
5395         /* don't set md_minor, it is determined by which /dev/md* was
5396          * openned
5397          */
5398         if (info->state & (1<<MD_SB_CLEAN))
5399                 mddev->recovery_cp = MaxSector;
5400         else
5401                 mddev->recovery_cp = 0;
5402         mddev->persistent    = ! info->not_persistent;
5403         mddev->external      = 0;
5404
5405         mddev->layout        = info->layout;
5406         mddev->chunk_sectors = info->chunk_size >> 9;
5407
5408         mddev->max_disks     = MD_SB_DISKS;
5409
5410         if (mddev->persistent)
5411                 mddev->flags         = 0;
5412         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5413
5414         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5415         mddev->bitmap_info.offset = 0;
5416
5417         mddev->reshape_position = MaxSector;
5418
5419         /*
5420          * Generate a 128 bit UUID
5421          */
5422         get_random_bytes(mddev->uuid, 16);
5423
5424         mddev->new_level = mddev->level;
5425         mddev->new_chunk_sectors = mddev->chunk_sectors;
5426         mddev->new_layout = mddev->layout;
5427         mddev->delta_disks = 0;
5428
5429         return 0;
5430 }
5431
5432 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5433 {
5434         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5435
5436         if (mddev->external_size)
5437                 return;
5438
5439         mddev->array_sectors = array_sectors;
5440 }
5441 EXPORT_SYMBOL(md_set_array_sectors);
5442
5443 static int update_size(mddev_t *mddev, sector_t num_sectors)
5444 {
5445         mdk_rdev_t *rdev;
5446         int rv;
5447         int fit = (num_sectors == 0);
5448
5449         if (mddev->pers->resize == NULL)
5450                 return -EINVAL;
5451         /* The "num_sectors" is the number of sectors of each device that
5452          * is used.  This can only make sense for arrays with redundancy.
5453          * linear and raid0 always use whatever space is available. We can only
5454          * consider changing this number if no resync or reconstruction is
5455          * happening, and if the new size is acceptable. It must fit before the
5456          * sb_start or, if that is <data_offset, it must fit before the size
5457          * of each device.  If num_sectors is zero, we find the largest size
5458          * that fits.
5459
5460          */
5461         if (mddev->sync_thread)
5462                 return -EBUSY;
5463         if (mddev->bitmap)
5464                 /* Sorry, cannot grow a bitmap yet, just remove it,
5465                  * grow, and re-add.
5466                  */
5467                 return -EBUSY;
5468         list_for_each_entry(rdev, &mddev->disks, same_set) {
5469                 sector_t avail = rdev->sectors;
5470
5471                 if (fit && (num_sectors == 0 || num_sectors > avail))
5472                         num_sectors = avail;
5473                 if (avail < num_sectors)
5474                         return -ENOSPC;
5475         }
5476         rv = mddev->pers->resize(mddev, num_sectors);
5477         if (!rv)
5478                 revalidate_disk(mddev->gendisk);
5479         return rv;
5480 }
5481
5482 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5483 {
5484         int rv;
5485         /* change the number of raid disks */
5486         if (mddev->pers->check_reshape == NULL)
5487                 return -EINVAL;
5488         if (raid_disks <= 0 ||
5489             (mddev->max_disks && raid_disks >= mddev->max_disks))
5490                 return -EINVAL;
5491         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5492                 return -EBUSY;
5493         mddev->delta_disks = raid_disks - mddev->raid_disks;
5494
5495         rv = mddev->pers->check_reshape(mddev);
5496         return rv;
5497 }
5498
5499
5500 /*
5501  * update_array_info is used to change the configuration of an
5502  * on-line array.
5503  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5504  * fields in the info are checked against the array.
5505  * Any differences that cannot be handled will cause an error.
5506  * Normally, only one change can be managed at a time.
5507  */
5508 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5509 {
5510         int rv = 0;
5511         int cnt = 0;
5512         int state = 0;
5513
5514         /* calculate expected state,ignoring low bits */
5515         if (mddev->bitmap && mddev->bitmap_info.offset)
5516                 state |= (1 << MD_SB_BITMAP_PRESENT);
5517
5518         if (mddev->major_version != info->major_version ||
5519             mddev->minor_version != info->minor_version ||
5520 /*          mddev->patch_version != info->patch_version || */
5521             mddev->ctime         != info->ctime         ||
5522             mddev->level         != info->level         ||
5523 /*          mddev->layout        != info->layout        || */
5524             !mddev->persistent   != info->not_persistent||
5525             mddev->chunk_sectors != info->chunk_size >> 9 ||
5526             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5527             ((state^info->state) & 0xfffffe00)
5528                 )
5529                 return -EINVAL;
5530         /* Check there is only one change */
5531         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5532                 cnt++;
5533         if (mddev->raid_disks != info->raid_disks)
5534                 cnt++;
5535         if (mddev->layout != info->layout)
5536                 cnt++;
5537         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5538                 cnt++;
5539         if (cnt == 0)
5540                 return 0;
5541         if (cnt > 1)
5542                 return -EINVAL;
5543
5544         if (mddev->layout != info->layout) {
5545                 /* Change layout
5546                  * we don't need to do anything at the md level, the
5547                  * personality will take care of it all.
5548                  */
5549                 if (mddev->pers->check_reshape == NULL)
5550                         return -EINVAL;
5551                 else {
5552                         mddev->new_layout = info->layout;
5553                         rv = mddev->pers->check_reshape(mddev);
5554                         if (rv)
5555                                 mddev->new_layout = mddev->layout;
5556                         return rv;
5557                 }
5558         }
5559         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5560                 rv = update_size(mddev, (sector_t)info->size * 2);
5561
5562         if (mddev->raid_disks    != info->raid_disks)
5563                 rv = update_raid_disks(mddev, info->raid_disks);
5564
5565         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5566                 if (mddev->pers->quiesce == NULL)
5567                         return -EINVAL;
5568                 if (mddev->recovery || mddev->sync_thread)
5569                         return -EBUSY;
5570                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5571                         /* add the bitmap */
5572                         if (mddev->bitmap)
5573                                 return -EEXIST;
5574                         if (mddev->bitmap_info.default_offset == 0)
5575                                 return -EINVAL;
5576                         mddev->bitmap_info.offset =
5577                                 mddev->bitmap_info.default_offset;
5578                         mddev->pers->quiesce(mddev, 1);
5579                         rv = bitmap_create(mddev);
5580                         if (rv)
5581                                 bitmap_destroy(mddev);
5582                         mddev->pers->quiesce(mddev, 0);
5583                 } else {
5584                         /* remove the bitmap */
5585                         if (!mddev->bitmap)
5586                                 return -ENOENT;
5587                         if (mddev->bitmap->file)
5588                                 return -EINVAL;
5589                         mddev->pers->quiesce(mddev, 1);
5590                         bitmap_destroy(mddev);
5591                         mddev->pers->quiesce(mddev, 0);
5592                         mddev->bitmap_info.offset = 0;
5593                 }
5594         }
5595         md_update_sb(mddev, 1);
5596         return rv;
5597 }
5598
5599 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5600 {
5601         mdk_rdev_t *rdev;
5602
5603         if (mddev->pers == NULL)
5604                 return -ENODEV;
5605
5606         rdev = find_rdev(mddev, dev);
5607         if (!rdev)
5608                 return -ENODEV;
5609
5610         md_error(mddev, rdev);
5611         return 0;
5612 }
5613
5614 /*
5615  * We have a problem here : there is no easy way to give a CHS
5616  * virtual geometry. We currently pretend that we have a 2 heads
5617  * 4 sectors (with a BIG number of cylinders...). This drives
5618  * dosfs just mad... ;-)
5619  */
5620 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5621 {
5622         mddev_t *mddev = bdev->bd_disk->private_data;
5623
5624         geo->heads = 2;
5625         geo->sectors = 4;
5626         geo->cylinders = mddev->array_sectors / 8;
5627         return 0;
5628 }
5629
5630 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5631                         unsigned int cmd, unsigned long arg)
5632 {
5633         int err = 0;
5634         void __user *argp = (void __user *)arg;
5635         mddev_t *mddev = NULL;
5636         int ro;
5637
5638         if (!capable(CAP_SYS_ADMIN))
5639                 return -EACCES;
5640
5641         /*
5642          * Commands dealing with the RAID driver but not any
5643          * particular array:
5644          */
5645         switch (cmd)
5646         {
5647                 case RAID_VERSION:
5648                         err = get_version(argp);
5649                         goto done;
5650
5651                 case PRINT_RAID_DEBUG:
5652                         err = 0;
5653                         md_print_devices();
5654                         goto done;
5655
5656 #ifndef MODULE
5657                 case RAID_AUTORUN:
5658                         err = 0;
5659                         autostart_arrays(arg);
5660                         goto done;
5661 #endif
5662                 default:;
5663         }
5664
5665         /*
5666          * Commands creating/starting a new array:
5667          */
5668
5669         mddev = bdev->bd_disk->private_data;
5670
5671         if (!mddev) {
5672                 BUG();
5673                 goto abort;
5674         }
5675
5676         err = mddev_lock(mddev);
5677         if (err) {
5678                 printk(KERN_INFO
5679                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5680                         err, cmd);
5681                 goto abort;
5682         }
5683
5684         switch (cmd)
5685         {
5686                 case SET_ARRAY_INFO:
5687                         {
5688                                 mdu_array_info_t info;
5689                                 if (!arg)
5690                                         memset(&info, 0, sizeof(info));
5691                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5692                                         err = -EFAULT;
5693                                         goto abort_unlock;
5694                                 }
5695                                 if (mddev->pers) {
5696                                         err = update_array_info(mddev, &info);
5697                                         if (err) {
5698                                                 printk(KERN_WARNING "md: couldn't update"
5699                                                        " array info. %d\n", err);
5700                                                 goto abort_unlock;
5701                                         }
5702                                         goto done_unlock;
5703                                 }
5704                                 if (!list_empty(&mddev->disks)) {
5705                                         printk(KERN_WARNING
5706                                                "md: array %s already has disks!\n",
5707                                                mdname(mddev));
5708                                         err = -EBUSY;
5709                                         goto abort_unlock;
5710                                 }
5711                                 if (mddev->raid_disks) {
5712                                         printk(KERN_WARNING
5713                                                "md: array %s already initialised!\n",
5714                                                mdname(mddev));
5715                                         err = -EBUSY;
5716                                         goto abort_unlock;
5717                                 }
5718                                 err = set_array_info(mddev, &info);
5719                                 if (err) {
5720                                         printk(KERN_WARNING "md: couldn't set"
5721                                                " array info. %d\n", err);
5722                                         goto abort_unlock;
5723                                 }
5724                         }
5725                         goto done_unlock;
5726
5727                 default:;
5728         }
5729
5730         /*
5731          * Commands querying/configuring an existing array:
5732          */
5733         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5734          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5735         if ((!mddev->raid_disks && !mddev->external)
5736             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5737             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5738             && cmd != GET_BITMAP_FILE) {
5739                 err = -ENODEV;
5740                 goto abort_unlock;
5741         }
5742
5743         /*
5744          * Commands even a read-only array can execute:
5745          */
5746         switch (cmd)
5747         {
5748                 case GET_ARRAY_INFO:
5749                         err = get_array_info(mddev, argp);
5750                         goto done_unlock;
5751
5752                 case GET_BITMAP_FILE:
5753                         err = get_bitmap_file(mddev, argp);
5754                         goto done_unlock;
5755
5756                 case GET_DISK_INFO:
5757                         err = get_disk_info(mddev, argp);
5758                         goto done_unlock;
5759
5760                 case RESTART_ARRAY_RW:
5761                         err = restart_array(mddev);
5762                         goto done_unlock;
5763
5764                 case STOP_ARRAY:
5765                         err = do_md_stop(mddev, 0, 1);
5766                         goto done_unlock;
5767
5768                 case STOP_ARRAY_RO:
5769                         err = md_set_readonly(mddev, 1);
5770                         goto done_unlock;
5771
5772                 case BLKROSET:
5773                         if (get_user(ro, (int __user *)(arg))) {
5774                                 err = -EFAULT;
5775                                 goto done_unlock;
5776                         }
5777                         err = -EINVAL;
5778
5779                         /* if the bdev is going readonly the value of mddev->ro
5780                          * does not matter, no writes are coming
5781                          */
5782                         if (ro)
5783                                 goto done_unlock;
5784
5785                         /* are we are already prepared for writes? */
5786                         if (mddev->ro != 1)
5787                                 goto done_unlock;
5788
5789                         /* transitioning to readauto need only happen for
5790                          * arrays that call md_write_start
5791                          */
5792                         if (mddev->pers) {
5793                                 err = restart_array(mddev);
5794                                 if (err == 0) {
5795                                         mddev->ro = 2;
5796                                         set_disk_ro(mddev->gendisk, 0);
5797                                 }
5798                         }
5799                         goto done_unlock;
5800         }
5801
5802         /*
5803          * The remaining ioctls are changing the state of the
5804          * superblock, so we do not allow them on read-only arrays.
5805          * However non-MD ioctls (e.g. get-size) will still come through
5806          * here and hit the 'default' below, so only disallow
5807          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5808          */
5809         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5810                 if (mddev->ro == 2) {
5811                         mddev->ro = 0;
5812                         sysfs_notify_dirent_safe(mddev->sysfs_state);
5813                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5814                         md_wakeup_thread(mddev->thread);
5815                 } else {
5816                         err = -EROFS;
5817                         goto abort_unlock;
5818                 }
5819         }
5820
5821         switch (cmd)
5822         {
5823                 case ADD_NEW_DISK:
5824                 {
5825                         mdu_disk_info_t info;
5826                         if (copy_from_user(&info, argp, sizeof(info)))
5827                                 err = -EFAULT;
5828                         else
5829                                 err = add_new_disk(mddev, &info);
5830                         goto done_unlock;
5831                 }
5832
5833                 case HOT_REMOVE_DISK:
5834                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5835                         goto done_unlock;
5836
5837                 case HOT_ADD_DISK:
5838                         err = hot_add_disk(mddev, new_decode_dev(arg));
5839                         goto done_unlock;
5840
5841                 case SET_DISK_FAULTY:
5842                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5843                         goto done_unlock;
5844
5845                 case RUN_ARRAY:
5846                         err = do_md_run(mddev);
5847                         goto done_unlock;
5848
5849                 case SET_BITMAP_FILE:
5850                         err = set_bitmap_file(mddev, (int)arg);
5851                         goto done_unlock;
5852
5853                 default:
5854                         err = -EINVAL;
5855                         goto abort_unlock;
5856         }
5857
5858 done_unlock:
5859 abort_unlock:
5860         if (mddev->hold_active == UNTIL_IOCTL &&
5861             err != -EINVAL)
5862                 mddev->hold_active = 0;
5863         mddev_unlock(mddev);
5864
5865         return err;
5866 done:
5867         if (err)
5868                 MD_BUG();
5869 abort:
5870         return err;
5871 }
5872 #ifdef CONFIG_COMPAT
5873 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5874                     unsigned int cmd, unsigned long arg)
5875 {
5876         switch (cmd) {
5877         case HOT_REMOVE_DISK:
5878         case HOT_ADD_DISK:
5879         case SET_DISK_FAULTY:
5880         case SET_BITMAP_FILE:
5881                 /* These take in integer arg, do not convert */
5882                 break;
5883         default:
5884                 arg = (unsigned long)compat_ptr(arg);
5885                 break;
5886         }
5887
5888         return md_ioctl(bdev, mode, cmd, arg);
5889 }
5890 #endif /* CONFIG_COMPAT */
5891
5892 static int md_open(struct block_device *bdev, fmode_t mode)
5893 {
5894         /*
5895          * Succeed if we can lock the mddev, which confirms that
5896          * it isn't being stopped right now.
5897          */
5898         mddev_t *mddev = mddev_find(bdev->bd_dev);
5899         int err;
5900
5901         if (mddev->gendisk != bdev->bd_disk) {
5902                 /* we are racing with mddev_put which is discarding this
5903                  * bd_disk.
5904                  */
5905                 mddev_put(mddev);
5906                 /* Wait until bdev->bd_disk is definitely gone */
5907                 flush_scheduled_work();
5908                 /* Then retry the open from the top */
5909                 return -ERESTARTSYS;
5910         }
5911         BUG_ON(mddev != bdev->bd_disk->private_data);
5912
5913         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5914                 goto out;
5915
5916         err = 0;
5917         atomic_inc(&mddev->openers);
5918         mutex_unlock(&mddev->open_mutex);
5919
5920         check_disk_size_change(mddev->gendisk, bdev);
5921  out:
5922         return err;
5923 }
5924
5925 static int md_release(struct gendisk *disk, fmode_t mode)
5926 {
5927         mddev_t *mddev = disk->private_data;
5928
5929         BUG_ON(!mddev);
5930         atomic_dec(&mddev->openers);
5931         mddev_put(mddev);
5932
5933         return 0;
5934 }
5935 static const struct block_device_operations md_fops =
5936 {
5937         .owner          = THIS_MODULE,
5938         .open           = md_open,
5939         .release        = md_release,
5940         .ioctl          = md_ioctl,
5941 #ifdef CONFIG_COMPAT
5942         .compat_ioctl   = md_compat_ioctl,
5943 #endif
5944         .getgeo         = md_getgeo,
5945 };
5946
5947 static int md_thread(void * arg)
5948 {
5949         mdk_thread_t *thread = arg;
5950
5951         /*
5952          * md_thread is a 'system-thread', it's priority should be very
5953          * high. We avoid resource deadlocks individually in each
5954          * raid personality. (RAID5 does preallocation) We also use RR and
5955          * the very same RT priority as kswapd, thus we will never get
5956          * into a priority inversion deadlock.
5957          *
5958          * we definitely have to have equal or higher priority than
5959          * bdflush, otherwise bdflush will deadlock if there are too
5960          * many dirty RAID5 blocks.
5961          */
5962
5963         allow_signal(SIGKILL);
5964         while (!kthread_should_stop()) {
5965
5966                 /* We need to wait INTERRUPTIBLE so that
5967                  * we don't add to the load-average.
5968                  * That means we need to be sure no signals are
5969                  * pending
5970                  */
5971                 if (signal_pending(current))
5972                         flush_signals(current);
5973
5974                 wait_event_interruptible_timeout
5975                         (thread->wqueue,
5976                          test_bit(THREAD_WAKEUP, &thread->flags)
5977                          || kthread_should_stop(),
5978                          thread->timeout);
5979
5980                 clear_bit(THREAD_WAKEUP, &thread->flags);
5981
5982                 thread->run(thread->mddev);
5983         }
5984
5985         return 0;
5986 }
5987
5988 void md_wakeup_thread(mdk_thread_t *thread)
5989 {
5990         if (thread) {
5991                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5992                 set_bit(THREAD_WAKEUP, &thread->flags);
5993                 wake_up(&thread->wqueue);
5994         }
5995 }
5996
5997 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5998                                  const char *name)
5999 {
6000         mdk_thread_t *thread;
6001
6002         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6003         if (!thread)
6004                 return NULL;
6005
6006         init_waitqueue_head(&thread->wqueue);
6007
6008         thread->run = run;
6009         thread->mddev = mddev;
6010         thread->timeout = MAX_SCHEDULE_TIMEOUT;
6011         thread->tsk = kthread_run(md_thread, thread,
6012                                   "%s_%s",
6013                                   mdname(thread->mddev),
6014                                   name ?: mddev->pers->name);
6015         if (IS_ERR(thread->tsk)) {
6016                 kfree(thread);
6017                 return NULL;
6018         }
6019         return thread;
6020 }
6021
6022 void md_unregister_thread(mdk_thread_t *thread)
6023 {
6024         if (!thread)
6025                 return;
6026         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6027
6028         kthread_stop(thread->tsk);
6029         kfree(thread);
6030 }
6031
6032 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6033 {
6034         if (!mddev) {
6035                 MD_BUG();
6036                 return;
6037         }
6038
6039         if (!rdev || test_bit(Faulty, &rdev->flags))
6040                 return;
6041
6042         if (mddev->external)
6043                 set_bit(Blocked, &rdev->flags);
6044 /*
6045         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6046                 mdname(mddev),
6047                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6048                 __builtin_return_address(0),__builtin_return_address(1),
6049                 __builtin_return_address(2),__builtin_return_address(3));
6050 */
6051         if (!mddev->pers)
6052                 return;
6053         if (!mddev->pers->error_handler)
6054                 return;
6055         mddev->pers->error_handler(mddev,rdev);
6056         if (mddev->degraded)
6057                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6058         sysfs_notify_dirent_safe(rdev->sysfs_state);
6059         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6060         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6061         md_wakeup_thread(mddev->thread);
6062         md_new_event_inintr(mddev);
6063 }
6064
6065 /* seq_file implementation /proc/mdstat */
6066
6067 static void status_unused(struct seq_file *seq)
6068 {
6069         int i = 0;
6070         mdk_rdev_t *rdev;
6071
6072         seq_printf(seq, "unused devices: ");
6073
6074         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6075                 char b[BDEVNAME_SIZE];
6076                 i++;
6077                 seq_printf(seq, "%s ",
6078                               bdevname(rdev->bdev,b));
6079         }
6080         if (!i)
6081                 seq_printf(seq, "<none>");
6082
6083         seq_printf(seq, "\n");
6084 }
6085
6086
6087 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6088 {
6089         sector_t max_sectors, resync, res;
6090         unsigned long dt, db;
6091         sector_t rt;
6092         int scale;
6093         unsigned int per_milli;
6094
6095         resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6096
6097         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6098                 max_sectors = mddev->resync_max_sectors;
6099         else
6100                 max_sectors = mddev->dev_sectors;
6101
6102         /*
6103          * Should not happen.
6104          */
6105         if (!max_sectors) {
6106                 MD_BUG();
6107                 return;
6108         }
6109         /* Pick 'scale' such that (resync>>scale)*1000 will fit
6110          * in a sector_t, and (max_sectors>>scale) will fit in a
6111          * u32, as those are the requirements for sector_div.
6112          * Thus 'scale' must be at least 10
6113          */
6114         scale = 10;
6115         if (sizeof(sector_t) > sizeof(unsigned long)) {
6116                 while ( max_sectors/2 > (1ULL<<(scale+32)))
6117                         scale++;
6118         }
6119         res = (resync>>scale)*1000;
6120         sector_div(res, (u32)((max_sectors>>scale)+1));
6121
6122         per_milli = res;
6123         {
6124                 int i, x = per_milli/50, y = 20-x;
6125                 seq_printf(seq, "[");
6126                 for (i = 0; i < x; i++)
6127                         seq_printf(seq, "=");
6128                 seq_printf(seq, ">");
6129                 for (i = 0; i < y; i++)
6130                         seq_printf(seq, ".");
6131                 seq_printf(seq, "] ");
6132         }
6133         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6134                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6135                     "reshape" :
6136                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6137                      "check" :
6138                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6139                       "resync" : "recovery"))),
6140                    per_milli/10, per_milli % 10,
6141                    (unsigned long long) resync/2,
6142                    (unsigned long long) max_sectors/2);
6143
6144         /*
6145          * dt: time from mark until now
6146          * db: blocks written from mark until now
6147          * rt: remaining time
6148          *
6149          * rt is a sector_t, so could be 32bit or 64bit.
6150          * So we divide before multiply in case it is 32bit and close
6151          * to the limit.
6152          * We scale the divisor (db) by 32 to avoid loosing precision
6153          * near the end of resync when the number of remaining sectors
6154          * is close to 'db'.
6155          * We then divide rt by 32 after multiplying by db to compensate.
6156          * The '+1' avoids division by zero if db is very small.
6157          */
6158         dt = ((jiffies - mddev->resync_mark) / HZ);
6159         if (!dt) dt++;
6160         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6161                 - mddev->resync_mark_cnt;
6162
6163         rt = max_sectors - resync;    /* number of remaining sectors */
6164         sector_div(rt, db/32+1);
6165         rt *= dt;
6166         rt >>= 5;
6167
6168         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6169                    ((unsigned long)rt % 60)/6);
6170
6171         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6172 }
6173
6174 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6175 {
6176         struct list_head *tmp;
6177         loff_t l = *pos;
6178         mddev_t *mddev;
6179
6180         if (l >= 0x10000)
6181                 return NULL;
6182         if (!l--)
6183                 /* header */
6184                 return (void*)1;
6185
6186         spin_lock(&all_mddevs_lock);
6187         list_for_each(tmp,&all_mddevs)
6188                 if (!l--) {
6189                         mddev = list_entry(tmp, mddev_t, all_mddevs);
6190                         mddev_get(mddev);
6191                         spin_unlock(&all_mddevs_lock);
6192                         return mddev;
6193                 }
6194         spin_unlock(&all_mddevs_lock);
6195         if (!l--)
6196                 return (void*)2;/* tail */
6197         return NULL;
6198 }
6199
6200 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6201 {
6202         struct list_head *tmp;
6203         mddev_t *next_mddev, *mddev = v;
6204
6205         ++*pos;
6206         if (v == (void*)2)
6207                 return NULL;
6208
6209         spin_lock(&all_mddevs_lock);
6210         if (v == (void*)1)
6211                 tmp = all_mddevs.next;
6212         else
6213                 tmp = mddev->all_mddevs.next;
6214         if (tmp != &all_mddevs)
6215                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6216         else {
6217                 next_mddev = (void*)2;
6218                 *pos = 0x10000;
6219         }
6220         spin_unlock(&all_mddevs_lock);
6221
6222         if (v != (void*)1)
6223                 mddev_put(mddev);
6224         return next_mddev;
6225
6226 }
6227
6228 static void md_seq_stop(struct seq_file *seq, void *v)
6229 {
6230         mddev_t *mddev = v;
6231
6232         if (mddev && v != (void*)1 && v != (void*)2)
6233                 mddev_put(mddev);
6234 }
6235
6236 struct mdstat_info {
6237         int event;
6238 };
6239
6240 static int md_seq_show(struct seq_file *seq, void *v)
6241 {
6242         mddev_t *mddev = v;
6243         sector_t sectors;
6244         mdk_rdev_t *rdev;
6245         struct mdstat_info *mi = seq->private;
6246         struct bitmap *bitmap;
6247
6248         if (v == (void*)1) {
6249                 struct mdk_personality *pers;
6250                 seq_printf(seq, "Personalities : ");
6251                 spin_lock(&pers_lock);
6252                 list_for_each_entry(pers, &pers_list, list)
6253                         seq_printf(seq, "[%s] ", pers->name);
6254
6255                 spin_unlock(&pers_lock);
6256                 seq_printf(seq, "\n");
6257                 mi->event = atomic_read(&md_event_count);
6258                 return 0;
6259         }
6260         if (v == (void*)2) {
6261                 status_unused(seq);
6262                 return 0;
6263         }
6264
6265         if (mddev_lock(mddev) < 0)
6266                 return -EINTR;
6267
6268         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6269                 seq_printf(seq, "%s : %sactive", mdname(mddev),
6270                                                 mddev->pers ? "" : "in");
6271                 if (mddev->pers) {
6272                         if (mddev->ro==1)
6273                                 seq_printf(seq, " (read-only)");
6274                         if (mddev->ro==2)
6275                                 seq_printf(seq, " (auto-read-only)");
6276                         seq_printf(seq, " %s", mddev->pers->name);
6277                 }
6278
6279                 sectors = 0;
6280                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6281                         char b[BDEVNAME_SIZE];
6282                         seq_printf(seq, " %s[%d]",
6283                                 bdevname(rdev->bdev,b), rdev->desc_nr);
6284                         if (test_bit(WriteMostly, &rdev->flags))
6285                                 seq_printf(seq, "(W)");
6286                         if (test_bit(Faulty, &rdev->flags)) {
6287                                 seq_printf(seq, "(F)");
6288                                 continue;
6289                         } else if (rdev->raid_disk < 0)
6290                                 seq_printf(seq, "(S)"); /* spare */
6291                         sectors += rdev->sectors;
6292                 }
6293
6294                 if (!list_empty(&mddev->disks)) {
6295                         if (mddev->pers)
6296                                 seq_printf(seq, "\n      %llu blocks",
6297                                            (unsigned long long)
6298                                            mddev->array_sectors / 2);
6299                         else
6300                                 seq_printf(seq, "\n      %llu blocks",
6301                                            (unsigned long long)sectors / 2);
6302                 }
6303                 if (mddev->persistent) {
6304                         if (mddev->major_version != 0 ||
6305                             mddev->minor_version != 90) {
6306                                 seq_printf(seq," super %d.%d",
6307                                            mddev->major_version,
6308                                            mddev->minor_version);
6309                         }
6310                 } else if (mddev->external)
6311                         seq_printf(seq, " super external:%s",
6312                                    mddev->metadata_type);
6313                 else
6314                         seq_printf(seq, " super non-persistent");
6315
6316                 if (mddev->pers) {
6317                         mddev->pers->status(seq, mddev);
6318                         seq_printf(seq, "\n      ");
6319                         if (mddev->pers->sync_request) {
6320                                 if (mddev->curr_resync > 2) {
6321                                         status_resync(seq, mddev);
6322                                         seq_printf(seq, "\n      ");
6323                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6324                                         seq_printf(seq, "\tresync=DELAYED\n      ");
6325                                 else if (mddev->recovery_cp < MaxSector)
6326                                         seq_printf(seq, "\tresync=PENDING\n      ");
6327                         }
6328                 } else
6329                         seq_printf(seq, "\n       ");
6330
6331                 if ((bitmap = mddev->bitmap)) {
6332                         unsigned long chunk_kb;
6333                         unsigned long flags;
6334                         spin_lock_irqsave(&bitmap->lock, flags);
6335                         chunk_kb = mddev->bitmap_info.chunksize >> 10;
6336                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6337                                 "%lu%s chunk",
6338                                 bitmap->pages - bitmap->missing_pages,
6339                                 bitmap->pages,
6340                                 (bitmap->pages - bitmap->missing_pages)
6341                                         << (PAGE_SHIFT - 10),
6342                                 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6343                                 chunk_kb ? "KB" : "B");
6344                         if (bitmap->file) {
6345                                 seq_printf(seq, ", file: ");
6346                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
6347                         }
6348
6349                         seq_printf(seq, "\n");
6350                         spin_unlock_irqrestore(&bitmap->lock, flags);
6351                 }
6352
6353                 seq_printf(seq, "\n");
6354         }
6355         mddev_unlock(mddev);
6356
6357         return 0;
6358 }
6359
6360 static const struct seq_operations md_seq_ops = {
6361         .start  = md_seq_start,
6362         .next   = md_seq_next,
6363         .stop   = md_seq_stop,
6364         .show   = md_seq_show,
6365 };
6366
6367 static int md_seq_open(struct inode *inode, struct file *file)
6368 {
6369         int error;
6370         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6371         if (mi == NULL)
6372                 return -ENOMEM;
6373
6374         error = seq_open(file, &md_seq_ops);
6375         if (error)
6376                 kfree(mi);
6377         else {
6378                 struct seq_file *p = file->private_data;
6379                 p->private = mi;
6380                 mi->event = atomic_read(&md_event_count);
6381         }
6382         return error;
6383 }
6384
6385 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6386 {
6387         struct seq_file *m = filp->private_data;
6388         struct mdstat_info *mi = m->private;
6389         int mask;
6390
6391         poll_wait(filp, &md_event_waiters, wait);
6392
6393         /* always allow read */
6394         mask = POLLIN | POLLRDNORM;
6395
6396         if (mi->event != atomic_read(&md_event_count))
6397                 mask |= POLLERR | POLLPRI;
6398         return mask;
6399 }
6400
6401 static const struct file_operations md_seq_fops = {
6402         .owner          = THIS_MODULE,
6403         .open           = md_seq_open,
6404         .read           = seq_read,
6405         .llseek         = seq_lseek,
6406         .release        = seq_release_private,
6407         .poll           = mdstat_poll,
6408 };
6409
6410 int register_md_personality(struct mdk_personality *p)
6411 {
6412         spin_lock(&pers_lock);
6413         list_add_tail(&p->list, &pers_list);
6414         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6415         spin_unlock(&pers_lock);
6416         return 0;
6417 }
6418
6419 int unregister_md_personality(struct mdk_personality *p)
6420 {
6421         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6422         spin_lock(&pers_lock);
6423         list_del_init(&p->list);
6424         spin_unlock(&pers_lock);
6425         return 0;
6426 }
6427
6428 static int is_mddev_idle(mddev_t *mddev, int init)
6429 {
6430         mdk_rdev_t * rdev;
6431         int idle;
6432         int curr_events;
6433
6434         idle = 1;
6435         rcu_read_lock();
6436         rdev_for_each_rcu(rdev, mddev) {
6437                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6438                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6439                               (int)part_stat_read(&disk->part0, sectors[1]) -
6440                               atomic_read(&disk->sync_io);
6441                 /* sync IO will cause sync_io to increase before the disk_stats
6442                  * as sync_io is counted when a request starts, and
6443                  * disk_stats is counted when it completes.
6444                  * So resync activity will cause curr_events to be smaller than
6445                  * when there was no such activity.
6446                  * non-sync IO will cause disk_stat to increase without
6447                  * increasing sync_io so curr_events will (eventually)
6448                  * be larger than it was before.  Once it becomes
6449                  * substantially larger, the test below will cause
6450                  * the array to appear non-idle, and resync will slow
6451                  * down.
6452                  * If there is a lot of outstanding resync activity when
6453                  * we set last_event to curr_events, then all that activity
6454                  * completing might cause the array to appear non-idle
6455                  * and resync will be slowed down even though there might
6456                  * not have been non-resync activity.  This will only
6457                  * happen once though.  'last_events' will soon reflect
6458                  * the state where there is little or no outstanding
6459                  * resync requests, and further resync activity will
6460                  * always make curr_events less than last_events.
6461                  *
6462                  */
6463                 if (init || curr_events - rdev->last_events > 64) {
6464                         rdev->last_events = curr_events;
6465                         idle = 0;
6466                 }
6467         }
6468         rcu_read_unlock();
6469         return idle;
6470 }
6471
6472 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6473 {
6474         /* another "blocks" (512byte) blocks have been synced */
6475         atomic_sub(blocks, &mddev->recovery_active);
6476         wake_up(&mddev->recovery_wait);
6477         if (!ok) {
6478                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6479                 md_wakeup_thread(mddev->thread);
6480                 // stop recovery, signal do_sync ....
6481         }
6482 }
6483
6484
6485 /* md_write_start(mddev, bi)
6486  * If we need to update some array metadata (e.g. 'active' flag
6487  * in superblock) before writing, schedule a superblock update
6488  * and wait for it to complete.
6489  */
6490 void md_write_start(mddev_t *mddev, struct bio *bi)
6491 {
6492         int did_change = 0;
6493         if (bio_data_dir(bi) != WRITE)
6494                 return;
6495
6496         BUG_ON(mddev->ro == 1);
6497         if (mddev->ro == 2) {
6498                 /* need to switch to read/write */
6499                 mddev->ro = 0;
6500                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6501                 md_wakeup_thread(mddev->thread);
6502                 md_wakeup_thread(mddev->sync_thread);
6503                 did_change = 1;
6504         }
6505         atomic_inc(&mddev->writes_pending);
6506         if (mddev->safemode == 1)
6507                 mddev->safemode = 0;
6508         if (mddev->in_sync) {
6509                 spin_lock_irq(&mddev->write_lock);
6510                 if (mddev->in_sync) {
6511                         mddev->in_sync = 0;
6512                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6513                         md_wakeup_thread(mddev->thread);
6514                         did_change = 1;
6515                 }
6516                 spin_unlock_irq(&mddev->write_lock);
6517         }
6518         if (did_change)
6519                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6520         wait_event(mddev->sb_wait,
6521                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6522                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6523 }
6524
6525 void md_write_end(mddev_t *mddev)
6526 {
6527         if (atomic_dec_and_test(&mddev->writes_pending)) {
6528                 if (mddev->safemode == 2)
6529                         md_wakeup_thread(mddev->thread);
6530                 else if (mddev->safemode_delay)
6531                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6532         }
6533 }
6534
6535 /* md_allow_write(mddev)
6536  * Calling this ensures that the array is marked 'active' so that writes
6537  * may proceed without blocking.  It is important to call this before
6538  * attempting a GFP_KERNEL allocation while holding the mddev lock.
6539  * Must be called with mddev_lock held.
6540  *
6541  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6542  * is dropped, so return -EAGAIN after notifying userspace.
6543  */
6544 int md_allow_write(mddev_t *mddev)
6545 {
6546         if (!mddev->pers)
6547                 return 0;
6548         if (mddev->ro)
6549                 return 0;
6550         if (!mddev->pers->sync_request)
6551                 return 0;
6552
6553         spin_lock_irq(&mddev->write_lock);
6554         if (mddev->in_sync) {
6555                 mddev->in_sync = 0;
6556                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6557                 if (mddev->safemode_delay &&
6558                     mddev->safemode == 0)
6559                         mddev->safemode = 1;
6560                 spin_unlock_irq(&mddev->write_lock);
6561                 md_update_sb(mddev, 0);
6562                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6563         } else
6564                 spin_unlock_irq(&mddev->write_lock);
6565
6566         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6567                 return -EAGAIN;
6568         else
6569                 return 0;
6570 }
6571 EXPORT_SYMBOL_GPL(md_allow_write);
6572
6573 #define SYNC_MARKS      10
6574 #define SYNC_MARK_STEP  (3*HZ)
6575 void md_do_sync(mddev_t *mddev)
6576 {
6577         mddev_t *mddev2;
6578         unsigned int currspeed = 0,
6579                  window;
6580         sector_t max_sectors,j, io_sectors;
6581         unsigned long mark[SYNC_MARKS];
6582         sector_t mark_cnt[SYNC_MARKS];
6583         int last_mark,m;
6584         struct list_head *tmp;
6585         sector_t last_check;
6586         int skipped = 0;
6587         mdk_rdev_t *rdev;
6588         char *desc;
6589
6590         /* just incase thread restarts... */
6591         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6592                 return;
6593         if (mddev->ro) /* never try to sync a read-only array */
6594                 return;
6595
6596         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6597                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6598                         desc = "data-check";
6599                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6600                         desc = "requested-resync";
6601                 else
6602                         desc = "resync";
6603         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6604                 desc = "reshape";
6605         else
6606                 desc = "recovery";
6607
6608         /* we overload curr_resync somewhat here.
6609          * 0 == not engaged in resync at all
6610          * 2 == checking that there is no conflict with another sync
6611          * 1 == like 2, but have yielded to allow conflicting resync to
6612          *              commense
6613          * other == active in resync - this many blocks
6614          *
6615          * Before starting a resync we must have set curr_resync to
6616          * 2, and then checked that every "conflicting" array has curr_resync
6617          * less than ours.  When we find one that is the same or higher
6618          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
6619          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6620          * This will mean we have to start checking from the beginning again.
6621          *
6622          */
6623
6624         do {
6625                 mddev->curr_resync = 2;
6626
6627         try_again:
6628                 if (kthread_should_stop())
6629                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6630
6631                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6632                         goto skip;
6633                 for_each_mddev(mddev2, tmp) {
6634                         if (mddev2 == mddev)
6635                                 continue;
6636                         if (!mddev->parallel_resync
6637                         &&  mddev2->curr_resync
6638                         &&  match_mddev_units(mddev, mddev2)) {
6639                                 DEFINE_WAIT(wq);
6640                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
6641                                         /* arbitrarily yield */
6642                                         mddev->curr_resync = 1;
6643                                         wake_up(&resync_wait);
6644                                 }
6645                                 if (mddev > mddev2 && mddev->curr_resync == 1)
6646                                         /* no need to wait here, we can wait the next
6647                                          * time 'round when curr_resync == 2
6648                                          */
6649                                         continue;
6650                                 /* We need to wait 'interruptible' so as not to
6651                                  * contribute to the load average, and not to
6652                                  * be caught by 'softlockup'
6653                                  */
6654                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6655                                 if (!kthread_should_stop() &&
6656                                     mddev2->curr_resync >= mddev->curr_resync) {
6657                                         printk(KERN_INFO "md: delaying %s of %s"
6658                                                " until %s has finished (they"
6659                                                " share one or more physical units)\n",
6660                                                desc, mdname(mddev), mdname(mddev2));
6661                                         mddev_put(mddev2);
6662                                         if (signal_pending(current))
6663                                                 flush_signals(current);
6664                                         schedule();
6665                                         finish_wait(&resync_wait, &wq);
6666                                         goto try_again;
6667                                 }
6668                                 finish_wait(&resync_wait, &wq);
6669                         }
6670                 }
6671         } while (mddev->curr_resync < 2);
6672
6673         j = 0;
6674         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6675                 /* resync follows the size requested by the personality,
6676                  * which defaults to physical size, but can be virtual size
6677                  */
6678                 max_sectors = mddev->resync_max_sectors;
6679                 mddev->resync_mismatches = 0;
6680                 /* we don't use the checkpoint if there's a bitmap */
6681                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6682                         j = mddev->resync_min;
6683                 else if (!mddev->bitmap)
6684                         j = mddev->recovery_cp;
6685
6686         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6687                 max_sectors = mddev->dev_sectors;
6688         else {
6689                 /* recovery follows the physical size of devices */
6690                 max_sectors = mddev->dev_sectors;
6691                 j = MaxSector;
6692                 rcu_read_lock();
6693                 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6694                         if (rdev->raid_disk >= 0 &&
6695                             !test_bit(Faulty, &rdev->flags) &&
6696                             !test_bit(In_sync, &rdev->flags) &&
6697                             rdev->recovery_offset < j)
6698                                 j = rdev->recovery_offset;
6699                 rcu_read_unlock();
6700         }
6701
6702         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6703         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
6704                 " %d KB/sec/disk.\n", speed_min(mddev));
6705         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6706                "(but not more than %d KB/sec) for %s.\n",
6707                speed_max(mddev), desc);
6708
6709         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6710
6711         io_sectors = 0;
6712         for (m = 0; m < SYNC_MARKS; m++) {
6713                 mark[m] = jiffies;
6714                 mark_cnt[m] = io_sectors;
6715         }
6716         last_mark = 0;
6717         mddev->resync_mark = mark[last_mark];
6718         mddev->resync_mark_cnt = mark_cnt[last_mark];
6719
6720         /*
6721          * Tune reconstruction:
6722          */
6723         window = 32*(PAGE_SIZE/512);
6724         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6725                 window/2,(unsigned long long) max_sectors/2);
6726
6727         atomic_set(&mddev->recovery_active, 0);
6728         last_check = 0;
6729
6730         if (j>2) {
6731                 printk(KERN_INFO
6732                        "md: resuming %s of %s from checkpoint.\n",
6733                        desc, mdname(mddev));
6734                 mddev->curr_resync = j;
6735         }
6736         mddev->curr_resync_completed = mddev->curr_resync;
6737
6738         while (j < max_sectors) {
6739                 sector_t sectors;
6740
6741                 skipped = 0;
6742
6743                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6744                     ((mddev->curr_resync > mddev->curr_resync_completed &&
6745                       (mddev->curr_resync - mddev->curr_resync_completed)
6746                       > (max_sectors >> 4)) ||
6747                      (j - mddev->curr_resync_completed)*2
6748                      >= mddev->resync_max - mddev->curr_resync_completed
6749                             )) {
6750                         /* time to update curr_resync_completed */
6751                         blk_unplug(mddev->queue);
6752                         wait_event(mddev->recovery_wait,
6753                                    atomic_read(&mddev->recovery_active) == 0);
6754                         mddev->curr_resync_completed =
6755                                 mddev->curr_resync;
6756                         if (mddev->persistent)
6757                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6758                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6759                 }
6760
6761                 while (j >= mddev->resync_max && !kthread_should_stop()) {
6762                         /* As this condition is controlled by user-space,
6763                          * we can block indefinitely, so use '_interruptible'
6764                          * to avoid triggering warnings.
6765                          */
6766                         flush_signals(current); /* just in case */
6767                         wait_event_interruptible(mddev->recovery_wait,
6768                                                  mddev->resync_max > j
6769                                                  || kthread_should_stop());
6770                 }
6771
6772                 if (kthread_should_stop())
6773                         goto interrupted;
6774
6775                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6776                                                   currspeed < speed_min(mddev));
6777                 if (sectors == 0) {
6778                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6779                         goto out;
6780                 }
6781
6782                 if (!skipped) { /* actual IO requested */
6783                         io_sectors += sectors;
6784                         atomic_add(sectors, &mddev->recovery_active);
6785                 }
6786
6787                 j += sectors;
6788                 if (j>1) mddev->curr_resync = j;
6789                 mddev->curr_mark_cnt = io_sectors;
6790                 if (last_check == 0)
6791                         /* this is the earliers that rebuilt will be
6792                          * visible in /proc/mdstat
6793                          */
6794                         md_new_event(mddev);
6795
6796                 if (last_check + window > io_sectors || j == max_sectors)
6797                         continue;
6798
6799                 last_check = io_sectors;
6800
6801                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6802                         break;
6803
6804         repeat:
6805                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6806                         /* step marks */
6807                         int next = (last_mark+1) % SYNC_MARKS;
6808
6809                         mddev->resync_mark = mark[next];
6810                         mddev->resync_mark_cnt = mark_cnt[next];
6811                         mark[next] = jiffies;
6812                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6813                         last_mark = next;
6814                 }
6815
6816
6817                 if (kthread_should_stop())
6818                         goto interrupted;
6819
6820
6821                 /*
6822                  * this loop exits only if either when we are slower than
6823                  * the 'hard' speed limit, or the system was IO-idle for
6824                  * a jiffy.
6825                  * the system might be non-idle CPU-wise, but we only care
6826                  * about not overloading the IO subsystem. (things like an
6827                  * e2fsck being done on the RAID array should execute fast)
6828                  */
6829                 blk_unplug(mddev->queue);
6830                 cond_resched();
6831
6832                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6833                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6834
6835                 if (currspeed > speed_min(mddev)) {
6836                         if ((currspeed > speed_max(mddev)) ||
6837                                         !is_mddev_idle(mddev, 0)) {
6838                                 msleep(500);
6839                                 goto repeat;
6840                         }
6841                 }
6842         }
6843         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6844         /*
6845          * this also signals 'finished resyncing' to md_stop
6846          */
6847  out:
6848         blk_unplug(mddev->queue);
6849
6850         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6851
6852         /* tell personality that we are finished */
6853         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6854
6855         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6856             mddev->curr_resync > 2) {
6857                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6858                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6859                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6860                                         printk(KERN_INFO
6861                                                "md: checkpointing %s of %s.\n",
6862                                                desc, mdname(mddev));
6863                                         mddev->recovery_cp = mddev->curr_resync;
6864                                 }
6865                         } else
6866                                 mddev->recovery_cp = MaxSector;
6867                 } else {
6868                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6869                                 mddev->curr_resync = MaxSector;
6870                         rcu_read_lock();
6871                         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6872                                 if (rdev->raid_disk >= 0 &&
6873                                     mddev->delta_disks >= 0 &&
6874                                     !test_bit(Faulty, &rdev->flags) &&
6875                                     !test_bit(In_sync, &rdev->flags) &&
6876                                     rdev->recovery_offset < mddev->curr_resync)
6877                                         rdev->recovery_offset = mddev->curr_resync;
6878                         rcu_read_unlock();
6879                 }
6880         }
6881         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6882
6883  skip:
6884         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6885                 /* We completed so min/max setting can be forgotten if used. */
6886                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6887                         mddev->resync_min = 0;
6888                 mddev->resync_max = MaxSector;
6889         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6890                 mddev->resync_min = mddev->curr_resync_completed;
6891         mddev->curr_resync = 0;
6892         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6893                 mddev->curr_resync_completed = 0;
6894         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6895         wake_up(&resync_wait);
6896         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6897         md_wakeup_thread(mddev->thread);
6898         return;
6899
6900  interrupted:
6901         /*
6902          * got a signal, exit.
6903          */
6904         printk(KERN_INFO
6905                "md: md_do_sync() got signal ... exiting\n");
6906         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6907         goto out;
6908
6909 }
6910 EXPORT_SYMBOL_GPL(md_do_sync);
6911
6912
6913 static int remove_and_add_spares(mddev_t *mddev)
6914 {
6915         mdk_rdev_t *rdev;
6916         int spares = 0;
6917
6918         mddev->curr_resync_completed = 0;
6919
6920         list_for_each_entry(rdev, &mddev->disks, same_set)
6921                 if (rdev->raid_disk >= 0 &&
6922                     !test_bit(Blocked, &rdev->flags) &&
6923                     (test_bit(Faulty, &rdev->flags) ||
6924                      ! test_bit(In_sync, &rdev->flags)) &&
6925                     atomic_read(&rdev->nr_pending)==0) {
6926                         if (mddev->pers->hot_remove_disk(
6927                                     mddev, rdev->raid_disk)==0) {
6928                                 char nm[20];
6929                                 sprintf(nm,"rd%d", rdev->raid_disk);
6930                                 sysfs_remove_link(&mddev->kobj, nm);
6931                                 rdev->raid_disk = -1;
6932                         }
6933                 }
6934
6935         if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6936                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6937                         if (rdev->raid_disk >= 0 &&
6938                             !test_bit(In_sync, &rdev->flags) &&
6939                             !test_bit(Blocked, &rdev->flags))
6940                                 spares++;
6941                         if (rdev->raid_disk < 0
6942                             && !test_bit(Faulty, &rdev->flags)) {
6943                                 rdev->recovery_offset = 0;
6944                                 if (mddev->pers->
6945                                     hot_add_disk(mddev, rdev) == 0) {
6946                                         char nm[20];
6947                                         sprintf(nm, "rd%d", rdev->raid_disk);
6948                                         if (sysfs_create_link(&mddev->kobj,
6949                                                               &rdev->kobj, nm))
6950                                                 /* failure here is OK */;
6951                                         spares++;
6952                                         md_new_event(mddev);
6953                                         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6954                                 } else
6955                                         break;
6956                         }
6957                 }
6958         }
6959         return spares;
6960 }
6961 /*
6962  * This routine is regularly called by all per-raid-array threads to
6963  * deal with generic issues like resync and super-block update.
6964  * Raid personalities that don't have a thread (linear/raid0) do not
6965  * need this as they never do any recovery or update the superblock.
6966  *
6967  * It does not do any resync itself, but rather "forks" off other threads
6968  * to do that as needed.
6969  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6970  * "->recovery" and create a thread at ->sync_thread.
6971  * When the thread finishes it sets MD_RECOVERY_DONE
6972  * and wakeups up this thread which will reap the thread and finish up.
6973  * This thread also removes any faulty devices (with nr_pending == 0).
6974  *
6975  * The overall approach is:
6976  *  1/ if the superblock needs updating, update it.
6977  *  2/ If a recovery thread is running, don't do anything else.
6978  *  3/ If recovery has finished, clean up, possibly marking spares active.
6979  *  4/ If there are any faulty devices, remove them.
6980  *  5/ If array is degraded, try to add spares devices
6981  *  6/ If array has spares or is not in-sync, start a resync thread.
6982  */
6983 void md_check_recovery(mddev_t *mddev)
6984 {
6985         mdk_rdev_t *rdev;
6986
6987
6988         if (mddev->bitmap)
6989                 bitmap_daemon_work(mddev);
6990
6991         if (mddev->ro)
6992                 return;
6993
6994         if (signal_pending(current)) {
6995                 if (mddev->pers->sync_request && !mddev->external) {
6996                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6997                                mdname(mddev));
6998                         mddev->safemode = 2;
6999                 }
7000                 flush_signals(current);
7001         }
7002
7003         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7004                 return;
7005         if ( ! (
7006                 (mddev->flags && !mddev->external) ||
7007                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7008                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7009                 (mddev->external == 0 && mddev->safemode == 1) ||
7010                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7011                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7012                 ))
7013                 return;
7014
7015         if (mddev_trylock(mddev)) {
7016                 int spares = 0;
7017
7018                 if (mddev->ro) {
7019                         /* Only thing we do on a ro array is remove
7020                          * failed devices.
7021                          */
7022                         remove_and_add_spares(mddev);
7023                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7024                         goto unlock;
7025                 }
7026
7027                 if (!mddev->external) {
7028                         int did_change = 0;
7029                         spin_lock_irq(&mddev->write_lock);
7030                         if (mddev->safemode &&
7031                             !atomic_read(&mddev->writes_pending) &&
7032                             !mddev->in_sync &&
7033                             mddev->recovery_cp == MaxSector) {
7034                                 mddev->in_sync = 1;
7035                                 did_change = 1;
7036                                 if (mddev->persistent)
7037                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7038                         }
7039                         if (mddev->safemode == 1)
7040                                 mddev->safemode = 0;
7041                         spin_unlock_irq(&mddev->write_lock);
7042                         if (did_change)
7043                                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7044                 }
7045
7046                 if (mddev->flags)
7047                         md_update_sb(mddev, 0);
7048
7049                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7050                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7051                         /* resync/recovery still happening */
7052                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7053                         goto unlock;
7054                 }
7055                 if (mddev->sync_thread) {
7056                         /* resync has finished, collect result */
7057                         md_unregister_thread(mddev->sync_thread);
7058                         mddev->sync_thread = NULL;
7059                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7060                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7061                                 /* success...*/
7062                                 /* activate any spares */
7063                                 if (mddev->pers->spare_active(mddev))
7064                                         sysfs_notify(&mddev->kobj, NULL,
7065                                                      "degraded");
7066                         }
7067                         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7068                             mddev->pers->finish_reshape)
7069                                 mddev->pers->finish_reshape(mddev);
7070                         md_update_sb(mddev, 1);
7071
7072                         /* if array is no-longer degraded, then any saved_raid_disk
7073                          * information must be scrapped
7074                          */
7075                         if (!mddev->degraded)
7076                                 list_for_each_entry(rdev, &mddev->disks, same_set)
7077                                         rdev->saved_raid_disk = -1;
7078
7079                         mddev->recovery = 0;
7080                         /* flag recovery needed just to double check */
7081                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7082                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7083                         md_new_event(mddev);
7084                         goto unlock;
7085                 }
7086                 /* Set RUNNING before clearing NEEDED to avoid
7087                  * any transients in the value of "sync_action".
7088                  */
7089                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7090                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7091                 /* Clear some bits that don't mean anything, but
7092                  * might be left set
7093                  */
7094                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7095                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7096
7097                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7098                         goto unlock;
7099                 /* no recovery is running.
7100                  * remove any failed drives, then
7101                  * add spares if possible.
7102                  * Spare are also removed and re-added, to allow
7103                  * the personality to fail the re-add.
7104                  */
7105
7106                 if (mddev->reshape_position != MaxSector) {
7107                         if (mddev->pers->check_reshape == NULL ||
7108                             mddev->pers->check_reshape(mddev) != 0)
7109                                 /* Cannot proceed */
7110                                 goto unlock;
7111                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7112                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7113                 } else if ((spares = remove_and_add_spares(mddev))) {
7114                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7115                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7116                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7117                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7118                 } else if (mddev->recovery_cp < MaxSector) {
7119                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7120                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7121                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7122                         /* nothing to be done ... */
7123                         goto unlock;
7124
7125                 if (mddev->pers->sync_request) {
7126                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7127                                 /* We are adding a device or devices to an array
7128                                  * which has the bitmap stored on all devices.
7129                                  * So make sure all bitmap pages get written
7130                                  */
7131                                 bitmap_write_all(mddev->bitmap);
7132                         }
7133                         mddev->sync_thread = md_register_thread(md_do_sync,
7134                                                                 mddev,
7135                                                                 "resync");
7136                         if (!mddev->sync_thread) {
7137                                 printk(KERN_ERR "%s: could not start resync"
7138                                         " thread...\n",
7139                                         mdname(mddev));
7140                                 /* leave the spares where they are, it shouldn't hurt */
7141                                 mddev->recovery = 0;
7142                         } else
7143                                 md_wakeup_thread(mddev->sync_thread);
7144                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7145                         md_new_event(mddev);
7146                 }
7147         unlock:
7148                 if (!mddev->sync_thread) {
7149                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7150                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7151                                                &mddev->recovery))
7152                                 if (mddev->sysfs_action)
7153                                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7154                 }
7155                 mddev_unlock(mddev);
7156         }
7157 }
7158
7159 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7160 {
7161         sysfs_notify_dirent_safe(rdev->sysfs_state);
7162         wait_event_timeout(rdev->blocked_wait,
7163                            !test_bit(Blocked, &rdev->flags),
7164                            msecs_to_jiffies(5000));
7165         rdev_dec_pending(rdev, mddev);
7166 }
7167 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7168
7169 static int md_notify_reboot(struct notifier_block *this,
7170                             unsigned long code, void *x)
7171 {
7172         struct list_head *tmp;
7173         mddev_t *mddev;
7174
7175         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7176
7177                 printk(KERN_INFO "md: stopping all md devices.\n");
7178
7179                 for_each_mddev(mddev, tmp)
7180                         if (mddev_trylock(mddev)) {
7181                                 /* Force a switch to readonly even array
7182                                  * appears to still be in use.  Hence
7183                                  * the '100'.
7184                                  */
7185                                 md_set_readonly(mddev, 100);
7186                                 mddev_unlock(mddev);
7187                         }
7188                 /*
7189                  * certain more exotic SCSI devices are known to be
7190                  * volatile wrt too early system reboots. While the
7191                  * right place to handle this issue is the given
7192                  * driver, we do want to have a safe RAID driver ...
7193                  */
7194                 mdelay(1000*1);
7195         }
7196         return NOTIFY_DONE;
7197 }
7198
7199 static struct notifier_block md_notifier = {
7200         .notifier_call  = md_notify_reboot,
7201         .next           = NULL,
7202         .priority       = INT_MAX, /* before any real devices */
7203 };
7204
7205 static void md_geninit(void)
7206 {
7207         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7208
7209         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7210 }
7211
7212 static int __init md_init(void)
7213 {
7214         if (register_blkdev(MD_MAJOR, "md"))
7215                 return -1;
7216         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7217                 unregister_blkdev(MD_MAJOR, "md");
7218                 return -1;
7219         }
7220         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7221                             md_probe, NULL, NULL);
7222         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7223                             md_probe, NULL, NULL);
7224
7225         register_reboot_notifier(&md_notifier);
7226         raid_table_header = register_sysctl_table(raid_root_table);
7227
7228         md_geninit();
7229         return 0;
7230 }
7231
7232
7233 #ifndef MODULE
7234
7235 /*
7236  * Searches all registered partitions for autorun RAID arrays
7237  * at boot time.
7238  */
7239
7240 static LIST_HEAD(all_detected_devices);
7241 struct detected_devices_node {
7242         struct list_head list;
7243         dev_t dev;
7244 };
7245
7246 void md_autodetect_dev(dev_t dev)
7247 {
7248         struct detected_devices_node *node_detected_dev;
7249
7250         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7251         if (node_detected_dev) {
7252                 node_detected_dev->dev = dev;
7253                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7254         } else {
7255                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7256                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7257         }
7258 }
7259
7260
7261 static void autostart_arrays(int part)
7262 {
7263         mdk_rdev_t *rdev;
7264         struct detected_devices_node *node_detected_dev;
7265         dev_t dev;
7266         int i_scanned, i_passed;
7267
7268         i_scanned = 0;
7269         i_passed = 0;
7270
7271         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7272
7273         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7274                 i_scanned++;
7275                 node_detected_dev = list_entry(all_detected_devices.next,
7276                                         struct detected_devices_node, list);
7277                 list_del(&node_detected_dev->list);
7278                 dev = node_detected_dev->dev;
7279                 kfree(node_detected_dev);
7280                 rdev = md_import_device(dev,0, 90);
7281                 if (IS_ERR(rdev))
7282                         continue;
7283
7284                 if (test_bit(Faulty, &rdev->flags)) {
7285                         MD_BUG();
7286                         continue;
7287                 }
7288                 set_bit(AutoDetected, &rdev->flags);
7289                 list_add(&rdev->same_set, &pending_raid_disks);
7290                 i_passed++;
7291         }
7292
7293         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7294                                                 i_scanned, i_passed);
7295
7296         autorun_devices(part);
7297 }
7298
7299 #endif /* !MODULE */
7300
7301 static __exit void md_exit(void)
7302 {
7303         mddev_t *mddev;
7304         struct list_head *tmp;
7305
7306         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7307         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7308
7309         unregister_blkdev(MD_MAJOR,"md");
7310         unregister_blkdev(mdp_major, "mdp");
7311         unregister_reboot_notifier(&md_notifier);
7312         unregister_sysctl_table(raid_table_header);
7313         remove_proc_entry("mdstat", NULL);
7314         for_each_mddev(mddev, tmp) {
7315                 export_array(mddev);
7316                 mddev->hold_active = 0;
7317         }
7318 }
7319
7320 subsys_initcall(md_init);
7321 module_exit(md_exit)
7322
7323 static int get_ro(char *buffer, struct kernel_param *kp)
7324 {
7325         return sprintf(buffer, "%d", start_readonly);
7326 }
7327 static int set_ro(const char *val, struct kernel_param *kp)
7328 {
7329         char *e;
7330         int num = simple_strtoul(val, &e, 10);
7331         if (*val && (*e == '\0' || *e == '\n')) {
7332                 start_readonly = num;
7333                 return 0;
7334         }
7335         return -EINVAL;
7336 }
7337
7338 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7339 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7340
7341 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7342
7343 EXPORT_SYMBOL(register_md_personality);
7344 EXPORT_SYMBOL(unregister_md_personality);
7345 EXPORT_SYMBOL(md_error);
7346 EXPORT_SYMBOL(md_done_sync);
7347 EXPORT_SYMBOL(md_write_start);
7348 EXPORT_SYMBOL(md_write_end);
7349 EXPORT_SYMBOL(md_register_thread);
7350 EXPORT_SYMBOL(md_unregister_thread);
7351 EXPORT_SYMBOL(md_wakeup_thread);
7352 EXPORT_SYMBOL(md_check_recovery);
7353 MODULE_LICENSE("GPL");
7354 MODULE_DESCRIPTION("MD RAID framework");
7355 MODULE_ALIAS("md");
7356 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);